Merge remote branch 'origin/master' into radeon-rewrite

Conflicts:
	configure.ac
	src/mesa/drivers/dri/r200/r200_context.c
	src/mesa/drivers/dri/r300/r300_render.c

214 files changed, 69940 insertions, 0 deletions

diff --git a/src/gallium/auxiliary/Makefile b/src/gallium/auxiliary/Makefile
new file mode 100644
index 00000000000..eaa0f2fe4e8
--- /dev/null
+++ b/src/gallium/auxiliary/Makefile
@@ -0,0 +1,20 @@
+TOP = ../../..
+include $(TOP)/configs/current
+
+
+SUBDIRS = $(GALLIUM_AUXILIARY_DIRS)
+
+
+default: subdirs
+
+
+subdirs:
+	@for dir in $(SUBDIRS) ; do \
+		if [ -d $$dir ] ; then \
+			(cd $$dir && $(MAKE)) || exit 1 ; \
+		fi \
+	done
+
+
+clean:
+	rm -f `find . -name \*.[oa]`
diff --git a/src/gallium/auxiliary/cso_cache/Makefile b/src/gallium/auxiliary/cso_cache/Makefile
new file mode 100644
index 00000000000..6bd66020887
--- /dev/null
+++ b/src/gallium/auxiliary/cso_cache/Makefile
@@ -0,0 +1,14 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = cso_cache
+
+C_SOURCES = \
+	cso_context.c \
+	cso_cache.c \
+	cso_hash.c
+
+include ../../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/auxiliary/cso_cache/SConscript b/src/gallium/auxiliary/cso_cache/SConscript
new file mode 100644
index 00000000000..651e68a191a
--- /dev/null
+++ b/src/gallium/auxiliary/cso_cache/SConscript
@@ -0,0 +1,11 @@
+Import('*')
+
+cso_cache = env.ConvenienceLibrary(
+	target = 'cso_cache',
+	source = [
+		'cso_context.c',
+		'cso_cache.c',
+		'cso_hash.c',
+	])
+
+auxiliaries.insert(0, cso_cache)
diff --git a/src/gallium/auxiliary/cso_cache/cso_cache.c b/src/gallium/auxiliary/cso_cache/cso_cache.c
new file mode 100644
index 00000000000..6b1754ea002
--- /dev/null
+++ b/src/gallium/auxiliary/cso_cache/cso_cache.c
@@ -0,0 +1,406 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/* Authors:  Zack Rusin <[email protected]>
+ */
+
+#include "pipe/p_debug.h"
+
+#include "util/u_memory.h"
+
+#include "cso_cache.h"
+#include "cso_hash.h"
+
+
+struct cso_cache {
+   struct cso_hash *blend_hash;
+   struct cso_hash *depth_stencil_hash;
+   struct cso_hash *fs_hash;
+   struct cso_hash *vs_hash;
+   struct cso_hash *rasterizer_hash;
+   struct cso_hash *sampler_hash;
+   int    max_size;
+
+   cso_sanitize_callback sanitize_cb;
+   void                 *sanitize_data;
+};
+
+#if 1
+static unsigned hash_key(const void *key, unsigned key_size)
+{
+   unsigned *ikey = (unsigned *)key;
+   unsigned hash = 0, i;
+
+   assert(key_size % 4 == 0);
+
+   /* I'm sure this can be improved on:
+    */
+   for (i = 0; i < key_size/4; i++)
+      hash ^= ikey[i];
+
+   return hash;
+}
+#else
+static unsigned hash_key(const unsigned char *p, int n)
+{
+   unsigned h = 0;
+   unsigned g;
+
+   while (n--) {
+      h = (h << 4) + *p++;
+      if ((g = (h & 0xf0000000)) != 0)
+         h ^= g >> 23;
+      h &= ~g;
+   }
+   return h;
+}
+#endif
+
+unsigned cso_construct_key(void *item, int item_size)
+{
+   return hash_key((item), item_size);
+}
+
+static struct cso_hash *_cso_hash_for_type(struct cso_cache *sc, enum cso_cache_type type)
+{
+   struct cso_hash *hash = 0;
+
+   switch(type) {
+   case CSO_BLEND:
+      hash = sc->blend_hash;
+      break;
+   case CSO_SAMPLER:
+      hash = sc->sampler_hash;
+      break;
+   case CSO_DEPTH_STENCIL_ALPHA:
+      hash = sc->depth_stencil_hash;
+      break;
+   case CSO_RASTERIZER:
+      hash = sc->rasterizer_hash;
+      break;
+   case CSO_FRAGMENT_SHADER:
+      hash = sc->fs_hash;
+      break;
+   case CSO_VERTEX_SHADER:
+      hash = sc->vs_hash;
+      break;
+   }
+
+   return hash;
+}
+
+static int _cso_size_for_type(enum cso_cache_type type)
+{
+   switch(type) {
+   case CSO_BLEND:
+      return sizeof(struct pipe_blend_state);
+   case CSO_SAMPLER:
+      return sizeof(struct pipe_sampler_state);
+   case CSO_DEPTH_STENCIL_ALPHA:
+      return sizeof(struct pipe_depth_stencil_alpha_state);
+   case CSO_RASTERIZER:
+      return sizeof(struct pipe_rasterizer_state);
+   case CSO_FRAGMENT_SHADER:
+      return sizeof(struct pipe_shader_state);
+   case CSO_VERTEX_SHADER:
+      return sizeof(struct pipe_shader_state);
+   }
+   return 0;
+}
+
+
+static void delete_blend_state(void *state, void *data)
+{
+   struct cso_blend *cso = (struct cso_blend *)state;
+   if (cso->delete_state)
+      cso->delete_state(cso->context, cso->data);
+   FREE(state);
+}
+
+static void delete_depth_stencil_state(void *state, void *data)
+{
+   struct cso_depth_stencil_alpha *cso = (struct cso_depth_stencil_alpha *)state;
+   if (cso->delete_state)
+      cso->delete_state(cso->context, cso->data);
+   FREE(state);
+}
+
+static void delete_sampler_state(void *state, void *data)
+{
+   struct cso_sampler *cso = (struct cso_sampler *)state;
+   if (cso->delete_state)
+      cso->delete_state(cso->context, cso->data);
+   FREE(state);
+}
+
+static void delete_rasterizer_state(void *state, void *data)
+{
+   struct cso_rasterizer *cso = (struct cso_rasterizer *)state;
+   if (cso->delete_state)
+      cso->delete_state(cso->context, cso->data);
+   FREE(state);
+}
+
+static void delete_fs_state(void *state, void *data)
+{
+   struct cso_fragment_shader *cso = (struct cso_fragment_shader *)state;
+   if (cso->delete_state)
+      cso->delete_state(cso->context, cso->data);
+   FREE(state);
+}
+
+static void delete_vs_state(void *state, void *data)
+{
+   struct cso_vertex_shader *cso = (struct cso_vertex_shader *)state;
+   if (cso->delete_state)
+      cso->delete_state(cso->context, cso->data);
+   FREE(state);
+}
+
+
+static INLINE void delete_cso(void *state, enum cso_cache_type type)
+{
+   switch (type) {
+   case CSO_BLEND:
+      delete_blend_state(state, 0);
+      break;
+   case CSO_SAMPLER:
+      delete_sampler_state(state, 0);
+      break;
+   case CSO_DEPTH_STENCIL_ALPHA:
+      delete_depth_stencil_state(state, 0);
+      break;
+   case CSO_RASTERIZER:
+      delete_rasterizer_state(state, 0);
+      break;
+   case CSO_FRAGMENT_SHADER:
+      delete_fs_state(state, 0);
+      break;
+   case CSO_VERTEX_SHADER:
+      delete_vs_state(state, 0);
+      break;
+   default:
+      assert(0);
+      FREE(state);
+   }
+}
+
+
+static INLINE void sanitize_hash(struct cso_cache *sc,
+                                 struct cso_hash *hash,
+                                 enum cso_cache_type type,
+                                 int max_size)
+{
+   if (sc->sanitize_cb)
+      sc->sanitize_cb(hash, type, max_size, sc->sanitize_data);
+}
+
+
+static INLINE void sanitize_cb(struct cso_hash *hash, enum cso_cache_type type,
+                               int max_size, void *user_data)
+{
+   /* if we're approach the maximum size, remove fourth of the entries
+    * otherwise every subsequent call will go through the same */
+   int hash_size = cso_hash_size(hash);
+   int max_entries = (max_size > hash_size) ? max_size : hash_size;
+   int to_remove =  (max_size < max_entries) * max_entries/4;
+   if (hash_size > max_size)
+      to_remove += hash_size - max_size;
+   while (to_remove) {
+      /*remove elements until we're good */
+      /*fixme: currently we pick the nodes to remove at random*/
+      struct cso_hash_iter iter = cso_hash_first_node(hash);
+      void  *cso = cso_hash_take(hash, cso_hash_iter_key(iter));
+      delete_cso(cso, type);
+      --to_remove;
+   }
+}
+
+struct cso_hash_iter
+cso_insert_state(struct cso_cache *sc,
+                 unsigned hash_key, enum cso_cache_type type,
+                 void *state)
+{
+   struct cso_hash *hash = _cso_hash_for_type(sc, type);
+   sanitize_hash(sc, hash, type, sc->max_size);
+
+   return cso_hash_insert(hash, hash_key, state);
+}
+
+struct cso_hash_iter
+cso_find_state(struct cso_cache *sc,
+               unsigned hash_key, enum cso_cache_type type)
+{
+   struct cso_hash *hash = _cso_hash_for_type(sc, type);
+
+   return cso_hash_find(hash, hash_key);
+}
+
+
+void *cso_hash_find_data_from_template( struct cso_hash *hash,
+				        unsigned hash_key, 
+				        void *templ,
+				        int size )
+{
+   struct cso_hash_iter iter = cso_hash_find(hash, hash_key);
+   while (!cso_hash_iter_is_null(iter)) {
+      void *iter_data = cso_hash_iter_data(iter);
+      if (!memcmp(iter_data, templ, size)) {
+	 /* We found a match
+	  */
+         return iter_data;
+      }
+      iter = cso_hash_iter_next(iter);
+   }
+   return NULL;
+}
+
+
+struct cso_hash_iter cso_find_state_template(struct cso_cache *sc,
+                                             unsigned hash_key, enum cso_cache_type type,
+                                             void *templ)
+{
+   struct cso_hash_iter iter = cso_find_state(sc, hash_key, type);
+   int size = _cso_size_for_type(type);
+   while (!cso_hash_iter_is_null(iter)) {
+      void *iter_data = cso_hash_iter_data(iter);
+      if (!memcmp(iter_data, templ, size))
+         return iter;
+      iter = cso_hash_iter_next(iter);
+   }
+   return iter;
+}
+
+void * cso_take_state(struct cso_cache *sc,
+                      unsigned hash_key, enum cso_cache_type type)
+{
+   struct cso_hash *hash = _cso_hash_for_type(sc, type);
+   return cso_hash_take(hash, hash_key);
+}
+
+struct cso_cache *cso_cache_create(void)
+{
+   struct cso_cache *sc = MALLOC_STRUCT(cso_cache);
+   if (sc == NULL)
+      return NULL;
+
+   sc->max_size           = 4096;
+   sc->blend_hash         = cso_hash_create();
+   sc->sampler_hash       = cso_hash_create();
+   sc->depth_stencil_hash = cso_hash_create();
+   sc->rasterizer_hash    = cso_hash_create();
+   sc->fs_hash            = cso_hash_create();
+   sc->vs_hash            = cso_hash_create();
+   sc->sanitize_cb        = sanitize_cb;
+   sc->sanitize_data      = 0;
+
+   return sc;
+}
+
+void cso_for_each_state(struct cso_cache *sc, enum cso_cache_type type,
+                        cso_state_callback func, void *user_data)
+{
+   struct cso_hash *hash = 0;
+   struct cso_hash_iter iter;
+
+   switch (type) {
+   case CSO_BLEND:
+      hash = sc->blend_hash;
+      break;
+   case CSO_SAMPLER:
+      hash = sc->sampler_hash;
+      break;
+   case CSO_DEPTH_STENCIL_ALPHA:
+      hash = sc->depth_stencil_hash;
+      break;
+   case CSO_RASTERIZER:
+      hash = sc->rasterizer_hash;
+      break;
+   case CSO_FRAGMENT_SHADER:
+      hash = sc->fs_hash;
+      break;
+   case CSO_VERTEX_SHADER:
+      hash = sc->vs_hash;
+      break;
+   }
+
+   iter = cso_hash_first_node(hash);
+   while (!cso_hash_iter_is_null(iter)) {
+      void *state = cso_hash_iter_data(iter);
+      iter = cso_hash_iter_next(iter);
+      if (state) {
+         func(state, user_data);
+      }
+   }
+}
+
+void cso_cache_delete(struct cso_cache *sc)
+{
+   assert(sc);
+   /* delete driver data */
+   cso_for_each_state(sc, CSO_BLEND, delete_blend_state, 0);
+   cso_for_each_state(sc, CSO_DEPTH_STENCIL_ALPHA, delete_depth_stencil_state, 0);
+   cso_for_each_state(sc, CSO_FRAGMENT_SHADER, delete_fs_state, 0);
+   cso_for_each_state(sc, CSO_VERTEX_SHADER, delete_vs_state, 0);
+   cso_for_each_state(sc, CSO_RASTERIZER, delete_rasterizer_state, 0);
+   cso_for_each_state(sc, CSO_SAMPLER, delete_sampler_state, 0);
+
+   cso_hash_delete(sc->blend_hash);
+   cso_hash_delete(sc->sampler_hash);
+   cso_hash_delete(sc->depth_stencil_hash);
+   cso_hash_delete(sc->rasterizer_hash);
+   cso_hash_delete(sc->fs_hash);
+   cso_hash_delete(sc->vs_hash);
+   FREE(sc);
+}
+
+void cso_set_maximum_cache_size(struct cso_cache *sc, int number)
+{
+   sc->max_size = number;
+
+   sanitize_hash(sc, sc->blend_hash, CSO_BLEND, sc->max_size);
+   sanitize_hash(sc, sc->depth_stencil_hash, CSO_DEPTH_STENCIL_ALPHA,
+                 sc->max_size);
+   sanitize_hash(sc, sc->fs_hash, CSO_FRAGMENT_SHADER, sc->max_size);
+   sanitize_hash(sc, sc->vs_hash, CSO_VERTEX_SHADER, sc->max_size);
+   sanitize_hash(sc, sc->rasterizer_hash, CSO_RASTERIZER, sc->max_size);
+   sanitize_hash(sc, sc->sampler_hash, CSO_SAMPLER, sc->max_size);
+}
+
+int cso_maximum_cache_size(const struct cso_cache *sc)
+{
+   return sc->max_size;
+}
+
+void cso_cache_set_sanitize_callback(struct cso_cache *sc,
+                                     cso_sanitize_callback cb,
+                                     void *user_data)
+{
+   sc->sanitize_cb   = cb;
+   sc->sanitize_data = user_data;
+}
+
diff --git a/src/gallium/auxiliary/cso_cache/cso_cache.h b/src/gallium/auxiliary/cso_cache/cso_cache.h
new file mode 100644
index 00000000000..6b5c230e8f2
--- /dev/null
+++ b/src/gallium/auxiliary/cso_cache/cso_cache.h
@@ -0,0 +1,176 @@
+/**************************************************************************
+ *
+ * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /**
+  * @file
+  * Constant State Object (CSO) cache.
+  *
+  * The basic idea is that the states are created via the
+  * create_state/bind_state/delete_state semantics. The driver is expected to
+  * perform as much of the Gallium state translation to whatever its internal
+  * representation is during the create call. Gallium then has a caching
+  * mechanism where it stores the created states. When the pipeline needs an
+  * actual state change, a bind call is issued. In the bind call the driver
+  * gets its already translated representation.
+  *
+  * Those semantics mean that the driver doesn't do the repeated translations
+  * of states on every frame, but only once, when a new state is actually
+  * created.
+  *
+  * Even on hardware that doesn't do any kind of state cache, it makes the
+  * driver look a lot neater, plus it avoids all the redundant state
+  * translations on every frame.
+  *
+  * Currently our constant state objects are:
+  * - alpha test
+  * - blend
+  * - depth stencil
+  * - fragment shader
+  * - rasterizer (old setup)
+  * - sampler
+  * - vertex shader
+  *
+  * Things that are not constant state objects include:
+  * - blend_color
+  * - clip_state
+  * - clear_color_state
+  * - constant_buffer
+  * - feedback_state
+  * - framebuffer_state
+  * - polygon_stipple
+  * - scissor_state
+  * - texture_state
+  * - viewport_state
+  *
+  * @author Zack Rusin <[email protected]>
+  */
+
+#ifndef CSO_CACHE_H
+#define CSO_CACHE_H
+
+#include "pipe/p_context.h"
+#include "pipe/p_state.h"
+
+/* cso_hash.h is necessary for cso_hash_iter, as MSVC requires structures
+ * returned by value to be fully defined */
+#include "cso_hash.h"
+
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+enum cso_cache_type {
+   CSO_BLEND,
+   CSO_SAMPLER,
+   CSO_DEPTH_STENCIL_ALPHA,
+   CSO_RASTERIZER,
+   CSO_FRAGMENT_SHADER,
+   CSO_VERTEX_SHADER
+};
+
+typedef void (*cso_state_callback)(void *ctx, void *obj);
+
+typedef void (*cso_sanitize_callback)(struct cso_hash *hash,
+                                      enum cso_cache_type type,
+                                      int max_size,
+                                      void *user_data);
+
+struct cso_cache;
+
+struct cso_blend {
+   struct pipe_blend_state state;
+   void *data;
+   cso_state_callback delete_state;
+   struct pipe_context *context;
+};
+
+struct cso_depth_stencil_alpha {
+   struct pipe_depth_stencil_alpha_state state;
+   void *data;
+   cso_state_callback delete_state;
+   struct pipe_context *context;
+};
+
+struct cso_rasterizer {
+   struct pipe_rasterizer_state state;
+   void *data;
+   cso_state_callback delete_state;
+   struct pipe_context *context;
+};
+
+struct cso_fragment_shader {
+   struct pipe_shader_state state;
+   void *data;
+   cso_state_callback delete_state;
+   struct pipe_context *context;
+};
+
+struct cso_vertex_shader {
+   struct pipe_shader_state state;
+   void *data;
+   cso_state_callback delete_state;
+   struct pipe_context *context;
+};
+
+struct cso_sampler {
+   struct pipe_sampler_state state;
+   void *data;
+   cso_state_callback delete_state;
+   struct pipe_context *context;
+};
+
+unsigned cso_construct_key(void *item, int item_size);
+
+struct cso_cache *cso_cache_create(void);
+void cso_cache_delete(struct cso_cache *sc);
+
+void cso_cache_set_sanitize_callback(struct cso_cache *sc,
+                                     cso_sanitize_callback cb,
+                                     void *user_data);
+
+struct cso_hash_iter cso_insert_state(struct cso_cache *sc,
+                                      unsigned hash_key, enum cso_cache_type type,
+                                      void *state);
+struct cso_hash_iter cso_find_state(struct cso_cache *sc,
+                                    unsigned hash_key, enum cso_cache_type type);
+struct cso_hash_iter cso_find_state_template(struct cso_cache *sc,
+                                             unsigned hash_key, enum cso_cache_type type,
+                                             void *templ);
+void cso_for_each_state(struct cso_cache *sc, enum cso_cache_type type,
+                        cso_state_callback func, void *user_data);
+void * cso_take_state(struct cso_cache *sc, unsigned hash_key,
+                      enum cso_cache_type type);
+
+void cso_set_maximum_cache_size(struct cso_cache *sc, int number);
+int cso_maximum_cache_size(const struct cso_cache *sc);
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.c b/src/gallium/auxiliary/cso_cache/cso_context.c
new file mode 100644
index 00000000000..a9157aad711
--- /dev/null
+++ b/src/gallium/auxiliary/cso_cache/cso_context.c
@@ -0,0 +1,866 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /**
+  * @file
+  * 
+  * Wrap the cso cache & hash mechanisms in a simplified
+  * pipe-driver-specific interface.
+  *
+  * @author Zack Rusin <[email protected]>
+  * @author Keith Whitwell <[email protected]>
+  */
+
+#include "pipe/p_state.h"
+#include "util/u_memory.h"
+#include "pipe/p_inlines.h"
+#include "tgsi/tgsi_parse.h"
+
+#include "cso_cache/cso_context.h"
+#include "cso_cache/cso_cache.h"
+#include "cso_cache/cso_hash.h"
+
+struct cso_context {
+   struct pipe_context *pipe;
+   struct cso_cache *cache;
+
+   struct {
+      void *samplers[PIPE_MAX_SAMPLERS];
+      unsigned nr_samplers;
+   } hw;
+
+   void *samplers[PIPE_MAX_SAMPLERS];
+   unsigned nr_samplers;
+
+   void *samplers_saved[PIPE_MAX_SAMPLERS];
+   unsigned nr_samplers_saved;
+
+   struct pipe_texture *textures[PIPE_MAX_SAMPLERS];
+   uint nr_textures;
+
+   struct pipe_texture *textures_saved[PIPE_MAX_SAMPLERS];
+   uint nr_textures_saved;
+
+   /** Current and saved state.
+    * The saved state is used as a 1-deep stack.
+    */
+   void *blend, *blend_saved;
+   void *depth_stencil, *depth_stencil_saved;
+   void *rasterizer, *rasterizer_saved;
+   void *fragment_shader, *fragment_shader_saved;
+   void *vertex_shader, *vertex_shader_saved;
+
+   struct pipe_framebuffer_state fb, fb_saved;
+   struct pipe_viewport_state vp, vp_saved;
+   struct pipe_blend_color blend_color;
+};
+
+
+static void
+free_framebuffer_state(struct pipe_framebuffer_state *fb);
+
+
+static boolean delete_blend_state(struct cso_context *ctx, void *state)
+{
+   struct cso_blend *cso = (struct cso_blend *)state;
+
+   if (ctx->blend == cso->data)
+      return FALSE;
+
+   if (cso->delete_state)
+      cso->delete_state(cso->context, cso->data);
+   FREE(state);
+   return TRUE;
+}
+
+static boolean delete_depth_stencil_state(struct cso_context *ctx, void *state)
+{
+   struct cso_depth_stencil_alpha *cso = (struct cso_depth_stencil_alpha *)state;
+
+   if (ctx->depth_stencil == cso->data)
+      return FALSE;
+
+   if (cso->delete_state)
+      cso->delete_state(cso->context, cso->data);
+   FREE(state);
+
+   return TRUE;
+}
+
+static boolean delete_sampler_state(struct cso_context *ctx, void *state)
+{
+   struct cso_sampler *cso = (struct cso_sampler *)state;
+   if (cso->delete_state)
+      cso->delete_state(cso->context, cso->data);
+   FREE(state);
+   return TRUE;
+}
+
+static boolean delete_rasterizer_state(struct cso_context *ctx, void *state)
+{
+   struct cso_rasterizer *cso = (struct cso_rasterizer *)state;
+
+   if (ctx->rasterizer == cso->data)
+      return FALSE;
+   if (cso->delete_state)
+      cso->delete_state(cso->context, cso->data);
+   FREE(state);
+   return TRUE;
+}
+
+static boolean delete_fs_state(struct cso_context *ctx, void *state)
+{
+   struct cso_fragment_shader *cso = (struct cso_fragment_shader *)state;
+   if (ctx->fragment_shader == cso->data)
+      return FALSE;
+   if (cso->delete_state)
+      cso->delete_state(cso->context, cso->data);
+   FREE(state);
+   return TRUE;
+}
+
+static boolean delete_vs_state(struct cso_context *ctx, void *state)
+{
+   struct cso_vertex_shader *cso = (struct cso_vertex_shader *)state;
+   if (ctx->vertex_shader == cso->data)
+      return TRUE;
+   if (cso->delete_state)
+      cso->delete_state(cso->context, cso->data);
+   FREE(state);
+   return FALSE;
+}
+
+
+static INLINE boolean delete_cso(struct cso_context *ctx,
+                                 void *state, enum cso_cache_type type)
+{
+   switch (type) {
+   case CSO_BLEND:
+      return delete_blend_state(ctx, state);
+      break;
+   case CSO_SAMPLER:
+      return delete_sampler_state(ctx, state);
+      break;
+   case CSO_DEPTH_STENCIL_ALPHA:
+      return delete_depth_stencil_state(ctx, state);
+      break;
+   case CSO_RASTERIZER:
+      return delete_rasterizer_state(ctx, state);
+      break;
+   case CSO_FRAGMENT_SHADER:
+      return delete_fs_state(ctx, state);
+      break;
+   case CSO_VERTEX_SHADER:
+      return delete_vs_state(ctx, state);
+      break;
+   default:
+      assert(0);
+      FREE(state);
+   }
+   return FALSE;
+}
+
+static INLINE void sanitize_hash(struct cso_hash *hash, enum cso_cache_type type,
+                                 int max_size, void *user_data)
+{
+   struct cso_context *ctx = (struct cso_context *)user_data;
+   /* if we're approach the maximum size, remove fourth of the entries
+    * otherwise every subsequent call will go through the same */
+   int hash_size = cso_hash_size(hash);
+   int max_entries = (max_size > hash_size) ? max_size : hash_size;
+   int to_remove =  (max_size < max_entries) * max_entries/4;
+   struct cso_hash_iter iter = cso_hash_first_node(hash);
+   if (hash_size > max_size)
+      to_remove += hash_size - max_size;
+   while (to_remove) {
+      /*remove elements until we're good */
+      /*fixme: currently we pick the nodes to remove at random*/
+      void *cso = cso_hash_iter_data(iter);
+      if (delete_cso(ctx, cso, type)) {
+         iter = cso_hash_erase(hash, iter);
+         --to_remove;
+      } else
+         iter = cso_hash_iter_next(iter);
+   }
+}
+
+
+struct cso_context *cso_create_context( struct pipe_context *pipe )
+{
+   struct cso_context *ctx = CALLOC_STRUCT(cso_context);
+   if (ctx == NULL)
+      goto out;
+
+   ctx->cache = cso_cache_create();
+   if (ctx->cache == NULL)
+      goto out;
+   cso_cache_set_sanitize_callback(ctx->cache,
+                                   sanitize_hash,
+                                   ctx);
+
+   ctx->pipe = pipe;
+
+   /* Enable for testing: */
+   if (0) cso_set_maximum_cache_size( ctx->cache, 4 );
+
+   return ctx;
+
+out:
+   cso_destroy_context( ctx );      
+   return NULL;
+}
+
+
+/**
+ * Prior to context destruction, this function unbinds all state objects.
+ */
+void cso_release_all( struct cso_context *ctx )
+{
+   unsigned i;
+   
+   if (ctx->pipe) {
+      ctx->pipe->bind_blend_state( ctx->pipe, NULL );
+      ctx->pipe->bind_rasterizer_state( ctx->pipe, NULL );
+      ctx->pipe->bind_sampler_states( ctx->pipe, 0, NULL );
+      ctx->pipe->bind_depth_stencil_alpha_state( ctx->pipe, NULL );
+      ctx->pipe->bind_fs_state( ctx->pipe, NULL );
+      ctx->pipe->bind_vs_state( ctx->pipe, NULL );
+   }
+
+   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+      pipe_texture_reference(&ctx->textures[i], NULL);
+      pipe_texture_reference(&ctx->textures_saved[i], NULL);
+   }
+
+   free_framebuffer_state(&ctx->fb);
+   free_framebuffer_state(&ctx->fb_saved);
+
+   if (ctx->cache) {
+      cso_cache_delete( ctx->cache );
+      ctx->cache = NULL;
+   }
+}
+
+
+void cso_destroy_context( struct cso_context *ctx )
+{
+   if (ctx) {
+      //cso_release_all( ctx );
+      FREE( ctx );
+   }
+}
+
+
+/* Those function will either find the state of the given template
+ * in the cache or they will create a new state from the given
+ * template, insert it in the cache and return it.
+ */
+
+/*
+ * If the driver returns 0 from the create method then they will assign
+ * the data member of the cso to be the template itself.
+ */
+
+enum pipe_error cso_set_blend(struct cso_context *ctx,
+                              const struct pipe_blend_state *templ)
+{
+   unsigned hash_key = cso_construct_key((void*)templ, sizeof(struct pipe_blend_state));
+   struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
+                                                       hash_key, CSO_BLEND,
+                                                       (void*)templ);
+   void *handle;
+
+   if (cso_hash_iter_is_null(iter)) {
+      struct cso_blend *cso = MALLOC(sizeof(struct cso_blend));
+      if (!cso)
+         return PIPE_ERROR_OUT_OF_MEMORY;
+
+      memcpy(&cso->state, templ, sizeof(*templ));
+      cso->data = ctx->pipe->create_blend_state(ctx->pipe, &cso->state);
+      cso->delete_state = (cso_state_callback)ctx->pipe->delete_blend_state;
+      cso->context = ctx->pipe;
+
+      iter = cso_insert_state(ctx->cache, hash_key, CSO_BLEND, cso);
+      if (cso_hash_iter_is_null(iter)) {
+         FREE(cso);
+         return PIPE_ERROR_OUT_OF_MEMORY;
+      }
+
+      handle = cso->data;
+   }
+   else {
+      handle = ((struct cso_blend *)cso_hash_iter_data(iter))->data;
+   }
+
+   if (ctx->blend != handle) {
+      ctx->blend = handle;
+      ctx->pipe->bind_blend_state(ctx->pipe, handle);
+   }
+   return PIPE_OK;
+}
+
+void cso_save_blend(struct cso_context *ctx)
+{
+   assert(!ctx->blend_saved);
+   ctx->blend_saved = ctx->blend;
+}
+
+void cso_restore_blend(struct cso_context *ctx)
+{
+   if (ctx->blend != ctx->blend_saved) {
+      ctx->blend = ctx->blend_saved;
+      ctx->pipe->bind_blend_state(ctx->pipe, ctx->blend_saved);
+   }
+   ctx->blend_saved = NULL;
+}
+
+
+
+enum pipe_error cso_single_sampler(struct cso_context *ctx,
+                                   unsigned idx,
+                                   const struct pipe_sampler_state *templ)
+{
+   void *handle = NULL;
+
+   if (templ != NULL) {
+      unsigned hash_key = cso_construct_key((void*)templ, sizeof(struct pipe_sampler_state));
+      struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
+                                                          hash_key, CSO_SAMPLER,
+                                                          (void*)templ);
+
+      if (cso_hash_iter_is_null(iter)) {
+         struct cso_sampler *cso = MALLOC(sizeof(struct cso_sampler));
+         if (!cso)
+            return PIPE_ERROR_OUT_OF_MEMORY;
+
+         memcpy(&cso->state, templ, sizeof(*templ));
+         cso->data = ctx->pipe->create_sampler_state(ctx->pipe, &cso->state);
+         cso->delete_state = (cso_state_callback)ctx->pipe->delete_sampler_state;
+         cso->context = ctx->pipe;
+
+         iter = cso_insert_state(ctx->cache, hash_key, CSO_SAMPLER, cso);
+         if (cso_hash_iter_is_null(iter)) {
+            FREE(cso);
+            return PIPE_ERROR_OUT_OF_MEMORY;
+         }
+
+         handle = cso->data;
+      }
+      else {
+         handle = ((struct cso_sampler *)cso_hash_iter_data(iter))->data;
+      }
+   }
+
+   ctx->samplers[idx] = handle;
+   return PIPE_OK;
+}
+
+void cso_single_sampler_done( struct cso_context *ctx )
+{
+   unsigned i;
+
+   /* find highest non-null sampler */
+   for (i = PIPE_MAX_SAMPLERS; i > 0; i--) {
+      if (ctx->samplers[i - 1] != NULL)
+         break;
+   }
+
+   ctx->nr_samplers = i;
+
+   if (ctx->hw.nr_samplers != ctx->nr_samplers ||
+       memcmp(ctx->hw.samplers,
+              ctx->samplers,
+              ctx->nr_samplers * sizeof(void *)) != 0) 
+   {
+      memcpy(ctx->hw.samplers, ctx->samplers, ctx->nr_samplers * sizeof(void *));
+      ctx->hw.nr_samplers = ctx->nr_samplers;
+
+      ctx->pipe->bind_sampler_states(ctx->pipe, ctx->nr_samplers, ctx->samplers);
+   }
+}
+
+/*
+ * If the function encouters any errors it will return the
+ * last one. Done to always try to set as many samplers
+ * as possible.
+ */
+enum pipe_error cso_set_samplers( struct cso_context *ctx,
+                                  unsigned nr,
+                                  const struct pipe_sampler_state **templates )
+{
+   unsigned i;
+   enum pipe_error temp, error = PIPE_OK;
+
+   /* TODO: fastpath
+    */
+
+   for (i = 0; i < nr; i++) {
+      temp = cso_single_sampler( ctx, i, templates[i] );
+      if (temp != PIPE_OK)
+         error = temp;
+   }
+
+   for ( ; i < ctx->nr_samplers; i++) {
+      temp = cso_single_sampler( ctx, i, NULL );
+      if (temp != PIPE_OK)
+         error = temp;
+   }
+
+   cso_single_sampler_done( ctx );
+
+   return error;
+}
+
+void cso_save_samplers(struct cso_context *ctx)
+{
+   ctx->nr_samplers_saved = ctx->nr_samplers;
+   memcpy(ctx->samplers_saved, ctx->samplers, sizeof(ctx->samplers));
+}
+
+void cso_restore_samplers(struct cso_context *ctx)
+{
+   ctx->nr_samplers = ctx->nr_samplers_saved;
+   memcpy(ctx->samplers, ctx->samplers_saved, sizeof(ctx->samplers));
+   cso_single_sampler_done( ctx );
+}
+
+
+enum pipe_error cso_set_sampler_textures( struct cso_context *ctx,
+                                          uint count,
+                                          struct pipe_texture **textures )
+{
+   uint i;
+
+   ctx->nr_textures = count;
+
+   for (i = 0; i < count; i++)
+      pipe_texture_reference(&ctx->textures[i], textures[i]);
+   for ( ; i < PIPE_MAX_SAMPLERS; i++)
+      pipe_texture_reference(&ctx->textures[i], NULL);
+
+   ctx->pipe->set_sampler_textures(ctx->pipe, count, textures);
+
+   return PIPE_OK;
+}
+
+void cso_save_sampler_textures( struct cso_context *ctx )
+{
+   uint i;
+
+   ctx->nr_textures_saved = ctx->nr_textures;
+   for (i = 0; i < ctx->nr_textures; i++) {
+      assert(!ctx->textures_saved[i]);
+      pipe_texture_reference(&ctx->textures_saved[i], ctx->textures[i]);
+   }
+}
+
+void cso_restore_sampler_textures( struct cso_context *ctx )
+{
+   uint i;
+
+   ctx->nr_textures = ctx->nr_textures_saved;
+
+   for (i = 0; i < ctx->nr_textures; i++) {
+      pipe_texture_reference(&ctx->textures[i], NULL);
+      ctx->textures[i] = ctx->textures_saved[i];
+      ctx->textures_saved[i] = NULL;
+   }
+   for ( ; i < PIPE_MAX_SAMPLERS; i++)
+      pipe_texture_reference(&ctx->textures[i], NULL);
+
+   ctx->pipe->set_sampler_textures(ctx->pipe, ctx->nr_textures, ctx->textures);
+
+   ctx->nr_textures_saved = 0;
+}
+
+
+
+enum pipe_error cso_set_depth_stencil_alpha(struct cso_context *ctx,
+                                            const struct pipe_depth_stencil_alpha_state *templ)
+{
+   unsigned hash_key = cso_construct_key((void*)templ,
+                                         sizeof(struct pipe_depth_stencil_alpha_state));
+   struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
+                                                       hash_key, 
+						       CSO_DEPTH_STENCIL_ALPHA,
+                                                       (void*)templ);
+   void *handle;
+
+   if (cso_hash_iter_is_null(iter)) {
+      struct cso_depth_stencil_alpha *cso = MALLOC(sizeof(struct cso_depth_stencil_alpha));
+      if (!cso)
+         return PIPE_ERROR_OUT_OF_MEMORY;
+
+      memcpy(&cso->state, templ, sizeof(*templ));
+      cso->data = ctx->pipe->create_depth_stencil_alpha_state(ctx->pipe, &cso->state);
+      cso->delete_state = (cso_state_callback)ctx->pipe->delete_depth_stencil_alpha_state;
+      cso->context = ctx->pipe;
+
+      iter = cso_insert_state(ctx->cache, hash_key, CSO_DEPTH_STENCIL_ALPHA, cso);
+      if (cso_hash_iter_is_null(iter)) {
+         FREE(cso);
+         return PIPE_ERROR_OUT_OF_MEMORY;
+      }
+
+      handle = cso->data;
+   }
+   else {
+      handle = ((struct cso_depth_stencil_alpha *)cso_hash_iter_data(iter))->data;
+   }
+
+   if (ctx->depth_stencil != handle) {
+      ctx->depth_stencil = handle;
+      ctx->pipe->bind_depth_stencil_alpha_state(ctx->pipe, handle);
+   }
+   return PIPE_OK;
+}
+
+void cso_save_depth_stencil_alpha(struct cso_context *ctx)
+{
+   assert(!ctx->depth_stencil_saved);
+   ctx->depth_stencil_saved = ctx->depth_stencil;
+}
+
+void cso_restore_depth_stencil_alpha(struct cso_context *ctx)
+{
+   if (ctx->depth_stencil != ctx->depth_stencil_saved) {
+      ctx->depth_stencil = ctx->depth_stencil_saved;
+      ctx->pipe->bind_depth_stencil_alpha_state(ctx->pipe, ctx->depth_stencil_saved);
+   }
+   ctx->depth_stencil_saved = NULL;
+}
+
+
+
+enum pipe_error cso_set_rasterizer(struct cso_context *ctx,
+                                   const struct pipe_rasterizer_state *templ)
+{
+   unsigned hash_key = cso_construct_key((void*)templ,
+                                         sizeof(struct pipe_rasterizer_state));
+   struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
+                                                       hash_key, CSO_RASTERIZER,
+                                                       (void*)templ);
+   void *handle = NULL;
+
+   if (cso_hash_iter_is_null(iter)) {
+      struct cso_rasterizer *cso = MALLOC(sizeof(struct cso_rasterizer));
+      if (!cso)
+         return PIPE_ERROR_OUT_OF_MEMORY;
+
+      memcpy(&cso->state, templ, sizeof(*templ));
+      cso->data = ctx->pipe->create_rasterizer_state(ctx->pipe, &cso->state);
+      cso->delete_state = (cso_state_callback)ctx->pipe->delete_rasterizer_state;
+      cso->context = ctx->pipe;
+
+      iter = cso_insert_state(ctx->cache, hash_key, CSO_RASTERIZER, cso);
+      if (cso_hash_iter_is_null(iter)) {
+         FREE(cso);
+         return PIPE_ERROR_OUT_OF_MEMORY;
+      }
+
+      handle = cso->data;
+   }
+   else {
+      handle = ((struct cso_rasterizer *)cso_hash_iter_data(iter))->data;
+   }
+
+   if (ctx->rasterizer != handle) {
+      ctx->rasterizer = handle;
+      ctx->pipe->bind_rasterizer_state(ctx->pipe, handle);
+   }
+   return PIPE_OK;
+}
+
+void cso_save_rasterizer(struct cso_context *ctx)
+{
+   assert(!ctx->rasterizer_saved);
+   ctx->rasterizer_saved = ctx->rasterizer;
+}
+
+void cso_restore_rasterizer(struct cso_context *ctx)
+{
+   if (ctx->rasterizer != ctx->rasterizer_saved) {
+      ctx->rasterizer = ctx->rasterizer_saved;
+      ctx->pipe->bind_rasterizer_state(ctx->pipe, ctx->rasterizer_saved);
+   }
+   ctx->rasterizer_saved = NULL;
+}
+
+
+
+enum pipe_error cso_set_fragment_shader_handle(struct cso_context *ctx,
+                                               void *handle )
+{
+   if (ctx->fragment_shader != handle) {
+      ctx->fragment_shader = handle;
+      ctx->pipe->bind_fs_state(ctx->pipe, handle);
+   }
+   return PIPE_OK;
+}
+
+void cso_delete_fragment_shader(struct cso_context *ctx, void *handle )
+{
+   if (handle == ctx->fragment_shader) {
+      /* unbind before deleting */
+      ctx->pipe->bind_fs_state(ctx->pipe, NULL);
+      ctx->fragment_shader = NULL;
+   }
+   ctx->pipe->delete_fs_state(ctx->pipe, handle);
+}
+
+/* Not really working:
+ */
+#if 0
+enum pipe_error cso_set_fragment_shader(struct cso_context *ctx,
+                                        const struct pipe_shader_state *templ)
+{
+   const struct tgsi_token *tokens = templ->tokens;
+   unsigned num_tokens = tgsi_num_tokens(tokens);
+   size_t tokens_size = num_tokens*sizeof(struct tgsi_token);
+   unsigned hash_key = cso_construct_key((void*)tokens, tokens_size);
+   struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
+                                                       hash_key, 
+                                                       CSO_FRAGMENT_SHADER,
+                                                       (void*)tokens);
+   void *handle = NULL;
+
+   if (cso_hash_iter_is_null(iter)) {
+      struct cso_fragment_shader *cso = MALLOC(sizeof(struct cso_fragment_shader) + tokens_size);
+      struct tgsi_token *cso_tokens = (struct tgsi_token *)((char *)cso + sizeof(*cso));
+
+      if (!cso)
+         return PIPE_ERROR_OUT_OF_MEMORY;
+
+      memcpy(cso_tokens, tokens, tokens_size);
+      cso->state.tokens = cso_tokens;
+      cso->data = ctx->pipe->create_fs_state(ctx->pipe, &cso->state);
+      cso->delete_state = (cso_state_callback)ctx->pipe->delete_fs_state;
+      cso->context = ctx->pipe;
+
+      iter = cso_insert_state(ctx->cache, hash_key, CSO_FRAGMENT_SHADER, cso);
+      if (cso_hash_iter_is_null(iter)) {
+         FREE(cso);
+         return PIPE_ERROR_OUT_OF_MEMORY;
+      }
+
+      handle = cso->data;
+   }
+   else {
+      handle = ((struct cso_fragment_shader *)cso_hash_iter_data(iter))->data;
+   }
+
+   return cso_set_fragment_shader_handle( ctx, handle );
+}
+#endif
+
+void cso_save_fragment_shader(struct cso_context *ctx)
+{
+   assert(!ctx->fragment_shader_saved);
+   ctx->fragment_shader_saved = ctx->fragment_shader;
+}
+
+void cso_restore_fragment_shader(struct cso_context *ctx)
+{
+   if (ctx->fragment_shader_saved != ctx->fragment_shader) {
+      ctx->pipe->bind_fs_state(ctx->pipe, ctx->fragment_shader_saved);
+      ctx->fragment_shader = ctx->fragment_shader_saved;
+   }
+   ctx->fragment_shader_saved = NULL;
+}
+
+
+enum pipe_error cso_set_vertex_shader_handle(struct cso_context *ctx,
+                                             void *handle )
+{
+   if (ctx->vertex_shader != handle) {
+      ctx->vertex_shader = handle;
+      ctx->pipe->bind_vs_state(ctx->pipe, handle);
+   }
+   return PIPE_OK;
+}
+
+void cso_delete_vertex_shader(struct cso_context *ctx, void *handle )
+{
+   if (handle == ctx->vertex_shader) {
+      /* unbind before deleting */
+      ctx->pipe->bind_vs_state(ctx->pipe, NULL);
+      ctx->vertex_shader = NULL;
+   }
+   ctx->pipe->delete_vs_state(ctx->pipe, handle);
+}
+
+
+/* Not really working:
+ */
+#if 0
+enum pipe_error cso_set_vertex_shader(struct cso_context *ctx,
+                                      const struct pipe_shader_state *templ)
+{
+   unsigned hash_key = cso_construct_key((void*)templ,
+                                         sizeof(struct pipe_shader_state));
+   struct cso_hash_iter iter = cso_find_state_template(ctx->cache,
+                                                       hash_key, CSO_VERTEX_SHADER,
+                                                       (void*)templ);
+   void *handle = NULL;
+
+   if (cso_hash_iter_is_null(iter)) {
+      struct cso_vertex_shader *cso = MALLOC(sizeof(struct cso_vertex_shader));
+
+      if (!cso)
+         return PIPE_ERROR_OUT_OF_MEMORY;
+
+      memcpy(cso->state, templ, sizeof(*templ));
+      cso->data = ctx->pipe->create_vs_state(ctx->pipe, &cso->state);
+      cso->delete_state = (cso_state_callback)ctx->pipe->delete_vs_state;
+      cso->context = ctx->pipe;
+
+      iter = cso_insert_state(ctx->cache, hash_key, CSO_VERTEX_SHADER, cso);
+      if (cso_hash_iter_is_null(iter)) {
+         FREE(cso);
+         return PIPE_ERROR_OUT_OF_MEMORY;
+      }
+
+      handle = cso->data;
+   }
+   else {
+      handle = ((struct cso_vertex_shader *)cso_hash_iter_data(iter))->data;
+   }
+
+   return cso_set_vertex_shader_handle( ctx, handle );
+}
+#endif
+
+
+
+void cso_save_vertex_shader(struct cso_context *ctx)
+{
+   assert(!ctx->vertex_shader_saved);
+   ctx->vertex_shader_saved = ctx->vertex_shader;
+}
+
+void cso_restore_vertex_shader(struct cso_context *ctx)
+{
+   if (ctx->vertex_shader_saved != ctx->vertex_shader) {
+      ctx->pipe->bind_vs_state(ctx->pipe, ctx->vertex_shader_saved);
+      ctx->vertex_shader = ctx->vertex_shader_saved;
+   }
+   ctx->vertex_shader_saved = NULL;
+}
+
+
+/**
+ * Copy framebuffer state from src to dst with refcounting of surfaces.
+ */
+static void
+copy_framebuffer_state(struct pipe_framebuffer_state *dst,
+                       const struct pipe_framebuffer_state *src)
+{
+   uint i;
+
+   dst->width = src->width;
+   dst->height = src->height;
+   dst->nr_cbufs = src->nr_cbufs;
+   for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
+      pipe_surface_reference(&dst->cbufs[i], src->cbufs[i]);
+   }
+   pipe_surface_reference(&dst->zsbuf, src->zsbuf);
+}
+
+
+static void
+free_framebuffer_state(struct pipe_framebuffer_state *fb)
+{
+   uint i;
+
+   for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
+      pipe_surface_reference(&fb->cbufs[i], NULL);
+   }
+   pipe_surface_reference(&fb->zsbuf, NULL);
+}
+
+
+enum pipe_error cso_set_framebuffer(struct cso_context *ctx,
+                                    const struct pipe_framebuffer_state *fb)
+{
+   if (memcmp(&ctx->fb, fb, sizeof(*fb)) != 0) {
+      copy_framebuffer_state(&ctx->fb, fb);
+      ctx->pipe->set_framebuffer_state(ctx->pipe, fb);
+   }
+   return PIPE_OK;
+}
+
+void cso_save_framebuffer(struct cso_context *ctx)
+{
+   copy_framebuffer_state(&ctx->fb_saved, &ctx->fb);
+}
+
+void cso_restore_framebuffer(struct cso_context *ctx)
+{
+   if (memcmp(&ctx->fb, &ctx->fb_saved, sizeof(ctx->fb))) {
+      copy_framebuffer_state(&ctx->fb, &ctx->fb_saved);
+      ctx->pipe->set_framebuffer_state(ctx->pipe, &ctx->fb);
+      free_framebuffer_state(&ctx->fb_saved);
+   }
+}
+
+
+enum pipe_error cso_set_viewport(struct cso_context *ctx,
+                                 const struct pipe_viewport_state *vp)
+{
+   if (memcmp(&ctx->vp, vp, sizeof(*vp))) {
+      ctx->vp = *vp;
+      ctx->pipe->set_viewport_state(ctx->pipe, vp);
+   }
+   return PIPE_OK;
+}
+
+void cso_save_viewport(struct cso_context *ctx)
+{
+   ctx->vp_saved = ctx->vp;
+}
+
+
+void cso_restore_viewport(struct cso_context *ctx)
+{
+   if (memcmp(&ctx->vp, &ctx->vp_saved, sizeof(ctx->vp))) {
+      ctx->vp = ctx->vp_saved;
+      ctx->pipe->set_viewport_state(ctx->pipe, &ctx->vp);
+   }
+}
+
+
+
+
+enum pipe_error cso_set_blend_color(struct cso_context *ctx,
+                                    const struct pipe_blend_color *bc)
+{
+   if (memcmp(&ctx->blend_color, bc, sizeof(ctx->blend_color))) {
+      ctx->blend_color = *bc;
+      ctx->pipe->set_blend_color(ctx->pipe, bc);
+   }
+   return PIPE_OK;
+}
diff --git a/src/gallium/auxiliary/cso_cache/cso_context.h b/src/gallium/auxiliary/cso_cache/cso_context.h
new file mode 100644
index 00000000000..b04e98bfa12
--- /dev/null
+++ b/src/gallium/auxiliary/cso_cache/cso_context.h
@@ -0,0 +1,145 @@
+/**************************************************************************
+ *
+ * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+
+#ifndef CSO_CONTEXT_H
+#define CSO_CONTEXT_H
+
+#include "pipe/p_context.h"
+#include "pipe/p_state.h"
+#include "pipe/p_error.h"
+
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+struct cso_context;
+
+struct cso_context *cso_create_context( struct pipe_context *pipe );
+
+void cso_release_all( struct cso_context *ctx );
+
+void cso_destroy_context( struct cso_context *cso );
+
+
+
+enum pipe_error cso_set_blend( struct cso_context *cso,
+                               const struct pipe_blend_state *blend );
+void cso_save_blend(struct cso_context *cso);
+void cso_restore_blend(struct cso_context *cso);
+
+
+
+enum pipe_error cso_set_depth_stencil_alpha( struct cso_context *cso,
+                                             const struct pipe_depth_stencil_alpha_state *dsa );
+void cso_save_depth_stencil_alpha(struct cso_context *cso);
+void cso_restore_depth_stencil_alpha(struct cso_context *cso);
+
+
+
+enum pipe_error cso_set_rasterizer( struct cso_context *cso,
+                                    const struct pipe_rasterizer_state *rasterizer );
+void cso_save_rasterizer(struct cso_context *cso);
+void cso_restore_rasterizer(struct cso_context *cso);
+
+
+
+enum pipe_error cso_set_samplers( struct cso_context *cso,
+                                  unsigned count,
+                                  const struct pipe_sampler_state **states );
+void cso_save_samplers(struct cso_context *cso);
+void cso_restore_samplers(struct cso_context *cso);
+
+/* Alternate interface to support state trackers that like to modify
+ * samplers one at a time:
+ */
+enum pipe_error cso_single_sampler( struct cso_context *cso,
+                                    unsigned nr,
+                                    const struct pipe_sampler_state *states );
+
+void cso_single_sampler_done( struct cso_context *cso );
+
+
+
+enum pipe_error cso_set_sampler_textures( struct cso_context *cso,
+                                          uint count,
+                                          struct pipe_texture **textures );
+void cso_save_sampler_textures( struct cso_context *cso );
+void cso_restore_sampler_textures( struct cso_context *cso );
+
+
+
+/* These aren't really sensible -- most of the time the api provides
+ * object semantics for shaders anyway, and the cases where it doesn't
+ * (eg mesa's internall-generated texenv programs), it will be up to
+ * the state tracker to implement their own specialized caching.
+ */
+enum pipe_error cso_set_fragment_shader_handle(struct cso_context *ctx,
+                                               void *handle );
+void cso_delete_fragment_shader(struct cso_context *ctx, void *handle );
+/*
+enum pipe_error cso_set_fragment_shader( struct cso_context *cso,
+                                         const struct pipe_shader_state *shader );
+*/
+void cso_save_fragment_shader(struct cso_context *cso);
+void cso_restore_fragment_shader(struct cso_context *cso);
+
+
+enum pipe_error cso_set_vertex_shader_handle(struct cso_context *ctx,
+                                             void *handle );
+void cso_delete_vertex_shader(struct cso_context *ctx, void *handle );
+/*
+enum pipe_error cso_set_vertex_shader( struct cso_context *cso,
+                                       const struct pipe_shader_state *shader );
+*/
+void cso_save_vertex_shader(struct cso_context *cso);
+void cso_restore_vertex_shader(struct cso_context *cso);
+
+
+
+enum pipe_error cso_set_framebuffer(struct cso_context *cso,
+                                    const struct pipe_framebuffer_state *fb);
+void cso_save_framebuffer(struct cso_context *cso);
+void cso_restore_framebuffer(struct cso_context *cso);
+
+
+enum pipe_error cso_set_viewport(struct cso_context *cso,
+                                 const struct pipe_viewport_state *vp);
+void cso_save_viewport(struct cso_context *cso);
+void cso_restore_viewport(struct cso_context *cso);
+
+
+enum pipe_error cso_set_blend_color(struct cso_context *cso,
+                                    const struct pipe_blend_color *bc);
+
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif
diff --git a/src/gallium/auxiliary/cso_cache/cso_hash.c b/src/gallium/auxiliary/cso_cache/cso_hash.c
new file mode 100644
index 00000000000..4e7664f9bf0
--- /dev/null
+++ b/src/gallium/auxiliary/cso_cache/cso_hash.c
@@ -0,0 +1,439 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin <[email protected]>
+  */
+
+#include "pipe/p_debug.h"
+#include "util/u_memory.h"
+
+#include "cso_hash.h"
+
+#define MAX(a, b) ((a > b) ? (a) : (b))
+
+static const int MinNumBits = 4;
+
+static const unsigned char prime_deltas[] = {
+   0,  0,  1,  3,  1,  5,  3,  3,  1,  9,  7,  5,  3,  9, 25,  3,
+   1, 21,  3, 21,  7, 15,  9,  5,  3, 29, 15,  0,  0,  0,  0,  0
+};
+
+static int primeForNumBits(int numBits)
+{
+   return (1 << numBits) + prime_deltas[numBits];
+}
+
+/*
+    Returns the smallest integer n such that
+    primeForNumBits(n) >= hint.
+*/
+static int countBits(int hint)
+{
+   int numBits = 0;
+   int bits = hint;
+
+   while (bits > 1) {
+      bits >>= 1;
+      numBits++;
+   }
+
+   if (numBits >= (int)sizeof(prime_deltas)) {
+      numBits = sizeof(prime_deltas) - 1;
+   } else if (primeForNumBits(numBits) < hint) {
+      ++numBits;
+   }
+   return numBits;
+}
+
+struct cso_node {
+   struct cso_node *next;
+   unsigned key;
+   void *value;
+};
+
+struct cso_hash_data {
+   struct cso_node *fakeNext;
+   struct cso_node **buckets;
+   int size;
+   int nodeSize;
+   short userNumBits;
+   short numBits;
+   int numBuckets;
+};
+
+struct cso_hash {
+   union {
+      struct cso_hash_data *d;
+      struct cso_node      *e;
+   } data;
+};
+
+static void *cso_data_allocate_node(struct cso_hash_data *hash)
+{
+   return MALLOC(hash->nodeSize);
+}
+
+static void cso_free_node(struct cso_node *node)
+{
+   FREE(node);
+}
+
+static struct cso_node *
+cso_hash_create_node(struct cso_hash *hash,
+                      unsigned akey, void *avalue,
+                      struct cso_node **anextNode)
+{
+   struct cso_node *node = cso_data_allocate_node(hash->data.d);
+
+   if (!node)
+      return NULL;
+
+   node->key = akey;
+   node->value = avalue;
+
+   node->next = (struct cso_node*)(*anextNode);
+   *anextNode = node;
+   ++hash->data.d->size;
+   return node;
+}
+
+static void cso_data_rehash(struct cso_hash_data *hash, int hint)
+{
+   if (hint < 0) {
+      hint = countBits(-hint);
+      if (hint < MinNumBits)
+         hint = MinNumBits;
+      hash->userNumBits = (short)hint;
+      while (primeForNumBits(hint) < (hash->size >> 1))
+         ++hint;
+   } else if (hint < MinNumBits) {
+      hint = MinNumBits;
+   }
+
+   if (hash->numBits != hint) {
+      struct cso_node *e = (struct cso_node *)(hash);
+      struct cso_node **oldBuckets = hash->buckets;
+      int oldNumBuckets = hash->numBuckets;
+      int  i = 0;
+
+      hash->numBits = (short)hint;
+      hash->numBuckets = primeForNumBits(hint);
+      hash->buckets = MALLOC(sizeof(struct cso_node*) * hash->numBuckets);
+      for (i = 0; i < hash->numBuckets; ++i)
+         hash->buckets[i] = e;
+
+      for (i = 0; i < oldNumBuckets; ++i) {
+         struct cso_node *firstNode = oldBuckets[i];
+         while (firstNode != e) {
+            unsigned h = firstNode->key;
+            struct cso_node *lastNode = firstNode;
+            struct cso_node *afterLastNode;
+            struct cso_node **beforeFirstNode;
+            
+            while (lastNode->next != e && lastNode->next->key == h)
+               lastNode = lastNode->next;
+
+            afterLastNode = lastNode->next;
+            beforeFirstNode = &hash->buckets[h % hash->numBuckets];
+            while (*beforeFirstNode != e)
+               beforeFirstNode = &(*beforeFirstNode)->next;
+            lastNode->next = *beforeFirstNode;
+            *beforeFirstNode = firstNode;
+            firstNode = afterLastNode;
+         }
+      }
+      FREE(oldBuckets);
+   }
+}
+
+static void cso_data_might_grow(struct cso_hash_data *hash)
+{
+   if (hash->size >= hash->numBuckets)
+      cso_data_rehash(hash, hash->numBits + 1);
+}
+
+static void cso_data_has_shrunk(struct cso_hash_data *hash)
+{
+   if (hash->size <= (hash->numBuckets >> 3) &&
+       hash->numBits > hash->userNumBits) {
+      int max = MAX(hash->numBits-2, hash->userNumBits);
+      cso_data_rehash(hash,  max);
+   }
+}
+
+static struct cso_node *cso_data_first_node(struct cso_hash_data *hash)
+{
+   struct cso_node *e = (struct cso_node *)(hash);
+   struct cso_node **bucket = hash->buckets;
+   int n = hash->numBuckets;
+   while (n--) {
+      if (*bucket != e)
+         return *bucket;
+      ++bucket;
+   }
+   return e;
+}
+
+static struct cso_node **cso_hash_find_node(struct cso_hash *hash, unsigned akey)
+{
+   struct cso_node **node;
+
+   if (hash->data.d->numBuckets) {
+      node = (struct cso_node **)(&hash->data.d->buckets[akey % hash->data.d->numBuckets]);
+      assert(*node == hash->data.e || (*node)->next);
+      while (*node != hash->data.e && (*node)->key != akey)
+         node = &(*node)->next;
+   } else {
+      node = (struct cso_node **)((const struct cso_node * const *)(&hash->data.e));
+   }
+   return node;
+}
+
+struct cso_hash_iter cso_hash_insert(struct cso_hash *hash,
+                                       unsigned key, void *data)
+{
+   cso_data_might_grow(hash->data.d);
+
+   {
+      struct cso_node **nextNode = cso_hash_find_node(hash, key);
+      struct cso_node *node = cso_hash_create_node(hash, key, data, nextNode);
+      if (!node) {
+         struct cso_hash_iter null_iter = {hash, 0};
+         return null_iter;
+      }
+
+      {
+         struct cso_hash_iter iter = {hash, node};
+         return iter;
+      }
+   }
+}
+
+struct cso_hash * cso_hash_create(void)
+{
+   struct cso_hash *hash = MALLOC_STRUCT(cso_hash);
+   if (!hash)
+      return NULL;
+
+   hash->data.d = MALLOC_STRUCT(cso_hash_data);
+   if (!hash->data.d) {
+      FREE(hash);
+      return NULL;
+   }
+
+   hash->data.d->fakeNext = 0;
+   hash->data.d->buckets = 0;
+   hash->data.d->size = 0;
+   hash->data.d->nodeSize = sizeof(struct cso_node);
+   hash->data.d->userNumBits = (short)MinNumBits;
+   hash->data.d->numBits = 0;
+   hash->data.d->numBuckets = 0;
+
+   return hash;
+}
+
+void cso_hash_delete(struct cso_hash *hash)
+{
+   struct cso_node *e_for_x = (struct cso_node *)(hash->data.d);
+   struct cso_node **bucket = (struct cso_node **)(hash->data.d->buckets);
+   int n = hash->data.d->numBuckets;
+   while (n--) {
+      struct cso_node *cur = *bucket++;
+      while (cur != e_for_x) {
+         struct cso_node *next = cur->next;
+         cso_free_node(cur);
+         cur = next;
+      }
+   }
+   FREE(hash->data.d->buckets);
+   FREE(hash->data.d);
+   FREE(hash);
+}
+
+struct cso_hash_iter cso_hash_find(struct cso_hash *hash,
+                                     unsigned key)
+{
+   struct cso_node **nextNode = cso_hash_find_node(hash, key);
+   struct cso_hash_iter iter = {hash, *nextNode};
+   return iter;
+}
+
+unsigned cso_hash_iter_key(struct cso_hash_iter iter)
+{
+   if (!iter.node || iter.hash->data.e == iter.node)
+      return 0;
+   return iter.node->key;
+}
+
+void * cso_hash_iter_data(struct cso_hash_iter iter)
+{
+   if (!iter.node || iter.hash->data.e == iter.node)
+      return 0;
+   return iter.node->value;
+}
+
+static struct cso_node *cso_hash_data_next(struct cso_node *node)
+{
+   union {
+      struct cso_node *next;
+      struct cso_node *e;
+      struct cso_hash_data *d;
+   } a;
+   int start;
+   struct cso_node **bucket;
+   int n;
+
+   a.next = node->next;
+   if (!a.next) {
+      debug_printf("iterating beyond the last element\n");
+      return 0;
+   }
+   if (a.next->next)
+      return a.next;
+
+   start = (node->key % a.d->numBuckets) + 1;
+   bucket = a.d->buckets + start;
+   n = a.d->numBuckets - start;
+   while (n--) {
+      if (*bucket != a.e)
+         return *bucket;
+      ++bucket;
+   }
+   return a.e;
+}
+
+
+static struct cso_node *cso_hash_data_prev(struct cso_node *node)
+{
+   union {
+      struct cso_node *e;
+      struct cso_hash_data *d;
+   } a;
+   int start;
+   struct cso_node *sentinel;
+   struct cso_node **bucket;
+
+   a.e = node;
+   while (a.e->next)
+      a.e = a.e->next;
+
+   if (node == a.e)
+      start = a.d->numBuckets - 1;
+   else
+      start = node->key % a.d->numBuckets;
+
+   sentinel = node;
+   bucket = a.d->buckets + start;
+   while (start >= 0) {
+      if (*bucket != sentinel) {
+         struct cso_node *prev = *bucket;
+         while (prev->next != sentinel)
+            prev = prev->next;
+         return prev;
+      }
+
+      sentinel = a.e;
+      --bucket;
+      --start;
+   }
+   debug_printf("iterating backward beyond first element\n");
+   return a.e;
+}
+
+struct cso_hash_iter cso_hash_iter_next(struct cso_hash_iter iter)
+{
+   struct cso_hash_iter next = {iter.hash, cso_hash_data_next(iter.node)};
+   return next;
+}
+
+int cso_hash_iter_is_null(struct cso_hash_iter iter)
+{
+   if (!iter.node || iter.node == iter.hash->data.e)
+      return 1;
+   return 0;
+}
+
+void * cso_hash_take(struct cso_hash *hash,
+                      unsigned akey)
+{
+   struct cso_node **node = cso_hash_find_node(hash, akey);
+   if (*node != hash->data.e) {
+      void *t = (*node)->value;
+      struct cso_node *next = (*node)->next;
+      cso_free_node(*node);
+      *node = next;
+      --hash->data.d->size;
+      cso_data_has_shrunk(hash->data.d);
+      return t;
+   }
+   return 0;
+}
+
+struct cso_hash_iter cso_hash_iter_prev(struct cso_hash_iter iter)
+{
+   struct cso_hash_iter prev = {iter.hash,
+                                 cso_hash_data_prev(iter.node)};
+   return prev;
+}
+
+struct cso_hash_iter cso_hash_first_node(struct cso_hash *hash)
+{
+   struct cso_hash_iter iter = {hash, cso_data_first_node(hash->data.d)};
+   return iter;
+}
+
+int cso_hash_size(struct cso_hash *hash)
+{
+   return hash->data.d->size;
+}
+
+struct cso_hash_iter cso_hash_erase(struct cso_hash *hash, struct cso_hash_iter iter)
+{
+   struct cso_hash_iter ret = iter;
+   struct cso_node *node = iter.node;
+   struct cso_node **node_ptr;
+
+   if (node == hash->data.e)
+      return iter;
+
+   ret = cso_hash_iter_next(ret);
+   node_ptr = (struct cso_node**)(&hash->data.d->buckets[node->key % hash->data.d->numBuckets]);
+   while (*node_ptr != node)
+      node_ptr = &(*node_ptr)->next;
+   *node_ptr = node->next;
+   cso_free_node(node);
+   --hash->data.d->size;
+   return ret;
+}
+
+boolean cso_hash_contains(struct cso_hash *hash, unsigned key)
+{
+   struct cso_node **node = cso_hash_find_node(hash, key);
+   return (*node != hash->data.e);
+}
diff --git a/src/gallium/auxiliary/cso_cache/cso_hash.h b/src/gallium/auxiliary/cso_cache/cso_hash.h
new file mode 100644
index 00000000000..5891c325fa5
--- /dev/null
+++ b/src/gallium/auxiliary/cso_cache/cso_hash.h
@@ -0,0 +1,129 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Hash table implementation.
+ * 
+ * This file provides a hash implementation that is capable of dealing
+ * with collisions. It stores colliding entries in linked list. All
+ * functions operating on the hash return an iterator. The iterator
+ * itself points to the collision list. If there wasn't any collision
+ * the list will have just one entry, otherwise client code should
+ * iterate over the entries to find the exact entry among ones that
+ * had the same key (e.g. memcmp could be used on the data to check
+ * that)
+ * 
+ * @author Zack Rusin <[email protected]>
+ */
+
+#ifndef CSO_HASH_H
+#define CSO_HASH_H
+
+#include "pipe/p_compiler.h"
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+
+struct cso_hash;
+struct cso_node;
+
+
+struct cso_hash_iter {
+   struct cso_hash *hash;
+   struct cso_node  *node;
+};
+
+
+struct cso_hash *cso_hash_create(void);
+void             cso_hash_delete(struct cso_hash *hash);
+
+
+int              cso_hash_size(struct cso_hash *hash);
+
+
+/**
+ * Adds a data with the given key to the hash. If entry with the given
+ * key is already in the hash, this current entry is instered before it
+ * in the collision list.
+ * Function returns iterator pointing to the inserted item in the hash.
+ */
+struct cso_hash_iter cso_hash_insert(struct cso_hash *hash, unsigned key,
+                                     void *data);
+/**
+ * Removes the item pointed to by the current iterator from the hash.
+ * Note that the data itself is not erased and if it was a malloc'ed pointer
+ * it will have to be freed after calling this function by the callee.
+ * Function returns iterator pointing to the item after the removed one in
+ * the hash.
+ */
+struct cso_hash_iter cso_hash_erase(struct cso_hash *hash, struct cso_hash_iter iter);
+
+void  *cso_hash_take(struct cso_hash *hash, unsigned key);
+
+
+
+struct cso_hash_iter cso_hash_first_node(struct cso_hash *hash);
+
+/**
+ * Return an iterator pointing to the first entry in the collision list.
+ */
+struct cso_hash_iter cso_hash_find(struct cso_hash *hash, unsigned key);
+
+/**
+ * Returns true if a value with the given key exists in the hash
+ */
+boolean   cso_hash_contains(struct cso_hash *hash, unsigned key);
+
+
+int       cso_hash_iter_is_null(struct cso_hash_iter iter);
+unsigned  cso_hash_iter_key(struct cso_hash_iter iter);
+void     *cso_hash_iter_data(struct cso_hash_iter iter);
+
+
+struct cso_hash_iter cso_hash_iter_next(struct cso_hash_iter iter);
+struct cso_hash_iter cso_hash_iter_prev(struct cso_hash_iter iter);
+
+
+/**
+ * Convenience routine to iterate over the collision list while doing a memory
+ * comparison to see which entry in the list is a direct copy of our template
+ * and returns that entry.
+ */
+void *cso_hash_find_data_from_template( struct cso_hash *hash,
+				        unsigned hash_key,
+				        void *templ,
+				        int size );
+
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif
diff --git a/src/gallium/auxiliary/draw/Makefile b/src/gallium/auxiliary/draw/Makefile
new file mode 100644
index 00000000000..bdbf5a08ede
--- /dev/null
+++ b/src/gallium/auxiliary/draw/Makefile
@@ -0,0 +1,50 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = draw
+
+C_SOURCES = \
+	draw_context.c \
+	draw_pipe.c \
+	draw_pipe_aaline.c \
+	draw_pipe_aapoint.c \
+	draw_pipe_clip.c \
+	draw_pipe_cull.c \
+	draw_pipe_flatshade.c \
+	draw_pipe_offset.c \
+	draw_pipe_pstipple.c \
+	draw_pipe_stipple.c \
+	draw_pipe_twoside.c \
+	draw_pipe_unfilled.c \
+	draw_pipe_util.c \
+	draw_pipe_validate.c \
+	draw_pipe_vbuf.c \
+	draw_pipe_wide_line.c \
+	draw_pipe_wide_point.c \
+	draw_pt.c \
+	draw_pt_elts.c \
+	draw_pt_emit.c \
+	draw_pt_fetch.c \
+	draw_pt_fetch_emit.c \
+	draw_pt_fetch_shade_emit.c \
+	draw_pt_fetch_shade_pipeline.c \
+	draw_pt_post_vs.c \
+        draw_pt_util.c \
+        draw_pt_varray.c \
+	draw_pt_vcache.c \
+	draw_vertex.c \
+	draw_vs.c \
+	draw_vs_varient.c \
+	draw_vs_aos.c \
+	draw_vs_aos_io.c \
+	draw_vs_aos_machine.c \
+	draw_vs_exec.c \
+	draw_vs_llvm.c \
+	draw_vs_ppc.c  \
+	draw_vs_sse.c 
+
+
+include ../../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/auxiliary/draw/SConscript b/src/gallium/auxiliary/draw/SConscript
new file mode 100644
index 00000000000..5f05aa324a5
--- /dev/null
+++ b/src/gallium/auxiliary/draw/SConscript
@@ -0,0 +1,46 @@
+Import('*')
+
+draw = env.ConvenienceLibrary(
+	target = 'draw',
+	source = [
+		'draw_context.c',
+		'draw_pipe.c',
+		'draw_pipe_aaline.c',
+		'draw_pipe_aapoint.c',
+		'draw_pipe_clip.c',
+		'draw_pipe_cull.c',
+		'draw_pipe_flatshade.c',
+		'draw_pipe_offset.c',
+		'draw_pipe_pstipple.c',
+		'draw_pipe_stipple.c',
+		'draw_pipe_twoside.c',
+		'draw_pipe_unfilled.c',
+		'draw_pipe_util.c',
+		'draw_pipe_validate.c',
+		'draw_pipe_vbuf.c',
+		'draw_pipe_wide_line.c',
+		'draw_pipe_wide_point.c',
+		'draw_pt.c',
+		'draw_pt_elts.c',
+		'draw_pt_emit.c',
+		'draw_pt_fetch.c',
+		'draw_pt_fetch_emit.c',
+		'draw_pt_fetch_shade_emit.c',
+		'draw_pt_fetch_shade_pipeline.c',
+		'draw_pt_post_vs.c',
+		'draw_pt_util.c',
+		'draw_pt_varray.c',
+		'draw_pt_vcache.c',
+		'draw_vertex.c',
+		'draw_vs.c',
+		'draw_vs_aos.c',
+		'draw_vs_aos_io.c',
+		'draw_vs_aos_machine.c',
+		'draw_vs_exec.c',
+		'draw_vs_llvm.c',
+		'draw_vs_ppc.c',
+		'draw_vs_sse.c',
+		'draw_vs_varient.c'
+	])
+
+auxiliaries.insert(0, draw)
diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c
new file mode 100644
index 00000000000..7bd4a2e2218
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_context.c
@@ -0,0 +1,429 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  */
+
+
+#include "util/u_memory.h"
+#include "util/u_math.h"
+#include "draw_context.h"
+#include "draw_vbuf.h"
+#include "draw_vs.h"
+#include "draw_pt.h"
+#include "draw_pipe.h"
+
+
+struct draw_context *draw_create( void )
+{
+   struct draw_context *draw = CALLOC_STRUCT( draw_context );
+   if (draw == NULL)
+      goto fail;
+
+   ASSIGN_4V( draw->plane[0], -1,  0,  0, 1 );
+   ASSIGN_4V( draw->plane[1],  1,  0,  0, 1 );
+   ASSIGN_4V( draw->plane[2],  0, -1,  0, 1 );
+   ASSIGN_4V( draw->plane[3],  0,  1,  0, 1 );
+   ASSIGN_4V( draw->plane[4],  0,  0,  1, 1 ); /* yes these are correct */
+   ASSIGN_4V( draw->plane[5],  0,  0, -1, 1 ); /* mesa's a bit wonky */
+   draw->nr_planes = 6;
+
+
+   draw->reduced_prim = ~0; /* != any of PIPE_PRIM_x */
+
+
+   if (!draw_pipeline_init( draw ))
+      goto fail;
+
+   if (!draw_pt_init( draw ))
+      goto fail;
+
+   if (!draw_vs_init( draw ))
+      goto fail;
+
+   return draw;
+
+fail:
+   draw_destroy( draw );   
+   return NULL;
+}
+
+
+void draw_destroy( struct draw_context *draw )
+{
+   if (!draw)
+      return;
+
+
+
+   /* Not so fast -- we're just borrowing this at the moment.
+    * 
+   if (draw->render)
+      draw->render->destroy( draw->render );
+   */
+
+   draw_pipeline_destroy( draw );
+   draw_pt_destroy( draw );
+   draw_vs_destroy( draw );
+
+   FREE( draw );
+}
+
+
+
+void draw_flush( struct draw_context *draw )
+{
+   draw_do_flush( draw, DRAW_FLUSH_BACKEND );
+}
+
+
+/**
+ * Specify the Minimum Resolvable Depth factor for polygon offset.
+ * This factor potentially depends on the number of Z buffer bits,
+ * the rasterization algorithm and the arithmetic performed on Z
+ * values between vertex shading and rasterization.  It will vary
+ * from one driver to another.
+ */
+void draw_set_mrd(struct draw_context *draw, double mrd)
+{
+   draw->mrd = mrd;
+}
+
+
+/**
+ * Register new primitive rasterization/rendering state.
+ * This causes the drawing pipeline to be rebuilt.
+ */
+void draw_set_rasterizer_state( struct draw_context *draw,
+                                const struct pipe_rasterizer_state *raster )
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+
+   draw->rasterizer = raster;
+   draw->bypass_clipping =
+      ((draw->rasterizer && draw->rasterizer->bypass_clipping) ||
+       draw->driver.bypass_clipping);
+}
+
+
+void draw_set_driver_clipping( struct draw_context *draw,
+                               boolean bypass_clipping )
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+
+   draw->driver.bypass_clipping = bypass_clipping;
+   draw->bypass_clipping = (draw->rasterizer->bypass_clipping || 
+                            draw->driver.bypass_clipping);
+}
+
+
+/** 
+ * Plug in the primitive rendering/rasterization stage (which is the last
+ * stage in the drawing pipeline).
+ * This is provided by the device driver.
+ */
+void draw_set_rasterize_stage( struct draw_context *draw,
+                               struct draw_stage *stage )
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+
+   draw->pipeline.rasterize = stage;
+}
+
+
+/**
+ * Set the draw module's clipping state.
+ */
+void draw_set_clip_state( struct draw_context *draw,
+                          const struct pipe_clip_state *clip )
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+
+   assert(clip->nr <= PIPE_MAX_CLIP_PLANES);
+   memcpy(&draw->plane[6], clip->ucp, clip->nr * sizeof(clip->ucp[0]));
+   draw->nr_planes = 6 + clip->nr;
+}
+
+
+/**
+ * Set the draw module's viewport state.
+ */
+void draw_set_viewport_state( struct draw_context *draw,
+                              const struct pipe_viewport_state *viewport )
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+   draw->viewport = *viewport; /* struct copy */
+   draw->identity_viewport = (viewport->scale[0] == 1.0f &&
+                              viewport->scale[1] == 1.0f &&
+                              viewport->scale[2] == 1.0f &&
+                              viewport->scale[3] == 1.0f &&
+                              viewport->translate[0] == 0.0f &&
+                              viewport->translate[1] == 0.0f &&
+                              viewport->translate[2] == 0.0f &&
+                              viewport->translate[3] == 0.0f);
+
+   draw_vs_set_viewport( draw, viewport );
+}
+
+
+
+void
+draw_set_vertex_buffers(struct draw_context *draw,
+                        unsigned count,
+                        const struct pipe_vertex_buffer *buffers)
+{
+   assert(count <= PIPE_MAX_ATTRIBS);
+
+   memcpy(draw->pt.vertex_buffer, buffers, count * sizeof(buffers[0]));
+   draw->pt.nr_vertex_buffers = count;
+}
+
+
+void
+draw_set_vertex_elements(struct draw_context *draw,
+                         unsigned count,
+                         const struct pipe_vertex_element *elements)
+{
+   assert(count <= PIPE_MAX_ATTRIBS);
+
+   memcpy(draw->pt.vertex_element, elements, count * sizeof(elements[0]));
+   draw->pt.nr_vertex_elements = count;
+}
+
+
+/**
+ * Tell drawing context where to find mapped vertex buffers.
+ */
+void
+draw_set_mapped_vertex_buffer(struct draw_context *draw,
+                              unsigned attr, const void *buffer)
+{
+   draw->pt.user.vbuffer[attr] = buffer;
+}
+
+
+void
+draw_set_mapped_constant_buffer(struct draw_context *draw,
+                                const void *buffer, 
+                                unsigned size )
+{
+   draw->pt.user.constants = buffer;
+   draw_vs_set_constants( draw, (const float (*)[4])buffer, size );
+}
+
+
+/**
+ * Tells the draw module to draw points with triangles if their size
+ * is greater than this threshold.
+ */
+void
+draw_wide_point_threshold(struct draw_context *draw, float threshold)
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+   draw->pipeline.wide_point_threshold = threshold;
+}
+
+
+/**
+ * Tells the draw module to draw lines with triangles if their width
+ * is greater than this threshold.
+ */
+void
+draw_wide_line_threshold(struct draw_context *draw, float threshold)
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+   draw->pipeline.wide_line_threshold = threshold;
+}
+
+
+/**
+ * Tells the draw module whether or not to implement line stipple.
+ */
+void
+draw_enable_line_stipple(struct draw_context *draw, boolean enable)
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+   draw->pipeline.line_stipple = enable;
+}
+
+
+/**
+ * Tells draw module whether to convert points to quads for sprite mode.
+ */
+void
+draw_enable_point_sprites(struct draw_context *draw, boolean enable)
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+   draw->pipeline.point_sprite = enable;
+}
+
+
+void
+draw_set_force_passthrough( struct draw_context *draw, boolean enable )
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+   draw->force_passthrough = enable;
+}
+
+
+/**
+ * Ask the draw module for the location/slot of the given vertex attribute in
+ * a post-transformed vertex.
+ *
+ * With this function, drivers that use the draw module should have no reason
+ * to track the current vertex shader.
+ *
+ * Note that the draw module may sometimes generate vertices with extra
+ * attributes (such as texcoords for AA lines).  The driver can call this
+ * function to find those attributes.
+ *
+ * Zero is returned if the attribute is not found since this is
+ * a don't care / undefined situtation.  Returning -1 would be a bit more
+ * work for the drivers.
+ */
+int
+draw_find_vs_output(const struct draw_context *draw,
+                    uint semantic_name, uint semantic_index)
+{
+   const struct draw_vertex_shader *vs = draw->vs.vertex_shader;
+   uint i;
+   for (i = 0; i < vs->info.num_outputs; i++) {
+      if (vs->info.output_semantic_name[i] == semantic_name &&
+          vs->info.output_semantic_index[i] == semantic_index)
+         return i;
+   }
+
+   /* XXX there may be more than one extra vertex attrib.
+    * For example, simulated gl_FragCoord and gl_PointCoord.
+    */
+   if (draw->extra_vp_outputs.semantic_name == semantic_name &&
+       draw->extra_vp_outputs.semantic_index == semantic_index) {
+      return draw->extra_vp_outputs.slot;
+   }
+   return 0;
+}
+
+
+/**
+ * Return number of vertex shader outputs.
+ */
+uint
+draw_num_vs_outputs(const struct draw_context *draw)
+{
+   uint count = draw->vs.vertex_shader->info.num_outputs;
+   if (draw->extra_vp_outputs.slot > 0)
+      count++;
+   return count;
+}
+
+
+/**
+ * Provide TGSI sampler objects for vertex shaders that use texture fetches.
+ * This might only be used by software drivers for the time being.
+ */
+void
+draw_texture_samplers(struct draw_context *draw,
+                      uint num_samplers,
+                      struct tgsi_sampler **samplers)
+{
+   draw->vs.num_samplers = num_samplers;
+   draw->vs.samplers = samplers;
+}
+
+
+
+
+void draw_set_render( struct draw_context *draw, 
+		      struct vbuf_render *render )
+{
+   draw->render = render;
+}
+
+void draw_set_edgeflags( struct draw_context *draw,
+                         const unsigned *edgeflag )
+{
+   draw->pt.user.edgeflag = edgeflag;
+}
+
+
+
+
+/**
+ * Tell the drawing context about the index/element buffer to use
+ * (ala glDrawElements)
+ * If no element buffer is to be used (i.e. glDrawArrays) then this
+ * should be called with eltSize=0 and elements=NULL.
+ *
+ * \param draw  the drawing context
+ * \param eltSize  size of each element (1, 2 or 4 bytes)
+ * \param elements  the element buffer ptr
+ */
+void
+draw_set_mapped_element_buffer_range( struct draw_context *draw,
+                                      unsigned eltSize,
+                                      unsigned min_index,
+                                      unsigned max_index,
+                                      const void *elements )
+{
+   draw->pt.user.elts = elements;
+   draw->pt.user.eltSize = eltSize;
+   draw->pt.user.min_index = min_index;
+   draw->pt.user.max_index = max_index;
+}
+
+
+void
+draw_set_mapped_element_buffer( struct draw_context *draw,
+                                unsigned eltSize,
+                                const void *elements )
+{
+   draw->pt.user.elts = elements;
+   draw->pt.user.eltSize = eltSize;
+   draw->pt.user.min_index = 0;
+   draw->pt.user.max_index = 0xffffffff;
+}
+
+ 
+/* Revamp me please:
+ */
+void draw_do_flush( struct draw_context *draw, unsigned flags )
+{
+   if (!draw->suspend_flushing)
+   {
+      assert(!draw->flushing); /* catch inadvertant recursion */
+
+      draw->flushing = TRUE;
+
+      draw_pipeline_flush( draw, flags );
+
+      draw->reduced_prim = ~0; /* is reduced_prim needed any more? */
+      
+      draw->flushing = FALSE;
+   }
+}
diff --git a/src/gallium/auxiliary/draw/draw_context.h b/src/gallium/auxiliary/draw/draw_context.h
new file mode 100644
index 00000000000..d529e4e9a27
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_context.h
@@ -0,0 +1,182 @@
+
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * \brief  Public interface into the drawing module.
+ */
+
+/* Authors:  Keith Whitwell <[email protected]>
+ */
+
+
+#ifndef DRAW_CONTEXT_H
+#define DRAW_CONTEXT_H
+
+
+#include "pipe/p_state.h"
+
+
+struct pipe_context;
+struct draw_context;
+struct draw_stage;
+struct draw_vertex_shader;
+struct tgsi_sampler;
+
+
+struct draw_context *draw_create( void );
+
+void draw_destroy( struct draw_context *draw );
+
+void draw_set_viewport_state( struct draw_context *draw,
+                              const struct pipe_viewport_state *viewport );
+
+void draw_set_clip_state( struct draw_context *pipe,
+                          const struct pipe_clip_state *clip );
+
+void draw_set_rasterizer_state( struct draw_context *draw,
+                                const struct pipe_rasterizer_state *raster );
+
+void draw_set_rasterize_stage( struct draw_context *draw,
+                               struct draw_stage *stage );
+
+void draw_wide_point_threshold(struct draw_context *draw, float threshold);
+
+void draw_wide_line_threshold(struct draw_context *draw, float threshold);
+
+void draw_enable_line_stipple(struct draw_context *draw, boolean enable);
+
+void draw_enable_point_sprites(struct draw_context *draw, boolean enable);
+
+void draw_set_mrd(struct draw_context *draw, double mrd);
+
+boolean
+draw_install_aaline_stage(struct draw_context *draw, struct pipe_context *pipe);
+
+boolean
+draw_install_aapoint_stage(struct draw_context *draw, struct pipe_context *pipe);
+
+boolean
+draw_install_pstipple_stage(struct draw_context *draw, struct pipe_context *pipe);
+
+
+int
+draw_find_vs_output(const struct draw_context *draw,
+                    uint semantic_name, uint semantic_index);
+
+uint
+draw_num_vs_outputs(const struct draw_context *draw);
+
+
+void
+draw_texture_samplers(struct draw_context *draw,
+                      uint num_samplers,
+                      struct tgsi_sampler **samplers);
+
+
+
+/*
+ * Vertex shader functions
+ */
+
+struct draw_vertex_shader *
+draw_create_vertex_shader(struct draw_context *draw,
+                          const struct pipe_shader_state *shader);
+void draw_bind_vertex_shader(struct draw_context *draw,
+                             struct draw_vertex_shader *dvs);
+void draw_delete_vertex_shader(struct draw_context *draw,
+                               struct draw_vertex_shader *dvs);
+
+
+
+/*
+ * Vertex data functions
+ */
+
+void draw_set_vertex_buffers(struct draw_context *draw,
+                             unsigned count,
+                             const struct pipe_vertex_buffer *buffers);
+
+void draw_set_vertex_elements(struct draw_context *draw,
+			      unsigned count,
+                              const struct pipe_vertex_element *elements);
+
+void
+draw_set_mapped_element_buffer_range( struct draw_context *draw,
+                                      unsigned eltSize,
+                                      unsigned min_index,
+                                      unsigned max_index,
+                                      const void *elements );
+
+void draw_set_mapped_element_buffer( struct draw_context *draw,
+                                     unsigned eltSize, 
+                                     const void *elements );
+
+void draw_set_mapped_vertex_buffer(struct draw_context *draw,
+                                   unsigned attr, const void *buffer);
+
+void draw_set_mapped_constant_buffer(struct draw_context *draw,
+                                     const void *buffer,
+                                     unsigned size );
+
+void draw_set_edgeflags( struct draw_context *draw,
+                         const unsigned *edgeflag );
+
+
+/***********************************************************************
+ * draw_prim.c 
+ */
+
+void draw_arrays(struct draw_context *draw, unsigned prim,
+		 unsigned start, unsigned count);
+
+void draw_flush(struct draw_context *draw);
+
+
+/*******************************************************************************
+ * Driver backend interface 
+ */
+struct vbuf_render;
+void draw_set_render( struct draw_context *draw, 
+		      struct vbuf_render *render );
+
+void draw_set_driver_clipping( struct draw_context *draw,
+                               boolean bypass_clipping );
+
+void draw_set_force_passthrough( struct draw_context *draw, 
+                                 boolean enable );
+
+/*******************************************************************************
+ * Draw pipeline 
+ */
+boolean draw_need_pipeline(const struct draw_context *draw,
+                           const struct pipe_rasterizer_state *rasterizer,
+                           unsigned prim );
+
+
+
+#endif /* DRAW_CONTEXT_H */
diff --git a/src/gallium/auxiliary/draw/draw_pipe.c b/src/gallium/auxiliary/draw/draw_pipe.c
new file mode 100644
index 00000000000..3cde9d36d3a
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pipe.c
@@ -0,0 +1,287 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  */
+
+#include "draw/draw_private.h"
+#include "draw/draw_pipe.h"
+
+
+
+boolean draw_pipeline_init( struct draw_context *draw )
+{
+   /* create pipeline stages */
+   draw->pipeline.wide_line  = draw_wide_line_stage( draw );
+   draw->pipeline.wide_point = draw_wide_point_stage( draw );
+   draw->pipeline.stipple   = draw_stipple_stage( draw );
+   draw->pipeline.unfilled  = draw_unfilled_stage( draw );
+   draw->pipeline.twoside   = draw_twoside_stage( draw );
+   draw->pipeline.offset    = draw_offset_stage( draw );
+   draw->pipeline.clip      = draw_clip_stage( draw );
+   draw->pipeline.flatshade = draw_flatshade_stage( draw );
+   draw->pipeline.cull      = draw_cull_stage( draw );
+   draw->pipeline.validate  = draw_validate_stage( draw );
+   draw->pipeline.first     = draw->pipeline.validate;
+
+   if (!draw->pipeline.wide_line ||
+       !draw->pipeline.wide_point ||
+       !draw->pipeline.stipple ||
+       !draw->pipeline.unfilled ||
+       !draw->pipeline.twoside ||
+       !draw->pipeline.offset ||
+       !draw->pipeline.clip ||
+       !draw->pipeline.flatshade ||
+       !draw->pipeline.cull ||
+       !draw->pipeline.validate)
+      return FALSE;
+
+   /* these defaults are oriented toward the needs of softpipe */
+   draw->pipeline.wide_point_threshold = 1000000.0; /* infinity */
+   draw->pipeline.wide_line_threshold = 1.0;
+   draw->pipeline.line_stipple = TRUE;
+   draw->pipeline.point_sprite = TRUE;
+
+   return TRUE;
+}
+
+
+void draw_pipeline_destroy( struct draw_context *draw )
+{
+   if (draw->pipeline.wide_line)
+      draw->pipeline.wide_line->destroy( draw->pipeline.wide_line );
+   if (draw->pipeline.wide_point)
+      draw->pipeline.wide_point->destroy( draw->pipeline.wide_point );
+   if (draw->pipeline.stipple)
+      draw->pipeline.stipple->destroy( draw->pipeline.stipple );
+   if (draw->pipeline.unfilled)
+      draw->pipeline.unfilled->destroy( draw->pipeline.unfilled );
+   if (draw->pipeline.twoside)
+      draw->pipeline.twoside->destroy( draw->pipeline.twoside );
+   if (draw->pipeline.offset)
+      draw->pipeline.offset->destroy( draw->pipeline.offset );
+   if (draw->pipeline.clip)
+      draw->pipeline.clip->destroy( draw->pipeline.clip );
+   if (draw->pipeline.flatshade)
+      draw->pipeline.flatshade->destroy( draw->pipeline.flatshade );
+   if (draw->pipeline.cull)
+      draw->pipeline.cull->destroy( draw->pipeline.cull );
+   if (draw->pipeline.validate)
+      draw->pipeline.validate->destroy( draw->pipeline.validate );
+   if (draw->pipeline.aaline)
+      draw->pipeline.aaline->destroy( draw->pipeline.aaline );
+   if (draw->pipeline.aapoint)
+      draw->pipeline.aapoint->destroy( draw->pipeline.aapoint );
+   if (draw->pipeline.pstipple)
+      draw->pipeline.pstipple->destroy( draw->pipeline.pstipple );
+   if (draw->pipeline.rasterize)
+      draw->pipeline.rasterize->destroy( draw->pipeline.rasterize );
+}
+
+
+
+
+
+
+
+static void do_point( struct draw_context *draw,
+		      const char *v0 )
+{
+   struct prim_header prim;
+   
+   prim.flags = 0;
+   prim.pad = 0;
+   prim.v[0] = (struct vertex_header *)v0;
+
+   draw->pipeline.first->point( draw->pipeline.first, &prim );
+}
+
+
+static void do_line( struct draw_context *draw,
+                     ushort flags,
+		     const char *v0,
+		     const char *v1 )
+{
+   struct prim_header prim;
+   
+   prim.flags = flags;
+   prim.pad = 0;
+   prim.v[0] = (struct vertex_header *)v0;
+   prim.v[1] = (struct vertex_header *)v1;
+
+   draw->pipeline.first->line( draw->pipeline.first, &prim );
+}
+
+
+static void do_triangle( struct draw_context *draw,
+                         ushort flags,
+			 char *v0,
+			 char *v1,
+			 char *v2 )
+{
+   struct prim_header prim;
+   
+   prim.v[0] = (struct vertex_header *)v0;
+   prim.v[1] = (struct vertex_header *)v1;
+   prim.v[2] = (struct vertex_header *)v2;
+   prim.flags = flags;
+   prim.pad = 0;
+
+   draw->pipeline.first->tri( draw->pipeline.first, &prim );
+}
+
+
+
+
+/* Code to run the pipeline on a fairly arbitary collection of vertices.
+ *
+ * Vertex headers must be pre-initialized with the
+ * UNDEFINED_VERTEX_ID, this code will cause that id to become
+ * overwritten, so it may have to be reset if there is the intention
+ * to reuse the vertices.
+ *
+ * This code provides a callback to reset the vertex id's which the
+ * draw_vbuf.c code uses when it has to perform a flush.
+ */
+void draw_pipeline_run( struct draw_context *draw,
+                        unsigned prim,
+                        struct vertex_header *vertices,
+                        unsigned vertex_count,
+                        unsigned stride,
+                        const ushort *elts,
+                        unsigned count )
+{
+   char *verts = (char *)vertices;
+   unsigned i;
+
+   draw->pipeline.verts = verts;
+   draw->pipeline.vertex_stride = stride;
+   draw->pipeline.vertex_count = vertex_count;
+   
+   switch (prim) {
+   case PIPE_PRIM_POINTS:
+      for (i = 0; i < count; i++) 
+         do_point( draw, 
+                   verts + stride * elts[i] );
+      break;
+   case PIPE_PRIM_LINES:
+      for (i = 0; i+1 < count; i += 2) 
+         do_line( draw, 
+                  elts[i+0],  /* flags */
+                  verts + stride * (elts[i+0] & ~DRAW_PIPE_FLAG_MASK),
+                  verts + stride * elts[i+1]);
+      break;
+   case PIPE_PRIM_TRIANGLES:
+      for (i = 0; i+2 < count; i += 3)
+         do_triangle( draw, 
+                      elts[i+0],  /* flags */
+                      verts + stride * (elts[i+0] & ~DRAW_PIPE_FLAG_MASK),
+                      verts + stride * elts[i+1],
+                      verts + stride * elts[i+2]);
+      break;
+   }
+   
+   draw->pipeline.verts = NULL;
+   draw->pipeline.vertex_count = 0;
+}
+
+#define QUAD(i0,i1,i2,i3)                                        \
+   do_triangle( draw,                                            \
+                ( DRAW_PIPE_RESET_STIPPLE |                      \
+                  DRAW_PIPE_EDGE_FLAG_0 |                        \
+                  DRAW_PIPE_EDGE_FLAG_2 ),                       \
+                 verts + stride * ((i0) & ~DRAW_PIPE_FLAG_MASK), \
+                 verts + stride * (i1),                          \
+                verts + stride * (i3));                          \
+      do_triangle( draw,                                         \
+                   ( DRAW_PIPE_EDGE_FLAG_0 |                     \
+                     DRAW_PIPE_EDGE_FLAG_1 ),                    \
+                 verts + stride * ((i1) & ~DRAW_PIPE_FLAG_MASK), \
+                 verts + stride * (i2),                          \
+                 verts + stride * (i3))
+
+#define TRIANGLE(flags,i0,i1,i2)                                 \
+   do_triangle( draw,                                            \
+                flags,  /* flags */                              \
+                 verts + stride * ((i0) & ~DRAW_PIPE_FLAG_MASK), \
+                 verts + stride * (i1),                          \
+                 verts + stride * (i2))
+
+#define LINE(flags,i0,i1)                                   \
+   do_line( draw,                                           \
+            flags,                                          \
+            verts + stride * ((i0) & ~DRAW_PIPE_FLAG_MASK), \
+            verts + stride * (i1))
+
+#define POINT(i0)                               \
+   do_point( draw,                              \
+             verts + stride * i0 )
+
+#define FUNC pipe_run_linear
+#define ARGS                                    \
+    struct draw_context *draw,                  \
+    unsigned prim,                              \
+    struct vertex_header *vertices,             \
+    unsigned stride
+
+#define LOCAL_VARS                                           \
+   char *verts = (char *)vertices;                           \
+   boolean flatfirst = (draw->rasterizer->flatshade &&       \
+                        draw->rasterizer->flatshade_first);  \
+   unsigned i;                                               \
+   ushort flags
+
+#define FLUSH
+
+#include "draw_pt_decompose.h"
+
+void draw_pipeline_run_linear( struct draw_context *draw,
+                               unsigned prim,
+                               struct vertex_header *vertices,
+                               unsigned count,
+                               unsigned stride )
+{
+   char *verts = (char *)vertices;
+   draw->pipeline.verts = verts;
+   draw->pipeline.vertex_stride = stride;
+   draw->pipeline.vertex_count = count;
+
+   pipe_run_linear(draw, prim, vertices, stride, count);
+
+   draw->pipeline.verts = NULL;
+   draw->pipeline.vertex_count = 0;
+}
+
+
+void draw_pipeline_flush( struct draw_context *draw, 
+                          unsigned flags )
+{
+   draw->pipeline.first->flush( draw->pipeline.first, flags );
+   draw->pipeline.first = draw->pipeline.validate;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pipe.h b/src/gallium/auxiliary/draw/draw_pipe.h
new file mode 100644
index 00000000000..dbad8f98ac7
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pipe.h
@@ -0,0 +1,125 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  */
+
+#ifndef DRAW_PIPE_H
+#define DRAW_PIPE_H
+
+#include "pipe/p_compiler.h"
+#include "draw_private.h"       /* for sizeof(vertex_header) */
+
+
+/**
+ * Basic info for a point/line/triangle primitive.
+ */
+struct prim_header {
+   float det;                 /**< front/back face determinant */
+   ushort flags;
+   ushort pad;
+   struct vertex_header *v[3];  /**< 1 to 3 vertex pointers */
+};
+
+
+
+/**
+ * Base class for all primitive drawing stages.
+ */
+struct draw_stage
+{
+   struct draw_context *draw;   /**< parent context */
+
+   struct draw_stage *next;     /**< next stage in pipeline */
+
+   struct vertex_header **tmp;  /**< temp vert storage, such as for clipping */
+   unsigned nr_tmps;
+
+   void (*point)( struct draw_stage *,
+		  struct prim_header * );
+
+   void (*line)( struct draw_stage *,
+		 struct prim_header * );
+
+   void (*tri)( struct draw_stage *,
+		struct prim_header * );
+
+   void (*flush)( struct draw_stage *,
+		  unsigned flags );
+
+   void (*reset_stipple_counter)( struct draw_stage * );
+
+   void (*destroy)( struct draw_stage * );
+};
+
+
+extern struct draw_stage *draw_unfilled_stage( struct draw_context *context );
+extern struct draw_stage *draw_twoside_stage( struct draw_context *context );
+extern struct draw_stage *draw_offset_stage( struct draw_context *context );
+extern struct draw_stage *draw_clip_stage( struct draw_context *context );
+extern struct draw_stage *draw_flatshade_stage( struct draw_context *context );
+extern struct draw_stage *draw_cull_stage( struct draw_context *context );
+extern struct draw_stage *draw_stipple_stage( struct draw_context *context );
+extern struct draw_stage *draw_wide_line_stage( struct draw_context *context );
+extern struct draw_stage *draw_wide_point_stage( struct draw_context *context );
+extern struct draw_stage *draw_validate_stage( struct draw_context *context );
+
+
+extern void draw_free_temp_verts( struct draw_stage *stage );
+extern boolean draw_alloc_temp_verts( struct draw_stage *stage, unsigned nr );
+
+extern void draw_reset_vertex_ids( struct draw_context *draw );
+
+void draw_pipe_passthrough_tri(struct draw_stage *stage, struct prim_header *header);
+void draw_pipe_passthrough_line(struct draw_stage *stage, struct prim_header *header);
+void draw_pipe_passthrough_point(struct draw_stage *stage, struct prim_header *header);
+
+
+
+/**
+ * Get a writeable copy of a vertex.
+ * \param stage  drawing stage info
+ * \param vert  the vertex to copy (source)
+ * \param idx  index into stage's tmp[] array to put the copy (dest)
+ * \return  pointer to the copied vertex
+ */
+static INLINE struct vertex_header *
+dup_vert( struct draw_stage *stage,
+	  const struct vertex_header *vert,
+	  unsigned idx )
+{   
+   struct vertex_header *tmp = stage->tmp[idx];
+   const uint vsize = sizeof(struct vertex_header)
+      + stage->draw->vs.num_vs_outputs * 4 * sizeof(float);
+   memcpy(tmp, vert, vsize);
+   tmp->vertex_id = UNDEFINED_VERTEX_ID;
+   return tmp;
+}
+
+#endif
diff --git a/src/gallium/auxiliary/draw/draw_pipe_aaline.c b/src/gallium/auxiliary/draw/draw_pipe_aaline.c
new file mode 100644
index 00000000000..20841bb5d69
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pipe_aaline.c
@@ -0,0 +1,914 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * AA line stage:  AA lines are converted to texture mapped triangles.
+ *
+ * Authors:  Brian Paul
+ */
+
+
+#include "pipe/p_inlines.h"
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
+
+#include "tgsi/tgsi_transform.h"
+#include "tgsi/tgsi_dump.h"
+
+#include "draw_context.h"
+#include "draw_private.h"
+#include "draw_pipe.h"
+
+
+/**
+ * Max texture level for the alpha texture used for antialiasing
+ */
+#define MAX_TEXTURE_LEVEL  5   /* 32 x 32 */
+
+
+/**
+ * Subclass of pipe_shader_state to carry extra fragment shader info.
+ */
+struct aaline_fragment_shader
+{
+   struct pipe_shader_state state;
+   void *driver_fs;
+   void *aaline_fs;
+   void *aapoint_fs; /* not yet */
+   void *sprite_fs; /* not yet */
+   uint sampler_unit;
+   int generic_attrib;  /**< texcoord/generic used for texture */
+};
+
+
+/**
+ * Subclass of draw_stage
+ */
+struct aaline_stage
+{
+   struct draw_stage stage;
+
+   float half_line_width;
+
+   /** For AA lines, this is the vertex attrib slot for the new texcoords */
+   uint tex_slot;
+   /** position, not necessarily output zero */
+   uint pos_slot;
+
+   void *sampler_cso;
+   struct pipe_texture *texture;
+   uint num_samplers;
+   uint num_textures;
+
+
+   /*
+    * Currently bound state
+    */
+   struct aaline_fragment_shader *fs;
+   struct {
+      void *sampler[PIPE_MAX_SAMPLERS];
+      struct pipe_texture *texture[PIPE_MAX_SAMPLERS];
+   } state;
+
+   /*
+    * Driver interface/override functions
+    */
+   void * (*driver_create_fs_state)(struct pipe_context *,
+                                    const struct pipe_shader_state *);
+   void (*driver_bind_fs_state)(struct pipe_context *, void *);
+   void (*driver_delete_fs_state)(struct pipe_context *, void *);
+
+   void (*driver_bind_sampler_states)(struct pipe_context *, unsigned,
+                                      void **);
+   void (*driver_set_sampler_textures)(struct pipe_context *, unsigned,
+                                       struct pipe_texture **);
+
+   struct pipe_context *pipe;
+};
+
+
+
+/**
+ * Subclass of tgsi_transform_context, used for transforming the
+ * user's fragment shader to add the special AA instructions.
+ */
+struct aa_transform_context {
+   struct tgsi_transform_context base;
+   uint tempsUsed;  /**< bitmask */
+   int colorOutput; /**< which output is the primary color */
+   uint samplersUsed;  /**< bitfield of samplers used */
+   int freeSampler;  /** an available sampler for the pstipple */
+   int maxInput, maxGeneric;  /**< max input index found */
+   int colorTemp, texTemp;  /**< temp registers */
+   boolean firstInstruction;
+};
+
+
+/**
+ * TGSI declaration transform callback.
+ * Look for a free sampler, a free input attrib, and two free temp regs.
+ */
+static void
+aa_transform_decl(struct tgsi_transform_context *ctx,
+                  struct tgsi_full_declaration *decl)
+{
+   struct aa_transform_context *aactx = (struct aa_transform_context *) ctx;
+
+   if (decl->Declaration.File == TGSI_FILE_OUTPUT &&
+       decl->Semantic.SemanticName == TGSI_SEMANTIC_COLOR &&
+       decl->Semantic.SemanticIndex == 0) {
+      aactx->colorOutput = decl->DeclarationRange.First;
+   }
+   else if (decl->Declaration.File == TGSI_FILE_SAMPLER) {
+      uint i;
+      for (i = decl->DeclarationRange.First;
+           i <= decl->DeclarationRange.Last; i++) {
+         aactx->samplersUsed |= 1 << i;
+      }
+   }
+   else if (decl->Declaration.File == TGSI_FILE_INPUT) {
+      if ((int) decl->DeclarationRange.Last > aactx->maxInput)
+         aactx->maxInput = decl->DeclarationRange.Last;
+      if (decl->Semantic.SemanticName == TGSI_SEMANTIC_GENERIC &&
+           (int) decl->Semantic.SemanticIndex > aactx->maxGeneric) {
+         aactx->maxGeneric = decl->Semantic.SemanticIndex;
+      }
+   }
+   else if (decl->Declaration.File == TGSI_FILE_TEMPORARY) {
+      uint i;
+      for (i = decl->DeclarationRange.First;
+           i <= decl->DeclarationRange.Last; i++) {
+         aactx->tempsUsed |= (1 << i);
+      }
+   }
+
+   ctx->emit_declaration(ctx, decl);
+}
+
+
+/**
+ * Find the lowest zero bit in the given word, or -1 if bitfield is all ones.
+ */
+static int
+free_bit(uint bitfield)
+{
+   int i;
+   for (i = 0; i < 32; i++) {
+      if ((bitfield & (1 << i)) == 0)
+         return i;
+   }
+   return -1;
+}
+
+
+/**
+ * TGSI instruction transform callback.
+ * Replace writes to result.color w/ a temp reg.
+ * Upon END instruction, insert texture sampling code for antialiasing.
+ */
+static void
+aa_transform_inst(struct tgsi_transform_context *ctx,
+                  struct tgsi_full_instruction *inst)
+{
+   struct aa_transform_context *aactx = (struct aa_transform_context *) ctx;
+
+   if (aactx->firstInstruction) {
+      /* emit our new declarations before the first instruction */
+
+      struct tgsi_full_declaration decl;
+      uint i;
+
+      /* find free sampler */
+      aactx->freeSampler = free_bit(aactx->samplersUsed);
+      if (aactx->freeSampler >= PIPE_MAX_SAMPLERS)
+         aactx->freeSampler = PIPE_MAX_SAMPLERS - 1;
+
+      /* find two free temp regs */
+      for (i = 0; i < 32; i++) {
+         if ((aactx->tempsUsed & (1 << i)) == 0) {
+            /* found a free temp */
+            if (aactx->colorTemp < 0)
+               aactx->colorTemp  = i;
+            else if (aactx->texTemp < 0)
+               aactx->texTemp  = i;
+            else
+               break;
+         }
+      }
+      assert(aactx->colorTemp >= 0);
+      assert(aactx->texTemp >= 0);
+
+      /* declare new generic input/texcoord */
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_INPUT;
+      /* XXX this could be linear... */
+      decl.Declaration.Interpolate = TGSI_INTERPOLATE_PERSPECTIVE;
+      decl.Declaration.Semantic = 1;
+      decl.Semantic.SemanticName = TGSI_SEMANTIC_GENERIC;
+      decl.Semantic.SemanticIndex = aactx->maxGeneric + 1;
+      decl.DeclarationRange.First = 
+      decl.DeclarationRange.Last = aactx->maxInput + 1;
+      ctx->emit_declaration(ctx, &decl);
+
+      /* declare new sampler */
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_SAMPLER;
+      decl.DeclarationRange.First = 
+      decl.DeclarationRange.Last = aactx->freeSampler;
+      ctx->emit_declaration(ctx, &decl);
+
+      /* declare new temp regs */
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_TEMPORARY;
+      decl.DeclarationRange.First = 
+      decl.DeclarationRange.Last = aactx->texTemp;
+      ctx->emit_declaration(ctx, &decl);
+
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_TEMPORARY;
+      decl.DeclarationRange.First = 
+      decl.DeclarationRange.Last = aactx->colorTemp;
+      ctx->emit_declaration(ctx, &decl);
+
+      aactx->firstInstruction = FALSE;
+   }
+
+   if (inst->Instruction.Opcode == TGSI_OPCODE_END &&
+       aactx->colorOutput != -1) {
+      struct tgsi_full_instruction newInst;
+
+      /* TEX */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_TEX;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = aactx->texTemp;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.InstructionExtTexture.Texture = TGSI_TEXTURE_2D;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = aactx->maxInput + 1;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_SAMPLER;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = aactx->freeSampler;
+
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* MOV rgb */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_MOV;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_OUTPUT;
+      newInst.FullDstRegisters[0].DstRegister.Index = aactx->colorOutput;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_XYZ;
+      newInst.Instruction.NumSrcRegs = 1;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = aactx->colorTemp;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* MUL alpha */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_MUL;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_OUTPUT;
+      newInst.FullDstRegisters[0].DstRegister.Index = aactx->colorOutput;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_W;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = aactx->colorTemp;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = aactx->texTemp;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* END */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_END;
+      newInst.Instruction.NumDstRegs = 0;
+      newInst.Instruction.NumSrcRegs = 0;
+      ctx->emit_instruction(ctx, &newInst);
+   }
+   else {
+      /* Not an END instruction.
+       * Look for writes to result.color and replace with colorTemp reg.
+       */
+      uint i;
+
+      for (i = 0; i < inst->Instruction.NumDstRegs; i++) {
+         struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i];
+         if (dst->DstRegister.File == TGSI_FILE_OUTPUT &&
+             dst->DstRegister.Index == aactx->colorOutput) {
+            dst->DstRegister.File = TGSI_FILE_TEMPORARY;
+            dst->DstRegister.Index = aactx->colorTemp;
+         }
+      }
+
+      ctx->emit_instruction(ctx, inst);
+   }
+}
+
+
+/**
+ * Generate the frag shader we'll use for drawing AA lines.
+ * This will be the user's shader plus some texture/modulate instructions.
+ */
+static boolean
+generate_aaline_fs(struct aaline_stage *aaline)
+{
+   const struct pipe_shader_state *orig_fs = &aaline->fs->state;
+   struct pipe_shader_state aaline_fs;
+   struct aa_transform_context transform;
+
+#define MAX 1000
+
+   aaline_fs = *orig_fs; /* copy to init */
+   aaline_fs.tokens = MALLOC(sizeof(struct tgsi_token) * MAX);
+   if (aaline_fs.tokens == NULL)
+      return FALSE;
+
+   memset(&transform, 0, sizeof(transform));
+   transform.colorOutput = -1;
+   transform.maxInput = -1;
+   transform.maxGeneric = -1;
+   transform.colorTemp = -1;
+   transform.texTemp = -1;
+   transform.firstInstruction = TRUE;
+   transform.base.transform_instruction = aa_transform_inst;
+   transform.base.transform_declaration = aa_transform_decl;
+
+   tgsi_transform_shader(orig_fs->tokens,
+                         (struct tgsi_token *) aaline_fs.tokens,
+                         MAX, &transform.base);
+
+#if 0 /* DEBUG */
+   tgsi_dump(orig_fs->tokens, 0);
+   tgsi_dump(aaline_fs.tokens, 0);
+#endif
+
+   aaline->fs->sampler_unit = transform.freeSampler;
+
+   aaline->fs->aaline_fs
+      = aaline->driver_create_fs_state(aaline->pipe, &aaline_fs);
+   if (aaline->fs->aaline_fs == NULL)
+      return FALSE;
+
+   aaline->fs->generic_attrib = transform.maxGeneric + 1;
+   return TRUE;
+}
+
+
+/**
+ * Create the texture map we'll use for antialiasing the lines.
+ */
+static boolean
+aaline_create_texture(struct aaline_stage *aaline)
+{
+   struct pipe_context *pipe = aaline->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   struct pipe_texture texTemp;
+   uint level;
+
+   memset(&texTemp, 0, sizeof(texTemp));
+   texTemp.target = PIPE_TEXTURE_2D;
+   texTemp.format = PIPE_FORMAT_A8_UNORM; /* XXX verify supported by driver! */
+   texTemp.last_level = MAX_TEXTURE_LEVEL;
+   texTemp.width[0] = 1 << MAX_TEXTURE_LEVEL;
+   texTemp.height[0] = 1 << MAX_TEXTURE_LEVEL;
+   texTemp.depth[0] = 1;
+   pf_get_block(texTemp.format, &texTemp.block);
+
+   aaline->texture = screen->texture_create(screen, &texTemp);
+   if (!aaline->texture)
+      return FALSE;
+
+   /* Fill in mipmap images.
+    * Basically each level is solid opaque, except for the outermost
+    * texels which are zero.  Special case the 1x1 and 2x2 levels.
+    */
+   for (level = 0; level <= MAX_TEXTURE_LEVEL; level++) {
+      struct pipe_surface *surface;
+      const uint size = aaline->texture->width[level];
+      ubyte *data;
+      uint i, j;
+
+      assert(aaline->texture->width[level] == aaline->texture->height[level]);
+
+      /* This texture is new, no need to flush. 
+       */
+      surface = screen->get_tex_surface(screen, aaline->texture, 0, level, 0,
+                                        PIPE_BUFFER_USAGE_CPU_WRITE);
+      data = screen->surface_map(screen, surface, PIPE_BUFFER_USAGE_CPU_WRITE);
+      if (data == NULL)
+         return FALSE;
+
+      for (i = 0; i < size; i++) {
+         for (j = 0; j < size; j++) {
+            ubyte d;
+            if (size == 1) {
+               d = 255;
+            }
+            else if (size == 2) {
+               d = 200; /* tuneable */
+            }
+            else if (i == 0 || j == 0 || i == size - 1 || j == size - 1) {
+               d = 0;
+            }
+            else {
+               d = 255;
+            }
+            data[i * surface->stride + j] = d;
+         }
+      }
+
+      /* unmap */
+      screen->surface_unmap(screen, surface);
+      screen->tex_surface_release(screen, &surface);
+   }
+   return TRUE;
+}
+
+
+/**
+ * Create the sampler CSO that'll be used for antialiasing.
+ * By using a mipmapped texture, we don't have to generate a different
+ * texture image for each line size.
+ */
+static boolean
+aaline_create_sampler(struct aaline_stage *aaline)
+{
+   struct pipe_sampler_state sampler;
+   struct pipe_context *pipe = aaline->pipe;
+
+   memset(&sampler, 0, sizeof(sampler));
+   sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   sampler.min_mip_filter = PIPE_TEX_MIPFILTER_LINEAR;
+   sampler.min_img_filter = PIPE_TEX_FILTER_LINEAR;
+   sampler.mag_img_filter = PIPE_TEX_FILTER_LINEAR;
+   sampler.normalized_coords = 1;
+   sampler.min_lod = 0.0f;
+   sampler.max_lod = MAX_TEXTURE_LEVEL;
+
+   aaline->sampler_cso = pipe->create_sampler_state(pipe, &sampler);
+   if (aaline->sampler_cso == NULL)
+      return FALSE;
+
+   return TRUE;
+}
+
+
+/**
+ * When we're about to draw our first AA line in a batch, this function is
+ * called to tell the driver to bind our modified fragment shader.
+ */
+static boolean
+bind_aaline_fragment_shader(struct aaline_stage *aaline)
+{
+   struct draw_context *draw = aaline->stage.draw;
+
+   if (!aaline->fs->aaline_fs && 
+       !generate_aaline_fs(aaline))
+      return FALSE;
+
+   draw->suspend_flushing = TRUE;
+   aaline->driver_bind_fs_state(aaline->pipe, aaline->fs->aaline_fs);
+   draw->suspend_flushing = FALSE;
+
+   return TRUE;
+}
+
+
+
+static INLINE struct aaline_stage *
+aaline_stage( struct draw_stage *stage )
+{
+   return (struct aaline_stage *) stage;
+}
+
+
+/**
+ * Draw a wide line by drawing a quad, using geometry which will
+ * fullfill GL's antialiased line requirements.
+ */
+static void
+aaline_line(struct draw_stage *stage, struct prim_header *header)
+{
+   const struct aaline_stage *aaline = aaline_stage(stage);
+   const float half_width = aaline->half_line_width;
+   struct prim_header tri;
+   struct vertex_header *v[8];
+   uint texPos = aaline->tex_slot;
+   uint posPos = aaline->pos_slot;
+   float *pos, *tex;
+   float dx = header->v[1]->data[posPos][0] - header->v[0]->data[posPos][0];
+   float dy = header->v[1]->data[posPos][1] - header->v[0]->data[posPos][1];
+   double a = atan2(dy, dx);
+   float c_a = (float) cos(a), s_a = (float) sin(a);
+   uint i;
+
+   /* XXX the ends of lines aren't quite perfect yet, but probably passable */
+   dx = 0.5F * half_width;
+   dy = half_width;
+
+   /* allocate/dup new verts */
+   for (i = 0; i < 8; i++) {
+      v[i] = dup_vert(stage, header->v[i/4], i);
+   }
+
+   /*
+    * Quad strip for line from v0 to v1 (*=endpoints):
+    *
+    *  1   3                     5   7
+    *  +---+---------------------+---+
+    *  |                             |
+    *  | *v0                     v1* |
+    *  |                             |
+    *  +---+---------------------+---+
+    *  0   2                     4   6
+    */
+
+   /* new verts */
+   pos = v[0]->data[posPos];
+   pos[0] += (-dx * c_a -  dy * s_a);
+   pos[1] += (-dx * s_a +  dy * c_a);
+
+   pos = v[1]->data[posPos];
+   pos[0] += (-dx * c_a - -dy * s_a);
+   pos[1] += (-dx * s_a + -dy * c_a);
+
+   pos = v[2]->data[posPos];
+   pos[0] += ( dx * c_a -  dy * s_a);
+   pos[1] += ( dx * s_a +  dy * c_a);
+
+   pos = v[3]->data[posPos];
+   pos[0] += ( dx * c_a - -dy * s_a);
+   pos[1] += ( dx * s_a + -dy * c_a);
+
+   pos = v[4]->data[posPos];
+   pos[0] += (-dx * c_a -  dy * s_a);
+   pos[1] += (-dx * s_a +  dy * c_a);
+
+   pos = v[5]->data[posPos];
+   pos[0] += (-dx * c_a - -dy * s_a);
+   pos[1] += (-dx * s_a + -dy * c_a);
+
+   pos = v[6]->data[posPos];
+   pos[0] += ( dx * c_a -  dy * s_a);
+   pos[1] += ( dx * s_a +  dy * c_a);
+
+   pos = v[7]->data[posPos];
+   pos[0] += ( dx * c_a - -dy * s_a);
+   pos[1] += ( dx * s_a + -dy * c_a);
+
+   /* new texcoords */
+   tex = v[0]->data[texPos];
+   ASSIGN_4V(tex, 0, 0, 0, 1);
+
+   tex = v[1]->data[texPos];
+   ASSIGN_4V(tex, 0, 1, 0, 1);
+
+   tex = v[2]->data[texPos];
+   ASSIGN_4V(tex, .5, 0, 0, 1);
+
+   tex = v[3]->data[texPos];
+   ASSIGN_4V(tex, .5, 1, 0, 1);
+
+   tex = v[4]->data[texPos];
+   ASSIGN_4V(tex, .5, 0, 0, 1);
+
+   tex = v[5]->data[texPos];
+   ASSIGN_4V(tex, .5, 1, 0, 1);
+
+   tex = v[6]->data[texPos];
+   ASSIGN_4V(tex, 1, 0, 0, 1);
+
+   tex = v[7]->data[texPos];
+   ASSIGN_4V(tex, 1, 1, 0, 1);
+
+   /* emit 6 tris for the quad strip */
+   tri.v[0] = v[2];  tri.v[1] = v[1];  tri.v[2] = v[0];
+   stage->next->tri( stage->next, &tri );
+
+   tri.v[0] = v[3];  tri.v[1] = v[1];  tri.v[2] = v[2];
+   stage->next->tri( stage->next, &tri );
+
+   tri.v[0] = v[4];  tri.v[1] = v[3];  tri.v[2] = v[2];
+   stage->next->tri( stage->next, &tri );
+
+   tri.v[0] = v[5];  tri.v[1] = v[3];  tri.v[2] = v[4];
+   stage->next->tri( stage->next, &tri );
+
+   tri.v[0] = v[6];  tri.v[1] = v[5];  tri.v[2] = v[4];
+   stage->next->tri( stage->next, &tri );
+
+   tri.v[0] = v[7];  tri.v[1] = v[5];  tri.v[2] = v[6];
+   stage->next->tri( stage->next, &tri );
+}
+
+
+static void
+aaline_first_line(struct draw_stage *stage, struct prim_header *header)
+{
+   auto struct aaline_stage *aaline = aaline_stage(stage);
+   struct draw_context *draw = stage->draw;
+   struct pipe_context *pipe = aaline->pipe;
+   uint num_samplers;
+
+   assert(draw->rasterizer->line_smooth);
+
+   if (draw->rasterizer->line_width <= 3.0)
+      aaline->half_line_width = 1.5f;
+   else
+      aaline->half_line_width = 0.5f * draw->rasterizer->line_width;
+
+   /*
+    * Bind (generate) our fragprog, sampler and texture
+    */
+   if (!bind_aaline_fragment_shader(aaline)) {
+      stage->line = draw_pipe_passthrough_line;
+      stage->line(stage, header);
+      return;
+   }
+
+   /* update vertex attrib info */
+   aaline->tex_slot = draw->vs.num_vs_outputs;
+   aaline->pos_slot = draw->vs.position_output;
+
+   /* advertise the extra post-transformed vertex attribute */
+   draw->extra_vp_outputs.semantic_name = TGSI_SEMANTIC_GENERIC;
+   draw->extra_vp_outputs.semantic_index = aaline->fs->generic_attrib;
+   draw->extra_vp_outputs.slot = aaline->tex_slot;
+
+   /* how many samplers? */
+   /* we'll use sampler/texture[pstip->sampler_unit] for the stipple */
+   num_samplers = MAX2(aaline->num_textures, aaline->num_samplers);
+   num_samplers = MAX2(num_samplers, aaline->fs->sampler_unit + 1);
+
+   aaline->state.sampler[aaline->fs->sampler_unit] = aaline->sampler_cso;
+   pipe_texture_reference(&aaline->state.texture[aaline->fs->sampler_unit],
+                          aaline->texture);
+
+   draw->suspend_flushing = TRUE;
+   aaline->driver_bind_sampler_states(pipe, num_samplers, aaline->state.sampler);
+   aaline->driver_set_sampler_textures(pipe, num_samplers, aaline->state.texture);
+   draw->suspend_flushing = FALSE;
+
+   /* now really draw first line */
+   stage->line = aaline_line;
+   stage->line(stage, header);
+}
+
+
+static void
+aaline_flush(struct draw_stage *stage, unsigned flags)
+{
+   struct draw_context *draw = stage->draw;
+   struct aaline_stage *aaline = aaline_stage(stage);
+   struct pipe_context *pipe = aaline->pipe;
+
+   stage->line = aaline_first_line;
+   stage->next->flush( stage->next, flags );
+
+   /* restore original frag shader, texture, sampler state */
+   draw->suspend_flushing = TRUE;
+   aaline->driver_bind_fs_state(pipe, aaline->fs->driver_fs);
+   aaline->driver_bind_sampler_states(pipe, aaline->num_samplers,
+                                      aaline->state.sampler);
+   aaline->driver_set_sampler_textures(pipe, aaline->num_textures,
+                                       aaline->state.texture);
+   draw->suspend_flushing = FALSE;
+
+   draw->extra_vp_outputs.slot = 0;
+}
+
+
+static void
+aaline_reset_stipple_counter(struct draw_stage *stage)
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void
+aaline_destroy(struct draw_stage *stage)
+{
+   struct aaline_stage *aaline = aaline_stage(stage);
+   uint i;
+
+   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+      pipe_texture_reference(&aaline->state.texture[i], NULL);
+   }
+
+   if (aaline->sampler_cso)
+      aaline->pipe->delete_sampler_state(aaline->pipe, aaline->sampler_cso);
+
+   if (aaline->texture)
+      pipe_texture_release(&aaline->texture);
+
+   draw_free_temp_verts( stage );
+
+   FREE( stage );
+}
+
+
+static struct aaline_stage *
+draw_aaline_stage(struct draw_context *draw)
+{
+   struct aaline_stage *aaline = CALLOC_STRUCT(aaline_stage);
+   if (aaline == NULL)
+      return NULL;
+
+   if (!draw_alloc_temp_verts( &aaline->stage, 8 ))
+      goto fail;
+
+   aaline->stage.draw = draw;
+   aaline->stage.next = NULL;
+   aaline->stage.point = draw_pipe_passthrough_point;
+   aaline->stage.line = aaline_first_line;
+   aaline->stage.tri = draw_pipe_passthrough_tri;
+   aaline->stage.flush = aaline_flush;
+   aaline->stage.reset_stipple_counter = aaline_reset_stipple_counter;
+   aaline->stage.destroy = aaline_destroy;
+
+   return aaline;
+
+ fail:
+   if (aaline)
+      aaline_destroy(&aaline->stage);
+
+   return NULL;
+}
+
+
+static struct aaline_stage *
+aaline_stage_from_pipe(struct pipe_context *pipe)
+{
+   struct draw_context *draw = (struct draw_context *) pipe->draw;
+   return aaline_stage(draw->pipeline.aaline);
+}
+
+
+/**
+ * This function overrides the driver's create_fs_state() function and
+ * will typically be called by the state tracker.
+ */
+static void *
+aaline_create_fs_state(struct pipe_context *pipe,
+                       const struct pipe_shader_state *fs)
+{
+   struct aaline_stage *aaline = aaline_stage_from_pipe(pipe);
+   struct aaline_fragment_shader *aafs = CALLOC_STRUCT(aaline_fragment_shader);
+   if (aafs == NULL)
+      return NULL;
+
+   aafs->state = *fs;
+
+   /* pass-through */
+   aafs->driver_fs = aaline->driver_create_fs_state(aaline->pipe, fs);
+
+   return aafs;
+}
+
+
+static void
+aaline_bind_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct aaline_stage *aaline = aaline_stage_from_pipe(pipe);
+   struct aaline_fragment_shader *aafs = (struct aaline_fragment_shader *) fs;
+
+   /* save current */
+   aaline->fs = aafs;
+   /* pass-through */
+   aaline->driver_bind_fs_state(aaline->pipe,
+                                (aafs ? aafs->driver_fs : NULL));
+}
+
+
+static void
+aaline_delete_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct aaline_stage *aaline = aaline_stage_from_pipe(pipe);
+   struct aaline_fragment_shader *aafs = (struct aaline_fragment_shader *) fs;
+   /* pass-through */
+   aaline->driver_delete_fs_state(aaline->pipe, aafs->driver_fs);
+   FREE(aafs);
+}
+
+
+static void
+aaline_bind_sampler_states(struct pipe_context *pipe,
+                           unsigned num, void **sampler)
+{
+   struct aaline_stage *aaline = aaline_stage_from_pipe(pipe);
+
+   /* save current */
+   memcpy(aaline->state.sampler, sampler, num * sizeof(void *));
+   aaline->num_samplers = num;
+
+   /* pass-through */
+   aaline->driver_bind_sampler_states(aaline->pipe, num, sampler);
+}
+
+
+static void
+aaline_set_sampler_textures(struct pipe_context *pipe,
+                            unsigned num, struct pipe_texture **texture)
+{
+   struct aaline_stage *aaline = aaline_stage_from_pipe(pipe);
+   uint i;
+
+   /* save current */
+   for (i = 0; i < num; i++) {
+      pipe_texture_reference(&aaline->state.texture[i], texture[i]);
+   }
+   for ( ; i < PIPE_MAX_SAMPLERS; i++) {
+      pipe_texture_reference(&aaline->state.texture[i], NULL);
+   }
+   aaline->num_textures = num;
+
+   /* pass-through */
+   aaline->driver_set_sampler_textures(aaline->pipe, num, texture);
+}
+
+
+/**
+ * Called by drivers that want to install this AA line prim stage
+ * into the draw module's pipeline.  This will not be used if the
+ * hardware has native support for AA lines.
+ */
+boolean
+draw_install_aaline_stage(struct draw_context *draw, struct pipe_context *pipe)
+{
+   struct aaline_stage *aaline;
+
+   pipe->draw = (void *) draw;
+
+   /*
+    * Create / install AA line drawing / prim stage
+    */
+   aaline = draw_aaline_stage( draw );
+   if (!aaline)
+      goto fail;
+
+   aaline->pipe = pipe;
+
+   /* create special texture, sampler state */
+   if (!aaline_create_texture(aaline))
+      goto fail;
+
+   if (!aaline_create_sampler(aaline))
+      goto fail;
+
+   /* save original driver functions */
+   aaline->driver_create_fs_state = pipe->create_fs_state;
+   aaline->driver_bind_fs_state = pipe->bind_fs_state;
+   aaline->driver_delete_fs_state = pipe->delete_fs_state;
+
+   aaline->driver_bind_sampler_states = pipe->bind_sampler_states;
+   aaline->driver_set_sampler_textures = pipe->set_sampler_textures;
+
+   /* override the driver's functions */
+   pipe->create_fs_state = aaline_create_fs_state;
+   pipe->bind_fs_state = aaline_bind_fs_state;
+   pipe->delete_fs_state = aaline_delete_fs_state;
+
+   pipe->bind_sampler_states = aaline_bind_sampler_states;
+   pipe->set_sampler_textures = aaline_set_sampler_textures;
+   
+   /* Install once everything is known to be OK:
+    */
+   draw->pipeline.aaline = &aaline->stage;
+
+   return TRUE;
+
+ fail:
+   if (aaline)
+      aaline->stage.destroy( &aaline->stage );
+   
+   return FALSE;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pipe_aapoint.c b/src/gallium/auxiliary/draw/draw_pipe_aapoint.c
new file mode 100644
index 00000000000..2c1cacbdb47
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pipe_aapoint.c
@@ -0,0 +1,875 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * AA point stage:  AA points are converted to quads and rendered with a
+ * special fragment shader.  Another approach would be to use a texture
+ * map image of a point, but experiments indicate the quality isn't nearly
+ * as good as this approach.
+ *
+ * Note: this looks a lot like draw_aaline.c but there's actually little
+ * if any code that can be shared.
+ *
+ * Authors:  Brian Paul
+ */
+
+
+#include "pipe/p_inlines.h"
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "tgsi/tgsi_transform.h"
+#include "tgsi/tgsi_dump.h"
+
+#include "util/u_math.h"
+#include "util/u_memory.h"
+
+#include "draw_context.h"
+#include "draw_vs.h"
+#include "draw_pipe.h"
+
+
+/*
+ * Enabling NORMALIZE might give _slightly_ better results.
+ * Basically, it controls whether we compute distance as d=sqrt(x*x+y*y) or
+ * d=x*x+y*y.  Since we're working with a unit circle, the later seems
+ * close enough and saves some costly instructions.
+ */
+#define NORMALIZE 0
+
+
+/**
+ * Subclass of pipe_shader_state to carry extra fragment shader info.
+ */
+struct aapoint_fragment_shader
+{
+   struct pipe_shader_state state;
+   void *driver_fs;   /**< the regular shader */
+   void *aapoint_fs;  /**< the aa point-augmented shader */
+   int generic_attrib; /**< The generic input attrib/texcoord we'll use */
+};
+
+
+/**
+ * Subclass of draw_stage
+ */
+struct aapoint_stage
+{
+   struct draw_stage stage;
+
+   int psize_slot;
+   float radius;
+
+   /** this is the vertex attrib slot for the new texcoords */
+   uint tex_slot;
+   uint pos_slot;
+
+   /*
+    * Currently bound state
+    */
+   struct aapoint_fragment_shader *fs;
+
+   /*
+    * Driver interface/override functions
+    */
+   void * (*driver_create_fs_state)(struct pipe_context *,
+                                    const struct pipe_shader_state *);
+   void (*driver_bind_fs_state)(struct pipe_context *, void *);
+   void (*driver_delete_fs_state)(struct pipe_context *, void *);
+
+   struct pipe_context *pipe;
+};
+
+
+
+/**
+ * Subclass of tgsi_transform_context, used for transforming the
+ * user's fragment shader to add the special AA instructions.
+ */
+struct aa_transform_context {
+   struct tgsi_transform_context base;
+   uint tempsUsed;  /**< bitmask */
+   int colorOutput; /**< which output is the primary color */
+   int maxInput, maxGeneric;  /**< max input index found */
+   int tmp0, colorTemp;  /**< temp registers */
+   boolean firstInstruction;
+};
+
+
+/**
+ * TGSI declaration transform callback.
+ * Look for two free temp regs and available input reg for new texcoords.
+ */
+static void
+aa_transform_decl(struct tgsi_transform_context *ctx,
+                  struct tgsi_full_declaration *decl)
+{
+   struct aa_transform_context *aactx = (struct aa_transform_context *) ctx;
+
+   if (decl->Declaration.File == TGSI_FILE_OUTPUT &&
+       decl->Semantic.SemanticName == TGSI_SEMANTIC_COLOR &&
+       decl->Semantic.SemanticIndex == 0) {
+      aactx->colorOutput = decl->DeclarationRange.First;
+   }
+   else if (decl->Declaration.File == TGSI_FILE_INPUT) {
+      if ((int) decl->DeclarationRange.Last > aactx->maxInput)
+         aactx->maxInput = decl->DeclarationRange.Last;
+      if (decl->Semantic.SemanticName == TGSI_SEMANTIC_GENERIC &&
+           (int) decl->Semantic.SemanticIndex > aactx->maxGeneric) {
+         aactx->maxGeneric = decl->Semantic.SemanticIndex;
+      }
+   }
+   else if (decl->Declaration.File == TGSI_FILE_TEMPORARY) {
+      uint i;
+      for (i = decl->DeclarationRange.First;
+           i <= decl->DeclarationRange.Last; i++) {
+         aactx->tempsUsed |= (1 << i);
+      }
+   }
+
+   ctx->emit_declaration(ctx, decl);
+}
+
+
+/**
+ * TGSI instruction transform callback.
+ * Replace writes to result.color w/ a temp reg.
+ * Upon END instruction, insert texture sampling code for antialiasing.
+ */
+static void
+aa_transform_inst(struct tgsi_transform_context *ctx,
+                  struct tgsi_full_instruction *inst)
+{
+   struct aa_transform_context *aactx = (struct aa_transform_context *) ctx;
+   struct tgsi_full_instruction newInst;
+
+   if (aactx->firstInstruction) {
+      /* emit our new declarations before the first instruction */
+
+      struct tgsi_full_declaration decl;
+      const int texInput = aactx->maxInput + 1;
+      int tmp0;
+      uint i;
+
+      /* find two free temp regs */
+      for (i = 0; i < 32; i++) {
+         if ((aactx->tempsUsed & (1 << i)) == 0) {
+            /* found a free temp */
+            if (aactx->tmp0 < 0)
+               aactx->tmp0 = i;
+            else if (aactx->colorTemp < 0)
+               aactx->colorTemp = i;
+            else
+               break;
+         }
+      }
+
+      assert(aactx->colorTemp != aactx->tmp0);
+
+      tmp0 = aactx->tmp0;
+
+      /* declare new generic input/texcoord */
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_INPUT;
+      /* XXX this could be linear... */
+      decl.Declaration.Interpolate = TGSI_INTERPOLATE_PERSPECTIVE;
+      decl.Declaration.Semantic = 1;
+      decl.Semantic.SemanticName = TGSI_SEMANTIC_GENERIC;
+      decl.Semantic.SemanticIndex = aactx->maxGeneric + 1;
+      decl.DeclarationRange.First = 
+      decl.DeclarationRange.Last = texInput;
+      ctx->emit_declaration(ctx, &decl);
+
+      /* declare new temp regs */
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_TEMPORARY;
+      decl.DeclarationRange.First = 
+      decl.DeclarationRange.Last = tmp0;
+      ctx->emit_declaration(ctx, &decl);
+
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_TEMPORARY;
+      decl.DeclarationRange.First = 
+      decl.DeclarationRange.Last = aactx->colorTemp;
+      ctx->emit_declaration(ctx, &decl);
+
+      aactx->firstInstruction = FALSE;
+
+
+      /*
+       * Emit code to compute fragment coverage, kill if outside point radius
+       *
+       * Temp reg0 usage:
+       *  t0.x = distance of fragment from center point
+       *  t0.y = boolean, is t0.x > 1.0, also misc temp usage
+       *  t0.z = temporary for computing 1/(1-k) value
+       *  t0.w = final coverage value
+       */
+
+      /* MUL t0.xy, tex, tex;  # compute x^2, y^2 */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_MUL;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_XY;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = texInput;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_INPUT;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = texInput;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* ADD t0.x, t0.x, t0.y;  # x^2 + y^2 */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_ADD;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_X;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = tmp0;
+      newInst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y;
+      ctx->emit_instruction(ctx, &newInst);
+
+#if NORMALIZE  /* OPTIONAL normalization of length */
+      /* RSQ t0.x, t0.x; */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_RSQ;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X;
+      newInst.Instruction.NumSrcRegs = 1;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* RCP t0.x, t0.x; */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_RCP;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_X;
+      newInst.Instruction.NumSrcRegs = 1;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+      ctx->emit_instruction(ctx, &newInst);
+#endif
+
+      /* SGT t0.y, t0.xxxx, t0.wwww;  # bool b = d > 1 (NOTE t0.w == 1) */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_SGT;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_Y;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_INPUT;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = texInput;
+      newInst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_W;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* KIL -tmp0.yyyy;   # if -tmp0.y < 0, KILL */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_KIL;
+      newInst.Instruction.NumDstRegs = 0;
+      newInst.Instruction.NumSrcRegs = 1;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_Y;
+      newInst.FullSrcRegisters[0].SrcRegister.Negate = 1;
+      ctx->emit_instruction(ctx, &newInst);
+
+
+      /* compute coverage factor = (1-d)/(1-k) */
+
+      /* SUB t0.z, tex.w, tex.z;  # m = 1 - k */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_SUB;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_Z;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = texInput;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_W;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_INPUT;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = texInput;
+      newInst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Z;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* RCP t0.z, t0.z;  # t0.z = 1 / m */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_RCP;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_Z;
+      newInst.Instruction.NumSrcRegs = 1;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_Z;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* SUB t0.y, 1, t0.x;  # d = 1 - d */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_SUB;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_Y;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = texInput;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_W;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = tmp0;
+      newInst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* MUL t0.w, t0.y, t0.z;   # coverage = d * m */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_MUL;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_W;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_Y;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = tmp0;
+      newInst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_Z;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* SLE t0.y, t0.x, tex.z;  # bool b = distance <= k */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_SLE;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_Y;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_X;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_INPUT;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = texInput;
+      newInst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_Z;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* CMP t0.w, -t0.y, tex.w, t0.w;
+       *  # if -t0.y < 0 then
+       *       t0.w = 1
+       *    else
+       *       t0.w = t0.w
+       */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_CMP;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = tmp0;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_W;
+      newInst.Instruction.NumSrcRegs = 3;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = tmp0;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleX = TGSI_SWIZZLE_Y;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleY = TGSI_SWIZZLE_Y;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleZ = TGSI_SWIZZLE_Y;
+      newInst.FullSrcRegisters[0].SrcRegister.SwizzleW = TGSI_SWIZZLE_Y;
+      newInst.FullSrcRegisters[0].SrcRegister.Negate = 1;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_INPUT;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = texInput;
+      newInst.FullSrcRegisters[1].SrcRegister.SwizzleX = TGSI_SWIZZLE_W;
+      newInst.FullSrcRegisters[1].SrcRegister.SwizzleY = TGSI_SWIZZLE_W;
+      newInst.FullSrcRegisters[1].SrcRegister.SwizzleZ = TGSI_SWIZZLE_W;
+      newInst.FullSrcRegisters[1].SrcRegister.SwizzleW = TGSI_SWIZZLE_W;
+      newInst.FullSrcRegisters[2].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[2].SrcRegister.Index = tmp0;
+      newInst.FullSrcRegisters[2].SrcRegister.SwizzleX = TGSI_SWIZZLE_W;
+      newInst.FullSrcRegisters[2].SrcRegister.SwizzleY = TGSI_SWIZZLE_W;
+      newInst.FullSrcRegisters[2].SrcRegister.SwizzleZ = TGSI_SWIZZLE_W;
+      newInst.FullSrcRegisters[2].SrcRegister.SwizzleW = TGSI_SWIZZLE_W;
+      ctx->emit_instruction(ctx, &newInst);
+
+   }
+
+   if (inst->Instruction.Opcode == TGSI_OPCODE_END) {
+      /* add alpha modulation code at tail of program */
+
+      /* MOV result.color.xyz, colorTemp; */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_MOV;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_OUTPUT;
+      newInst.FullDstRegisters[0].DstRegister.Index = aactx->colorOutput;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_XYZ;
+      newInst.Instruction.NumSrcRegs = 1;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = aactx->colorTemp;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* MUL result.color.w, colorTemp, tmp0.w; */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_MUL;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_OUTPUT;
+      newInst.FullDstRegisters[0].DstRegister.Index = aactx->colorOutput;
+      newInst.FullDstRegisters[0].DstRegister.WriteMask = TGSI_WRITEMASK_W;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = aactx->colorTemp;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = aactx->tmp0;
+      ctx->emit_instruction(ctx, &newInst);
+   }
+   else {
+      /* Not an END instruction.
+       * Look for writes to result.color and replace with colorTemp reg.
+       */
+      uint i;
+
+      for (i = 0; i < inst->Instruction.NumDstRegs; i++) {
+         struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i];
+         if (dst->DstRegister.File == TGSI_FILE_OUTPUT &&
+             dst->DstRegister.Index == aactx->colorOutput) {
+            dst->DstRegister.File = TGSI_FILE_TEMPORARY;
+            dst->DstRegister.Index = aactx->colorTemp;
+         }
+      }
+   }
+
+   ctx->emit_instruction(ctx, inst);
+}
+
+
+/**
+ * Generate the frag shader we'll use for drawing AA points.
+ * This will be the user's shader plus some texture/modulate instructions.
+ */
+static boolean
+generate_aapoint_fs(struct aapoint_stage *aapoint)
+{
+   const struct pipe_shader_state *orig_fs = &aapoint->fs->state;
+   struct pipe_shader_state aapoint_fs;
+   struct aa_transform_context transform;
+
+#define MAX 1000
+
+   aapoint_fs = *orig_fs; /* copy to init */
+   aapoint_fs.tokens = MALLOC(sizeof(struct tgsi_token) * MAX);
+   if (aapoint_fs.tokens == NULL)
+      return FALSE;
+
+   memset(&transform, 0, sizeof(transform));
+   transform.colorOutput = -1;
+   transform.maxInput = -1;
+   transform.maxGeneric = -1;
+   transform.colorTemp = -1;
+   transform.tmp0 = -1;
+   transform.firstInstruction = TRUE;
+   transform.base.transform_instruction = aa_transform_inst;
+   transform.base.transform_declaration = aa_transform_decl;
+
+   tgsi_transform_shader(orig_fs->tokens,
+                         (struct tgsi_token *) aapoint_fs.tokens,
+                         MAX, &transform.base);
+
+#if 0 /* DEBUG */
+   printf("draw_aapoint, orig shader:\n");
+   tgsi_dump(orig_fs->tokens, 0);
+   printf("draw_aapoint, new shader:\n");
+   tgsi_dump(aapoint_fs.tokens, 0);
+#endif
+
+   aapoint->fs->aapoint_fs
+      = aapoint->driver_create_fs_state(aapoint->pipe, &aapoint_fs);
+   if (aapoint->fs->aapoint_fs == NULL)
+      return FALSE;
+
+   aapoint->fs->generic_attrib = transform.maxGeneric + 1;
+
+   return TRUE;
+}
+
+
+/**
+ * When we're about to draw our first AA point in a batch, this function is
+ * called to tell the driver to bind our modified fragment shader.
+ */
+static boolean
+bind_aapoint_fragment_shader(struct aapoint_stage *aapoint)
+{
+   struct draw_context *draw = aapoint->stage.draw;
+
+   if (!aapoint->fs->aapoint_fs &&
+       !generate_aapoint_fs(aapoint))
+      return FALSE;
+
+   draw->suspend_flushing = TRUE;
+   aapoint->driver_bind_fs_state(aapoint->pipe, aapoint->fs->aapoint_fs);
+   draw->suspend_flushing = FALSE;
+
+   return TRUE;
+}
+
+
+
+static INLINE struct aapoint_stage *
+aapoint_stage( struct draw_stage *stage )
+{
+   return (struct aapoint_stage *) stage;
+}
+
+
+
+
+/**
+ * Draw an AA point by drawing a quad.
+ */
+static void
+aapoint_point(struct draw_stage *stage, struct prim_header *header)
+{
+   const struct aapoint_stage *aapoint = aapoint_stage(stage);
+   struct prim_header tri;
+   struct vertex_header *v[4];
+   uint texPos = aapoint->tex_slot;
+   uint pos_slot = aapoint->pos_slot;
+   float radius, *pos, *tex;
+   uint i;
+   float k;
+
+   if (aapoint->psize_slot >= 0) {
+      radius = 0.5f * header->v[0]->data[aapoint->psize_slot][0];
+   }
+   else {
+      radius = aapoint->radius;
+   }
+
+   /*
+    * Note: the texcoords (generic attrib, really) we use are special:
+    * The S and T components simply vary from -1 to +1.
+    * The R component is k, below.
+    * The Q component is 1.0 and will used as a handy constant in the
+    * fragment shader.
+    */
+
+   /*
+    * k is the threshold distance from the point's center at which
+    * we begin alpha attenuation (the coverage value).
+    * Operating within a unit circle, we'll compute the fragment's
+    * distance 'd' from the center point using the texcoords.
+    * IF d > 1.0 THEN
+    *    KILL fragment
+    * ELSE IF d > k THEN
+    *    compute coverage in [0,1] proportional to d in [k, 1].
+    * ELSE
+    *    coverage = 1.0;  // full coverage
+    * ENDIF
+    *
+    * Note: the ELSEIF and ELSE clauses are actually implemented with CMP to
+    * avoid using IF/ELSE/ENDIF TGSI opcodes.
+    */
+
+#if !NORMALIZE
+   k = 1.0f / radius;
+   k = 1.0f - 2.0f * k + k * k;
+#else
+   k = 1.0f - 1.0f / radius;
+#endif
+
+   /* allocate/dup new verts */
+   for (i = 0; i < 4; i++) {
+      v[i] = dup_vert(stage, header->v[0], i);
+   }
+
+   /* new verts */
+   pos = v[0]->data[pos_slot];
+   pos[0] -= radius;
+   pos[1] -= radius;
+
+   pos = v[1]->data[pos_slot];
+   pos[0] += radius;
+   pos[1] -= radius;
+
+   pos = v[2]->data[pos_slot];
+   pos[0] += radius;
+   pos[1] += radius;
+
+   pos = v[3]->data[pos_slot];
+   pos[0] -= radius;
+   pos[1] += radius;
+
+   /* new texcoords */
+   tex = v[0]->data[texPos];
+   ASSIGN_4V(tex, -1, -1, k, 1);
+
+   tex = v[1]->data[texPos];
+   ASSIGN_4V(tex,  1, -1, k, 1);
+
+   tex = v[2]->data[texPos];
+   ASSIGN_4V(tex,  1,  1, k, 1);
+
+   tex = v[3]->data[texPos];
+   ASSIGN_4V(tex, -1,  1, k, 1);
+
+   /* emit 2 tris for the quad strip */
+   tri.v[0] = v[0];
+   tri.v[1] = v[1];
+   tri.v[2] = v[2];
+   stage->next->tri( stage->next, &tri );
+
+   tri.v[0] = v[0];
+   tri.v[1] = v[2];
+   tri.v[2] = v[3];
+   stage->next->tri( stage->next, &tri );
+}
+
+
+static void
+aapoint_first_point(struct draw_stage *stage, struct prim_header *header)
+{
+   auto struct aapoint_stage *aapoint = aapoint_stage(stage);
+   struct draw_context *draw = stage->draw;
+
+   assert(draw->rasterizer->point_smooth);
+
+   if (draw->rasterizer->point_size <= 2.0)
+      aapoint->radius = 1.0;
+   else
+      aapoint->radius = 0.5f * draw->rasterizer->point_size;
+
+   /*
+    * Bind (generate) our fragprog.
+    */
+   bind_aapoint_fragment_shader(aapoint);
+
+   /* update vertex attrib info */
+   aapoint->tex_slot = draw->vs.num_vs_outputs;
+   assert(aapoint->tex_slot > 0); /* output[0] is vertex pos */
+
+   aapoint->pos_slot = draw->vs.position_output;
+
+   draw->extra_vp_outputs.semantic_name = TGSI_SEMANTIC_GENERIC;
+   draw->extra_vp_outputs.semantic_index = aapoint->fs->generic_attrib;
+   draw->extra_vp_outputs.slot = aapoint->tex_slot;
+
+   /* find psize slot in post-transform vertex */
+   aapoint->psize_slot = -1;
+   if (draw->rasterizer->point_size_per_vertex) {
+      /* find PSIZ vertex output */
+      const struct draw_vertex_shader *vs = draw->vs.vertex_shader;
+      uint i;
+      for (i = 0; i < vs->info.num_outputs; i++) {
+         if (vs->info.output_semantic_name[i] == TGSI_SEMANTIC_PSIZE) {
+            aapoint->psize_slot = i;
+            break;
+         }
+      }
+   }
+
+   /* now really draw first point */
+   stage->point = aapoint_point;
+   stage->point(stage, header);
+}
+
+
+static void
+aapoint_flush(struct draw_stage *stage, unsigned flags)
+{
+   struct draw_context *draw = stage->draw;
+   struct aapoint_stage *aapoint = aapoint_stage(stage);
+   struct pipe_context *pipe = aapoint->pipe;
+
+   stage->point = aapoint_first_point;
+   stage->next->flush( stage->next, flags );
+
+   /* restore original frag shader */
+   draw->suspend_flushing = TRUE;
+   aapoint->driver_bind_fs_state(pipe, aapoint->fs->driver_fs);
+   draw->suspend_flushing = FALSE;
+
+   draw->extra_vp_outputs.slot = 0;
+}
+
+
+static void
+aapoint_reset_stipple_counter(struct draw_stage *stage)
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void
+aapoint_destroy(struct draw_stage *stage)
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+static struct aapoint_stage *
+draw_aapoint_stage(struct draw_context *draw)
+{
+   struct aapoint_stage *aapoint = CALLOC_STRUCT(aapoint_stage);
+   if (aapoint == NULL)
+      goto fail;
+
+   if (!draw_alloc_temp_verts( &aapoint->stage, 4 ))
+      goto fail;
+
+   aapoint->stage.draw = draw;
+   aapoint->stage.next = NULL;
+   aapoint->stage.point = aapoint_first_point;
+   aapoint->stage.line = draw_pipe_passthrough_line;
+   aapoint->stage.tri = draw_pipe_passthrough_tri;
+   aapoint->stage.flush = aapoint_flush;
+   aapoint->stage.reset_stipple_counter = aapoint_reset_stipple_counter;
+   aapoint->stage.destroy = aapoint_destroy;
+
+   return aapoint;
+
+ fail:
+   if (aapoint)
+      aapoint_destroy(&aapoint->stage);
+
+   return NULL;
+
+}
+
+
+static struct aapoint_stage *
+aapoint_stage_from_pipe(struct pipe_context *pipe)
+{
+   struct draw_context *draw = (struct draw_context *) pipe->draw;
+   return aapoint_stage(draw->pipeline.aapoint);
+}
+
+
+/**
+ * This function overrides the driver's create_fs_state() function and
+ * will typically be called by the state tracker.
+ */
+static void *
+aapoint_create_fs_state(struct pipe_context *pipe,
+                       const struct pipe_shader_state *fs)
+{
+   struct aapoint_stage *aapoint = aapoint_stage_from_pipe(pipe);
+   struct aapoint_fragment_shader *aafs = CALLOC_STRUCT(aapoint_fragment_shader);
+   if (aafs == NULL) 
+      return NULL;
+
+   aafs->state = *fs;
+
+   /* pass-through */
+   aafs->driver_fs = aapoint->driver_create_fs_state(aapoint->pipe, fs);
+
+   return aafs;
+}
+
+
+static void
+aapoint_bind_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct aapoint_stage *aapoint = aapoint_stage_from_pipe(pipe);
+   struct aapoint_fragment_shader *aafs = (struct aapoint_fragment_shader *) fs;
+   /* save current */
+   aapoint->fs = aafs;
+   /* pass-through */
+   aapoint->driver_bind_fs_state(aapoint->pipe,
+                                 (aafs ? aafs->driver_fs : NULL));
+}
+
+
+static void
+aapoint_delete_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct aapoint_stage *aapoint = aapoint_stage_from_pipe(pipe);
+   struct aapoint_fragment_shader *aafs = (struct aapoint_fragment_shader *) fs;
+   /* pass-through */
+   aapoint->driver_delete_fs_state(aapoint->pipe, aafs->driver_fs);
+   FREE(aafs);
+}
+
+
+/**
+ * Called by drivers that want to install this AA point prim stage
+ * into the draw module's pipeline.  This will not be used if the
+ * hardware has native support for AA points.
+ */
+boolean
+draw_install_aapoint_stage(struct draw_context *draw,
+                           struct pipe_context *pipe)
+{
+   struct aapoint_stage *aapoint;
+
+   pipe->draw = (void *) draw;
+
+   /*
+    * Create / install AA point drawing / prim stage
+    */
+   aapoint = draw_aapoint_stage( draw );
+   if (aapoint == NULL)
+      goto fail;
+
+   aapoint->pipe = pipe;
+
+   /* save original driver functions */
+   aapoint->driver_create_fs_state = pipe->create_fs_state;
+   aapoint->driver_bind_fs_state = pipe->bind_fs_state;
+   aapoint->driver_delete_fs_state = pipe->delete_fs_state;
+
+   /* override the driver's functions */
+   pipe->create_fs_state = aapoint_create_fs_state;
+   pipe->bind_fs_state = aapoint_bind_fs_state;
+   pipe->delete_fs_state = aapoint_delete_fs_state;
+
+   draw->pipeline.aapoint = &aapoint->stage;
+
+   return TRUE;
+
+ fail:
+   if (aapoint)
+      aapoint->stage.destroy( &aapoint->stage );
+
+   return FALSE;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pipe_clip.c b/src/gallium/auxiliary/draw/draw_pipe_clip.c
new file mode 100644
index 00000000000..3265dcd154a
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pipe_clip.c
@@ -0,0 +1,515 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * \brief  Clipping stage
+ *
+ * \author  Keith Whitwell <[email protected]>
+ */
+
+
+#include "util/u_memory.h"
+#include "util/u_math.h"
+
+#include "pipe/p_shader_tokens.h"
+
+#include "draw_vs.h"
+#include "draw_pipe.h"
+
+
+#ifndef IS_NEGATIVE
+#define IS_NEGATIVE(X) ((X) < 0.0)
+#endif
+
+#ifndef DIFFERENT_SIGNS
+#define DIFFERENT_SIGNS(x, y) ((x) * (y) <= 0.0F && (x) - (y) != 0.0F)
+#endif
+
+#ifndef MAX_CLIPPED_VERTICES
+#define MAX_CLIPPED_VERTICES ((2 * (6 + PIPE_MAX_CLIP_PLANES))+1)
+#endif
+
+
+
+struct clipper {
+   struct draw_stage stage;      /**< base class */
+
+   /* Basically duplicate some of the flatshading logic here:
+    */
+   boolean flat;
+   uint num_color_attribs;
+   uint color_attribs[4];  /* front/back primary/secondary colors */
+
+   float (*plane)[4];
+};
+
+
+/* This is a bit confusing:
+ */
+static INLINE struct clipper *clipper_stage( struct draw_stage *stage )
+{
+   return (struct clipper *)stage;
+}
+
+
+#define LINTERP(T, OUT, IN) ((OUT) + (T) * ((IN) - (OUT)))
+
+
+/* All attributes are float[4], so this is easy:
+ */
+static void interp_attr( float *fdst,
+			 float t,
+			 const float *fin,
+			 const float *fout )
+{  
+   fdst[0] = LINTERP( t, fout[0], fin[0] );
+   fdst[1] = LINTERP( t, fout[1], fin[1] );
+   fdst[2] = LINTERP( t, fout[2], fin[2] );
+   fdst[3] = LINTERP( t, fout[3], fin[3] );
+}
+
+static void copy_colors( struct draw_stage *stage,
+			 struct vertex_header *dst,
+			 const struct vertex_header *src )
+{
+   const struct clipper *clipper = clipper_stage(stage);
+   uint i;
+   for (i = 0; i < clipper->num_color_attribs; i++) {
+      const uint attr = clipper->color_attribs[i];
+      COPY_4FV(dst->data[attr], src->data[attr]);
+   }
+}
+
+
+
+/* Interpolate between two vertices to produce a third.  
+ */
+static void interp( const struct clipper *clip,
+		    struct vertex_header *dst,
+		    float t,
+		    const struct vertex_header *out, 
+		    const struct vertex_header *in )
+{
+   const unsigned nr_attrs = clip->stage.draw->vs.num_vs_outputs;
+   const unsigned pos_attr = clip->stage.draw->vs.position_output;
+   unsigned j;
+
+   /* Vertex header.
+    */
+   {
+      dst->clipmask = 0;
+      dst->edgeflag = 0;        /* will get overwritten later */
+      dst->pad = 0;
+      dst->vertex_id = UNDEFINED_VERTEX_ID;
+   }
+
+   /* Clip coordinates:  interpolate normally
+    */
+   {
+      interp_attr(dst->clip, t, in->clip, out->clip);
+   }
+
+   /* Do the projective divide and insert window coordinates:
+    */
+   {
+      const float *pos = dst->clip;
+      const float *scale = clip->stage.draw->viewport.scale;
+      const float *trans = clip->stage.draw->viewport.translate;
+      const float oow = 1.0f / pos[3];
+
+      dst->data[pos_attr][0] = pos[0] * oow * scale[0] + trans[0];
+      dst->data[pos_attr][1] = pos[1] * oow * scale[1] + trans[1];
+      dst->data[pos_attr][2] = pos[2] * oow * scale[2] + trans[2];
+      dst->data[pos_attr][3] = oow;
+   }
+
+   /* Other attributes
+    */
+   for (j = 0; j < nr_attrs; j++) {
+      if (j != pos_attr)
+         interp_attr(dst->data[j], t, in->data[j], out->data[j]);
+   }
+}
+
+
+static void emit_poly( struct draw_stage *stage,
+		       struct vertex_header **inlist,
+		       unsigned n,
+		       const struct prim_header *origPrim)
+{
+   struct prim_header header;
+   unsigned i;
+
+   const ushort edge_first  = DRAW_PIPE_EDGE_FLAG_2;
+   const ushort edge_middle = DRAW_PIPE_EDGE_FLAG_0;
+   const ushort edge_last   = DRAW_PIPE_EDGE_FLAG_1;
+
+   /* later stages may need the determinant, but only the sign matters */
+   header.det = origPrim->det;
+   header.flags = DRAW_PIPE_RESET_STIPPLE | edge_first | edge_middle;
+   header.pad = 0;
+
+   for (i = 2; i < n; i++, header.flags = edge_middle) {
+      header.v[0] = inlist[i-1];
+      header.v[1] = inlist[i];
+      header.v[2] = inlist[0];	/* keep in v[2] for flatshading */
+
+      if (i == n-1)
+        header.flags |= edge_last;
+
+      if (0) {
+         const struct draw_vertex_shader *vs = stage->draw->vs.vertex_shader;
+         uint j, k;
+         debug_printf("Clipped tri:\n");
+         for (j = 0; j < 3; j++) {
+            for (k = 0; k < vs->info.num_outputs; k++) {
+               debug_printf("  Vert %d: Attr %d:  %f %f %f %f\n", j, k,
+                            header.v[j]->data[k][0],
+                            header.v[j]->data[k][1],
+                            header.v[j]->data[k][2],
+                            header.v[j]->data[k][3]);
+            }
+         }
+      }
+
+      stage->next->tri( stage->next, &header );
+   }
+}
+
+static INLINE float
+dot4(const float *a, const float *b)
+{
+   return (a[0]*b[0] +
+           a[1]*b[1] +
+           a[2]*b[2] +
+           a[3]*b[3]);
+}
+
+
+/* Clip a triangle against the viewport and user clip planes.
+ */
+static void
+do_clip_tri( struct draw_stage *stage, 
+	     struct prim_header *header,
+	     unsigned clipmask )
+{
+   struct clipper *clipper = clipper_stage( stage );
+   struct vertex_header *a[MAX_CLIPPED_VERTICES];
+   struct vertex_header *b[MAX_CLIPPED_VERTICES];
+   struct vertex_header **inlist = a;
+   struct vertex_header **outlist = b;
+   unsigned tmpnr = 0;
+   unsigned n = 3;
+   unsigned i;
+
+   inlist[0] = header->v[0];
+   inlist[1] = header->v[1];
+   inlist[2] = header->v[2];
+
+   while (clipmask && n >= 3) {
+      const unsigned plane_idx = ffs(clipmask)-1;
+      const float *plane = clipper->plane[plane_idx];
+      struct vertex_header *vert_prev = inlist[0];
+      float dp_prev = dot4( vert_prev->clip, plane );
+      unsigned outcount = 0;
+
+      clipmask &= ~(1<<plane_idx);
+
+      inlist[n] = inlist[0]; /* prevent rotation of vertices */
+
+      for (i = 1; i <= n; i++) {
+	 struct vertex_header *vert = inlist[i];
+
+	 float dp = dot4( vert->clip, plane );
+
+	 if (!IS_NEGATIVE(dp_prev)) {
+	    outlist[outcount++] = vert_prev;
+	 }
+
+	 if (DIFFERENT_SIGNS(dp, dp_prev)) {
+	    struct vertex_header *new_vert = clipper->stage.tmp[tmpnr++];
+	    outlist[outcount++] = new_vert;
+
+	    if (IS_NEGATIVE(dp)) {
+	       /* Going out of bounds.  Avoid division by zero as we
+		* know dp != dp_prev from DIFFERENT_SIGNS, above.
+		*/
+	       float t = dp / (dp - dp_prev);
+	       interp( clipper, new_vert, t, vert, vert_prev );
+	       
+	       /* Force edgeflag true in this case:
+		*/
+	       new_vert->edgeflag = 1;
+	    } else {
+	       /* Coming back in.
+		*/
+	       float t = dp_prev / (dp_prev - dp);
+	       interp( clipper, new_vert, t, vert_prev, vert );
+
+	       /* Copy starting vert's edgeflag:
+		*/
+	       new_vert->edgeflag = vert_prev->edgeflag;
+	    }
+	 }
+
+	 vert_prev = vert;
+	 dp_prev = dp;
+      }
+
+      {
+	 struct vertex_header **tmp = inlist;
+	 inlist = outlist;
+	 outlist = tmp;
+	 n = outcount;
+      }
+   }
+
+   /* If flat-shading, copy color to new provoking vertex.
+    */
+   if (clipper->flat && inlist[0] != header->v[2]) {
+      if (1) {
+	 inlist[0] = dup_vert(stage, inlist[0], tmpnr++);
+      }
+
+      copy_colors(stage, inlist[0], header->v[2]);
+   }
+
+
+
+   /* Emit the polygon as triangles to the setup stage:
+    */
+   if (n >= 3)
+      emit_poly( stage, inlist, n, header );
+}
+
+
+/* Clip a line against the viewport and user clip planes.
+ */
+static void
+do_clip_line( struct draw_stage *stage,
+	      struct prim_header *header,
+	      unsigned clipmask )
+{
+   const struct clipper *clipper = clipper_stage( stage );
+   struct vertex_header *v0 = header->v[0];
+   struct vertex_header *v1 = header->v[1];
+   const float *pos0 = v0->clip;
+   const float *pos1 = v1->clip;
+   float t0 = 0.0F;
+   float t1 = 0.0F;
+   struct prim_header newprim;
+
+   while (clipmask) {
+      const unsigned plane_idx = ffs(clipmask)-1;
+      const float *plane = clipper->plane[plane_idx];
+      const float dp0 = dot4( pos0, plane );
+      const float dp1 = dot4( pos1, plane );
+
+      if (dp1 < 0.0F) {
+	 float t = dp1 / (dp1 - dp0);
+         t1 = MAX2(t1, t);
+      } 
+
+      if (dp0 < 0.0F) {
+	 float t = dp0 / (dp0 - dp1);
+         t0 = MAX2(t0, t);
+      }
+
+      if (t0 + t1 >= 1.0F)
+	 return; /* discard */
+
+      clipmask &= ~(1 << plane_idx);  /* turn off this plane's bit */
+   }
+
+   if (v0->clipmask) {
+      interp( clipper, stage->tmp[0], t0, v0, v1 );
+
+      if (clipper->flat)
+	 copy_colors(stage, stage->tmp[0], v0);
+
+      newprim.v[0] = stage->tmp[0];
+   }
+   else {
+      newprim.v[0] = v0;
+   }
+
+   if (v1->clipmask) {
+      interp( clipper, stage->tmp[1], t1, v1, v0 );
+      newprim.v[1] = stage->tmp[1];
+   }
+   else {
+      newprim.v[1] = v1;
+   }
+
+   stage->next->line( stage->next, &newprim );
+}
+
+
+static void
+clip_point( struct draw_stage *stage, 
+	    struct prim_header *header )
+{
+   if (header->v[0]->clipmask == 0) 
+      stage->next->point( stage->next, header );
+}
+
+
+static void
+clip_line( struct draw_stage *stage,
+	   struct prim_header *header )
+{
+   unsigned clipmask = (header->v[0]->clipmask | 
+                        header->v[1]->clipmask);
+   
+   if (clipmask == 0) {
+      /* no clipping needed */
+      stage->next->line( stage->next, header );
+   }
+   else if ((header->v[0]->clipmask &
+             header->v[1]->clipmask) == 0) {
+      do_clip_line(stage, header, clipmask);
+   }
+   /* else, totally clipped */
+}
+
+
+static void
+clip_tri( struct draw_stage *stage,
+	  struct prim_header *header )
+{
+   unsigned clipmask = (header->v[0]->clipmask | 
+                        header->v[1]->clipmask | 
+                        header->v[2]->clipmask);
+   
+   if (clipmask == 0) {
+      /* no clipping needed */
+      stage->next->tri( stage->next, header );
+   }
+   else if ((header->v[0]->clipmask & 
+             header->v[1]->clipmask & 
+             header->v[2]->clipmask) == 0) {
+      do_clip_tri(stage, header, clipmask);
+   }
+}
+
+/* Update state.  Could further delay this until we hit the first
+ * primitive that really requires clipping.
+ */
+static void 
+clip_init_state( struct draw_stage *stage )
+{
+   struct clipper *clipper = clipper_stage( stage );
+
+   clipper->flat = stage->draw->rasterizer->flatshade ? TRUE : FALSE;
+
+   if (clipper->flat) {
+      const struct draw_vertex_shader *vs = stage->draw->vs.vertex_shader;
+      uint i;
+
+      clipper->num_color_attribs = 0;
+      for (i = 0; i < vs->info.num_outputs; i++) {
+	 if (vs->info.output_semantic_name[i] == TGSI_SEMANTIC_COLOR ||
+	     vs->info.output_semantic_name[i] == TGSI_SEMANTIC_BCOLOR) {
+	    clipper->color_attribs[clipper->num_color_attribs++] = i;
+	 }
+      }
+   }
+   
+   stage->tri = clip_tri;
+   stage->line = clip_line;
+}
+
+
+
+static void clip_first_tri( struct draw_stage *stage,
+			    struct prim_header *header )
+{
+   clip_init_state( stage );
+   stage->tri( stage, header );
+}
+
+static void clip_first_line( struct draw_stage *stage,
+			     struct prim_header *header )
+{
+   clip_init_state( stage );
+   stage->line( stage, header );
+}
+
+
+static void clip_flush( struct draw_stage *stage, 
+			     unsigned flags )
+{
+   stage->tri = clip_first_tri;
+   stage->line = clip_first_line;
+   stage->next->flush( stage->next, flags );
+}
+
+
+static void clip_reset_stipple_counter( struct draw_stage *stage )
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void clip_destroy( struct draw_stage *stage )
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+/**
+ * Allocate a new clipper stage.
+ * \return pointer to new stage object
+ */
+struct draw_stage *draw_clip_stage( struct draw_context *draw )
+{
+   struct clipper *clipper = CALLOC_STRUCT(clipper);
+   if (clipper == NULL)
+      goto fail;
+
+   if (!draw_alloc_temp_verts( &clipper->stage, MAX_CLIPPED_VERTICES+1 ))
+      goto fail;
+
+   clipper->stage.draw = draw;
+   clipper->stage.point = clip_point;
+   clipper->stage.line = clip_first_line;
+   clipper->stage.tri = clip_first_tri;
+   clipper->stage.flush = clip_flush;
+   clipper->stage.reset_stipple_counter = clip_reset_stipple_counter;
+   clipper->stage.destroy = clip_destroy;
+
+   clipper->plane = draw->plane;
+
+   return &clipper->stage;
+
+ fail:
+   if (clipper)
+      clipper->stage.destroy( &clipper->stage );
+
+   return NULL;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pipe_cull.c b/src/gallium/auxiliary/draw/draw_pipe_cull.c
new file mode 100644
index 00000000000..053be5f050d
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pipe_cull.c
@@ -0,0 +1,147 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * \brief  Drawing stage for polygon culling
+ */
+
+/* Authors:  Keith Whitwell <[email protected]>
+ */
+
+
+#include "util/u_memory.h"
+#include "pipe/p_defines.h"
+#include "draw_pipe.h"
+
+
+struct cull_stage {
+   struct draw_stage stage;
+   unsigned winding;  /**< which winding(s) to cull (one of PIPE_WINDING_x) */
+};
+
+
+static INLINE struct cull_stage *cull_stage( struct draw_stage *stage )
+{
+   return (struct cull_stage *)stage;
+}
+
+
+
+
+static void cull_tri( struct draw_stage *stage,
+		      struct prim_header *header )
+{
+   const unsigned pos = stage->draw->vs.position_output;
+
+   /* Window coords: */
+   const float *v0 = header->v[0]->data[pos];
+   const float *v1 = header->v[1]->data[pos];
+   const float *v2 = header->v[2]->data[pos];
+
+   /* edge vectors e = v0 - v2, f = v1 - v2 */
+   const float ex = v0[0] - v2[0];
+   const float ey = v0[1] - v2[1];
+   const float fx = v1[0] - v2[0];
+   const float fy = v1[1] - v2[1];
+   
+   /* det = cross(e,f).z */
+   header->det = ex * fy - ey * fx;
+
+   if (header->det != 0) {
+      /* if (det < 0 then Z points toward camera and triangle is 
+       * counter-clockwise winding.
+       */
+      unsigned winding = (header->det < 0) ? PIPE_WINDING_CCW : PIPE_WINDING_CW;
+
+      if ((winding & cull_stage(stage)->winding) == 0) {
+         /* triangle is not culled, pass to next stage */
+	 stage->next->tri( stage->next, header );
+      }
+   }
+}
+
+static void cull_first_tri( struct draw_stage *stage, 
+			    struct prim_header *header )
+{
+   struct cull_stage *cull = cull_stage(stage);
+
+   cull->winding = stage->draw->rasterizer->cull_mode;
+
+   stage->tri = cull_tri;
+   stage->tri( stage, header );
+}
+
+
+
+static void cull_flush( struct draw_stage *stage, unsigned flags )
+{
+   stage->tri = cull_first_tri;
+   stage->next->flush( stage->next, flags );
+}
+
+static void cull_reset_stipple_counter( struct draw_stage *stage )
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void cull_destroy( struct draw_stage *stage )
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+/**
+ * Create a new polygon culling stage.
+ */
+struct draw_stage *draw_cull_stage( struct draw_context *draw )
+{
+   struct cull_stage *cull = CALLOC_STRUCT(cull_stage);
+   if (cull == NULL)
+      goto fail;
+
+   if (!draw_alloc_temp_verts( &cull->stage, 0 ))
+      goto fail;
+
+   cull->stage.draw = draw;
+   cull->stage.next = NULL;
+   cull->stage.point = draw_pipe_passthrough_point;
+   cull->stage.line = draw_pipe_passthrough_line;
+   cull->stage.tri = cull_first_tri;
+   cull->stage.flush = cull_flush;
+   cull->stage.reset_stipple_counter = cull_reset_stipple_counter;
+   cull->stage.destroy = cull_destroy;
+
+   return &cull->stage;
+
+ fail:
+   if (cull)
+      cull->stage.destroy( &cull->stage );
+
+   return NULL;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pipe_flatshade.c b/src/gallium/auxiliary/draw/draw_pipe_flatshade.c
new file mode 100644
index 00000000000..43d1fecc4dd
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pipe_flatshade.c
@@ -0,0 +1,281 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <[email protected]>
+ */
+
+#include "util/u_math.h"
+#include "util/u_memory.h"
+
+#include "pipe/p_shader_tokens.h"
+#include "draw_vs.h"
+#include "draw_pipe.h"
+
+
+/** subclass of draw_stage */
+struct flat_stage
+{
+   struct draw_stage stage;
+
+   uint num_color_attribs;
+   uint color_attribs[2];  /* front/back primary colors */
+
+   uint num_spec_attribs;
+   uint spec_attribs[2];  /* front/back secondary colors */
+};
+
+#define COPY_3FV( DST, SRC )         \
+do {                                \
+   (DST)[0] = (SRC)[0];             \
+   (DST)[1] = (SRC)[1];             \
+   (DST)[2] = (SRC)[2];             \
+} while (0)
+
+
+static INLINE struct flat_stage *
+flat_stage(struct draw_stage *stage)
+{
+   return (struct flat_stage *) stage;
+}
+
+
+/** Copy all the color attributes from 'src' vertex to 'dst' vertex */
+static INLINE void copy_colors( struct draw_stage *stage,
+                                struct vertex_header *dst,
+                                const struct vertex_header *src )
+{
+   const struct flat_stage *flat = flat_stage(stage);
+   uint i;
+
+   for (i = 0; i < flat->num_color_attribs; i++) {
+      const uint attr = flat->color_attribs[i];
+      COPY_4FV(dst->data[attr], src->data[attr]);
+   }
+
+   for (i = 0; i < flat->num_spec_attribs; i++) {
+      const uint attr = flat->spec_attribs[i];
+      COPY_3FV(dst->data[attr], src->data[attr]);
+   }
+}
+
+
+/** Copy all the color attributes from src vertex to dst0 & dst1 vertices */
+static INLINE void copy_colors2( struct draw_stage *stage,
+                                 struct vertex_header *dst0,
+                                 struct vertex_header *dst1,
+                                 const struct vertex_header *src )
+{
+   const struct flat_stage *flat = flat_stage(stage);
+   uint i;
+   for (i = 0; i < flat->num_color_attribs; i++) {
+      const uint attr = flat->color_attribs[i];
+      COPY_4FV(dst0->data[attr], src->data[attr]);
+      COPY_4FV(dst1->data[attr], src->data[attr]);
+   }
+
+   for (i = 0; i < flat->num_spec_attribs; i++) {
+      const uint attr = flat->spec_attribs[i];
+      COPY_3FV(dst0->data[attr], src->data[attr]);
+      COPY_3FV(dst1->data[attr], src->data[attr]);
+   }
+}
+
+
+/**
+ * Flatshade tri.  Required for clipping and when unfilled tris are
+ * active, otherwise handled by hardware.
+ */
+static void flatshade_tri_0( struct draw_stage *stage,
+                             struct prim_header *header )
+{
+   struct prim_header tmp;
+
+   tmp.det = header->det;
+   tmp.flags = header->flags;
+   tmp.pad = header->pad;
+   tmp.v[0] = header->v[0];
+   tmp.v[1] = dup_vert(stage, header->v[1], 0);
+   tmp.v[2] = dup_vert(stage, header->v[2], 1);
+
+   copy_colors2(stage, tmp.v[1], tmp.v[2], tmp.v[0]);
+   
+   stage->next->tri( stage->next, &tmp );
+}
+
+
+static void flatshade_tri_2( struct draw_stage *stage,
+                             struct prim_header *header )
+{
+   struct prim_header tmp;
+
+   tmp.det = header->det;
+   tmp.flags = header->flags;
+   tmp.pad = header->pad;
+   tmp.v[0] = dup_vert(stage, header->v[0], 0);
+   tmp.v[1] = dup_vert(stage, header->v[1], 1);
+   tmp.v[2] = header->v[2];
+
+   copy_colors2(stage, tmp.v[0], tmp.v[1], tmp.v[2]);
+   
+   stage->next->tri( stage->next, &tmp );
+}
+
+
+
+
+
+/**
+ * Flatshade line.  Required for clipping.
+ */
+static void flatshade_line_0( struct draw_stage *stage,
+                              struct prim_header *header )
+{
+   struct prim_header tmp;
+
+   tmp.v[0] = header->v[0];
+   tmp.v[1] = dup_vert(stage, header->v[1], 0);
+
+   copy_colors(stage, tmp.v[1], tmp.v[0]);
+   
+   stage->next->line( stage->next, &tmp );
+}
+
+static void flatshade_line_1( struct draw_stage *stage,
+                              struct prim_header *header )
+{
+   struct prim_header tmp;
+
+   tmp.v[0] = dup_vert(stage, header->v[0], 0);
+   tmp.v[1] = header->v[1];
+
+   copy_colors(stage, tmp.v[0], tmp.v[1]);
+   
+   stage->next->line( stage->next, &tmp );
+}
+
+
+
+
+static void flatshade_init_state( struct draw_stage *stage )
+{
+   struct flat_stage *flat = flat_stage(stage);
+   const struct draw_vertex_shader *vs = stage->draw->vs.vertex_shader;
+   uint i;
+
+   /* Find which vertex shader outputs are colors, make a list */
+   flat->num_color_attribs = 0;
+   flat->num_spec_attribs = 0;
+   for (i = 0; i < vs->info.num_outputs; i++) {
+      if (vs->info.output_semantic_name[i] == TGSI_SEMANTIC_COLOR ||
+          vs->info.output_semantic_name[i] == TGSI_SEMANTIC_BCOLOR) {
+         if (vs->info.output_semantic_index[i] == 0)
+            flat->color_attribs[flat->num_color_attribs++] = i;
+         else
+            flat->spec_attribs[flat->num_spec_attribs++] = i;
+      }
+   }
+
+   /* Choose flatshade routine according to provoking vertex:
+    */
+   if (stage->draw->rasterizer->flatshade_first) {
+      stage->line = flatshade_line_0;
+      stage->tri = flatshade_tri_0;
+   }
+   else {
+      stage->line = flatshade_line_1;
+      stage->tri = flatshade_tri_2;
+   }
+}
+
+static void flatshade_first_tri( struct draw_stage *stage,
+				 struct prim_header *header )
+{
+   flatshade_init_state( stage );
+   stage->tri( stage, header );
+}
+
+static void flatshade_first_line( struct draw_stage *stage,
+				  struct prim_header *header )
+{
+   flatshade_init_state( stage );
+   stage->line( stage, header );
+}
+
+
+static void flatshade_flush( struct draw_stage *stage, 
+			     unsigned flags )
+{
+   stage->tri = flatshade_first_tri;
+   stage->line = flatshade_first_line;
+   stage->next->flush( stage->next, flags );
+}
+
+
+static void flatshade_reset_stipple_counter( struct draw_stage *stage )
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void flatshade_destroy( struct draw_stage *stage )
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+/**
+ * Create flatshading drawing stage.
+ */
+struct draw_stage *draw_flatshade_stage( struct draw_context *draw )
+{
+   struct flat_stage *flatshade = CALLOC_STRUCT(flat_stage);
+   if (flatshade == NULL)
+      goto fail;
+
+   if (!draw_alloc_temp_verts( &flatshade->stage, 2 ))
+      goto fail;
+
+   flatshade->stage.draw = draw;
+   flatshade->stage.next = NULL;
+   flatshade->stage.point = draw_pipe_passthrough_point;
+   flatshade->stage.line = flatshade_first_line;
+   flatshade->stage.tri = flatshade_first_tri;
+   flatshade->stage.flush = flatshade_flush;
+   flatshade->stage.reset_stipple_counter = flatshade_reset_stipple_counter;
+   flatshade->stage.destroy = flatshade_destroy;
+
+   return &flatshade->stage;
+
+ fail:
+   if (flatshade)
+      flatshade->stage.destroy( &flatshade->stage );
+
+   return NULL;
+}
+
+
diff --git a/src/gallium/auxiliary/draw/draw_pipe_offset.c b/src/gallium/auxiliary/draw/draw_pipe_offset.c
new file mode 100644
index 00000000000..62c31532d01
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pipe_offset.c
@@ -0,0 +1,184 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * \brief  polygon offset state
+ *
+ * \author  Keith Whitwell <[email protected]>
+ * \author  Brian Paul
+ */
+
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "draw_pipe.h"
+
+
+
+struct offset_stage {
+   struct draw_stage stage;
+
+   float scale;
+   float units;
+};
+
+
+
+static INLINE struct offset_stage *offset_stage( struct draw_stage *stage )
+{
+   return (struct offset_stage *) stage;
+}
+
+
+
+
+
+/**
+ * Offset tri Z.  Some hardware can handle this, but not usually when
+ * doing unfilled rendering.
+ */
+static void do_offset_tri( struct draw_stage *stage,
+			   struct prim_header *header )
+{
+   const unsigned pos = stage->draw->vs.position_output;
+   struct offset_stage *offset = offset_stage(stage);   
+   float inv_det = 1.0f / header->det;
+
+   /* Window coords:
+    */
+   float *v0 = header->v[0]->data[pos];
+   float *v1 = header->v[1]->data[pos];
+   float *v2 = header->v[2]->data[pos];
+
+   /* edge vectors e = v0 - v2, f = v1 - v2 */
+   float ex = v0[0] - v2[0];
+   float ey = v0[1] - v2[1];
+   float ez = v0[2] - v2[2];
+   float fx = v1[0] - v2[0];
+   float fy = v1[1] - v2[1];
+   float fz = v1[2] - v2[2];
+
+   /* (a,b) = cross(e,f).xy */
+   float a = ey*fz - ez*fy;
+   float b = ez*fx - ex*fz;
+
+   float dzdx = fabsf(a * inv_det);
+   float dzdy = fabsf(b * inv_det);
+
+   float zoffset = offset->units + MAX2(dzdx, dzdy) * offset->scale;
+
+   /*
+    * Note: we're applying the offset and clamping per-vertex.
+    * Ideally, the offset is applied per-fragment prior to fragment shading.
+    */
+   v0[2] = CLAMP(v0[2] + zoffset, 0.0f, 1.0f);
+   v1[2] = CLAMP(v1[2] + zoffset, 0.0f, 1.0f);
+   v2[2] = CLAMP(v2[2] + zoffset, 0.0f, 1.0f);
+
+   stage->next->tri( stage->next, header );
+}
+
+
+static void offset_tri( struct draw_stage *stage,
+			struct prim_header *header )
+{
+   struct prim_header tmp;
+
+   tmp.det = header->det;
+   tmp.flags = header->flags;
+   tmp.pad = header->pad;
+   tmp.v[0] = dup_vert(stage, header->v[0], 0);
+   tmp.v[1] = dup_vert(stage, header->v[1], 1);
+   tmp.v[2] = dup_vert(stage, header->v[2], 2);
+
+   do_offset_tri( stage, &tmp );
+}
+
+
+static void offset_first_tri( struct draw_stage *stage, 
+			      struct prim_header *header )
+{
+   struct offset_stage *offset = offset_stage(stage);
+
+   offset->units = (float) (stage->draw->rasterizer->offset_units * stage->draw->mrd);
+   offset->scale = stage->draw->rasterizer->offset_scale;
+
+   stage->tri = offset_tri;
+   stage->tri( stage, header );
+}
+
+
+
+
+static void offset_flush( struct draw_stage *stage,
+			  unsigned flags )
+{
+   stage->tri = offset_first_tri;
+   stage->next->flush( stage->next, flags );
+}
+
+
+static void offset_reset_stipple_counter( struct draw_stage *stage )
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void offset_destroy( struct draw_stage *stage )
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+/**
+ * Create polygon offset drawing stage.
+ */
+struct draw_stage *draw_offset_stage( struct draw_context *draw )
+{
+   struct offset_stage *offset = CALLOC_STRUCT(offset_stage);
+   if (offset == NULL)
+      goto fail;
+
+   draw_alloc_temp_verts( &offset->stage, 3 );
+
+   offset->stage.draw = draw;
+   offset->stage.next = NULL;
+   offset->stage.point = draw_pipe_passthrough_point;
+   offset->stage.line = draw_pipe_passthrough_line;
+   offset->stage.tri = offset_first_tri;
+   offset->stage.flush = offset_flush;
+   offset->stage.reset_stipple_counter = offset_reset_stipple_counter;
+   offset->stage.destroy = offset_destroy;
+
+   return &offset->stage;
+
+ fail:
+   if (offset)
+      offset->stage.destroy( &offset->stage );
+
+   return NULL;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pipe_pstipple.c b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c
new file mode 100644
index 00000000000..a0f9716dac2
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pipe_pstipple.c
@@ -0,0 +1,770 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * Polygon stipple stage:  implement polygon stipple with texture map and
+ * fragment program.  The fragment program samples the texture and does
+ * a fragment kill for the stipple-failing fragments.
+ *
+ * Authors:  Brian Paul
+ */
+
+
+#include "pipe/p_inlines.h"
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "util/u_math.h"
+#include "util/u_memory.h"
+
+#include "tgsi/tgsi_transform.h"
+#include "tgsi/tgsi_dump.h"
+
+#include "draw_context.h"
+#include "draw_pipe.h"
+
+
+
+/**
+ * Subclass of pipe_shader_state to carry extra fragment shader info.
+ */
+struct pstip_fragment_shader
+{
+   struct pipe_shader_state state;
+   void *driver_fs;
+   void *pstip_fs;
+   uint sampler_unit;
+};
+
+
+/**
+ * Subclass of draw_stage
+ */
+struct pstip_stage
+{
+   struct draw_stage stage;
+
+   void *sampler_cso;
+   struct pipe_texture *texture;
+   uint num_samplers;
+   uint num_textures;
+
+   /*
+    * Currently bound state
+    */
+   struct pstip_fragment_shader *fs;
+   struct {
+      void *samplers[PIPE_MAX_SAMPLERS];
+      struct pipe_texture *textures[PIPE_MAX_SAMPLERS];
+      const struct pipe_poly_stipple *stipple;
+   } state;
+
+   /*
+    * Driver interface/override functions
+    */
+   void * (*driver_create_fs_state)(struct pipe_context *,
+                                    const struct pipe_shader_state *);
+   void (*driver_bind_fs_state)(struct pipe_context *, void *);
+   void (*driver_delete_fs_state)(struct pipe_context *, void *);
+
+   void (*driver_bind_sampler_states)(struct pipe_context *, unsigned, void **);
+
+   void (*driver_set_sampler_textures)(struct pipe_context *, unsigned,
+                                       struct pipe_texture **);
+
+   void (*driver_set_polygon_stipple)(struct pipe_context *,
+                                      const struct pipe_poly_stipple *);
+
+   struct pipe_context *pipe;
+};
+
+
+
+/**
+ * Subclass of tgsi_transform_context, used for transforming the
+ * user's fragment shader to add the special AA instructions.
+ */
+struct pstip_transform_context {
+   struct tgsi_transform_context base;
+   uint tempsUsed;  /**< bitmask */
+   int wincoordInput;
+   int maxInput;
+   uint samplersUsed;  /**< bitfield of samplers used */
+   int freeSampler;  /** an available sampler for the pstipple */
+   int texTemp;  /**< temp registers */
+   int numImmed;
+   boolean firstInstruction;
+};
+
+
+/**
+ * TGSI declaration transform callback.
+ * Look for a free sampler, a free input attrib, and two free temp regs.
+ */
+static void
+pstip_transform_decl(struct tgsi_transform_context *ctx,
+                     struct tgsi_full_declaration *decl)
+{
+   struct pstip_transform_context *pctx = (struct pstip_transform_context *) ctx;
+
+   if (decl->Declaration.File == TGSI_FILE_SAMPLER) {
+      uint i;
+      for (i = decl->DeclarationRange.First;
+           i <= decl->DeclarationRange.Last; i++) {
+         pctx->samplersUsed |= 1 << i;
+      }
+   }
+   else if (decl->Declaration.File == TGSI_FILE_INPUT) {
+      pctx->maxInput = MAX2(pctx->maxInput, (int) decl->DeclarationRange.Last);
+      if (decl->Semantic.SemanticName == TGSI_SEMANTIC_POSITION)
+         pctx->wincoordInput = (int) decl->DeclarationRange.First;
+   }
+   else if (decl->Declaration.File == TGSI_FILE_TEMPORARY) {
+      uint i;
+      for (i = decl->DeclarationRange.First;
+           i <= decl->DeclarationRange.Last; i++) {
+         pctx->tempsUsed |= (1 << i);
+      }
+   }
+
+   ctx->emit_declaration(ctx, decl);
+}
+
+
+static void
+pstip_transform_immed(struct tgsi_transform_context *ctx,
+                      struct tgsi_full_immediate *immed)
+{
+   struct pstip_transform_context *pctx = (struct pstip_transform_context *) ctx;
+   pctx->numImmed++;
+}
+
+
+/**
+ * Find the lowest zero bit in the given word, or -1 if bitfield is all ones.
+ */
+static int
+free_bit(uint bitfield)
+{
+   int i;
+   for (i = 0; i < 32; i++) {
+      if ((bitfield & (1 << i)) == 0)
+         return i;
+   }
+   return -1;
+}
+
+
+/**
+ * TGSI instruction transform callback.
+ * Replace writes to result.color w/ a temp reg.
+ * Upon END instruction, insert texture sampling code for antialiasing.
+ */
+static void
+pstip_transform_inst(struct tgsi_transform_context *ctx,
+                     struct tgsi_full_instruction *inst)
+{
+   struct pstip_transform_context *pctx = (struct pstip_transform_context *) ctx;
+
+   if (pctx->firstInstruction) {
+      /* emit our new declarations before the first instruction */
+
+      struct tgsi_full_declaration decl;
+      struct tgsi_full_instruction newInst;
+      uint i;
+      int wincoordInput;
+
+      /* find free sampler */
+      pctx->freeSampler = free_bit(pctx->samplersUsed);
+      if (pctx->freeSampler >= PIPE_MAX_SAMPLERS)
+         pctx->freeSampler = PIPE_MAX_SAMPLERS - 1;
+
+      if (pctx->wincoordInput < 0)
+         wincoordInput = pctx->maxInput + 1;
+      else
+         wincoordInput = pctx->wincoordInput;
+
+      /* find one free temp reg */
+      for (i = 0; i < 32; i++) {
+         if ((pctx->tempsUsed & (1 << i)) == 0) {
+            /* found a free temp */
+            if (pctx->texTemp < 0)
+               pctx->texTemp  = i;
+            else
+               break;
+         }
+      }
+      assert(pctx->texTemp >= 0);
+
+      if (pctx->wincoordInput < 0) {
+         /* declare new position input reg */
+         decl = tgsi_default_full_declaration();
+         decl.Declaration.File = TGSI_FILE_INPUT;
+         decl.Declaration.Interpolate = TGSI_INTERPOLATE_LINEAR; /* XXX? */
+         decl.Declaration.Semantic = 1;
+         decl.Semantic.SemanticName = TGSI_SEMANTIC_POSITION;
+         decl.Semantic.SemanticIndex = 0;
+         decl.DeclarationRange.First = 
+            decl.DeclarationRange.Last = wincoordInput;
+         ctx->emit_declaration(ctx, &decl);
+      }
+
+      /* declare new sampler */
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_SAMPLER;
+      decl.DeclarationRange.First = 
+      decl.DeclarationRange.Last = pctx->freeSampler;
+      ctx->emit_declaration(ctx, &decl);
+
+      /* declare new temp regs */
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_TEMPORARY;
+      decl.DeclarationRange.First = 
+      decl.DeclarationRange.Last = pctx->texTemp;
+      ctx->emit_declaration(ctx, &decl);
+
+      /* emit immediate = {1/32, 1/32, 1, 1}
+       * The index/position of this immediate will be pctx->numImmed
+       */
+      {
+         static const float value[4] = { 1.0/32, 1.0/32, 1.0, 1.0 };
+         struct tgsi_full_immediate immed;
+         uint size = 4;
+         immed = tgsi_default_full_immediate();
+         immed.Immediate.NrTokens = 1 + size; /* one for the token itself */
+         immed.u.Pointer = (void *) value;
+         ctx->emit_immediate(ctx, &immed);
+      }
+
+      pctx->firstInstruction = FALSE;
+
+
+      /* 
+       * Insert new MUL/TEX/KILP instructions at start of program
+       * Take gl_FragCoord, divide by 32 (stipple size), sample the
+       * texture and kill fragment if needed.
+       *
+       * We'd like to use non-normalized texcoords to index into a RECT
+       * texture, but we can only use GL_REPEAT wrap mode with normalized
+       * texcoords.  Darn.
+       */
+
+      /* MUL texTemp, INPUT[wincoord], 1/32; */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_MUL;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = pctx->texTemp;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = wincoordInput;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_IMMEDIATE;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = pctx->numImmed;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* TEX texTemp, texTemp, sampler; */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_TEX;
+      newInst.Instruction.NumDstRegs = 1;
+      newInst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullDstRegisters[0].DstRegister.Index = pctx->texTemp;
+      newInst.Instruction.NumSrcRegs = 2;
+      newInst.InstructionExtTexture.Texture = TGSI_TEXTURE_2D;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = pctx->texTemp;
+      newInst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_SAMPLER;
+      newInst.FullSrcRegisters[1].SrcRegister.Index = pctx->freeSampler;
+      ctx->emit_instruction(ctx, &newInst);
+
+      /* KIL -texTemp;   # if -texTemp < 0, KILL fragment */
+      newInst = tgsi_default_full_instruction();
+      newInst.Instruction.Opcode = TGSI_OPCODE_KIL;
+      newInst.Instruction.NumDstRegs = 0;
+      newInst.Instruction.NumSrcRegs = 1;
+      newInst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_TEMPORARY;
+      newInst.FullSrcRegisters[0].SrcRegister.Index = pctx->texTemp;
+      newInst.FullSrcRegisters[0].SrcRegister.Negate = 1;
+      ctx->emit_instruction(ctx, &newInst);
+   }
+
+   /* emit this instruction */
+   ctx->emit_instruction(ctx, inst);
+}
+
+
+/**
+ * Generate the frag shader we'll use for doing polygon stipple.
+ * This will be the user's shader prefixed with a TEX and KIL instruction.
+ */
+static boolean
+generate_pstip_fs(struct pstip_stage *pstip)
+{
+   const struct pipe_shader_state *orig_fs = &pstip->fs->state;
+   /*struct draw_context *draw = pstip->stage.draw;*/
+   struct pipe_shader_state pstip_fs;
+   struct pstip_transform_context transform;
+
+#define MAX 1000
+
+   pstip_fs = *orig_fs; /* copy to init */
+   pstip_fs.tokens = MALLOC(sizeof(struct tgsi_token) * MAX);
+   if (pstip_fs.tokens == NULL)
+      return FALSE;
+
+   memset(&transform, 0, sizeof(transform));
+   transform.wincoordInput = -1;
+   transform.maxInput = -1;
+   transform.texTemp = -1;
+   transform.firstInstruction = TRUE;
+   transform.base.transform_instruction = pstip_transform_inst;
+   transform.base.transform_declaration = pstip_transform_decl;
+   transform.base.transform_immediate = pstip_transform_immed;
+
+   tgsi_transform_shader(orig_fs->tokens,
+                         (struct tgsi_token *) pstip_fs.tokens,
+                         MAX, &transform.base);
+
+#if 0 /* DEBUG */
+   tgsi_dump(orig_fs->tokens, 0);
+   tgsi_dump(pstip_fs.tokens, 0);
+#endif
+
+   pstip->fs->sampler_unit = transform.freeSampler;
+   assert(pstip->fs->sampler_unit < PIPE_MAX_SAMPLERS);
+
+   pstip->fs->pstip_fs = pstip->driver_create_fs_state(pstip->pipe, &pstip_fs);
+
+   return TRUE;
+}
+
+
+/**
+ * Load texture image with current stipple pattern.
+ */
+static void
+pstip_update_texture(struct pstip_stage *pstip)
+{
+   static const uint bit31 = 1 << 31;
+   struct pipe_context *pipe = pstip->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   struct pipe_surface *surface;
+   const uint *stipple = pstip->state.stipple->stipple;
+   uint i, j;
+   ubyte *data;
+
+   /* XXX: want to avoid flushing just because we use stipple: 
+    */
+   pipe->flush( pipe, PIPE_FLUSH_TEXTURE_CACHE, NULL );
+
+   surface = screen->get_tex_surface(screen, pstip->texture, 0, 0, 0,
+                                     PIPE_BUFFER_USAGE_CPU_WRITE);
+   data = screen->surface_map(screen, surface,
+                              PIPE_BUFFER_USAGE_CPU_WRITE);
+
+   /*
+    * Load alpha texture.
+    * Note: 0 means keep the fragment, 255 means kill it.
+    * We'll negate the texel value and use KILP which kills if value
+    * is negative.
+    */
+   for (i = 0; i < 32; i++) {
+      for (j = 0; j < 32; j++) {
+         if (stipple[i] & (bit31 >> j)) {
+            /* fragment "on" */
+            data[i * surface->stride + j] = 0;
+         }
+         else {
+            /* fragment "off" */
+            data[i * surface->stride + j] = 255;
+         }
+      }
+   }
+
+   /* unmap */
+   screen->surface_unmap(screen, surface);
+   screen->tex_surface_release(screen, &surface);
+}
+
+
+/**
+ * Create the texture map we'll use for stippling.
+ */
+static boolean
+pstip_create_texture(struct pstip_stage *pstip)
+{
+   struct pipe_context *pipe = pstip->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   struct pipe_texture texTemp;
+
+   memset(&texTemp, 0, sizeof(texTemp));
+   texTemp.target = PIPE_TEXTURE_2D;
+   texTemp.format = PIPE_FORMAT_A8_UNORM; /* XXX verify supported by driver! */
+   texTemp.last_level = 0;
+   texTemp.width[0] = 32;
+   texTemp.height[0] = 32;
+   texTemp.depth[0] = 1;
+   pf_get_block(texTemp.format, &texTemp.block);
+
+   pstip->texture = screen->texture_create(screen, &texTemp);
+   if (pstip->texture == NULL)
+      return FALSE;
+
+   return TRUE;
+}
+
+
+/**
+ * Create the sampler CSO that'll be used for stippling.
+ */
+static boolean
+pstip_create_sampler(struct pstip_stage *pstip)
+{
+   struct pipe_sampler_state sampler;
+   struct pipe_context *pipe = pstip->pipe;
+
+   memset(&sampler, 0, sizeof(sampler));
+   sampler.wrap_s = PIPE_TEX_WRAP_REPEAT;
+   sampler.wrap_t = PIPE_TEX_WRAP_REPEAT;
+   sampler.wrap_r = PIPE_TEX_WRAP_REPEAT;
+   sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
+   sampler.min_img_filter = PIPE_TEX_FILTER_NEAREST;
+   sampler.mag_img_filter = PIPE_TEX_FILTER_NEAREST;
+   sampler.normalized_coords = 1;
+   sampler.min_lod = 0.0f;
+   sampler.max_lod = 0.0f;
+
+   pstip->sampler_cso = pipe->create_sampler_state(pipe, &sampler);
+   if (pstip->sampler_cso == NULL)
+      return FALSE;
+   
+   return TRUE;
+}
+
+
+/**
+ * When we're about to draw our first stipple polygon in a batch, this function
+ * is called to tell the driver to bind our modified fragment shader.
+ */
+static boolean
+bind_pstip_fragment_shader(struct pstip_stage *pstip)
+{
+   struct draw_context *draw = pstip->stage.draw;
+   if (!pstip->fs->pstip_fs &&
+       !generate_pstip_fs(pstip))
+      return FALSE;
+
+   draw->suspend_flushing = TRUE;
+   pstip->driver_bind_fs_state(pstip->pipe, pstip->fs->pstip_fs);
+   draw->suspend_flushing = FALSE;
+   return TRUE;
+}
+
+
+static INLINE struct pstip_stage *
+pstip_stage( struct draw_stage *stage )
+{
+   return (struct pstip_stage *) stage;
+}
+
+
+static void
+pstip_first_tri(struct draw_stage *stage, struct prim_header *header)
+{
+   struct pstip_stage *pstip = pstip_stage(stage);
+   struct pipe_context *pipe = pstip->pipe;
+   struct draw_context *draw = stage->draw;
+   uint num_samplers;
+
+   assert(stage->draw->rasterizer->poly_stipple_enable);
+
+   /* bind our fragprog */
+   if (!bind_pstip_fragment_shader(pstip)) {
+      stage->tri = draw_pipe_passthrough_tri;
+      stage->tri(stage, header);
+      return;
+   }
+      
+
+   /* how many samplers? */
+   /* we'll use sampler/texture[pstip->sampler_unit] for the stipple */
+   num_samplers = MAX2(pstip->num_textures, pstip->num_samplers);
+   num_samplers = MAX2(num_samplers, pstip->fs->sampler_unit + 1);
+
+   /* plug in our sampler, texture */
+   pstip->state.samplers[pstip->fs->sampler_unit] = pstip->sampler_cso;
+   pipe_texture_reference(&pstip->state.textures[pstip->fs->sampler_unit],
+                          pstip->texture);
+
+   assert(num_samplers <= PIPE_MAX_SAMPLERS);
+
+   draw->suspend_flushing = TRUE;
+   pstip->driver_bind_sampler_states(pipe, num_samplers, pstip->state.samplers);
+   pstip->driver_set_sampler_textures(pipe, num_samplers, pstip->state.textures);
+   draw->suspend_flushing = FALSE;
+
+   /* now really draw first triangle */
+   stage->tri = draw_pipe_passthrough_tri;
+   stage->tri(stage, header);
+}
+
+
+static void
+pstip_flush(struct draw_stage *stage, unsigned flags)
+{
+   struct draw_context *draw = stage->draw;
+   struct pstip_stage *pstip = pstip_stage(stage);
+   struct pipe_context *pipe = pstip->pipe;
+
+   stage->tri = pstip_first_tri;
+   stage->next->flush( stage->next, flags );
+
+   /* restore original frag shader, texture, sampler state */
+   draw->suspend_flushing = TRUE;
+   pstip->driver_bind_fs_state(pipe, pstip->fs->driver_fs);
+   pstip->driver_bind_sampler_states(pipe, pstip->num_samplers,
+                                     pstip->state.samplers);
+   pstip->driver_set_sampler_textures(pipe, pstip->num_textures,
+                                      pstip->state.textures);
+   draw->suspend_flushing = FALSE;
+}
+
+
+static void
+pstip_reset_stipple_counter(struct draw_stage *stage)
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void
+pstip_destroy(struct draw_stage *stage)
+{
+   struct pstip_stage *pstip = pstip_stage(stage);
+   uint i;
+
+   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+      pipe_texture_reference(&pstip->state.textures[i], NULL);
+   }
+
+   pstip->pipe->delete_sampler_state(pstip->pipe, pstip->sampler_cso);
+
+   pipe_texture_release(&pstip->texture);
+
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+static struct pstip_stage *
+draw_pstip_stage(struct draw_context *draw)
+{
+   struct pstip_stage *pstip = CALLOC_STRUCT(pstip_stage);
+
+   draw_alloc_temp_verts( &pstip->stage, 8 );
+
+   pstip->stage.draw = draw;
+   pstip->stage.next = NULL;
+   pstip->stage.point = draw_pipe_passthrough_point;
+   pstip->stage.line = draw_pipe_passthrough_line;
+   pstip->stage.tri = pstip_first_tri;
+   pstip->stage.flush = pstip_flush;
+   pstip->stage.reset_stipple_counter = pstip_reset_stipple_counter;
+   pstip->stage.destroy = pstip_destroy;
+
+   return pstip;
+}
+
+
+static struct pstip_stage *
+pstip_stage_from_pipe(struct pipe_context *pipe)
+{
+   struct draw_context *draw = (struct draw_context *) pipe->draw;
+   return pstip_stage(draw->pipeline.pstipple);
+}
+
+
+/**
+ * This function overrides the driver's create_fs_state() function and
+ * will typically be called by the state tracker.
+ */
+static void *
+pstip_create_fs_state(struct pipe_context *pipe,
+                       const struct pipe_shader_state *fs)
+{
+   struct pstip_stage *pstip = pstip_stage_from_pipe(pipe);
+   struct pstip_fragment_shader *aafs = CALLOC_STRUCT(pstip_fragment_shader);
+
+   if (aafs) {
+      aafs->state = *fs;
+
+      /* pass-through */
+      aafs->driver_fs = pstip->driver_create_fs_state(pstip->pipe, fs);
+   }
+
+   return aafs;
+}
+
+
+static void
+pstip_bind_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct pstip_stage *pstip = pstip_stage_from_pipe(pipe);
+   struct pstip_fragment_shader *aafs = (struct pstip_fragment_shader *) fs;
+   /* save current */
+   pstip->fs = aafs;
+   /* pass-through */
+   pstip->driver_bind_fs_state(pstip->pipe,
+                               (aafs ? aafs->driver_fs : NULL));
+}
+
+
+static void
+pstip_delete_fs_state(struct pipe_context *pipe, void *fs)
+{
+   struct pstip_stage *pstip = pstip_stage_from_pipe(pipe);
+   struct pstip_fragment_shader *aafs = (struct pstip_fragment_shader *) fs;
+   /* pass-through */
+   pstip->driver_delete_fs_state(pstip->pipe, aafs->driver_fs);
+   FREE(aafs);
+}
+
+
+static void
+pstip_bind_sampler_states(struct pipe_context *pipe,
+                          unsigned num, void **sampler)
+{
+   struct pstip_stage *pstip = pstip_stage_from_pipe(pipe);
+   uint i;
+
+   /* save current */
+   memcpy(pstip->state.samplers, sampler, num * sizeof(void *));
+   for (i = num; i < PIPE_MAX_SAMPLERS; i++) {
+      pstip->state.samplers[i] = NULL;
+   }
+
+   pstip->num_samplers = num;
+   /* pass-through */
+   pstip->driver_bind_sampler_states(pstip->pipe, num, sampler);
+}
+
+
+static void
+pstip_set_sampler_textures(struct pipe_context *pipe,
+                           unsigned num, struct pipe_texture **texture)
+{
+   struct pstip_stage *pstip = pstip_stage_from_pipe(pipe);
+   uint i;
+
+   /* save current */
+   for (i = 0; i < num; i++) {
+      pipe_texture_reference(&pstip->state.textures[i], texture[i]);
+   }
+   for (; i < PIPE_MAX_SAMPLERS; i++) {
+      pipe_texture_reference(&pstip->state.textures[i], NULL);
+   }
+
+   pstip->num_textures = num;
+
+   /* pass-through */
+   pstip->driver_set_sampler_textures(pstip->pipe, num, texture);
+}
+
+
+static void
+pstip_set_polygon_stipple(struct pipe_context *pipe,
+                          const struct pipe_poly_stipple *stipple)
+{
+   struct pstip_stage *pstip = pstip_stage_from_pipe(pipe);
+
+   /* save current */
+   pstip->state.stipple = stipple;
+
+   /* pass-through */
+   pstip->driver_set_polygon_stipple(pstip->pipe, stipple);
+
+   pstip_update_texture(pstip);
+}
+
+
+/**
+ * Called by drivers that want to install this polygon stipple stage
+ * into the draw module's pipeline.  This will not be used if the
+ * hardware has native support for polygon stipple.
+ */
+boolean
+draw_install_pstipple_stage(struct draw_context *draw,
+                            struct pipe_context *pipe)
+{
+   struct pstip_stage *pstip;
+
+   pipe->draw = (void *) draw;
+
+   /*
+    * Create / install pgon stipple drawing / prim stage
+    */
+   pstip = draw_pstip_stage( draw );
+   if (pstip == NULL)
+      goto fail;
+
+   draw->pipeline.pstipple = &pstip->stage;
+
+   pstip->pipe = pipe;
+
+   /* create special texture, sampler state */
+   if (!pstip_create_texture(pstip))
+      goto fail;
+
+   if (!pstip_create_sampler(pstip))
+      goto fail;
+
+   /* save original driver functions */
+   pstip->driver_create_fs_state = pipe->create_fs_state;
+   pstip->driver_bind_fs_state = pipe->bind_fs_state;
+   pstip->driver_delete_fs_state = pipe->delete_fs_state;
+
+   pstip->driver_bind_sampler_states = pipe->bind_sampler_states;
+   pstip->driver_set_sampler_textures = pipe->set_sampler_textures;
+   pstip->driver_set_polygon_stipple = pipe->set_polygon_stipple;
+
+   /* override the driver's functions */
+   pipe->create_fs_state = pstip_create_fs_state;
+   pipe->bind_fs_state = pstip_bind_fs_state;
+   pipe->delete_fs_state = pstip_delete_fs_state;
+
+   pipe->bind_sampler_states = pstip_bind_sampler_states;
+   pipe->set_sampler_textures = pstip_set_sampler_textures;
+   pipe->set_polygon_stipple = pstip_set_polygon_stipple;
+
+   return TRUE;
+
+ fail:
+   if (pstip)
+      pstip->stage.destroy( &pstip->stage );
+
+   return FALSE;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pipe_stipple.c b/src/gallium/auxiliary/draw/draw_pipe_stipple.c
new file mode 100644
index 00000000000..b65e2aa1021
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pipe_stipple.c
@@ -0,0 +1,250 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <[email protected]>
+ */
+
+/* Implement line stipple by cutting lines up into smaller lines.
+ * There are hundreds of ways to implement line stipple, this is one
+ * choice that should work in all situations, requires no state
+ * manipulations, but with a penalty in terms of large amounts of
+ * generated geometry.
+ */
+
+
+#include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
+
+#include "draw/draw_pipe.h"
+
+
+/** Subclass of draw_stage */
+struct stipple_stage {
+   struct draw_stage stage;
+   float counter;
+   uint pattern;
+   uint factor;
+};
+
+
+static INLINE struct stipple_stage *
+stipple_stage(struct draw_stage *stage)
+{
+   return (struct stipple_stage *) stage;
+}
+
+
+/**
+ * Compute interpolated vertex attributes for 'dst' at position 't' 
+ * between 'v0' and 'v1'.
+ * XXX using linear interpolation for all attribs at this time.
+ */
+static void
+screen_interp( struct draw_context *draw,
+               struct vertex_header *dst,
+               float t,
+               const struct vertex_header *v0, 
+               const struct vertex_header *v1 )
+{
+   uint attr;
+   for (attr = 0; attr < draw->vs.num_vs_outputs; attr++) {
+      const float *val0 = v0->data[attr];
+      const float *val1 = v1->data[attr];
+      float *newv = dst->data[attr];
+      uint i;
+      for (i = 0; i < 4; i++) {
+         newv[i] = val0[i] + t * (val1[i] - val0[i]);
+      }
+   }
+}
+
+
+static void
+emit_segment(struct draw_stage *stage, struct prim_header *header,
+             float t0, float t1)
+{
+   struct vertex_header *v0new = dup_vert(stage, header->v[0], 0);
+   struct vertex_header *v1new = dup_vert(stage, header->v[1], 1);
+   struct prim_header newprim = *header;
+
+   if (t0 > 0.0) {
+      screen_interp( stage->draw, v0new, t0, header->v[0], header->v[1] );
+      newprim.v[0] = v0new;
+   }
+
+   if (t1 < 1.0) {
+      screen_interp( stage->draw, v1new, t1, header->v[0], header->v[1] );
+      newprim.v[1] = v1new;
+   }
+
+   stage->next->line( stage->next, &newprim );
+}
+
+
+static INLINE unsigned
+stipple_test(int counter, ushort pattern, int factor)
+{
+   int b = (counter / factor) & 0xf;
+   return (1 << b) & pattern;
+}
+
+
+static void
+stipple_line(struct draw_stage *stage, struct prim_header *header)
+{
+   struct stipple_stage *stipple = stipple_stage(stage);
+   struct vertex_header *v0 = header->v[0];
+   struct vertex_header *v1 = header->v[1];
+   const unsigned pos = stage->draw->vs.position_output;
+   const float *pos0 = v0->data[pos];
+   const float *pos1 = v1->data[pos];
+   float start = 0;
+   int state = 0;
+
+   float x0 = pos0[0];
+   float x1 = pos1[0];
+   float y0 = pos0[1];
+   float y1 = pos1[1];
+
+   float dx = x0 > x1 ? x0 - x1 : x1 - x0;
+   float dy = y0 > y1 ? y0 - y1 : y1 - y0;
+
+   float length = MAX2(dx, dy);
+   int i;
+
+   if (header->flags & DRAW_PIPE_RESET_STIPPLE)
+      stipple->counter = 0;
+
+
+   /* XXX ToDo: intead of iterating pixel-by-pixel, use a look-up table.
+    */
+   for (i = 0; i < length; i++) {
+      int result = stipple_test( (int) stipple->counter+i,
+                                 (ushort) stipple->pattern, stipple->factor );
+      if (result != state) {
+         /* changing from "off" to "on" or vice versa */
+	 if (state) {
+	    if (start != i) {
+               /* finishing an "on" segment */
+	       emit_segment( stage, header, start / length, i / length );
+            }
+	 }
+	 else {
+            /* starting an "on" segment */
+	    start = (float) i;
+	 }
+	 state = result;	   
+      }
+   }
+
+   if (state && start < length)
+      emit_segment( stage, header, start / length, 1.0 );
+
+   stipple->counter += length;
+}
+
+
+static void
+reset_stipple_counter(struct draw_stage *stage)
+{
+   struct stipple_stage *stipple = stipple_stage(stage);
+   stipple->counter = 0;
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+static void
+stipple_reset_point(struct draw_stage *stage, struct prim_header *header)
+{
+   struct stipple_stage *stipple = stipple_stage(stage);
+   stipple->counter = 0;
+   stage->next->point(stage->next, header);
+}
+
+static void
+stipple_reset_tri(struct draw_stage *stage, struct prim_header *header)
+{
+   struct stipple_stage *stipple = stipple_stage(stage);
+   stipple->counter = 0;
+   stage->next->tri(stage->next, header);
+}
+
+
+static void
+stipple_first_line(struct draw_stage *stage, 
+		   struct prim_header *header)
+{
+   struct stipple_stage *stipple = stipple_stage(stage);
+   struct draw_context *draw = stage->draw;
+
+   stipple->pattern = draw->rasterizer->line_stipple_pattern;
+   stipple->factor = draw->rasterizer->line_stipple_factor + 1;
+
+   stage->line = stipple_line;
+   stage->line( stage, header );
+}
+
+
+static void
+stipple_flush(struct draw_stage *stage, unsigned flags)
+{
+   stage->line = stipple_first_line;
+   stage->next->flush( stage->next, flags );
+}
+
+
+
+
+static void 
+stipple_destroy( struct draw_stage *stage )
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+/**
+ * Create line stippler stage
+ */
+struct draw_stage *draw_stipple_stage( struct draw_context *draw )
+{
+   struct stipple_stage *stipple = CALLOC_STRUCT(stipple_stage);
+
+   draw_alloc_temp_verts( &stipple->stage, 2 );
+
+   stipple->stage.draw = draw;
+   stipple->stage.next = NULL;
+   stipple->stage.point = stipple_reset_point;
+   stipple->stage.line = stipple_first_line;
+   stipple->stage.tri = stipple_reset_tri;
+   stipple->stage.reset_stipple_counter = reset_stipple_counter;
+   stipple->stage.flush = stipple_flush;
+   stipple->stage.destroy = stipple_destroy;
+
+   return &stipple->stage;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pipe_twoside.c b/src/gallium/auxiliary/draw/draw_pipe_twoside.c
new file mode 100644
index 00000000000..c329d923390
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pipe_twoside.c
@@ -0,0 +1,199 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <[email protected]>
+ */
+
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
+#include "draw_vs.h"
+#include "draw_pipe.h"
+
+struct twoside_stage {
+   struct draw_stage stage;
+   float sign;         /**< +1 or -1 */
+   uint attrib_front0, attrib_back0;
+   uint attrib_front1, attrib_back1;
+};
+
+
+static INLINE struct twoside_stage *twoside_stage( struct draw_stage *stage )
+{
+   return (struct twoside_stage *)stage;
+}
+
+
+
+
+/**
+ * Copy back color(s) to front color(s).
+ */
+static INLINE struct vertex_header *
+copy_bfc( struct twoside_stage *twoside, 
+          const struct vertex_header *v,
+          unsigned idx )
+{   
+   struct vertex_header *tmp = dup_vert( &twoside->stage, v, idx );
+   
+   if (twoside->attrib_back0) {
+      COPY_4FV(tmp->data[twoside->attrib_front0],
+               tmp->data[twoside->attrib_back0]);
+   }
+   if (twoside->attrib_back1) {
+      COPY_4FV(tmp->data[twoside->attrib_front1],
+               tmp->data[twoside->attrib_back1]);
+   }
+
+   return tmp;
+}
+
+
+/* Twoside tri:
+ */
+static void twoside_tri( struct draw_stage *stage,
+			 struct prim_header *header )
+{
+   struct twoside_stage *twoside = twoside_stage(stage);
+
+   if (header->det * twoside->sign < 0.0) {
+      /* this is a back-facing triangle */
+      struct prim_header tmp;
+
+      tmp.det = header->det;
+      tmp.flags = header->flags;
+      tmp.pad = header->pad;
+      /* copy back attribs to front attribs */
+      tmp.v[0] = copy_bfc(twoside, header->v[0], 0);
+      tmp.v[1] = copy_bfc(twoside, header->v[1], 1);
+      tmp.v[2] = copy_bfc(twoside, header->v[2], 2);
+
+      stage->next->tri( stage->next, &tmp );
+   }
+   else {
+      stage->next->tri( stage->next, header );
+   }
+}
+
+
+
+static void twoside_first_tri( struct draw_stage *stage, 
+			       struct prim_header *header )
+{
+   struct twoside_stage *twoside = twoside_stage(stage);
+   const struct draw_vertex_shader *vs = stage->draw->vs.vertex_shader;
+   uint i;
+
+   twoside->attrib_front0 = 0;
+   twoside->attrib_front1 = 0;
+   twoside->attrib_back0 = 0;
+   twoside->attrib_back1 = 0;
+
+   /* Find which vertex shader outputs are front/back colors */
+   for (i = 0; i < vs->info.num_outputs; i++) {
+      if (vs->info.output_semantic_name[i] == TGSI_SEMANTIC_COLOR) {
+         if (vs->info.output_semantic_index[i] == 0)
+            twoside->attrib_front0 = i;
+         else
+            twoside->attrib_front1 = i;
+      }
+      if (vs->info.output_semantic_name[i] == TGSI_SEMANTIC_BCOLOR) {
+         if (vs->info.output_semantic_index[i] == 0)
+            twoside->attrib_back0 = i;
+         else
+            twoside->attrib_back1 = i;
+      }
+   }
+
+   if (!twoside->attrib_back0)
+      twoside->attrib_front0 = 0;
+
+   if (!twoside->attrib_back1)
+      twoside->attrib_front1 = 0;
+
+   /*
+    * We'll multiply the primitive's determinant by this sign to determine
+    * if the triangle is back-facing (negative).
+    * sign = -1 for CCW, +1 for CW
+    */
+   twoside->sign = (stage->draw->rasterizer->front_winding == PIPE_WINDING_CCW) ? -1.0f : 1.0f;
+
+   stage->tri = twoside_tri;
+   stage->tri( stage, header );
+}
+
+
+static void twoside_flush( struct draw_stage *stage, unsigned flags )
+{
+   stage->tri = twoside_first_tri;
+   stage->next->flush( stage->next, flags );
+}
+
+
+static void twoside_reset_stipple_counter( struct draw_stage *stage )
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void twoside_destroy( struct draw_stage *stage )
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+/**
+ * Create twoside pipeline stage.
+ */
+struct draw_stage *draw_twoside_stage( struct draw_context *draw )
+{
+   struct twoside_stage *twoside = CALLOC_STRUCT(twoside_stage);
+   if (twoside == NULL)
+      goto fail;
+
+   if (!draw_alloc_temp_verts( &twoside->stage, 3 ))
+      goto fail;
+
+   twoside->stage.draw = draw;
+   twoside->stage.next = NULL;
+   twoside->stage.point = draw_pipe_passthrough_point;
+   twoside->stage.line = draw_pipe_passthrough_line;
+   twoside->stage.tri = twoside_first_tri;
+   twoside->stage.flush = twoside_flush;
+   twoside->stage.reset_stipple_counter = twoside_reset_stipple_counter;
+   twoside->stage.destroy = twoside_destroy;
+
+   return &twoside->stage;
+
+ fail:
+   if (twoside)
+      twoside->stage.destroy( &twoside->stage );
+
+   return NULL;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pipe_unfilled.c b/src/gallium/auxiliary/draw/draw_pipe_unfilled.c
new file mode 100644
index 00000000000..68835fd1a59
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pipe_unfilled.c
@@ -0,0 +1,203 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * \brief  Drawing stage for handling glPolygonMode(line/point).
+ * Convert triangles to points or lines as needed.
+ */
+
+/* Authors:  Keith Whitwell <[email protected]>
+ */
+
+#include "util/u_memory.h"
+#include "pipe/p_defines.h"
+#include "draw_private.h"
+#include "draw_pipe.h"
+
+
+struct unfilled_stage {
+   struct draw_stage stage;
+
+   /** [0] = front face, [1] = back face.
+    * legal values:  PIPE_POLYGON_MODE_FILL, PIPE_POLYGON_MODE_LINE,
+    * and PIPE_POLYGON_MODE_POINT,
+    */
+   unsigned mode[2];
+};
+
+
+static INLINE struct unfilled_stage *unfilled_stage( struct draw_stage *stage )
+{
+   return (struct unfilled_stage *)stage;
+}
+
+
+
+static void point( struct draw_stage *stage,
+		   struct vertex_header *v0 )
+{
+   struct prim_header tmp;
+   tmp.v[0] = v0;
+   stage->next->point( stage->next, &tmp );
+}
+
+static void line( struct draw_stage *stage,
+		  struct vertex_header *v0,
+		  struct vertex_header *v1 )
+{
+   struct prim_header tmp;
+   tmp.v[0] = v0;
+   tmp.v[1] = v1;
+   stage->next->line( stage->next, &tmp );
+}
+
+
+static void points( struct draw_stage *stage,
+		    struct prim_header *header )
+{
+   struct vertex_header *v0 = header->v[0];
+   struct vertex_header *v1 = header->v[1];
+   struct vertex_header *v2 = header->v[2];
+
+   if ((header->flags & DRAW_PIPE_EDGE_FLAG_0) && v0->edgeflag) point( stage, v0 );
+   if ((header->flags & DRAW_PIPE_EDGE_FLAG_1) && v1->edgeflag) point( stage, v1 );
+   if ((header->flags & DRAW_PIPE_EDGE_FLAG_2) && v2->edgeflag) point( stage, v2 );
+}
+
+
+static void lines( struct draw_stage *stage,
+		   struct prim_header *header )
+{
+   struct vertex_header *v0 = header->v[0];
+   struct vertex_header *v1 = header->v[1];
+   struct vertex_header *v2 = header->v[2];
+
+   if (header->flags & DRAW_PIPE_RESET_STIPPLE)
+      stage->next->reset_stipple_counter( stage->next );
+
+   if ((header->flags & DRAW_PIPE_EDGE_FLAG_2) && v2->edgeflag) line( stage, v2, v0 );
+   if ((header->flags & DRAW_PIPE_EDGE_FLAG_0) && v0->edgeflag) line( stage, v0, v1 );
+   if ((header->flags & DRAW_PIPE_EDGE_FLAG_1) && v1->edgeflag) line( stage, v1, v2 );
+}
+
+
+/* Unfilled tri:  
+ *
+ * Note edgeflags in the vertex struct is not sufficient as we will
+ * need to manipulate them when decomposing primitives.  
+ * 
+ * We currently keep the vertex edgeflag and primitive edgeflag mask
+ * separate until the last possible moment.
+ */
+static void unfilled_tri( struct draw_stage *stage,
+			  struct prim_header *header )
+{
+   struct unfilled_stage *unfilled = unfilled_stage(stage);
+   unsigned mode = unfilled->mode[header->det >= 0.0];
+  
+   switch (mode) {
+   case PIPE_POLYGON_MODE_FILL:
+      stage->next->tri( stage->next, header );
+      break;
+   case PIPE_POLYGON_MODE_LINE:
+      lines( stage, header );
+      break;
+   case PIPE_POLYGON_MODE_POINT:
+      points( stage, header );
+      break;
+   default:
+      assert(0);
+   }   
+}
+
+
+static void unfilled_first_tri( struct draw_stage *stage, 
+				struct prim_header *header )
+{
+   struct unfilled_stage *unfilled = unfilled_stage(stage);
+
+   unfilled->mode[0] = stage->draw->rasterizer->fill_ccw; /* front */
+   unfilled->mode[1] = stage->draw->rasterizer->fill_cw;  /* back */
+
+   stage->tri = unfilled_tri;
+   stage->tri( stage, header );
+}
+
+
+
+static void unfilled_flush( struct draw_stage *stage,
+			    unsigned flags )
+{
+   stage->next->flush( stage->next, flags );
+
+   stage->tri = unfilled_first_tri;
+}
+
+
+static void unfilled_reset_stipple_counter( struct draw_stage *stage )
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void unfilled_destroy( struct draw_stage *stage )
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+/**
+ * Create unfilled triangle stage.
+ */
+struct draw_stage *draw_unfilled_stage( struct draw_context *draw )
+{
+   struct unfilled_stage *unfilled = CALLOC_STRUCT(unfilled_stage);
+   if (unfilled == NULL)
+      goto fail;
+
+   if (!draw_alloc_temp_verts( &unfilled->stage, 0 ))
+      goto fail;
+
+   unfilled->stage.draw = draw;
+   unfilled->stage.next = NULL;
+   unfilled->stage.tmp = NULL;
+   unfilled->stage.point = draw_pipe_passthrough_point;
+   unfilled->stage.line = draw_pipe_passthrough_line;
+   unfilled->stage.tri = unfilled_first_tri;
+   unfilled->stage.flush = unfilled_flush;
+   unfilled->stage.reset_stipple_counter = unfilled_reset_stipple_counter;
+   unfilled->stage.destroy = unfilled_destroy;
+
+   return &unfilled->stage;
+
+ fail:
+   if (unfilled)
+      unfilled->stage.destroy( &unfilled->stage );
+
+   return NULL;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pipe_util.c b/src/gallium/auxiliary/draw/draw_pipe_util.c
new file mode 100644
index 00000000000..e22e5fed0c6
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pipe_util.c
@@ -0,0 +1,137 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  */
+
+#include "util/u_memory.h"
+#include "draw/draw_private.h"
+#include "draw/draw_pipe.h"
+
+
+
+void
+draw_pipe_passthrough_point(struct draw_stage *stage, struct prim_header *header)
+{
+   stage->next->point(stage->next, header);
+}
+
+void
+draw_pipe_passthrough_line(struct draw_stage *stage, struct prim_header *header)
+{
+   stage->next->line(stage->next, header);
+}
+
+void
+draw_pipe_passthrough_tri(struct draw_stage *stage, struct prim_header *header)
+{
+   stage->next->tri(stage->next, header);
+}
+
+
+
+
+
+/* This is only used for temporary verts.
+ */
+#define MAX_VERTEX_SIZE ((2 + PIPE_MAX_SHADER_OUTPUTS) * 4 * sizeof(float))
+
+
+/**
+ * Allocate space for temporary post-transform vertices, such as for clipping.
+ */
+boolean draw_alloc_temp_verts( struct draw_stage *stage, unsigned nr )
+{
+   assert(!stage->tmp);
+
+   stage->tmp = NULL;
+   stage->nr_tmps = nr;
+
+   if (nr != 0)
+   {
+      unsigned i;
+      ubyte *store = (ubyte *) MALLOC( MAX_VERTEX_SIZE * nr );
+
+      if (store == NULL)
+         return FALSE;
+
+      stage->tmp = (struct vertex_header **) MALLOC( sizeof(struct vertex_header *) * nr );
+      if (stage->tmp == NULL) {
+         FREE(store);
+         return FALSE;
+      }
+         
+      for (i = 0; i < nr; i++)
+         stage->tmp[i] = (struct vertex_header *)(store + i * MAX_VERTEX_SIZE);
+   }
+
+   return TRUE;
+}
+
+
+void draw_free_temp_verts( struct draw_stage *stage )
+{
+   if (stage->tmp) {
+      FREE( stage->tmp[0] );
+      FREE( stage->tmp );
+      stage->tmp = NULL;
+   }
+}
+
+
+/* Reset vertex ids.  This is basically a type of flush.
+ *
+ * Called only from draw_pipe_vbuf.c
+ */
+void draw_reset_vertex_ids(struct draw_context *draw)
+{
+   struct draw_stage *stage = draw->pipeline.first;
+   
+   while (stage) {
+      unsigned i;
+
+      for (i = 0; i < stage->nr_tmps; i++)
+	 stage->tmp[i]->vertex_id = UNDEFINED_VERTEX_ID;
+
+      stage = stage->next;
+   }
+
+   if (draw->pipeline.verts)
+   {
+      unsigned i;
+      char *verts = draw->pipeline.verts;
+      unsigned stride = draw->pipeline.vertex_stride;
+
+      for (i = 0; i < draw->pipeline.vertex_count; i++) {
+         ((struct vertex_header *)verts)->vertex_id = UNDEFINED_VERTEX_ID;
+         verts += stride;
+      }
+   }
+}
+
diff --git a/src/gallium/auxiliary/draw/draw_pipe_validate.c b/src/gallium/auxiliary/draw/draw_pipe_validate.c
new file mode 100644
index 00000000000..03e842ce082
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pipe_validate.c
@@ -0,0 +1,330 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <[email protected]>
+ */
+
+#include "util/u_memory.h"
+#include "pipe/p_defines.h"
+#include "draw_private.h"
+#include "draw_pipe.h"
+#include "draw_context.h"
+#include "draw_vbuf.h"
+
+static boolean points( unsigned prim )
+{
+   return (prim == PIPE_PRIM_POINTS);
+}
+
+static boolean lines( unsigned prim )
+{
+   return (prim == PIPE_PRIM_LINES ||
+           prim == PIPE_PRIM_LINE_STRIP ||
+           prim == PIPE_PRIM_LINE_LOOP);
+}
+
+static boolean triangles( unsigned prim )
+{
+   return prim >= PIPE_PRIM_TRIANGLES;
+}
+
+/**
+ * Default version of a function to check if we need any special
+ * pipeline stages, or whether prims/verts can go through untouched.
+ * Don't test for bypass clipping or vs modes, this function is just
+ * about the primitive pipeline stages.
+ *
+ * This can be overridden by the driver.
+ */
+boolean
+draw_need_pipeline(const struct draw_context *draw,
+                   const struct pipe_rasterizer_state *rasterizer,
+                   unsigned int prim )
+{
+   /* If the driver has overridden this, use that version: 
+    */
+   if (draw->render &&
+       draw->render->need_pipeline) 
+   {
+      return draw->render->need_pipeline( draw->render,
+                                          rasterizer,
+                                          prim );
+   }
+
+   /* Don't have to worry about triangles turning into lines/points
+    * and triggering the pipeline, because we have to trigger the
+    * pipeline *anyway* if unfilled mode is active.
+    */
+   if (lines(prim)) 
+   {
+      /* line stipple */
+      if (rasterizer->line_stipple_enable && draw->pipeline.line_stipple)
+         return TRUE;
+
+      /* wide lines */
+      if (rasterizer->line_width > draw->pipeline.wide_line_threshold)
+         return TRUE;
+
+      /* AA lines */
+      if (rasterizer->line_smooth && draw->pipeline.aaline)
+         return TRUE;
+   }
+
+   if (points(prim))
+   {
+      /* large points */
+      if (rasterizer->point_size > draw->pipeline.wide_point_threshold)
+         return TRUE;
+
+      /* AA points */
+      if (rasterizer->point_smooth && draw->pipeline.aapoint)
+         return TRUE;
+
+      /* point sprites */
+      if (rasterizer->point_sprite && draw->pipeline.point_sprite)
+         return TRUE;
+   }
+
+
+   if (triangles(prim)) 
+   {
+      /* polygon stipple */
+      if (rasterizer->poly_stipple_enable && draw->pipeline.pstipple)
+         return TRUE;
+
+      /* unfilled polygons */
+      if (rasterizer->fill_cw != PIPE_POLYGON_MODE_FILL ||
+          rasterizer->fill_ccw != PIPE_POLYGON_MODE_FILL)
+         return TRUE;
+      
+      /* polygon offset */
+      if (rasterizer->offset_cw || rasterizer->offset_ccw)
+         return TRUE;
+
+      /* two-side lighting */
+      if (rasterizer->light_twoside)
+         return TRUE;
+   }
+
+   /* polygon cull - this is difficult - hardware can cull just fine
+    * most of the time (though sometimes CULL_NEITHER is unsupported.
+    * 
+    * Generally this isn't a reason to require the pipeline, though.
+    *
+   if (rasterizer->cull_mode)
+      return TRUE;
+    */
+
+   return FALSE;
+}
+
+
+
+/**
+ * Rebuild the rendering pipeline.
+ */
+static struct draw_stage *validate_pipeline( struct draw_stage *stage )
+{
+   struct draw_context *draw = stage->draw;
+   struct draw_stage *next = draw->pipeline.rasterize;
+   int need_det = 0;
+   int precalc_flat = 0;
+   boolean wide_lines, wide_points;
+
+   /* Set the validate's next stage to the rasterize stage, so that it
+    * can be found later if needed for flushing.
+    */
+   stage->next = next;
+
+   /* drawing wide lines? */
+   wide_lines = (draw->rasterizer->line_width > draw->pipeline.wide_line_threshold
+                 && !draw->rasterizer->line_smooth);
+
+   /* drawing large points? */
+   if (draw->rasterizer->point_sprite && draw->pipeline.point_sprite)
+      wide_points = TRUE;
+   else if (draw->rasterizer->point_smooth && draw->pipeline.aapoint)
+      wide_points = FALSE;
+   else if (draw->rasterizer->point_size > draw->pipeline.wide_point_threshold)
+      wide_points = TRUE;
+   else
+      wide_points = FALSE;
+
+   /*
+    * NOTE: we build up the pipeline in end-to-start order.
+    *
+    * TODO: make the current primitive part of the state and build
+    * shorter pipelines for lines & points.
+    */
+
+   if (draw->rasterizer->line_smooth && draw->pipeline.aaline) {
+      draw->pipeline.aaline->next = next;
+      next = draw->pipeline.aaline;
+   }
+
+   if (draw->rasterizer->point_smooth && draw->pipeline.aapoint) {
+      draw->pipeline.aapoint->next = next;
+      next = draw->pipeline.aapoint;
+   }
+
+   if (wide_lines) {
+      draw->pipeline.wide_line->next = next;
+      next = draw->pipeline.wide_line;
+      precalc_flat = 1;
+   }
+
+   if (wide_points || draw->rasterizer->point_sprite) {
+      draw->pipeline.wide_point->next = next;
+      next = draw->pipeline.wide_point;
+   }
+
+   if (draw->rasterizer->line_stipple_enable && draw->pipeline.line_stipple) {
+      draw->pipeline.stipple->next = next;
+      next = draw->pipeline.stipple;
+      precalc_flat = 1;		/* only needed for lines really */
+   }
+
+   if (draw->rasterizer->poly_stipple_enable
+       && draw->pipeline.pstipple) {
+      draw->pipeline.pstipple->next = next;
+      next = draw->pipeline.pstipple;
+   }
+
+   if (draw->rasterizer->fill_cw != PIPE_POLYGON_MODE_FILL ||
+       draw->rasterizer->fill_ccw != PIPE_POLYGON_MODE_FILL) {
+      draw->pipeline.unfilled->next = next;
+      next = draw->pipeline.unfilled;
+      precalc_flat = 1;		/* only needed for triangles really */
+      need_det = 1;
+   }
+
+   if (draw->rasterizer->flatshade && precalc_flat) {
+      draw->pipeline.flatshade->next = next;
+      next = draw->pipeline.flatshade;
+   }
+	 
+   if (draw->rasterizer->offset_cw ||
+       draw->rasterizer->offset_ccw) {
+      draw->pipeline.offset->next = next;
+      next = draw->pipeline.offset;
+      need_det = 1;
+   }
+
+   if (draw->rasterizer->light_twoside) {
+      draw->pipeline.twoside->next = next;
+      next = draw->pipeline.twoside;
+      need_det = 1;
+   }
+
+   /* Always run the cull stage as we calculate determinant there
+    * also.  
+    *
+    * This can actually be a win as culling out the triangles can lead
+    * to less work emitting vertices, smaller vertex buffers, etc.
+    * It's difficult to say whether this will be true in general.
+    */
+   if (need_det || draw->rasterizer->cull_mode) {
+      draw->pipeline.cull->next = next;
+      next = draw->pipeline.cull;
+   }
+
+   /* Clip stage
+    */
+   if (!draw->bypass_clipping)
+   {
+      draw->pipeline.clip->next = next;
+      next = draw->pipeline.clip;
+   }
+
+   
+   draw->pipeline.first = next;
+   return next;
+}
+
+static void validate_tri( struct draw_stage *stage, 
+			  struct prim_header *header )
+{
+   struct draw_stage *pipeline = validate_pipeline( stage );
+   pipeline->tri( pipeline, header );
+}
+
+static void validate_line( struct draw_stage *stage, 
+			   struct prim_header *header )
+{
+   struct draw_stage *pipeline = validate_pipeline( stage );
+   pipeline->line( pipeline, header );
+}
+
+static void validate_point( struct draw_stage *stage, 
+			    struct prim_header *header )
+{
+   struct draw_stage *pipeline = validate_pipeline( stage );
+   pipeline->point( pipeline, header );
+}
+
+static void validate_reset_stipple_counter( struct draw_stage *stage )
+{
+   struct draw_stage *pipeline = validate_pipeline( stage );
+   pipeline->reset_stipple_counter( pipeline );
+}
+
+static void validate_flush( struct draw_stage *stage, 
+			    unsigned flags )
+{
+   /* May need to pass a backend flush on to the rasterize stage.
+    */
+   if (stage->next)
+      stage->next->flush( stage->next, flags );
+}
+
+
+static void validate_destroy( struct draw_stage *stage )
+{
+   FREE( stage );
+}
+
+
+/**
+ * Create validate pipeline stage.
+ */
+struct draw_stage *draw_validate_stage( struct draw_context *draw )
+{
+   struct draw_stage *stage = CALLOC_STRUCT(draw_stage);
+   if (stage == NULL)
+      return NULL;
+
+   stage->draw = draw;
+   stage->next = NULL;
+   stage->point = validate_point;
+   stage->line = validate_line;
+   stage->tri = validate_tri;
+   stage->flush = validate_flush;
+   stage->reset_stipple_counter = validate_reset_stipple_counter;
+   stage->destroy = validate_destroy;
+
+   return stage;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
new file mode 100644
index 00000000000..9153bc2f865
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
@@ -0,0 +1,497 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Vertex buffer drawing stage.
+ * 
+ * \author Jos� Fonseca <[email protected]>
+ * \author Keith Whitwell <[email protected]>
+ */
+
+
+#include "pipe/p_debug.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
+
+#include "draw_vbuf.h"
+#include "draw_private.h"
+#include "draw_vertex.h"
+#include "draw_pipe.h"
+#include "translate/translate.h"
+#include "translate/translate_cache.h"
+
+
+/**
+ * Vertex buffer emit stage.
+ */
+struct vbuf_stage {
+   struct draw_stage stage; /**< This must be first (base class) */
+
+   struct vbuf_render *render;
+   
+   const struct vertex_info *vinfo;
+   
+   /** Vertex size in bytes */
+   unsigned vertex_size;
+
+   struct translate *translate;
+   
+   /* FIXME: we have no guarantee that 'unsigned' is 32bit */
+
+   /** Vertices in hardware format */
+   unsigned *vertices;
+   unsigned *vertex_ptr;
+   unsigned max_vertices;
+   unsigned nr_vertices;
+   
+   /** Indices */
+   ushort *indices;
+   unsigned max_indices;
+   unsigned nr_indices;
+
+   /* Cache point size somewhere it's address won't change:
+    */
+   float point_size;
+
+   struct translate_cache *cache;
+};
+
+
+/**
+ * Basically a cast wrapper.
+ */
+static INLINE struct vbuf_stage *
+vbuf_stage( struct draw_stage *stage )
+{
+   assert(stage);
+   return (struct vbuf_stage *)stage;
+}
+
+
+static void vbuf_flush_indices( struct vbuf_stage *vbuf );
+static void vbuf_flush_vertices( struct vbuf_stage *vbuf );
+static void vbuf_alloc_vertices( struct vbuf_stage *vbuf );
+
+
+static INLINE boolean 
+overflow( void *map, void *ptr, unsigned bytes, unsigned bufsz )
+{
+   unsigned long used = (unsigned long) ((char *)ptr - (char *)map);
+   return (used + bytes) > bufsz;
+}
+
+
+static INLINE void 
+check_space( struct vbuf_stage *vbuf, unsigned nr )
+{
+   if (vbuf->nr_vertices + nr > vbuf->max_vertices ) {
+      vbuf_flush_vertices(vbuf);
+      vbuf_alloc_vertices(vbuf);
+   }
+
+   if (vbuf->nr_indices + nr > vbuf->max_indices )
+      vbuf_flush_indices(vbuf);
+}
+
+
+
+
+/**
+ * Extract the needed fields from post-transformed vertex and emit
+ * a hardware(driver) vertex.
+ * Recall that the vertices are constructed by the 'draw' module and
+ * have a couple of slots at the beginning (1-dword header, 4-dword
+ * clip pos) that we ignore here.  We only use the vertex->data[] fields.
+ */
+static INLINE ushort 
+emit_vertex( struct vbuf_stage *vbuf,
+             struct vertex_header *vertex )
+{
+   if(vertex->vertex_id == UNDEFINED_VERTEX_ID) {      
+      /* Hmm - vertices are emitted one at a time - better make sure
+       * set_buffer is efficient.  Consider a special one-shot mode for
+       * translate.
+       */
+      /* Note: we really do want data[0] here, not data[pos]: 
+       */
+      vbuf->translate->set_buffer(vbuf->translate, 0, vertex->data[0], 0);
+      vbuf->translate->run(vbuf->translate, 0, 1, vbuf->vertex_ptr);
+
+      if (0) draw_dump_emitted_vertex(vbuf->vinfo, (uint8_t *)vbuf->vertex_ptr);
+      
+      vbuf->vertex_ptr += vbuf->vertex_size/4;
+      vertex->vertex_id = vbuf->nr_vertices++;
+   }
+
+   return (ushort)vertex->vertex_id;
+}
+
+
+static void 
+vbuf_tri( struct draw_stage *stage,
+          struct prim_header *prim )
+{
+   struct vbuf_stage *vbuf = vbuf_stage( stage );
+   unsigned i;
+
+   check_space( vbuf, 3 );
+
+   for (i = 0; i < 3; i++) {
+      vbuf->indices[vbuf->nr_indices++] = emit_vertex( vbuf, prim->v[i] );
+   }
+}
+
+
+static void 
+vbuf_line( struct draw_stage *stage, 
+           struct prim_header *prim )
+{
+   struct vbuf_stage *vbuf = vbuf_stage( stage );
+   unsigned i;
+
+   check_space( vbuf, 2 );
+
+   for (i = 0; i < 2; i++) {
+      vbuf->indices[vbuf->nr_indices++] = emit_vertex( vbuf, prim->v[i] );
+   }   
+}
+
+
+static void 
+vbuf_point( struct draw_stage *stage, 
+            struct prim_header *prim )
+{
+   struct vbuf_stage *vbuf = vbuf_stage( stage );
+
+   check_space( vbuf, 1 );
+
+   vbuf->indices[vbuf->nr_indices++] = emit_vertex( vbuf, prim->v[0] );
+}
+
+
+
+
+/**
+ * Set the prim type for subsequent vertices.
+ * This may result in a new vertex size.  The existing vbuffer (if any)
+ * will be flushed if needed and a new one allocated.
+ */
+static void
+vbuf_set_prim( struct vbuf_stage *vbuf, uint prim )
+{
+   struct translate_key hw_key;
+   unsigned dst_offset;
+   unsigned i;
+
+   vbuf->render->set_primitive(vbuf->render, prim);
+
+   /* Must do this after set_primitive() above:
+    * 
+    * XXX: need some state managment to track when this needs to be
+    * recalculated.  The driver should tell us whether there was a
+    * state change.
+    */
+   vbuf->vinfo = vbuf->render->get_vertex_info(vbuf->render);
+
+   if (vbuf->vertex_size != vbuf->vinfo->size * sizeof(float)) {
+      vbuf_flush_vertices(vbuf);
+      vbuf->vertex_size = vbuf->vinfo->size * sizeof(float);
+   }
+
+   /* Translate from pipeline vertices to hw vertices.
+    */
+   dst_offset = 0;
+
+   for (i = 0; i < vbuf->vinfo->num_attribs; i++) {
+      unsigned emit_sz = 0;
+      unsigned src_buffer = 0;
+      unsigned output_format;
+      unsigned src_offset = (vbuf->vinfo->attrib[i].src_index * 4 * sizeof(float) );
+
+      switch (vbuf->vinfo->attrib[i].emit) {
+      case EMIT_4F:
+	 output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+	 emit_sz = 4 * sizeof(float);
+	 break;
+      case EMIT_3F:
+	 output_format = PIPE_FORMAT_R32G32B32_FLOAT;
+	 emit_sz = 3 * sizeof(float);
+	 break;
+      case EMIT_2F:
+	 output_format = PIPE_FORMAT_R32G32_FLOAT;
+	 emit_sz = 2 * sizeof(float);
+	 break;
+      case EMIT_1F:
+	 output_format = PIPE_FORMAT_R32_FLOAT;
+	 emit_sz = 1 * sizeof(float);
+	 break;
+      case EMIT_1F_PSIZE:
+	 output_format = PIPE_FORMAT_R32_FLOAT;
+	 emit_sz = 1 * sizeof(float);
+	 src_buffer = 1;
+	 src_offset = 0;
+	 break;
+      case EMIT_4UB:
+	 output_format = PIPE_FORMAT_B8G8R8A8_UNORM;
+	 emit_sz = 4 * sizeof(ubyte);
+         break;
+      default:
+	 assert(0);
+	 output_format = PIPE_FORMAT_NONE;
+	 emit_sz = 0;
+	 break;
+      }
+      
+      hw_key.element[i].input_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+      hw_key.element[i].input_buffer = src_buffer;
+      hw_key.element[i].input_offset = src_offset;
+      hw_key.element[i].output_format = output_format;
+      hw_key.element[i].output_offset = dst_offset;
+
+      dst_offset += emit_sz;
+   }
+
+   hw_key.nr_elements = vbuf->vinfo->num_attribs;
+   hw_key.output_stride = vbuf->vinfo->size * 4;
+
+   /* Don't bother with caching at this stage:
+    */
+   if (!vbuf->translate ||
+       translate_key_compare(&vbuf->translate->key, &hw_key) != 0) 
+   {
+      translate_key_sanitize(&hw_key);
+      vbuf->translate = translate_cache_find(vbuf->cache, &hw_key);
+
+      vbuf->translate->set_buffer(vbuf->translate, 1, &vbuf->point_size, 0);
+   }
+
+   vbuf->point_size = vbuf->stage.draw->rasterizer->point_size;
+
+   /* Allocate new buffer?
+    */
+   if (!vbuf->vertices)
+      vbuf_alloc_vertices(vbuf);
+}
+
+
+static void 
+vbuf_first_tri( struct draw_stage *stage,
+                struct prim_header *prim )
+{
+   struct vbuf_stage *vbuf = vbuf_stage( stage );
+
+   vbuf_flush_indices( vbuf );   
+   stage->tri = vbuf_tri;
+   vbuf_set_prim(vbuf, PIPE_PRIM_TRIANGLES);
+   stage->tri( stage, prim );
+}
+
+
+static void 
+vbuf_first_line( struct draw_stage *stage,
+                 struct prim_header *prim )
+{
+   struct vbuf_stage *vbuf = vbuf_stage( stage );
+
+   vbuf_flush_indices( vbuf );
+   stage->line = vbuf_line;
+   vbuf_set_prim(vbuf, PIPE_PRIM_LINES);
+   stage->line( stage, prim );
+}
+
+
+static void 
+vbuf_first_point( struct draw_stage *stage,
+                  struct prim_header *prim )
+{
+   struct vbuf_stage *vbuf = vbuf_stage( stage );
+
+   vbuf_flush_indices( vbuf );
+   stage->point = vbuf_point;
+   vbuf_set_prim(vbuf, PIPE_PRIM_POINTS);
+   stage->point( stage, prim );
+}
+
+
+static void 
+vbuf_flush_indices( struct vbuf_stage *vbuf ) 
+{
+   if(!vbuf->nr_indices)
+      return;
+   
+   assert((uint) (vbuf->vertex_ptr - vbuf->vertices) == 
+          vbuf->nr_vertices * vbuf->vertex_size / sizeof(unsigned));
+
+   vbuf->render->draw(vbuf->render, vbuf->indices, vbuf->nr_indices);
+   
+   vbuf->nr_indices = 0;
+}
+
+
+/**
+ * Flush existing vertex buffer and allocate a new one.
+ * 
+ * XXX: We separate flush-on-index-full and flush-on-vb-full, but may 
+ * raise issues uploading vertices if the hardware wants to flush when
+ * we flush.
+ */
+static void 
+vbuf_flush_vertices( struct vbuf_stage *vbuf )
+{
+   if(vbuf->vertices) {      
+      vbuf_flush_indices(vbuf);
+      
+      /* Reset temporary vertices ids */
+      if(vbuf->nr_vertices)
+	 draw_reset_vertex_ids( vbuf->stage.draw );
+      
+      /* Free the vertex buffer */
+      vbuf->render->release_vertices(vbuf->render,
+                                     vbuf->vertices,
+                                     vbuf->vertex_size,
+                                     vbuf->nr_vertices);
+      vbuf->max_vertices = vbuf->nr_vertices = 0;
+      vbuf->vertex_ptr = vbuf->vertices = NULL;
+      
+   }
+}
+   
+
+static void 
+vbuf_alloc_vertices( struct vbuf_stage *vbuf )
+{
+   assert(!vbuf->nr_indices);
+   assert(!vbuf->vertices);
+   
+   /* Allocate a new vertex buffer */
+   vbuf->max_vertices = vbuf->render->max_vertex_buffer_bytes / vbuf->vertex_size;
+
+   /* even number */
+   vbuf->max_vertices = vbuf->max_vertices & ~1;
+
+   if(vbuf->max_vertices >= UNDEFINED_VERTEX_ID)
+      vbuf->max_vertices = UNDEFINED_VERTEX_ID - 1;
+
+   /* Must always succeed -- driver gives us a
+    * 'max_vertex_buffer_bytes' which it guarantees it can allocate,
+    * and it will flush itself if necessary to do so.  If this does
+    * fail, we are basically without usable hardware.
+    */
+   vbuf->vertices = (uint *) vbuf->render->allocate_vertices(vbuf->render,
+							     (ushort) vbuf->vertex_size,
+							     (ushort) vbuf->max_vertices);
+   vbuf->vertex_ptr = vbuf->vertices;
+}
+
+
+
+static void 
+vbuf_flush( struct draw_stage *stage, unsigned flags )
+{
+   struct vbuf_stage *vbuf = vbuf_stage( stage );
+
+   vbuf_flush_indices( vbuf );
+
+   stage->point = vbuf_first_point;
+   stage->line = vbuf_first_line;
+   stage->tri = vbuf_first_tri;
+
+   if (flags & DRAW_FLUSH_BACKEND)
+      vbuf_flush_vertices( vbuf );
+}
+
+
+static void 
+vbuf_reset_stipple_counter( struct draw_stage *stage )
+{
+   /* XXX: Need to do something here for hardware with linestipple.
+    */
+   (void) stage;
+}
+
+
+static void vbuf_destroy( struct draw_stage *stage )
+{
+   struct vbuf_stage *vbuf = vbuf_stage( stage );
+
+   if(vbuf->indices)
+      align_free( vbuf->indices );
+   
+   if (vbuf->render)
+      vbuf->render->destroy( vbuf->render );
+
+   if (vbuf->cache)
+      translate_cache_destroy(vbuf->cache);
+
+   FREE( stage );
+}
+
+
+/**
+ * Create a new primitive vbuf/render stage.
+ */
+struct draw_stage *draw_vbuf_stage( struct draw_context *draw,
+                                    struct vbuf_render *render )
+{
+   struct vbuf_stage *vbuf = CALLOC_STRUCT(vbuf_stage);
+   if (vbuf == NULL)
+      goto fail;
+   
+   vbuf->stage.draw = draw;
+   vbuf->stage.point = vbuf_first_point;
+   vbuf->stage.line = vbuf_first_line;
+   vbuf->stage.tri = vbuf_first_tri;
+   vbuf->stage.flush = vbuf_flush;
+   vbuf->stage.reset_stipple_counter = vbuf_reset_stipple_counter;
+   vbuf->stage.destroy = vbuf_destroy;
+   
+   vbuf->render = render;
+   vbuf->max_indices = MAX2(render->max_indices, UNDEFINED_VERTEX_ID-1);
+
+   vbuf->indices = (ushort *) align_malloc( vbuf->max_indices * 
+					    sizeof(vbuf->indices[0]), 
+					    16 );
+   if (!vbuf->indices)
+      goto fail;
+
+   vbuf->cache = translate_cache_create();
+   if (!vbuf->cache) 
+      goto fail;
+      
+   
+   vbuf->vertices = NULL;
+   vbuf->vertex_ptr = vbuf->vertices;
+   
+   return &vbuf->stage;
+
+ fail:
+   if (vbuf)
+      vbuf_destroy(&vbuf->stage);
+   
+   return NULL;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pipe_wide_line.c b/src/gallium/auxiliary/draw/draw_pipe_wide_line.c
new file mode 100644
index 00000000000..184e363594d
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pipe_wide_line.c
@@ -0,0 +1,180 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <[email protected]>
+ */
+
+#include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "draw_private.h"
+#include "draw_pipe.h"
+
+
+struct wideline_stage {
+   struct draw_stage stage;
+
+   float half_line_width;
+};
+
+
+
+static INLINE struct wideline_stage *wideline_stage( struct draw_stage *stage )
+{
+   return (struct wideline_stage *)stage;
+}
+
+
+
+/**
+ * Draw a wide line by drawing a quad (two triangles).
+ * XXX need to disable polygon stipple.
+ */
+static void wideline_line( struct draw_stage *stage,
+                           struct prim_header *header )
+{
+   /*const struct wideline_stage *wide = wideline_stage(stage);*/
+   const unsigned pos = stage->draw->vs.position_output;
+   const float half_width = 0.5f * stage->draw->rasterizer->line_width;
+
+   struct prim_header tri;
+
+   struct vertex_header *v0 = dup_vert(stage, header->v[0], 0);
+   struct vertex_header *v1 = dup_vert(stage, header->v[0], 1);
+   struct vertex_header *v2 = dup_vert(stage, header->v[1], 2);
+   struct vertex_header *v3 = dup_vert(stage, header->v[1], 3);
+
+   float *pos0 = v0->data[pos];
+   float *pos1 = v1->data[pos];
+   float *pos2 = v2->data[pos];
+   float *pos3 = v3->data[pos];
+
+   const float dx = fabsf(pos0[0] - pos2[0]);
+   const float dy = fabsf(pos0[1] - pos2[1]);
+
+   /* small tweak to meet GL specification */
+   const float bias = 0.125f;
+
+   /*
+    * Draw wide line as a quad (two tris) by "stretching" the line along
+    * X or Y.
+    * We need to tweak coords in several ways to be conformant here.
+    */
+
+   if (dx > dy) {
+      /* x-major line */
+      pos0[1] = pos0[1] - half_width - bias;
+      pos1[1] = pos1[1] + half_width - bias;
+      pos2[1] = pos2[1] - half_width - bias;
+      pos3[1] = pos3[1] + half_width - bias;
+      if (pos0[0] < pos2[0]) {
+         /* left to right line */
+         pos0[0] -= 0.5f;
+         pos1[0] -= 0.5f;
+         pos2[0] -= 0.5f;
+         pos3[0] -= 0.5f;
+      }
+      else {
+         /* right to left line */
+         pos0[0] += 0.5f;
+         pos1[0] += 0.5f;
+         pos2[0] += 0.5f;
+         pos3[0] += 0.5f;
+      }
+   }
+   else {
+      /* y-major line */
+      pos0[0] = pos0[0] - half_width + bias;
+      pos1[0] = pos1[0] + half_width + bias;
+      pos2[0] = pos2[0] - half_width + bias;
+      pos3[0] = pos3[0] + half_width + bias;
+      if (pos0[1] < pos2[1]) {
+         /* top to bottom line */
+         pos0[1] -= 0.5f;
+         pos1[1] -= 0.5f;
+         pos2[1] -= 0.5f;
+         pos3[1] -= 0.5f;
+      }
+      else {
+         /* bottom to top line */
+         pos0[1] += 0.5f;
+         pos1[1] += 0.5f;
+         pos2[1] += 0.5f;
+         pos3[1] += 0.5f;
+      }
+   }
+
+   tri.det = header->det;  /* only the sign matters */
+   tri.v[0] = v0;
+   tri.v[1] = v2;
+   tri.v[2] = v3;
+   stage->next->tri( stage->next, &tri );
+
+   tri.v[0] = v0;
+   tri.v[1] = v3;
+   tri.v[2] = v1;
+   stage->next->tri( stage->next, &tri );
+}
+
+
+static void wideline_flush( struct draw_stage *stage, unsigned flags )
+{
+   stage->next->flush( stage->next, flags );
+}
+
+
+static void wideline_reset_stipple_counter( struct draw_stage *stage )
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void wideline_destroy( struct draw_stage *stage )
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+struct draw_stage *draw_wide_line_stage( struct draw_context *draw )
+{
+   struct wideline_stage *wide = CALLOC_STRUCT(wideline_stage);
+
+   draw_alloc_temp_verts( &wide->stage, 4 );
+
+   wide->stage.draw = draw;
+   wide->stage.next = NULL;
+   wide->stage.point = draw_pipe_passthrough_point;
+   wide->stage.line = wideline_line;
+   wide->stage.tri = draw_pipe_passthrough_tri;
+   wide->stage.flush = wideline_flush;
+   wide->stage.reset_stipple_counter = wideline_reset_stipple_counter;
+   wide->stage.destroy = wideline_destroy;
+
+   return &wide->stage;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pipe_wide_point.c b/src/gallium/auxiliary/draw/draw_pipe_wide_point.c
new file mode 100644
index 00000000000..e1af9e56a24
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pipe_wide_point.c
@@ -0,0 +1,297 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <[email protected]>
+ */
+
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_shader_tokens.h"
+#include "draw_vs.h"
+#include "draw_pipe.h"
+
+
+struct widepoint_stage {
+   struct draw_stage stage;
+
+   float half_point_size;
+   float point_size_min;
+   float point_size_max;
+
+   float xbias;
+   float ybias;
+
+   uint texcoord_slot[PIPE_MAX_SHADER_OUTPUTS];
+   uint texcoord_mode[PIPE_MAX_SHADER_OUTPUTS];
+   uint num_texcoords;
+
+   int psize_slot;
+
+   int point_coord_fs_input;  /**< input for pointcoord (and fog) */
+};
+
+
+
+static INLINE struct widepoint_stage *
+widepoint_stage( struct draw_stage *stage )
+{
+   return (struct widepoint_stage *)stage;
+}
+
+
+
+
+/**
+ * Set the vertex texcoords for sprite mode.
+ * Coords may be left untouched or set to a right-side-up or upside-down
+ * orientation.
+ */
+static void set_texcoords(const struct widepoint_stage *wide,
+                          struct vertex_header *v, const float tc[4])
+{
+   uint i;
+   for (i = 0; i < wide->num_texcoords; i++) {
+      if (wide->texcoord_mode[i] != PIPE_SPRITE_COORD_NONE) {
+         uint j = wide->texcoord_slot[i];
+         v->data[j][0] = tc[0];
+         if (wide->texcoord_mode[i] == PIPE_SPRITE_COORD_LOWER_LEFT)
+            v->data[j][1] = 1.0f - tc[1];
+         else
+            v->data[j][1] = tc[1];
+         v->data[j][2] = tc[2];
+         v->data[j][3] = tc[3];
+      }
+   }
+
+   if (wide->point_coord_fs_input >= 0) {
+      /* put gl_PointCoord into extra vertex output's zw components */
+      uint k = wide->stage.draw->extra_vp_outputs.slot;
+      v->data[k][2] = tc[0];
+      v->data[k][3] = tc[1];
+   }
+}
+
+
+/* If there are lots of sprite points (and why wouldn't there be?) it
+ * would probably be more sensible to change hardware setup to
+ * optimize this rather than doing the whole thing in software like
+ * this.
+ */
+static void widepoint_point( struct draw_stage *stage,
+                             struct prim_header *header )
+{
+   const struct widepoint_stage *wide = widepoint_stage(stage);
+   const unsigned pos = stage->draw->vs.position_output;
+   const boolean sprite = (boolean) stage->draw->rasterizer->point_sprite;
+   float half_size;
+   float left_adj, right_adj, bot_adj, top_adj;
+
+   struct prim_header tri;
+
+   /* four dups of original vertex */
+   struct vertex_header *v0 = dup_vert(stage, header->v[0], 0);
+   struct vertex_header *v1 = dup_vert(stage, header->v[0], 1);
+   struct vertex_header *v2 = dup_vert(stage, header->v[0], 2);
+   struct vertex_header *v3 = dup_vert(stage, header->v[0], 3);
+
+   float *pos0 = v0->data[pos];
+   float *pos1 = v1->data[pos];
+   float *pos2 = v2->data[pos];
+   float *pos3 = v3->data[pos];
+
+   /* point size is either per-vertex or fixed size */
+   if (wide->psize_slot >= 0) {
+      half_size = header->v[0]->data[wide->psize_slot][0];
+
+      /* XXX: temporary -- do this in the vertex shader??
+       */
+      half_size = CLAMP(half_size,
+                        wide->point_size_min,
+                        wide->point_size_max);
+      
+      half_size *= 0.5f; 
+   }
+   else {
+      half_size = wide->half_point_size;
+   }
+
+   left_adj = -half_size + wide->xbias;
+   right_adj = half_size + wide->xbias;
+   bot_adj = half_size + wide->ybias;
+   top_adj = -half_size + wide->ybias;
+
+   pos0[0] += left_adj;
+   pos0[1] += top_adj;
+
+   pos1[0] += left_adj;
+   pos1[1] += bot_adj;
+
+   pos2[0] += right_adj;
+   pos2[1] += top_adj;
+
+   pos3[0] += right_adj;
+   pos3[1] += bot_adj;
+
+   if (sprite) {
+      static const float tex00[4] = { 0, 0, 0, 1 };
+      static const float tex01[4] = { 0, 1, 0, 1 };
+      static const float tex11[4] = { 1, 1, 0, 1 };
+      static const float tex10[4] = { 1, 0, 0, 1 };
+      set_texcoords( wide, v0, tex00 );
+      set_texcoords( wide, v1, tex01 );
+      set_texcoords( wide, v2, tex10 );
+      set_texcoords( wide, v3, tex11 );
+   }
+
+   tri.det = header->det;  /* only the sign matters */
+   tri.v[0] = v0;
+   tri.v[1] = v2;
+   tri.v[2] = v3;
+   stage->next->tri( stage->next, &tri );
+
+   tri.v[0] = v0;
+   tri.v[1] = v3;
+   tri.v[2] = v1;
+   stage->next->tri( stage->next, &tri );
+}
+
+
+static void widepoint_first_point( struct draw_stage *stage, 
+			      struct prim_header *header )
+{
+   struct widepoint_stage *wide = widepoint_stage(stage);
+   struct draw_context *draw = stage->draw;
+
+   wide->half_point_size = 0.5f * draw->rasterizer->point_size;
+   wide->point_size_min = draw->rasterizer->point_size_min;
+   wide->point_size_max = draw->rasterizer->point_size_max;
+   wide->xbias = 0.0;
+   wide->ybias = 0.0;
+
+   if (draw->rasterizer->gl_rasterization_rules) {
+      wide->xbias = 0.125;
+   }
+
+   /* XXX we won't know the real size if it's computed by the vertex shader! */
+   if ((draw->rasterizer->point_size > draw->pipeline.wide_point_threshold) ||
+       (draw->rasterizer->point_sprite && draw->pipeline.point_sprite)) {
+      stage->point = widepoint_point;
+   }
+   else {
+      stage->point = draw_pipe_passthrough_point;
+   }
+
+   if (draw->rasterizer->point_sprite) {
+      /* find vertex shader texcoord outputs */
+      const struct draw_vertex_shader *vs = draw->vs.vertex_shader;
+      uint i, j = 0;
+      for (i = 0; i < vs->info.num_outputs; i++) {
+         if (vs->info.output_semantic_name[i] == TGSI_SEMANTIC_GENERIC) {
+            wide->texcoord_slot[j] = i;
+            wide->texcoord_mode[j] = draw->rasterizer->sprite_coord_mode[j];
+            j++;
+         }
+      }
+      wide->num_texcoords = j;
+
+      /* find fragment shader PointCoord/Fog input */
+      wide->point_coord_fs_input = 0; /* XXX fix this! */
+
+      /* setup extra vp output (point coord implemented as a texcoord) */
+      draw->extra_vp_outputs.semantic_name = TGSI_SEMANTIC_GENERIC;
+      draw->extra_vp_outputs.semantic_index = 0;
+      draw->extra_vp_outputs.slot = draw->vs.num_vs_outputs;
+   }
+   else {
+      wide->point_coord_fs_input = -1;
+      draw->extra_vp_outputs.slot = 0;
+   }
+
+   wide->psize_slot = -1;
+   if (draw->rasterizer->point_size_per_vertex) {
+      /* find PSIZ vertex output */
+      const struct draw_vertex_shader *vs = draw->vs.vertex_shader;
+      uint i;
+      for (i = 0; i < vs->info.num_outputs; i++) {
+         if (vs->info.output_semantic_name[i] == TGSI_SEMANTIC_PSIZE) {
+            wide->psize_slot = i;
+            break;
+         }
+      }
+   }
+   
+   stage->point( stage, header );
+}
+
+
+static void widepoint_flush( struct draw_stage *stage, unsigned flags )
+{
+   stage->point = widepoint_first_point;
+   stage->next->flush( stage->next, flags );
+}
+
+
+static void widepoint_reset_stipple_counter( struct draw_stage *stage )
+{
+   stage->next->reset_stipple_counter( stage->next );
+}
+
+
+static void widepoint_destroy( struct draw_stage *stage )
+{
+   draw_free_temp_verts( stage );
+   FREE( stage );
+}
+
+
+struct draw_stage *draw_wide_point_stage( struct draw_context *draw )
+{
+   struct widepoint_stage *wide = CALLOC_STRUCT(widepoint_stage);
+   if (wide == NULL)
+      goto fail;
+
+   if (!draw_alloc_temp_verts( &wide->stage, 4 ))
+      goto fail;
+
+   wide->stage.draw = draw;
+   wide->stage.next = NULL;
+   wide->stage.point = widepoint_first_point;
+   wide->stage.line = draw_pipe_passthrough_line;
+   wide->stage.tri = draw_pipe_passthrough_tri;
+   wide->stage.flush = widepoint_flush;
+   wide->stage.reset_stipple_counter = widepoint_reset_stipple_counter;
+   wide->stage.destroy = widepoint_destroy;
+
+   return &wide->stage;
+
+ fail:
+   if (wide)
+      wide->stage.destroy( &wide->stage );
+   
+   return NULL;
+}
diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
new file mode 100644
index 00000000000..81e4eae401c
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -0,0 +1,318 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * Private data structures, etc for the draw module.
+ */
+
+
+/**
+ * Authors:
+ * Keith Whitwell <[email protected]>
+ * Brian Paul
+ */
+
+
+#ifndef DRAW_PRIVATE_H
+#define DRAW_PRIVATE_H
+
+
+#include "pipe/p_state.h"
+#include "pipe/p_defines.h"
+
+#include "tgsi/tgsi_exec.h"
+#include "tgsi/tgsi_scan.h"
+
+
+struct pipe_context;
+struct gallivm_prog;
+struct gallivm_cpu_engine;
+struct draw_vertex_shader;
+struct draw_context;
+struct draw_stage;
+struct vbuf_render;
+
+
+/**
+ * Basic vertex info.
+ * Carry some useful information around with the vertices in the prim pipe.  
+ */
+struct vertex_header {
+   unsigned clipmask:12;
+   unsigned edgeflag:1;
+   unsigned pad:3;
+   unsigned vertex_id:16;
+
+   float clip[4];
+
+   /* This will probably become float (*data)[4] soon:
+    */
+   float data[][4];
+};
+
+/* NOTE: It should match vertex_id size above */
+#define UNDEFINED_VERTEX_ID 0xffff
+
+
+/**
+ * Private context for the drawing module.
+ */
+struct draw_context
+{
+   /** Drawing/primitive pipeline stages */
+   struct {
+      struct draw_stage *first;  /**< one of the following */
+
+      struct draw_stage *validate; 
+
+      /* stages (in logical order) */
+      struct draw_stage *flatshade;
+      struct draw_stage *clip;
+      struct draw_stage *cull;
+      struct draw_stage *twoside;
+      struct draw_stage *offset;
+      struct draw_stage *unfilled;
+      struct draw_stage *stipple;
+      struct draw_stage *aapoint;
+      struct draw_stage *aaline;
+      struct draw_stage *pstipple;
+      struct draw_stage *wide_line;
+      struct draw_stage *wide_point;
+      struct draw_stage *rasterize;
+
+      float wide_point_threshold; /**< convert pnts to tris if larger than this */
+      float wide_line_threshold;  /**< convert lines to tris if wider than this */
+      boolean line_stipple;       /**< do line stipple? */
+      boolean point_sprite;       /**< convert points to quads for sprites? */
+
+      /* Temporary storage while the pipeline is being run:
+       */
+      char *verts;
+      unsigned vertex_stride;
+      unsigned vertex_count;
+   } pipeline;
+
+
+   struct vbuf_render *render;
+
+   /* Support prototype passthrough path:
+    */
+   struct {
+      struct {
+         struct draw_pt_middle_end *fetch_emit;
+         struct draw_pt_middle_end *fetch_shade_emit;
+         struct draw_pt_middle_end *general;
+      } middle;
+
+      struct {
+         struct draw_pt_front_end *vcache;
+         struct draw_pt_front_end *varray;
+      } front;
+
+      struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
+      unsigned nr_vertex_buffers;
+
+      struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS];
+      unsigned nr_vertex_elements;
+
+      /* user-space vertex data, buffers */
+      struct {
+         const unsigned *edgeflag;
+
+         /** vertex element/index buffer (ex: glDrawElements) */
+         const void *elts;
+         /** bytes per index (0, 1, 2 or 4) */
+         unsigned eltSize;
+         unsigned min_index;
+         unsigned max_index;
+         
+         /** vertex arrays */
+         const void *vbuffer[PIPE_MAX_ATTRIBS];
+         
+         /** constant buffer (for vertex shader) */
+         const void *constants;
+      } user;
+
+      boolean test_fse;         /* enable FSE even though its not correct (eg for softpipe) */
+      boolean no_fse;           /* disable FSE even when it is correct */
+   } pt;
+
+   struct {
+      boolean bypass_clipping;
+      boolean bypass_vs;
+   } driver;
+
+   boolean flushing;         /**< debugging/sanity */
+   boolean suspend_flushing; /**< internally set */
+   boolean bypass_clipping;  /**< set if either api or driver bypass_clipping true */
+
+   boolean force_passthrough; /**< never clip or shade */
+
+   double mrd;  /**< minimum resolvable depth value, for polygon offset */
+
+   /* pipe state that we need: */
+   const struct pipe_rasterizer_state *rasterizer;
+   struct pipe_viewport_state viewport;
+   boolean identity_viewport;
+
+   struct {
+      struct draw_vertex_shader *vertex_shader;
+      uint num_vs_outputs;  /**< convenience, from vertex_shader */
+      uint position_output;
+
+      /** TGSI program interpreter runtime state */
+      struct tgsi_exec_machine machine;
+
+      uint num_samplers;
+      struct tgsi_sampler **samplers;
+
+      /* This (and the tgsi_exec_machine struct) probably need to be moved somewhere private.
+       */
+      struct gallivm_cpu_engine *engine;   
+
+      /* Here's another one:
+       */
+      struct aos_machine *aos_machine; 
+
+
+      const float (*aligned_constants)[4];
+
+      const float (*aligned_constant_storage)[4];
+      unsigned const_storage_size;
+
+
+      struct translate *fetch;
+      struct translate_cache *fetch_cache;
+      struct translate *emit;
+      struct translate_cache *emit_cache;
+   } vs;
+
+   /* Clip derived state:
+    */
+   float plane[12][4];
+   unsigned nr_planes;
+
+   /* If a prim stage introduces new vertex attributes, they'll be stored here
+    */
+   struct {
+      uint semantic_name;
+      uint semantic_index;
+      int slot;
+   } extra_vp_outputs;
+
+   unsigned reduced_prim;
+
+   void *driver_private;
+};
+
+
+/*******************************************************************************
+ * Vertex shader code:
+ */
+boolean draw_vs_init( struct draw_context *draw );
+void draw_vs_destroy( struct draw_context *draw );
+
+void draw_vs_set_viewport( struct draw_context *, 
+                           const struct pipe_viewport_state * );
+
+void draw_vs_set_constants( struct draw_context *,
+                            const float (*constants)[4],
+                            unsigned size );
+
+
+
+
+/*******************************************************************************
+ * Vertex processing (was passthrough) code:
+ */
+boolean draw_pt_init( struct draw_context *draw );
+void draw_pt_destroy( struct draw_context *draw );
+void draw_pt_reset_vertex_ids( struct draw_context *draw );
+
+
+/*******************************************************************************
+ * Primitive processing (pipeline) code: 
+ */
+
+boolean draw_pipeline_init( struct draw_context *draw );
+void draw_pipeline_destroy( struct draw_context *draw );
+
+
+
+
+
+/* We use the top few bits in the elts[] parameter to convey a little
+ * API information.  This limits the number of vertices we can address
+ * to only 4096 -- if that becomes a problem, we can switch to 32-bit
+ * draw indices.
+ *
+ * These flags expected at first vertex of lines & triangles when
+ * unfilled and/or line stipple modes are operational.
+ */
+#define DRAW_PIPE_MAX_VERTICES  (0x1<<12)
+#define DRAW_PIPE_EDGE_FLAG_0   (0x1<<12)
+#define DRAW_PIPE_EDGE_FLAG_1   (0x2<<12)
+#define DRAW_PIPE_EDGE_FLAG_2   (0x4<<12)
+#define DRAW_PIPE_EDGE_FLAG_ALL (0x7<<12)
+#define DRAW_PIPE_RESET_STIPPLE (0x8<<12)
+#define DRAW_PIPE_FLAG_MASK     (0xf<<12)
+
+void draw_pipeline_run( struct draw_context *draw,
+                        unsigned prim,
+                        struct vertex_header *vertices,
+                        unsigned vertex_count,
+                        unsigned stride,
+                        const ushort *elts,
+                        unsigned count );
+
+void draw_pipeline_run_linear( struct draw_context *draw,
+                               unsigned prim,
+                               struct vertex_header *vertices,
+                               unsigned count,
+                               unsigned stride );
+
+
+
+void draw_pipeline_flush( struct draw_context *draw, 
+                          unsigned flags );
+
+
+
+/*******************************************************************************
+ * Flushing 
+ */
+
+#define DRAW_FLUSH_STATE_CHANGE              0x8
+#define DRAW_FLUSH_BACKEND                   0x10
+
+
+void draw_do_flush( struct draw_context *draw, unsigned flags );
+
+
+
+
+#endif /* DRAW_PRIVATE_H */
diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c
new file mode 100644
index 00000000000..4e5ffa09305
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pt.c
@@ -0,0 +1,323 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  */
+
+#include "draw/draw_context.h"
+#include "draw/draw_private.h"
+#include "draw/draw_pt.h"
+#include "draw/draw_vs.h"
+#include "tgsi/tgsi_dump.h"
+#include "util/u_math.h"
+
+static unsigned trim( unsigned count, unsigned first, unsigned incr )
+{
+   if (count < first)
+      return 0;
+   return count - (count - first) % incr; 
+}
+
+
+
+/* Overall we split things into:
+ *     - frontend -- prepare fetch_elts, draw_elts - eg vcache
+ *     - middle   -- fetch, shade, cliptest, viewport
+ *     - pipeline -- the prim pipeline: clipping, wide lines, etc 
+ *     - backend  -- the vbuf_render provided by the driver.
+ */
+static boolean
+draw_pt_arrays(struct draw_context *draw, 
+               unsigned prim,
+               unsigned start, 
+               unsigned count)
+{
+   struct draw_pt_front_end *frontend = NULL;
+   struct draw_pt_middle_end *middle = NULL;
+   unsigned opt = 0;
+
+   /* Sanitize primitive length:
+    */
+   {
+      unsigned first, incr;
+      draw_pt_split_prim(prim, &first, &incr);
+      count = trim(count, first, incr); 
+      if (count < first)
+         return TRUE;
+   }
+
+   if (!draw->force_passthrough) {
+      if (!draw->render) {
+         opt |= PT_PIPELINE;
+      }
+      
+      if (draw_need_pipeline(draw,
+                             draw->rasterizer,
+                             prim)) {
+         opt |= PT_PIPELINE;
+      }
+
+      if (!draw->bypass_clipping && !draw->pt.test_fse) {
+         opt |= PT_CLIPTEST;
+      }
+      
+      if (!draw->rasterizer->bypass_vs) {
+         opt |= PT_SHADE;
+      }
+   }
+      
+   if (opt == 0) 
+      middle = draw->pt.middle.fetch_emit;
+   else if (opt == PT_SHADE && !draw->pt.no_fse)
+      middle = draw->pt.middle.fetch_shade_emit;
+   else
+      middle = draw->pt.middle.general;
+
+
+   /* Pick the right frontend
+    */
+   if (draw->pt.user.elts || (opt & PT_PIPELINE)) {
+      frontend = draw->pt.front.vcache;
+   } else {
+      frontend = draw->pt.front.varray;
+   }
+
+   frontend->prepare( frontend, prim, middle, opt );
+
+   frontend->run(frontend, 
+                 draw_pt_elt_func(draw),
+                 draw_pt_elt_ptr(draw, start),
+                 count);
+
+   frontend->finish( frontend );
+
+   return TRUE;
+}
+
+
+boolean draw_pt_init( struct draw_context *draw )
+{
+   draw->pt.test_fse = debug_get_bool_option("DRAW_FSE", FALSE);
+   draw->pt.no_fse = debug_get_bool_option("DRAW_NO_FSE", FALSE);
+
+   draw->pt.front.vcache = draw_pt_vcache( draw );
+   if (!draw->pt.front.vcache)
+      return FALSE;
+
+   draw->pt.front.varray = draw_pt_varray(draw);
+   if (!draw->pt.front.varray)
+      return FALSE;
+
+   draw->pt.middle.fetch_emit = draw_pt_fetch_emit( draw );
+   if (!draw->pt.middle.fetch_emit)
+      return FALSE;
+
+   draw->pt.middle.fetch_shade_emit = draw_pt_middle_fse( draw );
+   if (!draw->pt.middle.fetch_shade_emit)
+      return FALSE;
+
+   draw->pt.middle.general = draw_pt_fetch_pipeline_or_emit( draw );
+   if (!draw->pt.middle.general)
+      return FALSE;
+
+   return TRUE;
+}
+
+
+void draw_pt_destroy( struct draw_context *draw )
+{
+   if (draw->pt.middle.general) {
+      draw->pt.middle.general->destroy( draw->pt.middle.general );
+      draw->pt.middle.general = NULL;
+   }
+
+   if (draw->pt.middle.fetch_emit) {
+      draw->pt.middle.fetch_emit->destroy( draw->pt.middle.fetch_emit );
+      draw->pt.middle.fetch_emit = NULL;
+   }
+
+   if (draw->pt.middle.fetch_shade_emit) {
+      draw->pt.middle.fetch_shade_emit->destroy( draw->pt.middle.fetch_shade_emit );
+      draw->pt.middle.fetch_shade_emit = NULL;
+   }
+
+   if (draw->pt.front.vcache) {
+      draw->pt.front.vcache->destroy( draw->pt.front.vcache );
+      draw->pt.front.vcache = NULL;
+   }
+
+   if (draw->pt.front.varray) {
+      draw->pt.front.varray->destroy( draw->pt.front.varray );
+      draw->pt.front.varray = NULL;
+   }
+}
+
+
+/**
+ * Debug- print the first 'count' vertices.
+ */
+static void
+draw_print_arrays(struct draw_context *draw, uint prim, int start, uint count)
+{
+   uint i;
+
+   debug_printf("Draw arrays(prim = %u, start = %u, count = %u)\n",
+                prim, start, count);
+
+   for (i = 0; i < count; i++) {
+      uint ii, j;
+
+      if (draw->pt.user.elts) {
+         /* indexed arrays */
+         switch (draw->pt.user.eltSize) {
+         case 1:
+            {
+               const ubyte *elem = (const ubyte *) draw->pt.user.elts;
+               ii = elem[start + i];
+            }
+            break;
+         case 2:
+            {
+               const ushort *elem = (const ushort *) draw->pt.user.elts;
+               ii = elem[start + i];
+            }
+            break;
+         case 4:
+            {
+               const uint *elem = (const uint *) draw->pt.user.elts;
+               ii = elem[start + i];
+            }
+            break;
+         default:
+            assert(0);
+         }
+         debug_printf("Element[%u + %u] -> Vertex %u:\n", start, i, ii);
+      }
+      else {
+         /* non-indexed arrays */
+         ii = start + i;
+         debug_printf("Vertex %u:\n", ii);
+      }
+
+      for (j = 0; j < draw->pt.nr_vertex_elements; j++) {
+         uint buf = draw->pt.vertex_element[j].vertex_buffer_index;
+         ubyte *ptr = (ubyte *) draw->pt.user.vbuffer[buf];
+         ptr += draw->pt.vertex_buffer[buf].stride * ii;
+         ptr += draw->pt.vertex_element[j].src_offset;
+
+         debug_printf("  Attr %u: ", j);
+         switch (draw->pt.vertex_element[j].src_format) {
+         case PIPE_FORMAT_R32_FLOAT:
+            {
+               float *v = (float *) ptr;
+               debug_printf("%f  @ %p\n", v[0], (void *) v);
+            }
+            break;
+         case PIPE_FORMAT_R32G32_FLOAT:
+            {
+               float *v = (float *) ptr;
+               debug_printf("%f %f  @ %p\n", v[0], v[1], (void *) v);
+            }
+            break;
+         case PIPE_FORMAT_R32G32B32_FLOAT:
+            {
+               float *v = (float *) ptr;
+               debug_printf("%f %f %f  @ %p\n", v[0], v[1], v[2], (void *) v);
+            }
+            break;
+         case PIPE_FORMAT_R32G32B32A32_FLOAT:
+            {
+               float *v = (float *) ptr;
+               debug_printf("%f %f %f %f  @ %p\n", v[0], v[1], v[2], v[3],
+                            (void *) v);
+            }
+            break;
+         default:
+            debug_printf("other format (fix me)\n");
+            ;
+         }
+      }
+   }
+}
+
+
+/**
+ * Draw vertex arrays
+ * This is the main entrypoint into the drawing module.
+ * \param prim  one of PIPE_PRIM_x
+ * \param start  index of first vertex to draw
+ * \param count  number of vertices to draw
+ */
+void
+draw_arrays(struct draw_context *draw, unsigned prim,
+            unsigned start, unsigned count)
+{
+   unsigned reduced_prim = draw_pt_reduced_prim(prim);
+   if (reduced_prim != draw->reduced_prim) {
+      draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+      draw->reduced_prim = reduced_prim;
+   }
+
+   if (0)
+      draw_print_arrays(draw, prim, start, MIN2(count, 20));
+
+#if 0
+   {
+      int i;
+      debug_printf("draw_arrays(prim=%u start=%u count=%u):\n",
+                   prim, start, count);
+      tgsi_dump(draw->vs.vertex_shader->state.tokens, 0);
+      debug_printf("Elements:\n");
+      for (i = 0; i < draw->pt.nr_vertex_elements; i++) {
+         debug_printf("  format=%s comps=%u\n",
+                      pf_name(draw->pt.vertex_element[i].src_format),
+                      draw->pt.vertex_element[i].nr_components);
+      }
+      debug_printf("Buffers:\n");
+      for (i = 0; i < draw->pt.nr_vertex_buffers; i++) {
+         debug_printf("  stride=%u offset=%u ptr=%p\n",
+                      draw->pt.vertex_buffer[i].stride,
+                      draw->pt.vertex_buffer[i].buffer_offset,
+                      draw->pt.user.vbuffer[i]);
+      }
+   }
+#endif
+
+   /* drawing done here: */
+   draw_pt_arrays(draw, prim, start, count);
+}
+
+boolean draw_pt_get_edgeflag( struct draw_context *draw,
+                              unsigned idx )
+{
+   if (draw->pt.user.edgeflag)
+      return (draw->pt.user.edgeflag[idx/32] & (1 << (idx%32))) != 0;
+   else
+      return 1;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pt.h b/src/gallium/auxiliary/draw/draw_pt.h
new file mode 100644
index 00000000000..c02f229110a
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pt.h
@@ -0,0 +1,235 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  */
+
+#ifndef DRAW_PT_H
+#define DRAW_PT_H
+
+#include "pipe/p_compiler.h"
+
+typedef unsigned (*pt_elt_func)( const void *elts, unsigned idx );
+
+struct draw_pt_middle_end;
+struct draw_context;
+
+
+#define PT_SHADE      0x1
+#define PT_CLIPTEST   0x2
+#define PT_PIPELINE   0x4
+#define PT_MAX_MIDDLE 0x8
+
+
+/* The "front end" - prepare sets of fetch, draw elements for the
+ * middle end.
+ *
+ * Currenly one version of this:
+ *    - vcache - catchall implementation, decomposes to TRI/LINE/POINT prims
+ * Later:
+ *    - varray, varray_split
+ *    - velement, velement_split
+ *
+ * Currenly only using the vcache version.
+ */
+struct draw_pt_front_end {
+   void (*prepare)( struct draw_pt_front_end *,
+                    unsigned prim,
+                    struct draw_pt_middle_end *,
+		    unsigned opt );
+
+   void (*run)( struct draw_pt_front_end *,
+                pt_elt_func elt_func,
+                const void *elt_ptr,
+                unsigned count );
+
+   void (*finish)( struct draw_pt_front_end * );
+   void (*destroy)( struct draw_pt_front_end * );
+};
+
+
+/* The "middle end" - prepares actual hardware vertices for the
+ * hardware backend.
+ *
+ * Currently two versions of this:
+ *     - fetch, vertex shade, cliptest, prim-pipeline
+ *     - fetch, emit (ie passthrough)
+ */
+struct draw_pt_middle_end {
+   void (*prepare)( struct draw_pt_middle_end *,
+                    unsigned prim,
+		    unsigned opt,
+                    unsigned *max_vertices );
+
+   void (*run)( struct draw_pt_middle_end *,
+                const unsigned *fetch_elts,
+                unsigned fetch_count,
+                const ushort *draw_elts,
+                unsigned draw_count );
+
+   void (*run_linear)(struct draw_pt_middle_end *,
+                      unsigned start,
+                      unsigned count);
+
+   /* Transform all vertices in a linear range and then draw them with
+    * the supplied element list.  May fail and return FALSE.
+    */
+   boolean (*run_linear_elts)( struct draw_pt_middle_end *,
+                            unsigned fetch_start,
+                            unsigned fetch_count,
+                            const ushort *draw_elts,
+                            unsigned draw_count );
+
+   int (*get_max_vertex_count)( struct draw_pt_middle_end * );
+
+   void (*finish)( struct draw_pt_middle_end * );
+   void (*destroy)( struct draw_pt_middle_end * );
+};
+
+
+/* The "back end" - supplied by the driver, defined in draw_vbuf.h.
+ */
+struct vbuf_render;
+struct vertex_header;
+
+
+/* Helper functions.
+ */
+pt_elt_func draw_pt_elt_func( struct draw_context *draw );
+const void *draw_pt_elt_ptr( struct draw_context *draw,
+                             unsigned start );
+
+/* Frontends: 
+ *
+ * Currently only the general-purpose vcache implementation, could add
+ * a special case for tiny vertex buffers.
+ */
+struct draw_pt_front_end *draw_pt_vcache( struct draw_context *draw );
+struct draw_pt_front_end *draw_pt_varray(struct draw_context *draw);
+
+
+/* Middle-ends:
+ *
+ * Currently one general-purpose case which can do all possibilities,
+ * at the slight expense of creating a vertex_header in some cases
+ * unecessarily.
+ *
+ * The special case fetch_emit code avoids pipeline vertices
+ * altogether and builds hardware vertices directly from API
+ * vertex_elements.
+ */
+struct draw_pt_middle_end *draw_pt_fetch_emit( struct draw_context *draw );
+struct draw_pt_middle_end *draw_pt_middle_fse( struct draw_context *draw );
+struct draw_pt_middle_end *draw_pt_fetch_pipeline_or_emit(struct draw_context *draw);
+
+
+/* More helpers:
+ */
+boolean draw_pt_get_edgeflag( struct draw_context *draw,
+                              unsigned idx );
+
+
+/*******************************************************************************
+ * HW vertex emit:
+ */
+struct pt_emit;
+
+void draw_pt_emit_prepare( struct pt_emit *emit,
+			   unsigned prim,
+                           unsigned *max_vertices );
+
+void draw_pt_emit( struct pt_emit *emit,
+		   const float (*vertex_data)[4],
+		   unsigned vertex_count,
+		   unsigned stride,
+		   const ushort *elts,
+		   unsigned count );
+
+void draw_pt_emit_linear( struct pt_emit *emit,
+                          const float (*vertex_data)[4],
+                          unsigned vertex_count,
+                          unsigned stride,
+                          unsigned start,
+                          unsigned count );
+
+void draw_pt_emit_destroy( struct pt_emit *emit );
+
+struct pt_emit *draw_pt_emit_create( struct draw_context *draw );
+
+
+/*******************************************************************************
+ * API vertex fetch:
+ */
+
+struct pt_fetch;
+void draw_pt_fetch_prepare( struct pt_fetch *fetch,
+			    unsigned vertex_size );
+
+void draw_pt_fetch_run( struct pt_fetch *fetch,
+			const unsigned *elts,
+			unsigned count,
+			char *verts );
+
+void draw_pt_fetch_run_linear( struct pt_fetch *fetch,
+                               unsigned start,
+                               unsigned count,
+                               char *verts );
+
+void draw_pt_fetch_destroy( struct pt_fetch *fetch );
+
+struct pt_fetch *draw_pt_fetch_create( struct draw_context *draw );
+
+/*******************************************************************************
+ * Post-VS: cliptest, rhw, viewport
+ */
+struct pt_post_vs;
+
+boolean draw_pt_post_vs_run( struct pt_post_vs *pvs,
+			     struct vertex_header *pipeline_verts,
+			     unsigned stride,
+			     unsigned count );
+
+void draw_pt_post_vs_prepare( struct pt_post_vs *pvs,
+			      boolean bypass_clipping,
+			      boolean identity_viewport,
+			      boolean opengl );
+
+struct pt_post_vs *draw_pt_post_vs_create( struct draw_context *draw );
+
+void draw_pt_post_vs_destroy( struct pt_post_vs *pvs );
+
+
+/*******************************************************************************
+ * Utils: 
+ */
+void draw_pt_split_prim(unsigned prim, unsigned *first, unsigned *incr);
+unsigned draw_pt_reduced_prim(unsigned prim);
+
+
+#endif
diff --git a/src/gallium/auxiliary/draw/draw_pt_decompose.h b/src/gallium/auxiliary/draw/draw_pt_decompose.h
new file mode 100644
index 00000000000..3fb06956878
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pt_decompose.h
@@ -0,0 +1,153 @@
+
+
+static void FUNC( ARGS,
+                  unsigned count )
+{
+   LOCAL_VARS;
+
+   switch (prim) {
+   case PIPE_PRIM_POINTS:
+      for (i = 0; i < count; i ++) {
+	 POINT( (i + 0) );
+      }
+      break;
+
+   case PIPE_PRIM_LINES:
+      for (i = 0; i+1 < count; i += 2) {
+         LINE( DRAW_PIPE_RESET_STIPPLE,
+               (i + 0),
+               (i + 1));
+      }
+      break;
+
+   case PIPE_PRIM_LINE_LOOP:
+      if (count >= 2) {
+         flags = DRAW_PIPE_RESET_STIPPLE;
+
+         for (i = 1; i < count; i++, flags = 0) {
+            LINE( flags,
+                  (i - 1),
+                  (i ));
+         }
+
+	 LINE( flags,
+               (i - 1),
+               (0 ));
+      }
+      break;
+
+   case PIPE_PRIM_LINE_STRIP:
+      flags = DRAW_PIPE_RESET_STIPPLE;
+      for (i = 1; i < count; i++, flags = 0) {
+         LINE( flags,
+               (i - 1),
+               (i ));
+      }
+      break;
+
+   case PIPE_PRIM_TRIANGLES:
+      for (i = 0; i+2 < count; i += 3) {
+         TRIANGLE( DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL,
+                   (i + 0),
+                   (i + 1),
+                   (i + 2 ));
+      }
+      break;
+
+   case PIPE_PRIM_TRIANGLE_STRIP:
+      if (flatfirst) {
+         for (i = 0; i+2 < count; i++) {
+            TRIANGLE( DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL,
+                      (i + 0),
+                      (i + 1 + (i&1)),
+                      (i + 2 - (i&1)));
+         }
+      }
+      else {
+         for (i = 0; i+2 < count; i++) {
+            TRIANGLE( DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL,
+                      (i + 0 + (i&1)),
+                      (i + 1 - (i&1)),
+                      (i + 2 ));
+         }
+      }
+      break;
+
+   case PIPE_PRIM_TRIANGLE_FAN:
+      if (count >= 3) {
+         if (flatfirst) {
+            for (i = 0; i+2 < count; i++) {
+               TRIANGLE( DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL,
+                         (i + 1),
+                         (i + 2),
+                         (0 ));
+            }
+         }
+         else {
+            for (i = 0; i+2 < count; i++) {
+               TRIANGLE( DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL,
+                         (0),
+                         (i + 1),
+                         (i + 2 ));
+            }
+         }
+      }
+      break;
+
+
+   case PIPE_PRIM_QUADS:
+      for (i = 0; i+3 < count; i += 4) {
+         QUAD( (i + 0),
+               (i + 1),
+               (i + 2),
+               (i + 3));
+      }
+      break;
+
+   case PIPE_PRIM_QUAD_STRIP:
+      for (i = 0; i+3 < count; i += 2) {
+         QUAD( (i + 2),
+               (i + 0),
+               (i + 1),
+               (i + 3));
+      }
+      break;
+
+   case PIPE_PRIM_POLYGON:
+      {
+         /* These bitflags look a little odd because we submit the
+          * vertices as (1,2,0) to satisfy flatshade requirements.
+          */
+         const ushort edge_first  = DRAW_PIPE_EDGE_FLAG_2;
+         const ushort edge_middle = DRAW_PIPE_EDGE_FLAG_0;
+         const ushort edge_last   = DRAW_PIPE_EDGE_FLAG_1;
+
+         flags = DRAW_PIPE_RESET_STIPPLE | edge_first | edge_middle;
+
+	 for (i = 0; i+2 < count; i++, flags = edge_middle) {
+
+            if (i + 3 == count)
+               flags |= edge_last;
+
+	    TRIANGLE( flags,
+                      (i + 1),
+                      (i + 2),
+                      (0));
+	 }
+      }
+      break;
+
+   default:
+      assert(0);
+      break;
+   }
+
+   FLUSH;
+}
+
+
+#undef TRIANGLE
+#undef QUAD
+#undef POINT
+#undef LINE
+#undef FUNC
diff --git a/src/gallium/auxiliary/draw/draw_pt_elts.c b/src/gallium/auxiliary/draw/draw_pt_elts.c
new file mode 100644
index 00000000000..b7780fb5073
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pt_elts.c
@@ -0,0 +1,88 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  */
+
+#include "draw/draw_pt.h"
+#include "draw/draw_private.h"
+
+/* Neat get_elt func that also works for varrays drawing by encoding
+ * the start value into a pointer.  
+ */
+
+static unsigned elt_uint( const void *elts, unsigned idx )
+{
+   return *(((const uint *)elts) + idx);
+}
+
+static unsigned elt_ushort( const void *elts, unsigned idx )
+{
+   return *(((const ushort *)elts) + idx);
+}
+
+static unsigned elt_ubyte( const void *elts, unsigned idx )
+{
+   return *(((const ubyte *)elts) + idx);
+}
+
+static unsigned elt_vert( const void *elts, unsigned idx )
+{
+   return (const ubyte *)elts - (const ubyte *)NULL + idx;
+}
+
+pt_elt_func draw_pt_elt_func( struct draw_context *draw )
+{
+   switch (draw->pt.user.eltSize) {
+   case 0: return &elt_vert;
+   case 1: return &elt_ubyte;
+   case 2: return &elt_ushort; 
+   case 4: return &elt_uint;
+   default: return NULL;
+   }
+}     
+
+const void *draw_pt_elt_ptr( struct draw_context *draw,
+                             unsigned start )
+{
+   const char *elts = draw->pt.user.elts;
+
+   switch (draw->pt.user.eltSize) {
+   case 0: 
+      return (const void *)(((const ubyte *)NULL) + start);
+   case 1: 
+      return (const void *)(((const ubyte *)elts) + start);
+   case 2: 
+      return (const void *)(((const ushort *)elts) + start);
+   case 4: 
+      return (const void *)(((const uint *)elts) + start);
+   default:
+      return NULL;
+   }
+}
diff --git a/src/gallium/auxiliary/draw/draw_pt_emit.c b/src/gallium/auxiliary/draw/draw_pt_emit.c
new file mode 100644
index 00000000000..232dfdaed2d
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pt_emit.c
@@ -0,0 +1,308 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include "util/u_memory.h"
+#include "draw/draw_context.h"
+#include "draw/draw_private.h"
+#include "draw/draw_vbuf.h"
+#include "draw/draw_vertex.h"
+#include "draw/draw_pt.h"
+#include "translate/translate.h"
+#include "translate/translate_cache.h"
+
+struct pt_emit {
+   struct draw_context *draw;
+
+   struct translate *translate;
+
+   struct translate_cache *cache;
+   unsigned prim;
+
+   const struct vertex_info *vinfo;
+};
+
+void draw_pt_emit_prepare( struct pt_emit *emit,
+			   unsigned prim,
+                           unsigned *max_vertices )
+{
+   struct draw_context *draw = emit->draw;
+   const struct vertex_info *vinfo;
+   unsigned dst_offset;
+   struct translate_key hw_key;
+   unsigned i;
+   boolean ok;
+   
+   /* XXX: need to flush to get prim_vbuf.c to release its allocation?? 
+    */
+   draw_do_flush( draw, DRAW_FLUSH_BACKEND );
+
+
+   /* XXX: may need to defensively reset this later on as clipping can
+    * clobber this state in the render backend.
+    */
+   emit->prim = prim;
+
+   ok = draw->render->set_primitive(draw->render, emit->prim);
+   if (!ok) {
+      assert(0);
+      return;
+   }
+
+   /* Must do this after set_primitive() above:
+    */
+   emit->vinfo = vinfo = draw->render->get_vertex_info(draw->render);
+
+
+   /* Translate from pipeline vertices to hw vertices.
+    */
+   dst_offset = 0;
+   for (i = 0; i < vinfo->num_attribs; i++) {
+      unsigned emit_sz = 0;
+      unsigned src_buffer = 0;
+      unsigned output_format;
+      unsigned src_offset = (vinfo->attrib[i].src_index * 4 * sizeof(float) );
+
+
+         
+      switch (vinfo->attrib[i].emit) {
+      case EMIT_4F:
+	 output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+	 emit_sz = 4 * sizeof(float);
+	 break;
+      case EMIT_3F:
+	 output_format = PIPE_FORMAT_R32G32B32_FLOAT;
+	 emit_sz = 3 * sizeof(float);
+	 break;
+      case EMIT_2F:
+	 output_format = PIPE_FORMAT_R32G32_FLOAT;
+	 emit_sz = 2 * sizeof(float);
+	 break;
+      case EMIT_1F:
+	 output_format = PIPE_FORMAT_R32_FLOAT;
+	 emit_sz = 1 * sizeof(float);
+	 break;
+      case EMIT_1F_PSIZE:
+	 output_format = PIPE_FORMAT_R32_FLOAT;
+	 emit_sz = 1 * sizeof(float);
+	 src_buffer = 1;
+	 src_offset = 0;
+	 break;
+      case EMIT_4UB:
+	 output_format = PIPE_FORMAT_B8G8R8A8_UNORM;
+	 emit_sz = 4 * sizeof(ubyte);
+         break;
+      default:
+	 assert(0);
+	 output_format = PIPE_FORMAT_NONE;
+	 emit_sz = 0;
+	 break;
+      }
+      
+      hw_key.element[i].input_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+      hw_key.element[i].input_buffer = src_buffer;
+      hw_key.element[i].input_offset = src_offset;
+      hw_key.element[i].output_format = output_format;
+      hw_key.element[i].output_offset = dst_offset;
+
+      dst_offset += emit_sz;
+   }
+
+   hw_key.nr_elements = vinfo->num_attribs;
+   hw_key.output_stride = vinfo->size * 4;
+
+   if (!emit->translate ||
+       translate_key_compare(&emit->translate->key, &hw_key) != 0)
+   {
+      translate_key_sanitize(&hw_key);
+      emit->translate = translate_cache_find(emit->cache, &hw_key);
+   }
+
+   *max_vertices = (draw->render->max_vertex_buffer_bytes / 
+                    (vinfo->size * 4));
+
+   /* even number */
+   *max_vertices = *max_vertices & ~1;
+}
+
+
+void draw_pt_emit( struct pt_emit *emit,
+		   const float (*vertex_data)[4],
+		   unsigned vertex_count,
+		   unsigned stride,
+		   const ushort *elts,
+		   unsigned count )
+{
+   struct draw_context *draw = emit->draw;
+   struct translate *translate = emit->translate;
+   struct vbuf_render *render = draw->render;
+   void *hw_verts;
+
+   /* XXX: need to flush to get prim_vbuf.c to release its allocation?? 
+    */
+   draw_do_flush( draw, DRAW_FLUSH_BACKEND );
+
+   if (vertex_count >= UNDEFINED_VERTEX_ID) {
+      assert(0);
+      return;
+   }
+
+   /* XXX: and work out some way to coordinate the render primitive
+    * between vbuf.c and here...
+    */
+   if (!draw->render->set_primitive(draw->render, emit->prim)) {
+      assert(0);
+      return;
+   }
+
+   hw_verts = render->allocate_vertices(render,
+					(ushort)translate->key.output_stride,
+					(ushort)vertex_count);
+   if (!hw_verts) {
+      assert(0);
+      return;
+   }
+
+   translate->set_buffer(translate, 
+			 0, 
+			 vertex_data,
+			 stride );
+
+   translate->set_buffer(translate, 
+			 1, 
+			 &draw->rasterizer->point_size,
+			 0);
+
+   translate->run( translate,
+		   0, 
+		   vertex_count,
+		   hw_verts );
+
+   render->draw(render,
+		elts,
+		count);
+
+   render->release_vertices(render,
+			    hw_verts,
+			    translate->key.output_stride,
+			    vertex_count);
+}
+
+
+void draw_pt_emit_linear(struct pt_emit *emit,
+                         const float (*vertex_data)[4],
+                         unsigned vertex_count,
+                         unsigned stride,
+                         unsigned start,
+                         unsigned count)
+{
+   struct draw_context *draw = emit->draw;
+   struct translate *translate = emit->translate;
+   struct vbuf_render *render = draw->render;
+   void *hw_verts;
+
+#if 0
+   debug_printf("Linear emit\n");
+#endif
+   /* XXX: need to flush to get prim_vbuf.c to release its allocation?? 
+    */
+   draw_do_flush( draw, DRAW_FLUSH_BACKEND );
+
+   if (count >= UNDEFINED_VERTEX_ID) {
+      assert(0);
+      return;
+   }
+
+   /* XXX: and work out some way to coordinate the render primitive
+    * between vbuf.c and here...
+    */
+   if (!draw->render->set_primitive(draw->render, emit->prim)) {
+      assert(0);
+      return;
+   }
+
+   hw_verts = render->allocate_vertices(render,
+					(ushort)translate->key.output_stride,
+					(ushort)count);
+   if (!hw_verts) {
+      assert(0);
+      return;
+   }
+
+   translate->set_buffer(translate, 0,
+			 vertex_data, stride);
+
+   translate->set_buffer(translate, 1,
+			 &draw->rasterizer->point_size,
+			 0);
+
+   translate->run(translate,
+                  0,
+                  vertex_count,
+                  hw_verts);
+
+   if (0) {
+      unsigned i;
+      for (i = 0; i < vertex_count; i++) {
+         debug_printf("\n\n%s vertex %d:\n", __FUNCTION__, i);
+         draw_dump_emitted_vertex( emit->vinfo, 
+                                   (const uint8_t *)hw_verts + 
+                                   translate->key.output_stride * i );
+      }
+   }
+
+
+   render->draw_arrays(render, start, count);
+
+   render->release_vertices(render,
+			    hw_verts,
+			    translate->key.output_stride,
+			    vertex_count);
+}
+
+struct pt_emit *draw_pt_emit_create( struct draw_context *draw )
+{
+   struct pt_emit *emit = CALLOC_STRUCT(pt_emit);
+   if (!emit)
+      return NULL;
+
+   emit->draw = draw;
+   emit->cache = translate_cache_create();
+   if (!emit->cache) {
+      FREE(emit);
+      return NULL;
+   }
+
+   return emit;
+}
+
+void draw_pt_emit_destroy( struct pt_emit *emit )
+{
+   if (emit->cache)
+      translate_cache_destroy(emit->cache);
+
+   FREE(emit);
+}
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch.c b/src/gallium/auxiliary/draw/draw_pt_fetch.c
new file mode 100644
index 00000000000..058caf7dcc3
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch.c
@@ -0,0 +1,228 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include "util/u_memory.h"
+#include "draw/draw_context.h"
+#include "draw/draw_private.h"
+#include "draw/draw_vbuf.h"
+#include "draw/draw_vertex.h"
+#include "draw/draw_pt.h"
+#include "translate/translate.h"
+#include "translate/translate_cache.h"
+
+
+struct pt_fetch {
+   struct draw_context *draw;
+
+   struct translate *translate;
+
+   unsigned vertex_size;
+   boolean need_edgeflags;
+
+   struct translate_cache *cache;
+};
+
+/* Perform the fetch from API vertex elements & vertex buffers, to a
+ * contiguous set of float[4] attributes as required for the
+ * vertex_shader->run_linear() method.
+ *
+ * This is used in all cases except pure passthrough
+ * (draw_pt_fetch_emit.c) which has its own version to translate
+ * directly to hw vertices.
+ *
+ */
+void draw_pt_fetch_prepare( struct pt_fetch *fetch,
+			    unsigned vertex_size )
+{
+   struct draw_context *draw = fetch->draw;
+   unsigned i, nr = 0;
+   unsigned dst_offset = 0;
+   struct translate_key key;
+
+   fetch->vertex_size = vertex_size;
+
+   /* Always emit/leave space for a vertex header.
+    *
+    * It's worth considering whether the vertex headers should contain
+    * a pointer to the 'data', rather than having it inline.
+    * Something to look at after we've fully switched over to the pt
+    * paths.
+    */
+   {
+      /* Need to set header->vertex_id = 0xffff somehow.
+       */
+      key.element[nr].input_format = PIPE_FORMAT_R32_FLOAT;
+      key.element[nr].input_buffer = draw->pt.nr_vertex_buffers;
+      key.element[nr].input_offset = 0;
+      key.element[nr].output_format = PIPE_FORMAT_R32_FLOAT;
+      key.element[nr].output_offset = dst_offset;
+      dst_offset += 1 * sizeof(float);
+      nr++;
+
+
+      /* Just leave the clip[] array untouched.
+       */
+      dst_offset += 4 * sizeof(float);
+   }
+      
+
+   for (i = 0; i < draw->pt.nr_vertex_elements; i++) {
+      key.element[nr].input_format = draw->pt.vertex_element[i].src_format;
+      key.element[nr].input_buffer = draw->pt.vertex_element[i].vertex_buffer_index;
+      key.element[nr].input_offset = draw->pt.vertex_element[i].src_offset;
+      key.element[nr].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+      key.element[nr].output_offset = dst_offset;
+
+      dst_offset += 4 * sizeof(float);
+      nr++;
+   }
+
+   assert(dst_offset <= vertex_size);
+
+   key.nr_elements = nr;
+   key.output_stride = vertex_size;
+
+
+   if (!fetch->translate ||
+       translate_key_compare(&fetch->translate->key, &key) != 0)
+   {
+      translate_key_sanitize(&key);
+      fetch->translate = translate_cache_find(fetch->cache, &key);
+
+      {
+	 static struct vertex_header vh = { 0, 1, 0, 0xffff };
+	 fetch->translate->set_buffer(fetch->translate,
+				      draw->pt.nr_vertex_buffers,
+				      &vh,
+				      0);
+      }
+   }
+
+   fetch->need_edgeflags = ((draw->rasterizer->fill_cw != PIPE_POLYGON_MODE_FILL ||
+                             draw->rasterizer->fill_ccw != PIPE_POLYGON_MODE_FILL) &&
+                            draw->pt.user.edgeflag);
+}
+
+
+
+
+void draw_pt_fetch_run( struct pt_fetch *fetch,
+			const unsigned *elts,
+			unsigned count,
+			char *verts )
+{
+   struct draw_context *draw = fetch->draw;
+   struct translate *translate = fetch->translate;
+   unsigned i;
+
+   for (i = 0; i < draw->pt.nr_vertex_buffers; i++) {
+      translate->set_buffer(translate, 
+			    i, 
+			    ((char *)draw->pt.user.vbuffer[i] + 
+			     draw->pt.vertex_buffer[i].buffer_offset),
+			    draw->pt.vertex_buffer[i].stride );
+   }
+
+   translate->run_elts( translate,
+			elts, 
+			count,
+			verts );
+
+   /* Edgeflags are hard to fit into a translate program, populate
+    * them separately if required.  In the setup above they are
+    * defaulted to one, so only need this if there is reason to change
+    * that default:
+    */
+   if (fetch->need_edgeflags) {
+      for (i = 0; i < count; i++) {
+         struct vertex_header *vh = (struct vertex_header *)(verts + i * fetch->vertex_size);
+         vh->edgeflag = draw_pt_get_edgeflag( draw, elts[i] );
+      }
+   }
+}
+
+
+void draw_pt_fetch_run_linear( struct pt_fetch *fetch,
+                               unsigned start,
+                               unsigned count,
+                               char *verts )
+{
+   struct draw_context *draw = fetch->draw;
+   struct translate *translate = fetch->translate;
+   unsigned i;
+
+   for (i = 0; i < draw->pt.nr_vertex_buffers; i++) {
+      translate->set_buffer(translate,
+			    i,
+			    ((char *)draw->pt.user.vbuffer[i] +
+			     draw->pt.vertex_buffer[i].buffer_offset),
+			    draw->pt.vertex_buffer[i].stride );
+   }
+
+   translate->run( translate,
+                   start,
+                   count,
+                   verts );
+
+   /* Edgeflags are hard to fit into a translate program, populate
+    * them separately if required.  In the setup above they are
+    * defaulted to one, so only need this if there is reason to change
+    * that default:
+    */
+   if (fetch->need_edgeflags) {
+      for (i = 0; i < count; i++) {
+         struct vertex_header *vh = (struct vertex_header *)(verts + i * fetch->vertex_size);
+         vh->edgeflag = draw_pt_get_edgeflag( draw, start + i );
+      }
+   }
+}
+
+
+struct pt_fetch *draw_pt_fetch_create( struct draw_context *draw )
+{
+   struct pt_fetch *fetch = CALLOC_STRUCT(pt_fetch);
+   if (!fetch)
+      return NULL;
+
+   fetch->draw = draw;
+   fetch->cache = translate_cache_create();
+   if (!fetch->cache) {
+      FREE(fetch);
+      return NULL;
+   }
+
+   return fetch;
+}
+
+void draw_pt_fetch_destroy( struct pt_fetch *fetch )
+{
+   if (fetch->cache)
+      translate_cache_destroy(fetch->cache);
+
+   FREE(fetch);
+}
+
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
new file mode 100644
index 00000000000..dcb7744b177
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
@@ -0,0 +1,426 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  */
+
+#include "util/u_memory.h"
+#include "draw/draw_context.h"
+#include "draw/draw_private.h"
+#include "draw/draw_vbuf.h"
+#include "draw/draw_vertex.h"
+#include "draw/draw_pt.h"
+#include "translate/translate.h"
+#include "translate/translate_cache.h"
+
+/* The simplest 'middle end' in the new vertex code.  
+ * 
+ * The responsibilities of a middle end are to:
+ *  - perform vertex fetch using
+ *       - draw vertex element/buffer state
+ *       - a list of fetch indices we received as an input
+ *  - run the vertex shader
+ *  - cliptest, 
+ *  - clip coord calculation 
+ *  - viewport transformation
+ *  - if necessary, run the primitive pipeline, passing it:
+ *       - a linear array of vertex_header vertices constructed here
+ *       - a set of draw indices we received as an input
+ *  - otherwise, drive the hw backend,
+ *       - allocate space for hardware format vertices
+ *       - translate the vertex-shader output vertices to hw format
+ *       - calling the backend draw functions.
+ *
+ * For convenience, we provide a helper function to drive the hardware
+ * backend given similar inputs to those required to run the pipeline.
+ *
+ * In the case of passthrough mode, many of these actions are disabled
+ * or noops, so we end up doing:
+ *
+ *  - perform vertex fetch
+ *  - drive the hw backend
+ *
+ * IE, basically just vertex fetch to post-vs-format vertices,
+ * followed by a call to the backend helper function.
+ */
+
+
+struct fetch_emit_middle_end {
+   struct draw_pt_middle_end base;
+   struct draw_context *draw;
+   
+   struct translate *translate;
+   const struct vertex_info *vinfo;
+
+   /* Cache point size somewhere it's address won't change:
+    */
+   float point_size;
+
+   struct translate_cache *cache;
+};
+
+
+
+
+static void fetch_emit_prepare( struct draw_pt_middle_end *middle,
+                                unsigned prim,
+				unsigned opt,
+                                unsigned *max_vertices )
+{
+   struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
+   struct draw_context *draw = feme->draw;
+   const struct vertex_info *vinfo;
+   unsigned i, dst_offset;
+   boolean ok;
+   struct translate_key key;
+
+
+   ok = draw->render->set_primitive( draw->render, 
+                                     prim );
+   if (!ok) {
+      assert(0);
+      return;
+   }
+   
+   /* Must do this after set_primitive() above:
+    */
+   vinfo = feme->vinfo = draw->render->get_vertex_info(draw->render);
+   
+   
+
+   /* Transform from API vertices to HW vertices, skipping the
+    * pipeline_vertex intermediate step.
+    */
+   dst_offset = 0;
+   memset(&key, 0, sizeof(key));
+
+   for (i = 0; i < vinfo->num_attribs; i++) {
+      const struct pipe_vertex_element *src = &draw->pt.vertex_element[vinfo->attrib[i].src_index];
+
+      unsigned emit_sz = 0;
+      unsigned input_format = src->src_format;
+      unsigned input_buffer = src->vertex_buffer_index;
+      unsigned input_offset = src->src_offset;
+      unsigned output_format;
+
+      switch (vinfo->attrib[i].emit) {
+      case EMIT_4F:
+	 output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+	 emit_sz = 4 * sizeof(float);
+         break;
+      case EMIT_3F:
+	 output_format = PIPE_FORMAT_R32G32B32_FLOAT;
+	 emit_sz = 3 * sizeof(float);
+         break;
+      case EMIT_2F:
+	 output_format = PIPE_FORMAT_R32G32_FLOAT;
+	 emit_sz = 2 * sizeof(float);
+         break;
+      case EMIT_1F:
+	 output_format = PIPE_FORMAT_R32_FLOAT;
+	 emit_sz = 1 * sizeof(float);
+         break;
+      case EMIT_1F_PSIZE:
+	 input_format = PIPE_FORMAT_R32_FLOAT;
+	 input_buffer = draw->pt.nr_vertex_buffers;
+	 input_offset = 0;
+	 output_format = PIPE_FORMAT_R32_FLOAT;
+	 emit_sz = 1 * sizeof(float);
+         break;
+      default:
+         assert(0);
+	 output_format = PIPE_FORMAT_NONE;
+	 emit_sz = 0;
+	 continue;
+      }
+
+      key.element[i].input_format = input_format;
+      key.element[i].input_buffer = input_buffer;
+      key.element[i].input_offset = input_offset;
+      key.element[i].output_format = output_format;
+      key.element[i].output_offset = dst_offset;
+      
+      dst_offset += emit_sz;
+   }
+
+   key.nr_elements = vinfo->num_attribs;
+   key.output_stride = vinfo->size * 4;
+
+   /* Don't bother with caching at this stage:
+    */
+   if (!feme->translate ||
+       translate_key_compare(&feme->translate->key, &key) != 0) 
+   {
+      translate_key_sanitize(&key);
+      feme->translate = translate_cache_find(feme->cache,
+                                             &key);
+
+
+      feme->translate->set_buffer(feme->translate, 
+				  draw->pt.nr_vertex_buffers, 
+				  &feme->point_size,
+				  0);
+   }
+   
+   feme->point_size = draw->rasterizer->point_size;
+
+   for (i = 0; i < draw->pt.nr_vertex_buffers; i++) {
+      feme->translate->set_buffer(feme->translate, 
+                                  i, 
+                                  ((char *)draw->pt.user.vbuffer[i] + 
+                                   draw->pt.vertex_buffer[i].buffer_offset),
+                                  draw->pt.vertex_buffer[i].stride );
+   }
+
+   *max_vertices = (draw->render->max_vertex_buffer_bytes / 
+                    (vinfo->size * 4));
+
+   /* Return an even number of verts.
+    * This prevents "parity" errors when splitting long triangle strips which
+    * can lead to front/back culling mix-ups.
+    * Every other triangle in a strip has an alternate front/back orientation
+    * so splitting at an odd position can cause the orientation of subsequent
+    * triangles to get reversed.
+    */
+   *max_vertices = *max_vertices & ~1;
+}
+
+
+
+
+
+static void fetch_emit_run( struct draw_pt_middle_end *middle,
+                            const unsigned *fetch_elts,
+                            unsigned fetch_count,
+                            const ushort *draw_elts,
+                            unsigned draw_count )
+{
+   struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
+   struct draw_context *draw = feme->draw;
+   void *hw_verts;
+   
+   /* XXX: need to flush to get prim_vbuf.c to release its allocation?? 
+    */
+   draw_do_flush( draw, DRAW_FLUSH_BACKEND );
+
+   if (fetch_count >= UNDEFINED_VERTEX_ID) {
+      assert(0);
+      return;
+   }
+
+   hw_verts = draw->render->allocate_vertices( draw->render,
+                                               (ushort)feme->translate->key.output_stride,
+                                               (ushort)fetch_count );
+   if (!hw_verts) {
+      assert(0);
+      return;
+   }
+         
+					
+   /* Single routine to fetch vertices and emit HW verts.
+    */
+   feme->translate->run_elts( feme->translate, 
+			      fetch_elts,
+			      fetch_count,
+			      hw_verts );
+
+   if (0) {
+      unsigned i;
+      for (i = 0; i < fetch_count; i++) {
+         debug_printf("\n\nvertex %d:\n", i);
+         draw_dump_emitted_vertex( feme->vinfo, 
+                                   (const uint8_t *)hw_verts + feme->vinfo->size * 4 * i );
+      }
+   }
+
+   /* XXX: Draw arrays path to avoid re-emitting index list again and
+    * again.
+    */
+   draw->render->draw( draw->render, 
+                       draw_elts, 
+                       draw_count );
+
+   /* Done -- that was easy, wasn't it: 
+    */
+   draw->render->release_vertices( draw->render, 
+                                   hw_verts, 
+                                   feme->translate->key.output_stride, 
+                                   fetch_count );
+
+}
+
+
+static void fetch_emit_run_linear( struct draw_pt_middle_end *middle,
+                                   unsigned start,
+                                   unsigned count )
+{
+   struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
+   struct draw_context *draw = feme->draw;
+   void *hw_verts;
+
+   /* XXX: need to flush to get prim_vbuf.c to release its allocation??
+    */
+   draw_do_flush( draw, DRAW_FLUSH_BACKEND );
+
+   if (count >= UNDEFINED_VERTEX_ID) {
+      assert(0);
+      return;
+   }
+
+   hw_verts = draw->render->allocate_vertices( draw->render,
+                                               (ushort)feme->translate->key.output_stride,
+                                               (ushort)count );
+   if (!hw_verts) {
+      assert(0);
+      return;
+   }
+
+   /* Single routine to fetch vertices and emit HW verts.
+    */
+   feme->translate->run( feme->translate,
+                         start,
+                         count,
+                         hw_verts );
+
+   if (0) {
+      unsigned i;
+      for (i = 0; i < count; i++) {
+         debug_printf("\n\nvertex %d:\n", i);
+         draw_dump_emitted_vertex( feme->vinfo,
+                                   (const uint8_t *)hw_verts + feme->vinfo->size * 4 * i );
+      }
+   }
+
+   /* XXX: Draw arrays path to avoid re-emitting index list again and
+    * again.
+    */
+   draw->render->draw_arrays( draw->render,
+                              0, /*start*/
+                              count );
+
+   /* Done -- that was easy, wasn't it:
+    */
+   draw->render->release_vertices( draw->render,
+                                   hw_verts,
+                                   feme->translate->key.output_stride,
+                                   count );
+
+}
+
+
+static boolean fetch_emit_run_linear_elts( struct draw_pt_middle_end *middle,
+                                        unsigned start,
+                                        unsigned count,
+                                        const ushort *draw_elts,
+                                        unsigned draw_count )
+{
+   struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
+   struct draw_context *draw = feme->draw;
+   void *hw_verts;
+
+   /* XXX: need to flush to get prim_vbuf.c to release its allocation??
+    */
+   draw_do_flush( draw, DRAW_FLUSH_BACKEND );
+
+   if (count >= UNDEFINED_VERTEX_ID)
+      return FALSE;
+
+   hw_verts = draw->render->allocate_vertices( draw->render,
+                                               (ushort)feme->translate->key.output_stride,
+                                               (ushort)count );
+   if (!hw_verts) 
+      return FALSE;
+
+   /* Single routine to fetch vertices and emit HW verts.
+    */
+   feme->translate->run( feme->translate,
+                         start,
+                         count,
+                         hw_verts );
+
+   /* XXX: Draw arrays path to avoid re-emitting index list again and
+    * again.
+    */
+   draw->render->draw( draw->render, 
+                       draw_elts, 
+                       draw_count );
+
+   /* Done -- that was easy, wasn't it:
+    */
+   draw->render->release_vertices( draw->render,
+                                   hw_verts,
+                                   feme->translate->key.output_stride,
+                                   count );
+
+   return TRUE;
+}
+
+
+
+
+static void fetch_emit_finish( struct draw_pt_middle_end *middle )
+{
+   /* nothing to do */
+}
+
+static void fetch_emit_destroy( struct draw_pt_middle_end *middle )
+{
+   struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
+
+   if (feme->cache)
+      translate_cache_destroy(feme->cache);
+
+   FREE(middle);
+}
+
+
+struct draw_pt_middle_end *draw_pt_fetch_emit( struct draw_context *draw )
+{
+   struct fetch_emit_middle_end *fetch_emit = CALLOC_STRUCT( fetch_emit_middle_end );
+   if (fetch_emit == NULL)
+      return NULL;
+
+   fetch_emit->cache = translate_cache_create();
+   if (!fetch_emit->cache) {
+      FREE(fetch_emit);
+      return NULL;
+   }
+
+   fetch_emit->base.prepare    = fetch_emit_prepare;
+   fetch_emit->base.run        = fetch_emit_run;
+   fetch_emit->base.run_linear = fetch_emit_run_linear;
+   fetch_emit->base.run_linear_elts = fetch_emit_run_linear_elts;
+   fetch_emit->base.finish     = fetch_emit_finish;
+   fetch_emit->base.destroy    = fetch_emit_destroy;
+
+   fetch_emit->draw = draw;
+     
+   return &fetch_emit->base;
+}
+
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
new file mode 100644
index 00000000000..84ffe3296af
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
@@ -0,0 +1,422 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  */
+
+
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "draw/draw_context.h"
+#include "draw/draw_private.h"
+#include "draw/draw_vbuf.h"
+#include "draw/draw_vertex.h"
+#include "draw/draw_pt.h"
+#include "draw/draw_vs.h"
+
+#include "translate/translate.h"
+
+struct fetch_shade_emit;
+
+
+/* Prototype fetch, shade, emit-hw-verts all in one go.
+ */
+struct fetch_shade_emit {
+   struct draw_pt_middle_end base;
+   struct draw_context *draw;
+
+
+   /* Temporaries:
+    */
+   const float *constants;
+   unsigned pitch[PIPE_MAX_ATTRIBS];
+   const ubyte *src[PIPE_MAX_ATTRIBS];
+   unsigned prim;
+
+   struct draw_vs_varient_key key;
+   struct draw_vs_varient *active;
+
+
+   const struct vertex_info *vinfo;
+};
+
+
+
+			       
+static void fse_prepare( struct draw_pt_middle_end *middle,
+                         unsigned prim, 
+                         unsigned opt,
+                         unsigned *max_vertices )
+{
+   struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
+   struct draw_context *draw = fse->draw;
+   unsigned num_vs_inputs = draw->vs.vertex_shader->info.num_inputs;
+   const struct vertex_info *vinfo;
+   unsigned i;
+   unsigned nr_vbs = 0;
+   
+
+   if (!draw->render->set_primitive( draw->render, 
+                                     prim )) {
+      assert(0);
+      return;
+   }
+
+   /* Must do this after set_primitive() above:
+    */
+   fse->vinfo = vinfo = draw->render->get_vertex_info(draw->render);
+   
+
+
+   fse->key.output_stride = vinfo->size * 4;
+   fse->key.nr_outputs = vinfo->num_attribs;
+   fse->key.nr_inputs = num_vs_inputs;
+
+   fse->key.nr_elements = MAX2(fse->key.nr_outputs,     /* outputs - translate to hw format */
+                               fse->key.nr_inputs);     /* inputs - fetch from api format */
+
+   fse->key.viewport = !draw->identity_viewport;
+   fse->key.clip = !draw->bypass_clipping;
+   fse->key.const_vbuffers = 0;
+
+   memset(fse->key.element, 0, 
+          fse->key.nr_elements * sizeof(fse->key.element[0]));
+
+   for (i = 0; i < num_vs_inputs; i++) {
+      const struct pipe_vertex_element *src = &draw->pt.vertex_element[i];
+      fse->key.element[i].in.format = src->src_format;
+
+      /* Consider ignoring these, ie make generated programs
+       * independent of this state:
+       */
+      fse->key.element[i].in.buffer = src->vertex_buffer_index;
+      fse->key.element[i].in.offset = src->src_offset;
+      nr_vbs = MAX2(nr_vbs, src->vertex_buffer_index + 1);
+   }
+   
+   for (i = 0; i < 5 && i < nr_vbs; i++) {
+      if (draw->pt.vertex_buffer[i].stride == 0)
+         fse->key.const_vbuffers |= (1<<i);
+   }
+
+   if (0) debug_printf("%s: lookup const_vbuffers: %x\n", __FUNCTION__, fse->key.const_vbuffers);
+   
+   {
+      unsigned dst_offset = 0;
+
+      for (i = 0; i < vinfo->num_attribs; i++) {
+         unsigned emit_sz = 0;
+
+         switch (vinfo->attrib[i].emit) {
+         case EMIT_4F:
+            emit_sz = 4 * sizeof(float);
+            break;
+         case EMIT_3F:
+            emit_sz = 3 * sizeof(float);
+            break;
+         case EMIT_2F:
+            emit_sz = 2 * sizeof(float);
+            break;
+         case EMIT_1F:
+            emit_sz = 1 * sizeof(float);
+            break;
+         case EMIT_1F_PSIZE:
+            emit_sz = 1 * sizeof(float);
+            break;
+         case EMIT_4UB:
+            emit_sz = 4 * sizeof(ubyte);
+            break;
+         default:
+            assert(0);
+            break;
+         }
+
+         /* The elements in the key correspond to vertex shader output
+          * numbers, not to positions in the hw vertex description --
+          * that's handled by the output_offset field.
+          */
+         fse->key.element[i].out.format = vinfo->attrib[i].emit;
+         fse->key.element[i].out.vs_output = vinfo->attrib[i].src_index;
+         fse->key.element[i].out.offset = dst_offset;
+      
+         dst_offset += emit_sz;
+         assert(fse->key.output_stride >= dst_offset);
+      }
+   }
+
+   
+   fse->active = draw_vs_lookup_varient( draw->vs.vertex_shader, 
+                                         &fse->key );
+
+   if (!fse->active) {
+      assert(0);
+      return ;
+   }
+
+   if (0) debug_printf("%s: found const_vbuffers: %x\n", __FUNCTION__, 
+                       fse->active->key.const_vbuffers);
+
+   /* Now set buffer pointers:
+    */
+   for (i = 0; i < draw->pt.nr_vertex_buffers; i++) {
+      fse->active->set_buffer( fse->active, 
+                               i, 
+                               ((const ubyte *) draw->pt.user.vbuffer[i] + 
+                                draw->pt.vertex_buffer[i].buffer_offset),
+                              draw->pt.vertex_buffer[i].stride );
+   }
+
+   *max_vertices = (draw->render->max_vertex_buffer_bytes / 
+                    (vinfo->size * 4));
+
+   /* Return an even number of verts.
+    * This prevents "parity" errors when splitting long triangle strips which
+    * can lead to front/back culling mix-ups.
+    * Every other triangle in a strip has an alternate front/back orientation
+    * so splitting at an odd position can cause the orientation of subsequent
+    * triangles to get reversed.
+    */
+   *max_vertices = *max_vertices & ~1;
+
+   /* Probably need to do this somewhere (or fix exec shader not to
+    * need it):
+    */
+   if (1) {
+      struct draw_vertex_shader *vs = draw->vs.vertex_shader;
+      vs->prepare(vs, draw);
+   }
+   
+
+   //return TRUE;
+}
+
+
+
+
+
+
+
+static void fse_run_linear( struct draw_pt_middle_end *middle, 
+                            unsigned start, 
+                            unsigned count )
+{
+   struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
+   struct draw_context *draw = fse->draw;
+   char *hw_verts;
+
+   /* XXX: need to flush to get prim_vbuf.c to release its allocation??
+    */
+   draw_do_flush( draw, DRAW_FLUSH_BACKEND );
+
+   if (count >= UNDEFINED_VERTEX_ID) {
+      assert(0);
+      return;
+   }
+
+   hw_verts = draw->render->allocate_vertices( draw->render,
+                                               (ushort)fse->key.output_stride,
+                                               (ushort)count );
+
+   if (!hw_verts) {
+      assert(0);
+      return;
+   }
+
+   /* Single routine to fetch vertices, run shader and emit HW verts.
+    * Clipping is done elsewhere -- either by the API or on hardware,
+    * or for some other reason not required...
+    */
+   fse->active->run_linear( fse->active, 
+                            start, count,
+                            hw_verts );
+
+   /* Draw arrays path to avoid re-emitting index list again and
+    * again.
+    */
+   draw->render->draw_arrays( draw->render,
+                              0,
+                              count );
+   
+   if (0) {
+      unsigned i;
+      for (i = 0; i < count; i++) {
+         debug_printf("\n\n%s vertex %d: (stride %d, offset %d)\n", __FUNCTION__, i,
+                      fse->key.output_stride,
+                      fse->key.output_stride * i);
+
+         draw_dump_emitted_vertex( fse->vinfo, 
+                                   (const uint8_t *)hw_verts + fse->key.output_stride * i );
+      }
+   }
+
+
+   draw->render->release_vertices( draw->render, 
+				   hw_verts, 
+				   fse->key.output_stride, 
+				   count );
+}
+
+
+static void
+fse_run(struct draw_pt_middle_end *middle,
+        const unsigned *fetch_elts,
+        unsigned fetch_count,
+        const ushort *draw_elts,
+        unsigned draw_count )
+{
+   struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
+   struct draw_context *draw = fse->draw;
+   void *hw_verts;
+   
+   /* XXX: need to flush to get prim_vbuf.c to release its allocation?? 
+    */
+   draw_do_flush( draw, DRAW_FLUSH_BACKEND );
+
+   if (fetch_count >= UNDEFINED_VERTEX_ID) {
+      assert(0);
+      return;
+   }
+
+   hw_verts = draw->render->allocate_vertices( draw->render,
+                                               (ushort)fse->key.output_stride,
+                                               (ushort)fetch_count );
+   if (!hw_verts) {
+      assert(0);
+      return;
+   }
+         
+					
+   /* Single routine to fetch vertices, run shader and emit HW verts.
+    */
+   fse->active->run_elts( fse->active, 
+                          fetch_elts,
+                          fetch_count,
+                          hw_verts );
+
+   draw->render->draw( draw->render, 
+                       draw_elts, 
+                       draw_count );
+
+   if (0) {
+      unsigned i;
+      for (i = 0; i < fetch_count; i++) {
+         debug_printf("\n\n%s vertex %d:\n", __FUNCTION__, i);
+         draw_dump_emitted_vertex( fse->vinfo, 
+                                   (const uint8_t *)hw_verts + 
+                                   fse->key.output_stride * i );
+      }
+   }
+
+
+   draw->render->release_vertices( draw->render, 
+                                   hw_verts, 
+                                   fse->key.output_stride, 
+                                   fetch_count );
+
+}
+
+
+
+static boolean fse_run_linear_elts( struct draw_pt_middle_end *middle, 
+                                 unsigned start, 
+                                 unsigned count,
+                                 const ushort *draw_elts,
+                                 unsigned draw_count )
+{
+   struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
+   struct draw_context *draw = fse->draw;
+   char *hw_verts;
+
+   /* XXX: need to flush to get prim_vbuf.c to release its allocation??
+    */
+   draw_do_flush( draw, DRAW_FLUSH_BACKEND );
+
+   if (count >= UNDEFINED_VERTEX_ID)
+      return FALSE;
+
+   hw_verts = draw->render->allocate_vertices( draw->render,
+                                               (ushort)fse->key.output_stride,
+                                               (ushort)count );
+
+   if (!hw_verts) {
+      return FALSE;
+   }
+
+   /* Single routine to fetch vertices, run shader and emit HW verts.
+    * Clipping is done elsewhere -- either by the API or on hardware,
+    * or for some other reason not required...
+    */
+   fse->active->run_linear( fse->active, 
+                            start, count,
+                            hw_verts );
+
+
+   draw->render->draw( draw->render, 
+                       draw_elts, 
+                       draw_count );
+   
+
+
+   draw->render->release_vertices( draw->render, 
+				   hw_verts, 
+				   fse->key.output_stride, 
+				   count );
+
+   return TRUE;
+}
+
+
+
+static void fse_finish( struct draw_pt_middle_end *middle )
+{
+}
+
+
+static void
+fse_destroy( struct draw_pt_middle_end *middle ) 
+{
+   FREE(middle);
+}
+
+struct draw_pt_middle_end *draw_pt_middle_fse( struct draw_context *draw )
+{
+   struct fetch_shade_emit *fse = CALLOC_STRUCT(fetch_shade_emit);
+   if (!fse)
+      return NULL;
+
+   fse->base.prepare = fse_prepare;
+   fse->base.run = fse_run;
+   fse->base.run_linear = fse_run_linear;
+   fse->base.run_linear_elts = fse_run_linear_elts;
+   fse->base.finish = fse_finish;
+   fse->base.destroy = fse_destroy;
+   fse->draw = draw;
+
+   return &fse->base;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c
new file mode 100644
index 00000000000..ec3b41c320a
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c
@@ -0,0 +1,396 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "draw/draw_context.h"
+#include "draw/draw_vbuf.h"
+#include "draw/draw_vertex.h"
+#include "draw/draw_pt.h"
+#include "draw/draw_vs.h"
+#include "translate/translate.h"
+
+
+struct fetch_pipeline_middle_end {
+   struct draw_pt_middle_end base;
+   struct draw_context *draw;
+
+   struct pt_emit *emit;
+   struct pt_fetch *fetch;
+   struct pt_post_vs *post_vs;
+
+   unsigned vertex_data_offset;
+   unsigned vertex_size;
+   unsigned prim;
+   unsigned opt;
+};
+
+
+static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle,
+                                    unsigned prim,
+				    unsigned opt,
+                                    unsigned *max_vertices )
+{
+   struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle;
+   struct draw_context *draw = fpme->draw;
+   struct draw_vertex_shader *vs = draw->vs.vertex_shader;
+
+   /* Add one to num_outputs because the pipeline occasionally tags on
+    * an additional texcoord, eg for AA lines.
+    */
+   unsigned nr = MAX2( vs->info.num_inputs,
+		       vs->info.num_outputs + 1 );
+
+   fpme->prim = prim;
+   fpme->opt = opt;
+
+   /* Always leave room for the vertex header whether we need it or
+    * not.  It's hard to get rid of it in particular because of the
+    * viewport code in draw_pt_post_vs.c.  
+    */
+   fpme->vertex_size = sizeof(struct vertex_header) + nr * 4 * sizeof(float);
+
+   
+
+   draw_pt_fetch_prepare( fpme->fetch, 
+			  fpme->vertex_size );
+
+   /* XXX: it's not really gl rasterization rules we care about here,
+    * but gl vs dx9 clip spaces.
+    */
+   draw_pt_post_vs_prepare( fpme->post_vs,
+			    (boolean)draw->bypass_clipping,
+			    (boolean)draw->identity_viewport,
+			    (boolean)draw->rasterizer->gl_rasterization_rules );
+			    
+
+   if (!(opt & PT_PIPELINE)) {
+      draw_pt_emit_prepare( fpme->emit, 
+			    prim,
+                            max_vertices );
+
+      *max_vertices = MAX2( *max_vertices,
+                            DRAW_PIPE_MAX_VERTICES );
+   }
+   else {
+      *max_vertices = DRAW_PIPE_MAX_VERTICES; 
+   }
+
+   /* return even number */
+   *max_vertices = *max_vertices & ~1;
+
+   /* No need to prepare the shader.
+    */
+   vs->prepare(vs, draw);
+}
+
+
+
+static void fetch_pipeline_run( struct draw_pt_middle_end *middle,
+                                const unsigned *fetch_elts,
+                                unsigned fetch_count,
+                                const ushort *draw_elts,
+                                unsigned draw_count )
+{
+   struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle;
+   struct draw_context *draw = fpme->draw;
+   struct draw_vertex_shader *shader = draw->vs.vertex_shader;
+   unsigned opt = fpme->opt;
+   unsigned alloc_count = align( fetch_count, 4 );
+
+   struct vertex_header *pipeline_verts = 
+      (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count);
+
+   if (!pipeline_verts) {
+      /* Not much we can do here - just skip the rendering.
+       */
+      assert(0);
+      return;
+   }
+
+   /* Fetch into our vertex buffer
+    */
+   draw_pt_fetch_run( fpme->fetch,
+		      fetch_elts, 
+		      fetch_count,
+		      (char *)pipeline_verts );
+
+   /* Run the shader, note that this overwrites the data[] parts of
+    * the pipeline verts.  If there is no shader, ie a bypass shader,
+    * then the inputs == outputs, and are already in the correct
+    * place.
+    */
+   if (opt & PT_SHADE)
+   {
+      shader->run_linear(shader, 
+			 (const float (*)[4])pipeline_verts->data,
+			 (      float (*)[4])pipeline_verts->data,
+			 (const float (*)[4])draw->pt.user.constants,
+			 fetch_count,
+			 fpme->vertex_size,
+			 fpme->vertex_size);
+   }
+
+   if (draw_pt_post_vs_run( fpme->post_vs,
+			    pipeline_verts,
+			    fetch_count,
+			    fpme->vertex_size ))
+   {
+      opt |= PT_PIPELINE;
+   }
+
+   /* Do we need to run the pipeline?
+    */
+   if (opt & PT_PIPELINE) {
+      draw_pipeline_run( fpme->draw,
+                         fpme->prim,
+                         pipeline_verts,
+                         fetch_count,
+                         fpme->vertex_size,
+                         draw_elts,
+                         draw_count );
+   }
+   else {
+      draw_pt_emit( fpme->emit,
+		    (const float (*)[4])pipeline_verts->data,
+		    fetch_count,
+		    fpme->vertex_size,
+		    draw_elts,
+		    draw_count );
+   }
+
+
+   FREE(pipeline_verts);
+}
+
+
+static void fetch_pipeline_linear_run( struct draw_pt_middle_end *middle,
+                                       unsigned start,
+                                       unsigned count)
+{
+   struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle;
+   struct draw_context *draw = fpme->draw;
+   struct draw_vertex_shader *shader = draw->vs.vertex_shader;
+   unsigned opt = fpme->opt;
+   unsigned alloc_count = align( count, 4 );
+
+   struct vertex_header *pipeline_verts =
+      (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count);
+
+   if (!pipeline_verts) {
+      /* Not much we can do here - just skip the rendering.
+       */
+      assert(0);
+      return;
+   }
+
+   /* Fetch into our vertex buffer
+    */
+   draw_pt_fetch_run_linear( fpme->fetch,
+                             start,
+                             count,
+                             (char *)pipeline_verts );
+
+   /* Run the shader, note that this overwrites the data[] parts of
+    * the pipeline verts.  If there is no shader, ie a bypass shader,
+    * then the inputs == outputs, and are already in the correct
+    * place.
+    */
+   if (opt & PT_SHADE)
+   {
+      shader->run_linear(shader,
+			 (const float (*)[4])pipeline_verts->data,
+			 (      float (*)[4])pipeline_verts->data,
+			 (const float (*)[4])draw->pt.user.constants,
+			 count,
+			 fpme->vertex_size,
+			 fpme->vertex_size);
+   }
+
+   if (draw_pt_post_vs_run( fpme->post_vs,
+			    pipeline_verts,
+			    count,
+			    fpme->vertex_size ))
+   {
+      opt |= PT_PIPELINE;
+   }
+
+   /* Do we need to run the pipeline?
+    */
+   if (opt & PT_PIPELINE) {
+      draw_pipeline_run_linear( fpme->draw,
+                                fpme->prim,
+                                pipeline_verts,
+                                count,
+                                fpme->vertex_size);
+   }
+   else {
+      draw_pt_emit_linear( fpme->emit,
+                           (const float (*)[4])pipeline_verts->data,
+                           count,
+                           fpme->vertex_size,
+                           0, /*start*/
+                           count );
+   }
+
+   FREE(pipeline_verts);
+}
+
+
+
+static boolean fetch_pipeline_linear_run_elts( struct draw_pt_middle_end *middle,
+                                            unsigned start,
+                                            unsigned count,
+                                            const ushort *draw_elts,
+                                            unsigned draw_count )
+{
+   struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle;
+   struct draw_context *draw = fpme->draw;
+   struct draw_vertex_shader *shader = draw->vs.vertex_shader;
+   unsigned opt = fpme->opt;
+   unsigned alloc_count = align( count, 4 );
+
+   struct vertex_header *pipeline_verts =
+      (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count);
+
+   if (!pipeline_verts) 
+      return FALSE;
+
+   /* Fetch into our vertex buffer
+    */
+   draw_pt_fetch_run_linear( fpme->fetch,
+                             start,
+                             count,
+                             (char *)pipeline_verts );
+
+   /* Run the shader, note that this overwrites the data[] parts of
+    * the pipeline verts.  If there is no shader, ie a bypass shader,
+    * then the inputs == outputs, and are already in the correct
+    * place.
+    */
+   if (opt & PT_SHADE)
+   {
+      shader->run_linear(shader,
+			 (const float (*)[4])pipeline_verts->data,
+			 (      float (*)[4])pipeline_verts->data,
+			 (const float (*)[4])draw->pt.user.constants,
+			 count,
+			 fpme->vertex_size,
+			 fpme->vertex_size);
+   }
+
+   if (draw_pt_post_vs_run( fpme->post_vs,
+			    pipeline_verts,
+			    count,
+			    fpme->vertex_size ))
+   {
+      opt |= PT_PIPELINE;
+   }
+
+   /* Do we need to run the pipeline?
+    */
+   if (opt & PT_PIPELINE) {
+      draw_pipeline_run( fpme->draw,
+                         fpme->prim,
+                         pipeline_verts,
+                         count,
+                         fpme->vertex_size,
+                         draw_elts,
+                         draw_count );
+   }
+   else {
+      draw_pt_emit( fpme->emit,
+		    (const float (*)[4])pipeline_verts->data,
+		    count,
+		    fpme->vertex_size,
+		    draw_elts,
+		    draw_count );
+   }
+
+   FREE(pipeline_verts);
+   return TRUE;
+}
+
+
+
+static void fetch_pipeline_finish( struct draw_pt_middle_end *middle )
+{
+   /* nothing to do */
+}
+
+static void fetch_pipeline_destroy( struct draw_pt_middle_end *middle )
+{
+   struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle;
+
+   if (fpme->fetch)
+      draw_pt_fetch_destroy( fpme->fetch );
+
+   if (fpme->emit)
+      draw_pt_emit_destroy( fpme->emit );
+
+   if (fpme->post_vs)
+      draw_pt_post_vs_destroy( fpme->post_vs );
+
+   FREE(middle);
+}
+
+
+struct draw_pt_middle_end *draw_pt_fetch_pipeline_or_emit( struct draw_context *draw )
+{
+   struct fetch_pipeline_middle_end *fpme = CALLOC_STRUCT( fetch_pipeline_middle_end );
+   if (!fpme)
+      goto fail;
+
+   fpme->base.prepare        = fetch_pipeline_prepare;
+   fpme->base.run            = fetch_pipeline_run;
+   fpme->base.run_linear     = fetch_pipeline_linear_run;
+   fpme->base.run_linear_elts = fetch_pipeline_linear_run_elts;
+   fpme->base.finish         = fetch_pipeline_finish;
+   fpme->base.destroy        = fetch_pipeline_destroy;
+
+   fpme->draw = draw;
+
+   fpme->fetch = draw_pt_fetch_create( draw );
+   if (!fpme->fetch)
+      goto fail;
+
+   fpme->post_vs = draw_pt_post_vs_create( draw );
+   if (!fpme->post_vs)
+      goto fail;
+
+   fpme->emit = draw_pt_emit_create( draw );
+   if (!fpme->emit) 
+      goto fail;
+
+   return &fpme->base;
+
+ fail:
+   if (fpme)
+      fetch_pipeline_destroy( &fpme->base );
+
+   return NULL;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pt_post_vs.c b/src/gallium/auxiliary/draw/draw_pt_post_vs.c
new file mode 100644
index 00000000000..96dc706b998
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pt_post_vs.c
@@ -0,0 +1,233 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include "util/u_memory.h"
+#include "pipe/p_context.h"
+#include "draw/draw_context.h"
+#include "draw/draw_private.h"
+#include "draw/draw_vbuf.h"
+#include "draw/draw_vertex.h"
+#include "draw/draw_pt.h"
+
+struct pt_post_vs {
+   struct draw_context *draw;
+
+   boolean (*run)( struct pt_post_vs *pvs,
+		struct vertex_header *vertices,
+		unsigned count,
+		unsigned stride );
+};
+
+
+
+static INLINE float
+dot4(const float *a, const float *b)
+{
+   return (a[0]*b[0] +
+           a[1]*b[1] +
+           a[2]*b[2] +
+           a[3]*b[3]);
+}
+
+
+
+static INLINE unsigned
+compute_clipmask_gl(const float *clip, /*const*/ float plane[][4], unsigned nr)
+{
+   unsigned mask = 0x0;
+   unsigned i;
+
+#if 0
+   debug_printf("compute clipmask %f %f %f %f\n",
+                clip[0], clip[1], clip[2], clip[3]);
+   assert(clip[3] != 0.0);
+#endif
+
+   /* Do the hardwired planes first:
+    */
+   if (-clip[0] + clip[3] < 0) mask |= (1<<0);
+   if ( clip[0] + clip[3] < 0) mask |= (1<<1);
+   if (-clip[1] + clip[3] < 0) mask |= (1<<2);
+   if ( clip[1] + clip[3] < 0) mask |= (1<<3);
+   if ( clip[2] + clip[3] < 0) mask |= (1<<4); /* match mesa clipplane numbering - for now */
+   if (-clip[2] + clip[3] < 0) mask |= (1<<5); /* match mesa clipplane numbering - for now */
+
+   /* Followed by any remaining ones:
+    */
+   for (i = 6; i < nr; i++) {
+      if (dot4(clip, plane[i]) < 0) 
+         mask |= (1<<i);
+   }
+
+   return mask;
+}
+
+
+/* The normal case - cliptest, rhw divide, viewport transform.
+ *
+ * Also handle identity viewport here at the expense of a few wasted
+ * instructions
+ */
+static boolean post_vs_cliptest_viewport_gl( struct pt_post_vs *pvs,
+					  struct vertex_header *vertices,
+					  unsigned count,
+					  unsigned stride )
+{
+   struct vertex_header *out = vertices;
+   const float *scale = pvs->draw->viewport.scale;
+   const float *trans = pvs->draw->viewport.translate;
+   const unsigned pos = pvs->draw->vs.position_output;
+   unsigned clipped = 0;
+   unsigned j;
+
+   if (0) debug_printf("%s\n");
+
+   for (j = 0; j < count; j++) {
+      float *position = out->data[pos];
+
+      out->clip[0] = position[0];
+      out->clip[1] = position[1];
+      out->clip[2] = position[2];
+      out->clip[3] = position[3];
+
+      out->vertex_id = 0xffff;
+      out->clipmask = compute_clipmask_gl(out->clip, 
+					  pvs->draw->plane,
+					  pvs->draw->nr_planes);
+      clipped += out->clipmask;
+
+      if (out->clipmask == 0)
+      {
+	 /* divide by w */
+	 float w = 1.0f / position[3];
+
+	 /* Viewport mapping */
+	 position[0] = position[0] * w * scale[0] + trans[0];
+	 position[1] = position[1] * w * scale[1] + trans[1];
+	 position[2] = position[2] * w * scale[2] + trans[2];
+	 position[3] = w;
+#if 0
+         debug_printf("post viewport: %f %f %f %f\n",
+                      position[0],
+                      position[1],
+                      position[2],
+                      position[3]);
+#endif
+      }
+
+      out = (struct vertex_header *)( (char *)out + stride );
+   }
+
+   return clipped != 0;
+}
+
+
+
+/* If bypass_clipping is set, skip cliptest and rhw divide.
+ */
+static boolean post_vs_viewport( struct pt_post_vs *pvs,
+			      struct vertex_header *vertices,
+			      unsigned count,
+			      unsigned stride )
+{
+   struct vertex_header *out = vertices;
+   const float *scale = pvs->draw->viewport.scale;
+   const float *trans = pvs->draw->viewport.translate;
+   const unsigned pos = pvs->draw->vs.position_output;
+   unsigned j;
+
+   if (0) debug_printf("%s\n", __FUNCTION__);
+   for (j = 0; j < count; j++) {
+      float *position = out->data[pos];
+
+      /* Viewport mapping only, no cliptest/rhw divide
+       */
+      position[0] = position[0] * scale[0] + trans[0];
+      position[1] = position[1] * scale[1] + trans[1];
+      position[2] = position[2] * scale[2] + trans[2];
+
+      out = (struct vertex_header *)((char *)out + stride);
+   }
+   
+   return FALSE;
+}
+
+
+/* If bypass_clipping is set and we have an identity viewport, nothing
+ * to do.
+ */
+static boolean post_vs_none( struct pt_post_vs *pvs,
+			     struct vertex_header *vertices,
+			     unsigned count,
+			     unsigned stride )
+{
+   if (0) debug_printf("%s\n", __FUNCTION__);
+   return FALSE;
+}
+
+boolean draw_pt_post_vs_run( struct pt_post_vs *pvs,
+			     struct vertex_header *pipeline_verts,
+			     unsigned count,
+			     unsigned stride )
+{
+   return pvs->run( pvs, pipeline_verts, count, stride );
+}
+
+
+void draw_pt_post_vs_prepare( struct pt_post_vs *pvs,
+			      boolean bypass_clipping,
+			      boolean identity_viewport,
+			      boolean opengl )
+{
+   if (bypass_clipping) {
+      if (identity_viewport)
+	 pvs->run = post_vs_none;
+      else
+	 pvs->run = post_vs_viewport;
+   }
+   else {
+      //if (opengl) 
+      pvs->run = post_vs_cliptest_viewport_gl;
+   }
+}
+
+
+struct pt_post_vs *draw_pt_post_vs_create( struct draw_context *draw )
+{
+   struct pt_post_vs *pvs = CALLOC_STRUCT( pt_post_vs );
+   if (!pvs)
+      return NULL;
+
+   pvs->draw = draw;
+   
+   return pvs;
+}
+
+void draw_pt_post_vs_destroy( struct pt_post_vs *pvs )
+{
+   FREE(pvs);
+}
diff --git a/src/gallium/auxiliary/draw/draw_pt_util.c b/src/gallium/auxiliary/draw/draw_pt_util.c
new file mode 100644
index 00000000000..3bc7939c556
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pt_util.c
@@ -0,0 +1,102 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  */
+
+#include "draw/draw_context.h"
+#include "draw/draw_private.h"
+#include "draw/draw_pt.h"
+
+void draw_pt_split_prim(unsigned prim, unsigned *first, unsigned *incr)
+{
+   switch (prim) {
+   case PIPE_PRIM_POINTS:
+      *first = 1;
+      *incr = 1;
+      break;
+   case PIPE_PRIM_LINES:
+      *first = 2;
+      *incr = 2;
+      break;
+   case PIPE_PRIM_LINE_STRIP:
+   case PIPE_PRIM_LINE_LOOP:
+      *first = 2;
+      *incr = 1;
+      break;
+   case PIPE_PRIM_TRIANGLES:
+      *first = 3;
+      *incr = 3;
+      break;
+   case PIPE_PRIM_TRIANGLE_STRIP:
+   case PIPE_PRIM_TRIANGLE_FAN:
+   case PIPE_PRIM_POLYGON:
+      *first = 3;
+      *incr = 1;
+      break;
+   case PIPE_PRIM_QUADS:
+      *first = 4;
+      *incr = 4;
+      break;
+   case PIPE_PRIM_QUAD_STRIP:
+      *first = 4;
+      *incr = 2;
+      break;
+   default:
+      assert(0);
+      *first = 0;
+      *incr = 1;		/* set to one so that count % incr works */
+      break;
+   }
+}
+
+
+unsigned draw_pt_reduced_prim(unsigned prim)
+{
+   switch (prim) {
+   case PIPE_PRIM_POINTS:
+      return PIPE_PRIM_POINTS;
+   case PIPE_PRIM_LINES:
+   case PIPE_PRIM_LINE_STRIP:
+   case PIPE_PRIM_LINE_LOOP:
+      return PIPE_PRIM_LINES;
+   case PIPE_PRIM_TRIANGLES:
+   case PIPE_PRIM_TRIANGLE_STRIP:
+   case PIPE_PRIM_TRIANGLE_FAN:
+   case PIPE_PRIM_POLYGON:
+   case PIPE_PRIM_QUADS:
+   case PIPE_PRIM_QUAD_STRIP:
+      return PIPE_PRIM_TRIANGLES;
+   default:
+      assert(0);
+      return PIPE_PRIM_POINTS;
+   }
+}
+
+
diff --git a/src/gallium/auxiliary/draw/draw_pt_varray.c b/src/gallium/auxiliary/draw/draw_pt_varray.c
new file mode 100644
index 00000000000..d0e16c9bc3c
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pt_varray.c
@@ -0,0 +1,193 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include "util/u_math.h"
+#include "util/u_memory.h"
+
+#include "draw/draw_context.h"
+#include "draw/draw_private.h"
+#include "draw/draw_pt.h"
+
+#define FETCH_MAX 256
+#define DRAW_MAX (FETCH_MAX+8)
+
+struct varray_frontend {
+   struct draw_pt_front_end base;
+   struct draw_context *draw;
+
+   ushort draw_elts[DRAW_MAX];
+   unsigned fetch_elts[FETCH_MAX];
+
+   unsigned driver_fetch_max;
+   unsigned fetch_max;
+
+   struct draw_pt_middle_end *middle;
+
+   unsigned input_prim;
+   unsigned output_prim;
+};
+
+
+static void varray_flush_linear(struct varray_frontend *varray,
+                                unsigned start, unsigned count)
+{
+   if (count) {
+      assert(varray->middle->run_linear);
+      varray->middle->run_linear(varray->middle, start, count);
+   }
+}
+
+static void varray_line_loop_segment(struct varray_frontend *varray,
+                                     unsigned start,
+                                     unsigned segment_start,
+                                     unsigned segment_count,
+                                     boolean end )
+{
+   assert(segment_count < varray->fetch_max);
+   if (segment_count >= 1) {
+      unsigned nr = 0, i;
+
+      for (i = 0; i < segment_count; i++) 
+         varray->fetch_elts[nr++] = start + segment_start + i;
+
+      if (end) 
+         varray->fetch_elts[nr++] = start;
+
+      assert(nr <= FETCH_MAX);
+
+      varray->middle->run(varray->middle, 
+                          varray->fetch_elts,
+                          nr,
+                          varray->draw_elts, /* ie. linear */
+                          nr);
+   }
+}
+
+
+
+static void varray_fan_segment(struct varray_frontend *varray,
+                               unsigned start, 
+                               unsigned segment_start,
+                               unsigned segment_count )
+{
+   assert(segment_count < varray->fetch_max);
+   if (segment_count >= 2) {
+      unsigned nr = 0, i;
+
+      if (segment_start != 0)
+         varray->fetch_elts[nr++] = start;
+
+      for (i = 0 ; i < segment_count; i++) 
+         varray->fetch_elts[nr++] = start + segment_start + i;
+
+      assert(nr <= FETCH_MAX);
+
+      varray->middle->run(varray->middle, 
+                          varray->fetch_elts,
+                          nr,
+                          varray->draw_elts, /* ie. linear */
+                          nr);
+   }
+}
+
+
+
+
+#define FUNC varray_run
+#include "draw_pt_varray_tmp_linear.h"
+
+static unsigned decompose_prim[PIPE_PRIM_POLYGON + 1] = {
+   PIPE_PRIM_POINTS,
+   PIPE_PRIM_LINES,
+   PIPE_PRIM_LINE_STRIP,        /* decomposed LINELOOP */
+   PIPE_PRIM_LINE_STRIP,
+   PIPE_PRIM_TRIANGLES,
+   PIPE_PRIM_TRIANGLE_STRIP,
+   PIPE_PRIM_TRIANGLE_FAN, 
+   PIPE_PRIM_QUADS,
+   PIPE_PRIM_QUAD_STRIP,
+   PIPE_PRIM_POLYGON
+};
+
+
+
+static void varray_prepare(struct draw_pt_front_end *frontend,
+                           unsigned prim,
+                           struct draw_pt_middle_end *middle,
+                           unsigned opt)
+{
+   struct varray_frontend *varray = (struct varray_frontend *)frontend;
+
+   varray->base.run = varray_run;
+
+   varray->input_prim = prim;
+   varray->output_prim = decompose_prim[prim];
+
+   varray->middle = middle;
+   middle->prepare(middle, varray->output_prim, opt, &varray->driver_fetch_max );
+
+   /* check that the max is even */
+   assert((varray->driver_fetch_max & 1) == 0);
+
+   varray->fetch_max = MIN2(FETCH_MAX, varray->driver_fetch_max);
+}
+
+
+
+
+static void varray_finish(struct draw_pt_front_end *frontend)
+{
+   struct varray_frontend *varray = (struct varray_frontend *)frontend;
+   varray->middle->finish(varray->middle);
+   varray->middle = NULL;
+}
+
+static void varray_destroy(struct draw_pt_front_end *frontend)
+{
+   FREE(frontend);
+}
+
+
+struct draw_pt_front_end *draw_pt_varray(struct draw_context *draw)
+{
+   ushort i;
+   struct varray_frontend *varray = CALLOC_STRUCT(varray_frontend);
+   if (varray == NULL)
+      return NULL;
+
+   varray->base.prepare = varray_prepare;
+   varray->base.run     = NULL;
+   varray->base.finish  = varray_finish;
+   varray->base.destroy = varray_destroy;
+   varray->draw = draw;
+
+   for (i = 0; i < DRAW_MAX; i++) {
+      varray->draw_elts[i] = i;
+   }
+
+   return &varray->base;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h
new file mode 100644
index 00000000000..7c722457c3c
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h
@@ -0,0 +1,238 @@
+
+static void FUNC(struct draw_pt_front_end *frontend,
+                 pt_elt_func get_elt,
+                 const void *elts,
+                 unsigned count)
+{
+   struct varray_frontend *varray = (struct varray_frontend *)frontend;
+   struct draw_context *draw = varray->draw;
+   unsigned start = (unsigned)elts;
+
+   boolean flatfirst = (draw->rasterizer->flatshade &&
+                        draw->rasterizer->flatshade_first);
+   unsigned i, j;
+   ushort flags;
+   unsigned first, incr;
+
+   varray->fetch_start = start;
+
+   draw_pt_split_prim(varray->input_prim, &first, &incr);
+
+#if 0
+   debug_printf("%s (%d) %d/%d\n", __FUNCTION__,
+                varray->input_prim,
+                start, count);
+#endif
+
+   switch (varray->input_prim) {
+   case PIPE_PRIM_POINTS:
+      for (j = 0; j + first <= count; j += i) {
+         unsigned end = MIN2(FETCH_MAX, count - j);
+         end -= (end % incr);
+         for (i = 0; i < end; i++) {
+            POINT(varray, i + 0);
+         }
+         i = end;
+         fetch_init(varray, end);
+         varray_flush(varray);
+      }
+      break;
+
+   case PIPE_PRIM_LINES:
+      for (j = 0; j + first <= count; j += i) {
+         unsigned end = MIN2(FETCH_MAX, count - j);
+         end -= (end % incr);
+         for (i = 0; i+1 < end; i += 2) {
+            LINE(varray, DRAW_PIPE_RESET_STIPPLE,
+                 i + 0, i + 1);
+         }
+         i = end;
+         fetch_init(varray, end);
+         varray_flush(varray);
+      }
+      break;
+
+   case PIPE_PRIM_LINE_LOOP:
+      if (count >= 2) {
+         flags = DRAW_PIPE_RESET_STIPPLE;
+
+         for (j = 0; j + first <= count; j += i) {
+            unsigned end = MIN2(FETCH_MAX, count - j);
+            end -= (end % incr);
+            for (i = 1; i < end; i++, flags = 0) {
+               LINE(varray, flags, i - 1, i);
+            }
+            LINE(varray, flags, i - 1, 0);
+            i = end;
+            fetch_init(varray, end);
+            varray_flush(varray);
+         }
+      }
+      break;
+
+   case PIPE_PRIM_LINE_STRIP:
+      flags = DRAW_PIPE_RESET_STIPPLE;
+      for (j = 0; j + first <= count; j += i) {
+         unsigned end = MIN2(FETCH_MAX, count - j);
+         end -= (end % incr);
+         for (i = 1; i < end; i++, flags = 0) {
+            LINE(varray, flags, i - 1, i);
+         }
+         i = end;
+         fetch_init(varray, end);
+         varray_flush(varray);
+      }
+      break;
+
+   case PIPE_PRIM_TRIANGLES:
+      for (j = 0; j + first <= count; j += i) {
+         unsigned end = MIN2(FETCH_MAX, count - j);
+         end -= (end % incr);
+         for (i = 0; i+2 < end; i += 3) {
+            TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL,
+                     i + 0, i + 1, i + 2);
+         }
+         i = end;
+         fetch_init(varray, end);
+         varray_flush(varray);
+      }
+      break;
+
+   case PIPE_PRIM_TRIANGLE_STRIP:
+      if (flatfirst) {
+         for (j = 0; j + first <= count; j += i) {
+            unsigned end = MIN2(FETCH_MAX, count - j);
+            end -= (end % incr);
+            for (i = 0; i+2 < end; i++) {
+               TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL,
+                        i + 0, i + 1 + (i&1), i + 2 - (i&1));
+            }
+            i = end;
+            fetch_init(varray, end);
+            varray_flush(varray);
+            if (j + first + i <= count) {
+               varray->fetch_start -= 2;
+               i -= 2;
+            }
+         }
+      }
+      else {
+         for (j = 0; j + first <= count; j += i) {
+            unsigned end = MIN2(FETCH_MAX, count - j);
+            end -= (end  % incr);
+            for (i = 0; i + 2 < end; i++) {
+               TRIANGLE(varray, DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL,
+                        i + 0 + (i&1), i + 1 - (i&1), i + 2);
+            }
+            i = end;
+            fetch_init(varray, end);
+            varray_flush(varray);
+            if (j + first + i <= count) {
+               varray->fetch_start -= 2;
+               i -= 2;
+            }
+         }
+      }
+      break;
+
+   case PIPE_PRIM_TRIANGLE_FAN:
+      if (count >= 3) {
+         if (flatfirst) {
+            flags = DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL;
+            for (j = 0; j + first <= count; j += i) {
+               unsigned end = MIN2(FETCH_MAX, count - j);
+               end -= (end % incr);
+               for (i = 0; i+2 < end; i++) {
+                  TRIANGLE(varray, flags, i + 1, i + 2, 0);
+               }
+               i = end;
+               fetch_init(varray, end);
+               varray_flush(varray);
+            }
+         }
+         else {
+            flags = DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL;
+            for (j = 0; j + first <= count; j += i) {
+               unsigned end = MIN2(FETCH_MAX, count - j);
+               end -= (end % incr);
+               for (i = 0; i+2 < end; i++) {
+                  TRIANGLE(varray, flags, 0, i + 1, i + 2);
+               }
+               i = end;
+               fetch_init(varray, end);
+               varray_flush(varray);
+            }
+         }
+      }
+      break;
+
+   case PIPE_PRIM_QUADS:
+      for (j = 0; j + first <= count; j += i) {
+         unsigned end = MIN2(FETCH_MAX, count - j);
+         end -= (end % incr);
+         for (i = 0; i+3 < end; i += 4) {
+            QUAD(varray, i + 0, i + 1, i + 2, i + 3);
+         }
+         i = end;
+         fetch_init(varray, end);
+         varray_flush(varray);
+      }
+      break;
+
+   case PIPE_PRIM_QUAD_STRIP:
+      for (j = 0; j + first <= count; j += i) {
+         unsigned end = MIN2(FETCH_MAX, count - j);
+         end -= (end % incr);
+         for (i = 0; i+3 < end; i += 2) {
+            QUAD(varray, i + 2, i + 0, i + 1, i + 3);
+         }
+         i = end;
+         fetch_init(varray, end);
+         varray_flush(varray);
+         if (j + first + i <= count) {
+            varray->fetch_start -= 2;
+            i -= 2;
+         }
+      }
+      break;
+
+   case PIPE_PRIM_POLYGON:
+   {
+      /* These bitflags look a little odd because we submit the
+       * vertices as (1,2,0) to satisfy flatshade requirements.
+       */
+      const ushort edge_first  = DRAW_PIPE_EDGE_FLAG_2;
+      const ushort edge_middle = DRAW_PIPE_EDGE_FLAG_0;
+      const ushort edge_last   = DRAW_PIPE_EDGE_FLAG_1;
+
+      flags = DRAW_PIPE_RESET_STIPPLE | edge_first | edge_middle;
+      for (j = 0; j + first <= count; j += i) {
+         unsigned end = MIN2(FETCH_MAX, count - j);
+         end -= (end % incr);
+         for (i = 0; i+2 < end; i++, flags = edge_middle) {
+
+            if (i + 3 == count)
+               flags |= edge_last;
+
+            TRIANGLE(varray, flags, i + 1, i + 2, 0);
+         }
+         i = end;
+         fetch_init(varray, end);
+         varray_flush(varray);
+      }
+   }
+   break;
+
+   default:
+      assert(0);
+      break;
+   }
+
+   varray_flush(varray);
+}
+
+#undef TRIANGLE
+#undef QUAD
+#undef POINT
+#undef LINE
+#undef FUNC
diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h
new file mode 100644
index 00000000000..55a8e6521dc
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp_linear.h
@@ -0,0 +1,91 @@
+static unsigned trim( unsigned count, unsigned first, unsigned incr )
+{
+   return count - (count - first) % incr; 
+}
+
+static void FUNC(struct draw_pt_front_end *frontend,
+                 pt_elt_func get_elt,
+                 const void *elts,
+                 unsigned count)
+{
+   struct varray_frontend *varray = (struct varray_frontend *)frontend;
+   unsigned start = (unsigned)elts;
+
+   unsigned j;
+   unsigned first, incr;
+
+   draw_pt_split_prim(varray->input_prim, &first, &incr);
+   
+   /* Sanitize primitive length:
+    */
+   count = trim(count, first, incr); 
+   if (count < first)
+      return;
+
+#if 0
+   debug_printf("%s (%d) %d/%d\n", __FUNCTION__,
+                varray->input_prim,
+                start, count);
+#endif
+
+   switch (varray->input_prim) {
+   case PIPE_PRIM_POINTS:
+   case PIPE_PRIM_LINES:
+   case PIPE_PRIM_TRIANGLES:
+   case PIPE_PRIM_LINE_STRIP:
+   case PIPE_PRIM_TRIANGLE_STRIP:
+   case PIPE_PRIM_QUADS:
+   case PIPE_PRIM_QUAD_STRIP:
+      for (j = 0; j < count;) {
+         unsigned remaining = count - j;
+         unsigned nr = trim( MIN2(varray->driver_fetch_max, remaining), first, incr );
+         varray_flush_linear(varray, start + j, nr);
+         j += nr;
+         if (nr != remaining) 
+            j -= (first - incr);
+      }
+      break;
+
+   case PIPE_PRIM_LINE_LOOP:
+      /* Always have to decompose as we've stated that this will be
+       * emitted as a line-strip.
+       */
+      for (j = 0; j < count;) {
+         unsigned remaining = count - j;
+         unsigned nr = trim( MIN2(varray->fetch_max-1, remaining), first, incr );
+         varray_line_loop_segment(varray, start, j, nr, nr == remaining);
+         j += nr;
+         if (nr != remaining) 
+            j -= (first - incr);
+      }
+      break;
+
+
+   case PIPE_PRIM_POLYGON:
+   case PIPE_PRIM_TRIANGLE_FAN: 
+      if (count < varray->driver_fetch_max) {
+         varray_flush_linear(varray, start, count);
+      }
+      else {
+         for ( j = 0; j < count;) {
+            unsigned remaining = count - j;
+            unsigned nr = trim( MIN2(varray->fetch_max-1, remaining), first, incr );
+            varray_fan_segment(varray, start, j, nr);
+            j += nr;
+            if (nr != remaining) 
+               j -= (first - incr);
+         }
+      }
+      break;
+
+   default:
+      assert(0);
+      break;
+   }
+}
+
+#undef TRIANGLE
+#undef QUAD
+#undef POINT
+#undef LINE
+#undef FUNC
diff --git a/src/gallium/auxiliary/draw/draw_pt_vcache.c b/src/gallium/auxiliary/draw/draw_pt_vcache.c
new file mode 100644
index 00000000000..5d268a22264
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pt_vcache.c
@@ -0,0 +1,501 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  */
+
+#include "util/u_memory.h"
+#include "draw/draw_context.h"
+#include "draw/draw_private.h"
+#include "draw/draw_pt.h"
+
+
+#define CACHE_MAX 256
+#define FETCH_MAX 256
+#define DRAW_MAX (16*1024)
+
+struct vcache_frontend {
+   struct draw_pt_front_end base;
+   struct draw_context *draw;
+
+   unsigned in[CACHE_MAX];
+   ushort out[CACHE_MAX];
+
+   ushort draw_elts[DRAW_MAX];
+   unsigned fetch_elts[FETCH_MAX];
+
+   unsigned draw_count;
+   unsigned fetch_count;
+   unsigned fetch_max;
+   
+   struct draw_pt_middle_end *middle;
+
+   unsigned input_prim;
+   unsigned output_prim;
+
+   unsigned middle_prim;
+   unsigned opt;
+};
+
+static INLINE void 
+vcache_flush( struct vcache_frontend *vcache )
+{
+   if (vcache->middle_prim != vcache->output_prim) {
+      vcache->middle_prim = vcache->output_prim;
+      vcache->middle->prepare( vcache->middle, 
+                               vcache->middle_prim, 
+                               vcache->opt, 
+                               &vcache->fetch_max );
+   }
+
+   if (vcache->draw_count) {
+      vcache->middle->run( vcache->middle,
+                           vcache->fetch_elts,
+                           vcache->fetch_count,
+                           vcache->draw_elts,
+                           vcache->draw_count );
+   }
+
+   memset(vcache->in, ~0, sizeof(vcache->in));
+   vcache->fetch_count = 0;
+   vcache->draw_count = 0;
+}
+
+static INLINE void 
+vcache_check_flush( struct vcache_frontend *vcache )
+{
+   if ( vcache->draw_count + 6 >= DRAW_MAX ||
+        vcache->fetch_count + 4 >= FETCH_MAX )
+   {
+      vcache_flush( vcache );
+   }
+}
+
+
+static INLINE void 
+vcache_elt( struct vcache_frontend *vcache,
+            unsigned felt,
+            ushort flags )
+{
+   unsigned idx = felt % CACHE_MAX;
+
+   if (vcache->in[idx] != felt) {
+      assert(vcache->fetch_count < FETCH_MAX);
+
+      vcache->in[idx] = felt;
+      vcache->out[idx] = (ushort)vcache->fetch_count;
+      vcache->fetch_elts[vcache->fetch_count++] = felt;
+   }
+
+   vcache->draw_elts[vcache->draw_count++] = vcache->out[idx] | flags;
+}
+
+
+                   
+static INLINE void 
+vcache_triangle( struct vcache_frontend *vcache,
+                 unsigned i0,
+                 unsigned i1,
+                 unsigned i2 )
+{
+   vcache_elt(vcache, i0, 0);
+   vcache_elt(vcache, i1, 0);
+   vcache_elt(vcache, i2, 0);
+   vcache_check_flush(vcache);
+}
+
+			  
+static INLINE void 
+vcache_triangle_flags( struct vcache_frontend *vcache,
+                       ushort flags,
+                       unsigned i0,
+                       unsigned i1,
+                       unsigned i2 )
+{
+   vcache_elt(vcache, i0, flags);
+   vcache_elt(vcache, i1, 0);
+   vcache_elt(vcache, i2, 0);
+   vcache_check_flush(vcache);
+}
+
+static INLINE void 
+vcache_line( struct vcache_frontend *vcache,
+             unsigned i0,
+             unsigned i1 )
+{
+   vcache_elt(vcache, i0, 0);
+   vcache_elt(vcache, i1, 0);
+   vcache_check_flush(vcache);
+}
+
+
+static INLINE void 
+vcache_line_flags( struct vcache_frontend *vcache,
+                   ushort flags,
+                   unsigned i0,
+                   unsigned i1 )
+{
+   vcache_elt(vcache, i0, flags);
+   vcache_elt(vcache, i1, 0);
+   vcache_check_flush(vcache);
+}
+
+
+static INLINE void 
+vcache_point( struct vcache_frontend *vcache,
+              unsigned i0 )
+{
+   vcache_elt(vcache, i0, 0);
+   vcache_check_flush(vcache);
+}
+
+static INLINE void 
+vcache_quad( struct vcache_frontend *vcache,
+             unsigned i0,
+             unsigned i1,
+             unsigned i2,
+             unsigned i3 )
+{
+   vcache_triangle( vcache, i0, i1, i3 );
+   vcache_triangle( vcache, i1, i2, i3 );
+}
+
+static INLINE void 
+vcache_ef_quad( struct vcache_frontend *vcache,
+                unsigned i0,
+                unsigned i1,
+                unsigned i2,
+                unsigned i3 )
+{
+   vcache_triangle_flags( vcache,
+                          ( DRAW_PIPE_RESET_STIPPLE |
+                            DRAW_PIPE_EDGE_FLAG_0 |
+                            DRAW_PIPE_EDGE_FLAG_2 ),
+                          i0, i1, i3 );
+
+   vcache_triangle_flags( vcache,
+                          ( DRAW_PIPE_EDGE_FLAG_0 |
+                            DRAW_PIPE_EDGE_FLAG_1 ),
+                          i1, i2, i3 );
+}
+
+/* At least for now, we're back to using a template include file for
+ * this.  The two paths aren't too different though - it may be
+ * possible to reunify them.
+ */
+#define TRIANGLE(vc,flags,i0,i1,i2) vcache_triangle_flags(vc,flags,i0,i1,i2)
+#define QUAD(vc,i0,i1,i2,i3)        vcache_ef_quad(vc,i0,i1,i2,i3)
+#define LINE(vc,flags,i0,i1)        vcache_line_flags(vc,flags,i0,i1)
+#define POINT(vc,i0)                vcache_point(vc,i0)
+#define FUNC vcache_run_extras
+#include "draw_pt_vcache_tmp.h"
+
+#define TRIANGLE(vc,flags,i0,i1,i2) vcache_triangle(vc,i0,i1,i2)
+#define QUAD(vc,i0,i1,i2,i3)        vcache_quad(vc,i0,i1,i2,i3)
+#define LINE(vc,flags,i0,i1)        vcache_line(vc,i0,i1)
+#define POINT(vc,i0)                vcache_point(vc,i0)
+#define FUNC vcache_run
+#include "draw_pt_vcache_tmp.h"
+
+static INLINE void 
+rebase_uint_elts( const unsigned *src,
+                  unsigned count,
+                  int delta,
+                  ushort *dest )
+{
+   unsigned i;
+
+   for (i = 0; i < count; i++) 
+      dest[i] = (ushort)(src[i] + delta);
+}
+
+static INLINE void 
+rebase_ushort_elts( const ushort *src,
+                    unsigned count,
+                    int delta,
+                                ushort *dest )
+{
+   unsigned i;
+
+   for (i = 0; i < count; i++) 
+      dest[i] = (ushort)(src[i] + delta);
+}
+
+static INLINE void 
+rebase_ubyte_elts( const ubyte *src,
+                   unsigned count,
+                   int delta,
+                   ushort *dest )
+{
+   unsigned i;
+
+   for (i = 0; i < count; i++) 
+      dest[i] = (ushort)(src[i] + delta);
+}
+
+
+
+static INLINE void 
+translate_uint_elts( const unsigned *src,
+                     unsigned count,
+                     ushort *dest )
+{
+   unsigned i;
+
+   for (i = 0; i < count; i++) 
+      dest[i] = (ushort)(src[i]);
+}
+
+static INLINE void 
+translate_ushort_elts( const ushort *src,
+                       unsigned count,
+                       ushort *dest )
+{
+   unsigned i;
+
+   for (i = 0; i < count; i++) 
+      dest[i] = (ushort)(src[i]);
+}
+
+static INLINE void 
+translate_ubyte_elts( const ubyte *src,
+                      unsigned count,
+                      ushort *dest )
+{
+   unsigned i;
+
+   for (i = 0; i < count; i++) 
+      dest[i] = (ushort)(src[i]);
+}
+
+
+
+
+#if 0
+static INLINE enum pipe_format 
+format_from_get_elt( pt_elt_func get_elt )
+{
+   switch (draw->pt.user.eltSize) {
+   case 1: return PIPE_FORMAT_R8_UNORM;
+   case 2: return PIPE_FORMAT_R16_UNORM;
+   case 4: return PIPE_FORMAT_R32_UNORM;
+   default: return PIPE_FORMAT_NONE;
+   }
+}
+#endif
+
+static INLINE void 
+vcache_check_run( struct draw_pt_front_end *frontend, 
+                  pt_elt_func get_elt,
+                  const void *elts,
+                  unsigned draw_count )
+{
+   struct vcache_frontend *vcache = (struct vcache_frontend *)frontend; 
+   struct draw_context *draw = vcache->draw;
+   unsigned min_index = draw->pt.user.min_index;
+   unsigned max_index = draw->pt.user.max_index;
+   unsigned index_size = draw->pt.user.eltSize;
+   unsigned fetch_count = max_index + 1 - min_index;
+   const ushort *transformed_elts;
+   ushort *storage = NULL;
+   boolean ok = FALSE;
+
+
+   if (0) debug_printf("fetch_count %d fetch_max %d draw_count %d\n", fetch_count, 
+                       vcache->fetch_max,
+                       draw_count);
+      
+   if (max_index == 0xffffffff ||
+       fetch_count > draw_count) {
+      if (0) debug_printf("fail\n");
+      goto fail;
+   }
+      
+   if (vcache->middle_prim != vcache->input_prim) {
+      vcache->middle_prim = vcache->input_prim;
+      vcache->middle->prepare( vcache->middle, 
+                               vcache->middle_prim, 
+                               vcache->opt, 
+                               &vcache->fetch_max );
+   }
+
+
+   if (min_index == 0 &&
+       index_size == 2) 
+   {
+      transformed_elts = (const ushort *)elts;
+   }
+   else 
+   {
+      storage = MALLOC( draw_count * sizeof(ushort) );
+      if (!storage)
+         goto fail;
+      
+      if (min_index == 0) {
+         switch(index_size) {
+         case 1:
+            translate_ubyte_elts( (const ubyte *)elts,
+                                  draw_count,
+                                  storage );
+            break;
+
+         case 2:
+            translate_ushort_elts( (const ushort *)elts,
+                                   draw_count,
+                                   storage );
+            break;
+
+         case 4:
+            translate_uint_elts( (const uint *)elts,
+                                 draw_count,
+                                 storage );
+            break;
+
+         default:
+            assert(0);
+            return;
+         }
+      }
+      else {
+         switch(index_size) {
+         case 1:
+            rebase_ubyte_elts( (const ubyte *)elts,
+                                  draw_count,
+                                  0 - (int)min_index,
+                                  storage );
+            break;
+
+         case 2:
+            rebase_ushort_elts( (const ushort *)elts,
+                                   draw_count,
+                                   0 - (int)min_index,
+                                   storage );
+            break;
+
+         case 4:
+            rebase_uint_elts( (const uint *)elts,
+                                 draw_count,
+                                 0 - (int)min_index,
+                                 storage );
+            break;
+
+         default:
+            assert(0);
+            return;
+         }
+      }
+      transformed_elts = storage;
+   }
+
+   if (fetch_count < UNDEFINED_VERTEX_ID)
+      ok = vcache->middle->run_linear_elts( vcache->middle,
+                                            min_index, /* start */
+                                            fetch_count,
+                                            transformed_elts,
+                                            draw_count );
+   
+   FREE(storage);
+
+   if (ok)
+      return;
+
+   debug_printf("failed to execute atomic draw elts for %d/%d, splitting up\n",
+                fetch_count, draw_count);
+
+ fail:
+   vcache_run( frontend, get_elt, elts, draw_count );
+}
+
+
+
+
+static void 
+vcache_prepare( struct draw_pt_front_end *frontend,
+                unsigned prim,
+                struct draw_pt_middle_end *middle,
+                unsigned opt )
+{
+   struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
+
+   if (opt & PT_PIPELINE)
+   {
+      vcache->base.run = vcache_run_extras;
+   }
+   else 
+   {
+      vcache->base.run = vcache_check_run;
+   }
+
+   vcache->input_prim = prim;
+   vcache->output_prim = draw_pt_reduced_prim(prim);
+
+   vcache->middle = middle;
+   vcache->opt = opt;
+
+   /* Have to run prepare here, but try and guess a good prim for
+    * doing so:
+    */
+   vcache->middle_prim = (opt & PT_PIPELINE) ? vcache->output_prim : vcache->input_prim;
+   middle->prepare( middle, vcache->middle_prim, opt, &vcache->fetch_max );
+}
+
+
+
+
+static void 
+vcache_finish( struct draw_pt_front_end *frontend )
+{
+   struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
+   vcache->middle->finish( vcache->middle );
+   vcache->middle = NULL;
+}
+
+static void 
+vcache_destroy( struct draw_pt_front_end *frontend )
+{
+   FREE(frontend);
+}
+
+
+struct draw_pt_front_end *draw_pt_vcache( struct draw_context *draw )
+{
+   struct vcache_frontend *vcache = CALLOC_STRUCT( vcache_frontend );
+   if (vcache == NULL)
+      return NULL;
+ 
+   vcache->base.prepare = vcache_prepare;
+   vcache->base.run     = NULL;
+   vcache->base.finish  = vcache_finish;
+   vcache->base.destroy = vcache_destroy;
+   vcache->draw = draw;
+   
+   memset(vcache->in, ~0, sizeof(vcache->in));
+  
+   return &vcache->base;
+}
diff --git a/src/gallium/auxiliary/draw/draw_pt_vcache_tmp.h b/src/gallium/auxiliary/draw/draw_pt_vcache_tmp.h
new file mode 100644
index 00000000000..ec05bbeab40
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_pt_vcache_tmp.h
@@ -0,0 +1,177 @@
+
+
+static void FUNC( struct draw_pt_front_end *frontend, 
+                  pt_elt_func get_elt,
+                  const void *elts,
+                  unsigned count )
+{
+   struct vcache_frontend *vcache = (struct vcache_frontend *)frontend;
+   struct draw_context *draw = vcache->draw;
+
+   boolean flatfirst = (draw->rasterizer->flatshade && 
+                        draw->rasterizer->flatshade_first);
+   unsigned i;
+   ushort flags;
+
+   if (0) debug_printf("%s %d\n", __FUNCTION__, count);
+
+
+   switch (vcache->input_prim) {
+   case PIPE_PRIM_POINTS:
+      for (i = 0; i < count; i ++) {
+	 POINT( vcache,
+                get_elt(elts, i + 0) );
+      }
+      break;
+
+   case PIPE_PRIM_LINES:
+      for (i = 0; i+1 < count; i += 2) {
+         LINE( vcache, 
+               DRAW_PIPE_RESET_STIPPLE,
+               get_elt(elts, i + 0),
+               get_elt(elts, i + 1));
+      }
+      break;
+
+   case PIPE_PRIM_LINE_LOOP:  
+      if (count >= 2) {
+         flags = DRAW_PIPE_RESET_STIPPLE;
+
+         for (i = 1; i < count; i++, flags = 0) {
+            LINE( vcache, 
+                  flags,
+                  get_elt(elts, i - 1),
+                  get_elt(elts, i ));
+         }
+
+	 LINE( vcache, 
+               flags,
+               get_elt(elts, i - 1),
+               get_elt(elts, 0 ));
+      }
+      break;
+
+   case PIPE_PRIM_LINE_STRIP:
+      flags = DRAW_PIPE_RESET_STIPPLE;
+      for (i = 1; i < count; i++, flags = 0) {
+         LINE( vcache, 
+               flags,
+               get_elt(elts, i - 1),
+               get_elt(elts, i ));
+      }
+      break;
+
+   case PIPE_PRIM_TRIANGLES:
+      for (i = 0; i+2 < count; i += 3) {
+         TRIANGLE( vcache,
+                   DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, 
+                   get_elt(elts, i + 0),
+                   get_elt(elts, i + 1),
+                   get_elt(elts, i + 2 ));
+      }
+      break;
+
+   case PIPE_PRIM_TRIANGLE_STRIP:
+      if (flatfirst) {
+         for (i = 0; i+2 < count; i++) {
+            TRIANGLE( vcache,
+                      DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, 
+                      get_elt(elts, i + 0),
+                      get_elt(elts, i + 1 + (i&1)),
+                      get_elt(elts, i + 2 - (i&1)));
+         }
+      }
+      else {
+         for (i = 0; i+2 < count; i++) {
+            TRIANGLE( vcache,
+                      DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, 
+                      get_elt(elts, i + 0 + (i&1)),
+                      get_elt(elts, i + 1 - (i&1)),
+                      get_elt(elts, i + 2 ));
+         }
+      }
+      break;
+
+   case PIPE_PRIM_TRIANGLE_FAN:
+      if (count >= 3) {
+         if (flatfirst) {
+            for (i = 0; i+2 < count; i++) {
+               TRIANGLE( vcache,
+                         DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, 
+                         get_elt(elts, i + 1),
+                         get_elt(elts, i + 2),
+                         get_elt(elts, 0 ));
+            }
+         }
+         else {
+            for (i = 0; i+2 < count; i++) {
+               TRIANGLE( vcache,
+                         DRAW_PIPE_RESET_STIPPLE | DRAW_PIPE_EDGE_FLAG_ALL, 
+                         get_elt(elts, 0),
+                         get_elt(elts, i + 1),
+                         get_elt(elts, i + 2 ));
+            }
+         }
+      }
+      break;
+
+
+   case PIPE_PRIM_QUADS:
+      for (i = 0; i+3 < count; i += 4) {
+         QUAD( vcache,
+               get_elt(elts, i + 0),
+               get_elt(elts, i + 1),
+               get_elt(elts, i + 2),
+               get_elt(elts, i + 3));
+      }
+      break;
+
+   case PIPE_PRIM_QUAD_STRIP:
+      for (i = 0; i+3 < count; i += 2) {
+         QUAD( vcache,
+               get_elt(elts, i + 2),
+               get_elt(elts, i + 0),
+               get_elt(elts, i + 1),
+               get_elt(elts, i + 3));
+      }
+      break;
+
+   case PIPE_PRIM_POLYGON:
+      {
+         /* These bitflags look a little odd because we submit the
+          * vertices as (1,2,0) to satisfy flatshade requirements.  
+          */
+         const ushort edge_first  = DRAW_PIPE_EDGE_FLAG_2;
+         const ushort edge_middle = DRAW_PIPE_EDGE_FLAG_0;
+         const ushort edge_last   = DRAW_PIPE_EDGE_FLAG_1;
+
+         flags = DRAW_PIPE_RESET_STIPPLE | edge_first | edge_middle;
+
+	 for (i = 0; i+2 < count; i++, flags = edge_middle) {
+
+            if (i + 3 == count)
+               flags |= edge_last;
+
+	    TRIANGLE( vcache,
+                      flags,
+                      get_elt(elts, i + 1),
+                      get_elt(elts, i + 2),
+                      get_elt(elts, 0));
+	 }
+      }
+      break;
+
+   default:
+      assert(0);
+      break;
+   }
+   
+   vcache_flush( vcache );
+}
+
+
+#undef TRIANGLE
+#undef QUAD
+#undef POINT
+#undef LINE
+#undef FUNC
diff --git a/src/gallium/auxiliary/draw/draw_vbuf.h b/src/gallium/auxiliary/draw/draw_vbuf.h
new file mode 100644
index 00000000000..a1c4c14445b
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vbuf.h
@@ -0,0 +1,127 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Vertex buffer drawing stage.
+ * 
+ * \author Keith Whitwell <[email protected]>
+ * \author Jose Fonseca <[email protected]>
+ */
+
+#ifndef DRAW_VBUF_H_
+#define DRAW_VBUF_H_
+
+
+#include "pipe/p_compiler.h"
+
+
+struct pipe_rasterizer_state;
+struct draw_context;
+struct vertex_info;
+
+
+/**
+ * Interface for hardware vertex buffer rendering.
+ */
+struct vbuf_render {
+
+   /**
+    * Driver limits.  May be tuned lower to improve cache hits on
+    * index list.
+    */
+   unsigned max_indices;
+   unsigned max_vertex_buffer_bytes;
+
+   /**
+    * Query if the hardware driver needs assistance for a particular
+    * combination of rasterizer state and primitive.
+    *
+    * Currently optional.
+    */
+   boolean (*need_pipeline)(const struct vbuf_render *render,
+                            const struct pipe_rasterizer_state *rasterizer,
+                            unsigned int prim );
+
+
+   /**
+    * Get the hardware vertex format.
+    * 
+    * XXX: have this in draw_context instead?
+    */ 
+   const struct vertex_info *(*get_vertex_info)( struct vbuf_render * );
+	 
+   /**
+    * Request a destination for vertices.
+    * Hardware renderers will use ttm memory, others will just malloc
+    * something.
+    */
+   void *(*allocate_vertices)( struct vbuf_render *,
+			       ushort vertex_size,
+			       ushort nr_vertices );
+
+   /**
+    * Notify the renderer of the current primitive when it changes.
+    * Must succeed for TRIANGLES, LINES and POINTS.  Other prims at
+    * the discretion of the driver, for the benefit of the passthrough
+    * path.
+    */
+   boolean (*set_primitive)( struct vbuf_render *, unsigned prim );
+
+   /**
+    * DrawElements, note indices are ushort.  The driver must complete
+    * this call, if necessary splitting the index list itself.
+    */
+   void (*draw)( struct vbuf_render *,
+		 const ushort *indices,
+		 uint nr_indices );
+
+   /* Draw Arrays path too.
+    */
+   void (*draw_arrays)( struct vbuf_render *,
+			unsigned start,
+			uint nr );
+
+   /**
+    * Called when vbuf is done with this set of vertices:
+    */
+   void (*release_vertices)( struct vbuf_render *,
+			     void *vertices, 
+			     unsigned vertex_size,
+			     unsigned vertices_used );
+
+   void (*destroy)( struct vbuf_render * );
+};
+
+
+
+struct draw_stage *
+draw_vbuf_stage( struct draw_context *draw,
+                 struct vbuf_render *render );
+
+
+#endif /*DRAW_VBUF_H_*/
diff --git a/src/gallium/auxiliary/draw/draw_vertex.c b/src/gallium/auxiliary/draw/draw_vertex.c
new file mode 100644
index 00000000000..3214213e445
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vertex.c
@@ -0,0 +1,129 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/*
+ * Functions for specifying the post-transformation vertex layout.
+ *
+ * Author:
+ *    Brian Paul
+ *    Keith Whitwell
+ */
+
+
+#include "draw/draw_private.h"
+#include "draw/draw_vertex.h"
+
+
+/**
+ * Compute the size of a vertex, in dwords/floats, to update the
+ * vinfo->size field.
+ */
+void
+draw_compute_vertex_size(struct vertex_info *vinfo)
+{
+   uint i;
+
+   vinfo->size = 0;
+   for (i = 0; i < vinfo->num_attribs; i++) {
+      switch (vinfo->attrib[i].emit) {
+      case EMIT_OMIT:
+         break;
+      case EMIT_4UB:
+         /* fall-through */
+      case EMIT_1F_PSIZE:
+         /* fall-through */
+      case EMIT_1F:
+         vinfo->size += 1;
+         break;
+      case EMIT_2F:
+         vinfo->size += 2;
+         break;
+      case EMIT_3F:
+         vinfo->size += 3;
+         break;
+      case EMIT_4F:
+         vinfo->size += 4;
+         break;
+      default:
+         assert(0);
+      }
+   }
+}
+
+
+void
+draw_dump_emitted_vertex(const struct vertex_info *vinfo, const uint8_t *data)
+{
+   unsigned i, j;
+
+   for (i = 0; i < vinfo->num_attribs; i++) {
+      j = vinfo->attrib[i].src_index;
+      switch (vinfo->attrib[i].emit) {
+      case EMIT_OMIT:
+         debug_printf("EMIT_OMIT:");
+         break;
+      case EMIT_1F:
+         debug_printf("EMIT_1F:\t");
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         break;
+      case EMIT_1F_PSIZE:
+         debug_printf("EMIT_1F_PSIZE:\t");
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         break;
+      case EMIT_2F:
+         debug_printf("EMIT_2F:\t");
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         break;
+      case EMIT_3F:
+         debug_printf("EMIT_3F:\t");
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         data += sizeof(float);
+         break;
+      case EMIT_4F:
+         debug_printf("EMIT_4F:\t");
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         debug_printf("%f ", *(float *)data); data += sizeof(float);
+         break;
+      case EMIT_4UB:
+         debug_printf("EMIT_4UB:\t");
+         debug_printf("%u ", *data++);
+         debug_printf("%u ", *data++);
+         debug_printf("%u ", *data++);
+         debug_printf("%u ", *data++);
+         break;
+      default:
+         assert(0);
+      }
+      debug_printf("\n");
+   }
+   debug_printf("\n");
+}
diff --git a/src/gallium/auxiliary/draw/draw_vertex.h b/src/gallium/auxiliary/draw/draw_vertex.h
new file mode 100644
index 00000000000..c143cf23723
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vertex.h
@@ -0,0 +1,163 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * Post-transform vertex format info.  The vertex_info struct is used by
+ * the draw_vbuf code to emit hardware-specific vertex layouts into hw
+ * vertex buffers.
+ *
+ * Author:
+ *    Brian Paul
+ */
+
+
+#ifndef DRAW_VERTEX_H
+#define DRAW_VERTEX_H
+
+
+#include "pipe/p_state.h"
+
+
+/**
+ * Vertex attribute emit modes
+ */
+enum attrib_emit {
+   EMIT_OMIT,      /**< don't emit the attribute */
+   EMIT_1F,
+   EMIT_1F_PSIZE,  /**< insert constant point size */
+   EMIT_2F,
+   EMIT_3F,
+   EMIT_4F,
+   EMIT_4UB  /**< XXX may need variations for RGBA vs BGRA, etc */
+};
+
+
+/**
+ * Attribute interpolation mode
+ */
+enum interp_mode {
+   INTERP_NONE,      /**< never interpolate vertex header info */
+   INTERP_POS,       /**< special case for frag position */
+   INTERP_CONSTANT,
+   INTERP_LINEAR,
+   INTERP_PERSPECTIVE
+};
+
+
+/**
+ * Information about hardware/rasterization vertex layout.
+ */
+struct vertex_info
+{
+   uint num_attribs;
+   uint hwfmt[4];      /**< hardware format info for this format */
+   uint size;          /**< total vertex size in dwords */
+   
+   /* Keep this small and at the end of the struct to allow quick
+    * memcmp() comparisons.
+    */
+   struct {
+      unsigned interp_mode:4;      /**< INTERP_x */
+      unsigned emit:4;             /**< EMIT_x */
+      unsigned src_index:8;          /**< map to post-xform attribs */
+   } attrib[PIPE_MAX_SHADER_INPUTS];
+};
+
+static INLINE int
+draw_vinfo_size( const struct vertex_info *a )
+{
+   return ((const char *)&a->attrib[a->num_attribs] -
+           (const char *)a);
+}
+
+static INLINE int
+draw_vinfo_compare( const struct vertex_info *a,
+                    const struct vertex_info *b )
+{
+   unsigned sizea = draw_vinfo_size( a );
+   return memcmp( a, b, sizea );
+}
+
+static INLINE void
+draw_vinfo_copy( struct vertex_info *dst,
+                 const struct vertex_info *src )
+{
+   unsigned size = draw_vinfo_size( src );
+   memcpy( dst, src, size );
+}
+
+
+
+/**
+ * Add another attribute to the given vertex_info object.
+ * \param src_index  indicates which post-transformed vertex attrib slot
+ *                   corresponds to this attribute.
+ * \return slot in which the attribute was added
+ */
+static INLINE uint
+draw_emit_vertex_attr(struct vertex_info *vinfo,
+                      enum attrib_emit emit, 
+                      enum interp_mode interp, /* only used by softpipe??? */
+                      uint src_index)
+{
+   const uint n = vinfo->num_attribs;
+   assert(n < PIPE_MAX_SHADER_INPUTS);
+   vinfo->attrib[n].emit = emit;
+   vinfo->attrib[n].interp_mode = interp;
+   vinfo->attrib[n].src_index = src_index;
+   vinfo->num_attribs++;
+   return n;
+}
+
+
+extern void draw_compute_vertex_size(struct vertex_info *vinfo);
+
+void draw_dump_emitted_vertex(const struct vertex_info *vinfo, 
+                              const uint8_t *data);
+
+
+static INLINE unsigned draw_translate_vinfo_format(unsigned format )
+{
+   switch (format) {
+   case EMIT_1F:
+   case EMIT_1F_PSIZE:
+      return PIPE_FORMAT_R32_FLOAT;
+   case EMIT_2F:
+      return PIPE_FORMAT_R32G32_FLOAT;
+   case EMIT_3F:
+      return PIPE_FORMAT_R32G32B32_FLOAT;
+   case EMIT_4F:
+      return PIPE_FORMAT_R32G32B32A32_FLOAT;
+   case EMIT_4UB:
+      return PIPE_FORMAT_R8G8B8A8_UNORM;
+   default:
+      return PIPE_FORMAT_NONE;
+   }
+}
+
+
+#endif /* DRAW_VERTEX_H */
diff --git a/src/gallium/auxiliary/draw/draw_vs.c b/src/gallium/auxiliary/draw/draw_vs.c
new file mode 100644
index 00000000000..c057cd67fde
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs.c
@@ -0,0 +1,267 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  *   Brian Paul
+  */
+
+#include "util/u_math.h"
+#include "util/u_memory.h"
+
+#include "pipe/p_shader_tokens.h"
+
+#include "draw_private.h"
+#include "draw_context.h"
+#include "draw_vs.h"
+
+#include "translate/translate.h"
+#include "translate/translate_cache.h"
+
+
+
+
+void draw_vs_set_constants( struct draw_context *draw,
+                            const float (*constants)[4],
+                            unsigned size )
+{
+   if (((uintptr_t)constants) & 0xf) {
+      if (size > draw->vs.const_storage_size) {
+         if (draw->vs.aligned_constant_storage)
+            align_free((void *)draw->vs.aligned_constant_storage);
+         draw->vs.aligned_constant_storage = align_malloc( size, 16 );
+      }
+      memcpy( (void*)draw->vs.aligned_constant_storage,
+              constants, 
+              size );
+      constants = draw->vs.aligned_constant_storage;
+   }
+      
+   draw->vs.aligned_constants = constants;
+   draw_vs_aos_machine_constants( draw->vs.aos_machine, constants );
+}
+
+
+void draw_vs_set_viewport( struct draw_context *draw,
+                           const struct pipe_viewport_state *viewport )
+{
+   draw_vs_aos_machine_viewport( draw->vs.aos_machine, viewport );
+}
+
+
+
+struct draw_vertex_shader *
+draw_create_vertex_shader(struct draw_context *draw,
+                          const struct pipe_shader_state *shader)
+{
+   struct draw_vertex_shader *vs;
+
+   vs = draw_create_vs_llvm( draw, shader );
+   if (!vs) {
+      vs = draw_create_vs_sse( draw, shader );
+      if (!vs) {
+         vs = draw_create_vs_ppc( draw, shader );
+         if (!vs) {
+            vs = draw_create_vs_exec( draw, shader );
+         }
+      }
+   }
+
+   if (vs)
+   {
+      uint i;
+      for (i = 0; i < vs->info.num_outputs; i++) {
+         if (vs->info.output_semantic_name[i] == TGSI_SEMANTIC_POSITION &&
+             vs->info.output_semantic_index[i] == 0)
+            vs->position_output = i;
+      }
+   }
+
+   assert(vs);
+   return vs;
+}
+
+
+void
+draw_bind_vertex_shader(struct draw_context *draw,
+                        struct draw_vertex_shader *dvs)
+{
+   draw_do_flush( draw, DRAW_FLUSH_STATE_CHANGE );
+   
+   if (dvs) 
+   {
+      draw->vs.vertex_shader = dvs;
+      draw->vs.num_vs_outputs = dvs->info.num_outputs;
+      draw->vs.position_output = dvs->position_output;
+      dvs->prepare( dvs, draw );
+   }
+   else {
+      draw->vs.vertex_shader = NULL;
+      draw->vs.num_vs_outputs = 0;
+   }
+}
+
+
+void
+draw_delete_vertex_shader(struct draw_context *draw,
+                          struct draw_vertex_shader *dvs)
+{
+   unsigned i;
+
+   for (i = 0; i < dvs->nr_varients; i++) 
+      dvs->varient[i]->destroy( dvs->varient[i] );
+
+   dvs->nr_varients = 0;
+
+   dvs->delete( dvs );
+}
+
+
+
+boolean 
+draw_vs_init( struct draw_context *draw )
+{
+   tgsi_exec_machine_init(&draw->vs.machine);
+
+   /* FIXME: give this machine thing a proper constructor:
+    */
+   draw->vs.machine.Inputs = align_malloc(PIPE_MAX_ATTRIBS * sizeof(struct tgsi_exec_vector), 16);
+   if (!draw->vs.machine.Inputs)
+      return FALSE;
+
+   draw->vs.machine.Outputs = align_malloc(PIPE_MAX_ATTRIBS * sizeof(struct tgsi_exec_vector), 16);
+   if (!draw->vs.machine.Outputs)
+      return FALSE;
+
+   draw->vs.emit_cache = translate_cache_create();
+   if (!draw->vs.emit_cache) 
+      return FALSE;
+      
+   draw->vs.fetch_cache = translate_cache_create();
+   if (!draw->vs.fetch_cache) 
+      return FALSE;
+
+   draw->vs.aos_machine = draw_vs_aos_machine();
+#ifdef PIPE_ARCH_X86
+   if (!draw->vs.aos_machine)
+      return FALSE;
+#endif
+      
+   return TRUE;
+}
+
+void
+draw_vs_destroy( struct draw_context *draw )
+{
+   if (draw->vs.machine.Inputs)
+      align_free(draw->vs.machine.Inputs);
+
+   if (draw->vs.machine.Outputs)
+      align_free(draw->vs.machine.Outputs);
+
+   if (draw->vs.fetch_cache)
+      translate_cache_destroy(draw->vs.fetch_cache);
+
+   if (draw->vs.emit_cache)
+      translate_cache_destroy(draw->vs.emit_cache);
+
+   if (draw->vs.aos_machine)
+      draw_vs_aos_machine_destroy(draw->vs.aos_machine);
+
+   if (draw->vs.aligned_constant_storage)
+      align_free((void*)draw->vs.aligned_constant_storage);
+
+   tgsi_exec_machine_free_data(&draw->vs.machine);
+
+}
+
+
+struct draw_vs_varient *
+draw_vs_lookup_varient( struct draw_vertex_shader *vs,
+                        const struct draw_vs_varient_key *key )
+{
+   struct draw_vs_varient *varient;
+   unsigned i;
+
+   /* Lookup existing varient: 
+    */
+   for (i = 0; i < vs->nr_varients; i++)
+      if (draw_vs_varient_key_compare(key, &vs->varient[i]->key) == 0)
+         return vs->varient[i];
+   
+   /* Else have to create a new one: 
+    */
+   varient = vs->create_varient( vs, key );
+   if (varient == NULL)
+      return NULL;
+
+   /* Add it to our list, could be smarter: 
+    */
+   if (vs->nr_varients < Elements(vs->varient)) {
+      vs->varient[vs->nr_varients++] = varient;
+   }
+   else {
+      vs->last_varient++;
+      vs->last_varient %= Elements(vs->varient);
+      vs->varient[vs->last_varient]->destroy(vs->varient[vs->last_varient]);
+      vs->varient[vs->last_varient] = varient;
+   }
+
+   /* Done 
+    */
+   return varient;
+}
+
+
+struct translate *
+draw_vs_get_fetch( struct draw_context *draw,
+                   struct translate_key *key )
+{
+   if (!draw->vs.fetch ||
+       translate_key_compare(&draw->vs.fetch->key, key) != 0) 
+   {
+      translate_key_sanitize(key);
+      draw->vs.fetch = translate_cache_find(draw->vs.fetch_cache, key);
+   }
+   
+   return draw->vs.fetch;
+}
+
+struct translate *
+draw_vs_get_emit( struct draw_context *draw,
+                  struct translate_key *key )
+{
+   if (!draw->vs.emit ||
+       translate_key_compare(&draw->vs.emit->key, key) != 0) 
+   {
+      translate_key_sanitize(key);
+      draw->vs.emit = translate_cache_find(draw->vs.emit_cache, key);
+   }
+   
+   return draw->vs.emit;
+}
diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h
new file mode 100644
index 00000000000..89ae158751a
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs.h
@@ -0,0 +1,224 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <[email protected]>
+ */
+
+#ifndef DRAW_VS_H
+#define DRAW_VS_H
+
+#include "draw_context.h"
+#include "draw_private.h"
+
+
+struct draw_context;
+struct pipe_shader_state;
+
+struct draw_varient_input 
+{
+   enum pipe_format format;
+   unsigned buffer;
+   unsigned offset; 
+};
+
+struct draw_varient_output
+{
+   enum pipe_format format;     /* output format */
+   unsigned vs_output:8;        /* which vertex shader output is this? */
+   unsigned offset:24;          /* offset into output vertex */
+};
+
+struct draw_varient_element {
+   struct draw_varient_input in;
+   struct draw_varient_output out;
+};
+
+struct draw_vs_varient_key {
+   unsigned output_stride;
+   unsigned nr_elements:8;      /* max2(nr_inputs, nr_outputs) */
+   unsigned nr_inputs:8;
+   unsigned nr_outputs:8;
+   unsigned viewport:1;
+   unsigned clip:1;
+   unsigned const_vbuffers:5;
+   struct draw_varient_element element[PIPE_MAX_ATTRIBS];
+};
+
+struct draw_vs_varient;
+
+
+struct draw_vs_varient {
+   struct draw_vs_varient_key key;
+
+   struct draw_vertex_shader *vs;
+
+   void (*set_buffer)( struct draw_vs_varient *,
+                      unsigned i,
+                      const void *ptr,
+                      unsigned stride );
+
+   void (PIPE_CDECL *run_linear)( struct draw_vs_varient *shader,
+                                  unsigned start,
+                                  unsigned count,
+                                  void *output_buffer );
+
+   void (PIPE_CDECL *run_elts)( struct draw_vs_varient *shader,
+                                const unsigned *elts,
+                                unsigned count,
+                                void *output_buffer );
+
+   void (*destroy)( struct draw_vs_varient * );
+};
+
+
+/**
+ * Private version of the compiled vertex_shader
+ */
+struct draw_vertex_shader {
+   struct draw_context *draw;
+
+   /* This member will disappear shortly:
+    */
+   struct pipe_shader_state   state;
+
+   struct tgsi_shader_info info;
+   unsigned position_output;
+
+   /* Extracted from shader:
+    */
+   const float (*immediates)[4];
+
+   /* 
+    */
+   struct draw_vs_varient *varient[16];
+   unsigned nr_varients;
+   unsigned last_varient;
+   struct draw_vs_varient *(*create_varient)( struct draw_vertex_shader *shader,
+                                              const struct draw_vs_varient_key *key );
+
+
+   void (*prepare)( struct draw_vertex_shader *shader,
+		    struct draw_context *draw );
+
+   /* Run the shader - this interface will get cleaned up in the
+    * future:
+    */
+   void (*run_linear)( struct draw_vertex_shader *shader,
+		       const float (*input)[4],
+		       float (*output)[4],
+		       const float (*constants)[4],
+		       unsigned count,
+		       unsigned input_stride,
+		       unsigned output_stride );
+
+
+   void (*delete)( struct draw_vertex_shader * );
+};
+
+
+struct draw_vs_varient *
+draw_vs_lookup_varient( struct draw_vertex_shader *base,
+                        const struct draw_vs_varient_key *key );
+
+
+/********************************************************************************
+ * Internal functions:
+ */
+
+struct draw_vertex_shader *
+draw_create_vs_exec(struct draw_context *draw,
+		    const struct pipe_shader_state *templ);
+
+struct draw_vertex_shader *
+draw_create_vs_sse(struct draw_context *draw,
+		   const struct pipe_shader_state *templ);
+
+struct draw_vertex_shader *
+draw_create_vs_ppc(struct draw_context *draw,
+		   const struct pipe_shader_state *templ);
+
+struct draw_vertex_shader *
+draw_create_vs_llvm(struct draw_context *draw,
+		    const struct pipe_shader_state *templ);
+
+
+
+struct draw_vs_varient_key;
+struct draw_vertex_shader;
+
+struct draw_vs_varient *draw_vs_varient_aos_sse( struct draw_vertex_shader *vs,
+                                                 const struct draw_vs_varient_key *key );
+
+
+
+/********************************************************************************
+ * Helpers for vs implementations that don't do their own fetch/emit varients.
+ * Means these can be shared between shaders.
+ */
+struct translate;
+struct translate_key;
+
+struct translate *draw_vs_get_fetch( struct draw_context *draw,
+                                     struct translate_key *key );
+
+
+struct translate *draw_vs_get_emit( struct draw_context *draw,
+                                    struct translate_key *key );
+
+struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
+                                                 const struct draw_vs_varient_key *key );
+
+
+
+static INLINE int draw_vs_varient_keysize( const struct draw_vs_varient_key *key )
+{
+   return 2 * sizeof(int) + key->nr_elements * sizeof(struct draw_varient_element);
+}
+
+static INLINE int draw_vs_varient_key_compare( const struct draw_vs_varient_key *a,
+                                         const struct draw_vs_varient_key *b )
+{
+   int keysize = draw_vs_varient_keysize(a);
+   return memcmp(a, b, keysize);
+}
+
+
+struct aos_machine *draw_vs_aos_machine( void );
+void draw_vs_aos_machine_destroy( struct aos_machine *machine );
+
+void draw_vs_aos_machine_constants( struct aos_machine *machine,
+                                    const float (*constants)[4] );
+
+void draw_vs_aos_machine_viewport( struct aos_machine *machine,
+                                   const struct pipe_viewport_state *viewport );
+
+
+#define MAX_TGSI_VERTICES 4
+   
+
+
+#endif
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
new file mode 100644
index 00000000000..0c693a4a65c
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -0,0 +1,2252 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  6.3
+ *
+ * Copyright (C) 1999-2004  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/**
+ * Translate tgsi vertex programs to x86/x87/SSE/SSE2 machine code
+ * using the rtasm runtime assembler.  Based on the old
+ * t_vb_arb_program_sse.c
+ */
+
+
+#include "util/u_memory.h"
+#include "util/u_math.h"
+#include "pipe/p_shader_tokens.h"
+#include "pipe/p_debug.h"
+#include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_util.h"
+#include "tgsi/tgsi_exec.h"
+#include "tgsi/tgsi_dump.h"
+
+#include "draw_vs.h"
+#include "draw_vs_aos.h"
+
+#include "rtasm/rtasm_x86sse.h"
+
+#ifdef PIPE_ARCH_X86
+#define DISASSEM 0
+#define FAST_MATH 1
+
+static const char *files[] =
+{
+   "NULL",
+   "CONST",
+   "IN",
+   "OUT",
+   "TEMP",
+   "SAMP",
+   "ADDR",
+   "IMM",
+   "INTERNAL",
+};
+
+static INLINE boolean eq( struct x86_reg a,
+			    struct x86_reg b )
+{
+   return (a.file == b.file &&
+	   a.idx == b.idx &&
+	   a.mod == b.mod &&
+	   a.disp == b.disp);
+}
+      
+struct x86_reg aos_get_x86( struct aos_compilation *cp,
+                            unsigned which_reg, /* quick hack */
+                            unsigned value )
+{
+   struct x86_reg reg;
+
+   if (which_reg == 0)
+      reg = cp->temp_EBP;
+   else
+      reg = cp->tmp_EAX;
+
+   if (cp->x86_reg[which_reg] != value) {
+      unsigned offset;
+
+      switch (value) {
+      case X86_IMMEDIATES:
+         assert(which_reg == 0);
+         offset = Offset(struct aos_machine, immediates);
+         break;
+      case X86_CONSTANTS:
+         assert(which_reg == 1);
+         offset = Offset(struct aos_machine, constants);
+         break;
+      case X86_BUFFERS:
+         assert(which_reg == 0);
+         offset = Offset(struct aos_machine, buffer);
+         break;
+      default:
+         assert(0);
+         offset = 0;
+      }
+
+
+      x86_mov(cp->func, reg, 
+              x86_make_disp(cp->machine_EDX, offset));
+
+      cp->x86_reg[which_reg] = value;
+   }
+
+   return reg;
+}
+
+
+static struct x86_reg get_reg_ptr(struct aos_compilation *cp,
+                                  unsigned file,
+				  unsigned idx )
+{
+   struct x86_reg ptr = cp->machine_EDX;
+
+   switch (file) {
+   case TGSI_FILE_INPUT:
+      assert(idx < MAX_INPUTS);
+      return x86_make_disp(ptr, Offset(struct aos_machine, input[idx]));
+
+   case TGSI_FILE_OUTPUT:
+      return x86_make_disp(ptr, Offset(struct aos_machine, output[idx]));
+
+   case TGSI_FILE_TEMPORARY:
+      assert(idx < MAX_TEMPS);
+      return x86_make_disp(ptr, Offset(struct aos_machine, temp[idx]));
+
+   case AOS_FILE_INTERNAL:
+      assert(idx < MAX_INTERNALS);
+      return x86_make_disp(ptr, Offset(struct aos_machine, internal[idx]));
+
+   case TGSI_FILE_IMMEDIATE: 
+      assert(idx < MAX_IMMEDIATES);  /* just a sanity check */
+      return x86_make_disp(aos_get_x86(cp, 0, X86_IMMEDIATES), idx * 4 * sizeof(float));
+
+   case TGSI_FILE_CONSTANT: 
+      assert(idx < MAX_CONSTANTS);  /* just a sanity check */
+      return x86_make_disp(aos_get_x86(cp, 1, X86_CONSTANTS), idx * 4 * sizeof(float));
+
+   default:
+      ERROR(cp, "unknown reg file");
+      return x86_make_reg(0,0);
+   }
+}
+		
+
+
+#define X87_CW_EXCEPTION_INV_OP       (1<<0)
+#define X87_CW_EXCEPTION_DENORM_OP    (1<<1)
+#define X87_CW_EXCEPTION_ZERO_DIVIDE  (1<<2)
+#define X87_CW_EXCEPTION_OVERFLOW     (1<<3)
+#define X87_CW_EXCEPTION_UNDERFLOW    (1<<4)
+#define X87_CW_EXCEPTION_PRECISION    (1<<5)
+#define X87_CW_PRECISION_SINGLE       (0<<8)
+#define X87_CW_PRECISION_RESERVED     (1<<8)
+#define X87_CW_PRECISION_DOUBLE       (2<<8)
+#define X87_CW_PRECISION_DOUBLE_EXT   (3<<8)
+#define X87_CW_PRECISION_MASK         (3<<8)
+#define X87_CW_ROUND_NEAREST          (0<<10)
+#define X87_CW_ROUND_DOWN             (1<<10)
+#define X87_CW_ROUND_UP               (2<<10)
+#define X87_CW_ROUND_ZERO             (3<<10)
+#define X87_CW_ROUND_MASK             (3<<10)
+#define X87_CW_INFINITY               (1<<12)
+
+
+
+
+static void spill( struct aos_compilation *cp, unsigned idx )
+{
+   if (!cp->xmm[idx].dirty ||
+       (cp->xmm[idx].file != TGSI_FILE_INPUT && /* inputs are fetched into xmm & set dirty */
+        cp->xmm[idx].file != TGSI_FILE_OUTPUT &&
+        cp->xmm[idx].file != TGSI_FILE_TEMPORARY)) {
+      ERROR(cp, "invalid spill");
+      return;
+   }
+   else {
+      struct x86_reg oldval = get_reg_ptr(cp,
+                                          cp->xmm[idx].file,
+                                          cp->xmm[idx].idx);
+     
+      if (0) debug_printf("\nspill %s[%d]", 
+                          files[cp->xmm[idx].file],
+                          cp->xmm[idx].idx);
+ 
+      assert(cp->xmm[idx].dirty);
+      sse_movaps(cp->func, oldval, x86_make_reg(file_XMM, idx));
+      cp->xmm[idx].dirty = 0;
+   }
+}
+
+
+void aos_spill_all( struct aos_compilation *cp )
+{
+   unsigned i;
+
+   for (i = 0; i < 8; i++) {
+      if (cp->xmm[i].dirty) 
+         spill(cp, i);
+      aos_release_xmm_reg(cp, i);
+   }
+}
+
+
+static struct x86_reg get_xmm_writable( struct aos_compilation *cp,
+                                        struct x86_reg reg )
+{
+   if (reg.file != file_XMM ||
+       cp->xmm[reg.idx].file != TGSI_FILE_NULL)
+   {
+      struct x86_reg tmp = aos_get_xmm_reg(cp);
+      sse_movaps(cp->func, tmp, reg);
+      reg = tmp;
+   }
+
+   cp->xmm[reg.idx].last_used = cp->insn_counter;
+   return reg;
+}
+
+static struct x86_reg get_xmm( struct aos_compilation *cp,
+                               struct x86_reg reg )
+{
+   if (reg.file != file_XMM) 
+   {
+      struct x86_reg tmp = aos_get_xmm_reg(cp);
+      sse_movaps(cp->func, tmp, reg);
+      reg = tmp;
+   }
+
+   cp->xmm[reg.idx].last_used = cp->insn_counter;
+   return reg;
+}
+
+
+/* Allocate an empty xmm register, either as a temporary or later to
+ * "adopt" as a shader reg.
+ */
+struct x86_reg aos_get_xmm_reg( struct aos_compilation *cp )
+{
+   unsigned i;
+   unsigned oldest = 0;
+   boolean found = FALSE;
+
+   for (i = 0; i < 8; i++) 
+      if (cp->xmm[i].last_used != cp->insn_counter &&
+          cp->xmm[i].file == TGSI_FILE_NULL) {
+	 oldest = i;
+         found = TRUE;
+      }
+
+   if (!found) {
+      for (i = 0; i < 8; i++) 
+         if (cp->xmm[i].last_used < cp->xmm[oldest].last_used)
+            oldest = i;
+   }
+
+   /* Need to write out the old value?
+    */
+   if (cp->xmm[oldest].dirty) 
+      spill(cp, oldest);
+
+   assert(cp->xmm[oldest].last_used != cp->insn_counter);
+
+   cp->xmm[oldest].file = TGSI_FILE_NULL;
+   cp->xmm[oldest].idx = 0;
+   cp->xmm[oldest].dirty = 0;
+   cp->xmm[oldest].last_used = cp->insn_counter;
+   return x86_make_reg(file_XMM, oldest);
+}
+
+void aos_release_xmm_reg( struct aos_compilation *cp,
+                          unsigned idx )
+{
+   cp->xmm[idx].file = TGSI_FILE_NULL;
+   cp->xmm[idx].idx = 0;
+   cp->xmm[idx].dirty = 0;
+   cp->xmm[idx].last_used = 0;
+}
+
+
+
+     
+/* Mark an xmm reg as holding the current copy of a shader reg.
+ */
+void aos_adopt_xmm_reg( struct aos_compilation *cp,
+                        struct x86_reg reg,
+                        unsigned file,
+                        unsigned idx,
+                        unsigned dirty )
+{
+   unsigned i;
+
+   if (reg.file != file_XMM) {
+      assert(0);
+      return;
+   }
+
+
+   /* If any xmm reg thinks it holds this shader reg, break the
+    * illusion.
+    */
+   for (i = 0; i < 8; i++) {
+      if (cp->xmm[i].file == file && 
+          cp->xmm[i].idx == idx) 
+      {
+         /* If an xmm reg is already holding this shader reg, take into account its
+          * dirty flag...
+          */
+         dirty |= cp->xmm[i].dirty;
+         aos_release_xmm_reg(cp, i);
+      }
+   }
+
+   cp->xmm[reg.idx].file = file;
+   cp->xmm[reg.idx].idx = idx;
+   cp->xmm[reg.idx].dirty = dirty;
+   cp->xmm[reg.idx].last_used = cp->insn_counter;
+}
+
+
+/* Return a pointer to the in-memory copy of the reg, making sure it is uptodate.
+ */
+static struct x86_reg aos_get_shader_reg_ptr( struct aos_compilation *cp, 
+                                              unsigned file,
+                                              unsigned idx )
+{
+   unsigned i;
+
+   /* Ensure the in-memory copy of this reg is up-to-date
+    */
+   for (i = 0; i < 8; i++) {
+      if (cp->xmm[i].file == file && 
+          cp->xmm[i].idx == idx &&
+          cp->xmm[i].dirty) {
+         spill(cp, i);
+      }
+   }
+
+   return get_reg_ptr( cp, file, idx );
+}
+
+
+/* As above, but return a pointer.  Note - this pointer may alias
+ * those returned by get_arg_ptr().
+ */
+static struct x86_reg get_dst_ptr( struct aos_compilation *cp, 
+                                   const struct tgsi_full_dst_register *dst )
+{
+   unsigned file = dst->DstRegister.File;
+   unsigned idx = dst->DstRegister.Index;
+   unsigned i;
+   
+
+   /* Ensure in-memory copy of this reg is up-to-date and invalidate
+    * any xmm copies.
+    */
+   for (i = 0; i < 8; i++) {
+      if (cp->xmm[i].file == file &&
+          cp->xmm[i].idx == idx)
+      {
+         if (cp->xmm[i].dirty) 
+            spill(cp, i);
+         
+         aos_release_xmm_reg(cp, i);
+      }
+   }
+
+   return get_reg_ptr( cp, file, idx );
+}
+
+
+
+
+
+/* Return an XMM reg if the argument is resident, otherwise return a
+ * base+offset pointer to the saved value.
+ */
+struct x86_reg aos_get_shader_reg( struct aos_compilation *cp, 
+                                   unsigned file,
+                                   unsigned idx )
+{
+   unsigned i;
+
+   for (i = 0; i < 8; i++) {
+      if (cp->xmm[i].file == file &&
+	  cp->xmm[i].idx  == idx) 
+      {
+	 cp->xmm[i].last_used = cp->insn_counter;
+	 return x86_make_reg(file_XMM, i);
+      }
+   }
+
+   /* If not found in the XMM register file, return an indirect
+    * reference to the in-memory copy:
+    */
+   return get_reg_ptr( cp, file, idx );
+}
+
+
+
+static struct x86_reg aos_get_shader_reg_xmm( struct aos_compilation *cp, 
+                                              unsigned file,
+                                              unsigned idx )
+{
+   struct x86_reg reg = get_xmm( cp,
+                                 aos_get_shader_reg( cp, file, idx ) );
+
+   aos_adopt_xmm_reg( cp,
+                      reg,
+                      file,
+                      idx,
+                      FALSE );
+   
+   return reg;
+}
+
+
+
+struct x86_reg aos_get_internal_xmm( struct aos_compilation *cp,
+                                     unsigned imm )
+{
+   return aos_get_shader_reg_xmm( cp, AOS_FILE_INTERNAL, imm );
+}
+
+
+struct x86_reg aos_get_internal( struct aos_compilation *cp,
+                                 unsigned imm )
+{
+   return aos_get_shader_reg( cp, AOS_FILE_INTERNAL, imm );
+}
+
+
+
+
+
+/* Emulate pshufd insn in regular SSE, if necessary:
+ */
+static void emit_pshufd( struct aos_compilation *cp,
+			 struct x86_reg dst,
+			 struct x86_reg arg0,
+			 ubyte shuf )
+{
+   if (cp->have_sse2) {
+      sse2_pshufd(cp->func, dst, arg0, shuf);
+   }
+   else {
+      if (!eq(dst, arg0)) 
+	 sse_movaps(cp->func, dst, arg0);
+
+      sse_shufps(cp->func, dst, dst, shuf);
+   }
+}
+
+/* load masks (pack into negs??)
+ * pshufd - shuffle according to writemask
+ * and - result, mask
+ * nand - dest, mask
+ * or - dest, result
+ */
+static boolean mask_write( struct aos_compilation *cp,
+                           struct x86_reg dst,
+                           struct x86_reg result,
+                           unsigned mask )
+{
+   struct x86_reg imm_swz = aos_get_internal_xmm(cp, IMM_SWZ);
+   struct x86_reg tmp = aos_get_xmm_reg(cp);
+   
+   emit_pshufd(cp, tmp, imm_swz, 
+               SHUF((mask & 1) ? 2 : 3,
+                    (mask & 2) ? 2 : 3,
+                    (mask & 4) ? 2 : 3,
+                    (mask & 8) ? 2 : 3));
+
+   sse_andps(cp->func, dst, tmp);
+   sse_andnps(cp->func, tmp, result);
+   sse_orps(cp->func, dst, tmp);
+
+   aos_release_xmm_reg(cp, tmp.idx);
+   return TRUE;
+}
+
+
+
+
+/* Helper for writemask:
+ */
+static boolean emit_shuf_copy2( struct aos_compilation *cp,
+				  struct x86_reg dst,
+				  struct x86_reg arg0,
+				  struct x86_reg arg1,
+				  ubyte shuf )
+{
+   struct x86_reg tmp = aos_get_xmm_reg(cp);
+
+   emit_pshufd(cp, dst, arg1, shuf);
+   emit_pshufd(cp, tmp, arg0, shuf);
+   sse_shufps(cp->func, dst, tmp, SHUF(X, Y, Z, W));
+   emit_pshufd(cp, dst, dst, shuf);
+
+   aos_release_xmm_reg(cp, tmp.idx);
+   return TRUE;
+}
+
+
+
+#define SSE_SWIZZLE_NOOP ((0<<0) | (1<<2) | (2<<4) | (3<<6))
+
+
+/* Locate a source register and perform any required (simple) swizzle.  
+ * 
+ * Just fail on complex swizzles at this point.
+ */
+static struct x86_reg fetch_src( struct aos_compilation *cp, 
+                                 const struct tgsi_full_src_register *src ) 
+{
+   struct x86_reg arg0 = aos_get_shader_reg(cp, 
+                                            src->SrcRegister.File, 
+                                            src->SrcRegister.Index);
+   unsigned i;
+   ubyte swz = 0;
+   unsigned negs = 0;
+   unsigned abs = 0;
+
+   for (i = 0; i < 4; i++) {
+      unsigned swizzle = tgsi_util_get_full_src_register_extswizzle( src, i );
+      unsigned neg = tgsi_util_get_full_src_register_sign_mode( src, i );
+
+      switch (swizzle) {
+      case TGSI_EXTSWIZZLE_ZERO:
+      case TGSI_EXTSWIZZLE_ONE:
+         ERROR(cp, "not supporting full swizzles yet in tgsi_aos_sse2");
+         break;
+
+      default:
+         swz |= (swizzle & 0x3) << (i * 2);
+         break;
+      }
+
+      switch (neg) {
+      case TGSI_UTIL_SIGN_TOGGLE:
+         negs |= (1<<i);
+         break;
+         
+      case TGSI_UTIL_SIGN_KEEP:
+         break;
+
+      case TGSI_UTIL_SIGN_CLEAR:
+         abs |= (1<<i);
+         break;
+
+      default:
+         ERROR(cp, "unsupported sign-mode");
+         break;
+      }
+   }
+
+   if (swz != SSE_SWIZZLE_NOOP || negs != 0 || abs != 0) {
+      struct x86_reg dst = aos_get_xmm_reg(cp);
+
+      if (swz != SSE_SWIZZLE_NOOP)
+         emit_pshufd(cp, dst, arg0, swz);
+      else
+         sse_movaps(cp->func, dst, arg0);
+
+      if (negs && negs != 0xf) {
+         struct x86_reg imm_swz = aos_get_internal_xmm(cp, IMM_SWZ);
+         struct x86_reg tmp = aos_get_xmm_reg(cp);
+
+         /* Load 1,-1,0,0
+          * Use neg as arg to pshufd
+          * Multiply
+          */
+         emit_pshufd(cp, tmp, imm_swz, 
+                     SHUF((negs & 1) ? 1 : 0,
+                          (negs & 2) ? 1 : 0,
+                          (negs & 4) ? 1 : 0,
+                          (negs & 8) ? 1 : 0));
+         sse_mulps(cp->func, dst, tmp);
+
+         aos_release_xmm_reg(cp, tmp.idx);
+      }
+      else if (negs) {
+         struct x86_reg imm_negs = aos_get_internal_xmm(cp, IMM_NEGS);
+         sse_mulps(cp->func, dst, imm_negs);
+      }
+
+
+      if (abs && abs != 0xf) {
+         ERROR(cp, "unsupported partial abs");
+      }
+      else if (abs) {
+         struct x86_reg neg = aos_get_internal(cp, IMM_NEGS);
+         struct x86_reg tmp = aos_get_xmm_reg(cp);
+
+         sse_movaps(cp->func, tmp, dst);
+         sse_mulps(cp->func, tmp, neg);
+         sse_maxps(cp->func, dst, tmp);
+
+         aos_release_xmm_reg(cp, tmp.idx);
+      }
+
+      return dst;
+   }
+      
+   return arg0;
+}
+
+static void x87_fld_src( struct aos_compilation *cp, 
+                         const struct tgsi_full_src_register *src,
+                         unsigned channel ) 
+{
+   struct x86_reg arg0 = aos_get_shader_reg_ptr(cp, 
+                                                src->SrcRegister.File, 
+                                                src->SrcRegister.Index);
+
+   unsigned swizzle = tgsi_util_get_full_src_register_extswizzle( src, channel );
+   unsigned neg = tgsi_util_get_full_src_register_sign_mode( src, channel );
+
+   switch (swizzle) {
+   case TGSI_EXTSWIZZLE_ZERO:
+      x87_fldz( cp->func );
+      break;
+
+   case TGSI_EXTSWIZZLE_ONE:
+      x87_fld1( cp->func );
+      break;
+
+   default:
+      x87_fld( cp->func, x86_make_disp(arg0, (swizzle & 3) * sizeof(float)) );
+      break;
+   }
+   
+
+   switch (neg) {
+   case TGSI_UTIL_SIGN_TOGGLE:
+      /* Flip the sign:
+       */
+      x87_fchs( cp->func );
+      break;
+         
+   case TGSI_UTIL_SIGN_KEEP:
+      break;
+
+   case TGSI_UTIL_SIGN_CLEAR:
+      x87_fabs( cp->func );
+      break;
+
+   case TGSI_UTIL_SIGN_SET:
+      x87_fabs( cp->func );
+      x87_fchs( cp->func );
+      break;
+
+   default:
+      ERROR(cp, "unsupported sign-mode");
+      break;
+   }
+}
+
+
+
+
+
+
+/* Used to implement write masking.  This and most of the other instructions
+ * here would be easier to implement if there had been a translation
+ * to a 2 argument format (dst/arg0, arg1) at the shader level before
+ * attempting to translate to x86/sse code.
+ */
+static void store_dest( struct aos_compilation *cp, 
+                        const struct tgsi_full_dst_register *reg,
+                        struct x86_reg result )
+{
+   struct x86_reg dst;
+
+   switch (reg->DstRegister.WriteMask) {
+   case 0:
+      return;
+   
+   case TGSI_WRITEMASK_XYZW:
+      aos_adopt_xmm_reg(cp, 
+                        get_xmm_writable(cp, result), 
+                        reg->DstRegister.File,
+                        reg->DstRegister.Index,
+                        TRUE);
+      return;
+   default: 
+      break;
+   }
+
+   dst = aos_get_shader_reg_xmm(cp, 
+                                reg->DstRegister.File,
+                                reg->DstRegister.Index);
+
+   switch (reg->DstRegister.WriteMask) {
+   case TGSI_WRITEMASK_X:
+      sse_movss(cp->func, dst, get_xmm(cp, result));
+      break;
+      
+   case TGSI_WRITEMASK_ZW:
+      sse_shufps(cp->func, dst, get_xmm(cp, result), SHUF(X, Y, Z, W));
+      break;
+
+   case TGSI_WRITEMASK_XY: 
+      result = get_xmm_writable(cp, result);
+      sse_shufps(cp->func, result, dst, SHUF(X, Y, Z, W));
+      dst = result;
+      break;
+
+   case TGSI_WRITEMASK_YZW: 
+      result = get_xmm_writable(cp, result);
+      sse_movss(cp->func, result, dst);
+      dst = result;
+      break;
+
+   default:
+      mask_write(cp, dst, result, reg->DstRegister.WriteMask);
+      break;
+   }
+
+   aos_adopt_xmm_reg(cp, 
+                     dst, 
+                     reg->DstRegister.File,
+                     reg->DstRegister.Index,
+                     TRUE);
+
+}
+
+static void inject_scalar( struct aos_compilation *cp,
+                           struct x86_reg dst,
+                           struct x86_reg result,
+                           ubyte swizzle )
+{
+   sse_shufps(cp->func, dst, dst, swizzle);
+   sse_movss(cp->func, dst, result);
+   sse_shufps(cp->func, dst, dst, swizzle);
+}
+
+
+static void store_scalar_dest( struct aos_compilation *cp, 
+                               const struct tgsi_full_dst_register *reg,
+                               struct x86_reg result )
+{
+   unsigned writemask = reg->DstRegister.WriteMask;
+   struct x86_reg dst;
+
+   if (writemask != TGSI_WRITEMASK_X &&
+       writemask != TGSI_WRITEMASK_Y &&
+       writemask != TGSI_WRITEMASK_Z &&
+       writemask != TGSI_WRITEMASK_W &&
+       writemask != 0) 
+   {
+      result = get_xmm_writable(cp, result); /* already true, right? */
+      sse_shufps(cp->func, result, result, SHUF(X,X,X,X));
+      store_dest(cp, reg, result);
+      return;
+   }
+
+   result = get_xmm(cp, result);
+   dst = aos_get_shader_reg_xmm(cp, 
+                                reg->DstRegister.File,
+                                reg->DstRegister.Index);
+
+
+
+   switch (reg->DstRegister.WriteMask) {
+   case TGSI_WRITEMASK_X:
+      sse_movss(cp->func, dst, result);
+      break;
+
+   case TGSI_WRITEMASK_Y:
+      inject_scalar(cp, dst, result, SHUF(Y, X, Z, W));
+      break;
+
+   case TGSI_WRITEMASK_Z:
+      inject_scalar(cp, dst, result, SHUF(Z, Y, X, W));
+      break;
+
+   case TGSI_WRITEMASK_W:
+      inject_scalar(cp, dst, result, SHUF(W, Y, Z, X));
+      break;
+
+   default:
+      break;
+   }
+
+   aos_adopt_xmm_reg(cp, 
+                     dst, 
+                     reg->DstRegister.File,
+                     reg->DstRegister.Index,
+                     TRUE);
+}
+   
+
+
+static void x87_fst_or_nop( struct x86_function *func,
+                            unsigned writemask,
+                            unsigned channel,
+                            struct x86_reg ptr )
+{
+   assert(ptr.file == file_REG32);
+   if (writemask & (1<<channel)) 
+      x87_fst( func, x86_make_disp(ptr, channel * sizeof(float)) );
+}
+
+static void x87_fstp_or_pop( struct x86_function *func,
+                             unsigned writemask,
+                             unsigned channel,
+                             struct x86_reg ptr )
+{
+   assert(ptr.file == file_REG32);
+   if (writemask & (1<<channel)) 
+      x87_fstp( func, x86_make_disp(ptr, channel * sizeof(float)) );
+   else
+      x87_fstp( func, x86_make_reg( file_x87, 0 ));
+}
+
+
+
+/* 
+ */
+static void x87_fstp_dest4( struct aos_compilation *cp,
+                            const struct tgsi_full_dst_register *dst )
+{
+   struct x86_reg ptr = get_dst_ptr(cp, dst); 
+   unsigned writemask = dst->DstRegister.WriteMask;
+
+   x87_fst_or_nop(cp->func, writemask, 0, ptr);
+   x87_fst_or_nop(cp->func, writemask, 1, ptr);
+   x87_fst_or_nop(cp->func, writemask, 2, ptr);
+   x87_fstp_or_pop(cp->func, writemask, 3, ptr);
+}
+
+/* Save current x87 state and put it into single precision mode.
+ */
+static void save_fpu_state( struct aos_compilation *cp )
+{
+   x87_fnstcw( cp->func, x86_make_disp(cp->machine_EDX, 
+                                       Offset(struct aos_machine, fpu_restore)));
+}
+
+static void restore_fpu_state( struct aos_compilation *cp )
+{
+   x87_fnclex(cp->func);
+   x87_fldcw( cp->func, x86_make_disp(cp->machine_EDX, 
+                                      Offset(struct aos_machine, fpu_restore)));
+}
+
+static void set_fpu_round_neg_inf( struct aos_compilation *cp )
+{
+   if (cp->fpucntl != FPU_RND_NEG) {
+      cp->fpucntl = FPU_RND_NEG;
+      x87_fnclex(cp->func);
+      x87_fldcw( cp->func, x86_make_disp(cp->machine_EDX, 
+                                         Offset(struct aos_machine, fpu_rnd_neg_inf)));
+   }
+}
+
+static void set_fpu_round_nearest( struct aos_compilation *cp )
+{
+   if (cp->fpucntl != FPU_RND_NEAREST) {
+      cp->fpucntl = FPU_RND_NEAREST;
+      x87_fnclex(cp->func);
+      x87_fldcw( cp->func, x86_make_disp(cp->machine_EDX, 
+                                         Offset(struct aos_machine, fpu_rnd_nearest)));
+   }
+}
+
+#if 0
+static void x87_emit_ex2( struct aos_compilation *cp )
+{
+   struct x86_reg st0 = x86_make_reg(file_x87, 0);
+   struct x86_reg st1 = x86_make_reg(file_x87, 1);
+   int stack = cp->func->x87_stack;
+
+//   set_fpu_round_neg_inf( cp );
+
+   x87_fld(cp->func, st0);      /* a a */
+   x87_fprndint( cp->func );	/* int(a) a*/
+   x87_fsubr(cp->func, st1, st0);    /* int(a) frc(a) */
+   x87_fxch(cp->func, st1);     /* frc(a) int(a) */
+   x87_f2xm1(cp->func);         /* (2^frc(a))-1 int(a) */
+   x87_fld1(cp->func);          /* 1 (2^frc(a))-1 int(a) */
+   x87_faddp(cp->func, st1);	/* 2^frac(a) int(a)  */
+   x87_fscale(cp->func);	/* (2^frac(a)*2^int(int(a))) int(a) */
+                                /* 2^a int(a) */
+   x87_fstp(cp->func, st1);     /* 2^a */
+
+   assert( stack == cp->func->x87_stack);
+      
+}
+#endif
+
+#if 0
+static void PIPE_CDECL print_reg( const char *msg,
+                                  const float *reg )
+{
+   debug_printf("%s: %f %f %f %f\n", msg, reg[0], reg[1], reg[2], reg[3]);
+}
+#endif
+
+#if 0
+static void emit_print( struct aos_compilation *cp,
+                        const char *message, /* must point to a static string! */
+                        unsigned file,
+                        unsigned idx )
+{
+   struct x86_reg ecx = x86_make_reg( file_REG32, reg_CX );
+   struct x86_reg arg = aos_get_shader_reg_ptr( cp, file, idx );
+   unsigned i;
+
+   /* There shouldn't be anything on the x87 stack.  Can add this
+    * capacity later if need be.
+    */
+   assert(cp->func->x87_stack == 0);
+
+   /* For absolute correctness, need to spill/invalidate all XMM regs
+    * too.  We're obviously not concerned about performance on this
+    * debug path, so here goes:
+    */
+   for (i = 0; i < 8; i++) {
+      if (cp->xmm[i].dirty) 
+         spill(cp, i);
+
+      aos_release_xmm_reg(cp, i);
+   }
+
+   /* Push caller-save (ie scratch) regs.  
+    */
+   x86_cdecl_caller_push_regs( cp->func );
+
+
+   /* Push the arguments:
+    */
+   x86_lea( cp->func, ecx, arg );
+   x86_push( cp->func, ecx );
+   x86_push_imm32( cp->func, (int)message );
+
+   /* Call the helper.  Could call debug_printf directly, but
+    * print_reg is a nice place to put a breakpoint if need be.
+    */
+   x86_mov_reg_imm( cp->func, ecx, (int)print_reg );
+   x86_call( cp->func, ecx );
+   x86_pop( cp->func, ecx );
+   x86_pop( cp->func, ecx );
+
+   /* Pop caller-save regs 
+    */
+   x86_cdecl_caller_pop_regs( cp->func );
+
+   /* Done... 
+    */
+}
+#endif
+
+/**
+ * The traditional instructions.  All operate on internal registers
+ * and ignore write masks and swizzling issues.
+ */
+
+static boolean emit_ABS( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) 
+{
+   struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+   struct x86_reg neg = aos_get_internal(cp, IMM_NEGS);
+   struct x86_reg tmp = aos_get_xmm_reg(cp);
+
+   sse_movaps(cp->func, tmp, arg0);
+   sse_mulps(cp->func, tmp, neg);
+   sse_maxps(cp->func, tmp, arg0);
+   
+   store_dest(cp, &op->FullDstRegisters[0], tmp);
+   return TRUE;
+}
+
+static boolean emit_ADD( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+   struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+   struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]);
+   struct x86_reg dst = get_xmm_writable(cp, arg0);
+
+   sse_addps(cp->func, dst, arg1);
+
+   store_dest(cp, &op->FullDstRegisters[0], dst);
+   return TRUE;
+}
+
+static boolean emit_COS( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) 
+{
+   x87_fld_src(cp, &op->FullSrcRegisters[0], 0);
+   x87_fcos(cp->func);
+   x87_fstp_dest4(cp, &op->FullDstRegisters[0]);
+   return TRUE;
+}
+
+/* The dotproduct instructions don't really do that well in sse:
+ * XXX: produces wrong results -- disabled.
+ */
+static boolean emit_DP3( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+   struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+   struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]);
+   struct x86_reg tmp = aos_get_xmm_reg(cp); 
+   struct x86_reg dst = get_xmm_writable(cp, arg0);
+
+   sse_mulps(cp->func, dst, arg1);
+   /* Now the hard bit: sum the first 3 values:
+    */ 
+   sse_movhlps(cp->func, tmp, dst);
+   sse_addss(cp->func, dst, tmp); /* a*x+c*z, b*y, ?, ? */
+   emit_pshufd(cp, tmp, dst, SHUF(Y,X,W,Z));
+   sse_addss(cp->func, dst, tmp);
+   
+   aos_release_xmm_reg(cp, tmp.idx);
+   store_scalar_dest(cp, &op->FullDstRegisters[0], dst);
+   return TRUE;
+}
+
+static boolean emit_DP4( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+   struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+   struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]);
+   struct x86_reg tmp = aos_get_xmm_reg(cp);      
+   struct x86_reg dst = get_xmm_writable(cp, arg0);
+
+   sse_mulps(cp->func, dst, arg1);
+   
+   /* Now the hard bit: sum the values:
+    */ 
+   sse_movhlps(cp->func, tmp, dst);
+   sse_addps(cp->func, dst, tmp); /* a*x+c*z, b*y+d*w, a*x+c*z, b*y+d*w */
+   emit_pshufd(cp, tmp, dst, SHUF(Y,X,W,Z));
+   sse_addss(cp->func, dst, tmp);
+
+   aos_release_xmm_reg(cp, tmp.idx);
+   store_scalar_dest(cp, &op->FullDstRegisters[0], dst);
+   return TRUE;
+}
+
+static boolean emit_DPH( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+   struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+   struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]);
+   struct x86_reg tmp = aos_get_xmm_reg(cp);
+   struct x86_reg dst = get_xmm_writable(cp, arg0);
+
+   sse_mulps(cp->func, dst, arg1);
+
+   /* Now the hard bit: sum the values (from DP3):
+    */ 
+   sse_movhlps(cp->func, tmp, dst);
+   sse_addss(cp->func, dst, tmp); /* a*x+c*z, b*y, ?, ? */
+   emit_pshufd(cp, tmp, dst, SHUF(Y,X,W,Z));
+   sse_addss(cp->func, dst, tmp);
+   emit_pshufd(cp, tmp, arg1, SHUF(W,W,W,W));
+   sse_addss(cp->func, dst, tmp);
+
+   aos_release_xmm_reg(cp, tmp.idx);
+   store_scalar_dest(cp, &op->FullDstRegisters[0], dst);
+   return TRUE;
+}
+
+static boolean emit_DST( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+    struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+    struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]);
+    struct x86_reg dst = aos_get_xmm_reg(cp);
+    struct x86_reg tmp = aos_get_xmm_reg(cp);
+    struct x86_reg ones = aos_get_internal(cp, IMM_ONES);
+
+/*    dst[0] = 1.0     * 1.0F; */
+/*    dst[1] = arg0[1] * arg1[1]; */
+/*    dst[2] = arg0[2] * 1.0; */
+/*    dst[3] = 1.0     * arg1[3]; */
+
+    emit_shuf_copy2(cp, dst, arg0, ones, SHUF(X,W,Z,Y));
+    emit_shuf_copy2(cp, tmp, arg1, ones, SHUF(X,Z,Y,W));
+    sse_mulps(cp->func, dst, tmp);
+
+    aos_release_xmm_reg(cp, tmp.idx);
+    store_dest(cp, &op->FullDstRegisters[0], dst);
+    return TRUE;
+}
+
+static boolean emit_LG2( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) 
+{
+   x87_fld1(cp->func);		/* 1 */
+   x87_fld_src(cp, &op->FullSrcRegisters[0], 0);	/* a0 1 */
+   x87_fyl2x(cp->func);	/* log2(a0) */
+   x87_fstp_dest4(cp, &op->FullDstRegisters[0]);
+   return TRUE;
+}
+
+#if 0
+static boolean emit_EX2( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) 
+{
+   x87_fld_src(cp, &op->FullSrcRegisters[0], 0);
+   x87_emit_ex2(cp);
+   x87_fstp_dest4(cp, &op->FullDstRegisters[0]);
+   return TRUE;
+}
+#endif
+
+
+static boolean emit_FLR( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) 
+{
+   struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); 
+   unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask;
+   int i;
+
+   set_fpu_round_neg_inf( cp );
+
+   /* Load all sources first to avoid aliasing
+    */
+   for (i = 3; i >= 0; i--) {
+      if (writemask & (1<<i)) {
+         x87_fld_src(cp, &op->FullSrcRegisters[0], i);   
+      }
+   }
+
+   for (i = 0; i < 4; i++) {
+      if (writemask & (1<<i)) {
+         x87_fprndint( cp->func );   
+         x87_fstp(cp->func, x86_make_disp(dst, i*4));
+      }
+   }
+
+   return TRUE;
+}
+
+
+static boolean emit_RND( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) 
+{
+   struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); 
+   unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask;
+   int i;
+
+   set_fpu_round_nearest( cp );
+
+   /* Load all sources first to avoid aliasing
+    */
+   for (i = 3; i >= 0; i--) {
+      if (writemask & (1<<i)) {
+         x87_fld_src(cp, &op->FullSrcRegisters[0], i);   
+      }
+   }
+
+   for (i = 0; i < 4; i++) {
+      if (writemask & (1<<i)) {
+         x87_fprndint( cp->func );   
+         x87_fstp(cp->func, x86_make_disp(dst, i*4));
+      }
+   }
+
+   return TRUE;
+}
+
+
+static boolean emit_FRC( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) 
+{
+   struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); 
+   struct x86_reg st0 = x86_make_reg(file_x87, 0);
+   struct x86_reg st1 = x86_make_reg(file_x87, 1);
+   unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask;
+   int i;
+
+   set_fpu_round_neg_inf( cp );
+
+   /* suck all the source values onto the stack before writing out any
+    * dst, which may alias...
+    */
+   for (i = 3; i >= 0; i--) {
+      if (writemask & (1<<i)) {
+         x87_fld_src(cp, &op->FullSrcRegisters[0], i);   
+      }
+   }
+
+   for (i = 0; i < 4; i++) {
+      if (writemask & (1<<i)) {
+         x87_fld(cp->func, st0);     /* a a */
+         x87_fprndint( cp->func );   /* flr(a) a */
+         x87_fsubp(cp->func, st1);  /* frc(a) */
+         x87_fstp(cp->func, x86_make_disp(dst, i*4));
+      }
+   }
+
+   return TRUE;
+}
+
+
+
+
+
+
+static boolean emit_LIT( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+   struct x86_reg ecx = x86_make_reg( file_REG32, reg_CX );
+   unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask;
+   unsigned lit_count = cp->lit_count++;
+   struct x86_reg result, arg0;
+   unsigned i;
+
+#if 1
+   /* For absolute correctness, need to spill/invalidate all XMM regs
+    * too.  
+    */
+   for (i = 0; i < 8; i++) {
+      if (cp->xmm[i].dirty) 
+         spill(cp, i);
+      aos_release_xmm_reg(cp, i);
+   }
+#endif
+
+   if (writemask != TGSI_WRITEMASK_XYZW) 
+      result = x86_make_disp(cp->machine_EDX, Offset(struct aos_machine, tmp[0]));
+   else 
+      result = get_dst_ptr(cp, &op->FullDstRegisters[0]);    
+
+   
+   arg0 = fetch_src( cp, &op->FullSrcRegisters[0] );
+   if (arg0.file == file_XMM) {
+      struct x86_reg tmp = x86_make_disp(cp->machine_EDX, 
+                                         Offset(struct aos_machine, tmp[1]));
+      sse_movaps( cp->func, tmp, arg0 );
+      arg0 = tmp;
+   }
+                  
+      
+
+   /* Push caller-save (ie scratch) regs.  
+    */
+   x86_cdecl_caller_push_regs( cp->func );
+
+   /* Push the arguments:
+    */
+   x86_push_imm32( cp->func, lit_count );
+
+   x86_lea( cp->func, ecx, arg0 );
+   x86_push( cp->func, ecx );
+
+   x86_lea( cp->func, ecx, result );
+   x86_push( cp->func, ecx );
+
+   x86_push( cp->func, cp->machine_EDX );
+
+   if (lit_count < MAX_LIT_INFO) {
+      x86_mov( cp->func, ecx, x86_make_disp( cp->machine_EDX, 
+                                             Offset(struct aos_machine, lit_info) + 
+                                             lit_count * sizeof(struct lit_info) + 
+                                             Offset(struct lit_info, func)));
+   }
+   else {
+      x86_mov_reg_imm( cp->func, ecx, (int)aos_do_lit );
+   }
+
+   x86_call( cp->func, ecx );
+            
+   x86_pop( cp->func, ecx );    /* fixme... */
+   x86_pop( cp->func, ecx );
+   x86_pop( cp->func, ecx );
+   x86_pop( cp->func, ecx );
+
+   x86_cdecl_caller_pop_regs( cp->func );
+
+   if (writemask != TGSI_WRITEMASK_XYZW) {
+      store_dest( cp, 
+                  &op->FullDstRegisters[0],
+                  get_xmm_writable( cp, result ) );
+   }
+
+   return TRUE;
+}
+
+#if 0   
+static boolean emit_inline_LIT( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+   struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); 
+   unsigned writemask = op->FullDstRegisters[0].DstRegister.WriteMask;
+
+   if (writemask & TGSI_WRITEMASK_YZ) {
+      struct x86_reg st1 = x86_make_reg(file_x87, 1);
+      struct x86_reg st2 = x86_make_reg(file_x87, 2);
+
+      /* a1' = a1 <= 0 ? 1 : a1;  
+       */
+      x87_fldz(cp->func);                           /* 1 0  */
+#if 1
+      x87_fld1(cp->func);                           /* 1 0  */
+#else
+      /* Correct but slow due to fp exceptions generated in fyl2x - fix me.
+       */
+      x87_fldz(cp->func);                           /* 1 0  */
+#endif
+      x87_fld_src(cp, &op->FullSrcRegisters[0], 1); /* a1 1 0  */
+      x87_fcomi(cp->func, st2);	                    /* a1 1 0  */
+      x87_fcmovb(cp->func, st1);                    /* a1' 1 0  */
+      x87_fstp(cp->func, st1);                      /* a1' 0  */
+      x87_fstp(cp->func, st1);                      /* a1'  */
+
+      x87_fld_src(cp, &op->FullSrcRegisters[0], 3); /* a3 a1'  */
+      x87_fxch(cp->func, st1);                      /* a1' a3  */
+      
+
+      /* Compute pow(a1, a3)
+       */
+      x87_fyl2x(cp->func);	/* a3*log2(a1)      */
+      x87_emit_ex2( cp );       /* 2^(a3*log2(a1))   */
+
+
+      /* a0' = max2(a0, 0):
+       */
+      x87_fldz(cp->func);                           /* 0 r2 */
+      x87_fld_src(cp, &op->FullSrcRegisters[0], 0); /* a0 0 r2 */
+      x87_fcomi(cp->func, st1);	
+      x87_fcmovb(cp->func, st1);                    /* a0' 0 r2 */
+
+      x87_fst_or_nop(cp->func, writemask, 1, dst); /* result[1] = a0' */
+
+      x87_fcomi(cp->func, st1);  /* a0' 0 r2 */
+      x87_fcmovnbe(cp->func, st2); /* r2' 0' r2 */
+
+      x87_fstp_or_pop(cp->func, writemask, 2, dst); /* 0 r2 */
+      x87_fpop(cp->func);       /* r2 */
+      x87_fpop(cp->func);
+   }
+
+   if (writemask & TGSI_WRITEMASK_XW) {
+      x87_fld1(cp->func);
+      x87_fst_or_nop(cp->func, writemask, 0, dst);
+      x87_fstp_or_pop(cp->func, writemask, 3, dst);
+   }
+
+   return TRUE;
+}
+#endif
+
+
+
+static boolean emit_MAX( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+   struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+   struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]);
+   struct x86_reg dst = get_xmm_writable(cp, arg0);
+
+   sse_maxps(cp->func, dst, arg1);
+
+   store_dest(cp, &op->FullDstRegisters[0], dst);
+   return TRUE;
+}
+
+
+static boolean emit_MIN( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+   struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+   struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]);
+   struct x86_reg dst = get_xmm_writable(cp, arg0);
+
+   sse_minps(cp->func, dst, arg1);
+
+   store_dest(cp, &op->FullDstRegisters[0], dst);
+   return TRUE;
+}
+
+static boolean emit_MOV( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+   struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+   struct x86_reg dst = get_xmm_writable(cp, arg0);
+
+   /* potentially nothing to do */
+
+   store_dest(cp, &op->FullDstRegisters[0], dst);
+   return TRUE;
+}
+
+static boolean emit_MUL( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+   struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+   struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]);
+   struct x86_reg dst = get_xmm_writable(cp, arg0);
+
+   sse_mulps(cp->func, dst, arg1);
+
+   store_dest(cp, &op->FullDstRegisters[0], dst);
+   return TRUE;
+}
+
+
+static boolean emit_MAD( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+   struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+   struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]);
+   struct x86_reg arg2 = fetch_src(cp, &op->FullSrcRegisters[2]);
+
+   /* If we can't clobber old contents of arg0, get a temporary & copy
+    * it there, then clobber it...
+    */
+   arg0 = get_xmm_writable(cp, arg0);
+
+   sse_mulps(cp->func, arg0, arg1);
+   sse_addps(cp->func, arg0, arg2);
+   store_dest(cp, &op->FullDstRegisters[0], arg0);
+   return TRUE;
+}
+
+
+
+/* A wrapper for powf().
+ * Makes sure it is cdecl and operates on floats.
+ */
+static float PIPE_CDECL _powerf( float x, float y )
+{
+#if FAST_MATH
+   return util_fast_pow(x, y);
+#else
+   return powf( x, y );
+#endif
+}
+
+#if FAST_MATH
+static float PIPE_CDECL _exp2(float x)
+{
+   return util_fast_exp2(x);
+}
+#endif
+
+
+/* Really not sufficient -- need to check for conditions that could
+ * generate inf/nan values, which will slow things down hugely.
+ */
+static boolean emit_POW( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) 
+{
+#if 0
+   x87_fld_src(cp, &op->FullSrcRegisters[1], 0);  /* a1.x */
+   x87_fld_src(cp, &op->FullSrcRegisters[0], 0);	/* a0.x a1.x */
+   x87_fyl2x(cp->func);	                                /* a1*log2(a0) */
+
+   x87_emit_ex2( cp );		/* 2^(a1*log2(a0)) */
+
+   x87_fstp_dest4(cp, &op->FullDstRegisters[0]);
+#else
+   uint i;
+
+   /* For absolute correctness, need to spill/invalidate all XMM regs
+    * too.  
+    */
+   for (i = 0; i < 8; i++) {
+      if (cp->xmm[i].dirty) 
+         spill(cp, i);
+      aos_release_xmm_reg(cp, i);
+   }
+
+   /* Push caller-save (ie scratch) regs.  
+    */
+   x86_cdecl_caller_push_regs( cp->func );
+
+   x86_lea( cp->func, cp->stack_ESP, x86_make_disp(cp->stack_ESP, -8) );
+
+   x87_fld_src( cp, &op->FullSrcRegisters[1], 0 );
+   x87_fstp( cp->func, x86_make_disp( cp->stack_ESP, 4 ) );
+   x87_fld_src( cp, &op->FullSrcRegisters[0], 0 );
+   x87_fstp( cp->func, x86_make_disp( cp->stack_ESP, 0 ) );
+
+   /* tmp_EAX has been pushed & will be restored below */
+   x86_mov_reg_imm( cp->func, cp->tmp_EAX, (unsigned long) _powerf );
+   x86_call( cp->func, cp->tmp_EAX );
+
+   x86_lea( cp->func, cp->stack_ESP, x86_make_disp(cp->stack_ESP, 8) );
+
+   x86_cdecl_caller_pop_regs( cp->func );
+
+   /* Note retval on x87 stack:
+    */
+   cp->func->x87_stack++;
+
+   x87_fstp_dest4( cp, &op->FullDstRegisters[0] );
+#endif
+   return TRUE;
+}
+
+
+#if FAST_MATH
+static boolean emit_EXPBASE2( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) 
+{
+   uint i;
+
+   /* For absolute correctness, need to spill/invalidate all XMM regs
+    * too.  
+    */
+   for (i = 0; i < 8; i++) {
+      if (cp->xmm[i].dirty) 
+         spill(cp, i);
+      aos_release_xmm_reg(cp, i);
+   }
+
+   /* Push caller-save (ie scratch) regs.  
+    */
+   x86_cdecl_caller_push_regs( cp->func );
+
+   x86_lea( cp->func, cp->stack_ESP, x86_make_disp(cp->stack_ESP, -4) );
+
+   x87_fld_src( cp, &op->FullSrcRegisters[0], 0 );
+   x87_fstp( cp->func, x86_make_disp( cp->stack_ESP, 0 ) );
+
+   /* tmp_EAX has been pushed & will be restored below */
+   x86_mov_reg_imm( cp->func, cp->tmp_EAX, (unsigned long) _exp2 );
+   x86_call( cp->func, cp->tmp_EAX );
+
+   x86_lea( cp->func, cp->stack_ESP, x86_make_disp(cp->stack_ESP, 4) );
+
+   x86_cdecl_caller_pop_regs( cp->func );
+
+   /* Note retval on x87 stack:
+    */
+   cp->func->x87_stack++;
+
+   x87_fstp_dest4( cp, &op->FullDstRegisters[0] );
+
+   return TRUE;
+}
+#endif
+
+
+static boolean emit_RCP( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+   struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+   struct x86_reg dst = aos_get_xmm_reg(cp);
+
+   if (cp->have_sse2) {
+      sse2_rcpss(cp->func, dst, arg0);
+      /* extend precision here...
+       */
+   }
+   else {
+      struct x86_reg ones = aos_get_internal(cp, IMM_ONES);
+      sse_movss(cp->func, dst, ones);
+      sse_divss(cp->func, dst, arg0);
+   }
+
+   store_scalar_dest(cp, &op->FullDstRegisters[0], dst);
+   return TRUE;
+}
+
+
+/* Although rsqrtps() and rcpps() are low precision on some/all SSE
+ * implementations, it is possible to improve its precision at
+ * fairly low cost, using a newton/raphson step, as below:
+ * 
+ * x1 = 2 * rcpps(a) - a * rcpps(a) * rcpps(a)
+ * x1 = 0.5 * rsqrtps(a) * [3.0 - (a * rsqrtps(a))* rsqrtps(a)]
+ * or:
+ *   x1 = rsqrtps(a) * [1.5 - .5 * a * rsqrtps(a) * rsqrtps(a)]
+ * 
+ *
+ * See: http://softwarecommunity.intel.com/articles/eng/1818.htm
+ */
+static boolean emit_RSQ( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+
+   if (0) {
+      struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+      struct x86_reg r = aos_get_xmm_reg(cp);
+      sse_rsqrtss(cp->func, r, arg0);
+      store_scalar_dest(cp, &op->FullDstRegisters[0], r);
+      return TRUE;
+   }
+   else {
+      struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+      struct x86_reg r = aos_get_xmm_reg(cp);
+
+      struct x86_reg neg_half       = get_reg_ptr( cp, AOS_FILE_INTERNAL, IMM_RSQ );
+      struct x86_reg one_point_five = x86_make_disp( neg_half, 4 );
+      struct x86_reg src            = get_xmm_writable( cp, arg0 );
+      
+      sse_rsqrtss( cp->func, r, src  );             /* rsqrtss(a) */
+      sse_mulss(   cp->func, src, neg_half  );      /* -.5 * a */
+      sse_mulss(   cp->func, src,  r );             /* -.5 * a * r */
+      sse_mulss(   cp->func, src,  r );             /* -.5 * a * r * r */
+      sse_addss(   cp->func, src, one_point_five ); /* 1.5 - .5 * a * r * r */
+      sse_mulss(   cp->func, r,  src );             /* r * (1.5 - .5 * a * r * r) */
+
+      store_scalar_dest(cp, &op->FullDstRegisters[0], r);
+      return TRUE;
+   }
+}
+
+
+static boolean emit_SGE( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+   struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+   struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]);
+   struct x86_reg ones = aos_get_internal(cp, IMM_ONES);
+   struct x86_reg dst = get_xmm_writable(cp, arg0);
+
+   sse_cmpps(cp->func, dst, arg1, cc_NotLessThan);
+   sse_andps(cp->func, dst, ones);
+
+   store_dest(cp, &op->FullDstRegisters[0], dst);
+   return TRUE;
+}
+
+static boolean emit_SIN( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) 
+{
+   x87_fld_src(cp, &op->FullSrcRegisters[0], 0);
+   x87_fsin(cp->func);
+   x87_fstp_dest4(cp, &op->FullDstRegisters[0]);
+   return TRUE;
+}
+
+
+
+static boolean emit_SLT( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+   struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+   struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]);
+   struct x86_reg ones = aos_get_internal(cp, IMM_ONES);
+   struct x86_reg dst = get_xmm_writable(cp, arg0);
+   
+   sse_cmpps(cp->func, dst, arg1, cc_LessThan);
+   sse_andps(cp->func, dst, ones);
+
+   store_dest(cp, &op->FullDstRegisters[0], dst);
+   return TRUE;
+}
+
+static boolean emit_SUB( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) 
+{
+   struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+   struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]);
+   struct x86_reg dst = get_xmm_writable(cp, arg0);
+
+   sse_subps(cp->func, dst, arg1);
+
+   store_dest(cp, &op->FullDstRegisters[0], dst);
+   return TRUE;
+}
+
+static boolean emit_TRUNC( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
+{
+   struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+   struct x86_reg tmp0 = aos_get_xmm_reg(cp);
+
+   sse2_cvttps2dq(cp->func, tmp0, arg0);
+   sse2_cvtdq2ps(cp->func, tmp0, tmp0);
+
+   store_dest(cp, &op->FullDstRegisters[0], tmp0);
+   return TRUE;
+}
+
+static boolean emit_XPD( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) 
+{
+   struct x86_reg arg0 = fetch_src(cp, &op->FullSrcRegisters[0]);
+   struct x86_reg arg1 = fetch_src(cp, &op->FullSrcRegisters[1]);
+   struct x86_reg tmp0 = aos_get_xmm_reg(cp);
+   struct x86_reg tmp1 = aos_get_xmm_reg(cp);
+
+   emit_pshufd(cp, tmp1, arg1, SHUF(Y, Z, X, W));
+   sse_mulps(cp->func, tmp1, arg0);
+   emit_pshufd(cp, tmp0, arg0, SHUF(Y, Z, X, W));
+   sse_mulps(cp->func, tmp0, arg1);
+   sse_subps(cp->func, tmp1, tmp0);
+   sse_shufps(cp->func, tmp1, tmp1, SHUF(Y, Z, X, W));
+
+/*    dst[2] = arg0[0] * arg1[1] - arg0[1] * arg1[0]; */
+/*    dst[0] = arg0[1] * arg1[2] - arg0[2] * arg1[1]; */
+/*    dst[1] = arg0[2] * arg1[0] - arg0[0] * arg1[2]; */
+/*    dst[3] is undef */
+
+
+   aos_release_xmm_reg(cp, tmp0.idx);
+   store_dest(cp, &op->FullDstRegisters[0], tmp1);
+   return TRUE;
+}
+
+
+
+static boolean
+emit_instruction( struct aos_compilation *cp,
+                  struct tgsi_full_instruction *inst )
+{
+   x87_assert_stack_empty(cp->func);
+
+   switch( inst->Instruction.Opcode ) {
+   case TGSI_OPCODE_MOV:
+      return emit_MOV( cp, inst );
+
+   case TGSI_OPCODE_LIT:
+      return emit_LIT(cp, inst);
+
+   case TGSI_OPCODE_RCP:
+      return emit_RCP(cp, inst);
+
+   case TGSI_OPCODE_RSQ:
+      return emit_RSQ(cp, inst);
+
+   case TGSI_OPCODE_EXP:
+      /*return emit_EXP(cp, inst);*/
+      return FALSE;
+
+   case TGSI_OPCODE_LOG:
+      /*return emit_LOG(cp, inst);*/
+      return FALSE;
+
+   case TGSI_OPCODE_MUL:
+      return emit_MUL(cp, inst);
+
+   case TGSI_OPCODE_ADD:
+      return emit_ADD(cp, inst);
+
+   case TGSI_OPCODE_DP3:
+      return emit_DP3(cp, inst);
+
+   case TGSI_OPCODE_DP4:
+      return emit_DP4(cp, inst);
+
+   case TGSI_OPCODE_DST:
+      return emit_DST(cp, inst);
+
+   case TGSI_OPCODE_MIN:
+      return emit_MIN(cp, inst);
+
+   case TGSI_OPCODE_MAX:
+      return emit_MAX(cp, inst);
+
+   case TGSI_OPCODE_SLT:
+      return emit_SLT(cp, inst);
+
+   case TGSI_OPCODE_SGE:
+      return emit_SGE(cp, inst);
+
+   case TGSI_OPCODE_MAD:
+      return emit_MAD(cp, inst);
+
+   case TGSI_OPCODE_SUB:
+      return emit_SUB(cp, inst);
+ 
+   case TGSI_OPCODE_LERP:
+//      return emit_LERP(cp, inst);
+      return FALSE;
+
+   case TGSI_OPCODE_FRAC:
+      return emit_FRC(cp, inst);
+
+   case TGSI_OPCODE_CLAMP:
+//      return emit_CLAMP(cp, inst);
+      return FALSE;
+
+   case TGSI_OPCODE_FLOOR:
+      return emit_FLR(cp, inst);
+
+   case TGSI_OPCODE_ROUND:
+      return emit_RND(cp, inst);
+
+   case TGSI_OPCODE_EXPBASE2:
+#if FAST_MATH
+      return emit_EXPBASE2(cp, inst);
+#elif 0
+      /* this seems to fail for "larger" exponents.
+       * See glean tvertProg1's EX2 test.
+       */
+      return emit_EX2(cp, inst);
+#else
+      return FALSE;
+#endif
+
+   case TGSI_OPCODE_LOGBASE2:
+      return emit_LG2(cp, inst);
+
+   case TGSI_OPCODE_POWER:
+      return emit_POW(cp, inst);
+
+   case TGSI_OPCODE_CROSSPRODUCT:
+      return emit_XPD(cp, inst);
+
+   case TGSI_OPCODE_ABS:
+      return emit_ABS(cp, inst);
+
+   case TGSI_OPCODE_DPH:
+      return emit_DPH(cp, inst);
+
+   case TGSI_OPCODE_COS:
+      return emit_COS(cp, inst);
+
+   case TGSI_OPCODE_SIN:
+      return emit_SIN(cp, inst);
+
+   case TGSI_OPCODE_TRUNC:
+      return emit_TRUNC(cp, inst);
+
+   case TGSI_OPCODE_END:
+      return TRUE;
+
+   default:
+      return FALSE;
+   }
+}
+
+
+static boolean emit_viewport( struct aos_compilation *cp )
+{
+   struct x86_reg pos = aos_get_shader_reg_xmm(cp, 
+                                               TGSI_FILE_OUTPUT, 
+                                               cp->vaos->draw->vs.position_output );
+
+   struct x86_reg scale = x86_make_disp(cp->machine_EDX, 
+                                        Offset(struct aos_machine, scale));
+
+   struct x86_reg translate = x86_make_disp(cp->machine_EDX, 
+                                        Offset(struct aos_machine, translate));
+
+   sse_mulps(cp->func, pos, scale);
+   sse_addps(cp->func, pos, translate);
+
+   aos_adopt_xmm_reg( cp,
+                      pos,
+                      TGSI_FILE_OUTPUT,
+                      cp->vaos->draw->vs.position_output,
+                      TRUE );
+   return TRUE;
+}
+
+
+/* This is useful to be able to see the results on softpipe.  Doesn't
+ * do proper clipping, just assumes the backend can do it during
+ * rasterization -- for debug only...
+ */
+static boolean emit_rhw_viewport( struct aos_compilation *cp )
+{
+   struct x86_reg tmp = aos_get_xmm_reg(cp);
+   struct x86_reg pos = aos_get_shader_reg_xmm(cp, 
+                                               TGSI_FILE_OUTPUT, 
+                                               cp->vaos->draw->vs.position_output);
+
+   struct x86_reg scale = x86_make_disp(cp->machine_EDX, 
+                                        Offset(struct aos_machine, scale));
+
+   struct x86_reg translate = x86_make_disp(cp->machine_EDX, 
+                                        Offset(struct aos_machine, translate));
+
+
+
+   emit_pshufd(cp, tmp, pos, SHUF(W, W, W, W));
+   sse2_rcpss(cp->func, tmp, tmp);
+   sse_shufps(cp->func, tmp, tmp, SHUF(X, X, X, X));
+   
+   sse_mulps(cp->func, pos, scale);
+   sse_mulps(cp->func, pos, tmp);
+   sse_addps(cp->func, pos, translate);
+
+   /* Set pos[3] = w 
+    */
+   mask_write(cp, pos, tmp, TGSI_WRITEMASK_W);
+
+   aos_adopt_xmm_reg( cp,
+                      pos,
+                      TGSI_FILE_OUTPUT,
+                      cp->vaos->draw->vs.position_output,
+                      TRUE );
+   return TRUE;
+}
+
+
+#if 0
+static boolean note_immediate( struct aos_compilation *cp,
+                               struct tgsi_full_immediate *imm )
+{
+   unsigned pos = cp->num_immediates++;
+   unsigned j;
+
+   for (j = 0; j < imm->Immediate.NrTokens - 1; j++) {
+      cp->vaos->machine->immediate[pos][j] = imm->u.ImmediateFloat32[j].Float;
+   }
+
+   return TRUE;
+}
+#endif
+
+
+
+
+static void find_last_write_outputs( struct aos_compilation *cp )
+{
+   struct tgsi_parse_context parse;
+   unsigned this_instruction = 0;
+   unsigned i;
+
+   tgsi_parse_init( &parse, cp->vaos->base.vs->state.tokens );
+
+   while (!tgsi_parse_end_of_tokens( &parse )) {
+      
+      tgsi_parse_token( &parse );
+
+      if (parse.FullToken.Token.Type != TGSI_TOKEN_TYPE_INSTRUCTION) 
+         continue;
+
+      for (i = 0; i < TGSI_FULL_MAX_DST_REGISTERS; i++) {
+         if (parse.FullToken.FullInstruction.FullDstRegisters[i].DstRegister.File ==
+             TGSI_FILE_OUTPUT) 
+         {
+            unsigned idx = parse.FullToken.FullInstruction.FullDstRegisters[i].DstRegister.Index;
+            cp->output_last_write[idx] = this_instruction;
+         }
+      }
+
+      this_instruction++;
+   }
+
+   tgsi_parse_free( &parse );
+}
+
+
+#define ARG_MACHINE    1
+#define ARG_START_ELTS 2
+#define ARG_COUNT      3
+#define ARG_OUTBUF     4
+
+
+static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
+                                     boolean linear )
+{ 
+   struct tgsi_parse_context parse;
+   struct aos_compilation cp;
+   unsigned fixup, label;
+
+   util_init_math();
+
+   tgsi_parse_init( &parse, varient->base.vs->state.tokens );
+
+   memset(&cp, 0, sizeof(cp));
+
+   cp.insn_counter = 1;
+   cp.vaos = varient;
+   cp.have_sse2 = 1;
+   cp.func = &varient->func[ linear ? 0 : 1 ];
+
+   cp.tmp_EAX       = x86_make_reg(file_REG32, reg_AX);
+   cp.idx_EBX      = x86_make_reg(file_REG32, reg_BX);
+   cp.outbuf_ECX    = x86_make_reg(file_REG32, reg_CX);
+   cp.machine_EDX   = x86_make_reg(file_REG32, reg_DX);
+   cp.count_ESI     = x86_make_reg(file_REG32, reg_SI);
+   cp.temp_EBP     = x86_make_reg(file_REG32, reg_BP);
+   cp.stack_ESP = x86_make_reg( file_REG32, reg_SP );
+
+   x86_init_func(cp.func);
+
+   find_last_write_outputs(&cp);
+
+   x86_push(cp.func, cp.idx_EBX);
+   x86_push(cp.func, cp.count_ESI);
+   x86_push(cp.func, cp.temp_EBP);
+
+
+   /* Load arguments into regs:
+    */
+   x86_mov(cp.func, cp.machine_EDX, x86_fn_arg(cp.func, ARG_MACHINE));
+   x86_mov(cp.func, cp.idx_EBX, x86_fn_arg(cp.func, ARG_START_ELTS));
+   x86_mov(cp.func, cp.count_ESI, x86_fn_arg(cp.func, ARG_COUNT));
+   x86_mov(cp.func, cp.outbuf_ECX, x86_fn_arg(cp.func, ARG_OUTBUF));
+
+
+   /* Compare count to zero and possibly bail.
+    */
+   x86_xor(cp.func, cp.tmp_EAX, cp.tmp_EAX);
+   x86_cmp(cp.func, cp.count_ESI, cp.tmp_EAX);
+   fixup = x86_jcc_forward(cp.func, cc_E);
+
+
+   save_fpu_state( &cp );
+   set_fpu_round_nearest( &cp );
+
+   aos_init_inputs( &cp, linear );
+
+   cp.x86_reg[0] = 0;
+   cp.x86_reg[1] = 0;
+   
+   /* Note address for loop jump 
+    */
+   label = x86_get_label(cp.func);
+   {
+      /* Fetch inputs...  TODO:  fetch lazily...
+       */
+      if (!aos_fetch_inputs( &cp, linear ))
+         goto fail;
+
+      /* Emit the shader:
+       */
+      while( !tgsi_parse_end_of_tokens( &parse ) && !cp.error ) 
+      {
+         tgsi_parse_token( &parse );
+
+         switch (parse.FullToken.Token.Type) {
+         case TGSI_TOKEN_TYPE_IMMEDIATE:
+#if 0
+            if (!note_immediate( &cp, &parse.FullToken.FullImmediate ))
+               goto fail;
+#endif
+            break;
+
+         case TGSI_TOKEN_TYPE_INSTRUCTION:
+            if (DISASSEM)
+               tgsi_dump_instruction( &parse.FullToken.FullInstruction, cp.insn_counter );
+
+            if (!emit_instruction( &cp, &parse.FullToken.FullInstruction ))
+               goto fail;
+            break;
+         }
+
+         x87_assert_stack_empty(cp.func);
+         cp.insn_counter++;
+
+         if (DISASSEM)
+            debug_printf("\n");
+      }
+
+   
+      {
+         unsigned i;
+         for (i = 0; i < 8; i++) {
+            if (cp.xmm[i].file != TGSI_FILE_OUTPUT) {
+               cp.xmm[i].file = TGSI_FILE_NULL;
+               cp.xmm[i].dirty = 0;
+            }
+         }
+      }
+
+      if (cp.error)
+         goto fail;
+
+      if (cp.vaos->base.key.clip) {
+         /* not really handling clipping, just do the rhw so we can
+          * see the results...
+          */
+         emit_rhw_viewport(&cp); 
+      }
+      else if (cp.vaos->base.key.viewport) {
+         emit_viewport(&cp);
+      }
+
+      /* Emit output...  TODO: do this eagerly after the last write to a
+       * given output.
+       */
+      if (!aos_emit_outputs( &cp ))
+         goto fail;
+
+
+      /* Next vertex:
+       */
+      x86_lea(cp.func, 
+              cp.outbuf_ECX, 
+              x86_make_disp(cp.outbuf_ECX, 
+                            cp.vaos->base.key.output_stride));
+
+      /* Incr index
+       */   
+      aos_incr_inputs( &cp, linear );
+   }
+   /* decr count, loop if not zero
+    */
+   x86_dec(cp.func, cp.count_ESI);
+   x86_jcc(cp.func, cc_NZ, label);
+
+   restore_fpu_state(&cp);
+
+   /* Land forward jump here:
+    */
+   x86_fixup_fwd_jump(cp.func, fixup);
+
+   /* Exit mmx state?
+    */
+   if (cp.func->need_emms)
+      mmx_emms(cp.func);
+
+   x86_pop(cp.func, cp.temp_EBP);
+   x86_pop(cp.func, cp.count_ESI);
+   x86_pop(cp.func, cp.idx_EBX);
+
+   x87_assert_stack_empty(cp.func);
+   x86_ret(cp.func);
+
+   tgsi_parse_free( &parse );
+   return !cp.error;
+
+ fail:
+   tgsi_parse_free( &parse );
+   return FALSE;
+}
+
+
+
+static void vaos_set_buffer( struct draw_vs_varient *varient,
+                             unsigned buf,
+                             const void *ptr,
+                             unsigned stride )
+{
+   struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
+
+   if (buf < vaos->nr_vb) {
+      vaos->buffer[buf].base_ptr = (char *)ptr;
+      vaos->buffer[buf].stride = stride;
+   }
+
+   if (0) debug_printf("%s %d/%d: %p %d\n", __FUNCTION__, buf, vaos->nr_vb, ptr, stride);
+}
+
+
+
+static void PIPE_CDECL vaos_run_elts( struct draw_vs_varient *varient,
+                                      const unsigned *elts,
+                                      unsigned count,
+                                      void *output_buffer )
+{
+   struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
+   struct aos_machine *machine = vaos->draw->vs.aos_machine;
+
+   if (0) debug_printf("%s %d\n", __FUNCTION__, count);
+
+   machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
+   machine->constants = vaos->draw->vs.aligned_constants;
+   machine->immediates = vaos->base.vs->immediates;
+   machine->buffer = vaos->buffer;
+
+   vaos->gen_run_elts( machine,
+                       elts,
+                       count,
+                       output_buffer );
+}
+
+static void PIPE_CDECL vaos_run_linear( struct draw_vs_varient *varient,
+                                        unsigned start,
+                                        unsigned count,
+                                        void *output_buffer )
+{
+   struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
+   struct aos_machine *machine = vaos->draw->vs.aos_machine;
+
+   if (0) debug_printf("%s %d %d const: %x\n", __FUNCTION__, start, count, 
+                       vaos->base.key.const_vbuffers);
+
+   machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
+   machine->constants = vaos->draw->vs.aligned_constants;
+   machine->immediates = vaos->base.vs->immediates;
+   machine->buffer = vaos->buffer;
+
+   vaos->gen_run_linear( machine,
+                         start,
+                         count,
+                         output_buffer );
+
+   /* Sanity spot checks to make sure we didn't trash our constants */
+   assert(machine->internal[IMM_ONES][0] == 1.0f);
+   assert(machine->internal[IMM_IDENTITY][0] == 0.0f);
+   assert(machine->internal[IMM_NEGS][0] == -1.0f);
+}
+
+
+
+static void vaos_destroy( struct draw_vs_varient *varient )
+{
+   struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
+
+   FREE( vaos->buffer );
+
+   x86_release_func( &vaos->func[0] );
+   x86_release_func( &vaos->func[1] );
+
+   FREE(vaos);
+}
+
+
+
+static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
+                                                 const struct draw_vs_varient_key *key )
+{
+   unsigned i;
+   struct draw_vs_varient_aos_sse *vaos = CALLOC_STRUCT(draw_vs_varient_aos_sse);
+
+   if (!vaos)
+      goto fail;
+   
+   vaos->base.key = *key;
+   vaos->base.vs = vs;
+   vaos->base.set_buffer = vaos_set_buffer;
+   vaos->base.destroy = vaos_destroy;
+   vaos->base.run_linear = vaos_run_linear;
+   vaos->base.run_elts = vaos_run_elts;
+
+   vaos->draw = vs->draw;
+
+   for (i = 0; i < key->nr_inputs; i++) 
+      vaos->nr_vb = MAX2( vaos->nr_vb, key->element[i].in.buffer + 1 );
+
+   vaos->buffer = MALLOC( vaos->nr_vb * sizeof(vaos->buffer[0]) );
+   if (!vaos->buffer)
+      goto fail;
+
+   if (0)
+      debug_printf("nr_vb: %d const: %x\n", vaos->nr_vb, vaos->base.key.const_vbuffers);
+
+#if 0
+   tgsi_dump(vs->state.tokens, 0);
+#endif
+
+   if (!build_vertex_program( vaos, TRUE ))
+      goto fail;
+
+   if (!build_vertex_program( vaos, FALSE ))
+      goto fail;
+
+   vaos->gen_run_linear = (vaos_run_linear_func)x86_get_func(&vaos->func[0]);
+   if (!vaos->gen_run_linear)
+      goto fail;
+
+   vaos->gen_run_elts = (vaos_run_elts_func)x86_get_func(&vaos->func[1]);
+   if (!vaos->gen_run_elts)
+      goto fail;
+
+   return &vaos->base;
+
+ fail:
+   if (vaos && vaos->buffer)
+      FREE(vaos->buffer);
+
+   if (vaos)
+      x86_release_func( &vaos->func[0] );
+
+   if (vaos)
+      x86_release_func( &vaos->func[1] );
+
+   FREE(vaos);
+   
+   return NULL;
+}
+
+
+struct draw_vs_varient *draw_vs_varient_aos_sse( struct draw_vertex_shader *vs,
+                                                 const struct draw_vs_varient_key *key )
+{
+   struct draw_vs_varient *varient = varient_aos_sse( vs, key );
+
+   if (varient == NULL) {
+      varient = draw_vs_varient_generic( vs, key );
+   }
+
+   return varient;
+}
+
+
+
+#endif /* PIPE_ARCH_X86 */
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
new file mode 100644
index 00000000000..264387517b0
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -0,0 +1,253 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <[email protected]>
+ */
+
+#ifndef DRAW_VS_AOS_H
+#define DRAW_VS_AOS_H
+
+#include "pipe/p_config.h"
+
+#ifdef PIPE_ARCH_X86
+
+struct tgsi_token;
+struct x86_function;
+
+#include "pipe/p_state.h"
+#include "rtasm/rtasm_x86sse.h"
+
+
+
+
+
+#define X    0
+#define Y    1
+#define Z    2
+#define W    3
+
+#define MAX_INPUTS     PIPE_MAX_ATTRIBS
+#define MAX_OUTPUTS    PIPE_MAX_SHADER_OUTPUTS
+#define MAX_TEMPS      TGSI_EXEC_NUM_TEMPS
+#define MAX_CONSTANTS  1024  /** only used for sanity checking */
+#define MAX_IMMEDIATES 1024  /** only used for sanity checking */
+#define MAX_INTERNALS  8     /** see IMM_x values below */
+
+#define AOS_FILE_INTERNAL TGSI_FILE_COUNT
+
+#define FPU_RND_NEG    1
+#define FPU_RND_NEAREST 2
+
+struct aos_machine;
+typedef void (PIPE_CDECL *lit_func)( struct aos_machine *,
+                                    float *result,
+                                    const float *in,
+                                    unsigned count );
+
+void PIPE_CDECL aos_do_lit( struct aos_machine *machine,
+                            float *result,
+                            const float *in,
+                            unsigned count );
+
+struct shine_tab {
+   float exponent;
+   float values[258];
+   unsigned last_used;
+};
+
+struct lit_info {
+   lit_func func;
+   struct shine_tab *shine_tab;
+};
+
+#define MAX_SHINE_TAB    4
+#define MAX_LIT_INFO     16
+
+struct aos_buffer {
+   const void *base_ptr;
+   unsigned stride;
+   void *ptr;                   /* updated per vertex */
+};
+
+
+
+
+/* This is the temporary storage used by all the aos_sse vs varients.
+ * Create one per context and reuse by passing a pointer in at
+ * vs_varient creation??
+ */
+struct aos_machine {
+   float input    [MAX_INPUTS    ][4];
+   float output   [MAX_OUTPUTS   ][4];
+   float temp     [MAX_TEMPS     ][4];
+   float internal [MAX_INTERNALS ][4];
+
+   float scale[4];              /* viewport */
+   float translate[4];          /* viewport */
+
+   float tmp[2][4];             /* scratch space for LIT */
+
+   struct shine_tab shine_tab[MAX_SHINE_TAB];
+   struct lit_info  lit_info[MAX_LIT_INFO];
+   unsigned now;
+   
+
+   ushort fpu_rnd_nearest;
+   ushort fpu_rnd_neg_inf;
+   ushort fpu_restore;
+   ushort fpucntl;              /* one of FPU_* above */
+
+   const float (*immediates)[4];     /* points to shader data */
+   const float (*constants)[4];      /* points to draw data */
+
+   const struct aos_buffer *buffer; /* points to ? */
+};
+
+
+
+
+struct aos_compilation {
+   struct x86_function *func;
+   struct draw_vs_varient_aos_sse *vaos;
+
+   unsigned insn_counter;
+   unsigned num_immediates;
+   unsigned count;
+   unsigned lit_count;
+
+   struct {
+      unsigned idx:16;
+      unsigned file:8;
+      unsigned dirty:8;
+      unsigned last_used;
+   } xmm[8];
+
+   unsigned x86_reg[2];                /* one of X86_* */
+
+   boolean input_fetched[PIPE_MAX_ATTRIBS];
+   unsigned output_last_write[PIPE_MAX_ATTRIBS];
+
+   boolean have_sse2;
+   boolean error;
+   short fpucntl;
+
+   /* these are actually known values, but putting them in a struct
+    * like this is helpful to keep them in sync across the file.
+    */
+   struct x86_reg tmp_EAX;
+   struct x86_reg idx_EBX;     /* either start+i or &elt[i] */
+   struct x86_reg outbuf_ECX;
+   struct x86_reg machine_EDX;
+   struct x86_reg count_ESI;    /* decrements to zero */
+   struct x86_reg temp_EBP;
+   struct x86_reg stack_ESP;
+};
+
+struct x86_reg aos_get_xmm_reg( struct aos_compilation *cp );
+void aos_release_xmm_reg( struct aos_compilation *cp, unsigned idx );
+
+void aos_adopt_xmm_reg( struct aos_compilation *cp,
+                        struct x86_reg reg,
+                        unsigned file,
+                        unsigned idx,
+                        unsigned dirty );
+
+void aos_spill_all( struct aos_compilation *cp );
+
+struct x86_reg aos_get_shader_reg( struct aos_compilation *cp, 
+                                   unsigned file,
+                                   unsigned idx );
+
+boolean aos_init_inputs( struct aos_compilation *cp, boolean linear );
+boolean aos_fetch_inputs( struct aos_compilation *cp, boolean linear );
+boolean aos_incr_inputs( struct aos_compilation *cp, boolean linear );
+
+boolean aos_emit_outputs( struct aos_compilation *cp );
+
+
+#define IMM_ONES     0              /* 1, 1,1,1 */
+#define IMM_SWZ      1              /* 1,-1,0, 0xffffffff */
+#define IMM_IDENTITY 2              /* 0, 0,0,1 */
+#define IMM_INV_255  3              /* 1/255, 1/255, 1/255, 1/255 */
+#define IMM_255      4              /* 255, 255, 255, 255 */
+#define IMM_NEGS     5              /* -1,-1,-1,-1 */
+#define IMM_RSQ      6              /* -.5,1.5,_,_ */
+#define IMM_PSIZE    7              /* not really an immediate - updated each run */
+
+struct x86_reg aos_get_internal( struct aos_compilation *cp,
+                                 unsigned imm );
+struct x86_reg aos_get_internal_xmm( struct aos_compilation *cp,
+                                     unsigned imm );
+
+
+#define ERROR(cp, msg)                                                  \
+do {                                                                    \
+   if (0) debug_printf("%s: x86 translation failed: %s\n", __FUNCTION__, msg); \
+   cp->error = 1;                                                       \
+} while (0)
+
+
+#define X86_NULL       0
+#define X86_IMMEDIATES 1
+#define X86_CONSTANTS  2
+#define X86_BUFFERS    3
+
+struct x86_reg aos_get_x86( struct aos_compilation *cp,
+                            unsigned which_reg,
+                            unsigned value );
+
+
+typedef void (PIPE_CDECL *vaos_run_elts_func)( struct aos_machine *,
+                                               const unsigned *elts,
+                                               unsigned count,
+                                               void *output_buffer);
+
+typedef void (PIPE_CDECL *vaos_run_linear_func)( struct aos_machine *,
+                                                unsigned start,
+                                                unsigned count,
+                                                void *output_buffer);
+
+
+struct draw_vs_varient_aos_sse {
+   struct draw_vs_varient base;
+   struct draw_context *draw;
+
+   struct aos_buffer *buffer;
+   unsigned nr_vb;
+
+   vaos_run_linear_func gen_run_linear;
+   vaos_run_elts_func gen_run_elts;
+
+
+   struct x86_function func[2];
+};
+
+
+#endif
+
+#endif 
+
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_io.c b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
new file mode 100644
index 00000000000..39f75b50b76
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
@@ -0,0 +1,462 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "util/u_memory.h"
+#include "pipe/p_shader_tokens.h"
+#include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_util.h"
+#include "tgsi/tgsi_exec.h"
+#include "draw_vs.h"
+#include "draw_vs_aos.h"
+#include "draw_vertex.h"
+
+#include "rtasm/rtasm_x86sse.h"
+
+#ifdef PIPE_ARCH_X86
+
+/* Note - don't yet have to worry about interacting with the code in
+ * draw_vs_aos.c as there is no intermingling of generated code...
+ * That may have to change, we'll see.
+ */
+static void emit_load_R32G32B32A32( struct aos_compilation *cp, 			   
+				    struct x86_reg data,
+				    struct x86_reg src_ptr )
+{
+   sse_movups(cp->func, data, src_ptr);
+}
+
+static void emit_load_R32G32B32( struct aos_compilation *cp, 			   
+				 struct x86_reg data,
+				 struct x86_reg src_ptr )
+{
+#if 1
+   sse_movss(cp->func, data, x86_make_disp(src_ptr, 8));
+   /* data = z ? ? ? */
+   sse_shufps(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ), SHUF(X,Y,Z,W) );
+   /* data = z ? 0 1 */
+   sse_shufps(cp->func, data, data, SHUF(Y,Z,X,W) );
+   /* data = ? 0 z 1 */
+   sse_movlps(cp->func, data, src_ptr);
+   /* data = x y z 1 */
+#else
+   sse_movups(cp->func, data, src_ptr);
+   /* data = x y z ? */
+   sse2_pshufd(cp->func, data, data, SHUF(W,X,Y,Z) );
+   /* data = ? x y z */
+   sse_movss(cp->func, data, aos_get_internal_xmm( cp, IMM_ONES ) );
+   /* data = 1 x y z */
+   sse2_pshufd(cp->func, data, data, SHUF(Y,Z,W,X) );
+   /* data = x y z 1 */
+#endif
+}
+
+static void emit_load_R32G32( struct aos_compilation *cp, 
+			   struct x86_reg data,
+			   struct x86_reg src_ptr )
+{
+   sse_movups(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ) );
+   sse_movlps(cp->func, data, src_ptr);
+}
+
+
+static void emit_load_R32( struct aos_compilation *cp, 
+			   struct x86_reg data,
+			   struct x86_reg src_ptr )
+{
+   sse_movss(cp->func, data, src_ptr);
+   sse_orps(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ) );
+}
+
+
+static void emit_load_R8G8B8A8_UNORM( struct aos_compilation *cp,
+				       struct x86_reg data,
+				       struct x86_reg src_ptr )
+{
+   sse_movss(cp->func, data, src_ptr);
+   sse2_punpcklbw(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ));
+   sse2_punpcklbw(cp->func, data, aos_get_internal_xmm( cp, IMM_IDENTITY ));
+   sse2_cvtdq2ps(cp->func, data, data);
+   sse_mulps(cp->func, data, aos_get_internal(cp, IMM_INV_255));
+}
+
+
+
+/* Extended swizzles?  Maybe later.
+ */  
+static void emit_swizzle( struct aos_compilation *cp,
+			  struct x86_reg dest,
+			  struct x86_reg src,
+			  ubyte shuffle )
+{
+   sse_shufps(cp->func, dest, src, shuffle);
+}
+
+
+
+static boolean get_buffer_ptr( struct aos_compilation *cp,
+                               boolean linear,
+                               unsigned buf_idx,
+                               struct x86_reg elt,
+                               struct x86_reg ptr)
+{
+   struct x86_reg buf = x86_make_disp(aos_get_x86( cp, 0, X86_BUFFERS ), 
+                                      buf_idx * sizeof(struct aos_buffer));
+
+   struct x86_reg buf_stride = x86_make_disp(buf, 
+                                             Offset(struct aos_buffer, stride));
+   if (linear) {
+      struct x86_reg buf_ptr = x86_make_disp(buf, 
+                                             Offset(struct aos_buffer, ptr));
+
+
+      /* Calculate pointer to current attrib:
+       */
+      x86_mov(cp->func, ptr, buf_ptr);
+      x86_mov(cp->func, elt, buf_stride);
+      x86_add(cp->func, elt, ptr);
+      if (buf_idx == 0) sse_prefetchnta(cp->func, x86_make_disp(elt, 192));
+      x86_mov(cp->func, buf_ptr, elt);
+   }
+   else {
+      struct x86_reg buf_base_ptr = x86_make_disp(buf, 
+                                                  Offset(struct aos_buffer, base_ptr));
+
+
+      /* Calculate pointer to current attrib:
+       */
+      x86_mov(cp->func, ptr, buf_stride);
+      x86_imul(cp->func, ptr, elt);
+      x86_add(cp->func, ptr, buf_base_ptr);
+   }
+
+   cp->insn_counter++;
+
+   return TRUE;
+}
+
+
+static boolean load_input( struct aos_compilation *cp,
+                           unsigned idx,
+                           struct x86_reg bufptr )
+{
+   unsigned format = cp->vaos->base.key.element[idx].in.format;
+   unsigned offset = cp->vaos->base.key.element[idx].in.offset;
+   struct x86_reg dataXMM = aos_get_xmm_reg(cp);
+
+   /* Figure out source pointer address:
+    */
+   struct x86_reg src = x86_make_disp(bufptr, offset);
+
+   aos_adopt_xmm_reg( cp,
+                      dataXMM,
+                      TGSI_FILE_INPUT,
+                      idx,
+                      TRUE );
+
+   switch (format) {
+   case PIPE_FORMAT_R32_FLOAT:
+      emit_load_R32(cp, dataXMM, src);
+      break;
+   case PIPE_FORMAT_R32G32_FLOAT:
+      emit_load_R32G32(cp, dataXMM, src);
+      break;
+   case PIPE_FORMAT_R32G32B32_FLOAT:
+      emit_load_R32G32B32(cp, dataXMM, src);
+      break;
+   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+      emit_load_R32G32B32A32(cp, dataXMM, src);
+      break;
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      emit_load_R8G8B8A8_UNORM(cp, dataXMM, src);
+      emit_swizzle(cp, dataXMM, dataXMM, SHUF(Z,Y,X,W));
+      break;
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      emit_load_R8G8B8A8_UNORM(cp, dataXMM, src);
+      break;
+   default:
+      ERROR(cp, "unhandled input format");
+      return FALSE;
+   }
+
+   return TRUE;
+}
+
+static boolean load_inputs( struct aos_compilation *cp,
+                            unsigned buffer,
+                            struct x86_reg ptr )
+{
+   unsigned i;
+
+   for (i = 0; i < cp->vaos->base.key.nr_inputs; i++) {
+      if (cp->vaos->base.key.element[i].in.buffer == buffer) {
+
+         if (!load_input( cp, i, ptr ))
+            return FALSE;
+
+         cp->insn_counter++;
+      }
+   }
+   
+   return TRUE;
+}
+
+boolean aos_init_inputs( struct aos_compilation *cp, boolean linear )
+{
+   unsigned i;
+   for (i = 0; i < cp->vaos->nr_vb; i++) {
+      struct x86_reg buf = x86_make_disp(aos_get_x86( cp, 0, X86_BUFFERS ), 
+                                         i * sizeof(struct aos_buffer));
+
+      struct x86_reg buf_base_ptr = x86_make_disp(buf, 
+                                                  Offset(struct aos_buffer, base_ptr));
+
+      if (cp->vaos->base.key.const_vbuffers & (1<<i)) {
+         struct x86_reg ptr = cp->tmp_EAX;
+
+         x86_mov(cp->func, ptr, buf_base_ptr);
+
+         /* Load all inputs for this constant vertex buffer
+          */
+         load_inputs( cp, i, x86_deref(ptr) );
+         
+         /* Then just force them out to aos_machine.input[]
+          */
+         aos_spill_all( cp );
+
+      }
+      else if (linear) {
+
+         struct x86_reg elt = cp->idx_EBX;
+         struct x86_reg ptr = cp->tmp_EAX;
+
+         struct x86_reg buf_stride = x86_make_disp(buf, 
+                                                   Offset(struct aos_buffer, stride));
+
+         struct x86_reg buf_ptr = x86_make_disp(buf, 
+                                                Offset(struct aos_buffer, ptr));
+
+
+         /* Calculate pointer to current attrib:
+          */
+         x86_mov(cp->func, ptr, buf_stride);
+         x86_imul(cp->func, ptr, elt);
+         x86_add(cp->func, ptr, buf_base_ptr);
+
+
+         /* In the linear case, keep the buffer pointer instead of the
+          * index number.
+          */
+         if (cp->vaos->nr_vb == 1) 
+            x86_mov( cp->func, elt, ptr );
+         else
+            x86_mov( cp->func, buf_ptr, ptr );
+
+         cp->insn_counter++;
+      }
+   }
+
+   return TRUE;
+}
+
+boolean aos_fetch_inputs( struct aos_compilation *cp, boolean linear )
+{
+   unsigned j;
+
+   for (j = 0; j < cp->vaos->nr_vb; j++) {
+      if (cp->vaos->base.key.const_vbuffers & (1<<j)) {
+         /* just retreive pre-transformed input */
+      }
+      else if (linear && cp->vaos->nr_vb == 1) {
+         load_inputs( cp, 0, cp->idx_EBX );
+      }
+      else {
+         struct x86_reg elt = linear ? cp->idx_EBX : x86_deref(cp->idx_EBX);
+         struct x86_reg ptr = cp->tmp_EAX;
+
+         if (!get_buffer_ptr( cp, linear, j, elt, ptr ))
+            return FALSE;
+
+         if (!load_inputs( cp, j, ptr ))
+            return FALSE;
+      }
+   }
+
+   return TRUE;
+}
+
+boolean aos_incr_inputs( struct aos_compilation *cp, boolean linear )
+{
+   if (linear && cp->vaos->nr_vb == 1) {
+      struct x86_reg stride = x86_make_disp(aos_get_x86( cp, 0, X86_BUFFERS ), 
+                                            (0 * sizeof(struct aos_buffer) + 
+                                             Offset(struct aos_buffer, stride)));
+
+      x86_add(cp->func, cp->idx_EBX, stride);
+      sse_prefetchnta(cp->func, x86_make_disp(cp->idx_EBX, 192));
+   }
+   else if (linear) {
+      /* Nothing to do */
+   } 
+   else {
+      x86_lea(cp->func, cp->idx_EBX, x86_make_disp(cp->idx_EBX, 4));
+   }
+
+   return TRUE;
+}
+
+
+
+
+
+
+static void emit_store_R32G32B32A32( struct aos_compilation *cp, 			   
+				     struct x86_reg dst_ptr,
+				     struct x86_reg dataXMM )
+{
+   sse_movups(cp->func, dst_ptr, dataXMM);
+}
+
+static void emit_store_R32G32B32( struct aos_compilation *cp, 
+				  struct x86_reg dst_ptr,
+				  struct x86_reg dataXMM )
+{
+   sse_movlps(cp->func, dst_ptr, dataXMM);
+   sse_shufps(cp->func, dataXMM, dataXMM, SHUF(Z,Z,Z,Z) ); /* NOTE! destructive */
+   sse_movss(cp->func, x86_make_disp(dst_ptr,8), dataXMM);
+}
+
+static void emit_store_R32G32( struct aos_compilation *cp, 
+			       struct x86_reg dst_ptr,
+			       struct x86_reg dataXMM )
+{
+   sse_movlps(cp->func, dst_ptr, dataXMM);
+}
+
+static void emit_store_R32( struct aos_compilation *cp, 
+			    struct x86_reg dst_ptr,
+			    struct x86_reg dataXMM )
+{
+   sse_movss(cp->func, dst_ptr, dataXMM);
+}
+
+
+
+static void emit_store_R8G8B8A8_UNORM( struct aos_compilation *cp,
+				       struct x86_reg dst_ptr,
+				       struct x86_reg dataXMM )
+{
+   sse_mulps(cp->func, dataXMM, aos_get_internal(cp, IMM_255));
+   sse2_cvtps2dq(cp->func, dataXMM, dataXMM);
+   sse2_packssdw(cp->func, dataXMM, dataXMM);
+   sse2_packuswb(cp->func, dataXMM, dataXMM);
+   sse_movss(cp->func, dst_ptr, dataXMM);
+}
+
+
+
+
+
+static boolean emit_output( struct aos_compilation *cp,
+                            struct x86_reg ptr,
+                            struct x86_reg dataXMM, 
+                            unsigned format )
+{
+   switch (format) {
+   case EMIT_1F:
+   case EMIT_1F_PSIZE:
+      emit_store_R32(cp, ptr, dataXMM);
+      break;
+   case EMIT_2F:
+      emit_store_R32G32(cp, ptr, dataXMM);
+      break;
+   case EMIT_3F:
+      emit_store_R32G32B32(cp, ptr, dataXMM);
+      break;
+   case EMIT_4F:
+      emit_store_R32G32B32A32(cp, ptr, dataXMM);
+      break;
+   case EMIT_4UB:
+      if (1) {
+         emit_swizzle(cp, dataXMM, dataXMM, SHUF(Z,Y,X,W));
+         emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM);
+      }
+      else {
+         emit_store_R8G8B8A8_UNORM(cp, ptr, dataXMM);
+      }
+      break;
+   default:
+      ERROR(cp, "unhandled output format");
+      return FALSE;
+   }
+
+   return TRUE;
+}
+
+
+
+boolean aos_emit_outputs( struct aos_compilation *cp )
+{
+   unsigned i;
+   
+   for (i = 0; i < cp->vaos->base.key.nr_outputs; i++) {
+      unsigned format = cp->vaos->base.key.element[i].out.format;
+      unsigned offset = cp->vaos->base.key.element[i].out.offset;
+      unsigned vs_output = cp->vaos->base.key.element[i].out.vs_output;
+
+      struct x86_reg data;
+
+      if (format == EMIT_1F_PSIZE) {
+         data = aos_get_internal_xmm( cp, IMM_PSIZE );
+      }
+      else {
+         data = aos_get_shader_reg( cp, 
+                                    TGSI_FILE_OUTPUT,
+                                    vs_output );
+      }
+
+      if (data.file != file_XMM) {
+         struct x86_reg tmp = aos_get_xmm_reg( cp );
+         sse_movaps(cp->func, tmp, data);
+         data = tmp;
+      }
+      
+      if (!emit_output( cp, 
+                        x86_make_disp( cp->outbuf_ECX, offset ),
+                        data, 
+                        format ))
+         return FALSE;
+
+      aos_release_xmm_reg( cp, data.idx );
+
+      cp->insn_counter++;
+   }
+
+   return TRUE;
+}
+
+#endif
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_machine.c b/src/gallium/auxiliary/draw/draw_vs_aos_machine.c
new file mode 100644
index 00000000000..b358bd2df47
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs_aos_machine.c
@@ -0,0 +1,324 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "pipe/p_config.h"
+
+
+#include "pipe/p_shader_tokens.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_util.h"
+#include "tgsi/tgsi_exec.h"
+#include "draw_vs.h"
+#include "draw_vs_aos.h"
+#include "draw_vertex.h"
+
+#ifdef PIPE_ARCH_X86
+
+#include "rtasm/rtasm_x86sse.h"
+
+
+#define X87_CW_EXCEPTION_INV_OP       (1<<0)
+#define X87_CW_EXCEPTION_DENORM_OP    (1<<1)
+#define X87_CW_EXCEPTION_ZERO_DIVIDE  (1<<2)
+#define X87_CW_EXCEPTION_OVERFLOW     (1<<3)
+#define X87_CW_EXCEPTION_UNDERFLOW    (1<<4)
+#define X87_CW_EXCEPTION_PRECISION    (1<<5)
+#define X87_CW_PRECISION_SINGLE       (0<<8)
+#define X87_CW_PRECISION_RESERVED     (1<<8)
+#define X87_CW_PRECISION_DOUBLE       (2<<8)
+#define X87_CW_PRECISION_DOUBLE_EXT   (3<<8)
+#define X87_CW_PRECISION_MASK         (3<<8)
+#define X87_CW_ROUND_NEAREST          (0<<10)
+#define X87_CW_ROUND_DOWN             (1<<10)
+#define X87_CW_ROUND_UP               (2<<10)
+#define X87_CW_ROUND_ZERO             (3<<10)
+#define X87_CW_ROUND_MASK             (3<<10)
+#define X87_CW_INFINITY               (1<<12)
+
+
+void PIPE_CDECL aos_do_lit( struct aos_machine *machine,
+                            float *result,
+                            const float *in,
+                            unsigned count )
+{
+   if (in[0] > 0) 
+   {
+      if (in[1] <= 0.0) 
+      {
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = 1.0;
+         result[3] = 1.0F;
+      }
+      else
+      {
+         const float epsilon = 1.0F / 256.0F;    
+         float exponent = CLAMP(in[3], -(128.0F - epsilon), (128.0F - epsilon));
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = powf(in[1], exponent);
+         result[3] = 1.0;
+      }
+   }
+   else 
+   {
+      result[0] = 1.0F;
+      result[1] = 0.0;
+      result[2] = 0.0;
+      result[3] = 1.0F;
+   }
+}
+
+
+static void PIPE_CDECL do_lit_lut( struct aos_machine *machine,
+                                   float *result,
+                                   const float *in,
+                                   unsigned count )
+{
+   if (in[0] > 0) 
+   {
+      if (in[1] <= 0.0) 
+      {
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = 1.0;
+         result[3] = 1.0F;
+         return;
+      }
+      
+      if (machine->lit_info[count].shine_tab->exponent != in[3]) {
+         machine->lit_info[count].func = aos_do_lit;
+         goto no_luck;
+      }
+
+      if (in[1] <= 1.0)
+      {
+         const float *tab = machine->lit_info[count].shine_tab->values;
+         float f = in[1] * 256;
+         int k = (int)f;
+         float frac = f - (float)k;
+         
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = tab[k] + frac*(tab[k+1]-tab[k]);
+         result[3] = 1.0;
+         return;
+      }
+      
+   no_luck:
+      {
+         const float epsilon = 1.0F / 256.0F;    
+         float exponent = CLAMP(in[3], -(128.0F - epsilon), (128.0F - epsilon));
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = powf(in[1], exponent);
+         result[3] = 1.0;
+      }
+   }
+   else 
+   {
+      result[0] = 1.0F;
+      result[1] = 0.0;
+      result[2] = 0.0;
+      result[3] = 1.0F;
+   }
+}
+
+
+static void do_populate_lut( struct shine_tab *tab,
+                             float unclamped_exponent )
+{
+   const float epsilon = 1.0F / 256.0F;    
+   float exponent = CLAMP(unclamped_exponent, -(128.0F - epsilon), (128.0F - epsilon));
+   unsigned i;
+
+   tab->exponent = unclamped_exponent; /* for later comparison */
+   
+   tab->values[0] = 0;
+   if (exponent == 0) {
+      for (i = 1; i < 258; i++) {
+         tab->values[i] = 1.0;
+      }      
+   }
+   else {
+      for (i = 1; i < 258; i++) {
+         tab->values[i] = powf((float)i * epsilon, exponent);
+      }
+   }
+}
+
+
+
+
+static void PIPE_CDECL populate_lut( struct aos_machine *machine,
+                                     float *result,
+                                     const float *in,
+                                     unsigned count )
+{
+   unsigned i, tab;
+
+   /* Search for an existing table for this value.  Note that without
+    * static analysis we don't really know if in[3] will be constant,
+    * but it usually is...
+    */
+   for (tab = 0; tab < 4; tab++) {
+      if (machine->shine_tab[tab].exponent == in[3]) {
+         goto found;
+      }
+   }
+
+   for (tab = 0, i = 1; i < 4; i++) {
+      if (machine->shine_tab[i].last_used < machine->shine_tab[tab].last_used)
+         tab = i;
+   }
+
+   if (machine->shine_tab[tab].last_used == machine->now) {
+      /* No unused tables (this is not a ffvertex program...).  Just
+       * call pow each time:
+       */
+      machine->lit_info[count].func = aos_do_lit;
+      machine->lit_info[count].func( machine, result, in, count );
+      return;
+   }
+   else {
+      do_populate_lut( &machine->shine_tab[tab], in[3] );
+   }
+
+ found:
+   machine->shine_tab[tab].last_used = machine->now;
+   machine->lit_info[count].shine_tab = &machine->shine_tab[tab];
+   machine->lit_info[count].func = do_lit_lut;
+   machine->lit_info[count].func( machine, result, in, count );
+}
+
+
+void draw_vs_aos_machine_constants( struct aos_machine *machine,
+                                    const float (*constants)[4] )
+{
+   machine->constants = constants;
+
+   {
+      unsigned i;
+      for (i = 0; i < MAX_LIT_INFO; i++) {
+         machine->lit_info[i].func = populate_lut;
+         machine->now++;
+      }
+   }
+}
+
+
+void draw_vs_aos_machine_viewport( struct aos_machine *machine,
+                                   const struct pipe_viewport_state *viewport )
+{
+   memcpy(machine->scale, viewport->scale, 4 * sizeof(float));
+   memcpy(machine->translate, viewport->translate, 4 * sizeof(float));
+}
+
+
+
+void draw_vs_aos_machine_destroy( struct aos_machine *machine )
+{
+   align_free(machine);
+}
+
+struct aos_machine *draw_vs_aos_machine( void )
+{
+   struct aos_machine *machine;
+   unsigned i;
+   float inv = 1.0f/255.0f;
+   float f255 = 255.0f;
+
+   machine = align_malloc(sizeof(struct aos_machine), 16);
+   if (!machine)
+      return NULL;
+
+   memset(machine, 0, sizeof(*machine));
+
+   ASSIGN_4V(machine->internal[IMM_SWZ],       1.0f,  -1.0f,  0.0f, 1.0f);
+   *(unsigned *)&machine->internal[IMM_SWZ][3] = 0xffffffff;
+
+   ASSIGN_4V(machine->internal[IMM_ONES],      1.0f,  1.0f,  1.0f,  1.0f);
+   ASSIGN_4V(machine->internal[IMM_NEGS],     -1.0f, -1.0f, -1.0f, -1.0f);
+   ASSIGN_4V(machine->internal[IMM_IDENTITY],  0.0f,  0.0f,  0.0f,  1.0f);
+   ASSIGN_4V(machine->internal[IMM_INV_255],   inv,   inv,   inv,   inv);
+   ASSIGN_4V(machine->internal[IMM_255],       f255,  f255,  f255,  f255);
+   ASSIGN_4V(machine->internal[IMM_RSQ],       -.5f,  1.5f,  0.0f,  0.0f);
+
+
+   machine->fpu_rnd_nearest = (X87_CW_EXCEPTION_INV_OP |
+                               X87_CW_EXCEPTION_DENORM_OP |
+                               X87_CW_EXCEPTION_ZERO_DIVIDE |
+                               X87_CW_EXCEPTION_OVERFLOW |
+                               X87_CW_EXCEPTION_UNDERFLOW |
+                               X87_CW_EXCEPTION_PRECISION |
+                               (1<<6) |
+                               X87_CW_ROUND_NEAREST |
+                               X87_CW_PRECISION_DOUBLE_EXT);
+
+   assert(machine->fpu_rnd_nearest == 0x37f);
+                               
+   machine->fpu_rnd_neg_inf = (X87_CW_EXCEPTION_INV_OP |
+                               X87_CW_EXCEPTION_DENORM_OP |
+                               X87_CW_EXCEPTION_ZERO_DIVIDE |
+                               X87_CW_EXCEPTION_OVERFLOW |
+                               X87_CW_EXCEPTION_UNDERFLOW |
+                               X87_CW_EXCEPTION_PRECISION |
+                               (1<<6) |
+                               X87_CW_ROUND_DOWN |
+                               X87_CW_PRECISION_DOUBLE_EXT);
+
+   for (i = 0; i < MAX_SHINE_TAB; i++)
+      do_populate_lut( &machine->shine_tab[i], 1.0f );
+
+   return machine;
+}
+
+#else
+
+void draw_vs_aos_machine_viewport( struct aos_machine *machine,
+                                   const struct pipe_viewport_state *viewport )
+{
+}
+
+void draw_vs_aos_machine_constants( struct aos_machine *machine,
+                                    const float (*constants)[4] )
+{
+}
+
+void draw_vs_aos_machine_destroy( struct aos_machine *machine )
+{
+}
+
+struct aos_machine *draw_vs_aos_machine( void )
+{
+   return NULL;
+}
+#endif
+
diff --git a/src/gallium/auxiliary/draw/draw_vs_exec.c b/src/gallium/auxiliary/draw/draw_vs_exec.c
new file mode 100644
index 00000000000..b3200df8112
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs_exec.c
@@ -0,0 +1,200 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  *   Brian Paul
+  */
+
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "draw_private.h"
+#include "draw_context.h"
+#include "draw_vs.h"
+
+#include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_scan.h"
+
+
+struct exec_vertex_shader {
+   struct draw_vertex_shader base;
+   struct tgsi_exec_machine *machine;
+};
+
+static struct exec_vertex_shader *exec_vertex_shader( struct draw_vertex_shader *vs )
+{
+   return (struct exec_vertex_shader *)vs;
+}
+
+
+/* Not required for run_linear.
+ */
+static void
+vs_exec_prepare( struct draw_vertex_shader *shader,
+		 struct draw_context *draw )
+{
+   struct exec_vertex_shader *evs = exec_vertex_shader(shader);
+
+   /* Specify the vertex program to interpret/execute.
+    * Avoid rebinding when possible.
+    */
+   if (evs->machine->Tokens != shader->state.tokens) {
+      tgsi_exec_machine_bind_shader(evs->machine,
+                                    shader->state.tokens,
+                                    draw->vs.num_samplers,
+                                    draw->vs.samplers);
+   }
+}
+
+
+
+
+/* Simplified vertex shader interface for the pt paths.  Given the
+ * complexity of code-generating all the above operations together,
+ * it's time to try doing all the other stuff separately.
+ */
+static void
+vs_exec_run_linear( struct draw_vertex_shader *shader,
+		    const float (*input)[4],
+		    float (*output)[4],
+		    const float (*constants)[4],
+		    unsigned count,
+		    unsigned input_stride,
+		    unsigned output_stride )
+{
+   struct exec_vertex_shader *evs = exec_vertex_shader(shader);
+   struct tgsi_exec_machine *machine = evs->machine;
+   unsigned int i, j;
+   unsigned slot;
+
+   machine->Consts = constants;
+
+   for (i = 0; i < count; i += MAX_TGSI_VERTICES) {
+      unsigned int max_vertices = MIN2(MAX_TGSI_VERTICES, count - i);
+
+      /* Swizzle inputs.  
+       */
+      for (j = 0; j < max_vertices; j++) {
+#if 0
+         debug_printf("%d) Input vert:\n", i + j);
+         for (slot = 0; slot < shader->info.num_inputs; slot++) {
+            debug_printf("\t%d: %f %f %f %f\n", slot,
+			 input[slot][0],
+			 input[slot][1],
+			 input[slot][2],
+			 input[slot][3]);
+         }
+#endif
+
+         for (slot = 0; slot < shader->info.num_inputs; slot++) {
+            machine->Inputs[slot].xyzw[0].f[j] = input[slot][0];
+            machine->Inputs[slot].xyzw[1].f[j] = input[slot][1];
+            machine->Inputs[slot].xyzw[2].f[j] = input[slot][2];
+            machine->Inputs[slot].xyzw[3].f[j] = input[slot][3];
+         }
+
+	 input = (const float (*)[4])((const char *)input + input_stride);
+      } 
+
+      tgsi_set_exec_mask(machine,
+                         1,
+                         max_vertices > 1,
+                         max_vertices > 2,
+                         max_vertices > 3);
+
+      /* run interpreter */
+      tgsi_exec_machine_run( machine );
+
+      /* Unswizzle all output results.  
+       */
+      for (j = 0; j < max_vertices; j++) {
+         for (slot = 0; slot < shader->info.num_outputs; slot++) {
+            output[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
+            output[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
+            output[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
+            output[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
+
+         }
+
+#if 0
+	 debug_printf("%d) Post xform vert:\n", i + j);
+	 for (slot = 0; slot < shader->info.num_outputs; slot++) {
+	    debug_printf("\t%d: %f %f %f %f\n", slot,
+			 output[slot][0],
+			 output[slot][1],
+			 output[slot][2],
+			 output[slot][3]);
+         }
+#endif
+
+	 output = (float (*)[4])((char *)output + output_stride);
+      } 
+
+   }
+}
+
+
+
+
+static void
+vs_exec_delete( struct draw_vertex_shader *dvs )
+{
+   FREE((void*) dvs->state.tokens);
+   FREE( dvs );
+}
+
+
+struct draw_vertex_shader *
+draw_create_vs_exec(struct draw_context *draw,
+		    const struct pipe_shader_state *state)
+{
+   struct exec_vertex_shader *vs = CALLOC_STRUCT( exec_vertex_shader );
+
+   if (vs == NULL) 
+      return NULL;
+
+   /* we make a private copy of the tokens */
+   vs->base.state.tokens = tgsi_dup_tokens(state->tokens);
+   if (!vs->base.state.tokens) {
+      FREE(vs);
+      return NULL;
+   }
+
+   tgsi_scan_shader(state->tokens, &vs->base.info);
+
+   vs->base.draw = draw;
+   vs->base.prepare = vs_exec_prepare;
+   vs->base.run_linear = vs_exec_run_linear;
+   vs->base.delete = vs_exec_delete;
+   vs->base.create_varient = draw_vs_varient_generic;
+   vs->machine = &draw->vs.machine;
+
+   return &vs->base;
+}
diff --git a/src/gallium/auxiliary/draw/draw_vs_llvm.c b/src/gallium/auxiliary/draw/draw_vs_llvm.c
new file mode 100644
index 00000000000..727977bc3af
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs_llvm.c
@@ -0,0 +1,161 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin
+  *   Keith Whitwell <[email protected]>
+  *   Brian Paul
+  */
+
+#include "util/u_memory.h"
+#include "pipe/p_shader_tokens.h"
+#include "draw_private.h"
+#include "draw_context.h"
+#include "draw_vs.h"
+
+#include "tgsi/tgsi_parse.h"
+
+#ifdef MESA_LLVM
+
+#include "gallivm/gallivm.h"
+
+struct draw_llvm_vertex_shader {
+   struct draw_vertex_shader base;
+   struct gallivm_prog *llvm_prog;
+   struct tgsi_exec_machine *machine;
+};
+
+
+static void
+vs_llvm_prepare( struct draw_vertex_shader *base,
+		 struct draw_context *draw )
+{
+}
+
+
+
+
+static void
+vs_llvm_run_linear( struct draw_vertex_shader *base,
+		   const float (*input)[4],
+		   float (*output)[4],
+		   const float (*constants)[4],
+		   unsigned count,
+		   unsigned input_stride,
+		   unsigned output_stride )
+{
+   struct draw_llvm_vertex_shader *shader =
+      (struct draw_llvm_vertex_shader *)base;
+
+   gallivm_cpu_vs_exec(shader->llvm_prog, shader->machine,
+                       input, base->info.num_inputs, output, base->info.num_outputs,
+                       constants, count, input_stride, output_stride);
+}
+
+
+
+static void
+vs_llvm_delete( struct draw_vertex_shader *base )
+{
+   struct draw_llvm_vertex_shader *shader =
+      (struct draw_llvm_vertex_shader *)base;
+
+   /* Do something to free compiled shader:
+    */
+
+   FREE( (void*) shader->base.state.tokens );
+   FREE( shader );
+}
+
+
+
+
+struct draw_vertex_shader *
+draw_create_vs_llvm(struct draw_context *draw,
+		    const struct pipe_shader_state *templ)
+{
+   struct draw_llvm_vertex_shader *vs;
+
+   vs = CALLOC_STRUCT( draw_llvm_vertex_shader );
+   if (vs == NULL)
+      return NULL;
+
+   /* we make a private copy of the tokens */
+   vs->base.state.tokens = tgsi_dup_tokens(templ->tokens);
+   if (!vs->base.state.tokens) {
+      FREE(vs);
+      return NULL;
+   }
+
+   tgsi_scan_shader(vs->base.state.tokens, &vs->base.info);
+
+   vs->base.draw = draw;
+   vs->base.prepare = vs_llvm_prepare;
+   vs->base.create_varient = draw_vs_varient_generic;
+   vs->base.run_linear = vs_llvm_run_linear;
+   vs->base.delete = vs_llvm_delete;
+   vs->machine = &draw->vs.machine;
+
+   {
+      struct gallivm_ir *ir = gallivm_ir_new(GALLIVM_VS);
+      gallivm_ir_set_layout(ir, GALLIVM_SOA);
+      gallivm_ir_set_components(ir, 4);
+      gallivm_ir_fill_from_tgsi(ir, vs->base.state.tokens);
+      vs->llvm_prog = gallivm_ir_compile(ir);
+      gallivm_ir_delete(ir);
+   }
+
+   draw->vs.engine = gallivm_global_cpu_engine();
+
+   /* XXX: Why are there two versions of this?  Shouldn't creating the
+    *      engine be a separate operation to compiling a shader?
+    */
+   if (!draw->vs.engine) {
+      draw->vs.engine = gallivm_cpu_engine_create(vs->llvm_prog);
+   }
+   else {
+      gallivm_cpu_jit_compile(draw->vs.engine, vs->llvm_prog);
+   }
+
+   return &vs->base;
+}
+
+
+
+
+
+#else
+
+struct draw_vertex_shader *
+draw_create_vs_llvm(struct draw_context *draw,
+                          const struct pipe_shader_state *shader)
+{
+   return NULL;
+}
+
+#endif
diff --git a/src/gallium/auxiliary/draw/draw_vs_ppc.c b/src/gallium/auxiliary/draw/draw_vs_ppc.c
new file mode 100644
index 00000000000..d35db57d571
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs_ppc.c
@@ -0,0 +1,244 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  *   Brian Paul
+  */
+
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "pipe/p_config.h"
+
+#include "draw_vs.h"
+
+#if defined(PIPE_ARCH_PPC)
+
+#include "pipe/p_shader_tokens.h"
+
+#include "draw_private.h"
+#include "draw_context.h"
+
+#include "rtasm/rtasm_cpu.h"
+#include "rtasm/rtasm_ppc.h"
+#include "tgsi/tgsi_ppc.h"
+#include "tgsi/tgsi_parse.h"
+
+
+
+typedef void (PIPE_CDECL *codegen_function) (float (*inputs)[4][4],
+                                             float (*outputs)[4][4],
+                                             float (*temps)[4][4],
+                                             float (*immeds)[4],
+                                             float (*consts)[4],
+                                             const float *builtins);
+
+
+struct draw_ppc_vertex_shader {
+   struct draw_vertex_shader base;
+   struct ppc_function ppc_program;
+
+   codegen_function func;
+};
+
+
+static void
+vs_ppc_prepare( struct draw_vertex_shader *base,
+		struct draw_context *draw )
+{
+   /* nothing */
+}
+
+
+/**
+ * Simplified vertex shader interface for the pt paths.  Given the
+ * complexity of code-generating all the above operations together,
+ * it's time to try doing all the other stuff separately.
+ */
+static void
+vs_ppc_run_linear( struct draw_vertex_shader *base,
+		   const float (*input)[4],
+		   float (*output)[4],
+		   const float (*constants)[4],
+		   unsigned count,
+		   unsigned input_stride,
+		   unsigned output_stride )
+{
+   struct draw_ppc_vertex_shader *shader = (struct draw_ppc_vertex_shader *)base;
+   unsigned int i;
+
+#define MAX_VERTICES 4
+
+   /* loop over verts */
+   for (i = 0; i < count; i += MAX_VERTICES) {
+      const uint max_vertices = MIN2(MAX_VERTICES, count - i);
+      float inputs_soa[PIPE_MAX_SHADER_INPUTS][4][4] ALIGN16_ATTRIB;
+      float outputs_soa[PIPE_MAX_SHADER_OUTPUTS][4][4] ALIGN16_ATTRIB;
+      float temps_soa[TGSI_EXEC_NUM_TEMPS][4][4] ALIGN16_ATTRIB;
+      uint attr;
+
+      /* convert (up to) four input verts to SoA format */
+      for (attr = 0; attr < base->info.num_inputs; attr++) {
+         const float *vIn = (const float *) input;
+         uint vert;
+         for (vert = 0; vert < max_vertices; vert++) {
+#if 0
+            if (attr==0)
+               printf("Input v%d a%d: %f %f %f %f\n",
+                      vert, attr, vIn[0], vIn[1], vIn[2], vIn[3]);
+#endif
+            inputs_soa[attr][0][vert] = vIn[attr * 4 + 0];
+            inputs_soa[attr][1][vert] = vIn[attr * 4 + 1];
+            inputs_soa[attr][2][vert] = vIn[attr * 4 + 2];
+            inputs_soa[attr][3][vert] = vIn[attr * 4 + 3];
+            vIn += input_stride / 4;
+         }
+      }
+
+      /* run compiled shader
+       */
+      shader->func(inputs_soa, outputs_soa, temps_soa,
+		   (float (*)[4]) shader->base.immediates,
+		   (float (*)[4]) constants,
+                   ppc_builtin_constants);
+
+      /* convert (up to) four output verts from SoA back to AoS format */
+      for (attr = 0; attr < base->info.num_outputs; attr++) {
+         float *vOut = (float *) output;
+         uint vert;
+         for (vert = 0; vert < max_vertices; vert++) {
+            vOut[attr * 4 + 0] = outputs_soa[attr][0][vert];
+            vOut[attr * 4 + 1] = outputs_soa[attr][1][vert];
+            vOut[attr * 4 + 2] = outputs_soa[attr][2][vert];
+            vOut[attr * 4 + 3] = outputs_soa[attr][3][vert];
+#if 0
+            if (attr==0)
+               printf("Output v%d a%d: %f %f %f %f\n",
+                      vert, attr, vOut[0], vOut[1], vOut[2], vOut[3]);
+#endif
+            vOut += output_stride / 4;
+         }
+      }
+
+      /* advance to next group of four input/output verts */
+      input = (const float (*)[4])((const char *)input + input_stride * max_vertices);
+      output = (float (*)[4])((char *)output + output_stride * max_vertices);
+   }
+}
+
+
+static void
+vs_ppc_delete( struct draw_vertex_shader *base )
+{
+   struct draw_ppc_vertex_shader *shader = (struct draw_ppc_vertex_shader *)base;
+   
+   ppc_release_func( &shader->ppc_program );
+
+   align_free( (void *) shader->base.immediates );
+
+   FREE( (void*) shader->base.state.tokens );
+   FREE( shader );
+}
+
+
+struct draw_vertex_shader *
+draw_create_vs_ppc(struct draw_context *draw,
+                   const struct pipe_shader_state *templ)
+{
+   struct draw_ppc_vertex_shader *vs;
+
+   vs = CALLOC_STRUCT( draw_ppc_vertex_shader );
+   if (vs == NULL) 
+      return NULL;
+
+   /* we make a private copy of the tokens */
+   vs->base.state.tokens = tgsi_dup_tokens(templ->tokens);
+   if (!vs->base.state.tokens)
+      goto fail;
+
+   tgsi_scan_shader(templ->tokens, &vs->base.info);
+
+   vs->base.draw = draw;
+#if 0
+   if (1)
+      vs->base.create_varient = draw_vs_varient_aos_ppc;
+   else
+#endif
+      vs->base.create_varient = draw_vs_varient_generic;
+   vs->base.prepare = vs_ppc_prepare;
+   vs->base.run_linear = vs_ppc_run_linear;
+   vs->base.delete = vs_ppc_delete;
+   
+   vs->base.immediates = align_malloc(TGSI_EXEC_NUM_IMMEDIATES * 4 *
+                                      sizeof(float), 16);
+
+   ppc_init_func( &vs->ppc_program );
+
+#if 0
+   ppc_print_code(&vs->ppc_program, TRUE);
+   ppc_indent(&vs->ppc_program, 8);
+#endif
+
+   if (!tgsi_emit_ppc( (struct tgsi_token *) vs->base.state.tokens,
+			&vs->ppc_program, 
+                       (float (*)[4]) vs->base.immediates, 
+                        TRUE )) 
+      goto fail;
+      
+   vs->func = (codegen_function) ppc_get_func( &vs->ppc_program );
+   if (!vs->func) {
+      goto fail;
+   }
+   
+   return &vs->base;
+
+fail:
+   /*
+   debug_error("tgsi_emit_ppc() failed, falling back to interpreter\n");
+   */
+
+   ppc_release_func( &vs->ppc_program );
+   
+   FREE(vs);
+   return NULL;
+}
+
+
+
+#else /* PIPE_ARCH_PPC */
+
+
+struct draw_vertex_shader *
+draw_create_vs_ppc( struct draw_context *draw,
+		    const struct pipe_shader_state *templ )
+{
+   return (void *) 0;
+}
+
+
+#endif /* PIPE_ARCH_PPC */
diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c
new file mode 100644
index 00000000000..77ba5152f9f
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs_sse.c
@@ -0,0 +1,226 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  *   Brian Paul
+  */
+
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "pipe/p_config.h"
+
+#include "draw_vs.h"
+
+#if defined(PIPE_ARCH_X86)
+
+#include "pipe/p_shader_tokens.h"
+
+#include "draw_private.h"
+#include "draw_context.h"
+
+#include "rtasm/rtasm_cpu.h"
+#include "rtasm/rtasm_x86sse.h"
+#include "tgsi/tgsi_sse2.h"
+#include "tgsi/tgsi_parse.h"
+
+#define SSE_MAX_VERTICES 4
+
+typedef void (PIPE_CDECL *codegen_function) (
+   const struct tgsi_exec_vector *input, /* 1 */
+   struct tgsi_exec_vector *output, /* 2 */
+   float (*constant)[4],        /* 3 */
+   struct tgsi_exec_vector *temporary, /* 4 */
+   float (*immediates)[4],      /* 5 */
+   const float (*aos_input)[4], /* 6 */
+   uint num_inputs,             /* 7 */
+   uint input_stride,           /* 8 */
+   float (*aos_output)[4],      /* 9 */
+   uint num_outputs,            /* 10 */
+   uint output_stride );        /* 11 */
+
+struct draw_sse_vertex_shader {
+   struct draw_vertex_shader base;
+   struct x86_function sse2_program;
+
+   codegen_function func;
+   
+   struct tgsi_exec_machine *machine;
+};
+
+
+static void
+vs_sse_prepare( struct draw_vertex_shader *base,
+		struct draw_context *draw )
+{
+}
+
+
+
+/* Simplified vertex shader interface for the pt paths.  Given the
+ * complexity of code-generating all the above operations together,
+ * it's time to try doing all the other stuff separately.
+ */
+static void
+vs_sse_run_linear( struct draw_vertex_shader *base,
+		   const float (*input)[4],
+		   float (*output)[4],
+		   const float (*constants)[4],
+		   unsigned count,
+		   unsigned input_stride,
+		   unsigned output_stride )
+{
+   struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base;
+   struct tgsi_exec_machine *machine = shader->machine;
+   unsigned int i;
+
+   /* By default, execute all channels.  XXX move this inside the loop
+    * below when we support shader conditionals/loops.
+    */
+   tgsi_set_exec_mask(machine, 1, 1, 1, 1);
+
+   for (i = 0; i < count; i += MAX_TGSI_VERTICES) {
+      unsigned int max_vertices = MIN2(MAX_TGSI_VERTICES, count - i);
+
+      if (max_vertices < 4) {
+         /* disable the unused execution channels */
+         tgsi_set_exec_mask(machine,
+                            1,
+                            max_vertices > 1,
+                            max_vertices > 2,
+                            0);
+      }
+
+      /* run compiled shader
+       */
+      shader->func(machine->Inputs,
+		   machine->Outputs,
+		   (float (*)[4])constants,
+		   machine->Temps,
+		   (float (*)[4])shader->base.immediates,
+                   input,
+                   base->info.num_inputs,
+                   input_stride,
+                   output,
+                   base->info.num_outputs,
+                   output_stride );
+
+      input = (const float (*)[4])((const char *)input + input_stride * max_vertices);
+      output = (float (*)[4])((char *)output + output_stride * max_vertices);
+   }
+}
+
+
+
+
+static void
+vs_sse_delete( struct draw_vertex_shader *base )
+{
+   struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base;
+   
+   x86_release_func( &shader->sse2_program );
+
+   align_free( (void *) shader->base.immediates );
+
+   FREE( (void*) shader->base.state.tokens );
+   FREE( shader );
+}
+
+
+struct draw_vertex_shader *
+draw_create_vs_sse(struct draw_context *draw,
+                          const struct pipe_shader_state *templ)
+{
+   struct draw_sse_vertex_shader *vs;
+
+   if (!rtasm_cpu_has_sse2())
+      return NULL;
+
+   vs = CALLOC_STRUCT( draw_sse_vertex_shader );
+   if (vs == NULL) 
+      return NULL;
+
+   /* we make a private copy of the tokens */
+   vs->base.state.tokens = tgsi_dup_tokens(templ->tokens);
+   if (!vs->base.state.tokens)
+      goto fail;
+
+   tgsi_scan_shader(templ->tokens, &vs->base.info);
+
+   vs->base.draw = draw;
+   if (1)
+      vs->base.create_varient = draw_vs_varient_aos_sse;
+   else
+      vs->base.create_varient = draw_vs_varient_generic;
+   vs->base.prepare = vs_sse_prepare;
+   vs->base.run_linear = vs_sse_run_linear;
+   vs->base.delete = vs_sse_delete;
+   
+   vs->base.immediates = align_malloc(TGSI_EXEC_NUM_IMMEDIATES * 4 *
+                                      sizeof(float), 16);
+
+   vs->machine = &draw->vs.machine;
+   
+   x86_init_func( &vs->sse2_program );
+
+   if (!tgsi_emit_sse2( (struct tgsi_token *) vs->base.state.tokens,
+			&vs->sse2_program, 
+                        (float (*)[4])vs->base.immediates, 
+                        TRUE )) 
+      goto fail;
+      
+   vs->func = (codegen_function) x86_get_func( &vs->sse2_program );
+   if (!vs->func) {
+      goto fail;
+   }
+   
+   return &vs->base;
+
+fail:
+   debug_error("tgsi_emit_sse2() failed, falling back to interpreter\n");
+
+   x86_release_func( &vs->sse2_program );
+   
+   FREE(vs);
+   return NULL;
+}
+
+
+
+#else
+
+struct draw_vertex_shader *
+draw_create_vs_sse( struct draw_context *draw,
+		    const struct pipe_shader_state *templ )
+{
+   return (void *) 0;
+}
+
+
+#endif
+
diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c
new file mode 100644
index 00000000000..7ee567d4789
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs_varient.c
@@ -0,0 +1,322 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  */
+
+#include "util/u_memory.h"
+#include "util/u_math.h"
+#include "draw/draw_context.h"
+#include "draw/draw_private.h"
+#include "draw/draw_vbuf.h"
+#include "draw/draw_vertex.h"
+#include "draw/draw_vs.h"
+#include "translate/translate.h"
+#include "translate/translate_cache.h"
+
+/* A first pass at incorporating vertex fetch/emit functionality into 
+ */
+struct draw_vs_varient_generic {
+   struct draw_vs_varient base;
+
+   struct draw_vertex_shader *shader;
+   struct draw_context *draw;
+   
+   /* Basic plan is to run these two translate functions before/after
+    * the vertex shader's existing run_linear() routine to simulate
+    * the inclusion of this functionality into the shader...  
+    * 
+    * Next will look at actually including it.
+    */
+   struct translate *fetch;
+   struct translate *emit;
+
+   unsigned temp_vertex_stride;
+};
+
+
+
+
+
+static void vsvg_set_buffer( struct draw_vs_varient *varient,
+                             unsigned buffer,
+                             const void *ptr,
+                             unsigned stride )
+{
+   struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
+
+   vsvg->fetch->set_buffer(vsvg->fetch, 
+                           buffer, 
+                           ptr, 
+                           stride);
+}
+
+
+/* Mainly for debug at this stage:
+ */
+static void do_rhw_viewport( struct draw_vs_varient_generic *vsvg,
+                             unsigned count,
+                             void *output_buffer )
+{
+   char *ptr = (char *)output_buffer;
+   const float *scale = vsvg->base.vs->draw->viewport.scale;
+   const float *trans = vsvg->base.vs->draw->viewport.translate;
+   unsigned stride = vsvg->temp_vertex_stride;
+   unsigned j;
+
+   ptr += vsvg->base.vs->position_output * 4 * sizeof(float);
+
+   for (j = 0; j < count; j++, ptr += stride) {
+      float *data = (float *)ptr;
+      float w = 1.0f / data[3];
+
+      data[0] = data[0] * w * scale[0] + trans[0];
+      data[1] = data[1] * w * scale[1] + trans[1];
+      data[2] = data[2] * w * scale[2] + trans[2];
+      data[3] = w;
+   }
+}
+
+static void do_viewport( struct draw_vs_varient_generic *vsvg,
+                         unsigned count,
+                         void *output_buffer )
+{
+   char *ptr = (char *)output_buffer;
+   const float *scale = vsvg->base.vs->draw->viewport.scale;
+   const float *trans = vsvg->base.vs->draw->viewport.translate;
+   unsigned stride = vsvg->temp_vertex_stride;
+   unsigned j;
+
+   ptr += vsvg->base.vs->position_output * 4 * sizeof(float);
+
+   for (j = 0; j < count; j++, ptr += stride) {
+      float *data = (float *)ptr;
+
+      data[0] = data[0] * scale[0] + trans[0];
+      data[1] = data[1] * scale[1] + trans[1];
+      data[2] = data[2] * scale[2] + trans[2];
+   }
+}
+                         
+
+static void PIPE_CDECL vsvg_run_elts( struct draw_vs_varient *varient,
+                                      const unsigned *elts,
+                                      unsigned count,
+                                      void *output_buffer)
+{
+   struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
+   unsigned temp_vertex_stride = vsvg->temp_vertex_stride;
+   void *temp_buffer = MALLOC( align(count,4) * temp_vertex_stride );
+   
+   if (0) debug_printf("%s %d \n", __FUNCTION__,  count);
+			
+   /* Want to do this in small batches for cache locality?
+    */
+   
+   vsvg->fetch->run_elts( vsvg->fetch, 
+                          elts,
+                          count,
+                          temp_buffer );
+
+   vsvg->base.vs->run_linear( vsvg->base.vs, 
+                              temp_buffer,
+                              temp_buffer,
+                              (const float (*)[4])vsvg->base.vs->draw->pt.user.constants,
+                              count,
+                              temp_vertex_stride, 
+                              temp_vertex_stride);
+
+
+   if (vsvg->base.key.clip) {
+      /* not really handling clipping, just do the rhw so we can
+       * see the results...
+       */
+      do_rhw_viewport( vsvg,
+                       count,
+                       temp_buffer );
+   }
+   else if (vsvg->base.key.viewport) {
+      do_viewport( vsvg,
+                   count,
+                   temp_buffer );
+   }
+
+
+   vsvg->emit->set_buffer( vsvg->emit,
+                           0, 
+                           temp_buffer,
+                           temp_vertex_stride );
+
+   vsvg->emit->set_buffer( vsvg->emit, 
+                           1,
+                           &vsvg->draw->rasterizer->point_size,
+                           0);
+
+   vsvg->emit->run( vsvg->emit,
+                    0, count,
+                    output_buffer );
+
+   FREE(temp_buffer);
+}
+
+
+static void PIPE_CDECL vsvg_run_linear( struct draw_vs_varient *varient,
+                                        unsigned start,
+                                        unsigned count,
+                                        void *output_buffer )
+{
+   struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
+   unsigned temp_vertex_stride = vsvg->temp_vertex_stride;
+   void *temp_buffer = MALLOC( align(count,4) * temp_vertex_stride );
+	
+   if (0) debug_printf("%s %d %d (sz %d, %d)\n", __FUNCTION__, start, count,
+                       vsvg->base.key.output_stride,
+                       temp_vertex_stride);
+
+   vsvg->fetch->run( vsvg->fetch, 
+                     start,
+                     count,
+                     temp_buffer );
+
+   vsvg->base.vs->run_linear( vsvg->base.vs, 
+                              temp_buffer,
+                              temp_buffer,
+                              (const float (*)[4])vsvg->base.vs->draw->pt.user.constants,
+                              count,
+                              temp_vertex_stride, 
+                              temp_vertex_stride);
+
+   if (vsvg->base.key.clip) {
+      /* not really handling clipping, just do the rhw so we can
+       * see the results...
+       */
+      do_rhw_viewport( vsvg,
+                       count,
+                       temp_buffer );
+   }
+   else if (vsvg->base.key.viewport) {
+      do_viewport( vsvg,
+                   count,
+                   temp_buffer );
+   }
+
+   vsvg->emit->set_buffer( vsvg->emit,
+                           0, 
+                           temp_buffer,
+                           temp_vertex_stride );
+   
+   vsvg->emit->set_buffer( vsvg->emit, 
+                           1,
+                           &vsvg->draw->rasterizer->point_size,
+                           0);
+   
+   vsvg->emit->run( vsvg->emit,
+                    0, count,
+                    output_buffer );
+
+   FREE(temp_buffer);
+}
+
+
+
+
+
+static void vsvg_destroy( struct draw_vs_varient *varient )
+{
+   FREE(varient);
+}
+
+
+struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
+                                                 const struct draw_vs_varient_key *key )
+{
+   unsigned i;
+   struct translate_key fetch, emit;
+
+   struct draw_vs_varient_generic *vsvg = CALLOC_STRUCT( draw_vs_varient_generic );
+   if (vsvg == NULL)
+      return NULL;
+
+   vsvg->base.key = *key;
+   vsvg->base.vs = vs;
+   vsvg->base.set_buffer    = vsvg_set_buffer;
+   vsvg->base.run_elts      = vsvg_run_elts;
+   vsvg->base.run_linear    = vsvg_run_linear;
+   vsvg->base.destroy       = vsvg_destroy;
+
+   vsvg->draw = vs->draw;
+
+   vsvg->temp_vertex_stride = MAX2(key->nr_inputs,
+                                   vsvg->base.vs->info.num_outputs) * 4 * sizeof(float);
+
+   /* Build free-standing fetch and emit functions:
+    */
+   fetch.nr_elements = key->nr_inputs;
+   fetch.output_stride = vsvg->temp_vertex_stride;
+   for (i = 0; i < key->nr_inputs; i++) {
+      fetch.element[i].input_format = key->element[i].in.format;
+      fetch.element[i].input_buffer = key->element[i].in.buffer;
+      fetch.element[i].input_offset = key->element[i].in.offset;
+      fetch.element[i].output_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+      fetch.element[i].output_offset = i * 4 * sizeof(float);
+      assert(fetch.element[i].output_offset < fetch.output_stride);
+   }
+
+
+   emit.nr_elements = key->nr_outputs;
+   emit.output_stride = key->output_stride;
+   for (i = 0; i < key->nr_outputs; i++) {
+      if (key->element[i].out.format != EMIT_1F_PSIZE)
+      {      
+         emit.element[i].input_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+         emit.element[i].input_buffer = 0;
+         emit.element[i].input_offset = key->element[i].out.vs_output * 4 * sizeof(float);
+         emit.element[i].output_format = draw_translate_vinfo_format(key->element[i].out.format);
+         emit.element[i].output_offset = key->element[i].out.offset;
+         assert(emit.element[i].input_offset <= fetch.output_stride);
+      }
+      else {
+         emit.element[i].input_format = PIPE_FORMAT_R32_FLOAT;
+         emit.element[i].input_buffer = 1;
+         emit.element[i].input_offset = 0;
+         emit.element[i].output_format = PIPE_FORMAT_R32_FLOAT;
+         emit.element[i].output_offset = key->element[i].out.offset;
+      }
+   }
+
+   vsvg->fetch = draw_vs_get_fetch( vs->draw, &fetch );
+   vsvg->emit = draw_vs_get_emit( vs->draw, &emit );
+
+   return &vsvg->base;
+}
+
+
+
+
+
diff --git a/src/gallium/auxiliary/gallivm/Makefile b/src/gallium/auxiliary/gallivm/Makefile
new file mode 100644
index 00000000000..5a96d94ec37
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/Makefile
@@ -0,0 +1,92 @@
+# -*-makefile-*-
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = gallivm
+
+
+GALLIVM_SOURCES = \
+        gallivm.cpp  \
+        gallivm_cpu.cpp \
+        instructions.cpp  \
+        loweringpass.cpp \
+        tgsitollvm.cpp \
+        storage.cpp \
+        storagesoa.cpp \
+        instructionssoa.cpp
+
+INC_SOURCES = gallivm_builtins.cpp gallivmsoabuiltins.cpp
+
+CPP_SOURCES = \
+	$(GALLIVM_SOURCES)
+
+C_SOURCES =
+ASM_SOURCES =
+
+OBJECTS = $(C_SOURCES:.c=.o) \
+          $(CPP_SOURCES:.cpp=.o) \
+	  $(ASM_SOURCES:.S=.o)
+
+### Include directories
+INCLUDES = \
+	-I. \
+	-I$(TOP)/src/gallium/drivers \
+	-I$(TOP)/src/gallium/auxiliary \
+	-I$(TOP)/src/gallium/include \
+	-I$(TOP)/src/mesa \
+	-I$(TOP)/include
+
+
+##### RULES #####
+
+.c.o:
+	$(CC) -c $(INCLUDES) $(LLVM_CFLAGS) $(CFLAGS) $(DRIVER_DEFINES) $< -o $@
+
+.cpp.o:
+	$(CXX) -c $(INCLUDES) $(LLVM_CXXFLAGS) $(CXXFLAGS) $(DRIVER_DEFINES) $< -o $@
+
+.S.o:
+	$(CC) -c $(INCLUDES) $(CFLAGS) $(DRIVER_DEFINES)  $< -o $@
+
+##### TARGETS #####
+
+default:: depend symlinks $(LIBNAME)
+
+
+$(LIBNAME): $(OBJECTS) Makefile
+	$(TOP)/bin/mklib -o $@ -static $(OBJECTS)
+
+
+depend: $(C_SOURCES) $(CPP_SOURCES) $(ASM_SOURCES) $(INC_SOURCES)
+	rm -f depend
+	touch depend
+	$(MKDEP) $(MKDEP_OPTIONS) $(DRIVER_DEFINES) $(INCLUDES) $(C_SOURCES) $(CPP_SOURCES) \
+		$(ASM_SOURCES) $(INC_SOURCES) 2> /dev/null
+
+
+gallivm_builtins.cpp: llvm_builtins.c
+	clang --emit-llvm < $< |llvm-as|opt -std-compile-opts > temp1.bin
+	(echo "static const unsigned char llvm_builtins_data[] = {"; od -txC temp1.bin | sed -e "s/^[0-9]*//" -e s"/ \([0-9a-f][0-9a-f]\)/0x\1,/g" -e"\$$d" | sed -e"\$$s/,$$/,0x00};/") >$@
+	rm temp1.bin
+
+gallivmsoabuiltins.cpp: soabuiltins.c
+	clang --emit-llvm < $< |llvm-as|opt -std-compile-opts > temp2.bin
+	(echo "static const unsigned char soabuiltins_data[] = {"; od -txC temp2.bin | sed -e "s/^[0-9]*//" -e s"/ \([0-9a-f][0-9a-f]\)/0x\1,/g" -e"\$$d" | sed -e"\$$s/,$$/,0x00};/") >$@
+	rm temp2.bin
+
+# Emacs tags
+tags:
+	etags `find . -name \*.[ch]` `find ../include`
+
+
+# Remove .o and backup files
+clean:
+	-rm -f *.o */*.o *~ *.so *~ server/*.o
+	-rm -f depend depend.bak
+	-rm -f gallivm_builtins.cpp
+	-rm -f gallivmsoabuiltins.cpp
+
+symlinks:
+
+
+include depend
diff --git a/src/gallium/auxiliary/gallivm/SConscript b/src/gallium/auxiliary/gallivm/SConscript
new file mode 100644
index 00000000000..c0aa51b90a9
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/SConscript
@@ -0,0 +1,16 @@
+Import('*')
+
+gallivm = env.ConvenienceLibrary(
+	target = 'gallivm',
+	source = [
+        'gallivm.cpp',
+        'gallivm_cpu.cpp',
+        'instructions.cpp',
+        'loweringpass.cpp',
+        'tgsitollvm.cpp',
+        'storage.cpp',
+        'storagesoa.cpp',
+        'instructionssoa.cpp',
+	])
+
+auxiliaries.insert(0, gallivm)
diff --git a/src/gallium/auxiliary/gallivm/gallivm.cpp b/src/gallium/auxiliary/gallivm/gallivm.cpp
new file mode 100644
index 00000000000..f4af5cc8ad5
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/gallivm.cpp
@@ -0,0 +1,332 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin [email protected]
+  */
+#ifdef MESA_LLVM
+
+#include "gallivm.h"
+#include "gallivm_p.h"
+
+#include "instructions.h"
+#include "loweringpass.h"
+#include "storage.h"
+#include "tgsitollvm.h"
+
+#include "pipe/p_context.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "tgsi/tgsi_exec.h"
+#include "tgsi/tgsi_dump.h"
+
+#include <llvm/Module.h>
+#include <llvm/CallingConv.h>
+#include <llvm/Constants.h>
+#include <llvm/DerivedTypes.h>
+#include <llvm/Instructions.h>
+#include <llvm/ModuleProvider.h>
+#include <llvm/Pass.h>
+#include <llvm/PassManager.h>
+#include <llvm/Attributes.h>
+#include <llvm/Support/PatternMatch.h>
+#include <llvm/ExecutionEngine/JIT.h>
+#include <llvm/ExecutionEngine/Interpreter.h>
+#include <llvm/ExecutionEngine/GenericValue.h>
+#include <llvm/Support/MemoryBuffer.h>
+#include <llvm/LinkAllPasses.h>
+#include <llvm/Analysis/Verifier.h>
+#include <llvm/Analysis/LoopPass.h>
+#include <llvm/Target/TargetData.h>
+#include <llvm/Bitcode/ReaderWriter.h>
+#include <llvm/Transforms/Utils/Cloning.h>
+
+#include <sstream>
+#include <fstream>
+#include <iostream>
+
+static int GLOBAL_ID = 0;
+
+using namespace llvm;
+
+static inline
+void AddStandardCompilePasses(PassManager &PM)
+{
+   PM.add(new LoweringPass());
+   PM.add(createVerifierPass());                  // Verify that input is correct
+
+   PM.add(createLowerSetJmpPass());          // Lower llvm.setjmp/.longjmp
+
+   //PM.add(createStripSymbolsPass(true));
+
+   PM.add(createRaiseAllocationsPass());     // call %malloc -> malloc inst
+   PM.add(createCFGSimplificationPass());    // Clean up disgusting code
+   PM.add(createPromoteMemoryToRegisterPass());// Kill useless allocas
+   PM.add(createGlobalOptimizerPass());      // Optimize out global vars
+   PM.add(createGlobalDCEPass());            // Remove unused fns and globs
+   PM.add(createIPConstantPropagationPass());// IP Constant Propagation
+   PM.add(createDeadArgEliminationPass());   // Dead argument elimination
+   PM.add(createInstructionCombiningPass()); // Clean up after IPCP & DAE
+   PM.add(createCFGSimplificationPass());    // Clean up after IPCP & DAE
+
+   PM.add(createPruneEHPass());              // Remove dead EH info
+
+   PM.add(createFunctionInliningPass());   // Inline small functions
+   PM.add(createArgumentPromotionPass());    // Scalarize uninlined fn args
+
+   PM.add(createTailDuplicationPass());      // Simplify cfg by copying code
+   PM.add(createInstructionCombiningPass()); // Cleanup for scalarrepl.
+   PM.add(createCFGSimplificationPass());    // Merge & remove BBs
+   PM.add(createScalarReplAggregatesPass()); // Break up aggregate allocas
+   PM.add(createInstructionCombiningPass()); // Combine silly seq's
+   PM.add(createCondPropagationPass());      // Propagate conditionals
+
+   PM.add(createTailCallEliminationPass());  // Eliminate tail calls
+   PM.add(createCFGSimplificationPass());    // Merge & remove BBs
+   PM.add(createReassociatePass());          // Reassociate expressions
+   PM.add(createLoopRotatePass());
+   PM.add(createLICMPass());                 // Hoist loop invariants
+   PM.add(createLoopUnswitchPass());         // Unswitch loops.
+   PM.add(createLoopIndexSplitPass());       // Index split loops.
+   PM.add(createInstructionCombiningPass()); // Clean up after LICM/reassoc
+   PM.add(createIndVarSimplifyPass());       // Canonicalize indvars
+   PM.add(createLoopUnrollPass());           // Unroll small loops
+   PM.add(createInstructionCombiningPass()); // Clean up after the unroller
+   PM.add(createGVNPass());                  // Remove redundancies
+   PM.add(createSCCPPass());                 // Constant prop with SCCP
+
+   // Run instcombine after redundancy elimination to exploit opportunities
+   // opened up by them.
+   PM.add(createInstructionCombiningPass());
+   PM.add(createCondPropagationPass());      // Propagate conditionals
+
+   PM.add(createDeadStoreEliminationPass()); // Delete dead stores
+   PM.add(createAggressiveDCEPass());        // SSA based 'Aggressive DCE'
+   PM.add(createCFGSimplificationPass());    // Merge & remove BBs
+   PM.add(createSimplifyLibCallsPass());     // Library Call Optimizations
+   PM.add(createDeadTypeEliminationPass());  // Eliminate dead types
+   PM.add(createConstantMergePass());        // Merge dup global constants
+}
+
+void gallivm_prog_delete(struct gallivm_prog *prog)
+{
+   delete prog->module;
+   prog->module = 0;
+   prog->function = 0;
+   free(prog);
+}
+
+static inline void
+constant_interpolation(float (*inputs)[16][4],
+                       const struct tgsi_interp_coef *coefs,
+                       unsigned attrib,
+                       unsigned chan)
+{
+   unsigned i;
+
+   for (i = 0; i < QUAD_SIZE; ++i) {
+      inputs[i][attrib][chan] = coefs[attrib].a0[chan];
+   }
+}
+
+static inline void
+linear_interpolation(float (*inputs)[16][4],
+                     const struct tgsi_interp_coef *coefs,
+                     unsigned attrib,
+                     unsigned chan)
+{
+   unsigned i;
+
+   for( i = 0; i < QUAD_SIZE; i++ ) {
+      const float x = inputs[i][0][0];
+      const float y = inputs[i][0][1];
+
+      inputs[i][attrib][chan] =
+         coefs[attrib].a0[chan] +
+         coefs[attrib].dadx[chan] * x +
+         coefs[attrib].dady[chan] * y;
+   }
+}
+
+static inline void
+perspective_interpolation(float (*inputs)[16][4],
+                          const struct tgsi_interp_coef *coefs,
+                          unsigned attrib,
+                          unsigned chan )
+{
+   unsigned i;
+
+   for( i = 0; i < QUAD_SIZE; i++ ) {
+      const float x = inputs[i][0][0];
+      const float y = inputs[i][0][1];
+      /* WPOS.w here is really 1/w */
+      const float w = 1.0f / inputs[i][0][3];
+      assert(inputs[i][0][3] != 0.0);
+
+      inputs[i][attrib][chan] =
+         (coefs[attrib].a0[chan] +
+          coefs[attrib].dadx[chan] * x +
+          coefs[attrib].dady[chan] * y) * w;
+   }
+}
+
+void gallivm_ir_dump(struct gallivm_ir *ir, const char *file_prefix)
+{
+   if (!ir || !ir->module)
+      return;
+
+   if (file_prefix) {
+      std::ostringstream stream;
+      stream << file_prefix;
+      stream << ir->id;
+      stream << ".ll";
+      std::string name = stream.str();
+      std::ofstream out(name.c_str());
+      if (!out) {
+         std::cerr<<"Can't open file : "<<stream.str()<<std::endl;;
+         return;
+      }
+      out << (*ir->module);
+      out.close();
+   } else {
+      const llvm::Module::FunctionListType &funcs = ir->module->getFunctionList();
+      llvm::Module::FunctionListType::const_iterator itr;
+      std::cout<<"; ---------- Start shader "<<ir->id<<std::endl;
+      for (itr = funcs.begin(); itr != funcs.end(); ++itr) {
+         const llvm::Function &func = (*itr);
+         std::string name = func.getName();
+         const llvm::Function *found = 0;
+         if (name.find("vs_shader") != std::string::npos ||
+             name.find("fs_shader") != std::string::npos ||
+             name.find("function") != std::string::npos)
+            found = &func;
+         if (found) {
+            std::cout<<*found<<std::endl;
+         }
+      }
+      std::cout<<"; ---------- End shader "<<ir->id<<std::endl;
+   }
+}
+
+
+void gallivm_prog_inputs_interpolate(struct gallivm_prog *prog,
+                                     float (*inputs)[16][4],
+                                     const struct tgsi_interp_coef *coef)
+{
+   for (int i = 0; i < prog->num_interp; ++i) {
+      const gallivm_interpolate &interp = prog->interpolators[i];
+      switch (interp.type) {
+      case TGSI_INTERPOLATE_CONSTANT:
+         constant_interpolation(inputs, coef, interp.attrib, interp.chan);
+         break;
+
+      case TGSI_INTERPOLATE_LINEAR:
+         linear_interpolation(inputs, coef, interp.attrib, interp.chan);
+         break;
+
+      case TGSI_INTERPOLATE_PERSPECTIVE:
+         perspective_interpolation(inputs, coef, interp.attrib, interp.chan);
+         break;
+
+      default:
+         assert( 0 );
+      }
+   }
+}
+
+
+struct gallivm_ir * gallivm_ir_new(enum gallivm_shader_type type)
+{
+   struct gallivm_ir *ir =
+      (struct gallivm_ir *)calloc(1, sizeof(struct gallivm_ir));
+   ++GLOBAL_ID;
+   ir->id   = GLOBAL_ID;
+   ir->type = type;
+
+   return ir;
+}
+
+void gallivm_ir_set_layout(struct gallivm_ir *ir,
+                           enum gallivm_vector_layout layout)
+{
+   ir->layout = layout;
+}
+
+void gallivm_ir_set_components(struct gallivm_ir *ir, int num)
+{
+   ir->num_components = num;
+}
+
+void gallivm_ir_fill_from_tgsi(struct gallivm_ir *ir,
+                               const struct tgsi_token *tokens)
+{
+   std::cout << "Creating llvm from: " <<std::endl;
+   tgsi_dump(tokens, 0);
+
+   llvm::Module *mod = tgsi_to_llvmir(ir, tokens);
+   ir->module = mod;
+   gallivm_ir_dump(ir, 0);
+}
+
+void gallivm_ir_delete(struct gallivm_ir *ir)
+{
+   delete ir->module;
+   free(ir);
+}
+
+struct gallivm_prog * gallivm_ir_compile(struct gallivm_ir *ir)
+{
+   struct gallivm_prog *prog =
+      (struct gallivm_prog *)calloc(1, sizeof(struct gallivm_prog));
+
+   std::cout << "Before optimizations:"<<std::endl;
+   ir->module->dump();
+   std::cout<<"-------------------------------"<<std::endl;
+
+   PassManager veri;
+   veri.add(createVerifierPass());
+   veri.run(*ir->module);
+   llvm::Module *mod = llvm::CloneModule(ir->module);
+   prog->num_consts = ir->num_consts;
+   memcpy(prog->interpolators, ir->interpolators, sizeof(prog->interpolators));
+   prog->num_interp = ir->num_interp;
+
+   /* Run optimization passes over it */
+   PassManager passes;
+   passes.add(new TargetData(mod));
+   AddStandardCompilePasses(passes);
+   passes.run(*mod);
+   prog->module = mod;
+
+   std::cout << "After optimizations:"<<std::endl;
+   mod->dump();
+
+   return prog;
+}
+
+#endif /* MESA_LLVM */
diff --git a/src/gallium/auxiliary/gallivm/gallivm.h b/src/gallium/auxiliary/gallivm/gallivm.h
new file mode 100644
index 00000000000..36a64a77471
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/gallivm.h
@@ -0,0 +1,118 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin [email protected]
+  */
+
+#ifndef GALLIVM_H
+#define GALLIVM_H
+
+/*
+  LLVM representation consists of two stages - layout independent
+  intermediate representation gallivm_ir and driver specific
+  gallivm_prog. TGSI is first being translated into gallivm_ir
+  after that driver can set number of options on gallivm_ir and
+  have it compiled into gallivm_prog. gallivm_prog can be either
+  executed (assuming there's LLVM JIT backend for the current
+  target) or machine code generation can be done (assuming there's
+  a LLVM code generator for thecurrent target)
+ */
+#if defined __cplusplus
+extern "C" {
+#endif
+
+#include "pipe/p_state.h"
+
+#ifdef MESA_LLVM
+
+struct tgsi_token;
+
+struct gallivm_ir;
+struct gallivm_prog;
+struct gallivm_cpu_engine;
+struct tgsi_interp_coef;
+struct tgsi_sampler;
+struct tgsi_exec_vector;
+
+enum gallivm_shader_type {
+   GALLIVM_VS,
+   GALLIVM_FS
+};
+
+enum gallivm_vector_layout {
+   GALLIVM_AOS,
+   GALLIVM_SOA
+};
+
+struct gallivm_ir *gallivm_ir_new(enum gallivm_shader_type type);
+void               gallivm_ir_set_layout(struct gallivm_ir *ir,
+                                         enum gallivm_vector_layout layout);
+void               gallivm_ir_set_components(struct gallivm_ir *ir, int num);
+void               gallivm_ir_fill_from_tgsi(struct gallivm_ir *ir,
+                                             const struct tgsi_token *tokens);
+void               gallivm_ir_delete(struct gallivm_ir *ir);
+
+
+struct gallivm_prog *gallivm_ir_compile(struct gallivm_ir *ir);
+
+void gallivm_prog_inputs_interpolate(struct gallivm_prog *prog,
+                                     float (*inputs)[PIPE_MAX_SHADER_INPUTS][4],
+                                     const struct tgsi_interp_coef *coefs);
+void gallivm_prog_dump(struct gallivm_prog *prog, const char *file_prefix);
+
+
+struct gallivm_cpu_engine *gallivm_cpu_engine_create(struct gallivm_prog *prog);
+struct gallivm_cpu_engine *gallivm_global_cpu_engine();
+int gallivm_cpu_vs_exec(struct gallivm_prog *prog,
+                        struct tgsi_exec_machine *machine,
+                        const float (*input)[4],
+                        unsigned num_inputs,
+                        float (*output)[4],
+                        unsigned num_outputs,
+                        const float (*constants)[4],
+                        unsigned count,
+                        unsigned input_stride,
+                        unsigned output_stride);
+int gallivm_cpu_fs_exec(struct gallivm_prog *prog,
+                        float x, float y,
+                        float (*dests)[PIPE_MAX_SHADER_INPUTS][4],
+                        float (*inputs)[PIPE_MAX_SHADER_INPUTS][4],
+                        float (*consts)[4],
+                        struct tgsi_sampler *samplers);
+void gallivm_cpu_jit_compile(struct gallivm_cpu_engine *ee, struct gallivm_prog *prog);
+void gallivm_cpu_engine_delete(struct gallivm_cpu_engine *ee);
+
+
+#endif /* MESA_LLVM */
+
+#if defined __cplusplus
+}
+#endif
+
+#endif
diff --git a/src/gallium/auxiliary/gallivm/gallivm_builtins.cpp b/src/gallium/auxiliary/gallivm/gallivm_builtins.cpp
new file mode 100644
index 00000000000..634bac01507
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/gallivm_builtins.cpp
@@ -0,0 +1,140 @@
+static const unsigned char llvm_builtins_data[] = {
+0x42,0x43,0xc0,0xde,0x21,0x0c,0x00,0x00,0x27,0x02,0x00,0x00,0x01,0x10,0x00,0x00,
+0x10,0x00,0x00,0x00,0x07,0x81,0x23,0x91,0x41,0xc8,0x04,0x49,0x06,0x10,0x32,0x39,
+0x92,0x01,0x84,0x0c,0x25,0x05,0x08,0x19,0x1e,0x04,0x8b,0x62,0x80,0x14,0x45,0x02,
+0x42,0x92,0x0b,0x42,0xa4,0x10,0x32,0x14,0x38,0x08,0x18,0x49,0x0a,0x32,0x44,0x24,
+0x48,0x0a,0x90,0x21,0x23,0x44,0x72,0x80,0x8c,0x14,0x21,0x86,0x0a,0x8a,0x0a,0x64,
+0x0c,0x1f,0x00,0x00,0x49,0x18,0x00,0x00,0x03,0x00,0x00,0x00,0x0b,0x84,0xff,0xff,
+0xff,0xff,0x1f,0xc0,0x00,0x00,0x00,0x00,0x51,0x20,0x00,0x00,0x12,0x00,0x00,0x00,
+0x32,0x22,0x48,0x09,0x20,0x65,0x82,0x84,0x00,0x26,0x45,0x48,0x05,0x09,0x26,0x45,
+0xc6,0x05,0x42,0x52,0x26,0x08,0xae,0x19,0x80,0x61,0x04,0x02,0x98,0x23,0x00,0x83,
+0x29,0x80,0x21,0x00,0xb2,0x73,0x04,0x01,0x51,0x8a,0xf4,0x08,0x92,0xa4,0x39,0x47,
+0x80,0x50,0x2b,0x03,0x00,0xa0,0x08,0x21,0x5c,0x46,0x2b,0x44,0x08,0x21,0xd4,0x40,
+0x14,0x01,0x80,0x11,0x80,0x22,0x88,0x00,0x13,0x30,0x7c,0xc0,0x03,0x3b,0xf8,0x05,
+0x3b,0xa0,0x83,0x36,0xa8,0x07,0x77,0x58,0x07,0x77,0x78,0x87,0x7b,0x70,0x87,0x36,
+0x60,0x87,0x74,0x70,0x87,0x7a,0xc0,0x87,0x36,0x38,0x07,0x77,0xa8,0x87,0x0d,0xf7,
+0x50,0x0e,0x6d,0x00,0x0f,0x7a,0x60,0x07,0x74,0xa0,0x07,0x76,0x40,0x07,0x7a,0x60,
+0x07,0x74,0xd0,0x06,0xe9,0x10,0x07,0x7a,0x80,0x07,0x7a,0x80,0x07,0x6d,0x90,0x0e,
+0x78,0xa0,0x07,0x78,0xa0,0x07,0x78,0xd0,0x06,0xe9,0x10,0x07,0x76,0xa0,0x07,0x71,
+0x60,0x07,0x7a,0x10,0x07,0x76,0xd0,0x06,0xe9,0x30,0x07,0x72,0xa0,0x07,0x73,0x20,
+0x07,0x7a,0x30,0x07,0x72,0xd0,0x06,0xe9,0x60,0x07,0x74,0xa0,0x07,0x76,0x40,0x07,
+0x7a,0x60,0x07,0x74,0xd0,0x06,0xe6,0x30,0x07,0x72,0xa0,0x07,0x73,0x20,0x07,0x7a,
+0x30,0x07,0x72,0xd0,0x06,0xe6,0x60,0x07,0x74,0xa0,0x07,0x76,0x40,0x07,0x7a,0x60,
+0x07,0x74,0xd0,0x06,0xf6,0x60,0x07,0x74,0xa0,0x07,0x76,0x40,0x07,0x7a,0x60,0x07,
+0x74,0xd0,0x06,0xf6,0x10,0x07,0x72,0x80,0x07,0x7a,0x10,0x07,0x72,0x80,0x07,0x7a,
+0x10,0x07,0x72,0x80,0x07,0x6d,0x10,0x0e,0x70,0xa0,0x07,0x70,0xa0,0x07,0x76,0x40,
+0x07,0x6d,0x60,0x0e,0x78,0x00,0x07,0x7a,0x10,0x07,0x72,0x80,0x07,0x7a,0x10,0x07,
+0x72,0x80,0x07,0x3a,0x0f,0x84,0x48,0x20,0x23,0x24,0x40,0x00,0x62,0x67,0x88,0x9f,
+0x19,0x92,0x24,0x00,0x10,0x04,0x00,0x00,0x00,0x43,0x92,0x04,0x08,0x00,0x00,0x00,
+0x00,0x60,0x48,0xa2,0x00,0x40,0x10,0x00,0x00,0x00,0x0c,0x49,0x16,0x00,0x08,0x02,
+0x00,0x00,0x80,0x21,0x89,0x02,0x00,0x41,0x00,0x00,0x00,0x30,0x24,0x61,0x80,0x00,
+0x00,0x00,0x00,0x00,0x86,0x24,0x07,0x10,0x00,0x00,0x00,0x00,0xc0,0x90,0x44,0x01,
+0x80,0x20,0x00,0x00,0x00,0x18,0x92,0x1c,0x40,0x00,0x00,0x00,0x00,0x00,0x43,0x12,
+0x05,0x00,0x82,0x00,0x00,0x00,0x60,0x48,0x52,0x00,0x40,0x10,0x00,0x00,0x00,0x64,
+0x81,0x00,0x00,0x00,0x10,0x00,0x00,0x00,0x32,0x1e,0x98,0x10,0x19,0x11,0x4c,0x90,
+0x8c,0x09,0x26,0x47,0xc6,0x04,0x43,0x8a,0x8a,0x59,0x8b,0x43,0x50,0xd2,0x09,0x02,
+0x81,0xd2,0x73,0x50,0xc9,0x0c,0x2a,0x99,0x41,0x25,0x33,0xa8,0x64,0x56,0x28,0x66,
+0x2d,0x0e,0x41,0xcf,0x2a,0x15,0x04,0x4a,0xcf,0x41,0x25,0x33,0xa8,0x64,0x06,0x95,
+0xcc,0xa0,0x92,0x59,0x01,0x00,0x00,0x00,0x53,0x82,0x26,0x0c,0x04,0x00,0x00,0x00,
+0x22,0x00,0x00,0x00,0x13,0x04,0x41,0x2c,0x10,0x00,0x00,0x00,0x05,0x00,0x00,0x00,
+0x04,0xc6,0x08,0x40,0x10,0x04,0xe1,0x70,0x18,0x23,0x00,0x41,0x10,0x84,0xc3,0x60,
+0x04,0x00,0x00,0x00,0xc3,0x0d,0xce,0x43,0x4c,0x37,0x3c,0x8e,0x34,0xdc,0x30,0x41,
+0xc2,0x74,0x03,0x34,0x51,0xc3,0x0d,0x4d,0x44,0x4c,0x37,0x44,0x8d,0x35,0x56,0x01,
+0x04,0xc3,0x55,0x21,0x16,0x0e,0x04,0x00,0x0f,0x00,0x00,0x00,0xd6,0x10,0x00,0xe6,
+0x10,0x04,0x76,0x81,0x00,0x3e,0x30,0x0c,0x91,0x4f,0x1b,0x05,0x21,0x30,0x8f,0x6d,
+0x13,0x48,0xe0,0x03,0xc3,0x10,0xf9,0xb4,0x55,0x20,0x81,0x0f,0x0c,0x43,0xe4,0xd7,
+0x66,0x41,0x08,0xcc,0xa3,0x1f,0x40,0x41,0x34,0x53,0x84,0x99,0xc4,0x20,0x30,0x8f,
+0x61,0x10,0x02,0xb0,0x2c,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,
+0x27,0x00,0x00,0x00,0x13,0x04,0x43,0x2c,0x10,0x00,0x00,0x00,0x08,0x00,0x00,0x00,
+0x24,0x8a,0xa0,0x0c,0x46,0x00,0x4a,0x80,0xc2,0x1c,0x84,0x55,0x55,0xd6,0x1c,0x84,
+0x45,0x51,0x16,0x81,0x19,0x80,0x11,0x80,0x31,0x02,0x10,0x04,0x41,0xfc,0x03,0x00,
+0x63,0x08,0x0d,0x34,0xdc,0x70,0x55,0xc2,0x2c,0x43,0x20,0x60,0x73,0x0c,0xd3,0x15,
+0x8d,0x21,0x34,0xd1,0x18,0x42,0xf3,0x8c,0x55,0x00,0x81,0xa0,0x6d,0x73,0x0c,0x19,
+0xe7,0x60,0x87,0x52,0x38,0x10,0x00,0x00,0x10,0x00,0x00,0x00,0x27,0x50,0x20,0x05,
+0xd1,0x0c,0x17,0x60,0x20,0xc5,0x74,0x10,0x8d,0x65,0x14,0x13,0xf3,0xd4,0xb4,0x6d,
+0x14,0x13,0xf3,0xd4,0xb8,0x69,0x14,0x13,0xf3,0xd4,0xb6,0x75,0x14,0x13,0xf3,0xd4,
+0xba,0x35,0x0c,0x13,0xf3,0xd8,0x05,0x31,0x31,0x8f,0x6e,0x1c,0x84,0x00,0x2c,0xcb,
+0x01,0x14,0x44,0x33,0x45,0x98,0x61,0x0c,0x02,0xf3,0x00,0x00,0x00,0x00,0x00,0x00,
+0x61,0x20,0x00,0x00,0x81,0x00,0x00,0x00,0x13,0x04,0x4d,0x2c,0x10,0x00,0x00,0x00,
+0x04,0x00,0x00,0x00,0x24,0xca,0x60,0x04,0xa0,0x04,0x8a,0x80,0xc2,0x0c,0x00,0x91,
+0x11,0x00,0x00,0x00,0x63,0x08,0x4d,0x64,0x16,0xc1,0xe1,0x86,0xab,0x22,0x66,0x19,
+0x02,0x01,0x1b,0x43,0x70,0xa2,0x59,0x82,0x61,0x0c,0xe1,0x89,0x66,0x09,0x86,0x81,
+0x0a,0x20,0x0b,0x34,0x61,0x8e,0x81,0xda,0xa2,0x31,0x84,0x46,0xb2,0x8e,0xe0,0x70,
+0x83,0x57,0x11,0xb3,0x0c,0x44,0xf1,0x8d,0x21,0x38,0xd2,0x2c,0x81,0x31,0x86,0xf0,
+0x48,0xb3,0x04,0xc6,0x40,0x05,0x00,0x06,0x44,0x18,0x14,0x73,0x0c,0x9c,0x18,0x48,
+0x63,0x08,0xcd,0x64,0x64,0x40,0x70,0xb8,0xa1,0x0c,0x2a,0x62,0x96,0xe1,0x40,0xcc,
+0x60,0x0c,0xc1,0x99,0x66,0x09,0x92,0x31,0x84,0x67,0x9a,0x25,0x48,0x06,0x2a,0x80,
+0x33,0x38,0xd0,0x00,0x99,0x63,0x18,0x83,0x34,0x98,0xc6,0x10,0x1a,0xc8,0xd6,0x80,
+0xe0,0x70,0x03,0x1b,0x54,0xc4,0x2c,0x83,0xb2,0xb4,0xc1,0x18,0x82,0x03,0xcd,0x12,
+0x30,0x63,0x08,0x0f,0x34,0x4b,0xc0,0x0c,0x54,0x00,0x6e,0xa0,0xbc,0xc1,0x32,0xc7,
+0xa0,0x06,0x70,0x00,0x61,0x1c,0x84,0x03,0x01,0x00,0x00,0x00,0x4e,0x00,0x00,0x00,
+0x76,0x52,0x4c,0xcc,0x73,0xd3,0x24,0x05,0x64,0xec,0xcd,0x8d,0xcc,0xe5,0x87,0x46,
+0xc6,0x50,0x8a,0x89,0x79,0xee,0xdb,0x54,0x8a,0x89,0x79,0xee,0xdd,0x1a,0x88,0x89,
+0x79,0x68,0x73,0x20,0x26,0xe6,0xa9,0xed,0x81,0x98,0x98,0xc7,0x36,0x0b,0x62,0x62,
+0x9e,0xdb,0x32,0x88,0x89,0x79,0x72,0xd3,0x20,0x26,0xe6,0xd9,0x8d,0x83,0x98,0x98,
+0xa7,0xb7,0x95,0x62,0x62,0x9e,0xbb,0x27,0x2d,0x20,0x63,0x6f,0x6e,0x64,0x2e,0x3a,
+0x34,0x35,0x56,0x62,0x08,0x4e,0x53,0xd9,0xba,0xb5,0x14,0x02,0xf3,0xe0,0xf5,0x25,
+0x2c,0x82,0xd3,0x0c,0xbe,0xe0,0x34,0xd3,0x8d,0x9b,0x88,0x21,0x38,0xcd,0x60,0xd7,
+0x24,0x01,0x63,0xec,0xcd,0x8d,0xcc,0x45,0x87,0x44,0x80,0x8c,0xbd,0xb9,0x91,0xb9,
+0xfc,0xc4,0xd0,0x90,0x02,0x8c,0xb1,0x37,0x37,0x32,0x97,0x1f,0x73,0x29,0x26,0xe6,
+0xc1,0x71,0x7b,0x29,0x26,0xe6,0xc1,0x77,0xfb,0x28,0x04,0xe6,0xa9,0x6f,0x52,0x01,
+0x32,0xf6,0xe6,0x46,0xe6,0xa2,0x13,0x73,0x63,0x18,0x83,0xc0,0x3c,0xb6,0x41,0x08,
+0x4e,0x33,0x58,0x47,0x31,0x31,0x4f,0x5d,0x1f,0xc3,0x22,0x38,0xcd,0xe0,0x0b,0x4e,
+0x33,0xe1,0xbc,0xa5,0x18,0x82,0xd3,0x0c,0x77,0x6e,0x20,0xc5,0xc4,0x3c,0xb5,0x4e,
+0x3a,0x40,0xc6,0xde,0xdc,0xc8,0x5c,0x7e,0x64,0x70,0x2c,0xa4,0x98,0x98,0xa7,0xee,
+0x6f,0x20,0x11,0x9c,0x66,0xf0,0x05,0xa7,0x99,0xec,0x82,0x10,0x9c,0xa6,0x32,0x93,
+0x42,0x60,0x1e,0x7b,0xb7,0x98,0x62,0x62,0x9e,0xbc,0x36,0x16,0x43,0x70,0x9a,0x0a,
+0xa7,0x6d,0xa4,0x98,0x98,0xc7,0xbe,0x8d,0xa4,0x98,0x98,0xc7,0xce,0x0d,0xc6,0x10,
+0x9c,0x66,0xc0,0x7b,0x12,0x02,0x32,0xf6,0xe6,0x46,0xe6,0xa2,0x33,0x13,0x73,0x06,
+0x8b,0xe0,0x34,0x83,0x2f,0x38,0xcd,0x64,0xd3,0x07,0x50,0x10,0xcd,0x14,0x61,0xe6,
+0x61,0x08,0x4e,0x53,0xd5,0x36,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,
+0x4a,0x00,0x00,0x00,0x13,0x04,0x41,0x2c,0x10,0x00,0x00,0x00,0x07,0x00,0x00,0x00,
+0x24,0xca,0x60,0x04,0xa0,0x04,0x8a,0x80,0xc2,0x0c,0x00,0xb9,0x61,0x0c,0x04,0x10,
+0x1e,0xe1,0x19,0xc6,0x40,0x02,0xe1,0x11,0x1e,0x00,0x00,0x00,0x63,0x08,0xcd,0x63,
+0x15,0xc1,0x31,0x84,0x06,0xb2,0x8b,0xe0,0x18,0x42,0x13,0x59,0x46,0x70,0x0c,0xa1,
+0x71,0x6c,0x23,0x38,0x16,0x02,0x04,0xc7,0x64,0x61,0x1a,0x37,0x16,0x01,0x04,0x48,
+0x35,0xc7,0x20,0x79,0xcf,0x58,0x04,0x10,0x20,0xd5,0x1c,0xc3,0x07,0x06,0xd0,0x58,
+0x04,0x10,0x20,0xd5,0x1c,0x43,0x18,0x88,0x41,0x34,0x16,0x01,0x04,0x48,0x35,0xc7,
+0x30,0x06,0x64,0xe0,0x98,0x47,0xd0,0xc0,0x80,0xa0,0x89,0x01,0x41,0x23,0x03,0x82,
+0x63,0x21,0x40,0x70,0x50,0x66,0x70,0x06,0x68,0x90,0x06,0x58,0x06,0xe1,0x40,0x00,
+0x25,0x00,0x00,0x00,0x56,0x52,0x4c,0xcc,0x73,0xd3,0x56,0x41,0x4c,0xcc,0x53,0xdb,
+0x05,0x31,0x31,0xcf,0x6d,0x19,0xc4,0xc4,0x3c,0xba,0x6d,0x10,0x13,0xf3,0xf4,0xd6,
+0x41,0x08,0xc0,0xb2,0x18,0x46,0x21,0x38,0x4d,0x85,0x9b,0x46,0x21,0x38,0x4d,0xb5,
+0x9b,0x8a,0x21,0x00,0xcb,0x82,0xdf,0x66,0x62,0x08,0x4e,0x53,0xdd,0xb7,0x9d,0x18,
+0x82,0xd3,0x54,0xb7,0x6e,0x28,0x86,0xe0,0x34,0xd5,0xdd,0xdb,0x47,0x31,0x31,0x4f,
+0x9d,0x9b,0x87,0x21,0x00,0xcb,0x52,0xdf,0x06,0x62,0x08,0xc0,0xb2,0xd4,0xbc,0x59,
+0x10,0x82,0xd3,0x54,0x96,0x62,0x08,0x4e,0x53,0xe1,0xb6,0x85,0x14,0x13,0xf3,0xd8,
+0xb4,0x8d,0x14,0x13,0xf3,0xd8,0xb9,0x89,0x18,0x02,0xb0,0x2c,0xf6,0x6d,0x24,0x86,
+0x00,0x2c,0x8b,0xcd,0x1b,0x87,0x21,0x38,0x4d,0x55,0xd3,0xd6,0x30,0x54,0xc0,0x72,
+0x00,0x05,0xd1,0x4c,0x11,0x06,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,
+0x19,0x00,0x00,0x00,0x13,0x04,0x41,0x2c,0x10,0x00,0x00,0x00,0x03,0x00,0x00,0x00,
+0x24,0x4a,0x60,0x04,0x80,0xc2,0x0c,0x00,0x00,0x00,0x00,0x00,0x63,0x08,0xcd,0x33,
+0x16,0x01,0x04,0x48,0x34,0xc7,0x00,0x49,0xcf,0x58,0x04,0x10,0x28,0xd1,0x1c,0xc3,
+0x44,0x39,0x58,0x85,0x03,0x01,0x00,0x00,0x0a,0x00,0x00,0x00,0x26,0x41,0x08,0xc0,
+0xb2,0x18,0x45,0x21,0x00,0xcb,0xb2,0x5b,0x04,0x31,0x31,0x8f,0x6d,0x13,0xc4,0xc4,
+0x3c,0xb9,0x35,0x0c,0x15,0xb0,0x58,0x05,0x31,0x31,0x4f,0x7f,0x00,0x05,0xd1,0x4c,
+0x11,0x06,0x00,0x00,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,0x1b,0x00,0x00,0x00,
+0x13,0x04,0x41,0x2c,0x10,0x00,0x00,0x00,0x03,0x00,0x00,0x00,0x24,0xca,0x60,0x04,
+0xa0,0x04,0x8a,0x80,0xc2,0x0c,0x00,0x00,0x63,0x08,0xcd,0x33,0x16,0x01,0x04,0xca,
+0x34,0xc7,0x20,0x51,0xcf,0x1c,0x43,0x45,0x41,0x73,0x0c,0x16,0x15,0xcd,0x31,0x5c,
+0x94,0x83,0x58,0x38,0x10,0x00,0x00,0x00,0x0b,0x00,0x00,0x00,0x76,0x51,0x4c,0xcc,
+0x53,0xdb,0x86,0x51,0x4c,0xcc,0x53,0xe7,0x36,0x41,0x4c,0xcc,0x63,0x5b,0x05,0x31,
+0x31,0x8f,0x6e,0x16,0xc4,0xc4,0x3c,0xbd,0x51,0x10,0x02,0xb0,0x2c,0xd6,0x30,0x54,
+0xc0,0x72,0x00,0x05,0xd1,0x4c,0x11,0x06,0x00,0x00,0x00,0x00,0x61,0x20,0x00,0x00,
+0x2c,0x00,0x00,0x00,0x13,0x04,0x45,0x2c,0x10,0x00,0x00,0x00,0x03,0x00,0x00,0x00,
+0x24,0xca,0xa0,0x04,0x46,0x00,0x8a,0x80,0xc0,0x08,0x00,0x00,0x63,0x08,0x0d,0x34,
+0xdc,0x30,0x49,0xc4,0x2c,0x03,0x11,0x50,0x63,0x08,0xcd,0x33,0xdc,0x50,0x49,0xc4,
+0x2c,0x03,0x21,0x58,0x63,0x08,0x4d,0x34,0xdc,0x70,0x49,0xc4,0x2c,0x03,0x31,0x60,
+0x63,0x08,0x8d,0x33,0xdc,0x90,0x49,0x84,0x69,0x22,0x70,0xc3,0x27,0x1c,0x08,0x00,
+0x17,0x00,0x00,0x00,0x96,0x51,0x4c,0xcc,0x53,0xdf,0x66,0x41,0x08,0xcc,0x83,0xdb,
+0x04,0x31,0x31,0x4f,0x6d,0x15,0xc4,0xc4,0x3c,0xb7,0x61,0x10,0x02,0xf3,0xf0,0x76,
+0x41,0x4c,0xcc,0xb3,0x1f,0x81,0x11,0x11,0x13,0x15,0x35,0x37,0x90,0x2c,0x4e,0xf4,
+0x47,0x87,0x54,0xd7,0x17,0x70,0x2c,0x4e,0xf4,0x47,0x87,0x74,0x02,0xc8,0xe2,0x44,
+0x7f,0x74,0x48,0xb9,0x69,0x14,0x02,0xf3,0xd4,0xb8,0x6d,0x18,0x11,0x31,0x55,0xc0,
+0x62,0x0d,0x43,0x05,0x2c,0x07,0x50,0x10,0xcd,0x14,0x61,0x46,0x31,0x08,0xcc,0x03,
+0x00,0x00,0x00,0x00,0x71,0x20,0x00,0x00,0x12,0x00,0x00,0x00,0x66,0x40,0x54,0x82,
+0x23,0x19,0xc3,0xa0,0x20,0x8b,0x1d,0x18,0x4f,0x84,0x34,0x53,0x61,0x03,0xc4,0xe3,
+0x58,0x85,0x05,0x14,0xbe,0x34,0x45,0xb5,0x21,0x10,0x82,0x23,0x15,0x46,0x30,0x2c,
+0xc8,0x64,0x02,0x06,0xf0,0x3c,0x91,0x73,0x19,0x00,0xe1,0x4b,0x53,0x64,0x0a,0x84,
+0x84,0x34,0x85,0x25,0x0c,0x92,0x20,0x59,0xc1,0x20,0x30,0x8f,0x2d,0x10,0x95,0x84,
+0x34,0x00,0x00,0x00,0x00,0x00,0x00,0x00,0x00};
diff --git a/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp b/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp
new file mode 100644
index 00000000000..1bd00a0c2a6
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/gallivm_cpu.cpp
@@ -0,0 +1,243 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin [email protected]
+  */
+#ifdef MESA_LLVM
+
+#include "gallivm.h"
+#include "gallivm_p.h"
+
+#include "instructions.h"
+#include "loweringpass.h"
+#include "storage.h"
+#include "tgsitollvm.h"
+
+#include "pipe/p_context.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "tgsi/tgsi_exec.h"
+#include "tgsi/tgsi_dump.h"
+
+#include "util/u_memory.h"
+#include "util/u_math.h"
+
+#include <llvm/Module.h>
+#include <llvm/CallingConv.h>
+#include <llvm/Constants.h>
+#include <llvm/DerivedTypes.h>
+#include <llvm/Instructions.h>
+#include <llvm/ModuleProvider.h>
+#include <llvm/Pass.h>
+#include <llvm/PassManager.h>
+#include <llvm/Attributes.h>
+#include <llvm/Support/PatternMatch.h>
+#include <llvm/ExecutionEngine/JIT.h>
+#include <llvm/ExecutionEngine/Interpreter.h>
+#include <llvm/ExecutionEngine/GenericValue.h>
+#include <llvm/Support/MemoryBuffer.h>
+#include <llvm/LinkAllPasses.h>
+#include <llvm/Analysis/Verifier.h>
+#include <llvm/Analysis/LoopPass.h>
+#include <llvm/Target/TargetData.h>
+#include <llvm/Bitcode/ReaderWriter.h>
+#include <llvm/Transforms/Utils/Cloning.h>
+
+#include <sstream>
+#include <fstream>
+#include <iostream>
+
+struct gallivm_cpu_engine {
+   llvm::ExecutionEngine *engine;
+};
+
+static struct gallivm_cpu_engine *CPU = 0;
+
+typedef int (*fragment_shader_runner)(float x, float y,
+                                      float (*dests)[16][4],
+                                      float (*inputs)[16][4],
+                                      int num_attribs,
+                                      float (*consts)[4], int num_consts,
+                                      struct tgsi_sampler *samplers);
+
+int gallivm_cpu_fs_exec(struct gallivm_prog *prog,
+                        float fx, float fy,
+                        float (*dests)[16][4],
+                        float (*inputs)[16][4],
+                        float (*consts)[4],
+                        struct tgsi_sampler *samplers)
+{
+   fragment_shader_runner runner = reinterpret_cast<fragment_shader_runner>(prog->function);
+   assert(runner);
+
+   return runner(fx, fy, dests, inputs, prog->num_interp,
+                 consts, prog->num_consts,
+                 samplers);
+}
+
+static inline llvm::Function *func_for_shader(struct gallivm_prog *prog)
+{
+   llvm::Module *mod = prog->module;
+   llvm::Function *func = 0;
+
+   switch (prog->type) {
+   case GALLIVM_VS:
+      func = mod->getFunction("vs_shader");
+      break;
+   case GALLIVM_FS:
+      func = mod->getFunction("fs_shader");
+      break;
+   default:
+      assert(!"Unknown shader type!");
+      break;
+   }
+   return func;
+}
+
+/*!
+  This function creates a CPU based execution engine for the given gallivm_prog.
+  gallivm_cpu_engine should be used as a singleton throughout the library. Before
+  executing gallivm_prog_exec one needs to call gallivm_cpu_jit_compile.
+  The gallivm_prog instance which is being passed to the constructor is being
+  automatically JIT compiled so one shouldn't call gallivm_cpu_jit_compile
+  with it again.
+ */
+struct gallivm_cpu_engine * gallivm_cpu_engine_create(struct gallivm_prog *prog)
+{
+   struct gallivm_cpu_engine *cpu = (struct gallivm_cpu_engine *)
+                                    calloc(1, sizeof(struct gallivm_cpu_engine));
+   llvm::Module *mod = static_cast<llvm::Module*>(prog->module);
+   llvm::ExistingModuleProvider *mp = new llvm::ExistingModuleProvider(mod);
+   llvm::ExecutionEngine *ee = llvm::ExecutionEngine::create(mp, false);
+   ee->DisableLazyCompilation();
+   cpu->engine = ee;
+
+   llvm::Function *func = func_for_shader(prog);
+
+   prog->function = ee->getPointerToFunction(func);
+   CPU = cpu;
+   return cpu;
+}
+
+
+/*!
+  This function JIT compiles the given gallivm_prog with the given cpu based execution engine.
+  The reference to the generated machine code entry point will be stored
+  in the gallivm_prog program. After executing this function one can call gallivm_prog_exec
+  in order to execute the gallivm_prog on the CPU.
+ */
+void gallivm_cpu_jit_compile(struct gallivm_cpu_engine *cpu, struct gallivm_prog *prog)
+{
+   llvm::Module *mod = static_cast<llvm::Module*>(prog->module);
+   llvm::ExistingModuleProvider *mp = new llvm::ExistingModuleProvider(mod);
+   llvm::ExecutionEngine *ee = cpu->engine;
+   assert(ee);
+   /*FIXME : why was this disabled ? we need it for pow/sqrt/... */
+   ee->DisableLazyCompilation(false);
+   ee->addModuleProvider(mp);
+
+   llvm::Function *func = func_for_shader(prog);
+   prog->function = ee->getPointerToFunction(func);
+}
+
+void gallivm_cpu_engine_delete(struct gallivm_cpu_engine *cpu)
+{
+   free(cpu);
+}
+
+struct gallivm_cpu_engine * gallivm_global_cpu_engine()
+{
+   return CPU;
+}
+
+
+typedef void (*vertex_shader_runner)(void *ainputs,
+                                     void *dests,
+                                     float (*aconsts)[4]);
+
+#define MAX_TGSI_VERTICES 4
+/*!
+  This function is used to execute the gallivm_prog in software. Before calling
+  this function the gallivm_prog has to be JIT compiled with the gallivm_cpu_jit_compile
+  function.
+ */
+int gallivm_cpu_vs_exec(struct gallivm_prog *prog,
+                        struct tgsi_exec_machine *machine,
+                        const float (*input)[4],
+                        unsigned num_inputs,
+                        float (*output)[4],
+                        unsigned num_outputs,
+                        const float (*constants)[4],
+                        unsigned count,
+                        unsigned input_stride,
+                        unsigned output_stride )
+{
+   unsigned int i, j;
+   unsigned slot;
+   vertex_shader_runner runner = reinterpret_cast<vertex_shader_runner>(prog->function);
+   assert(runner);
+
+   for (i = 0; i < count; i += MAX_TGSI_VERTICES) {
+      unsigned int max_vertices = MIN2(MAX_TGSI_VERTICES, count - i);
+
+      /* Swizzle inputs.
+       */
+      for (j = 0; j < max_vertices; j++) {
+	 for (slot = 0; slot < num_inputs; slot++) {
+	    machine->Inputs[slot].xyzw[0].f[j] = input[slot][0];
+	    machine->Inputs[slot].xyzw[1].f[j] = input[slot][1];
+	    machine->Inputs[slot].xyzw[2].f[j] = input[slot][2];
+	    machine->Inputs[slot].xyzw[3].f[j] = input[slot][3];
+	 }
+
+	 input = (const float (*)[4])((const char *)input + input_stride);
+      }
+
+      /* run shader */
+      runner(machine->Inputs,
+             machine->Outputs,
+             (float (*)[4]) constants);
+
+      /* Unswizzle all output results
+       */
+      for (j = 0; j < max_vertices; j++) {
+         for (slot = 0; slot < num_outputs; slot++) {
+            output[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
+            output[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
+            output[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
+            output[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
+         }
+         output = (float (*)[4])((char *)output + output_stride);
+      }
+   }
+
+   return 0;
+}
+
+#endif
diff --git a/src/gallium/auxiliary/gallivm/gallivm_p.h b/src/gallium/auxiliary/gallivm/gallivm_p.h
new file mode 100644
index 00000000000..d2c5852bdf7
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/gallivm_p.h
@@ -0,0 +1,110 @@
+#ifndef GALLIVM_P_H
+#define GALLIVM_P_H
+
+#ifdef MESA_LLVM
+
+#include "gallivm.h"
+#include "pipe/p_shader_tokens.h"
+#include "pipe/p_compiler.h"
+
+namespace llvm {
+   class Module;
+}
+
+#if defined __cplusplus
+extern "C" {
+#endif
+
+enum gallivm_shader_type;
+enum gallivm_vector_layout;
+
+struct gallivm_interpolate {
+   int attrib;
+   int chan;
+   int type;
+};
+
+struct gallivm_ir {
+   llvm::Module *module;
+   int id;
+   enum gallivm_shader_type type;
+   enum gallivm_vector_layout layout;
+   int num_components;
+   int   num_consts;
+
+   /* FIXME: this might not be enough for some shaders */
+   struct gallivm_interpolate interpolators[32*4];
+   int   num_interp;
+};
+
+struct gallivm_prog {
+   llvm::Module *module;
+   void *function;
+
+   int   id;
+   enum gallivm_shader_type type;
+
+   int   num_consts;
+
+   /* FIXME: this might not be enough for some shaders */
+   struct gallivm_interpolate interpolators[32*4];
+   int   num_interp;
+};
+
+static INLINE void gallivm_swizzle_components(int swizzle,
+                                              int *xc, int *yc,
+                                              int *zc, int *wc)
+{
+   int x = swizzle / 1000; swizzle -= x * 1000;
+   int y = swizzle / 100;  swizzle -= y * 100;
+   int z = swizzle / 10;   swizzle -= z * 10;
+   int w = swizzle;
+
+   if (xc) *xc = x;
+   if (yc) *yc = y;
+   if (zc) *zc = z;
+   if (wc) *wc = w;
+}
+
+static INLINE boolean gallivm_is_swizzle(int swizzle)
+{
+   const int NO_SWIZZLE = TGSI_SWIZZLE_X * 1000 + TGSI_SWIZZLE_Y * 100 +
+                          TGSI_SWIZZLE_Z * 10 + TGSI_SWIZZLE_W;
+   return swizzle != NO_SWIZZLE;
+}
+
+static INLINE int gallivm_x_swizzle(int swizzle)
+{
+   int x;
+   gallivm_swizzle_components(swizzle, &x, 0, 0, 0);
+   return x;
+}
+
+static INLINE int gallivm_y_swizzle(int swizzle)
+{
+   int y;
+   gallivm_swizzle_components(swizzle, 0, &y, 0, 0);
+   return y;
+}
+
+static INLINE int gallivm_z_swizzle(int swizzle)
+{
+   int z;
+   gallivm_swizzle_components(swizzle, 0, 0, &z, 0);
+   return z;
+}
+
+static INLINE int gallivm_w_swizzle(int swizzle)
+{
+   int w;
+   gallivm_swizzle_components(swizzle, 0, 0, 0, &w);
+   return w;
+}
+
+#if defined __cplusplus
+}
+#endif
+
+#endif /* MESA_LLVM */
+
+#endif
diff --git a/src/gallium/auxiliary/gallivm/instructions.cpp b/src/gallium/auxiliary/gallivm/instructions.cpp
new file mode 100644
index 00000000000..ee8162efce5
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/instructions.cpp
@@ -0,0 +1,1193 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin [email protected]
+  */
+#ifdef MESA_LLVM
+
+#include "instructions.h"
+
+#include "storage.h"
+
+#include "util/u_memory.h"
+
+#include <llvm/CallingConv.h>
+#include <llvm/Constants.h>
+#include <llvm/DerivedTypes.h>
+#include <llvm/Function.h>
+#include <llvm/InstrTypes.h>
+#include <llvm/Instructions.h>
+#include <llvm/Attributes.h>
+#include <llvm/Support/MemoryBuffer.h>
+#include <llvm/Bitcode/ReaderWriter.h>
+
+#include <sstream>
+#include <fstream>
+#include <iostream>
+
+using namespace llvm;
+
+#include "gallivm_builtins.cpp"
+
+#if 0
+llvm::Value *arrayFromChannels(std::vector<llvm::Value*> &vals)
+{
+   VectorType *vectorType = VectorType::get(Type::FloatTy, 4);
+   ArrayType *vectorArray = ArrayType::get(vectorType, 4);
+}
+#endif
+
+static inline std::string createFuncName(int label)
+{
+   std::ostringstream stream;
+   stream << "function";
+   stream << label;
+   return stream.str();
+}
+
+Instructions::Instructions(llvm::Module *mod, llvm::Function *func, llvm::BasicBlock *block,
+                           Storage *storage)
+   :  m_mod(mod), m_func(func), m_builder(block), m_idx(0),
+      m_storage(storage)
+{
+   m_floatVecType = VectorType::get(Type::FloatTy, 4);
+
+   m_llvmFSqrt = 0;
+   m_llvmFAbs  = 0;
+   m_llvmPow   = 0;
+   m_llvmFloor = 0;
+   m_llvmFlog  = 0;
+   m_llvmFexp  = 0;
+   m_llvmLit  = 0;
+   m_fmtPtr = 0;
+
+   MemoryBuffer *buffer = MemoryBuffer::getMemBuffer(
+      (const char*)&llvm_builtins_data[0],
+      (const char*)&llvm_builtins_data[Elements(llvm_builtins_data)-1]);
+   m_mod = ParseBitcodeFile(buffer);
+}
+
+llvm::BasicBlock * Instructions::currentBlock() const
+{
+   return m_builder.GetInsertBlock();
+}
+
+llvm::Value * Instructions::abs(llvm::Value *in)
+{
+   std::vector<llvm::Value*> vec = extractVector(in);
+   Value *xabs  = callFAbs(vec[0]);
+   Value *yabs  = callFAbs(vec[1]);
+   Value *zabs  = callFAbs(vec[2]);
+   Value *wabs  = callFAbs(vec[3]);
+   return vectorFromVals(xabs, yabs, zabs, wabs);
+}
+
+llvm::Value * Instructions::add(llvm::Value *in1, llvm::Value *in2)
+{
+   return m_builder.CreateAdd(in1, in2, name("add"));
+}
+
+llvm::Value * Instructions::arl(llvm::Value *in)
+{
+   return floor(in);
+}
+
+void Instructions::beginLoop()
+{
+   BasicBlock *begin = BasicBlock::Create(name("loop"), m_func,0);
+   BasicBlock *end = BasicBlock::Create(name("endloop"), m_func,0);
+
+   m_builder.CreateBr(begin);
+   Loop loop;
+   loop.begin = begin;
+   loop.end   = end;
+   m_builder.SetInsertPoint(begin);
+   m_loopStack.push(loop);
+}
+
+void Instructions::bgnSub(unsigned label)
+{
+   llvm::Function *func = findFunction(label);
+
+   Function::arg_iterator args = func->arg_begin();
+   Value *ptr_INPUT = args++;
+   ptr_INPUT->setName("INPUT");
+   m_storage->pushArguments(ptr_INPUT);
+
+   llvm::BasicBlock *entry = BasicBlock::Create("entry", func, 0);
+
+   m_func = func;
+   m_builder.SetInsertPoint(entry);
+}
+
+void Instructions::brk()
+{
+   assert(!m_loopStack.empty());
+   BasicBlock *unr = BasicBlock::Create(name("unreachable"), m_func,0);
+   m_builder.CreateBr(m_loopStack.top().end);
+   m_builder.SetInsertPoint(unr);
+}
+
+void Instructions::cal(int label, llvm::Value *input)
+{
+   std::vector<Value*> params;
+   params.push_back(input);
+   llvm::Function *func = findFunction(label);
+
+   m_builder.CreateCall(func, params.begin(), params.end());
+}
+
+llvm::Value * Instructions::ceil(llvm::Value *in)
+{
+   std::vector<llvm::Value*> vec = extractVector(in);
+   return vectorFromVals(callCeil(vec[0]), callCeil(vec[1]),
+                         callCeil(vec[2]), callCeil(vec[3]));
+}
+
+llvm::Value * Instructions::clamp(llvm::Value *in1)
+{
+   llvm::Value *zero = constVector(0.0f, 0.0f, 0.0f, 0.0f);
+   llvm::Value *one = constVector(1.0f, 1.0f, 1.0f, 1.0f);
+   return min( max(zero, in1), one);
+}
+
+llvm::Value * Instructions::cmp(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3)
+{
+   llvm::Function *func = m_mod->getFunction("cmp");
+   assert(func);
+
+   std::vector<Value*> params;
+   params.push_back(in1);
+   params.push_back(in2);
+   params.push_back(in3);
+   CallInst *call = m_builder.CreateCall(func, params.begin(), params.end(), name("cmpres"));
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::cnd(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3)
+{
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+   std::vector<llvm::Value*> vec3 = extractVector(in3);
+   Constant *half = ConstantFP::get(APFloat(0.5f));
+
+   Value *xcmp  = m_builder.CreateFCmpOGT(vec1[0], half, name("xcmp"));
+   Value *selx = m_builder.CreateSelect(xcmp, vec2[0], vec3[0],
+                                        name("selx"));
+
+   Value *ycmp  = m_builder.CreateFCmpOGT(vec1[1], half, name("ycmp"));
+   Value *sely = m_builder.CreateSelect(ycmp, vec2[1], vec3[1],
+                                        name("sely"));
+
+   Value *zcmp  = m_builder.CreateFCmpOGT(vec1[2], half, name("zcmp"));
+   Value *selz = m_builder.CreateSelect(zcmp, vec2[2], vec3[2],
+                                        name("selz"));
+
+   Value *wcmp  = m_builder.CreateFCmpOGT(vec1[3], half, name("wcmp"));
+   Value *selw = m_builder.CreateSelect(wcmp, vec2[3], vec3[3],
+                                        name("selw"));
+
+   return vectorFromVals(selx, sely, selz, selw);
+}
+
+llvm::Value * Instructions::cnd0(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3)
+{
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+   std::vector<llvm::Value*> vec3 = extractVector(in3);
+   Constant *zero = Constant::getNullValue(Type::FloatTy);
+
+   Value *xcmp  = m_builder.CreateFCmpOGE(vec1[0], zero, name("xcmp"));
+   Value *selx = m_builder.CreateSelect(xcmp, vec2[0], vec3[0],
+                                        name("selx"));
+
+   Value *ycmp  = m_builder.CreateFCmpOGE(vec1[1], zero, name("ycmp"));
+   Value *sely = m_builder.CreateSelect(ycmp, vec2[1], vec3[1],
+                                        name("sely"));
+
+   Value *zcmp  = m_builder.CreateFCmpOGE(vec1[2], zero, name("zcmp"));
+   Value *selz = m_builder.CreateSelect(zcmp, vec2[2], vec3[2],
+                                        name("selz"));
+
+   Value *wcmp  = m_builder.CreateFCmpOGE(vec1[3], zero, name("wcmp"));
+   Value *selw = m_builder.CreateSelect(wcmp, vec2[3], vec3[3],
+                                        name("selw"));
+
+   return vectorFromVals(selx, sely, selz, selw);
+}
+
+llvm::Value * Instructions::cos(llvm::Value *in)
+{
+#if 0
+   llvm::Function *func = m_mod->getFunction("vcos");
+   assert(func);
+
+   CallInst *call = m_builder.CreateCall(func, in, name("cosres"));
+   call->setTailCall(false);
+   return call;
+#else
+   std::vector<llvm::Value*> elems = extractVector(in);
+   Function *func = m_mod->getFunction("cosf");
+   assert(func);
+   CallInst *cos = m_builder.CreateCall(func, elems[0], name("cosres"));
+   cos->setCallingConv(CallingConv::C);
+   cos->setTailCall(true);
+   return vectorFromVals(cos, cos, cos, cos);
+#endif
+}
+
+llvm::Value * Instructions::cross(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *x1 = m_builder.CreateExtractElement(in1,
+                                              m_storage->constantInt(0),
+                                              name("x1"));
+   Value *y1 = m_builder.CreateExtractElement(in1,
+                                              m_storage->constantInt(1),
+                                              name("y1"));
+   Value *z1 = m_builder.CreateExtractElement(in1,
+                                              m_storage->constantInt(2),
+                                              name("z1"));
+
+   Value *x2 = m_builder.CreateExtractElement(in2,
+                                              m_storage->constantInt(0),
+                                              name("x2"));
+   Value *y2 = m_builder.CreateExtractElement(in2,
+                                              m_storage->constantInt(1),
+                                              name("y2"));
+   Value *z2 = m_builder.CreateExtractElement(in2,
+                                              m_storage->constantInt(2),
+                                              name("z2"));
+   Value *y1z2 = mul(y1, z2);
+   Value *z1y2 = mul(z1, y2);
+
+   Value *z1x2 = mul(z1, x2);
+   Value *x1z2 = mul(x1, z2);
+
+   Value *x1y2 = mul(x1, y2);
+   Value *y1x2 = mul(y1, x2);
+
+   return vectorFromVals(sub(y1z2, z1y2), sub(z1x2, x1z2), sub(x1y2, y1x2));
+}
+
+llvm::Value * Instructions::ddx(llvm::Value *in)
+{
+   // FIXME
+   assert(0);
+}
+
+llvm::Value * Instructions::ddy(llvm::Value *in)
+{
+   // FIXME
+   assert(0);
+}
+
+llvm::Value * Instructions::div(llvm::Value *in1, llvm::Value *in2)
+{
+   return m_builder.CreateFDiv(in1, in2, name("div"));
+}
+
+llvm::Value * Instructions::dot2add(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3)
+{
+   Value *mulRes = mul(in1, in2);
+   Value *x = m_builder.CreateExtractElement(mulRes,
+                                                          m_storage->constantInt(0),
+                                                          name("extractx"));
+   Value *y = m_builder.CreateExtractElement(mulRes,
+                                                          m_storage->constantInt(1),
+                                                          name("extracty"));
+   Value *z = m_builder.CreateExtractElement(in3,
+                                                          m_storage->constantInt(2),
+                                                          name("extractz"));
+   Value *xy = m_builder.CreateAdd(x, y,name("xy"));
+   Value *dot2add = m_builder.CreateAdd(xy, z, name("dot2add"));
+   return vectorFromVals(dot2add, dot2add, dot2add, dot2add);
+}
+
+llvm::Value * Instructions::dp2(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *mulRes = mul(in1, in2);
+   Value *x = m_builder.CreateExtractElement(mulRes,
+                                                          m_storage->constantInt(0),
+                                                          name("extractx"));
+   Value *y = m_builder.CreateExtractElement(mulRes,
+                                                          m_storage->constantInt(1),
+                                                          name("extracty"));
+   Value *xy = m_builder.CreateAdd(x, y,name("xy"));
+   return vectorFromVals(xy, xy, xy, xy);
+}
+
+llvm::Value * Instructions::dp3(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *mulRes = mul(in1, in2);
+   Value *x = m_builder.CreateExtractElement(mulRes,
+                                                          m_storage->constantInt(0),
+                                                          name("extractx"));
+   Value *y = m_builder.CreateExtractElement(mulRes,
+                                                          m_storage->constantInt(1),
+                                                          name("extracty"));
+   Value *z = m_builder.CreateExtractElement(mulRes,
+                                                          m_storage->constantInt(2),
+                                                          name("extractz"));
+   Value *xy = m_builder.CreateAdd(x, y,name("xy"));
+   Value *dot3 = m_builder.CreateAdd(xy, z, name("dot3"));
+   return vectorFromVals(dot3, dot3, dot3, dot3);
+}
+
+llvm::Value * Instructions::dp4(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *mulRes = mul(in1, in2);
+   std::vector<llvm::Value*> vec = extractVector(mulRes);
+   Value *xy = m_builder.CreateAdd(vec[0], vec[1], name("xy"));
+   Value *xyz = m_builder.CreateAdd(xy, vec[2], name("xyz"));
+   Value *dot4 = m_builder.CreateAdd(xyz, vec[3], name("dot4"));
+   return vectorFromVals(dot4, dot4, dot4, dot4);
+}
+
+llvm::Value * Instructions::dph(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *mulRes = mul(in1, in2);
+   std::vector<llvm::Value*> vec1 = extractVector(mulRes);
+   Value *xy = m_builder.CreateAdd(vec1[0], vec1[1], name("xy"));
+   Value *xyz = m_builder.CreateAdd(xy, vec1[2], name("xyz"));
+   Value *dph = m_builder.CreateAdd(xyz, vec1[3], name("dph"));
+   return vectorFromVals(dph, dph, dph, dph);
+}
+
+llvm::Value * Instructions::dst(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *y1 = m_builder.CreateExtractElement(in1,
+                                              m_storage->constantInt(1),
+                                              name("y1"));
+   Value *z = m_builder.CreateExtractElement(in1,
+                                             m_storage->constantInt(2),
+                                             name("z"));
+   Value *y2 = m_builder.CreateExtractElement(in2,
+                                              m_storage->constantInt(1),
+                                              name("y2"));
+   Value *w = m_builder.CreateExtractElement(in2,
+                                             m_storage->constantInt(3),
+                                             name("w"));
+   Value *ry = m_builder.CreateMul(y1, y2, name("tyuy"));
+   return vectorFromVals(ConstantFP::get(APFloat(1.f)),
+                         ry, z, w);
+}
+
+void Instructions::elseop()
+{
+   assert(!m_ifStack.empty());
+   BasicBlock *ifend = BasicBlock::Create(name("ifend"), m_func,0);
+   m_builder.CreateBr(ifend);
+   m_builder.SetInsertPoint(m_ifStack.top());
+   currentBlock()->setName(name("ifelse"));
+   m_ifStack.pop();
+   m_ifStack.push(ifend);
+}
+
+void Instructions::endif()
+{
+   assert(!m_ifStack.empty());
+   m_builder.CreateBr(m_ifStack.top());
+   m_builder.SetInsertPoint(m_ifStack.top());
+   m_ifStack.pop();
+}
+
+void Instructions::endLoop()
+{
+   assert(!m_loopStack.empty());
+   Loop loop = m_loopStack.top();
+   m_builder.CreateBr(loop.begin);
+   loop.end->moveAfter(currentBlock());
+   m_builder.SetInsertPoint(loop.end);
+   m_loopStack.pop();
+}
+
+void Instructions::end()
+{
+   m_builder.CreateRetVoid();
+}
+
+void Instructions::endSub()
+{
+   m_func = 0;
+   m_builder.SetInsertPoint(0);
+}
+
+llvm::Value * Instructions::exp(llvm::Value *in)
+{
+   std::vector<llvm::Value*> vec = extractVector(in);
+   return vectorFromVals(callFExp(vec[0]), callFExp(vec[1]),
+                             callFExp(vec[2]), callFExp(vec[3]));
+}
+
+llvm::Value * Instructions::ex2(llvm::Value *in)
+{
+   llvm::Value *val = callPow(ConstantFP::get(APFloat(2.f)),
+                              m_builder.CreateExtractElement(
+                                 in, m_storage->constantInt(0),
+                                 name("x1")));
+   return vectorFromVals(val, val, val, val);
+}
+
+llvm::Value * Instructions::floor(llvm::Value *in)
+{
+   std::vector<llvm::Value*> vec = extractVector(in);
+   return vectorFromVals(callFloor(vec[0]), callFloor(vec[1]),
+                         callFloor(vec[2]), callFloor(vec[3]));
+}
+
+llvm::Value * Instructions::frc(llvm::Value *in)
+{
+   llvm::Value *flr = floor(in);
+   return sub(in, flr);
+}
+
+void Instructions::ifop(llvm::Value *in)
+{
+   BasicBlock *ifthen = BasicBlock::Create(name("ifthen"), m_func,0);
+   BasicBlock *ifend = BasicBlock::Create(name("ifthenend"), m_func,0);
+
+   //BasicBlock *yblock = new BasicBlock(name("yblock"), m_func,0);
+   //BasicBlock *zblock = new BasicBlock(name("zblock"), m_func,0);
+   //BasicBlock *wblock = new BasicBlock(name("wblock"), m_func,0);
+
+   Constant *float0 = Constant::getNullValue(Type::FloatTy);
+
+   Value *x = m_builder.CreateExtractElement(in, m_storage->constantInt(0),
+                                             name("extractx"));
+   Value *xcmp = m_builder.CreateFCmpUNE(x, float0, name("xcmp"));
+   m_builder.CreateCondBr(xcmp, ifthen, ifend);
+   //m_builder.SetInsertPoint(yblock);
+
+   m_builder.SetInsertPoint(ifthen);
+   m_ifStack.push(ifend);
+}
+
+llvm::Value * Instructions::kil(llvm::Value *in)
+{
+   llvm::Function *func = m_mod->getFunction("kil");
+   assert(func);
+
+   CallInst *call = m_builder.CreateCall(func, in, name("kilpres"));
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::lerp(llvm::Value *in1, llvm::Value *in2,
+                                 llvm::Value *in3)
+{
+   llvm::Value *m = mul(in1, in2);
+   llvm::Value *vec1 = constVector(1.f, 1.f, 1.f, 1.f);
+   llvm::Value *s = sub(vec1, in1);
+   return add(m, mul(s, in3));
+}
+
+llvm::Value * Instructions::lg2(llvm::Value *in)
+{
+   std::vector<llvm::Value*> vec = extractVector(in);
+   llvm::Value *const_vec = constVector(1.442695f, 1.442695f,
+                                        1.442695f, 1.442695f);
+   return mul(vectorFromVals(callFLog(vec[0]), callFLog(vec[1]),
+                             callFLog(vec[2]), callFLog(vec[3])), const_vec);
+}
+
+llvm::Value * Instructions::lit(llvm::Value *in)
+{
+   if (!m_llvmLit) {
+      m_llvmLit = m_mod->getFunction("lit");
+   }
+   CallInst *call = m_builder.CreateCall(m_llvmLit, in, name("litres"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::log(llvm::Value *in)
+{
+   std::vector<llvm::Value*> vec = extractVector(in);
+   return vectorFromVals(callFLog(vec[0]), callFLog(vec[1]),
+                             callFLog(vec[2]), callFLog(vec[3]));
+}
+
+llvm::Value * Instructions::madd(llvm::Value *in1, llvm::Value *in2,
+                                 llvm::Value *in3)
+{
+   Value *mulRes = mul(in1, in2);
+   return add(mulRes, in3);
+}
+
+llvm::Value * Instructions::max(llvm::Value *in1, llvm::Value *in2)
+{
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+
+   Value *xcmp  = m_builder.CreateFCmpOGT(vec1[0], vec2[0],
+                                          name("xcmp"));
+   Value *selx = m_builder.CreateSelect(xcmp, vec1[0], vec2[0],
+                                        name("selx"));
+
+   Value *ycmp  = m_builder.CreateFCmpOGT(vec1[1], vec2[1],
+                                          name("ycmp"));
+   Value *sely = m_builder.CreateSelect(ycmp, vec1[1], vec2[1],
+                                        name("sely"));
+
+   Value *zcmp  = m_builder.CreateFCmpOGT(vec1[2], vec2[2],
+                                          name("zcmp"));
+   Value *selz = m_builder.CreateSelect(zcmp, vec1[2], vec2[2],
+                                        name("selz"));
+
+   Value *wcmp  = m_builder.CreateFCmpOGT(vec1[3], vec2[3],
+                                          name("wcmp"));
+   Value *selw = m_builder.CreateSelect(wcmp, vec1[3], vec2[3],
+                                        name("selw"));
+
+   return vectorFromVals(selx, sely, selz, selw);
+}
+
+llvm::Value * Instructions::min(llvm::Value *in1, llvm::Value *in2)
+{
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+
+   Value *xcmp  = m_builder.CreateFCmpOLT(vec1[0], vec2[0], name("xcmp"));
+   Value *selx = m_builder.CreateSelect(xcmp, vec1[0], vec2[0],
+                                        name("selx"));
+
+   Value *ycmp  = m_builder.CreateFCmpOLT(vec1[1], vec2[1], name("ycmp"));
+   Value *sely = m_builder.CreateSelect(ycmp, vec1[1], vec2[1],
+                                        name("sely"));
+
+   Value *zcmp  = m_builder.CreateFCmpOLT(vec1[2], vec2[2], name("zcmp"));
+   Value *selz = m_builder.CreateSelect(zcmp, vec1[2], vec2[2],
+                                        name("selz"));
+
+   Value *wcmp  = m_builder.CreateFCmpOLT(vec1[3], vec2[3], name("wcmp"));
+   Value *selw = m_builder.CreateSelect(wcmp, vec1[3], vec2[3],
+                                        name("selw"));
+
+   return vectorFromVals(selx, sely, selz, selw);
+}
+
+llvm::Value * Instructions::mul(llvm::Value *in1, llvm::Value *in2)
+{
+   return m_builder.CreateMul(in1, in2, name("mul"));
+}
+
+llvm::Value * Instructions::neg(llvm::Value *in)
+{
+   Value *neg = m_builder.CreateNeg(in, name("neg"));
+   return neg;
+}
+
+llvm::Value * Instructions::nrm(llvm::Value *in)
+{
+   llvm::Value *v = rsq(in);
+   return mul(v, in);
+}
+
+llvm::Value * Instructions::pow(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *x1 = m_builder.CreateExtractElement(in1,
+                                              m_storage->constantInt(0),
+                                              name("x1"));
+   Value *x2 = m_builder.CreateExtractElement(in2,
+                                              m_storage->constantInt(0),
+                                              name("x2"));
+   llvm::Value *val = callPow(x1, x2);
+   return vectorFromVals(val, val, val, val);
+}
+
+llvm::Value * Instructions::rcp(llvm::Value *in1)
+{
+   Value *x1 = m_builder.CreateExtractElement(in1,
+                                              m_storage->constantInt(0),
+                                              name("x1"));
+   Value *res = m_builder.CreateFDiv(ConstantFP::get(APFloat(1.f)),
+                                     x1, name("rcp"));
+   return vectorFromVals(res, res, res, res);
+}
+
+llvm::Value * Instructions::rsq(llvm::Value *in1)
+{
+   Value *x = m_builder.CreateExtractElement(in1,
+                                             m_storage->constantInt(0),
+                                             name("extractx"));
+   Value *abs  = callFAbs(x);
+   Value *sqrt = callFSqrt(abs);
+
+   Value *rsqrt = m_builder.CreateFDiv(ConstantFP::get(APFloat(1.f)),
+                                       sqrt,
+                                       name("rsqrt"));
+   return vectorFromVals(rsqrt, rsqrt, rsqrt, rsqrt);
+}
+
+llvm::Value * Instructions::scs(llvm::Value *in)
+{
+   llvm::Function *func = m_mod->getFunction("scs");
+   assert(func);
+
+   CallInst *call = m_builder.CreateCall(func, in, name("scsres"));
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::seq(llvm::Value *in1, llvm::Value *in2)
+{
+   Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f));
+   Constant *const0f = Constant::getNullValue(Type::FloatTy);
+
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+
+   Value *xcmp = m_builder.CreateFCmpOEQ(vec1[0], vec2[0], name("xcmp"));
+   Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel"));
+
+   Value *ycmp = m_builder.CreateFCmpOEQ(vec1[1], vec2[1], name("ycmp"));
+   Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel"));
+
+   Value *zcmp = m_builder.CreateFCmpOEQ(vec1[2], vec2[2], name("zcmp"));
+   Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel"));
+
+   Value *wcmp = m_builder.CreateFCmpOEQ(vec1[3], vec2[3], name("wcmp"));
+   Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel"));
+
+   return vectorFromVals(x, y, z, w);
+}
+
+llvm::Value * Instructions::sfl(llvm::Value *in1, llvm::Value *in2)
+{
+   Constant *const0f = Constant::getNullValue(Type::FloatTy);
+
+   return vectorFromVals(const0f, const0f, const0f, const0f);
+}
+
+llvm::Value * Instructions::sge(llvm::Value *in1, llvm::Value *in2)
+{
+   Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f));
+   Constant *const0f = Constant::getNullValue(Type::FloatTy);
+
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+
+   Value *xcmp = m_builder.CreateFCmpOGE(vec1[0], vec2[0], name("xcmp"));
+   Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel"));
+
+   Value *ycmp = m_builder.CreateFCmpOGE(vec1[1], vec2[1], name("ycmp"));
+   Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel"));
+
+   Value *zcmp = m_builder.CreateFCmpOGE(vec1[2], vec2[2], name("zcmp"));
+   Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel"));
+
+   Value *wcmp = m_builder.CreateFCmpOGE(vec1[3], vec2[3], name("wcmp"));
+   Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel"));
+
+   return vectorFromVals(x, y, z, w);
+}
+
+llvm::Value * Instructions::sgt(llvm::Value *in1, llvm::Value *in2)
+{
+   Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f));
+   Constant *const0f = Constant::getNullValue(Type::FloatTy);
+
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+   Value *xcmp = m_builder.CreateFCmpOGT(vec1[0], vec2[0], name("xcmp"));
+   Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel"));
+
+   Value *ycmp = m_builder.CreateFCmpOGT(vec1[1], vec2[1], name("ycmp"));
+   Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel"));
+
+   Value *zcmp = m_builder.CreateFCmpOGT(vec1[2], vec2[2], name("zcmp"));
+   Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel"));
+
+   Value *wcmp = m_builder.CreateFCmpOGT(vec1[3], vec2[3], name("wcmp"));
+   Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel"));
+
+   return vectorFromVals(x, y, z, w);
+}
+
+llvm::Value * Instructions::sin(llvm::Value *in)
+{
+   llvm::Function *func = m_mod->getFunction("vsin");
+   assert(func);
+
+   CallInst *call = m_builder.CreateCall(func, in, name("sinres"));
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::sle(llvm::Value *in1, llvm::Value *in2)
+{
+   Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f));
+   Constant *const0f = Constant::getNullValue(Type::FloatTy);
+
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+
+   Value *xcmp = m_builder.CreateFCmpOLE(vec1[0], vec2[0], name("xcmp"));
+   Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel"));
+
+   Value *ycmp = m_builder.CreateFCmpOLE(vec1[1], vec2[1], name("ycmp"));
+   Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel"));
+
+   Value *zcmp = m_builder.CreateFCmpOLE(vec1[2], vec2[2], name("zcmp"));
+   Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel"));
+
+   Value *wcmp = m_builder.CreateFCmpOLE(vec1[3], vec2[3], name("wcmp"));
+   Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel"));
+
+   return vectorFromVals(x, y, z, w);
+}
+
+llvm::Value * Instructions::slt(llvm::Value *in1, llvm::Value *in2)
+{
+   Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f));
+   Constant *const0f = Constant::getNullValue(Type::FloatTy);
+
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+
+   Value *xcmp = m_builder.CreateFCmpOLT(vec1[0], vec2[0], name("xcmp"));
+   Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel"));
+
+   Value *ycmp = m_builder.CreateFCmpOLT(vec1[1], vec2[1], name("ycmp"));
+   Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel"));
+
+   Value *zcmp = m_builder.CreateFCmpOLT(vec1[2], vec2[2], name("zcmp"));
+   Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel"));
+
+   Value *wcmp = m_builder.CreateFCmpOLT(vec1[3], vec2[3], name("wcmp"));
+   Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel"));
+
+   return vectorFromVals(x, y, z, w);
+}
+
+llvm::Value * Instructions::sne(llvm::Value *in1, llvm::Value *in2)
+{
+   Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f));
+   Constant *const0f = Constant::getNullValue(Type::FloatTy);
+
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+
+   Value *xcmp = m_builder.CreateFCmpONE(vec1[0], vec2[0], name("xcmp"));
+   Value *x = m_builder.CreateSelect(xcmp, const1f, const0f, name("xsel"));
+
+   Value *ycmp = m_builder.CreateFCmpONE(vec1[1], vec2[1], name("ycmp"));
+   Value *y = m_builder.CreateSelect(ycmp, const1f, const0f, name("ysel"));
+
+   Value *zcmp = m_builder.CreateFCmpONE(vec1[2], vec2[2], name("zcmp"));
+   Value *z = m_builder.CreateSelect(zcmp, const1f, const0f, name("zsel"));
+
+   Value *wcmp = m_builder.CreateFCmpONE(vec1[3], vec2[3], name("wcmp"));
+   Value *w = m_builder.CreateSelect(wcmp, const1f, const0f, name("wsel"));
+
+   return vectorFromVals(x, y, z, w);
+}
+
+llvm::Value * Instructions::str(llvm::Value *in1, llvm::Value *in2)
+{
+   Constant *const1f = ConstantFP::get(APFloat(1.000000e+00f));
+
+   return vectorFromVals(const1f, const1f, const1f, const1f);
+}
+
+llvm::Value * Instructions::sub(llvm::Value *in1, llvm::Value *in2)
+{
+   Value *res = m_builder.CreateSub(in1, in2, name("sub"));
+   return res;
+}
+
+llvm::Value * Instructions::trunc(llvm::Value *in)
+{
+   std::vector<llvm::Value*> vec = extractVector(in);
+   Value *icastx = m_builder.CreateFPToSI(vec[0], IntegerType::get(32),
+                                          name("ftoix"));
+   Value *icasty = m_builder.CreateFPToSI(vec[1], IntegerType::get(32),
+                                          name("ftoiy"));
+   Value *icastz = m_builder.CreateFPToSI(vec[2], IntegerType::get(32),
+                                          name("ftoiz"));
+   Value *icastw = m_builder.CreateFPToSI(vec[3], IntegerType::get(32),
+                                          name("ftoiw"));
+   Value *fx = m_builder.CreateSIToFP(icastx, Type::FloatTy,
+                                      name("fx"));
+   Value *fy = m_builder.CreateSIToFP(icasty, Type::FloatTy,
+                                      name("fy"));
+   Value *fz = m_builder.CreateSIToFP(icastz, Type::FloatTy,
+                                      name("fz"));
+   Value *fw = m_builder.CreateSIToFP(icastw, Type::FloatTy,
+                                      name("fw"));
+   return vectorFromVals(fx, fy, fz, fw);
+}
+
+llvm::Value * Instructions::x2d(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3)
+{
+   std::vector<llvm::Value*> vec1 = extractVector(in1);
+   std::vector<llvm::Value*> vec2 = extractVector(in2);
+   std::vector<llvm::Value*> vec3 = extractVector(in3);
+
+   Value *x2x3 = m_builder.CreateMul( vec2[0], vec3[0], name("x2x3"));
+   Value *y2y3 = m_builder.CreateMul( vec2[1], vec3[1], name("y2y3"));
+   Value *x1px2x3 = m_builder.CreateAdd (vec1[0], x2x3, name("x1 + x2x3"));
+   Value *x1px2x3py2y3 = m_builder.CreateAdd (x1px2x3, y2y3, name("x1 + x2x3 + y2y3"));
+
+   Value *x2z3 = m_builder.CreateMul( vec2[0], vec3[2], name("x2z3"));
+   Value *y2w3 = m_builder.CreateMul( vec2[1], vec3[3], name("y2w3"));
+   Value *y1px2z3 = m_builder.CreateAdd (vec1[1], x2z3, name("y1 + x2z3"));
+   Value *y1px2z3py2w3 = m_builder.CreateAdd (y1px2z3, y2w3, name("y1 + x2z3 + y2w3"));
+
+   return vectorFromVals(x1px2x3py2y3, y1px2z3py2w3, x1px2x3py2y3, y1px2z3py2w3);
+}
+
+void Instructions::printVector(llvm::Value *val)
+{
+   static const char *frmt = "Vector is [%f, %f, %f, %f]\x0A";
+
+   if (!m_fmtPtr) {
+      Constant *format = ConstantArray::get(frmt, true);
+      ArrayType *arrayTy = ArrayType::get(IntegerType::get(8), strlen(frmt) + 1);
+      GlobalVariable* globalFormat = new GlobalVariable(
+         /*Type=*/arrayTy,
+         /*isConstant=*/true,
+         /*Linkage=*/GlobalValue::InternalLinkage,
+         /*Initializer=*/0, // has initializer, specified below
+         /*Name=*/name(".str"),
+         m_mod);
+      globalFormat->setInitializer(format);
+
+      Constant* const_int0 = Constant::getNullValue(IntegerType::get(32));
+      std::vector<Constant*> const_ptr_21_indices;
+      const_ptr_21_indices.push_back(const_int0);
+      const_ptr_21_indices.push_back(const_int0);
+      m_fmtPtr = ConstantExpr::getGetElementPtr(globalFormat,
+                                                &const_ptr_21_indices[0], const_ptr_21_indices.size());
+   }
+
+   Function *func_printf = m_mod->getFunction("printf");
+   if (!func_printf)
+      func_printf = declarePrintf();
+   assert(func_printf);
+   std::vector<llvm::Value*> vec = extractVector(val);
+   Value *dx = m_builder.CreateFPExt(vec[0], Type::DoubleTy, name("dx"));
+   Value *dy = m_builder.CreateFPExt(vec[1], Type::DoubleTy, name("dy"));
+   Value *dz = m_builder.CreateFPExt(vec[2], Type::DoubleTy, name("dz"));
+   Value *dw = m_builder.CreateFPExt(vec[3], Type::DoubleTy, name("dw"));
+   std::vector<Value*> params;
+   params.push_back(m_fmtPtr);
+   params.push_back(dx);
+   params.push_back(dy);
+   params.push_back(dz);
+   params.push_back(dw);
+   CallInst *call = m_builder.CreateCall(func_printf, params.begin(), params.end(),
+                                         name("printf"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(true);
+}
+
+const char * Instructions::name(const char *prefix)
+{
+   ++m_idx;
+   snprintf(m_name, 32, "%s%d", prefix, m_idx);
+   return m_name;
+}
+
+llvm::Value * Instructions::callCeil(llvm::Value *val)
+{
+   if (!m_llvmCeil) {
+      // predeclare the intrinsic
+      std::vector<const Type*> ceilArgs;
+      ceilArgs.push_back(Type::FloatTy);
+      AttrListPtr ceilPal;
+      FunctionType* ceilType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/ceilArgs,
+         /*isVarArg=*/false);
+      m_llvmCeil = Function::Create(
+         /*Type=*/ceilType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"ceilf", m_mod);
+      m_llvmCeil->setCallingConv(CallingConv::C);
+      m_llvmCeil->setAttributes(ceilPal);
+   }
+   CallInst *call =  m_builder.CreateCall(m_llvmCeil, val,
+                                          name("ceilf"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value *Instructions::callFAbs(llvm::Value *val)
+{
+   if (!m_llvmFAbs) {
+      // predeclare the intrinsic
+      std::vector<const Type*> fabsArgs;
+      fabsArgs.push_back(Type::FloatTy);
+      AttrListPtr fabsPal;
+      FunctionType* fabsType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/fabsArgs,
+         /*isVarArg=*/false);
+      m_llvmFAbs = Function::Create(
+         /*Type=*/fabsType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"fabs", m_mod);
+      m_llvmFAbs->setCallingConv(CallingConv::C);
+      m_llvmFAbs->setAttributes(fabsPal);
+   }
+   CallInst *call = m_builder.CreateCall(m_llvmFAbs, val,
+                                         name("fabs"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::callFExp(llvm::Value *val)
+{
+   if (!m_llvmFexp) {
+      // predeclare the intrinsic
+      std::vector<const Type*> fexpArgs;
+      fexpArgs.push_back(Type::FloatTy);
+      AttrListPtr fexpPal;
+      FunctionType* fexpType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/fexpArgs,
+         /*isVarArg=*/false);
+      m_llvmFexp = Function::Create(
+         /*Type=*/fexpType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"expf", m_mod);
+      m_llvmFexp->setCallingConv(CallingConv::C);
+      m_llvmFexp->setAttributes(fexpPal);
+   }
+   CallInst *call = m_builder.CreateCall(m_llvmFexp, val,
+                                         name("expf"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::callFLog(llvm::Value *val)
+{
+   if (!m_llvmFlog) {
+      // predeclare the intrinsic
+      std::vector<const Type*> flogArgs;
+      flogArgs.push_back(Type::FloatTy);
+      AttrListPtr flogPal;
+      FunctionType* flogType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/flogArgs,
+         /*isVarArg=*/false);
+      m_llvmFlog = Function::Create(
+         /*Type=*/flogType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"logf", m_mod);
+      m_llvmFlog->setCallingConv(CallingConv::C);
+      m_llvmFlog->setAttributes(flogPal);
+   }
+   CallInst *call = m_builder.CreateCall(m_llvmFlog, val,
+                                         name("logf"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::callFloor(llvm::Value *val)
+{
+   if (!m_llvmFloor) {
+      // predeclare the intrinsic
+      std::vector<const Type*> floorArgs;
+      floorArgs.push_back(Type::FloatTy);
+      AttrListPtr floorPal;
+      FunctionType* floorType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/floorArgs,
+         /*isVarArg=*/false);
+      m_llvmFloor = Function::Create(
+         /*Type=*/floorType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"floorf", m_mod);
+      m_llvmFloor->setCallingConv(CallingConv::C);
+      m_llvmFloor->setAttributes(floorPal);
+   }
+   CallInst *call =  m_builder.CreateCall(m_llvmFloor, val,
+                                          name("floorf"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value *Instructions::callFSqrt(llvm::Value *val)
+{
+   if (!m_llvmFSqrt) {
+      // predeclare the intrinsic
+      std::vector<const Type*> fsqrtArgs;
+      fsqrtArgs.push_back(Type::FloatTy);
+      AttrListPtr fsqrtPal;
+      FunctionType* fsqrtType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/fsqrtArgs,
+         /*isVarArg=*/false);
+      m_llvmFSqrt = Function::Create(
+         /*Type=*/fsqrtType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"llvm.sqrt.f32", m_mod);
+      m_llvmFSqrt->setCallingConv(CallingConv::C);
+      m_llvmFSqrt->setAttributes(fsqrtPal);
+   }
+   CallInst *call = m_builder.CreateCall(m_llvmFSqrt, val,
+                                         name("sqrt"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::callPow(llvm::Value *val1, llvm::Value *val2)
+{
+   if (!m_llvmPow) {
+      // predeclare the intrinsic
+      std::vector<const Type*> powArgs;
+      powArgs.push_back(Type::FloatTy);
+      powArgs.push_back(Type::FloatTy);
+      AttrListPtr powPal;
+      FunctionType* powType = FunctionType::get(
+         /*Result=*/Type::FloatTy,
+         /*Params=*/powArgs,
+         /*isVarArg=*/false);
+      m_llvmPow = Function::Create(
+         /*Type=*/powType,
+         /*Linkage=*/GlobalValue::ExternalLinkage,
+         /*Name=*/"llvm.pow.f32", m_mod);
+      m_llvmPow->setCallingConv(CallingConv::C);
+      m_llvmPow->setAttributes(powPal);
+   }
+   std::vector<Value*> params;
+   params.push_back(val1);
+   params.push_back(val2);
+   CallInst *call = m_builder.CreateCall(m_llvmPow, params.begin(), params.end(),
+                                         name("pow"));
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+   return call;
+}
+
+llvm::Value * Instructions::vectorFromVals(llvm::Value *x, llvm::Value *y,
+                                           llvm::Value *z, llvm::Value *w)
+{
+   Constant *const_vec = Constant::getNullValue(m_floatVecType);
+   Value *res = m_builder.CreateInsertElement(const_vec, x,
+                                              m_storage->constantInt(0),
+                                              name("vecx"));
+   res = m_builder.CreateInsertElement(res, y, m_storage->constantInt(1),
+                               name("vecxy"));
+   res = m_builder.CreateInsertElement(res, z, m_storage->constantInt(2),
+                               name("vecxyz"));
+   if (w)
+      res = m_builder.CreateInsertElement(res, w, m_storage->constantInt(3),
+                                          name("vecxyzw"));
+   return res;
+}
+
+llvm::Value * Instructions::constVector(float x, float y, float z, float w)
+{
+   std::vector<Constant*> vec(4);
+   vec[0] = ConstantFP::get(APFloat(x));
+   vec[1] = ConstantFP::get(APFloat(y));
+   vec[2] = ConstantFP::get(APFloat(z));
+   vec[3] = ConstantFP::get(APFloat(w));
+   return ConstantVector::get(m_floatVecType, vec);
+}
+
+llvm::Function * Instructions::declarePrintf()
+{
+   std::vector<const Type*> args;
+   AttrListPtr params;
+   FunctionType* funcTy = FunctionType::get(
+      /*Result=*/IntegerType::get(32),
+      /*Params=*/args,
+      /*isVarArg=*/true);
+   Function* func_printf = Function::Create(
+      /*Type=*/funcTy,
+      /*Linkage=*/GlobalValue::ExternalLinkage,
+      /*Name=*/"printf", m_mod);
+   func_printf->setCallingConv(CallingConv::C);
+   func_printf->setAttributes(params);
+   return func_printf;
+}
+
+llvm::Function * Instructions::declareFunc(int label)
+{
+   PointerType *vecPtr = PointerType::getUnqual(m_floatVecType);
+   std::vector<const Type*> args;
+   args.push_back(vecPtr);
+   args.push_back(vecPtr);
+   args.push_back(vecPtr);
+   args.push_back(vecPtr);
+   AttrListPtr params;
+   FunctionType *funcType = FunctionType::get(
+      /*Result=*/Type::VoidTy,
+      /*Params=*/args,
+      /*isVarArg=*/false);
+   std::string name = createFuncName(label);
+   Function *func = Function::Create(
+      /*Type=*/funcType,
+      /*Linkage=*/GlobalValue::ExternalLinkage,
+      /*Name=*/name.c_str(), m_mod);
+   func->setCallingConv(CallingConv::C);
+   func->setAttributes(params);
+   return func;
+}
+
+llvm::Function * Instructions::findFunction(int label)
+{
+   llvm::Function *func = m_functions[label];
+   if (!func) {
+      func = declareFunc(label);
+      m_functions[label] = func;
+   }
+   return func;
+}
+
+std::vector<llvm::Value*> Instructions::extractVector(llvm::Value *vec)
+{
+   std::vector<llvm::Value*> elems(4);
+   elems[0] = m_builder.CreateExtractElement(vec, m_storage->constantInt(0),
+                                             name("x"));
+   elems[1] = m_builder.CreateExtractElement(vec, m_storage->constantInt(1),
+                                             name("y"));
+   elems[2] = m_builder.CreateExtractElement(vec, m_storage->constantInt(2),
+                                             name("z"));
+   elems[3] = m_builder.CreateExtractElement(vec, m_storage->constantInt(3),
+                                             name("w"));
+   return elems;
+}
+
+
+#endif //MESA_LLVM
+
+
diff --git a/src/gallium/auxiliary/gallivm/instructions.h b/src/gallium/auxiliary/gallivm/instructions.h
new file mode 100644
index 00000000000..e18571251ee
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/instructions.h
@@ -0,0 +1,175 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin [email protected]
+  */
+
+#ifndef INSTRUCTIONS_H
+#define INSTRUCTIONS_H
+
+#include <llvm/BasicBlock.h>
+#include <llvm/Module.h>
+#include <llvm/Value.h>
+#include <llvm/Support/IRBuilder.h>
+
+#include <map>
+#include <stack>
+
+namespace llvm {
+   class VectorType;
+   class Function;
+}
+
+class Storage;
+
+class Instructions
+{
+public:
+   Instructions(llvm::Module *mod, llvm::Function *func, llvm::BasicBlock *block,
+                Storage *storage);
+
+   llvm::BasicBlock *currentBlock() const;
+
+   llvm::Value *abs(llvm::Value *in1);
+   llvm::Value *add(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *arl(llvm::Value *in1);
+   void         beginLoop();
+   void         bgnSub(unsigned);
+   void         brk();
+   void         cal(int label, llvm::Value *input);
+   llvm::Value *ceil(llvm::Value *in);
+   llvm::Value *clamp(llvm::Value *in);
+   llvm::Value *cmp(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3);
+   llvm::Value *cnd(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3);
+   llvm::Value *cnd0(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3);
+   llvm::Value *cos(llvm::Value *in);
+   llvm::Value *cross(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *ddx(llvm::Value *in);
+   llvm::Value *ddy(llvm::Value *in);
+   llvm::Value *div(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *dot2add(llvm::Value *in, llvm::Value *in2, llvm::Value *in3);
+   llvm::Value *dp2(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *dp3(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *dp4(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *dph(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *dst(llvm::Value *in1, llvm::Value *in2);
+   void         elseop();
+   void         endif();
+   void         endLoop();
+   void         end();
+   void         endSub();
+   llvm::Value *exp(llvm::Value *in);
+   llvm::Value *ex2(llvm::Value *in);
+   llvm::Value *floor(llvm::Value *in);
+   llvm::Value *frc(llvm::Value *in);
+   void         ifop(llvm::Value *in);
+   llvm::Value *kil(llvm::Value *in);
+   llvm::Value *lerp(llvm::Value *in1, llvm::Value *in2,
+                     llvm::Value *in3);
+   llvm::Value *lg2(llvm::Value *in);
+   llvm::Value *lit(llvm::Value *in);
+   llvm::Value *log(llvm::Value *in);
+   llvm::Value *madd(llvm::Value *in1, llvm::Value *in2,
+                     llvm::Value *in3);
+   llvm::Value *max(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *min(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *mul(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *neg(llvm::Value *in);
+   llvm::Value *nrm(llvm::Value *in);
+   llvm::Value *pow(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *rcp(llvm::Value *in);
+   llvm::Value *rsq(llvm::Value *in);
+   llvm::Value *scs(llvm::Value *in);
+   llvm::Value *seq(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *sfl(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *sge(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *sgt(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *sin(llvm::Value *in);
+   llvm::Value *sle(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *slt(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *sne(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *str(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *sub(llvm::Value *in1, llvm::Value *in2);
+   llvm::Value *trunc(llvm::Value *in);
+   llvm::Value *x2d(llvm::Value *in1, llvm::Value *in2, llvm::Value *in3);
+
+   void printVector(llvm::Value *val);
+private:
+   const char *name(const char *prefix);
+
+   llvm::Value *callCeil(llvm::Value *val);
+   llvm::Value *callFAbs(llvm::Value *val);
+   llvm::Value *callFExp(llvm::Value *val);
+   llvm::Value *callFLog(llvm::Value *val);
+   llvm::Value *callFloor(llvm::Value *val);
+   llvm::Value *callFSqrt(llvm::Value *val);
+   llvm::Value *callPow(llvm::Value *val1, llvm::Value *val2);
+
+   llvm::Value *vectorFromVals(llvm::Value *x, llvm::Value *y,
+                               llvm::Value *z, llvm::Value *w=0);
+
+   llvm::Value *constVector(float x, float y, float z, float w);
+
+   llvm::Function *declarePrintf();
+   llvm::Function *declareFunc(int label);
+
+   llvm::Function *findFunction(int label);
+
+   std::vector<llvm::Value*> extractVector(llvm::Value *vec);
+private:
+   llvm::Module             *m_mod;
+   llvm::Function           *m_func;
+   char                      m_name[32];
+   llvm::IRBuilder<>         m_builder;
+   int                       m_idx;
+
+   llvm::VectorType *m_floatVecType;
+
+   llvm::Function   *m_llvmCeil;
+   llvm::Function   *m_llvmFSqrt;
+   llvm::Function   *m_llvmFAbs;
+   llvm::Function   *m_llvmPow;
+   llvm::Function   *m_llvmFloor;
+   llvm::Function   *m_llvmFlog;
+   llvm::Function   *m_llvmFexp;
+   llvm::Function   *m_llvmLit;
+
+   llvm::Constant   *m_fmtPtr;
+
+   std::stack<llvm::BasicBlock*> m_ifStack;
+   struct Loop {
+      llvm::BasicBlock *begin;
+      llvm::BasicBlock *end;
+   };
+   std::stack<Loop> m_loopStack;
+   std::map<int, llvm::Function*> m_functions;
+   Storage *m_storage;
+};
+
+#endif
diff --git a/src/gallium/auxiliary/gallivm/instructionssoa.cpp b/src/gallium/auxiliary/gallivm/instructionssoa.cpp
new file mode 100644
index 00000000000..925e948763e
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/instructionssoa.cpp
@@ -0,0 +1,523 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+#include "instructionssoa.h"
+
+#include "storagesoa.h"
+
+#include "pipe/p_shader_tokens.h"
+#include "util/u_memory.h"
+
+#include <llvm/CallingConv.h>
+#include <llvm/Constants.h>
+#include <llvm/Module.h>
+#include <llvm/Function.h>
+#include <llvm/Instructions.h>
+#include <llvm/Transforms/Utils/Cloning.h>
+#include <llvm/Attributes.h>
+#include <llvm/Support/MemoryBuffer.h>
+#include <llvm/Bitcode/ReaderWriter.h>
+
+
+#include <iostream>
+
+
+/* disable some warnings. this file is autogenerated */
+#if defined(__GNUC__)
+#pragma GCC diagnostic ignored "-Wunused-variable"
+#endif
+using namespace llvm;
+#include "gallivmsoabuiltins.cpp"
+#if defined(__GNUC__)
+#pragma GCC diagnostic warning "-Wunused-variable"
+#endif
+
+InstructionsSoa::InstructionsSoa(llvm::Module *mod, llvm::Function *func,
+                                 llvm::BasicBlock *block, StorageSoa *storage)
+   : m_builder(block),
+     m_storage(storage),
+     m_idx(0)
+{
+   createFunctionMap();
+   createBuiltins();
+}
+
+const char * InstructionsSoa::name(const char *prefix) const
+{
+   ++m_idx;
+   snprintf(m_name, 32, "%s%d", prefix, m_idx);
+   return m_name;
+}
+
+llvm::Value * InstructionsSoa::vectorFromVals(llvm::Value *x, llvm::Value *y,
+                                              llvm::Value *z, llvm::Value *w)
+{
+   VectorType  *vectorType = VectorType::get(Type::FloatTy, 4);
+   Constant *constVector = Constant::getNullValue(vectorType);
+   Value *res = m_builder.CreateInsertElement(constVector, x,
+                                              m_storage->constantInt(0),
+                                              name("vecx"));
+   res = m_builder.CreateInsertElement(res, y, m_storage->constantInt(1),
+                               name("vecxy"));
+   res = m_builder.CreateInsertElement(res, z, m_storage->constantInt(2),
+                               name("vecxyz"));
+   if (w)
+      res = m_builder.CreateInsertElement(res, w, m_storage->constantInt(3),
+                                          name("vecxyzw"));
+   return res;
+}
+
+void InstructionsSoa::end()
+{
+   m_builder.CreateRetVoid();
+}
+
+std::vector<llvm::Value*> InstructionsSoa::extractVector(llvm::Value *vector)
+{
+   std::vector<llvm::Value*> res(4);
+   res[0] = m_builder.CreateExtractElement(vector,
+                                           m_storage->constantInt(0),
+                                           name("extract1X"));
+   res[1] = m_builder.CreateExtractElement(vector,
+                                           m_storage->constantInt(1),
+                                           name("extract2X"));
+   res[2] = m_builder.CreateExtractElement(vector,
+                                           m_storage->constantInt(2),
+                                           name("extract3X"));
+   res[3] = m_builder.CreateExtractElement(vector,
+                                           m_storage->constantInt(3),
+                                           name("extract4X"));
+
+   return res;
+}
+
+llvm::IRBuilder<>* InstructionsSoa::getIRBuilder()
+{
+   return &m_builder;
+}
+
+void InstructionsSoa::createFunctionMap()
+{
+   m_functionsMap[TGSI_OPCODE_ABS]   = "abs";
+   m_functionsMap[TGSI_OPCODE_DP3]   = "dp3";
+   m_functionsMap[TGSI_OPCODE_DP4]   = "dp4";
+   m_functionsMap[TGSI_OPCODE_MIN]   = "min";
+   m_functionsMap[TGSI_OPCODE_MAX]   = "max";
+   m_functionsMap[TGSI_OPCODE_POWER] = "pow";
+   m_functionsMap[TGSI_OPCODE_LIT]   = "lit";
+   m_functionsMap[TGSI_OPCODE_RSQ]   = "rsq";
+   m_functionsMap[TGSI_OPCODE_SLT]   = "slt";
+}
+
+void InstructionsSoa::createDependencies()
+{
+   {
+      std::vector<std::string> powDeps(2);
+      powDeps[0] = "powf";
+      powDeps[1] = "powvec";
+      m_builtinDependencies["pow"] = powDeps;
+   }
+   {
+      std::vector<std::string> absDeps(2);
+      absDeps[0] = "fabsf";
+      absDeps[1] = "absvec";
+      m_builtinDependencies["abs"] = absDeps;
+   }
+   {
+      std::vector<std::string> maxDeps(1);
+      maxDeps[0] = "maxvec";
+      m_builtinDependencies["max"] = maxDeps;
+   }
+   {
+      std::vector<std::string> minDeps(1);
+      minDeps[0] = "minvec";
+      m_builtinDependencies["min"] = minDeps;
+   }
+   {
+      std::vector<std::string> litDeps(4);
+      litDeps[0] = "minvec";
+      litDeps[1] = "maxvec";
+      litDeps[2] = "powf";
+      litDeps[3] = "powvec";
+      m_builtinDependencies["lit"] = litDeps;
+   }
+   {
+      std::vector<std::string> rsqDeps(4);
+      rsqDeps[0] = "sqrtf";
+      rsqDeps[1] = "sqrtvec";
+      rsqDeps[2] = "fabsf";
+      rsqDeps[3] = "absvec";
+      m_builtinDependencies["rsq"] = rsqDeps;
+   }
+}
+
+llvm::Function * InstructionsSoa::function(int op)
+{
+    if (m_functions.find(op) != m_functions.end())
+       return m_functions[op];
+
+    std::string name = m_functionsMap[op];
+
+    std::cout <<"For op = "<<op<<", func is '"<<name<<"'"<<std::endl;
+
+    std::vector<std::string> deps = m_builtinDependencies[name];
+    for (unsigned int i = 0; i < deps.size(); ++i) {
+       llvm::Function *func = m_builtins->getFunction(deps[i]);
+       std::cout <<"\tinjecting dep = '"<<func->getName()<<"'"<<std::endl;
+       injectFunction(func);
+    }
+
+    llvm::Function *originalFunc = m_builtins->getFunction(name);
+    injectFunction(originalFunc, op);
+    return m_functions[op];
+}
+
+llvm::Module * InstructionsSoa::currentModule() const
+{
+   BasicBlock *block = m_builder.GetInsertBlock();
+   if (!block || !block->getParent())
+      return 0;
+
+   return block->getParent()->getParent();
+}
+
+void InstructionsSoa::createBuiltins()
+{
+   std::string ErrMsg;
+   MemoryBuffer *buffer = MemoryBuffer::getMemBuffer(
+      (const char*)&soabuiltins_data[0],
+      (const char*)&soabuiltins_data[Elements(soabuiltins_data) - 1]);
+   m_builtins = ParseBitcodeFile(buffer, &ErrMsg);
+   std::cout<<"Builtins created at "<<m_builtins<<" ("<<ErrMsg<<")"<<std::endl;
+   assert(m_builtins);
+   createDependencies();
+}
+
+
+std::vector<llvm::Value*> InstructionsSoa::abs(const std::vector<llvm::Value*> in1)
+{
+   llvm::Function *func = function(TGSI_OPCODE_ABS);
+   return callBuiltin(func, in1);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::add(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = m_builder.CreateAdd(in1[0], in2[0], name("addx"));
+   res[1] = m_builder.CreateAdd(in1[1], in2[1], name("addy"));
+   res[2] = m_builder.CreateAdd(in1[2], in2[2], name("addz"));
+   res[3] = m_builder.CreateAdd(in1[3], in2[3], name("addw"));
+
+   return res;
+}
+
+std::vector<llvm::Value*> InstructionsSoa::arl(const std::vector<llvm::Value*> in)
+{
+   std::vector<llvm::Value*> res(4);
+
+   //Extract x's
+   llvm::Value *x1 = m_builder.CreateExtractElement(in[0],
+                                                    m_storage->constantInt(0),
+                                                    name("extractX"));
+   //cast it to an unsigned int
+   x1 = m_builder.CreateFPToUI(x1, IntegerType::get(32), name("x1IntCast"));
+
+   res[0] = x1;//vectorFromVals(x1, x2, x3, x4);
+   //only x is valid. the others shouldn't be necessary
+   /*
+   res[1] = Constant::getNullValue(m_floatVecType);
+   res[2] = Constant::getNullValue(m_floatVecType);
+   res[3] = Constant::getNullValue(m_floatVecType);
+   */
+
+   return res;
+}
+
+std::vector<llvm::Value*> InstructionsSoa::dp3(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   llvm::Function *func = function(TGSI_OPCODE_DP3);
+   return callBuiltin(func, in1, in2);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::lit(const std::vector<llvm::Value*> in)
+{
+   llvm::Function *func = function(TGSI_OPCODE_LIT);
+   return callBuiltin(func, in);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::madd(const std::vector<llvm::Value*> in1,
+                                                const std::vector<llvm::Value*> in2,
+                                                const std::vector<llvm::Value*> in3)
+{
+   std::vector<llvm::Value*> res = mul(in1, in2);
+   return add(res, in3);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::max(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   llvm::Function *func = function(TGSI_OPCODE_MAX);
+   return callBuiltin(func, in1, in2);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::min(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   llvm::Function *func = function(TGSI_OPCODE_MIN);
+   return callBuiltin(func, in1, in2);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::mul(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = m_builder.CreateMul(in1[0], in2[0], name("mulx"));
+   res[1] = m_builder.CreateMul(in1[1], in2[1], name("muly"));
+   res[2] = m_builder.CreateMul(in1[2], in2[2], name("mulz"));
+   res[3] = m_builder.CreateMul(in1[3], in2[3], name("mulw"));
+
+   return res;
+}
+
+std::vector<llvm::Value*> InstructionsSoa::pow(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   llvm::Function *func = function(TGSI_OPCODE_POWER);
+   return callBuiltin(func, in1, in2);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::rsq(const std::vector<llvm::Value*> in)
+{
+   llvm::Function *func = function(TGSI_OPCODE_RSQ);
+   return callBuiltin(func, in);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::slt(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   llvm::Function *func = function(TGSI_OPCODE_SLT);
+   return callBuiltin(func, in1, in2);
+}
+
+std::vector<llvm::Value*> InstructionsSoa::sub(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = m_builder.CreateSub(in1[0], in2[0], name("subx"));
+   res[1] = m_builder.CreateSub(in1[1], in2[1], name("suby"));
+   res[2] = m_builder.CreateSub(in1[2], in2[2], name("subz"));
+   res[3] = m_builder.CreateSub(in1[3], in2[3], name("subw"));
+
+   return res;
+}
+
+void checkFunction(Function *func)
+{
+   for (Function::const_iterator BI = func->begin(), BE = func->end();
+        BI != BE; ++BI) {
+      const BasicBlock &BB = *BI;
+      for (BasicBlock::const_iterator II = BB.begin(), IE = BB.end();
+           II != IE; ++II) {
+         const Instruction &I = *II;
+         std::cout<< "Instr = "<<I;
+         for (unsigned op = 0, E = I.getNumOperands(); op != E; ++op) {
+            const Value *Op = I.getOperand(op);
+            std::cout<< "\top = "<<Op<<"("<<op<<")"<<std::endl;
+            //I->setOperand(op, V);
+  }
+      }
+   }
+}
+
+llvm::Value * InstructionsSoa::allocaTemp()
+{
+   VectorType *vector   = VectorType::get(Type::FloatTy, 4);
+   ArrayType  *vecArray = ArrayType::get(vector, 4);
+   AllocaInst *alloca = new AllocaInst(vecArray, name("tmpRes"),
+                                       m_builder.GetInsertBlock());
+
+   std::vector<Value*> indices;
+   indices.push_back(m_storage->constantInt(0));
+   indices.push_back(m_storage->constantInt(0));
+   GetElementPtrInst *getElem = GetElementPtrInst::Create(alloca,
+                                                          indices.begin(),
+                                                          indices.end(),
+                                                          name("allocaPtr"),
+                                                          m_builder.GetInsertBlock());
+   return getElem;
+}
+
+std::vector<llvm::Value*> InstructionsSoa::allocaToResult(llvm::Value *allocaPtr)
+{
+   GetElementPtrInst *xElemPtr =  GetElementPtrInst::Create(allocaPtr,
+                                                            m_storage->constantInt(0),
+                                                            name("xPtr"),
+                                                            m_builder.GetInsertBlock());
+   GetElementPtrInst *yElemPtr =  GetElementPtrInst::Create(allocaPtr,
+                                                            m_storage->constantInt(1),
+                                                            name("yPtr"),
+                                                            m_builder.GetInsertBlock());
+   GetElementPtrInst *zElemPtr =  GetElementPtrInst::Create(allocaPtr,
+                                                            m_storage->constantInt(2),
+                                                            name("zPtr"),
+                                                            m_builder.GetInsertBlock());
+   GetElementPtrInst *wElemPtr =  GetElementPtrInst::Create(allocaPtr,
+                                                            m_storage->constantInt(3),
+                                                            name("wPtr"),
+                                                            m_builder.GetInsertBlock());
+
+   std::vector<llvm::Value*> res(4);
+   res[0] = new LoadInst(xElemPtr, name("xRes"), false, m_builder.GetInsertBlock());
+   res[1] = new LoadInst(yElemPtr, name("yRes"), false, m_builder.GetInsertBlock());
+   res[2] = new LoadInst(zElemPtr, name("zRes"), false, m_builder.GetInsertBlock());
+   res[3] = new LoadInst(wElemPtr, name("wRes"), false, m_builder.GetInsertBlock());
+
+   return res;
+}
+
+std::vector<llvm::Value*> InstructionsSoa::dp4(const std::vector<llvm::Value*> in1,
+                                               const std::vector<llvm::Value*> in2)
+{
+   llvm::Function *func = function(TGSI_OPCODE_DP4);
+   return callBuiltin(func, in1, in2);
+}
+
+std::vector<Value*> InstructionsSoa::callBuiltin(llvm::Function *func, const std::vector<llvm::Value*> in1)
+{
+   std::vector<Value*> params;
+
+   llvm::Value *allocaPtr = allocaTemp();
+   params.push_back(allocaPtr);
+   params.push_back(in1[0]);
+   params.push_back(in1[1]);
+   params.push_back(in1[2]);
+   params.push_back(in1[3]);
+   CallInst *call = m_builder.CreateCall(func, params.begin(), params.end());
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+
+   return allocaToResult(allocaPtr);
+}
+
+std::vector<Value*> InstructionsSoa::callBuiltin(llvm::Function *func, const std::vector<llvm::Value*> in1,
+                                                 const std::vector<llvm::Value*> in2)
+{
+   std::vector<Value*> params;
+
+   llvm::Value *allocaPtr = allocaTemp();
+   params.push_back(allocaPtr);
+   params.push_back(in1[0]);
+   params.push_back(in1[1]);
+   params.push_back(in1[2]);
+   params.push_back(in1[3]);
+   params.push_back(in2[0]);
+   params.push_back(in2[1]);
+   params.push_back(in2[2]);
+   params.push_back(in2[3]);
+   CallInst *call = m_builder.CreateCall(func, params.begin(), params.end());
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+
+   return allocaToResult(allocaPtr);
+}
+
+std::vector<Value*> InstructionsSoa::callBuiltin(llvm::Function *func, const std::vector<llvm::Value*> in1,
+                                                 const std::vector<llvm::Value*> in2,
+                                                 const std::vector<llvm::Value*> in3)
+{
+   std::vector<Value*> params;
+
+   llvm::Value *allocaPtr = allocaTemp();
+   params.push_back(allocaPtr);
+   params.push_back(in1[0]);
+   params.push_back(in1[1]);
+   params.push_back(in1[2]);
+   params.push_back(in1[3]);
+   params.push_back(in2[0]);
+   params.push_back(in2[1]);
+   params.push_back(in2[2]);
+   params.push_back(in2[3]);
+   params.push_back(in3[0]);
+   params.push_back(in3[1]);
+   params.push_back(in3[2]);
+   params.push_back(in3[3]);
+   CallInst *call = m_builder.CreateCall(func, params.begin(), params.end());
+   call->setCallingConv(CallingConv::C);
+   call->setTailCall(false);
+
+   return allocaToResult(allocaPtr);
+}
+
+void InstructionsSoa::injectFunction(llvm::Function *originalFunc, int op)
+{
+   assert(originalFunc);
+   std::cout << "injecting function originalFunc " <<originalFunc->getName() <<std::endl;
+   if (op != TGSI_OPCODE_LAST) {
+      /* in this case it's possible the function has been already
+       * injected as part of the dependency chain, which gets
+       * injected below */
+      llvm::Function *func = currentModule()->getFunction(originalFunc->getName());
+      if (func) {
+         m_functions[op] = func;
+         return;
+      }
+   }
+   llvm::Function *func = 0;
+   if (originalFunc->isDeclaration()) {
+      func = Function::Create(originalFunc->getFunctionType(), GlobalValue::ExternalLinkage,
+                              originalFunc->getName(), currentModule());
+      func->setCallingConv(CallingConv::C);
+      const AttrListPtr pal;
+      func->setAttributes(pal);
+      currentModule()->dump();
+   } else {
+      DenseMap<const Value*, Value *> val;
+      val[m_builtins->getFunction("fabsf")] = currentModule()->getFunction("fabsf");
+      val[m_builtins->getFunction("powf")] = currentModule()->getFunction("powf");
+      val[m_builtins->getFunction("sqrtf")] = currentModule()->getFunction("sqrtf");
+      func = CloneFunction(originalFunc, val);
+#if 0
+      std::cout <<" replacing "<<m_builtins->getFunction("powf")
+                <<", with " <<currentModule()->getFunction("powf")<<std::endl;
+      std::cout<<"1111-------------------------------"<<std::endl;
+      checkFunction(originalFunc);
+      std::cout<<"2222-------------------------------"<<std::endl;
+      checkFunction(func);
+      std::cout <<"XXXX = " <<val[m_builtins->getFunction("powf")]<<std::endl;
+#endif
+      currentModule()->getFunctionList().push_back(func);
+   }
+   if (op != TGSI_OPCODE_LAST) {
+      m_functions[op] = func;
+   }
+}
+
+
diff --git a/src/gallium/auxiliary/gallivm/instructionssoa.h b/src/gallium/auxiliary/gallivm/instructionssoa.h
new file mode 100644
index 00000000000..d6831e0a6b9
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/instructionssoa.h
@@ -0,0 +1,116 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#ifndef INSTRUCTIONSSOA_H
+#define INSTRUCTIONSSOA_H
+
+#include <pipe/p_shader_tokens.h>
+#include <llvm/Support/IRBuilder.h>
+
+#include <map>
+#include <vector>
+
+namespace llvm {
+   class Module;
+   class Function;
+   class BasicBlock;
+   class Value;
+}
+class StorageSoa;
+
+class InstructionsSoa
+{
+public:
+   InstructionsSoa(llvm::Module *mod, llvm::Function *func,
+                   llvm::BasicBlock *block, StorageSoa *storage);
+
+   std::vector<llvm::Value*> abs(const std::vector<llvm::Value*> in1);
+   std::vector<llvm::Value*> arl(const std::vector<llvm::Value*> in);
+   std::vector<llvm::Value*> add(const std::vector<llvm::Value*> in1,
+                                 const std::vector<llvm::Value*> in2);
+   std::vector<llvm::Value*> dp3(const std::vector<llvm::Value*> in1,
+                                 const std::vector<llvm::Value*> in2);
+   std::vector<llvm::Value*> dp4(const std::vector<llvm::Value*> in1,
+                                 const std::vector<llvm::Value*> in2);
+   std::vector<llvm::Value*> lit(const std::vector<llvm::Value*> in);
+   std::vector<llvm::Value*> madd(const std::vector<llvm::Value*> in1,
+                                  const std::vector<llvm::Value*> in2,
+                                  const std::vector<llvm::Value*> in3);
+   std::vector<llvm::Value*> max(const std::vector<llvm::Value*> in1,
+                                 const std::vector<llvm::Value*> in2);
+   std::vector<llvm::Value*> min(const std::vector<llvm::Value*> in1,
+                                 const std::vector<llvm::Value*> in2);
+   std::vector<llvm::Value*> mul(const std::vector<llvm::Value*> in1,
+                                 const std::vector<llvm::Value*> in2);
+   std::vector<llvm::Value*> pow(const std::vector<llvm::Value*> in1,
+                                 const std::vector<llvm::Value*> in2);
+   std::vector<llvm::Value*> rsq(const std::vector<llvm::Value*> in1);
+   std::vector<llvm::Value*> slt(const std::vector<llvm::Value*> in1,
+                                 const std::vector<llvm::Value*> in2);
+   std::vector<llvm::Value*> sub(const std::vector<llvm::Value*> in1,
+                                 const std::vector<llvm::Value*> in2);
+   void         end();
+
+   std::vector<llvm::Value*> extractVector(llvm::Value *vector);
+   llvm::IRBuilder<>*  getIRBuilder();
+private:
+   const char * name(const char *prefix) const;
+   llvm::Value *vectorFromVals(llvm::Value *x, llvm::Value *y,
+                               llvm::Value *z, llvm::Value *w);
+   void createFunctionMap();
+   void createBuiltins();
+   void createDependencies();
+   llvm::Function *function(int);
+   llvm::Module *currentModule() const;
+   llvm::Value *allocaTemp();
+   std::vector<llvm::Value*> allocaToResult(llvm::Value *allocaPtr);
+   std::vector<llvm::Value*> callBuiltin(llvm::Function *func,
+                                         const std::vector<llvm::Value*> in1);
+   std::vector<llvm::Value*> callBuiltin(llvm::Function *func,
+                                         const std::vector<llvm::Value*> in1,
+                                         const std::vector<llvm::Value*> in2);
+   std::vector<llvm::Value*> callBuiltin(llvm::Function *func,
+                                         const std::vector<llvm::Value*> in1,
+                                         const std::vector<llvm::Value*> in2,
+                                         const std::vector<llvm::Value*> in3);
+   void injectFunction(llvm::Function *originalFunc, int op = TGSI_OPCODE_LAST);
+private:
+   llvm::IRBuilder<>  m_builder;
+   StorageSoa *m_storage;
+
+   std::map<int, std::string> m_functionsMap;
+   std::map<int, llvm::Function*> m_functions;
+   llvm::Module *m_builtins;
+   std::map<std::string, std::vector<std::string> > m_builtinDependencies;
+
+private:
+   mutable int  m_idx;
+   mutable char m_name[32];
+};
+
+
+#endif
diff --git a/src/gallium/auxiliary/gallivm/llvm_builtins.c b/src/gallium/auxiliary/gallivm/llvm_builtins.c
new file mode 100644
index 00000000000..d5a003a48b2
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/llvm_builtins.c
@@ -0,0 +1,114 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin [email protected]
+  */
+typedef __attribute__(( ext_vector_type(4) )) float float4;
+
+extern float powf(float a, float b);
+
+inline float approx(float a, float b)
+{
+    if (b < -128.0f) b = -128.0f;
+    if (b > 128.0f)   b = 128.0f;
+    if (a < 0) a = 0;
+    return powf(a, b);
+}
+
+inline float4 lit(float4 tmp)
+{
+    float4 result;
+    result.x = 1.0;
+    result.w = 1.0;
+    if (tmp.x > 0) {
+        result.y = tmp.x;
+        result.z = approx(tmp.y, tmp.w);
+    } else {
+        result.y = 0;
+        result.z = 0;
+    }
+    return result;
+}
+
+inline float4 cmp(float4 tmp0, float4 tmp1, float4 tmp2)
+{
+   float4 result;
+
+   result.x = (tmp0.x < 0.0) ? tmp1.x : tmp2.x;
+   result.y = (tmp0.y < 0.0) ? tmp1.y : tmp2.y;
+   result.z = (tmp0.z < 0.0) ? tmp1.z : tmp2.z;
+   result.w = (tmp0.w < 0.0) ? tmp1.w : tmp2.w;
+
+   return result;
+}
+
+extern float cosf(float  val);
+extern float sinf(float  val);
+
+inline float4 vcos(float4 val)
+{
+   float4 result;
+   printf("VEC IN   is %f %f %f %f\n", val.x, val.y, val.z, val.w);
+   result.x = cosf(val.x);
+   result.y = cosf(val.x);
+   result.z = cosf(val.x);
+   result.w = cosf(val.x);
+   printf("VEC OUT  is %f %f %f %f\n", result.x, result.y, result.z, result.w);
+   return result;
+}
+
+inline float4 scs(float4 val)
+{
+   float4 result;
+   float tmp = val.x;
+   result.x = cosf(tmp);
+   result.y = sinf(tmp);
+   return result;
+}
+
+
+inline float4 vsin(float4 val)
+{
+   float4 result;
+   float tmp = val.x;
+   float res = sinf(tmp);
+   result.x = res;
+   result.y = res;
+   result.z = res;
+   result.w = res;
+   return result;
+}
+
+inline int kil(float4 val)
+{
+   if (val.x < 0 || val.y < 0 || val.z < 0 || val.w < 0)
+      return 1;
+   else
+      return 0;
+}
diff --git a/src/gallium/auxiliary/gallivm/loweringpass.cpp b/src/gallium/auxiliary/gallivm/loweringpass.cpp
new file mode 100644
index 00000000000..556dbec3661
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/loweringpass.cpp
@@ -0,0 +1,17 @@
+#include "loweringpass.h"
+
+using namespace llvm;
+
+char LoweringPass::ID = 0;
+RegisterPass<LoweringPass> X("lowering", "Lowering Pass");
+
+LoweringPass::LoweringPass()
+   :  ModulePass((intptr_t)&ID)
+{
+}
+
+bool LoweringPass::runOnModule(Module &m)
+{
+   llvm::cerr << "Hello: " << m.getModuleIdentifier() << "\n";
+   return false;
+}
diff --git a/src/gallium/auxiliary/gallivm/loweringpass.h b/src/gallium/auxiliary/gallivm/loweringpass.h
new file mode 100644
index 00000000000..f62dcf6ba73
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/loweringpass.h
@@ -0,0 +1,15 @@
+#ifndef LOWERINGPASS_H
+#define LOWERINGPASS_H
+
+#include "llvm/Pass.h"
+#include "llvm/Module.h"
+
+struct LoweringPass : public llvm::ModulePass
+{
+   static char ID;
+   LoweringPass();
+
+   virtual bool runOnModule(llvm::Module &m);
+};
+
+#endif
diff --git a/src/gallium/auxiliary/gallivm/soabuiltins.c b/src/gallium/auxiliary/gallivm/soabuiltins.c
new file mode 100644
index 00000000000..cb85e1734ec
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/soabuiltins.c
@@ -0,0 +1,210 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * This file is compiled with clang into the LLVM bitcode
+  *
+  * Authors:
+  *   Zack Rusin [email protected]
+  */
+typedef __attribute__(( ext_vector_type(4) )) float float4;
+
+
+extern float fabsf(float val);
+
+/* helpers */
+
+float4 absvec(float4 vec)
+{
+   float4 res;
+   res.x = fabsf(vec.x);
+   res.y = fabsf(vec.y);
+   res.z = fabsf(vec.z);
+   res.w = fabsf(vec.w);
+
+   return res;
+}
+
+float4 maxvec(float4 a, float4 b)
+{
+   return (float4){(a.x > b.x) ? a.x : b.x,
+         (a.y > b.y) ? a.y : b.y,
+         (a.z > b.z) ? a.z : b.z,
+         (a.w > b.w) ? a.w : b.w};
+}
+
+float4 minvec(float4 a, float4 b)
+{
+   return (float4){(a.x < b.x) ? a.x : b.x,
+         (a.y < b.y) ? a.y : b.y,
+         (a.z < b.z) ? a.z : b.z,
+         (a.w < b.w) ? a.w : b.w};
+}
+
+extern float powf(float num, float p);
+extern float sqrtf(float x);
+
+float4 powvec(float4 vec, float4 q)
+{
+   float4 p;
+   p.x = powf(vec.x, q.x);
+   p.y = powf(vec.y, q.y);
+   p.z = powf(vec.z, q.z);
+   p.w = powf(vec.w, q.w);
+   return p;
+}
+
+float4 sqrtvec(float4 vec)
+{
+   float4 p;
+   p.x = sqrtf(vec.x);
+   p.y = sqrtf(vec.y);
+   p.z = sqrtf(vec.z);
+   p.w = sqrtf(vec.w);
+   return p;
+}
+
+float4 sltvec(float4 v1, float4 v2)
+{
+   float4 p;
+   p.x = (v1.x < v2.x) ? 1.0 : 0.0;
+   p.y = (v1.y < v2.y) ? 1.0 : 0.0;
+   p.z = (v1.z < v2.z) ? 1.0 : 0.0;
+   p.w = (v1.w < v2.w) ? 1.0 : 0.0;
+   return p;
+}
+
+
+/* instructions */
+
+void abs(float4 *res,
+         float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w)
+{
+   res[0] = absvec(tmp0x);
+   res[1] = absvec(tmp0y);
+   res[2] = absvec(tmp0z);
+   res[3] = absvec(tmp0w);
+}
+
+void dp3(float4 *res,
+         float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w,
+         float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w)
+{
+   float4 dot = (tmp0x * tmp1x) + (tmp0y * tmp1y) +
+                (tmp0z * tmp1z);
+
+   res[0] = dot;
+   res[1] = dot;
+   res[2] = dot;
+   res[3] = dot;
+}
+
+void dp4(float4 *res,
+         float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w,
+         float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w)
+{
+   float4 dot = (tmp0x * tmp1x) + (tmp0y * tmp1y) +
+                (tmp0z * tmp1z) + (tmp0w * tmp1w);
+
+   res[0] = dot;
+   res[1] = dot;
+   res[2] = dot;
+   res[3] = dot;
+}
+
+void lit(float4 *res,
+         float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w)
+{
+   const float4 zerovec = (float4) {0.0, 0.0, 0.0, 0.0};
+   const float4 min128 = (float4) {-128.f, -128.f, -128.f, -128.f};
+   const float4 plus128 = (float4) {128.f,  128.f,  128.f,  128.f};
+
+   res[0] = (float4){1.0, 1.0, 1.0, 1.0};
+   if (tmp0x.x > 0) {
+      float4 tmpy = maxvec(tmp0y, zerovec);
+      float4 tmpw = minvec(tmp0w, plus128);
+      tmpw = maxvec(tmpw, min128);
+      res[1] = tmp0x;
+      res[2] = powvec(tmpy, tmpw);
+   } else {
+      res[1] = zerovec;
+      res[2] = zerovec;
+   }
+   res[3] = (float4){1.0, 1.0, 1.0, 1.0};
+}
+
+void min(float4 *res,
+         float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w,
+         float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w)
+{
+   res[0] = minvec(tmp0x, tmp1x);
+   res[1] = minvec(tmp0y, tmp1y);
+   res[2] = minvec(tmp0z, tmp1z);
+   res[3] = minvec(tmp0w, tmp1w);
+}
+
+
+void max(float4 *res,
+         float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w,
+         float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w)
+{
+   res[0] = maxvec(tmp0x, tmp1x);
+   res[1] = maxvec(tmp0y, tmp1y);
+   res[2] = maxvec(tmp0z, tmp1z);
+   res[3] = maxvec(tmp0w, tmp1w);
+}
+
+void pow(float4 *res,
+         float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w,
+         float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w)
+{
+   res[0] = powvec(tmp0x, tmp1x);
+   res[1] = res[0];
+   res[2] = res[0];
+   res[3] = res[0];
+}
+
+void rsq(float4 *res,
+         float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w)
+{
+   const float4 onevec = (float4) {1., 1., 1., 1.};
+   res[0] = onevec/sqrtvec(absvec(tmp0x));
+   res[1] = onevec/sqrtvec(absvec(tmp0y));
+   res[2] = onevec/sqrtvec(absvec(tmp0z));
+   res[3] = onevec/sqrtvec(absvec(tmp0w));
+}
+
+void slt(float4 *res,
+         float4 tmp0x, float4 tmp0y, float4 tmp0z, float4 tmp0w,
+         float4 tmp1x, float4 tmp1y, float4 tmp1z, float4 tmp1w)
+{
+   res[0] = sltvec(tmp0x, tmp1x);
+   res[1] = sltvec(tmp0y, tmp1y);
+   res[2] = sltvec(tmp0z, tmp1z);
+   res[3] = sltvec(tmp0w, tmp1w);
+}
+
diff --git a/src/gallium/auxiliary/gallivm/storage.cpp b/src/gallium/auxiliary/gallivm/storage.cpp
new file mode 100644
index 00000000000..73df24c9769
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/storage.cpp
@@ -0,0 +1,364 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin [email protected]
+  */
+#ifdef MESA_LLVM
+
+#include "storage.h"
+
+#include "gallivm_p.h"
+
+#include "pipe/p_shader_tokens.h"
+#include <llvm/BasicBlock.h>
+#include <llvm/Module.h>
+#include <llvm/Value.h>
+
+#include <llvm/CallingConv.h>
+#include <llvm/Constants.h>
+#include <llvm/DerivedTypes.h>
+#include <llvm/InstrTypes.h>
+#include <llvm/Instructions.h>
+
+using namespace llvm;
+
+Storage::Storage(llvm::BasicBlock *block, llvm::Value *input)
+   : m_block(block),
+     m_INPUT(input),
+     m_addrs(32),
+     m_idx(0)
+{
+   m_floatVecType = VectorType::get(Type::FloatTy, 4);
+   m_intVecType   = VectorType::get(IntegerType::get(32), 4);
+
+   m_undefFloatVec = UndefValue::get(m_floatVecType);
+   m_undefIntVec   = UndefValue::get(m_intVecType);
+   m_extSwizzleVec = 0;
+
+   m_numConsts = 0;
+}
+
+//can only build vectors with all members in the [0, 9] range
+llvm::Constant *Storage::shuffleMask(int vec)
+{
+   if (!m_extSwizzleVec) {
+      std::vector<Constant*> elems;
+      elems.push_back(ConstantFP::get(APFloat(0.f)));
+      elems.push_back(ConstantFP::get(APFloat(1.f)));
+      elems.push_back(ConstantFP::get(APFloat(0.f)));
+      elems.push_back(ConstantFP::get(APFloat(1.f)));
+      m_extSwizzleVec = ConstantVector::get(m_floatVecType, elems);
+   }
+
+   if (m_intVecs.find(vec) != m_intVecs.end()) {
+      return m_intVecs[vec];
+   }
+   int origVec = vec;
+   Constant* const_vec = 0;
+   if (origVec == 0) {
+      const_vec = Constant::getNullValue(m_intVecType);
+   } else {
+      int x = gallivm_x_swizzle(vec);
+      int y = gallivm_y_swizzle(vec);
+      int z = gallivm_z_swizzle(vec);
+      int w = gallivm_w_swizzle(vec);
+      std::vector<Constant*> elems;
+      elems.push_back(constantInt(x));
+      elems.push_back(constantInt(y));
+      elems.push_back(constantInt(z));
+      elems.push_back(constantInt(w));
+      const_vec = ConstantVector::get(m_intVecType, elems);
+   }
+
+   m_intVecs[origVec] = const_vec;
+   return const_vec;
+}
+
+llvm::ConstantInt *Storage::constantInt(int idx)
+{
+   if (m_constInts.find(idx) != m_constInts.end()) {
+      return m_constInts[idx];
+   }
+   ConstantInt *const_int = ConstantInt::get(APInt(32,  idx));
+   m_constInts[idx] = const_int;
+   return const_int;
+}
+
+llvm::Value *Storage::inputElement(int idx, llvm::Value *indIdx)
+{
+   Value *val = element(InputsArg, idx, indIdx);
+   LoadInst *load = new LoadInst(val, name("input"), false, m_block);
+   load->setAlignment(8);
+
+   return load;
+}
+
+llvm::Value *Storage::constElement(int idx, llvm::Value *indIdx)
+{
+   m_numConsts = ((idx + 1) > m_numConsts) ? (idx + 1) : m_numConsts;
+
+   Value *elem = element(ConstsArg, idx, indIdx);
+   LoadInst *load = new LoadInst(elem, name("const"), false, m_block);
+   load->setAlignment(8);
+   return load;
+}
+
+llvm::Value *Storage::shuffleVector(llvm::Value *vec, int shuffle)
+{
+   Constant *mask = shuffleMask(shuffle);
+   ShuffleVectorInst *res =
+      new ShuffleVectorInst(vec, m_extSwizzleVec, mask,
+                            name("shuffle"), m_block);
+   return res;
+}
+
+
+llvm::Value *Storage::tempElement(int idx, llvm::Value *indIdx)
+{
+   Value *elem = element(TempsArg, idx, indIdx);
+
+   LoadInst *load = new LoadInst(elem, name("temp"), false, m_block);
+   load->setAlignment(8);
+
+   return load;
+}
+
+void Storage::setTempElement(int idx, llvm::Value *val, int mask)
+{
+   if (mask != TGSI_WRITEMASK_XYZW) {
+      llvm::Value *templ = 0;
+      if (m_tempWriteMap[idx])
+         templ = tempElement(idx);
+      val = maskWrite(val, mask, templ);
+   }
+   Value *elem = element(TempsArg, idx);
+   StoreInst *st = new StoreInst(val, elem, false, m_block);
+   st->setAlignment(8);
+   m_tempWriteMap[idx] = true;
+}
+
+void Storage::setOutputElement(int dstIdx, llvm::Value *val, int mask)
+{
+   if (mask != TGSI_WRITEMASK_XYZW) {
+      llvm::Value *templ = 0;
+      if (m_destWriteMap[dstIdx])
+         templ = outputElement(dstIdx);
+      val = maskWrite(val, mask, templ);
+   }
+
+   Value *elem = element(DestsArg, dstIdx);
+   StoreInst *st = new StoreInst(val, elem, false, m_block);
+   st->setAlignment(8);
+   m_destWriteMap[dstIdx] = true;
+}
+
+llvm::Value *Storage::maskWrite(llvm::Value *src, int mask, llvm::Value *templ)
+{
+   llvm::Value *dst = templ;
+   if (!dst)
+      dst = Constant::getNullValue(m_floatVecType);
+   if ((mask & TGSI_WRITEMASK_X)) {
+      llvm::Value *x = new ExtractElementInst(src, unsigned(0),
+                                              name("x"), m_block);
+      dst = InsertElementInst::Create(dst, x, unsigned(0),
+                                      name("dstx"), m_block);
+   }
+   if ((mask & TGSI_WRITEMASK_Y)) {
+      llvm::Value *y = new ExtractElementInst(src, unsigned(1),
+                                              name("y"), m_block);
+      dst = InsertElementInst::Create(dst, y, unsigned(1),
+                                      name("dsty"), m_block);
+   }
+   if ((mask & TGSI_WRITEMASK_Z)) {
+      llvm::Value *z = new ExtractElementInst(src, unsigned(2),
+                                              name("z"), m_block);
+      dst = InsertElementInst::Create(dst, z, unsigned(2),
+                                      name("dstz"), m_block);
+   }
+   if ((mask & TGSI_WRITEMASK_W)) {
+      llvm::Value *w = new ExtractElementInst(src, unsigned(3),
+                                              name("w"), m_block);
+      dst = InsertElementInst::Create(dst, w, unsigned(3),
+                                      name("dstw"), m_block);
+   }
+   return dst;
+}
+
+const char * Storage::name(const char *prefix)
+{
+   ++m_idx;
+   snprintf(m_name, 32, "%s%d", prefix, m_idx);
+   return m_name;
+}
+
+int Storage::numConsts() const
+{
+   return m_numConsts;
+}
+
+llvm::Value * Storage::addrElement(int idx) const
+{
+   Value *ret = m_addrs[idx];
+   if (!ret)
+      return m_undefFloatVec;
+   return ret;
+}
+
+void Storage::setAddrElement(int idx, llvm::Value *val, int mask)
+{
+   if (mask != TGSI_WRITEMASK_XYZW) {
+      llvm::Value *templ = m_addrs[idx];
+      val = maskWrite(val, mask, templ);
+   }
+   m_addrs[idx] = val;
+}
+
+llvm::Value * Storage::extractIndex(llvm::Value *vec)
+{
+   llvm::Value *x = new ExtractElementInst(vec, unsigned(0),
+                                           name("x"), m_block);
+   return new FPToSIInst(x, IntegerType::get(32), name("intidx"), m_block);
+}
+
+void Storage::setCurrentBlock(llvm::BasicBlock *block)
+{
+   m_block = block;
+}
+
+llvm::Value * Storage::outputElement(int idx, llvm::Value *indIdx)
+{
+   Value *elem = element(DestsArg, idx, indIdx);
+   LoadInst *load = new LoadInst(elem, name("output"), false, m_block);
+   load->setAlignment(8);
+
+   return load;
+}
+
+llvm::Value * Storage::inputPtr() const
+{
+   return m_INPUT;
+}
+
+void Storage::pushArguments(llvm::Value *input)
+{
+   m_argStack.push(m_INPUT);
+
+   m_INPUT = input;
+}
+
+void Storage::popArguments()
+{
+   m_INPUT = m_argStack.top();
+   m_argStack.pop();
+}
+
+void Storage::pushTemps()
+{
+   m_extSwizzleVec = 0;
+}
+
+void Storage::popTemps()
+{
+}
+
+llvm::Value * Storage::immediateElement(int idx)
+{
+   return m_immediates[idx];
+}
+
+void Storage::addImmediate(float *val)
+{
+   std::vector<Constant*> vec(4);
+   vec[0] = ConstantFP::get(APFloat(val[0]));
+   vec[1] = ConstantFP::get(APFloat(val[1]));
+   vec[2] = ConstantFP::get(APFloat(val[2]));
+   vec[3] = ConstantFP::get(APFloat(val[3]));
+   m_immediates.push_back(ConstantVector::get(m_floatVecType, vec));
+}
+
+
+llvm::Value * Storage::elemPtr(Args arg)
+{
+   std::vector<Value*> indices;
+   indices.push_back(constantInt(0));
+   indices.push_back(constantInt(static_cast<int>(arg)));
+   GetElementPtrInst *getElem = GetElementPtrInst::Create(m_INPUT,
+                                                          indices.begin(),
+                                                          indices.end(),
+                                                          name("input_ptr"),
+                                                          m_block);
+   return new LoadInst(getElem, name("input_field"), false, m_block);
+}
+
+llvm::Value * Storage::elemIdx(llvm::Value *ptr, int idx,
+                               llvm::Value *indIdx )
+{
+   GetElementPtrInst *getElem = 0;
+
+   if (indIdx) {
+      getElem = GetElementPtrInst::Create(ptr,
+                                      BinaryOperator::Create(Instruction::Add,
+                                                             indIdx,
+                                                             constantInt(idx),
+                                                             name("add"),
+                                                             m_block),
+                                      name("field"),
+                                      m_block);
+   } else {
+      getElem = GetElementPtrInst::Create(ptr,
+                                      constantInt(idx),
+                                      name("field"),
+                                      m_block);
+   }
+   return getElem;
+}
+
+llvm::Value * Storage::element(Args arg, int idx, llvm::Value *indIdx )
+{
+   Value *val = elemPtr(arg);
+   return elemIdx(val, idx, indIdx);
+}
+
+void Storage::setKilElement(llvm::Value *val)
+{
+   std::vector<Value*> indices;
+   indices.push_back(constantInt(0));
+   indices.push_back(constantInt(static_cast<int>(KilArg)));
+   GetElementPtrInst *elem = GetElementPtrInst::Create(m_INPUT,
+                                                   indices.begin(),
+                                                   indices.end(),
+                                                   name("kil_ptr"),
+                                                   m_block);
+   StoreInst *st = new StoreInst(val, elem, false, m_block);
+   st->setAlignment(8);
+}
+
+#endif //MESA_LLVM
+
+
diff --git a/src/gallium/auxiliary/gallivm/storage.h b/src/gallium/auxiliary/gallivm/storage.h
new file mode 100644
index 00000000000..8574f7554e3
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/storage.h
@@ -0,0 +1,133 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin [email protected]
+  */
+
+#ifndef STORAGE_H
+#define STORAGE_H
+
+#include <map>
+#include <set>
+#include <stack>
+#include <vector>
+
+namespace llvm {
+   class BasicBlock;
+   class Constant;
+   class ConstantInt;
+   class LoadInst;
+   class Value;
+   class VectorType;
+}
+
+class Storage
+{
+public:
+   Storage(llvm::BasicBlock *block,
+           llvm::Value *input);
+
+   llvm::Value *inputPtr() const;
+
+   void setCurrentBlock(llvm::BasicBlock *block);
+
+   llvm::ConstantInt *constantInt(int);
+   llvm::Constant *shuffleMask(int vec);
+   llvm::Value *inputElement(int idx, llvm::Value *indIdx =0);
+   llvm::Value *constElement(int idx, llvm::Value *indIdx =0);
+   llvm::Value *outputElement(int idx, llvm::Value *indIdx =0);
+   llvm::Value *tempElement(int idx, llvm::Value *indIdx =0);
+   llvm::Value *immediateElement(int idx);
+
+   void setOutputElement(int dstIdx, llvm::Value *val, int mask);
+   void setTempElement(int idx, llvm::Value *val, int mask);
+
+   llvm::Value *addrElement(int idx) const;
+   void setAddrElement(int idx, llvm::Value *val, int mask);
+
+   void setKilElement(llvm::Value *val);
+
+   llvm::Value *shuffleVector(llvm::Value *vec, int shuffle);
+
+   llvm::Value *extractIndex(llvm::Value *vec);
+
+   int numConsts() const;
+
+   void pushArguments(llvm::Value *input);
+   void popArguments();
+   void pushTemps();
+   void popTemps();
+
+   void addImmediate(float *val);
+
+private:
+   llvm::Value *maskWrite(llvm::Value *src, int mask, llvm::Value *templ);
+   const char *name(const char *prefix);
+
+   enum Args {
+      DestsArg   = 0,
+      InputsArg  = 1,
+      TempsArg   = 2,
+      ConstsArg  = 3,
+      KilArg     = 4
+   };
+   llvm::Value *elemPtr(Args arg);
+   llvm::Value *elemIdx(llvm::Value *ptr, int idx,
+                        llvm::Value *indIdx = 0);
+   llvm::Value *element(Args arg, int idx, llvm::Value *indIdx = 0);
+
+private:
+   llvm::BasicBlock *m_block;
+   llvm::Value *m_INPUT;
+
+   std::map<int, llvm::ConstantInt*> m_constInts;
+   std::map<int, llvm::Constant*>    m_intVecs;
+   std::vector<llvm::Value*>         m_addrs;
+   std::vector<llvm::Constant*>      m_immediates;
+
+   llvm::VectorType *m_floatVecType;
+   llvm::VectorType *m_intVecType;
+
+   char        m_name[32];
+   int         m_idx;
+
+   int         m_numConsts;
+
+   std::map<int, bool > m_destWriteMap;
+   std::map<int, bool > m_tempWriteMap;
+
+   llvm::Value      *m_undefFloatVec;
+   llvm::Value      *m_undefIntVec;
+   llvm::Value      *m_extSwizzleVec;
+
+   std::stack<llvm::Value*> m_argStack;
+   std::stack<std::vector<llvm::Value*> > m_tempStack;
+};
+
+#endif
diff --git a/src/gallium/auxiliary/gallivm/storagesoa.cpp b/src/gallium/auxiliary/gallivm/storagesoa.cpp
new file mode 100644
index 00000000000..e1e5cabcf55
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/storagesoa.cpp
@@ -0,0 +1,438 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include "storagesoa.h"
+
+#include "gallivm_p.h"
+
+#include "pipe/p_shader_tokens.h"
+#include "pipe/p_debug.h"
+
+#include <llvm/BasicBlock.h>
+#include <llvm/Module.h>
+#include <llvm/Value.h>
+
+#include <llvm/CallingConv.h>
+#include <llvm/Constants.h>
+#include <llvm/DerivedTypes.h>
+#include <llvm/InstrTypes.h>
+#include <llvm/Instructions.h>
+
+using namespace llvm;
+
+
+StorageSoa::StorageSoa(llvm::BasicBlock *block,
+                       llvm::Value *input,
+                       llvm::Value *output,
+                       llvm::Value *consts)
+   : m_block(block),
+     m_input(input),
+     m_output(output),
+     m_consts(consts),
+     m_immediates(0),
+     m_idx(0)
+{
+}
+
+void StorageSoa::addImmediate(float *vec)
+{
+   std::vector<float> vals(4);
+   vals[0] = vec[0];
+   vals[1] = vec[1];
+   vals[2] = vec[2];
+   vals[3] = vec[3];
+   m_immediatesToFlush.push_back(vals);
+}
+
+void StorageSoa::declareImmediates()
+{
+   if (m_immediatesToFlush.empty())
+      return;
+
+   VectorType *vectorType = VectorType::get(Type::FloatTy, 4);
+   ArrayType  *vectorChannels = ArrayType::get(vectorType, 4);
+   ArrayType  *arrayType = ArrayType::get(vectorChannels, m_immediatesToFlush.size());
+
+   m_immediates = new GlobalVariable(
+      /*Type=*/arrayType,
+      /*isConstant=*/false,
+      /*Linkage=*/GlobalValue::ExternalLinkage,
+      /*Initializer=*/0, // has initializer, specified below
+      /*Name=*/name("immediates"),
+      currentModule());
+
+   std::vector<Constant*> arrayVals;
+   for (unsigned int i = 0; i < m_immediatesToFlush.size(); ++i) {
+      std::vector<float> vec = m_immediatesToFlush[i];
+      std::vector<float> vals(4);
+      std::vector<Constant*> channelArray;
+
+      vals[0] = vec[0]; vals[1] = vec[1]; vals[2] = vec[2]; vals[3] = vec[3];
+      llvm::Constant *xChannel = createConstGlobalVector(vals);
+
+      vals[0] = vec[1]; vals[1] = vec[1]; vals[2] = vec[1]; vals[3] = vec[1];
+      llvm::Constant *yChannel = createConstGlobalVector(vals);
+
+      vals[0] = vec[2]; vals[1] = vec[2]; vals[2] = vec[2]; vals[3] = vec[2];
+      llvm::Constant *zChannel = createConstGlobalVector(vals);
+
+      vals[0] = vec[3]; vals[1] = vec[3]; vals[2] = vec[3]; vals[3] = vec[3];
+      llvm::Constant *wChannel = createConstGlobalVector(vals);
+      channelArray.push_back(xChannel);
+      channelArray.push_back(yChannel);
+      channelArray.push_back(zChannel);
+      channelArray.push_back(wChannel);
+      Constant *constChannels = ConstantArray::get(vectorChannels,
+                                                   channelArray);
+      arrayVals.push_back(constChannels);
+   }
+   Constant *constArray = ConstantArray::get(arrayType, arrayVals);
+   m_immediates->setInitializer(constArray);
+
+   m_immediatesToFlush.clear();
+}
+
+llvm::Value *StorageSoa::addrElement(int idx) const
+{
+   std::map<int, llvm::Value*>::const_iterator itr = m_addresses.find(idx);
+   if (itr == m_addresses.end()) {
+      debug_printf("Trying to access invalid shader 'address'\n");
+      return 0;
+   }
+   llvm::Value * res = (*itr).second;
+
+   res = new LoadInst(res, name("addr"), false, m_block);
+
+   return res;
+}
+
+std::vector<llvm::Value*> StorageSoa::inputElement(llvm::Value *idx)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = element(m_input, idx, 0);
+   res[1] = element(m_input, idx, 1);
+   res[2] = element(m_input, idx, 2);
+   res[3] = element(m_input, idx, 3);
+
+   return res;
+}
+
+llvm::Value* StorageSoa::unpackConstElement(llvm::IRBuilder<>* m_builder, llvm::Value* vector, int cc)
+{
+   std::vector<llvm::Value*> x(4);
+   x[0] = m_builder->CreateExtractElement(vector,
+                                           constantInt(cc),
+                                           name("x"));
+
+   VectorType  *vectorType = VectorType::get(Type::FloatTy, 4);
+   Constant *constVector = Constant::getNullValue(vectorType);
+   Value *res = m_builder->CreateInsertElement(constVector, x[0],
+                                              constantInt(0),
+                                              name("vecx"));
+   res = m_builder->CreateInsertElement(res, x[0], constantInt(1),
+                               name("vecxx"));
+   res = m_builder->CreateInsertElement(res, x[0], constantInt(2),
+                               name("vecxxx"));
+   res = m_builder->CreateInsertElement(res, x[0], constantInt(3),
+                               name("vecxxxx"));
+   return res;
+}
+
+std::vector<llvm::Value*> StorageSoa::constElement(llvm::IRBuilder<>* m_builder, llvm::Value *idx)
+{
+   llvm::Value* res;
+   std::vector<llvm::Value*> res2(4);
+   llvm::Value *xChannel;
+
+   xChannel = elementPointer(m_consts, idx, 0);
+
+   res = alignedArrayLoad(xChannel);
+
+   res2[0]=unpackConstElement(m_builder, res,0);
+   res2[1]=unpackConstElement(m_builder, res,1);
+   res2[2]=unpackConstElement(m_builder, res,2);
+   res2[3]=unpackConstElement(m_builder, res,3);
+
+   return res2;
+}
+
+std::vector<llvm::Value*> StorageSoa::outputElement(llvm::Value *idx)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = element(m_output, idx, 0);
+   res[1] = element(m_output, idx, 1);
+   res[2] = element(m_output, idx, 2);
+   res[3] = element(m_output, idx, 3);
+
+   return res;
+}
+
+std::vector<llvm::Value*> StorageSoa::tempElement(llvm::IRBuilder<>* m_builder, int idx)
+{
+   std::vector<llvm::Value*> res(4);
+   llvm::Value *temp = m_temps[idx];
+
+   res[0] = element(temp, constantInt(0), 0);
+   res[1] = element(temp, constantInt(0), 1);
+   res[2] = element(temp, constantInt(0), 2);
+   res[3] = element(temp, constantInt(0), 3);
+
+   return res;
+}
+
+std::vector<llvm::Value*> StorageSoa::immediateElement(llvm::Value *idx)
+{
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = element(m_immediates, idx, 0);
+   res[1] = element(m_immediates, idx, 1);
+   res[2] = element(m_immediates, idx, 2);
+   res[3] = element(m_immediates, idx, 3);
+
+   return res;
+}
+
+llvm::Value * StorageSoa::elementPointer(llvm::Value *ptr, llvm::Value *index,
+                                         int channel) const
+{
+   std::vector<Value*> indices;
+   if (m_immediates == ptr)
+      indices.push_back(constantInt(0));
+   indices.push_back(index);
+   indices.push_back(constantInt(channel));
+
+   GetElementPtrInst *getElem = GetElementPtrInst::Create(ptr,
+                                                          indices.begin(),
+                                                          indices.end(),
+                                                          name("ptr"),
+                                                          m_block);
+   return getElem;
+}
+
+llvm::Value * StorageSoa::element(llvm::Value *ptr, llvm::Value *index,
+                                  int channel) const
+{
+   llvm::Value *res = elementPointer(ptr, index, channel);
+   LoadInst *load = new LoadInst(res, name("element"), false, m_block);
+   //load->setAlignment(8);
+   return load;
+}
+
+const char * StorageSoa::name(const char *prefix) const
+{
+   ++m_idx;
+   snprintf(m_name, 32, "%s%d", prefix, m_idx);
+   return m_name;
+}
+
+llvm::ConstantInt * StorageSoa::constantInt(int idx) const
+{
+   if (m_constInts.find(idx) != m_constInts.end()) {
+      return m_constInts[idx];
+   }
+   ConstantInt *constInt = ConstantInt::get(APInt(32,  idx));
+   m_constInts[idx] = constInt;
+   return constInt;
+}
+
+llvm::Value *StorageSoa::alignedArrayLoad(llvm::Value *val)
+{
+   VectorType  *vectorType = VectorType::get(Type::FloatTy, 4);
+   PointerType *vectorPtr  = PointerType::get(vectorType, 0);
+
+   CastInst *cast = new BitCastInst(val, vectorPtr, name("toVector"), m_block);
+   LoadInst *load = new LoadInst(cast, name("alignLoad"), false, m_block);
+   load->setAlignment(8);
+   return load;
+}
+
+llvm::Module * StorageSoa::currentModule() const
+{
+    if (!m_block || !m_block->getParent())
+       return 0;
+
+    return m_block->getParent()->getParent();
+}
+
+llvm::Constant * StorageSoa::createConstGlobalFloat(const float val)
+{
+   Constant*c = ConstantFP::get(APFloat(val));
+   return c;
+}
+
+llvm::Constant * StorageSoa::createConstGlobalVector(const std::vector<float> &vec)
+{
+   VectorType *vectorType = VectorType::get(Type::FloatTy, 4);
+   std::vector<Constant*> immValues;
+   ConstantFP *constx = ConstantFP::get(APFloat(vec[0]));
+   ConstantFP *consty = ConstantFP::get(APFloat(vec[1]));
+   ConstantFP *constz = ConstantFP::get(APFloat(vec[2]));
+   ConstantFP *constw = ConstantFP::get(APFloat(vec[3]));
+   immValues.push_back(constx);
+   immValues.push_back(consty);
+   immValues.push_back(constz);
+   immValues.push_back(constw);
+   Constant  *constVector = ConstantVector::get(vectorType, immValues);
+
+   return constVector;
+}
+
+std::vector<llvm::Value*> StorageSoa::load(enum tgsi_file_type type, int idx, int swizzle,
+                                           llvm::IRBuilder<>* m_builder,llvm::Value *indIdx)
+{
+   std::vector<llvm::Value*> val(4);
+
+   //if we have an indirect index, always use that
+   //   if not use the integer offset to create one
+   llvm::Value *realIndex = 0;
+   if (indIdx)
+      realIndex = indIdx;
+   else
+      realIndex = constantInt(idx);
+   debug_printf("XXXXXXXXX realIdx = %p, indIdx = %p\n", realIndex, indIdx);
+
+   switch(type) {
+   case TGSI_FILE_INPUT:
+      val = inputElement(realIndex);
+      break;
+   case TGSI_FILE_OUTPUT:
+      val = outputElement(realIndex);
+      break;
+   case TGSI_FILE_TEMPORARY:
+      val = tempElement(m_builder, idx);
+      break;
+   case TGSI_FILE_CONSTANT:
+      val = constElement(m_builder, realIndex);
+      break;
+   case TGSI_FILE_IMMEDIATE:
+      val = immediateElement(realIndex);
+      break;
+   case TGSI_FILE_ADDRESS:
+      debug_printf("Address not handled in the load phase!\n");
+      assert(0);
+      break;
+   default:
+      debug_printf("Unknown load!\n");
+      assert(0);
+      break;
+   }
+   if (!gallivm_is_swizzle(swizzle))
+      return val;
+
+   std::vector<llvm::Value*> res(4);
+
+   res[0] = val[gallivm_x_swizzle(swizzle)];
+   res[1] = val[gallivm_y_swizzle(swizzle)];
+   res[2] = val[gallivm_z_swizzle(swizzle)];
+   res[3] = val[gallivm_w_swizzle(swizzle)];
+   return res;
+}
+
+llvm::Value * StorageSoa::allocaTemp(llvm::IRBuilder<>* m_builder)
+{
+   VectorType *vector   = VectorType::get(Type::FloatTy, 4);
+   ArrayType  *vecArray = ArrayType::get(vector, 4);
+   AllocaInst *alloca = new AllocaInst(vecArray, "temp",
+                                       m_builder->GetInsertBlock());
+
+   return alloca;
+}
+
+
+void StorageSoa::store(enum tgsi_file_type type, int idx, const std::vector<llvm::Value*> &val,
+                       int mask, llvm::IRBuilder<>* m_builder)
+{
+   llvm::Value *out = 0;
+   llvm::Value *realIndex = 0;
+   switch(type) {
+   case TGSI_FILE_OUTPUT:
+      out = m_output;
+      realIndex = constantInt(idx);
+      break;
+   case TGSI_FILE_TEMPORARY:
+      // if that temp doesn't already exist, alloca it
+      if (m_temps.find(idx) == m_temps.end())
+         m_temps[idx] = allocaTemp(m_builder);
+
+      out = m_temps[idx];
+
+      realIndex = constantInt(0);
+      break;
+   case TGSI_FILE_INPUT:
+      out = m_input;
+      realIndex = constantInt(idx);
+      break;
+   case TGSI_FILE_ADDRESS: {
+      llvm::Value *addr = m_addresses[idx];
+      if (!addr) {
+         addAddress(idx);
+         addr = m_addresses[idx];
+         assert(addr);
+      }
+      new StoreInst(val[0], addr, false, m_block);
+      return;
+      break;
+   }
+   default:
+      debug_printf("Can't save output of this type: %d !\n", type);
+      assert(0);
+      break;
+   }
+   if ((mask & TGSI_WRITEMASK_X)) {
+      llvm::Value *xChannel = elementPointer(out, realIndex, 0);
+      new StoreInst(val[0], xChannel, false, m_block);
+   }
+   if ((mask & TGSI_WRITEMASK_Y)) {
+      llvm::Value *yChannel = elementPointer(out, realIndex, 1);
+      new StoreInst(val[1], yChannel, false, m_block);
+   }
+   if ((mask & TGSI_WRITEMASK_Z)) {
+      llvm::Value *zChannel = elementPointer(out, realIndex, 2);
+      new StoreInst(val[2], zChannel, false, m_block);
+   }
+   if ((mask & TGSI_WRITEMASK_W)) {
+      llvm::Value *wChannel = elementPointer(out, realIndex, 3);
+      new StoreInst(val[3], wChannel, false, m_block);
+   }
+}
+
+void StorageSoa::addAddress(int idx)
+{
+   GlobalVariable *val = new GlobalVariable(
+      /*Type=*/IntegerType::get(32),
+      /*isConstant=*/false,
+      /*Linkage=*/GlobalValue::ExternalLinkage,
+      /*Initializer=*/0, // has initializer, specified below
+      /*Name=*/name("address"),
+      currentModule());
+   val->setInitializer(Constant::getNullValue(IntegerType::get(32)));
+
+   debug_printf("adding to %d\n", idx);
+   m_addresses[idx] = val;
+}
diff --git a/src/gallium/auxiliary/gallivm/storagesoa.h b/src/gallium/auxiliary/gallivm/storagesoa.h
new file mode 100644
index 00000000000..56886f85e7a
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/storagesoa.h
@@ -0,0 +1,107 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#ifndef STORAGESOA_H
+#define STORAGESOA_H
+
+#include <pipe/p_shader_tokens.h>
+#include <llvm/Support/IRBuilder.h>
+
+#include <vector>
+#include <list>
+#include <map>
+
+namespace llvm {
+   class BasicBlock;
+   class Constant;
+   class ConstantInt;
+   class GlobalVariable;
+   class LoadInst;
+   class Value;
+   class VectorType;
+   class Module;
+}
+
+class StorageSoa
+{
+public:
+   StorageSoa(llvm::BasicBlock *block,
+              llvm::Value *input,
+              llvm::Value *output,
+              llvm::Value *consts);
+
+
+   std::vector<llvm::Value*> load(enum tgsi_file_type type, int idx, int swizzle, 
+                                  llvm::IRBuilder<>* m_builder, llvm::Value *indIdx =0);
+   void store(enum tgsi_file_type type, int idx, const std::vector<llvm::Value*> &val,
+              int mask, llvm::IRBuilder<>* m_builder);
+
+   void addImmediate(float *vec);
+   void declareImmediates();
+
+   void addAddress(int idx);
+
+   llvm::Value  * addrElement(int idx) const;
+
+   llvm::ConstantInt *constantInt(int) const;
+private:
+   llvm::Value *elementPointer(llvm::Value *ptr, llvm::Value *indIdx,
+                               int channel) const;
+   llvm::Value *element(llvm::Value *ptr, llvm::Value *idx,
+                        int channel) const;
+   const char *name(const char *prefix) const;
+   llvm::Value  *alignedArrayLoad(llvm::Value *val);
+   llvm::Module *currentModule() const;
+   llvm::Constant  *createConstGlobalFloat(const float val);
+   llvm::Constant  *createConstGlobalVector(const std::vector<float> &vec);
+
+   std::vector<llvm::Value*> inputElement(llvm::Value *indIdx);
+   llvm::Value* unpackConstElement(llvm::IRBuilder<>* m_builder, llvm::Value *indIdx, int cc);
+   std::vector<llvm::Value*> constElement(llvm::IRBuilder<>* m_builder, llvm::Value *indIdx);
+   std::vector<llvm::Value*> outputElement(llvm::Value *indIdx);
+   std::vector<llvm::Value*> tempElement(llvm::IRBuilder<>* m_builder, int idx);
+   std::vector<llvm::Value*> immediateElement(llvm::Value *indIdx);
+private:
+   llvm::BasicBlock *m_block;
+
+   llvm::Value *m_input;
+   llvm::Value *m_output;
+   llvm::Value *m_consts;
+   std::map<int, llvm::Value*> m_temps;
+   llvm::GlobalVariable *m_immediates;
+
+   std::map<int, llvm::Value*> m_addresses;
+
+   std::vector<std::vector<float> > m_immediatesToFlush;
+   llvm::Value * allocaTemp(llvm::IRBuilder<>* m_builder);
+
+   mutable std::map<int, llvm::ConstantInt*> m_constInts;
+   mutable char        m_name[32];
+   mutable int         m_idx;
+};
+
+#endif
diff --git a/src/gallium/auxiliary/gallivm/tgsitollvm.cpp b/src/gallium/auxiliary/gallivm/tgsitollvm.cpp
new file mode 100644
index 00000000000..5b08200d142
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/tgsitollvm.cpp
@@ -0,0 +1,1164 @@
+#include "tgsitollvm.h"
+
+#include "gallivm.h"
+#include "gallivm_p.h"
+
+#include "storage.h"
+#include "instructions.h"
+#include "storagesoa.h"
+#include "instructionssoa.h"
+
+#include "pipe/p_shader_tokens.h"
+
+#include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_exec.h"
+#include "tgsi/tgsi_util.h"
+#include "tgsi/tgsi_build.h"
+#include "tgsi/tgsi_dump.h"
+
+
+#include <llvm/Module.h>
+#include <llvm/CallingConv.h>
+#include <llvm/Constants.h>
+#include <llvm/DerivedTypes.h>
+#include <llvm/Instructions.h>
+#include <llvm/ModuleProvider.h>
+#include <llvm/Pass.h>
+#include <llvm/PassManager.h>
+#include <llvm/Attributes.h>
+#include <llvm/Support/PatternMatch.h>
+#include <llvm/ExecutionEngine/JIT.h>
+#include <llvm/ExecutionEngine/Interpreter.h>
+#include <llvm/ExecutionEngine/GenericValue.h>
+#include <llvm/Support/MemoryBuffer.h>
+#include <llvm/LinkAllPasses.h>
+#include <llvm/Analysis/Verifier.h>
+#include <llvm/Analysis/LoopPass.h>
+#include <llvm/Target/TargetData.h>
+#include <llvm/Bitcode/ReaderWriter.h>
+#include <llvm/Transforms/Utils/Cloning.h>
+
+
+#include <sstream>
+#include <fstream>
+#include <iostream>
+
+using namespace llvm;
+
+static inline FunctionType *vertexShaderFunctionType()
+{
+   //Function takes three arguments,
+   // the calling code has to make sure the types it will
+   // pass are castable to the following:
+   // [4 x <4 x float>] inputs,
+   // [4 x <4 x float>] output,
+   // [4 x [1 x float]] consts,
+
+   std::vector<const Type*> funcArgs;
+   VectorType *vectorType = VectorType::get(Type::FloatTy, 4);
+   ArrayType *vectorArray = ArrayType::get(vectorType, 4);
+   PointerType *vectorArrayPtr = PointerType::get(vectorArray, 0);
+
+   ArrayType   *floatArray     = ArrayType::get(Type::FloatTy, 4);
+   ArrayType   *constsArray    = ArrayType::get(floatArray, 1);
+   PointerType *constsArrayPtr = PointerType::get(constsArray, 0);
+
+   funcArgs.push_back(vectorArrayPtr);//inputs
+   funcArgs.push_back(vectorArrayPtr);//output
+   funcArgs.push_back(constsArrayPtr);//consts
+
+   FunctionType *functionType = FunctionType::get(
+      /*Result=*/Type::VoidTy,
+      /*Params=*/funcArgs,
+      /*isVarArg=*/false);
+
+   return functionType;
+}
+
+static inline void
+add_interpolator(struct gallivm_ir *ir,
+                 struct gallivm_interpolate *interp)
+{
+   ir->interpolators[ir->num_interp] = *interp;
+   ++ir->num_interp;
+}
+
+static void
+translate_declaration(struct gallivm_ir *prog,
+                      llvm::Module *module,
+                      Storage *storage,
+                      struct tgsi_full_declaration *decl,
+                      struct tgsi_full_declaration *fd)
+{
+   if (decl->Declaration.File == TGSI_FILE_INPUT) {
+      unsigned first, last, mask;
+      uint interp_method;
+
+      first = decl->DeclarationRange.First;
+      last = decl->DeclarationRange.Last;
+      mask = decl->Declaration.UsageMask;
+
+      /* Do not touch WPOS.xy */
+      if (first == 0) {
+         mask &= ~TGSI_WRITEMASK_XY;
+         if (mask == TGSI_WRITEMASK_NONE) {
+            first++;
+            if (first > last) {
+               return;
+            }
+         }
+      }
+
+      interp_method = decl->Declaration.Interpolate;
+
+      if (mask == TGSI_WRITEMASK_XYZW) {
+         unsigned i, j;
+
+         for (i = first; i <= last; i++) {
+            for (j = 0; j < NUM_CHANNELS; j++) {
+               //interp( mach, i, j );
+               struct gallivm_interpolate interp;
+               interp.type = interp_method;
+               interp.attrib = i;
+               interp.chan = j;
+               add_interpolator(prog, &interp);
+            }
+         }
+      } else {
+         unsigned i, j;
+         for( j = 0; j < NUM_CHANNELS; j++ ) {
+            if( mask & (1 << j) ) {
+               for( i = first; i <= last; i++ ) {
+                  struct gallivm_interpolate interp;
+                  interp.type = interp_method;
+                  interp.attrib = i;
+                  interp.chan = j;
+                  add_interpolator(prog, &interp);
+               }
+            }
+         }
+      }
+   }
+}
+
+static void
+translate_declarationir(struct gallivm_ir *,
+                      llvm::Module *,
+                      StorageSoa *storage,
+                      struct tgsi_full_declaration *decl,
+                      struct tgsi_full_declaration *)
+{
+   if (decl->Declaration.File == TGSI_FILE_ADDRESS) {
+      int idx = decl->DeclarationRange.First;
+      storage->addAddress(idx);
+   }
+}
+
+static void
+translate_immediate(Storage *storage,
+                    struct tgsi_full_immediate *imm)
+{
+   float vec[4];
+   int i;
+   for (i = 0; i < imm->Immediate.NrTokens - 1; ++i) {
+      switch (imm->Immediate.DataType) {
+      case TGSI_IMM_FLOAT32:
+         vec[i] = imm->u.ImmediateFloat32[i].Float;
+         break;
+      default:
+         assert(0);
+      }
+   }
+   storage->addImmediate(vec);
+}
+
+
+static void
+translate_immediateir(StorageSoa *storage,
+                      struct tgsi_full_immediate *imm)
+{
+   float vec[4];
+   int i;
+   for (i = 0; i < imm->Immediate.NrTokens - 1; ++i) {
+      switch (imm->Immediate.DataType) {
+      case TGSI_IMM_FLOAT32:
+         vec[i] = imm->u.ImmediateFloat32[i].Float;
+         break;
+      default:
+         assert(0);
+      }
+   }
+   storage->addImmediate(vec);
+}
+
+static inline int
+swizzleInt(struct tgsi_full_src_register *src)
+{
+   int swizzle = 0;
+   int start = 1000;
+
+   for (int k = 0; k < 4; ++k) {
+      swizzle += tgsi_util_get_full_src_register_extswizzle(src, k) * start;
+      start /= 10;
+   }
+   return swizzle;
+}
+
+static inline llvm::Value *
+swizzleVector(llvm::Value *val, struct tgsi_full_src_register *src,
+              Storage *storage)
+{
+   int swizzle = swizzleInt(src);
+
+   if (gallivm_is_swizzle(swizzle)) {
+      /*fprintf(stderr, "XXXXXXXX swizzle = %d\n", swizzle);*/
+      val = storage->shuffleVector(val, swizzle);
+   }
+   return val;
+}
+
+static void
+translate_instruction(llvm::Module *module,
+                      Storage *storage,
+                      Instructions *instr,
+                      struct tgsi_full_instruction *inst,
+                      struct tgsi_full_instruction *fi,
+                      unsigned instno)
+{
+   llvm::Value *inputs[4];
+   inputs[0] = 0;
+   inputs[1] = 0;
+   inputs[2] = 0;
+   inputs[3] = 0;
+
+   for (int i = 0; i < inst->Instruction.NumSrcRegs; ++i) {
+      struct tgsi_full_src_register *src = &inst->FullSrcRegisters[i];
+      llvm::Value *val = 0;
+      llvm::Value *indIdx = 0;
+
+      if (src->SrcRegister.Indirect) {
+         indIdx = storage->addrElement(src->SrcRegisterInd.Index);
+         indIdx = storage->extractIndex(indIdx);
+      }
+      if (src->SrcRegister.File == TGSI_FILE_CONSTANT) {
+         val = storage->constElement(src->SrcRegister.Index, indIdx);
+      } else if (src->SrcRegister.File == TGSI_FILE_INPUT) {
+         val = storage->inputElement(src->SrcRegister.Index, indIdx);
+      } else if (src->SrcRegister.File == TGSI_FILE_TEMPORARY) {
+         val = storage->tempElement(src->SrcRegister.Index);
+      } else if (src->SrcRegister.File == TGSI_FILE_OUTPUT) {
+         val = storage->outputElement(src->SrcRegister.Index, indIdx);
+      } else if (src->SrcRegister.File == TGSI_FILE_IMMEDIATE) {
+         val = storage->immediateElement(src->SrcRegister.Index);
+      } else {
+         fprintf(stderr, "ERROR: not supported llvm source %d\n", src->SrcRegister.File);
+         return;
+      }
+
+      inputs[i] = swizzleVector(val, src, storage);
+   }
+
+   /*if (inputs[0])
+     instr->printVector(inputs[0]);
+     if (inputs[1])
+     instr->printVector(inputs[1]);*/
+   llvm::Value *out = 0;
+   switch (inst->Instruction.Opcode) {
+   case TGSI_OPCODE_ARL: {
+      out = instr->arl(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_MOV: {
+      out = inputs[0];
+   }
+      break;
+   case TGSI_OPCODE_LIT: {
+      out = instr->lit(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_RCP: {
+      out = instr->rcp(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_RSQ: {
+      out = instr->rsq(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_EXP: {
+      out = instr->exp(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_LOG: {
+      out = instr->log(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_MUL: {
+      out = instr->mul(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_ADD: {
+      out = instr->add(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_DP3: {
+      out = instr->dp3(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_DP4: {
+      out = instr->dp4(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_DST: {
+      out = instr->dst(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_MIN: {
+      out = instr->min(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_MAX: {
+      out = instr->max(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_SLT: {
+      out = instr->slt(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_SGE: {
+      out = instr->sge(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_MAD: {
+      out = instr->madd(inputs[0], inputs[1], inputs[2]);
+   }
+      break;
+   case TGSI_OPCODE_SUB: {
+      out = instr->sub(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_LERP: {
+      out = instr->lerp(inputs[0], inputs[1], inputs[2]);
+   }
+      break;
+   case TGSI_OPCODE_CND: {
+      out = instr->cnd(inputs[0], inputs[1], inputs[2]);
+   }
+      break;
+   case TGSI_OPCODE_CND0: {
+      out = instr->cnd0(inputs[0], inputs[1], inputs[2]);
+   }
+      break;
+   case TGSI_OPCODE_DOT2ADD: {
+      out = instr->dot2add(inputs[0], inputs[1], inputs[2]);
+   }
+      break;
+   case TGSI_OPCODE_INDEX:
+      break;
+   case TGSI_OPCODE_NEGATE: {
+      out = instr->neg(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_FRAC: {
+      out = instr->frc(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_CLAMP: {
+      out = instr->clamp(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_FLOOR: {
+      out = instr->floor(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_ROUND:
+      break;
+   case TGSI_OPCODE_EXPBASE2: {
+      out = instr->ex2(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_LOGBASE2: {
+      out = instr->lg2(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_POWER: {
+      out = instr->pow(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_CROSSPRODUCT: {
+      out = instr->cross(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_MULTIPLYMATRIX:
+      break;
+   case TGSI_OPCODE_ABS: {
+      out = instr->abs(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_RCC:
+      break;
+   case TGSI_OPCODE_DPH: {
+      out = instr->dph(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_COS: {
+      out = instr->cos(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_DDX: {
+      out = instr->ddx(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_DDY: {
+      out = instr->ddy(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_KILP:
+      break;
+   case TGSI_OPCODE_PK2H:
+      break;
+   case TGSI_OPCODE_PK2US:
+      break;
+   case TGSI_OPCODE_PK4B:
+      break;
+   case TGSI_OPCODE_PK4UB:
+      break;
+   case TGSI_OPCODE_RFL:
+      break;
+   case TGSI_OPCODE_SEQ: {
+      out = instr->seq(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_SFL: {
+      out = instr->sfl(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_SGT: {
+      out = instr->sgt(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_SIN: {
+      out = instr->sin(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_SLE: {
+      out = instr->sle(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_SNE: {
+      out = instr->sne(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_STR: {
+      out = instr->str(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_TEX:
+      break;
+   case TGSI_OPCODE_TXD:
+      break;
+   case TGSI_OPCODE_UP2H:
+      break;
+   case TGSI_OPCODE_UP2US:
+      break;
+   case TGSI_OPCODE_UP4B:
+      break;
+   case TGSI_OPCODE_UP4UB:
+      break;
+   case TGSI_OPCODE_X2D: {
+      out = instr->x2d(inputs[0], inputs[1], inputs[2]);
+   }
+      break;
+   case TGSI_OPCODE_ARA:
+      break;
+   case TGSI_OPCODE_ARR:
+      break;
+   case TGSI_OPCODE_BRA:
+      break;
+   case TGSI_OPCODE_CAL: {
+      instr->cal(inst->InstructionExtLabel.Label, storage->inputPtr());
+      return;
+   }
+      break;
+   case TGSI_OPCODE_RET: {
+      instr->end();
+      return;
+   }
+      break;
+   case TGSI_OPCODE_SSG:
+      break;
+   case TGSI_OPCODE_CMP: {
+      out = instr->cmp(inputs[0], inputs[1], inputs[2]);
+   }
+      break;
+   case TGSI_OPCODE_SCS: {
+      out = instr->scs(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_TXB:
+      break;
+   case TGSI_OPCODE_NRM4:
+   case TGSI_OPCODE_NRM: {
+      out = instr->nrm(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_DIV: {
+      out = instr->div(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_DP2: {
+      out = instr->dp2(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_TXL:
+      break;
+   case TGSI_OPCODE_BRK: {
+      instr->brk();
+      return;
+   }
+      break;
+   case TGSI_OPCODE_IF: {
+      instr->ifop(inputs[0]);
+      storage->setCurrentBlock(instr->currentBlock());
+      return;  //just update the state
+   }
+      break;
+   case TGSI_OPCODE_LOOP:
+      break;
+   case TGSI_OPCODE_REP:
+      break;
+   case TGSI_OPCODE_ELSE: {
+      instr->elseop();
+      storage->setCurrentBlock(instr->currentBlock());
+      return; //only state update
+   }
+      break;
+   case TGSI_OPCODE_ENDIF: {
+      instr->endif();
+      storage->setCurrentBlock(instr->currentBlock());
+      return; //just update the state
+   }
+      break;
+   case TGSI_OPCODE_ENDLOOP:
+      break;
+   case TGSI_OPCODE_ENDREP:
+      break;
+   case TGSI_OPCODE_PUSHA:
+      break;
+   case TGSI_OPCODE_POPA:
+      break;
+   case TGSI_OPCODE_CEIL:
+      break;
+   case TGSI_OPCODE_I2F:
+      break;
+   case TGSI_OPCODE_NOT:
+      break;
+   case TGSI_OPCODE_TRUNC: {
+      out = instr->trunc(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_SHL:
+      break;
+   case TGSI_OPCODE_SHR:
+      break;
+   case TGSI_OPCODE_AND:
+      break;
+   case TGSI_OPCODE_OR:
+      break;
+   case TGSI_OPCODE_MOD:
+      break;
+   case TGSI_OPCODE_XOR:
+      break;
+   case TGSI_OPCODE_SAD:
+      break;
+   case TGSI_OPCODE_TXF:
+      break;
+   case TGSI_OPCODE_TXQ:
+      break;
+   case TGSI_OPCODE_CONT:
+      break;
+   case TGSI_OPCODE_EMIT:
+      break;
+   case TGSI_OPCODE_ENDPRIM:
+      break;
+   case TGSI_OPCODE_BGNLOOP2: {
+      instr->beginLoop();
+      storage->setCurrentBlock(instr->currentBlock());
+      return;
+   }
+      break;
+   case TGSI_OPCODE_BGNSUB: {
+      instr->bgnSub(instno);
+      storage->setCurrentBlock(instr->currentBlock());
+      storage->pushTemps();
+      return;
+   }
+      break;
+   case TGSI_OPCODE_ENDLOOP2: {
+      instr->endLoop();
+      storage->setCurrentBlock(instr->currentBlock());
+      return;
+   }
+      break;
+   case TGSI_OPCODE_ENDSUB: {
+      instr->endSub();
+      storage->setCurrentBlock(instr->currentBlock());
+      storage->popArguments();
+      storage->popTemps();
+      return;
+   }
+      break;
+   case TGSI_OPCODE_NOISE1:
+      break;
+   case TGSI_OPCODE_NOISE2:
+      break;
+   case TGSI_OPCODE_NOISE3:
+      break;
+   case TGSI_OPCODE_NOISE4:
+      break;
+   case TGSI_OPCODE_NOP:
+      break;
+   case TGSI_OPCODE_M4X3:
+      break;
+   case TGSI_OPCODE_M3X4:
+      break;
+   case TGSI_OPCODE_M3X3:
+      break;
+   case TGSI_OPCODE_M3X2:
+      break;
+   case TGSI_OPCODE_CALLNZ:
+      break;
+   case TGSI_OPCODE_IFC:
+      break;
+   case TGSI_OPCODE_BREAKC:
+      break;
+   case TGSI_OPCODE_KIL: {
+      out = instr->kil(inputs[0]);
+      storage->setKilElement(out);
+      return;
+   }
+      break;
+   case TGSI_OPCODE_END:
+      instr->end();
+      return;
+      break;
+   default:
+      fprintf(stderr, "ERROR: Unknown opcode %d\n",
+              inst->Instruction.Opcode);
+      assert(0);
+      break;
+   }
+
+   if (!out) {
+      fprintf(stderr, "ERROR: unsupported opcode %d\n",
+              inst->Instruction.Opcode);
+      assert(!"Unsupported opcode");
+   }
+
+   /* # not sure if we need this */
+   switch( inst->Instruction.Saturate ) {
+   case TGSI_SAT_NONE:
+      break;
+   case TGSI_SAT_ZERO_ONE:
+      /*TXT( "_SAT" );*/
+      break;
+   case TGSI_SAT_MINUS_PLUS_ONE:
+      /*TXT( "_SAT[-1,1]" );*/
+      break;
+   default:
+      assert( 0 );
+   }
+
+   /* store results  */
+   for (int i = 0; i < inst->Instruction.NumDstRegs; ++i) {
+      struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i];
+
+      if (dst->DstRegister.File == TGSI_FILE_OUTPUT) {
+         storage->setOutputElement(dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
+      } else if (dst->DstRegister.File == TGSI_FILE_TEMPORARY) {
+         storage->setTempElement(dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
+      } else if (dst->DstRegister.File == TGSI_FILE_ADDRESS) {
+         storage->setAddrElement(dst->DstRegister.Index, out, dst->DstRegister.WriteMask);
+      } else {
+         fprintf(stderr, "ERROR: unsupported LLVM destination!");
+         assert(!"wrong destination");
+      }
+   }
+}
+
+
+static void
+translate_instructionir(llvm::Module *module,
+                        StorageSoa *storage,
+                        InstructionsSoa *instr,
+                        struct tgsi_full_instruction *inst,
+                        struct tgsi_full_instruction *fi,
+                        unsigned instno)
+{
+   std::vector< std::vector<llvm::Value*> > inputs(inst->Instruction.NumSrcRegs);
+
+   for (int i = 0; i < inst->Instruction.NumSrcRegs; ++i) {
+      struct tgsi_full_src_register *src = &inst->FullSrcRegisters[i];
+      std::vector<llvm::Value*> val;
+      llvm::Value *indIdx = 0;
+      int swizzle = swizzleInt(src);
+
+      if (src->SrcRegister.Indirect) {
+         indIdx = storage->addrElement(src->SrcRegisterInd.Index);
+      }
+      val = storage->load((enum tgsi_file_type)src->SrcRegister.File,
+                          src->SrcRegister.Index, swizzle, instr->getIRBuilder(), indIdx);
+
+      inputs[i] = val;
+   }
+
+   std::vector<llvm::Value*> out(4);
+   switch (inst->Instruction.Opcode) {
+   case TGSI_OPCODE_ARL: {
+      out = instr->arl(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_MOV: {
+      out = inputs[0];
+   }
+      break;
+   case TGSI_OPCODE_LIT: {
+      out = instr->lit(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_RCP: {
+   }
+      break;
+   case TGSI_OPCODE_RSQ: {
+      out = instr->rsq(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_EXP:
+      break;
+   case TGSI_OPCODE_LOG:
+      break;
+   case TGSI_OPCODE_MUL: {
+      out = instr->mul(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_ADD: {
+      out = instr->add(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_DP3: {
+      out = instr->dp3(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_DP4: {
+      out = instr->dp4(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_DST: {
+   }
+      break;
+   case TGSI_OPCODE_MIN: {
+      out = instr->min(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_MAX: {
+      out = instr->max(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_SLT: {
+      out = instr->slt(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_SGE: {
+   }
+      break;
+   case TGSI_OPCODE_MAD: {
+      out = instr->madd(inputs[0], inputs[1], inputs[2]);
+   }
+      break;
+   case TGSI_OPCODE_SUB: {
+      out = instr->sub(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_LERP: {
+   }
+      break;
+   case TGSI_OPCODE_CND:
+      break;
+   case TGSI_OPCODE_CND0:
+      break;
+   case TGSI_OPCODE_DOT2ADD:
+      break;
+   case TGSI_OPCODE_INDEX:
+      break;
+   case TGSI_OPCODE_NEGATE:
+      break;
+   case TGSI_OPCODE_FRAC: {
+   }
+      break;
+   case TGSI_OPCODE_CLAMP:
+      break;
+   case TGSI_OPCODE_FLOOR: {
+   }
+      break;
+   case TGSI_OPCODE_ROUND:
+      break;
+   case TGSI_OPCODE_EXPBASE2: {
+   }
+      break;
+   case TGSI_OPCODE_LOGBASE2: {
+   }
+      break;
+   case TGSI_OPCODE_POWER: {
+      out = instr->pow(inputs[0], inputs[1]);
+   }
+      break;
+   case TGSI_OPCODE_CROSSPRODUCT: {
+   }
+      break;
+   case TGSI_OPCODE_MULTIPLYMATRIX:
+      break;
+   case TGSI_OPCODE_ABS: {
+      out = instr->abs(inputs[0]);
+   }
+      break;
+   case TGSI_OPCODE_RCC:
+      break;
+   case TGSI_OPCODE_DPH: {
+   }
+      break;
+   case TGSI_OPCODE_COS: {
+   }
+      break;
+   case TGSI_OPCODE_DDX:
+      break;
+   case TGSI_OPCODE_DDY:
+      break;
+   case TGSI_OPCODE_KILP:
+      break;
+   case TGSI_OPCODE_PK2H:
+      break;
+   case TGSI_OPCODE_PK2US:
+      break;
+   case TGSI_OPCODE_PK4B:
+      break;
+   case TGSI_OPCODE_PK4UB:
+      break;
+   case TGSI_OPCODE_RFL:
+      break;
+   case TGSI_OPCODE_SEQ:
+      break;
+   case TGSI_OPCODE_SFL:
+      break;
+   case TGSI_OPCODE_SGT: {
+   }
+      break;
+   case TGSI_OPCODE_SIN: {
+   }
+      break;
+   case TGSI_OPCODE_SLE:
+      break;
+   case TGSI_OPCODE_SNE:
+      break;
+   case TGSI_OPCODE_STR:
+      break;
+   case TGSI_OPCODE_TEX:
+      break;
+   case TGSI_OPCODE_TXD:
+      break;
+   case TGSI_OPCODE_UP2H:
+      break;
+   case TGSI_OPCODE_UP2US:
+      break;
+   case TGSI_OPCODE_UP4B:
+      break;
+   case TGSI_OPCODE_UP4UB:
+      break;
+   case TGSI_OPCODE_X2D:
+      break;
+   case TGSI_OPCODE_ARA:
+      break;
+   case TGSI_OPCODE_ARR:
+      break;
+   case TGSI_OPCODE_BRA:
+      break;
+   case TGSI_OPCODE_CAL: {
+   }
+      break;
+   case TGSI_OPCODE_RET: {
+   }
+      break;
+   case TGSI_OPCODE_SSG:
+      break;
+   case TGSI_OPCODE_CMP: {
+   }
+      break;
+   case TGSI_OPCODE_SCS: {
+   }
+      break;
+   case TGSI_OPCODE_TXB:
+      break;
+   case TGSI_OPCODE_NRM:
+      break;
+   case TGSI_OPCODE_DIV:
+      break;
+   case TGSI_OPCODE_DP2:
+      break;
+   case TGSI_OPCODE_TXL:
+      break;
+   case TGSI_OPCODE_BRK: {
+   }
+      break;
+   case TGSI_OPCODE_IF: {
+   }
+      break;
+   case TGSI_OPCODE_LOOP:
+      break;
+   case TGSI_OPCODE_REP:
+      break;
+   case TGSI_OPCODE_ELSE: {
+   }
+      break;
+   case TGSI_OPCODE_ENDIF: {
+   }
+      break;
+   case TGSI_OPCODE_ENDLOOP:
+      break;
+   case TGSI_OPCODE_ENDREP:
+      break;
+   case TGSI_OPCODE_PUSHA:
+      break;
+   case TGSI_OPCODE_POPA:
+      break;
+   case TGSI_OPCODE_CEIL:
+      break;
+   case TGSI_OPCODE_I2F:
+      break;
+   case TGSI_OPCODE_NOT:
+      break;
+   case TGSI_OPCODE_TRUNC: {
+   }
+      break;
+   case TGSI_OPCODE_SHL:
+      break;
+   case TGSI_OPCODE_SHR:
+      break;
+   case TGSI_OPCODE_AND:
+      break;
+   case TGSI_OPCODE_OR:
+      break;
+   case TGSI_OPCODE_MOD:
+      break;
+   case TGSI_OPCODE_XOR:
+      break;
+   case TGSI_OPCODE_SAD:
+      break;
+   case TGSI_OPCODE_TXF:
+      break;
+   case TGSI_OPCODE_TXQ:
+      break;
+   case TGSI_OPCODE_CONT:
+      break;
+   case TGSI_OPCODE_EMIT:
+      break;
+   case TGSI_OPCODE_ENDPRIM:
+      break;
+   case TGSI_OPCODE_BGNLOOP2: {
+   }
+      break;
+   case TGSI_OPCODE_BGNSUB: {
+   }
+      break;
+   case TGSI_OPCODE_ENDLOOP2: {
+   }
+      break;
+   case TGSI_OPCODE_ENDSUB: {
+   }
+      break;
+   case TGSI_OPCODE_NOISE1:
+      break;
+   case TGSI_OPCODE_NOISE2:
+      break;
+   case TGSI_OPCODE_NOISE3:
+      break;
+   case TGSI_OPCODE_NOISE4:
+      break;
+   case TGSI_OPCODE_NOP:
+      break;
+   case TGSI_OPCODE_M4X3:
+      break;
+   case TGSI_OPCODE_M3X4:
+      break;
+   case TGSI_OPCODE_M3X3:
+      break;
+   case TGSI_OPCODE_M3X2:
+      break;
+   case TGSI_OPCODE_NRM4:
+      break;
+   case TGSI_OPCODE_CALLNZ:
+      break;
+   case TGSI_OPCODE_IFC:
+      break;
+   case TGSI_OPCODE_BREAKC:
+      break;
+   case TGSI_OPCODE_KIL: {
+   }
+      break;
+   case TGSI_OPCODE_END:
+      instr->end();
+      return;
+      break;
+   default:
+      fprintf(stderr, "ERROR: Unknown opcode %d\n",
+              inst->Instruction.Opcode);
+      assert(0);
+      break;
+   }
+
+   if (!out[0]) {
+      fprintf(stderr, "ERROR: unsupported opcode %d\n",
+              inst->Instruction.Opcode);
+      assert(!"Unsupported opcode");
+   }
+
+   /* store results  */
+   for (int i = 0; i < inst->Instruction.NumDstRegs; ++i) {
+      struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i];
+      storage->store((enum tgsi_file_type)dst->DstRegister.File,
+                     dst->DstRegister.Index, out, dst->DstRegister.WriteMask,
+		     instr->getIRBuilder() );
+   }
+}
+
+llvm::Module *
+tgsi_to_llvm(struct gallivm_ir *ir, const struct tgsi_token *tokens)
+{
+   llvm::Module *mod = new Module("shader");
+   struct tgsi_parse_context parse;
+   struct tgsi_full_instruction fi;
+   struct tgsi_full_declaration fd;
+   unsigned instno = 0;
+   Function* shader = mod->getFunction("execute_shader");
+   std::ostringstream stream;
+   if (ir->type == GALLIVM_VS) {
+      stream << "vs_shader";
+   } else {
+      stream << "fs_shader";
+   }
+   stream << ir->id;
+   std::string func_name = stream.str();
+   shader->setName(func_name.c_str());
+
+   Function::arg_iterator args = shader->arg_begin();
+   Value *ptr_INPUT = args++;
+   ptr_INPUT->setName("input");
+
+   BasicBlock *label_entry = BasicBlock::Create("entry", shader, 0);
+
+   tgsi_parse_init(&parse, tokens);
+
+   fi = tgsi_default_full_instruction();
+   fd = tgsi_default_full_declaration();
+   Storage storage(label_entry, ptr_INPUT);
+   Instructions instr(mod, shader, label_entry, &storage);
+   while(!tgsi_parse_end_of_tokens(&parse)) {
+      tgsi_parse_token(&parse);
+
+      switch (parse.FullToken.Token.Type) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         translate_declaration(ir, mod, &storage,
+                               &parse.FullToken.FullDeclaration,
+                               &fd);
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         translate_immediate(&storage,
+                             &parse.FullToken.FullImmediate);
+         break;
+
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         translate_instruction(mod, &storage, &instr,
+                               &parse.FullToken.FullInstruction,
+                               &fi, instno);
+         ++instno;
+         break;
+
+      default:
+         assert(0);
+      }
+   }
+
+   tgsi_parse_free(&parse);
+
+   ir->num_consts = storage.numConsts();
+   return mod;
+}
+
+llvm::Module * tgsi_to_llvmir(struct gallivm_ir *ir,
+                              const struct tgsi_token *tokens)
+{
+   llvm::Module *mod = new Module("shader");
+   struct tgsi_parse_context parse;
+   struct tgsi_full_instruction fi;
+   struct tgsi_full_declaration fd;
+   unsigned instno = 0;
+   std::ostringstream stream;
+   if (ir->type == GALLIVM_VS) {
+      stream << "vs_shader";
+   } else {
+      stream << "fs_shader";
+   }
+   //stream << ir->id;
+   std::string func_name = stream.str();
+   Function *shader = llvm::cast<Function>(mod->getOrInsertFunction(
+                                              func_name.c_str(),
+                                              vertexShaderFunctionType()));
+
+   Function::arg_iterator args = shader->arg_begin();
+   Value *input = args++;
+   input->setName("inputs");
+   Value *output = args++;
+   output->setName("outputs");
+   Value *consts = args++;
+   consts->setName("consts");
+
+   BasicBlock *label_entry = BasicBlock::Create("entry", shader, 0);
+
+   tgsi_parse_init(&parse, tokens);
+
+   fi = tgsi_default_full_instruction();
+   fd = tgsi_default_full_declaration();
+
+   StorageSoa storage(label_entry, input, output, consts);
+   InstructionsSoa instr(mod, shader, label_entry, &storage);
+
+   while(!tgsi_parse_end_of_tokens(&parse)) {
+      tgsi_parse_token(&parse);
+
+      switch (parse.FullToken.Token.Type) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         translate_declarationir(ir, mod, &storage,
+                                 &parse.FullToken.FullDeclaration,
+                                 &fd);
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         translate_immediateir(&storage,
+                             &parse.FullToken.FullImmediate);
+         break;
+
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         storage.declareImmediates();
+         translate_instructionir(mod, &storage, &instr,
+                                 &parse.FullToken.FullInstruction,
+                                 &fi, instno);
+         ++instno;
+         break;
+
+      default:
+         assert(0);
+      }
+   }
+
+   tgsi_parse_free(&parse);
+
+   return mod;
+}
diff --git a/src/gallium/auxiliary/gallivm/tgsitollvm.h b/src/gallium/auxiliary/gallivm/tgsitollvm.h
new file mode 100644
index 00000000000..7ada04d6299
--- /dev/null
+++ b/src/gallium/auxiliary/gallivm/tgsitollvm.h
@@ -0,0 +1,20 @@
+#ifndef TGSITOLLVM_H
+#define TGSITOLLVM_H
+
+
+namespace llvm {
+   class Module;
+}
+
+struct gallivm_ir;
+struct tgsi_token;
+
+
+llvm::Module * tgsi_to_llvm(struct gallivm_ir *ir,
+                            const struct tgsi_token *tokens);
+
+
+llvm::Module * tgsi_to_llvmir(struct gallivm_ir *ir,
+                              const struct tgsi_token *tokens);
+
+#endif
diff --git a/src/gallium/auxiliary/indices/Makefile b/src/gallium/auxiliary/indices/Makefile
new file mode 100644
index 00000000000..8fa61d265ea
--- /dev/null
+++ b/src/gallium/auxiliary/indices/Makefile
@@ -0,0 +1,16 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = indices
+
+C_SOURCES = \
+	u_indices_gen.c 
+
+include ../../Makefile.template
+
+u_indices_gen.c: u_indices_gen.py
+	python $< > $@
+
+
+symlinks:
+
diff --git a/src/gallium/auxiliary/indices/SConscript b/src/gallium/auxiliary/indices/SConscript
new file mode 100644
index 00000000000..65a43a9f648
--- /dev/null
+++ b/src/gallium/auxiliary/indices/SConscript
@@ -0,0 +1,17 @@
+Import('*')
+
+env.CodeGenerate(
+	target = 'u_indices_gen.c', 
+	script = 'u_indices_gen.py', 
+	source = [],
+	command = 'python $SCRIPT > $TARGET'
+)
+
+indices = env.ConvenienceLibrary(
+	target = 'indices',
+	source = [
+#               'u_indices.c',
+               'u_indices_gen.c',
+	])
+
+auxiliaries.insert(0, indices)
diff --git a/src/gallium/auxiliary/indices/u_indices.c b/src/gallium/auxiliary/indices/u_indices.c
new file mode 100644
index 00000000000..0cf7d88653c
--- /dev/null
+++ b/src/gallium/auxiliary/indices/u_indices.c
@@ -0,0 +1,253 @@
+/*
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#include "u_indices.h"
+#include "u_indices_priv.h"
+
+static void translate_memcpy_ushort( const void *in,
+                                     unsigned nr,
+                                     void *out )
+{
+   memcpy(out, in, nr*sizeof(short));
+}
+                              
+static void translate_memcpy_uint( const void *in,
+                                   unsigned nr,
+                                   void *out )
+{
+   memcpy(out, in, nr*sizeof(int));
+}
+                              
+
+int u_index_translator( unsigned hw_mask,
+                        unsigned prim,
+                        unsigned in_index_size,
+                        unsigned nr,
+                        unsigned in_pv,
+                        unsigned out_pv,
+                        unsigned *out_prim,
+                        unsigned *out_index_size,
+                        unsigned *out_nr,
+                        u_translate_func *out_translate )
+{
+   unsigned in_idx;
+   unsigned out_idx;
+   int ret = U_TRANSLATE_NORMAL;
+
+   u_index_init();
+
+   in_idx = in_size_idx(in_index_size);
+   *out_index_size = (in_index_size == 4) ? 4 : 2;
+   out_idx = out_size_idx(*out_index_size);
+
+   if ((hw_mask & (1<<prim)) && 
+       in_index_size == *out_index_size &&
+       in_pv == out_pv) 
+   {
+      if (in_index_size == 4)
+         *out_translate = translate_memcpy_uint;
+      else
+         *out_translate = translate_memcpy_ushort;
+
+      *out_prim = prim;
+      *out_nr = nr;
+
+      return U_TRANSLATE_MEMCPY;
+   }
+   else {
+      switch (prim) {
+      case PIPE_PRIM_POINTS:
+         *out_translate = translate[in_idx][out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_POINTS;
+         *out_nr = nr;
+         break;
+
+      case PIPE_PRIM_LINES:
+         *out_translate = translate[in_idx][out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_LINES;
+         *out_nr = nr;
+         break;
+
+      case PIPE_PRIM_LINE_STRIP:
+         *out_translate = translate[in_idx][out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_LINES;
+         *out_nr = (nr - 1) * 2;
+         break;
+
+      case PIPE_PRIM_LINE_LOOP:
+         *out_translate = translate[in_idx][out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_LINES;
+         *out_nr = nr * 2;
+         break;
+
+      case PIPE_PRIM_TRIANGLES:
+         *out_translate = translate[in_idx][out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_TRIANGLES;
+         *out_nr = nr;
+         break;
+
+      case PIPE_PRIM_TRIANGLE_STRIP:
+         *out_translate = translate[in_idx][out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_TRIANGLES;
+         *out_nr = (nr - 2) * 3;
+         break;
+
+      case PIPE_PRIM_TRIANGLE_FAN:
+         *out_translate = translate[in_idx][out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_TRIANGLES;
+         *out_nr = (nr - 2) * 3;
+         break;
+
+      case PIPE_PRIM_QUADS:
+         *out_translate = translate[in_idx][out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_TRIANGLES;
+         *out_nr = (nr / 4) * 6;
+         break;
+
+      case PIPE_PRIM_QUAD_STRIP:
+         *out_translate = translate[in_idx][out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_TRIANGLES;
+         *out_nr = (nr - 2) * 3;
+         break;
+
+      case PIPE_PRIM_POLYGON:
+         *out_translate = translate[in_idx][out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_TRIANGLES;
+         *out_nr = (nr - 2) * 3;
+         break;
+
+      default:
+         assert(0);
+         *out_translate = translate[in_idx][out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_POINTS;
+         *out_nr = nr;
+         return U_TRANSLATE_ERROR;
+      }
+   }
+
+   return ret;
+}
+
+
+
+
+
+int u_index_generator( unsigned hw_mask,
+                       unsigned prim,
+                       unsigned start,
+                       unsigned nr,
+                       unsigned in_pv,
+                       unsigned out_pv,
+                       unsigned *out_prim,
+                       unsigned *out_index_size,
+                       unsigned *out_nr,
+                       u_generate_func *out_generate )
+
+{
+   unsigned out_idx;
+
+   u_index_init();
+
+   *out_index_size = ((start + nr) > 0xfffe) ? 4 : 2;
+   out_idx = out_size_idx(*out_index_size);
+
+   if ((hw_mask & (1<<prim)) && 
+       (in_pv == out_pv)) {
+       
+      *out_generate = generate[out_idx][in_pv][out_pv][PIPE_PRIM_POINTS];
+      *out_prim = prim;
+      *out_nr = nr;
+      return U_GENERATE_LINEAR;
+   }
+   else {
+      switch (prim) {
+      case PIPE_PRIM_POINTS:
+         *out_generate = generate[out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_POINTS;
+         *out_nr = nr;
+         return U_GENERATE_REUSABLE;
+
+      case PIPE_PRIM_LINES:
+         *out_generate = generate[out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_LINES;
+         *out_nr = nr;
+         return U_GENERATE_REUSABLE;
+
+      case PIPE_PRIM_LINE_STRIP:
+         *out_generate = generate[out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_LINES;
+         *out_nr = (nr - 1) * 2;
+         return U_GENERATE_REUSABLE;
+
+      case PIPE_PRIM_LINE_LOOP:
+         *out_generate = generate[out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_LINES;
+         *out_nr = nr * 2;
+         return U_GENERATE_ONE_OFF;
+
+      case PIPE_PRIM_TRIANGLES:
+         *out_generate = generate[out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_TRIANGLES;
+         *out_nr = nr;
+         return U_GENERATE_REUSABLE;
+
+      case PIPE_PRIM_TRIANGLE_STRIP:
+         *out_generate = generate[out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_TRIANGLES;
+         *out_nr = (nr - 2) * 3;
+         return U_GENERATE_REUSABLE;
+
+      case PIPE_PRIM_TRIANGLE_FAN:
+         *out_generate = generate[out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_TRIANGLES;
+         *out_nr = (nr - 2) * 3;
+         return U_GENERATE_REUSABLE;
+
+      case PIPE_PRIM_QUADS:
+         *out_generate = generate[out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_TRIANGLES;
+         *out_nr = (nr / 4) * 6;
+         return U_GENERATE_REUSABLE;
+
+      case PIPE_PRIM_QUAD_STRIP:
+         *out_generate = generate[out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_TRIANGLES;
+         *out_nr = (nr - 2) * 3;
+         return U_GENERATE_REUSABLE;
+
+      case PIPE_PRIM_POLYGON:
+         *out_generate = generate[out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_TRIANGLES;
+         *out_nr = (nr - 2) * 3;
+         return U_GENERATE_REUSABLE;
+
+      default:
+         assert(0);
+         *out_generate = generate[out_idx][in_pv][out_pv][prim];
+         *out_prim = PIPE_PRIM_POINTS;
+         *out_nr = nr;
+         return U_TRANSLATE_ERROR;
+      }
+   }
+}
diff --git a/src/gallium/auxiliary/indices/u_indices.h b/src/gallium/auxiliary/indices/u_indices.h
new file mode 100644
index 00000000000..abf5a3037d1
--- /dev/null
+++ b/src/gallium/auxiliary/indices/u_indices.h
@@ -0,0 +1,83 @@
+/*
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#ifndef U_INDICES_H
+#define U_INDICES_H
+
+#include "pipe/p_compiler.h"
+
+#define PV_FIRST      0
+#define PV_LAST       1
+#define PV_COUNT      2
+
+typedef void (*u_translate_func)( const void *in,
+                                  unsigned nr,
+                                  void *out );
+
+typedef void (*u_generate_func)( unsigned nr,
+                                 void *out );
+
+
+/* Return codes describe the translate/generate operation.  Caller may
+ * be able to reuse translated indices under some circumstances.
+ */
+#define U_TRANSLATE_ERROR  -1
+#define U_TRANSLATE_NORMAL  1
+#define U_TRANSLATE_MEMCPY  2
+#define U_GENERATE_LINEAR   3
+#define U_GENERATE_REUSABLE 4
+#define U_GENERATE_ONE_OFF  5
+
+
+void u_index_init( void );
+
+int u_index_translator( unsigned hw_mask,
+                        unsigned prim,
+                        unsigned in_index_size,
+                        unsigned nr,
+                        unsigned in_pv,   /* API */
+                        unsigned out_pv,  /* hardware */
+                        unsigned *out_prim,
+                        unsigned *out_index_size,
+                        unsigned *out_nr,
+                        u_translate_func *out_translate );
+
+/* Note that even when generating it is necessary to know what the
+ * API's PV is, as the indices generated will depend on whether it is
+ * the same as hardware or not, and in the case of triangle strips,
+ * whether it is first or last.
+ */
+int u_index_generator( unsigned hw_mask,
+                       unsigned prim,
+                       unsigned start,
+                       unsigned nr,
+                       unsigned in_pv,   /* API */
+                       unsigned out_pv,  /* hardware */
+                       unsigned *out_prim,
+                       unsigned *out_index_size,
+                       unsigned *out_nr,
+                       u_generate_func *out_generate );
+
+
+#endif
diff --git a/src/gallium/auxiliary/indices/u_indices_gen.c b/src/gallium/auxiliary/indices/u_indices_gen.c
new file mode 100644
index 00000000000..4c05b3eedb3
--- /dev/null
+++ b/src/gallium/auxiliary/indices/u_indices_gen.c
@@ -0,0 +1,5129 @@
+/* File automatically generated by indices.py */
+
+/*
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+
+/**
+ * @file
+ * Functions to translate and generate index lists
+ */
+
+#include "indices/u_indices.h"
+#include "indices/u_indices_priv.h"
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_defines.h"
+#include "util/u_memory.h"
+
+
+static unsigned out_size_idx( unsigned index_size )
+{
+   switch (index_size) {
+   case 4: return OUT_UINT;
+   case 2: return OUT_USHORT;
+   default: assert(0); return OUT_USHORT;
+   }
+}
+
+static unsigned in_size_idx( unsigned index_size )
+{
+   switch (index_size) {
+   case 4: return IN_UINT;
+   case 2: return IN_USHORT;
+   case 1: return IN_UBYTE;
+   default: assert(0); return IN_UBYTE;
+   }
+}
+
+
+static u_translate_func translate[IN_COUNT][OUT_COUNT][PV_COUNT][PV_COUNT][PRIM_COUNT];
+static u_generate_func  generate[OUT_COUNT][PV_COUNT][PV_COUNT][PRIM_COUNT];
+
+
+
+static void generate_points_ushort_first2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (ushort)(i);
+   }
+}
+static void generate_lines_ushort_first2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (ushort)(i);
+      (out+i)[1] = (ushort)(i+1);
+   }
+}
+static void generate_linestrip_ushort_first2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (ushort)(i);
+      (out+j)[1] = (ushort)(i+1);
+   }
+}
+static void generate_lineloop_ushort_first2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (ushort)(i);
+      (out+j)[1] = (ushort)(i+1);
+   }
+      (out+j)[0] = (ushort)(i);
+      (out+j)[1] = (ushort)(0);
+}
+static void generate_tris_ushort_first2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (ushort)(i);
+      (out+i)[1] = (ushort)(i+1);
+      (out+i)[2] = (ushort)(i+2);
+   }
+}
+static void generate_tristrip_ushort_first2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)(i);
+      (out+j)[1] = (ushort)(i+1+(i&1));
+      (out+j)[2] = (ushort)(i+2-(i&1));
+   }
+}
+static void generate_trifan_ushort_first2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)(0);
+      (out+j)[1] = (ushort)(i+1);
+      (out+j)[2] = (ushort)(i+2);
+   }
+}
+static void generate_quads_ushort_first2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (ushort)(i+0);
+      (out+j+0)[1] = (ushort)(i+1);
+      (out+j+0)[2] = (ushort)(i+3);
+      (out+j+3)[0] = (ushort)(i+1);
+      (out+j+3)[1] = (ushort)(i+2);
+      (out+j+3)[2] = (ushort)(i+3);
+   }
+}
+static void generate_quadstrip_ushort_first2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (ushort)(i+2);
+      (out+j+0)[1] = (ushort)(i+0);
+      (out+j+0)[2] = (ushort)(i+3);
+      (out+j+3)[0] = (ushort)(i+0);
+      (out+j+3)[1] = (ushort)(i+1);
+      (out+j+3)[2] = (ushort)(i+3);
+   }
+}
+static void generate_polygon_ushort_first2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)(0);
+      (out+j)[1] = (ushort)(i+1);
+      (out+j)[2] = (ushort)(i+2);
+   }
+}
+static void generate_points_ushort_first2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (ushort)(i);
+   }
+}
+static void generate_lines_ushort_first2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (ushort)(i+1);
+      (out+i)[1] = (ushort)(i);
+   }
+}
+static void generate_linestrip_ushort_first2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (ushort)(i+1);
+      (out+j)[1] = (ushort)(i);
+   }
+}
+static void generate_lineloop_ushort_first2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (ushort)(i+1);
+      (out+j)[1] = (ushort)(i);
+   }
+      (out+j)[0] = (ushort)(0);
+      (out+j)[1] = (ushort)(i);
+}
+static void generate_tris_ushort_first2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (ushort)(i+1);
+      (out+i)[1] = (ushort)(i+2);
+      (out+i)[2] = (ushort)(i);
+   }
+}
+static void generate_tristrip_ushort_first2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)(i+1+(i&1));
+      (out+j)[1] = (ushort)(i+2-(i&1));
+      (out+j)[2] = (ushort)(i);
+   }
+}
+static void generate_trifan_ushort_first2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)(i+1);
+      (out+j)[1] = (ushort)(i+2);
+      (out+j)[2] = (ushort)(0);
+   }
+}
+static void generate_quads_ushort_first2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (ushort)(i+1);
+      (out+j+0)[1] = (ushort)(i+3);
+      (out+j+0)[2] = (ushort)(i+0);
+      (out+j+3)[0] = (ushort)(i+2);
+      (out+j+3)[1] = (ushort)(i+3);
+      (out+j+3)[2] = (ushort)(i+1);
+   }
+}
+static void generate_quadstrip_ushort_first2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (ushort)(i+0);
+      (out+j+0)[1] = (ushort)(i+3);
+      (out+j+0)[2] = (ushort)(i+2);
+      (out+j+3)[0] = (ushort)(i+1);
+      (out+j+3)[1] = (ushort)(i+3);
+      (out+j+3)[2] = (ushort)(i+0);
+   }
+}
+static void generate_polygon_ushort_first2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)(i+1);
+      (out+j)[1] = (ushort)(i+2);
+      (out+j)[2] = (ushort)(0);
+   }
+}
+static void generate_points_ushort_last2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (ushort)(i);
+   }
+}
+static void generate_lines_ushort_last2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (ushort)(i+1);
+      (out+i)[1] = (ushort)(i);
+   }
+}
+static void generate_linestrip_ushort_last2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (ushort)(i+1);
+      (out+j)[1] = (ushort)(i);
+   }
+}
+static void generate_lineloop_ushort_last2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (ushort)(i+1);
+      (out+j)[1] = (ushort)(i);
+   }
+      (out+j)[0] = (ushort)(0);
+      (out+j)[1] = (ushort)(i);
+}
+static void generate_tris_ushort_last2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (ushort)(i+2);
+      (out+i)[1] = (ushort)(i);
+      (out+i)[2] = (ushort)(i+1);
+   }
+}
+static void generate_tristrip_ushort_last2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)(i+2);
+      (out+j)[1] = (ushort)(i+(i&1));
+      (out+j)[2] = (ushort)(i+1-(i&1));
+   }
+}
+static void generate_trifan_ushort_last2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)(i+2);
+      (out+j)[1] = (ushort)(0);
+      (out+j)[2] = (ushort)(i+1);
+   }
+}
+static void generate_quads_ushort_last2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (ushort)(i+3);
+      (out+j+0)[1] = (ushort)(i+0);
+      (out+j+0)[2] = (ushort)(i+1);
+      (out+j+3)[0] = (ushort)(i+3);
+      (out+j+3)[1] = (ushort)(i+1);
+      (out+j+3)[2] = (ushort)(i+2);
+   }
+}
+static void generate_quadstrip_ushort_last2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (ushort)(i+3);
+      (out+j+0)[1] = (ushort)(i+2);
+      (out+j+0)[2] = (ushort)(i+0);
+      (out+j+3)[0] = (ushort)(i+3);
+      (out+j+3)[1] = (ushort)(i+0);
+      (out+j+3)[2] = (ushort)(i+1);
+   }
+}
+static void generate_polygon_ushort_last2first(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)(0);
+      (out+j)[1] = (ushort)(i+1);
+      (out+j)[2] = (ushort)(i+2);
+   }
+}
+static void generate_points_ushort_last2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (ushort)(i);
+   }
+}
+static void generate_lines_ushort_last2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (ushort)(i);
+      (out+i)[1] = (ushort)(i+1);
+   }
+}
+static void generate_linestrip_ushort_last2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (ushort)(i);
+      (out+j)[1] = (ushort)(i+1);
+   }
+}
+static void generate_lineloop_ushort_last2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (ushort)(i);
+      (out+j)[1] = (ushort)(i+1);
+   }
+      (out+j)[0] = (ushort)(i);
+      (out+j)[1] = (ushort)(0);
+}
+static void generate_tris_ushort_last2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (ushort)(i);
+      (out+i)[1] = (ushort)(i+1);
+      (out+i)[2] = (ushort)(i+2);
+   }
+}
+static void generate_tristrip_ushort_last2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)(i+(i&1));
+      (out+j)[1] = (ushort)(i+1-(i&1));
+      (out+j)[2] = (ushort)(i+2);
+   }
+}
+static void generate_trifan_ushort_last2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)(0);
+      (out+j)[1] = (ushort)(i+1);
+      (out+j)[2] = (ushort)(i+2);
+   }
+}
+static void generate_quads_ushort_last2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (ushort)(i+0);
+      (out+j+0)[1] = (ushort)(i+1);
+      (out+j+0)[2] = (ushort)(i+3);
+      (out+j+3)[0] = (ushort)(i+1);
+      (out+j+3)[1] = (ushort)(i+2);
+      (out+j+3)[2] = (ushort)(i+3);
+   }
+}
+static void generate_quadstrip_ushort_last2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (ushort)(i+2);
+      (out+j+0)[1] = (ushort)(i+0);
+      (out+j+0)[2] = (ushort)(i+3);
+      (out+j+3)[0] = (ushort)(i+0);
+      (out+j+3)[1] = (ushort)(i+1);
+      (out+j+3)[2] = (ushort)(i+3);
+   }
+}
+static void generate_polygon_ushort_last2last(
+    unsigned nr,
+    void *_out )
+{
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)(i+1);
+      (out+j)[1] = (ushort)(i+2);
+      (out+j)[2] = (ushort)(0);
+   }
+}
+static void generate_points_uint_first2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (uint)(i);
+   }
+}
+static void generate_lines_uint_first2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (uint)(i);
+      (out+i)[1] = (uint)(i+1);
+   }
+}
+static void generate_linestrip_uint_first2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (uint)(i);
+      (out+j)[1] = (uint)(i+1);
+   }
+}
+static void generate_lineloop_uint_first2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (uint)(i);
+      (out+j)[1] = (uint)(i+1);
+   }
+      (out+j)[0] = (uint)(i);
+      (out+j)[1] = (uint)(0);
+}
+static void generate_tris_uint_first2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (uint)(i);
+      (out+i)[1] = (uint)(i+1);
+      (out+i)[2] = (uint)(i+2);
+   }
+}
+static void generate_tristrip_uint_first2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)(i);
+      (out+j)[1] = (uint)(i+1+(i&1));
+      (out+j)[2] = (uint)(i+2-(i&1));
+   }
+}
+static void generate_trifan_uint_first2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)(0);
+      (out+j)[1] = (uint)(i+1);
+      (out+j)[2] = (uint)(i+2);
+   }
+}
+static void generate_quads_uint_first2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (uint)(i+0);
+      (out+j+0)[1] = (uint)(i+1);
+      (out+j+0)[2] = (uint)(i+3);
+      (out+j+3)[0] = (uint)(i+1);
+      (out+j+3)[1] = (uint)(i+2);
+      (out+j+3)[2] = (uint)(i+3);
+   }
+}
+static void generate_quadstrip_uint_first2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (uint)(i+2);
+      (out+j+0)[1] = (uint)(i+0);
+      (out+j+0)[2] = (uint)(i+3);
+      (out+j+3)[0] = (uint)(i+0);
+      (out+j+3)[1] = (uint)(i+1);
+      (out+j+3)[2] = (uint)(i+3);
+   }
+}
+static void generate_polygon_uint_first2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)(0);
+      (out+j)[1] = (uint)(i+1);
+      (out+j)[2] = (uint)(i+2);
+   }
+}
+static void generate_points_uint_first2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (uint)(i);
+   }
+}
+static void generate_lines_uint_first2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (uint)(i+1);
+      (out+i)[1] = (uint)(i);
+   }
+}
+static void generate_linestrip_uint_first2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (uint)(i+1);
+      (out+j)[1] = (uint)(i);
+   }
+}
+static void generate_lineloop_uint_first2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (uint)(i+1);
+      (out+j)[1] = (uint)(i);
+   }
+      (out+j)[0] = (uint)(0);
+      (out+j)[1] = (uint)(i);
+}
+static void generate_tris_uint_first2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (uint)(i+1);
+      (out+i)[1] = (uint)(i+2);
+      (out+i)[2] = (uint)(i);
+   }
+}
+static void generate_tristrip_uint_first2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)(i+1+(i&1));
+      (out+j)[1] = (uint)(i+2-(i&1));
+      (out+j)[2] = (uint)(i);
+   }
+}
+static void generate_trifan_uint_first2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)(i+1);
+      (out+j)[1] = (uint)(i+2);
+      (out+j)[2] = (uint)(0);
+   }
+}
+static void generate_quads_uint_first2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (uint)(i+1);
+      (out+j+0)[1] = (uint)(i+3);
+      (out+j+0)[2] = (uint)(i+0);
+      (out+j+3)[0] = (uint)(i+2);
+      (out+j+3)[1] = (uint)(i+3);
+      (out+j+3)[2] = (uint)(i+1);
+   }
+}
+static void generate_quadstrip_uint_first2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (uint)(i+0);
+      (out+j+0)[1] = (uint)(i+3);
+      (out+j+0)[2] = (uint)(i+2);
+      (out+j+3)[0] = (uint)(i+1);
+      (out+j+3)[1] = (uint)(i+3);
+      (out+j+3)[2] = (uint)(i+0);
+   }
+}
+static void generate_polygon_uint_first2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)(i+1);
+      (out+j)[1] = (uint)(i+2);
+      (out+j)[2] = (uint)(0);
+   }
+}
+static void generate_points_uint_last2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (uint)(i);
+   }
+}
+static void generate_lines_uint_last2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (uint)(i+1);
+      (out+i)[1] = (uint)(i);
+   }
+}
+static void generate_linestrip_uint_last2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (uint)(i+1);
+      (out+j)[1] = (uint)(i);
+   }
+}
+static void generate_lineloop_uint_last2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (uint)(i+1);
+      (out+j)[1] = (uint)(i);
+   }
+      (out+j)[0] = (uint)(0);
+      (out+j)[1] = (uint)(i);
+}
+static void generate_tris_uint_last2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (uint)(i+2);
+      (out+i)[1] = (uint)(i);
+      (out+i)[2] = (uint)(i+1);
+   }
+}
+static void generate_tristrip_uint_last2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)(i+2);
+      (out+j)[1] = (uint)(i+(i&1));
+      (out+j)[2] = (uint)(i+1-(i&1));
+   }
+}
+static void generate_trifan_uint_last2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)(i+2);
+      (out+j)[1] = (uint)(0);
+      (out+j)[2] = (uint)(i+1);
+   }
+}
+static void generate_quads_uint_last2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (uint)(i+3);
+      (out+j+0)[1] = (uint)(i+0);
+      (out+j+0)[2] = (uint)(i+1);
+      (out+j+3)[0] = (uint)(i+3);
+      (out+j+3)[1] = (uint)(i+1);
+      (out+j+3)[2] = (uint)(i+2);
+   }
+}
+static void generate_quadstrip_uint_last2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (uint)(i+3);
+      (out+j+0)[1] = (uint)(i+2);
+      (out+j+0)[2] = (uint)(i+0);
+      (out+j+3)[0] = (uint)(i+3);
+      (out+j+3)[1] = (uint)(i+0);
+      (out+j+3)[2] = (uint)(i+1);
+   }
+}
+static void generate_polygon_uint_last2first(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)(0);
+      (out+j)[1] = (uint)(i+1);
+      (out+j)[2] = (uint)(i+2);
+   }
+}
+static void generate_points_uint_last2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (uint)(i);
+   }
+}
+static void generate_lines_uint_last2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (uint)(i);
+      (out+i)[1] = (uint)(i+1);
+   }
+}
+static void generate_linestrip_uint_last2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (uint)(i);
+      (out+j)[1] = (uint)(i+1);
+   }
+}
+static void generate_lineloop_uint_last2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (uint)(i);
+      (out+j)[1] = (uint)(i+1);
+   }
+      (out+j)[0] = (uint)(i);
+      (out+j)[1] = (uint)(0);
+}
+static void generate_tris_uint_last2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (uint)(i);
+      (out+i)[1] = (uint)(i+1);
+      (out+i)[2] = (uint)(i+2);
+   }
+}
+static void generate_tristrip_uint_last2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)(i+(i&1));
+      (out+j)[1] = (uint)(i+1-(i&1));
+      (out+j)[2] = (uint)(i+2);
+   }
+}
+static void generate_trifan_uint_last2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)(0);
+      (out+j)[1] = (uint)(i+1);
+      (out+j)[2] = (uint)(i+2);
+   }
+}
+static void generate_quads_uint_last2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (uint)(i+0);
+      (out+j+0)[1] = (uint)(i+1);
+      (out+j+0)[2] = (uint)(i+3);
+      (out+j+3)[0] = (uint)(i+1);
+      (out+j+3)[1] = (uint)(i+2);
+      (out+j+3)[2] = (uint)(i+3);
+   }
+}
+static void generate_quadstrip_uint_last2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (uint)(i+2);
+      (out+j+0)[1] = (uint)(i+0);
+      (out+j+0)[2] = (uint)(i+3);
+      (out+j+3)[0] = (uint)(i+0);
+      (out+j+3)[1] = (uint)(i+1);
+      (out+j+3)[2] = (uint)(i+3);
+   }
+}
+static void generate_polygon_uint_last2last(
+    unsigned nr,
+    void *_out )
+{
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)(i+1);
+      (out+j)[1] = (uint)(i+2);
+      (out+j)[2] = (uint)(0);
+   }
+}
+static void translate_points_ubyte2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (ushort)in[i];
+   }
+}
+static void translate_lines_ubyte2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (ushort)in[i];
+      (out+i)[1] = (ushort)in[i+1];
+   }
+}
+static void translate_linestrip_ubyte2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[i+1];
+   }
+}
+static void translate_lineloop_ubyte2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[i+1];
+   }
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[0];
+}
+static void translate_tris_ubyte2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (ushort)in[i];
+      (out+i)[1] = (ushort)in[i+1];
+      (out+i)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_tristrip_ubyte2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[i+1+(i&1)];
+      (out+j)[2] = (ushort)in[i+2-(i&1)];
+   }
+}
+static void translate_trifan_ubyte2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i+1];
+      (out+j)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_quads_ubyte2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (ushort)in[i+0];
+      (out+j+0)[1] = (ushort)in[i+1];
+      (out+j+0)[2] = (ushort)in[i+3];
+      (out+j+3)[0] = (ushort)in[i+1];
+      (out+j+3)[1] = (ushort)in[i+2];
+      (out+j+3)[2] = (ushort)in[i+3];
+   }
+}
+static void translate_quadstrip_ubyte2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (ushort)in[i+2];
+      (out+j+0)[1] = (ushort)in[i+0];
+      (out+j+0)[2] = (ushort)in[i+3];
+      (out+j+3)[0] = (ushort)in[i+0];
+      (out+j+3)[1] = (ushort)in[i+1];
+      (out+j+3)[2] = (ushort)in[i+3];
+   }
+}
+static void translate_polygon_ubyte2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i+1];
+      (out+j)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_points_ubyte2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (ushort)in[i];
+   }
+}
+static void translate_lines_ubyte2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (ushort)in[i+1];
+      (out+i)[1] = (ushort)in[i];
+   }
+}
+static void translate_linestrip_ubyte2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i];
+   }
+}
+static void translate_lineloop_ubyte2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i];
+   }
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i];
+}
+static void translate_tris_ubyte2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (ushort)in[i+1];
+      (out+i)[1] = (ushort)in[i+2];
+      (out+i)[2] = (ushort)in[i];
+   }
+}
+static void translate_tristrip_ubyte2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+1+(i&1)];
+      (out+j)[1] = (ushort)in[i+2-(i&1)];
+      (out+j)[2] = (ushort)in[i];
+   }
+}
+static void translate_trifan_ubyte2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i+2];
+      (out+j)[2] = (ushort)in[0];
+   }
+}
+static void translate_quads_ubyte2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (ushort)in[i+1];
+      (out+j+0)[1] = (ushort)in[i+3];
+      (out+j+0)[2] = (ushort)in[i+0];
+      (out+j+3)[0] = (ushort)in[i+2];
+      (out+j+3)[1] = (ushort)in[i+3];
+      (out+j+3)[2] = (ushort)in[i+1];
+   }
+}
+static void translate_quadstrip_ubyte2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (ushort)in[i+0];
+      (out+j+0)[1] = (ushort)in[i+3];
+      (out+j+0)[2] = (ushort)in[i+2];
+      (out+j+3)[0] = (ushort)in[i+1];
+      (out+j+3)[1] = (ushort)in[i+3];
+      (out+j+3)[2] = (ushort)in[i+0];
+   }
+}
+static void translate_polygon_ubyte2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i+2];
+      (out+j)[2] = (ushort)in[0];
+   }
+}
+static void translate_points_ubyte2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (ushort)in[i];
+   }
+}
+static void translate_lines_ubyte2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (ushort)in[i+1];
+      (out+i)[1] = (ushort)in[i];
+   }
+}
+static void translate_linestrip_ubyte2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i];
+   }
+}
+static void translate_lineloop_ubyte2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i];
+   }
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i];
+}
+static void translate_tris_ubyte2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (ushort)in[i+2];
+      (out+i)[1] = (ushort)in[i];
+      (out+i)[2] = (ushort)in[i+1];
+   }
+}
+static void translate_tristrip_ubyte2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+2];
+      (out+j)[1] = (ushort)in[i+(i&1)];
+      (out+j)[2] = (ushort)in[i+1-(i&1)];
+   }
+}
+static void translate_trifan_ubyte2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+2];
+      (out+j)[1] = (ushort)in[0];
+      (out+j)[2] = (ushort)in[i+1];
+   }
+}
+static void translate_quads_ubyte2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (ushort)in[i+3];
+      (out+j+0)[1] = (ushort)in[i+0];
+      (out+j+0)[2] = (ushort)in[i+1];
+      (out+j+3)[0] = (ushort)in[i+3];
+      (out+j+3)[1] = (ushort)in[i+1];
+      (out+j+3)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_quadstrip_ubyte2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (ushort)in[i+3];
+      (out+j+0)[1] = (ushort)in[i+2];
+      (out+j+0)[2] = (ushort)in[i+0];
+      (out+j+3)[0] = (ushort)in[i+3];
+      (out+j+3)[1] = (ushort)in[i+0];
+      (out+j+3)[2] = (ushort)in[i+1];
+   }
+}
+static void translate_polygon_ubyte2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i+1];
+      (out+j)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_points_ubyte2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (ushort)in[i];
+   }
+}
+static void translate_lines_ubyte2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (ushort)in[i];
+      (out+i)[1] = (ushort)in[i+1];
+   }
+}
+static void translate_linestrip_ubyte2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[i+1];
+   }
+}
+static void translate_lineloop_ubyte2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[i+1];
+   }
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[0];
+}
+static void translate_tris_ubyte2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (ushort)in[i];
+      (out+i)[1] = (ushort)in[i+1];
+      (out+i)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_tristrip_ubyte2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+(i&1)];
+      (out+j)[1] = (ushort)in[i+1-(i&1)];
+      (out+j)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_trifan_ubyte2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i+1];
+      (out+j)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_quads_ubyte2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (ushort)in[i+0];
+      (out+j+0)[1] = (ushort)in[i+1];
+      (out+j+0)[2] = (ushort)in[i+3];
+      (out+j+3)[0] = (ushort)in[i+1];
+      (out+j+3)[1] = (ushort)in[i+2];
+      (out+j+3)[2] = (ushort)in[i+3];
+   }
+}
+static void translate_quadstrip_ubyte2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (ushort)in[i+2];
+      (out+j+0)[1] = (ushort)in[i+0];
+      (out+j+0)[2] = (ushort)in[i+3];
+      (out+j+3)[0] = (ushort)in[i+0];
+      (out+j+3)[1] = (ushort)in[i+1];
+      (out+j+3)[2] = (ushort)in[i+3];
+   }
+}
+static void translate_polygon_ubyte2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i+2];
+      (out+j)[2] = (ushort)in[0];
+   }
+}
+static void translate_points_ubyte2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (uint)in[i];
+   }
+}
+static void translate_lines_ubyte2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (uint)in[i];
+      (out+i)[1] = (uint)in[i+1];
+   }
+}
+static void translate_linestrip_ubyte2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[i+1];
+   }
+}
+static void translate_lineloop_ubyte2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[i+1];
+   }
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[0];
+}
+static void translate_tris_ubyte2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (uint)in[i];
+      (out+i)[1] = (uint)in[i+1];
+      (out+i)[2] = (uint)in[i+2];
+   }
+}
+static void translate_tristrip_ubyte2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[i+1+(i&1)];
+      (out+j)[2] = (uint)in[i+2-(i&1)];
+   }
+}
+static void translate_trifan_ubyte2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i+1];
+      (out+j)[2] = (uint)in[i+2];
+   }
+}
+static void translate_quads_ubyte2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (uint)in[i+0];
+      (out+j+0)[1] = (uint)in[i+1];
+      (out+j+0)[2] = (uint)in[i+3];
+      (out+j+3)[0] = (uint)in[i+1];
+      (out+j+3)[1] = (uint)in[i+2];
+      (out+j+3)[2] = (uint)in[i+3];
+   }
+}
+static void translate_quadstrip_ubyte2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (uint)in[i+2];
+      (out+j+0)[1] = (uint)in[i+0];
+      (out+j+0)[2] = (uint)in[i+3];
+      (out+j+3)[0] = (uint)in[i+0];
+      (out+j+3)[1] = (uint)in[i+1];
+      (out+j+3)[2] = (uint)in[i+3];
+   }
+}
+static void translate_polygon_ubyte2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i+1];
+      (out+j)[2] = (uint)in[i+2];
+   }
+}
+static void translate_points_ubyte2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (uint)in[i];
+   }
+}
+static void translate_lines_ubyte2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (uint)in[i+1];
+      (out+i)[1] = (uint)in[i];
+   }
+}
+static void translate_linestrip_ubyte2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i];
+   }
+}
+static void translate_lineloop_ubyte2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i];
+   }
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i];
+}
+static void translate_tris_ubyte2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (uint)in[i+1];
+      (out+i)[1] = (uint)in[i+2];
+      (out+i)[2] = (uint)in[i];
+   }
+}
+static void translate_tristrip_ubyte2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+1+(i&1)];
+      (out+j)[1] = (uint)in[i+2-(i&1)];
+      (out+j)[2] = (uint)in[i];
+   }
+}
+static void translate_trifan_ubyte2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i+2];
+      (out+j)[2] = (uint)in[0];
+   }
+}
+static void translate_quads_ubyte2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (uint)in[i+1];
+      (out+j+0)[1] = (uint)in[i+3];
+      (out+j+0)[2] = (uint)in[i+0];
+      (out+j+3)[0] = (uint)in[i+2];
+      (out+j+3)[1] = (uint)in[i+3];
+      (out+j+3)[2] = (uint)in[i+1];
+   }
+}
+static void translate_quadstrip_ubyte2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (uint)in[i+0];
+      (out+j+0)[1] = (uint)in[i+3];
+      (out+j+0)[2] = (uint)in[i+2];
+      (out+j+3)[0] = (uint)in[i+1];
+      (out+j+3)[1] = (uint)in[i+3];
+      (out+j+3)[2] = (uint)in[i+0];
+   }
+}
+static void translate_polygon_ubyte2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i+2];
+      (out+j)[2] = (uint)in[0];
+   }
+}
+static void translate_points_ubyte2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (uint)in[i];
+   }
+}
+static void translate_lines_ubyte2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (uint)in[i+1];
+      (out+i)[1] = (uint)in[i];
+   }
+}
+static void translate_linestrip_ubyte2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i];
+   }
+}
+static void translate_lineloop_ubyte2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i];
+   }
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i];
+}
+static void translate_tris_ubyte2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (uint)in[i+2];
+      (out+i)[1] = (uint)in[i];
+      (out+i)[2] = (uint)in[i+1];
+   }
+}
+static void translate_tristrip_ubyte2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+2];
+      (out+j)[1] = (uint)in[i+(i&1)];
+      (out+j)[2] = (uint)in[i+1-(i&1)];
+   }
+}
+static void translate_trifan_ubyte2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+2];
+      (out+j)[1] = (uint)in[0];
+      (out+j)[2] = (uint)in[i+1];
+   }
+}
+static void translate_quads_ubyte2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (uint)in[i+3];
+      (out+j+0)[1] = (uint)in[i+0];
+      (out+j+0)[2] = (uint)in[i+1];
+      (out+j+3)[0] = (uint)in[i+3];
+      (out+j+3)[1] = (uint)in[i+1];
+      (out+j+3)[2] = (uint)in[i+2];
+   }
+}
+static void translate_quadstrip_ubyte2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (uint)in[i+3];
+      (out+j+0)[1] = (uint)in[i+2];
+      (out+j+0)[2] = (uint)in[i+0];
+      (out+j+3)[0] = (uint)in[i+3];
+      (out+j+3)[1] = (uint)in[i+0];
+      (out+j+3)[2] = (uint)in[i+1];
+   }
+}
+static void translate_polygon_ubyte2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i+1];
+      (out+j)[2] = (uint)in[i+2];
+   }
+}
+static void translate_points_ubyte2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (uint)in[i];
+   }
+}
+static void translate_lines_ubyte2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (uint)in[i];
+      (out+i)[1] = (uint)in[i+1];
+   }
+}
+static void translate_linestrip_ubyte2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[i+1];
+   }
+}
+static void translate_lineloop_ubyte2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[i+1];
+   }
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[0];
+}
+static void translate_tris_ubyte2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (uint)in[i];
+      (out+i)[1] = (uint)in[i+1];
+      (out+i)[2] = (uint)in[i+2];
+   }
+}
+static void translate_tristrip_ubyte2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+(i&1)];
+      (out+j)[1] = (uint)in[i+1-(i&1)];
+      (out+j)[2] = (uint)in[i+2];
+   }
+}
+static void translate_trifan_ubyte2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i+1];
+      (out+j)[2] = (uint)in[i+2];
+   }
+}
+static void translate_quads_ubyte2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (uint)in[i+0];
+      (out+j+0)[1] = (uint)in[i+1];
+      (out+j+0)[2] = (uint)in[i+3];
+      (out+j+3)[0] = (uint)in[i+1];
+      (out+j+3)[1] = (uint)in[i+2];
+      (out+j+3)[2] = (uint)in[i+3];
+   }
+}
+static void translate_quadstrip_ubyte2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (uint)in[i+2];
+      (out+j+0)[1] = (uint)in[i+0];
+      (out+j+0)[2] = (uint)in[i+3];
+      (out+j+3)[0] = (uint)in[i+0];
+      (out+j+3)[1] = (uint)in[i+1];
+      (out+j+3)[2] = (uint)in[i+3];
+   }
+}
+static void translate_polygon_ubyte2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ubyte*in = (const ubyte*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i+2];
+      (out+j)[2] = (uint)in[0];
+   }
+}
+static void translate_points_ushort2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (ushort)in[i];
+   }
+}
+static void translate_lines_ushort2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (ushort)in[i];
+      (out+i)[1] = (ushort)in[i+1];
+   }
+}
+static void translate_linestrip_ushort2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[i+1];
+   }
+}
+static void translate_lineloop_ushort2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[i+1];
+   }
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[0];
+}
+static void translate_tris_ushort2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (ushort)in[i];
+      (out+i)[1] = (ushort)in[i+1];
+      (out+i)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_tristrip_ushort2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[i+1+(i&1)];
+      (out+j)[2] = (ushort)in[i+2-(i&1)];
+   }
+}
+static void translate_trifan_ushort2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i+1];
+      (out+j)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_quads_ushort2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (ushort)in[i+0];
+      (out+j+0)[1] = (ushort)in[i+1];
+      (out+j+0)[2] = (ushort)in[i+3];
+      (out+j+3)[0] = (ushort)in[i+1];
+      (out+j+3)[1] = (ushort)in[i+2];
+      (out+j+3)[2] = (ushort)in[i+3];
+   }
+}
+static void translate_quadstrip_ushort2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (ushort)in[i+2];
+      (out+j+0)[1] = (ushort)in[i+0];
+      (out+j+0)[2] = (ushort)in[i+3];
+      (out+j+3)[0] = (ushort)in[i+0];
+      (out+j+3)[1] = (ushort)in[i+1];
+      (out+j+3)[2] = (ushort)in[i+3];
+   }
+}
+static void translate_polygon_ushort2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i+1];
+      (out+j)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_points_ushort2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (ushort)in[i];
+   }
+}
+static void translate_lines_ushort2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (ushort)in[i+1];
+      (out+i)[1] = (ushort)in[i];
+   }
+}
+static void translate_linestrip_ushort2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i];
+   }
+}
+static void translate_lineloop_ushort2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i];
+   }
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i];
+}
+static void translate_tris_ushort2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (ushort)in[i+1];
+      (out+i)[1] = (ushort)in[i+2];
+      (out+i)[2] = (ushort)in[i];
+   }
+}
+static void translate_tristrip_ushort2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+1+(i&1)];
+      (out+j)[1] = (ushort)in[i+2-(i&1)];
+      (out+j)[2] = (ushort)in[i];
+   }
+}
+static void translate_trifan_ushort2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i+2];
+      (out+j)[2] = (ushort)in[0];
+   }
+}
+static void translate_quads_ushort2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (ushort)in[i+1];
+      (out+j+0)[1] = (ushort)in[i+3];
+      (out+j+0)[2] = (ushort)in[i+0];
+      (out+j+3)[0] = (ushort)in[i+2];
+      (out+j+3)[1] = (ushort)in[i+3];
+      (out+j+3)[2] = (ushort)in[i+1];
+   }
+}
+static void translate_quadstrip_ushort2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (ushort)in[i+0];
+      (out+j+0)[1] = (ushort)in[i+3];
+      (out+j+0)[2] = (ushort)in[i+2];
+      (out+j+3)[0] = (ushort)in[i+1];
+      (out+j+3)[1] = (ushort)in[i+3];
+      (out+j+3)[2] = (ushort)in[i+0];
+   }
+}
+static void translate_polygon_ushort2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i+2];
+      (out+j)[2] = (ushort)in[0];
+   }
+}
+static void translate_points_ushort2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (ushort)in[i];
+   }
+}
+static void translate_lines_ushort2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (ushort)in[i+1];
+      (out+i)[1] = (ushort)in[i];
+   }
+}
+static void translate_linestrip_ushort2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i];
+   }
+}
+static void translate_lineloop_ushort2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i];
+   }
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i];
+}
+static void translate_tris_ushort2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (ushort)in[i+2];
+      (out+i)[1] = (ushort)in[i];
+      (out+i)[2] = (ushort)in[i+1];
+   }
+}
+static void translate_tristrip_ushort2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+2];
+      (out+j)[1] = (ushort)in[i+(i&1)];
+      (out+j)[2] = (ushort)in[i+1-(i&1)];
+   }
+}
+static void translate_trifan_ushort2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+2];
+      (out+j)[1] = (ushort)in[0];
+      (out+j)[2] = (ushort)in[i+1];
+   }
+}
+static void translate_quads_ushort2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (ushort)in[i+3];
+      (out+j+0)[1] = (ushort)in[i+0];
+      (out+j+0)[2] = (ushort)in[i+1];
+      (out+j+3)[0] = (ushort)in[i+3];
+      (out+j+3)[1] = (ushort)in[i+1];
+      (out+j+3)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_quadstrip_ushort2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (ushort)in[i+3];
+      (out+j+0)[1] = (ushort)in[i+2];
+      (out+j+0)[2] = (ushort)in[i+0];
+      (out+j+3)[0] = (ushort)in[i+3];
+      (out+j+3)[1] = (ushort)in[i+0];
+      (out+j+3)[2] = (ushort)in[i+1];
+   }
+}
+static void translate_polygon_ushort2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i+1];
+      (out+j)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_points_ushort2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (ushort)in[i];
+   }
+}
+static void translate_lines_ushort2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (ushort)in[i];
+      (out+i)[1] = (ushort)in[i+1];
+   }
+}
+static void translate_linestrip_ushort2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[i+1];
+   }
+}
+static void translate_lineloop_ushort2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[i+1];
+   }
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[0];
+}
+static void translate_tris_ushort2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (ushort)in[i];
+      (out+i)[1] = (ushort)in[i+1];
+      (out+i)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_tristrip_ushort2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+(i&1)];
+      (out+j)[1] = (ushort)in[i+1-(i&1)];
+      (out+j)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_trifan_ushort2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i+1];
+      (out+j)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_quads_ushort2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (ushort)in[i+0];
+      (out+j+0)[1] = (ushort)in[i+1];
+      (out+j+0)[2] = (ushort)in[i+3];
+      (out+j+3)[0] = (ushort)in[i+1];
+      (out+j+3)[1] = (ushort)in[i+2];
+      (out+j+3)[2] = (ushort)in[i+3];
+   }
+}
+static void translate_quadstrip_ushort2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (ushort)in[i+2];
+      (out+j+0)[1] = (ushort)in[i+0];
+      (out+j+0)[2] = (ushort)in[i+3];
+      (out+j+3)[0] = (ushort)in[i+0];
+      (out+j+3)[1] = (ushort)in[i+1];
+      (out+j+3)[2] = (ushort)in[i+3];
+   }
+}
+static void translate_polygon_ushort2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i+2];
+      (out+j)[2] = (ushort)in[0];
+   }
+}
+static void translate_points_ushort2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (uint)in[i];
+   }
+}
+static void translate_lines_ushort2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (uint)in[i];
+      (out+i)[1] = (uint)in[i+1];
+   }
+}
+static void translate_linestrip_ushort2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[i+1];
+   }
+}
+static void translate_lineloop_ushort2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[i+1];
+   }
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[0];
+}
+static void translate_tris_ushort2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (uint)in[i];
+      (out+i)[1] = (uint)in[i+1];
+      (out+i)[2] = (uint)in[i+2];
+   }
+}
+static void translate_tristrip_ushort2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[i+1+(i&1)];
+      (out+j)[2] = (uint)in[i+2-(i&1)];
+   }
+}
+static void translate_trifan_ushort2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i+1];
+      (out+j)[2] = (uint)in[i+2];
+   }
+}
+static void translate_quads_ushort2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (uint)in[i+0];
+      (out+j+0)[1] = (uint)in[i+1];
+      (out+j+0)[2] = (uint)in[i+3];
+      (out+j+3)[0] = (uint)in[i+1];
+      (out+j+3)[1] = (uint)in[i+2];
+      (out+j+3)[2] = (uint)in[i+3];
+   }
+}
+static void translate_quadstrip_ushort2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (uint)in[i+2];
+      (out+j+0)[1] = (uint)in[i+0];
+      (out+j+0)[2] = (uint)in[i+3];
+      (out+j+3)[0] = (uint)in[i+0];
+      (out+j+3)[1] = (uint)in[i+1];
+      (out+j+3)[2] = (uint)in[i+3];
+   }
+}
+static void translate_polygon_ushort2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i+1];
+      (out+j)[2] = (uint)in[i+2];
+   }
+}
+static void translate_points_ushort2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (uint)in[i];
+   }
+}
+static void translate_lines_ushort2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (uint)in[i+1];
+      (out+i)[1] = (uint)in[i];
+   }
+}
+static void translate_linestrip_ushort2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i];
+   }
+}
+static void translate_lineloop_ushort2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i];
+   }
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i];
+}
+static void translate_tris_ushort2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (uint)in[i+1];
+      (out+i)[1] = (uint)in[i+2];
+      (out+i)[2] = (uint)in[i];
+   }
+}
+static void translate_tristrip_ushort2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+1+(i&1)];
+      (out+j)[1] = (uint)in[i+2-(i&1)];
+      (out+j)[2] = (uint)in[i];
+   }
+}
+static void translate_trifan_ushort2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i+2];
+      (out+j)[2] = (uint)in[0];
+   }
+}
+static void translate_quads_ushort2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (uint)in[i+1];
+      (out+j+0)[1] = (uint)in[i+3];
+      (out+j+0)[2] = (uint)in[i+0];
+      (out+j+3)[0] = (uint)in[i+2];
+      (out+j+3)[1] = (uint)in[i+3];
+      (out+j+3)[2] = (uint)in[i+1];
+   }
+}
+static void translate_quadstrip_ushort2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (uint)in[i+0];
+      (out+j+0)[1] = (uint)in[i+3];
+      (out+j+0)[2] = (uint)in[i+2];
+      (out+j+3)[0] = (uint)in[i+1];
+      (out+j+3)[1] = (uint)in[i+3];
+      (out+j+3)[2] = (uint)in[i+0];
+   }
+}
+static void translate_polygon_ushort2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i+2];
+      (out+j)[2] = (uint)in[0];
+   }
+}
+static void translate_points_ushort2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (uint)in[i];
+   }
+}
+static void translate_lines_ushort2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (uint)in[i+1];
+      (out+i)[1] = (uint)in[i];
+   }
+}
+static void translate_linestrip_ushort2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i];
+   }
+}
+static void translate_lineloop_ushort2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i];
+   }
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i];
+}
+static void translate_tris_ushort2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (uint)in[i+2];
+      (out+i)[1] = (uint)in[i];
+      (out+i)[2] = (uint)in[i+1];
+   }
+}
+static void translate_tristrip_ushort2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+2];
+      (out+j)[1] = (uint)in[i+(i&1)];
+      (out+j)[2] = (uint)in[i+1-(i&1)];
+   }
+}
+static void translate_trifan_ushort2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+2];
+      (out+j)[1] = (uint)in[0];
+      (out+j)[2] = (uint)in[i+1];
+   }
+}
+static void translate_quads_ushort2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (uint)in[i+3];
+      (out+j+0)[1] = (uint)in[i+0];
+      (out+j+0)[2] = (uint)in[i+1];
+      (out+j+3)[0] = (uint)in[i+3];
+      (out+j+3)[1] = (uint)in[i+1];
+      (out+j+3)[2] = (uint)in[i+2];
+   }
+}
+static void translate_quadstrip_ushort2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (uint)in[i+3];
+      (out+j+0)[1] = (uint)in[i+2];
+      (out+j+0)[2] = (uint)in[i+0];
+      (out+j+3)[0] = (uint)in[i+3];
+      (out+j+3)[1] = (uint)in[i+0];
+      (out+j+3)[2] = (uint)in[i+1];
+   }
+}
+static void translate_polygon_ushort2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i+1];
+      (out+j)[2] = (uint)in[i+2];
+   }
+}
+static void translate_points_ushort2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (uint)in[i];
+   }
+}
+static void translate_lines_ushort2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (uint)in[i];
+      (out+i)[1] = (uint)in[i+1];
+   }
+}
+static void translate_linestrip_ushort2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[i+1];
+   }
+}
+static void translate_lineloop_ushort2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[i+1];
+   }
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[0];
+}
+static void translate_tris_ushort2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (uint)in[i];
+      (out+i)[1] = (uint)in[i+1];
+      (out+i)[2] = (uint)in[i+2];
+   }
+}
+static void translate_tristrip_ushort2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+(i&1)];
+      (out+j)[1] = (uint)in[i+1-(i&1)];
+      (out+j)[2] = (uint)in[i+2];
+   }
+}
+static void translate_trifan_ushort2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i+1];
+      (out+j)[2] = (uint)in[i+2];
+   }
+}
+static void translate_quads_ushort2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (uint)in[i+0];
+      (out+j+0)[1] = (uint)in[i+1];
+      (out+j+0)[2] = (uint)in[i+3];
+      (out+j+3)[0] = (uint)in[i+1];
+      (out+j+3)[1] = (uint)in[i+2];
+      (out+j+3)[2] = (uint)in[i+3];
+   }
+}
+static void translate_quadstrip_ushort2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (uint)in[i+2];
+      (out+j+0)[1] = (uint)in[i+0];
+      (out+j+0)[2] = (uint)in[i+3];
+      (out+j+3)[0] = (uint)in[i+0];
+      (out+j+3)[1] = (uint)in[i+1];
+      (out+j+3)[2] = (uint)in[i+3];
+   }
+}
+static void translate_polygon_ushort2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const ushort*in = (const ushort*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i+2];
+      (out+j)[2] = (uint)in[0];
+   }
+}
+static void translate_points_uint2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (ushort)in[i];
+   }
+}
+static void translate_lines_uint2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (ushort)in[i];
+      (out+i)[1] = (ushort)in[i+1];
+   }
+}
+static void translate_linestrip_uint2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[i+1];
+   }
+}
+static void translate_lineloop_uint2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[i+1];
+   }
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[0];
+}
+static void translate_tris_uint2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (ushort)in[i];
+      (out+i)[1] = (ushort)in[i+1];
+      (out+i)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_tristrip_uint2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[i+1+(i&1)];
+      (out+j)[2] = (ushort)in[i+2-(i&1)];
+   }
+}
+static void translate_trifan_uint2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i+1];
+      (out+j)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_quads_uint2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (ushort)in[i+0];
+      (out+j+0)[1] = (ushort)in[i+1];
+      (out+j+0)[2] = (ushort)in[i+3];
+      (out+j+3)[0] = (ushort)in[i+1];
+      (out+j+3)[1] = (ushort)in[i+2];
+      (out+j+3)[2] = (ushort)in[i+3];
+   }
+}
+static void translate_quadstrip_uint2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (ushort)in[i+2];
+      (out+j+0)[1] = (ushort)in[i+0];
+      (out+j+0)[2] = (ushort)in[i+3];
+      (out+j+3)[0] = (ushort)in[i+0];
+      (out+j+3)[1] = (ushort)in[i+1];
+      (out+j+3)[2] = (ushort)in[i+3];
+   }
+}
+static void translate_polygon_uint2ushort_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i+1];
+      (out+j)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_points_uint2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (ushort)in[i];
+   }
+}
+static void translate_lines_uint2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (ushort)in[i+1];
+      (out+i)[1] = (ushort)in[i];
+   }
+}
+static void translate_linestrip_uint2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i];
+   }
+}
+static void translate_lineloop_uint2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i];
+   }
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i];
+}
+static void translate_tris_uint2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (ushort)in[i+1];
+      (out+i)[1] = (ushort)in[i+2];
+      (out+i)[2] = (ushort)in[i];
+   }
+}
+static void translate_tristrip_uint2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+1+(i&1)];
+      (out+j)[1] = (ushort)in[i+2-(i&1)];
+      (out+j)[2] = (ushort)in[i];
+   }
+}
+static void translate_trifan_uint2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i+2];
+      (out+j)[2] = (ushort)in[0];
+   }
+}
+static void translate_quads_uint2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (ushort)in[i+1];
+      (out+j+0)[1] = (ushort)in[i+3];
+      (out+j+0)[2] = (ushort)in[i+0];
+      (out+j+3)[0] = (ushort)in[i+2];
+      (out+j+3)[1] = (ushort)in[i+3];
+      (out+j+3)[2] = (ushort)in[i+1];
+   }
+}
+static void translate_quadstrip_uint2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (ushort)in[i+0];
+      (out+j+0)[1] = (ushort)in[i+3];
+      (out+j+0)[2] = (ushort)in[i+2];
+      (out+j+3)[0] = (ushort)in[i+1];
+      (out+j+3)[1] = (ushort)in[i+3];
+      (out+j+3)[2] = (ushort)in[i+0];
+   }
+}
+static void translate_polygon_uint2ushort_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i+2];
+      (out+j)[2] = (ushort)in[0];
+   }
+}
+static void translate_points_uint2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (ushort)in[i];
+   }
+}
+static void translate_lines_uint2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (ushort)in[i+1];
+      (out+i)[1] = (ushort)in[i];
+   }
+}
+static void translate_linestrip_uint2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i];
+   }
+}
+static void translate_lineloop_uint2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i];
+   }
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i];
+}
+static void translate_tris_uint2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (ushort)in[i+2];
+      (out+i)[1] = (ushort)in[i];
+      (out+i)[2] = (ushort)in[i+1];
+   }
+}
+static void translate_tristrip_uint2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+2];
+      (out+j)[1] = (ushort)in[i+(i&1)];
+      (out+j)[2] = (ushort)in[i+1-(i&1)];
+   }
+}
+static void translate_trifan_uint2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+2];
+      (out+j)[1] = (ushort)in[0];
+      (out+j)[2] = (ushort)in[i+1];
+   }
+}
+static void translate_quads_uint2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (ushort)in[i+3];
+      (out+j+0)[1] = (ushort)in[i+0];
+      (out+j+0)[2] = (ushort)in[i+1];
+      (out+j+3)[0] = (ushort)in[i+3];
+      (out+j+3)[1] = (ushort)in[i+1];
+      (out+j+3)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_quadstrip_uint2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (ushort)in[i+3];
+      (out+j+0)[1] = (ushort)in[i+2];
+      (out+j+0)[2] = (ushort)in[i+0];
+      (out+j+3)[0] = (ushort)in[i+3];
+      (out+j+3)[1] = (ushort)in[i+0];
+      (out+j+3)[2] = (ushort)in[i+1];
+   }
+}
+static void translate_polygon_uint2ushort_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i+1];
+      (out+j)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_points_uint2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (ushort)in[i];
+   }
+}
+static void translate_lines_uint2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (ushort)in[i];
+      (out+i)[1] = (ushort)in[i+1];
+   }
+}
+static void translate_linestrip_uint2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[i+1];
+   }
+}
+static void translate_lineloop_uint2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[i+1];
+   }
+      (out+j)[0] = (ushort)in[i];
+      (out+j)[1] = (ushort)in[0];
+}
+static void translate_tris_uint2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (ushort)in[i];
+      (out+i)[1] = (ushort)in[i+1];
+      (out+i)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_tristrip_uint2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+(i&1)];
+      (out+j)[1] = (ushort)in[i+1-(i&1)];
+      (out+j)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_trifan_uint2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[0];
+      (out+j)[1] = (ushort)in[i+1];
+      (out+j)[2] = (ushort)in[i+2];
+   }
+}
+static void translate_quads_uint2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (ushort)in[i+0];
+      (out+j+0)[1] = (ushort)in[i+1];
+      (out+j+0)[2] = (ushort)in[i+3];
+      (out+j+3)[0] = (ushort)in[i+1];
+      (out+j+3)[1] = (ushort)in[i+2];
+      (out+j+3)[2] = (ushort)in[i+3];
+   }
+}
+static void translate_quadstrip_uint2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (ushort)in[i+2];
+      (out+j+0)[1] = (ushort)in[i+0];
+      (out+j+0)[2] = (ushort)in[i+3];
+      (out+j+3)[0] = (ushort)in[i+0];
+      (out+j+3)[1] = (ushort)in[i+1];
+      (out+j+3)[2] = (ushort)in[i+3];
+   }
+}
+static void translate_polygon_uint2ushort_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  ushort *out = (ushort*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (ushort)in[i+1];
+      (out+j)[1] = (ushort)in[i+2];
+      (out+j)[2] = (ushort)in[0];
+   }
+}
+static void translate_points_uint2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (uint)in[i];
+   }
+}
+static void translate_lines_uint2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (uint)in[i];
+      (out+i)[1] = (uint)in[i+1];
+   }
+}
+static void translate_linestrip_uint2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[i+1];
+   }
+}
+static void translate_lineloop_uint2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[i+1];
+   }
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[0];
+}
+static void translate_tris_uint2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (uint)in[i];
+      (out+i)[1] = (uint)in[i+1];
+      (out+i)[2] = (uint)in[i+2];
+   }
+}
+static void translate_tristrip_uint2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[i+1+(i&1)];
+      (out+j)[2] = (uint)in[i+2-(i&1)];
+   }
+}
+static void translate_trifan_uint2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i+1];
+      (out+j)[2] = (uint)in[i+2];
+   }
+}
+static void translate_quads_uint2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (uint)in[i+0];
+      (out+j+0)[1] = (uint)in[i+1];
+      (out+j+0)[2] = (uint)in[i+3];
+      (out+j+3)[0] = (uint)in[i+1];
+      (out+j+3)[1] = (uint)in[i+2];
+      (out+j+3)[2] = (uint)in[i+3];
+   }
+}
+static void translate_quadstrip_uint2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (uint)in[i+2];
+      (out+j+0)[1] = (uint)in[i+0];
+      (out+j+0)[2] = (uint)in[i+3];
+      (out+j+3)[0] = (uint)in[i+0];
+      (out+j+3)[1] = (uint)in[i+1];
+      (out+j+3)[2] = (uint)in[i+3];
+   }
+}
+static void translate_polygon_uint2uint_first2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i+1];
+      (out+j)[2] = (uint)in[i+2];
+   }
+}
+static void translate_points_uint2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (uint)in[i];
+   }
+}
+static void translate_lines_uint2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (uint)in[i+1];
+      (out+i)[1] = (uint)in[i];
+   }
+}
+static void translate_linestrip_uint2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i];
+   }
+}
+static void translate_lineloop_uint2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i];
+   }
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i];
+}
+static void translate_tris_uint2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (uint)in[i+1];
+      (out+i)[1] = (uint)in[i+2];
+      (out+i)[2] = (uint)in[i];
+   }
+}
+static void translate_tristrip_uint2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+1+(i&1)];
+      (out+j)[1] = (uint)in[i+2-(i&1)];
+      (out+j)[2] = (uint)in[i];
+   }
+}
+static void translate_trifan_uint2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i+2];
+      (out+j)[2] = (uint)in[0];
+   }
+}
+static void translate_quads_uint2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (uint)in[i+1];
+      (out+j+0)[1] = (uint)in[i+3];
+      (out+j+0)[2] = (uint)in[i+0];
+      (out+j+3)[0] = (uint)in[i+2];
+      (out+j+3)[1] = (uint)in[i+3];
+      (out+j+3)[2] = (uint)in[i+1];
+   }
+}
+static void translate_quadstrip_uint2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (uint)in[i+0];
+      (out+j+0)[1] = (uint)in[i+3];
+      (out+j+0)[2] = (uint)in[i+2];
+      (out+j+3)[0] = (uint)in[i+1];
+      (out+j+3)[1] = (uint)in[i+3];
+      (out+j+3)[2] = (uint)in[i+0];
+   }
+}
+static void translate_polygon_uint2uint_first2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i+2];
+      (out+j)[2] = (uint)in[0];
+   }
+}
+static void translate_points_uint2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (uint)in[i];
+   }
+}
+static void translate_lines_uint2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (uint)in[i+1];
+      (out+i)[1] = (uint)in[i];
+   }
+}
+static void translate_linestrip_uint2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i];
+   }
+}
+static void translate_lineloop_uint2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i];
+   }
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i];
+}
+static void translate_tris_uint2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (uint)in[i+2];
+      (out+i)[1] = (uint)in[i];
+      (out+i)[2] = (uint)in[i+1];
+   }
+}
+static void translate_tristrip_uint2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+2];
+      (out+j)[1] = (uint)in[i+(i&1)];
+      (out+j)[2] = (uint)in[i+1-(i&1)];
+   }
+}
+static void translate_trifan_uint2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+2];
+      (out+j)[1] = (uint)in[0];
+      (out+j)[2] = (uint)in[i+1];
+   }
+}
+static void translate_quads_uint2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (uint)in[i+3];
+      (out+j+0)[1] = (uint)in[i+0];
+      (out+j+0)[2] = (uint)in[i+1];
+      (out+j+3)[0] = (uint)in[i+3];
+      (out+j+3)[1] = (uint)in[i+1];
+      (out+j+3)[2] = (uint)in[i+2];
+   }
+}
+static void translate_quadstrip_uint2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (uint)in[i+3];
+      (out+j+0)[1] = (uint)in[i+2];
+      (out+j+0)[2] = (uint)in[i+0];
+      (out+j+3)[0] = (uint)in[i+3];
+      (out+j+3)[1] = (uint)in[i+0];
+      (out+j+3)[2] = (uint)in[i+1];
+   }
+}
+static void translate_polygon_uint2uint_last2first(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i+1];
+      (out+j)[2] = (uint)in[i+2];
+   }
+}
+static void translate_points_uint2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i++) { 
+      (out+i)[0] = (uint)in[i];
+   }
+}
+static void translate_lines_uint2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=2) { 
+      (out+i)[0] = (uint)in[i];
+      (out+i)[1] = (uint)in[i+1];
+   }
+}
+static void translate_linestrip_uint2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[i+1];
+   }
+}
+static void translate_lineloop_uint2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr - 2; j+=2, i++) { 
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[i+1];
+   }
+      (out+j)[0] = (uint)in[i];
+      (out+j)[1] = (uint)in[0];
+}
+static void translate_tris_uint2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (i = 0; i < nr; i+=3) { 
+      (out+i)[0] = (uint)in[i];
+      (out+i)[1] = (uint)in[i+1];
+      (out+i)[2] = (uint)in[i+2];
+   }
+}
+static void translate_tristrip_uint2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+(i&1)];
+      (out+j)[1] = (uint)in[i+1-(i&1)];
+      (out+j)[2] = (uint)in[i+2];
+   }
+}
+static void translate_trifan_uint2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[0];
+      (out+j)[1] = (uint)in[i+1];
+      (out+j)[2] = (uint)in[i+2];
+   }
+}
+static void translate_quads_uint2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=4) { 
+      (out+j+0)[0] = (uint)in[i+0];
+      (out+j+0)[1] = (uint)in[i+1];
+      (out+j+0)[2] = (uint)in[i+3];
+      (out+j+3)[0] = (uint)in[i+1];
+      (out+j+3)[1] = (uint)in[i+2];
+      (out+j+3)[2] = (uint)in[i+3];
+   }
+}
+static void translate_quadstrip_uint2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=6, i+=2) { 
+      (out+j+0)[0] = (uint)in[i+2];
+      (out+j+0)[1] = (uint)in[i+0];
+      (out+j+0)[2] = (uint)in[i+3];
+      (out+j+3)[0] = (uint)in[i+0];
+      (out+j+3)[1] = (uint)in[i+1];
+      (out+j+3)[2] = (uint)in[i+3];
+   }
+}
+static void translate_polygon_uint2uint_last2last(
+    const void * _in,
+    unsigned nr,
+    void *_out )
+{
+  const uint*in = (const uint*)_in;
+  uint *out = (uint*)_out;
+  unsigned i, j;
+  (void)j;
+  for (j = i = 0; j < nr; j+=3, i++) { 
+      (out+j)[0] = (uint)in[i+1];
+      (out+j)[1] = (uint)in[i+2];
+      (out+j)[2] = (uint)in[0];
+   }
+}
+void u_index_init( void )
+{
+  static int firsttime = 1;
+  if (!firsttime) return;
+  firsttime = 0;
+generate[OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_POINTS] = generate_points_ushort_first2first;
+generate[OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINES] = generate_lines_ushort_first2first;
+generate[OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINE_STRIP] = generate_linestrip_ushort_first2first;
+generate[OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINE_LOOP] = generate_lineloop_ushort_first2first;
+generate[OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLES] = generate_tris_ushort_first2first;
+generate[OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLE_FAN] = generate_trifan_ushort_first2first;
+generate[OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLE_STRIP] = generate_tristrip_ushort_first2first;
+generate[OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_QUADS] = generate_quads_ushort_first2first;
+generate[OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_QUAD_STRIP] = generate_quadstrip_ushort_first2first;
+generate[OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_POLYGON] = generate_polygon_ushort_first2first;
+generate[OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_POINTS] = generate_points_ushort_first2last;
+generate[OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_LINES] = generate_lines_ushort_first2last;
+generate[OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_LINE_STRIP] = generate_linestrip_ushort_first2last;
+generate[OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_LINE_LOOP] = generate_lineloop_ushort_first2last;
+generate[OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLES] = generate_tris_ushort_first2last;
+generate[OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLE_FAN] = generate_trifan_ushort_first2last;
+generate[OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLE_STRIP] = generate_tristrip_ushort_first2last;
+generate[OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_QUADS] = generate_quads_ushort_first2last;
+generate[OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_QUAD_STRIP] = generate_quadstrip_ushort_first2last;
+generate[OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_POLYGON] = generate_polygon_ushort_first2last;
+generate[OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_POINTS] = generate_points_ushort_last2first;
+generate[OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_LINES] = generate_lines_ushort_last2first;
+generate[OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_LINE_STRIP] = generate_linestrip_ushort_last2first;
+generate[OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_LINE_LOOP] = generate_lineloop_ushort_last2first;
+generate[OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLES] = generate_tris_ushort_last2first;
+generate[OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLE_FAN] = generate_trifan_ushort_last2first;
+generate[OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLE_STRIP] = generate_tristrip_ushort_last2first;
+generate[OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_QUADS] = generate_quads_ushort_last2first;
+generate[OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_QUAD_STRIP] = generate_quadstrip_ushort_last2first;
+generate[OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_POLYGON] = generate_polygon_ushort_last2first;
+generate[OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_POINTS] = generate_points_ushort_last2last;
+generate[OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_LINES] = generate_lines_ushort_last2last;
+generate[OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_LINE_STRIP] = generate_linestrip_ushort_last2last;
+generate[OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_LINE_LOOP] = generate_lineloop_ushort_last2last;
+generate[OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLES] = generate_tris_ushort_last2last;
+generate[OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLE_FAN] = generate_trifan_ushort_last2last;
+generate[OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLE_STRIP] = generate_tristrip_ushort_last2last;
+generate[OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_QUADS] = generate_quads_ushort_last2last;
+generate[OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_QUAD_STRIP] = generate_quadstrip_ushort_last2last;
+generate[OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_POLYGON] = generate_polygon_ushort_last2last;
+generate[OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_POINTS] = generate_points_uint_first2first;
+generate[OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINES] = generate_lines_uint_first2first;
+generate[OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINE_STRIP] = generate_linestrip_uint_first2first;
+generate[OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINE_LOOP] = generate_lineloop_uint_first2first;
+generate[OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLES] = generate_tris_uint_first2first;
+generate[OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLE_FAN] = generate_trifan_uint_first2first;
+generate[OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLE_STRIP] = generate_tristrip_uint_first2first;
+generate[OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_QUADS] = generate_quads_uint_first2first;
+generate[OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_QUAD_STRIP] = generate_quadstrip_uint_first2first;
+generate[OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_POLYGON] = generate_polygon_uint_first2first;
+generate[OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_POINTS] = generate_points_uint_first2last;
+generate[OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_LINES] = generate_lines_uint_first2last;
+generate[OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_LINE_STRIP] = generate_linestrip_uint_first2last;
+generate[OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_LINE_LOOP] = generate_lineloop_uint_first2last;
+generate[OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLES] = generate_tris_uint_first2last;
+generate[OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLE_FAN] = generate_trifan_uint_first2last;
+generate[OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLE_STRIP] = generate_tristrip_uint_first2last;
+generate[OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_QUADS] = generate_quads_uint_first2last;
+generate[OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_QUAD_STRIP] = generate_quadstrip_uint_first2last;
+generate[OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_POLYGON] = generate_polygon_uint_first2last;
+generate[OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_POINTS] = generate_points_uint_last2first;
+generate[OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_LINES] = generate_lines_uint_last2first;
+generate[OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_LINE_STRIP] = generate_linestrip_uint_last2first;
+generate[OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_LINE_LOOP] = generate_lineloop_uint_last2first;
+generate[OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLES] = generate_tris_uint_last2first;
+generate[OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLE_FAN] = generate_trifan_uint_last2first;
+generate[OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLE_STRIP] = generate_tristrip_uint_last2first;
+generate[OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_QUADS] = generate_quads_uint_last2first;
+generate[OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_QUAD_STRIP] = generate_quadstrip_uint_last2first;
+generate[OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_POLYGON] = generate_polygon_uint_last2first;
+generate[OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_POINTS] = generate_points_uint_last2last;
+generate[OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_LINES] = generate_lines_uint_last2last;
+generate[OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_LINE_STRIP] = generate_linestrip_uint_last2last;
+generate[OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_LINE_LOOP] = generate_lineloop_uint_last2last;
+generate[OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLES] = generate_tris_uint_last2last;
+generate[OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLE_FAN] = generate_trifan_uint_last2last;
+generate[OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLE_STRIP] = generate_tristrip_uint_last2last;
+generate[OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_QUADS] = generate_quads_uint_last2last;
+generate[OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_QUAD_STRIP] = generate_quadstrip_uint_last2last;
+generate[OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_POLYGON] = generate_polygon_uint_last2last;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_POINTS] = translate_points_ubyte2ushort_first2first;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINES] = translate_lines_ubyte2ushort_first2first;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_ubyte2ushort_first2first;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_ubyte2ushort_first2first;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLES] = translate_tris_ubyte2ushort_first2first;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_ubyte2ushort_first2first;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_ubyte2ushort_first2first;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_QUADS] = translate_quads_ubyte2ushort_first2first;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_ubyte2ushort_first2first;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_POLYGON] = translate_polygon_ubyte2ushort_first2first;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_POINTS] = translate_points_ubyte2ushort_first2last;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_LINES] = translate_lines_ubyte2ushort_first2last;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_ubyte2ushort_first2last;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_ubyte2ushort_first2last;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLES] = translate_tris_ubyte2ushort_first2last;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_ubyte2ushort_first2last;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_ubyte2ushort_first2last;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_QUADS] = translate_quads_ubyte2ushort_first2last;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_ubyte2ushort_first2last;
+translate[IN_UBYTE][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_POLYGON] = translate_polygon_ubyte2ushort_first2last;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_POINTS] = translate_points_ubyte2ushort_last2first;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_LINES] = translate_lines_ubyte2ushort_last2first;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_ubyte2ushort_last2first;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_ubyte2ushort_last2first;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLES] = translate_tris_ubyte2ushort_last2first;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_ubyte2ushort_last2first;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_ubyte2ushort_last2first;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_QUADS] = translate_quads_ubyte2ushort_last2first;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_ubyte2ushort_last2first;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_POLYGON] = translate_polygon_ubyte2ushort_last2first;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_POINTS] = translate_points_ubyte2ushort_last2last;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_LINES] = translate_lines_ubyte2ushort_last2last;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_ubyte2ushort_last2last;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_ubyte2ushort_last2last;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLES] = translate_tris_ubyte2ushort_last2last;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_ubyte2ushort_last2last;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_ubyte2ushort_last2last;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_QUADS] = translate_quads_ubyte2ushort_last2last;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_ubyte2ushort_last2last;
+translate[IN_UBYTE][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_POLYGON] = translate_polygon_ubyte2ushort_last2last;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_POINTS] = translate_points_ubyte2uint_first2first;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINES] = translate_lines_ubyte2uint_first2first;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_ubyte2uint_first2first;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_ubyte2uint_first2first;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLES] = translate_tris_ubyte2uint_first2first;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_ubyte2uint_first2first;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_ubyte2uint_first2first;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_QUADS] = translate_quads_ubyte2uint_first2first;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_ubyte2uint_first2first;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_POLYGON] = translate_polygon_ubyte2uint_first2first;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_POINTS] = translate_points_ubyte2uint_first2last;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_LINES] = translate_lines_ubyte2uint_first2last;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_ubyte2uint_first2last;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_ubyte2uint_first2last;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLES] = translate_tris_ubyte2uint_first2last;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_ubyte2uint_first2last;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_ubyte2uint_first2last;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_QUADS] = translate_quads_ubyte2uint_first2last;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_ubyte2uint_first2last;
+translate[IN_UBYTE][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_POLYGON] = translate_polygon_ubyte2uint_first2last;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_POINTS] = translate_points_ubyte2uint_last2first;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_LINES] = translate_lines_ubyte2uint_last2first;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_ubyte2uint_last2first;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_ubyte2uint_last2first;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLES] = translate_tris_ubyte2uint_last2first;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_ubyte2uint_last2first;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_ubyte2uint_last2first;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_QUADS] = translate_quads_ubyte2uint_last2first;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_ubyte2uint_last2first;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_POLYGON] = translate_polygon_ubyte2uint_last2first;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_POINTS] = translate_points_ubyte2uint_last2last;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_LINES] = translate_lines_ubyte2uint_last2last;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_ubyte2uint_last2last;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_ubyte2uint_last2last;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLES] = translate_tris_ubyte2uint_last2last;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_ubyte2uint_last2last;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_ubyte2uint_last2last;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_QUADS] = translate_quads_ubyte2uint_last2last;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_ubyte2uint_last2last;
+translate[IN_UBYTE][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_POLYGON] = translate_polygon_ubyte2uint_last2last;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_POINTS] = translate_points_ushort2ushort_first2first;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINES] = translate_lines_ushort2ushort_first2first;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_ushort2ushort_first2first;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_ushort2ushort_first2first;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLES] = translate_tris_ushort2ushort_first2first;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_ushort2ushort_first2first;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_ushort2ushort_first2first;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_QUADS] = translate_quads_ushort2ushort_first2first;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_ushort2ushort_first2first;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_POLYGON] = translate_polygon_ushort2ushort_first2first;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_POINTS] = translate_points_ushort2ushort_first2last;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_LINES] = translate_lines_ushort2ushort_first2last;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_ushort2ushort_first2last;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_ushort2ushort_first2last;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLES] = translate_tris_ushort2ushort_first2last;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_ushort2ushort_first2last;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_ushort2ushort_first2last;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_QUADS] = translate_quads_ushort2ushort_first2last;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_ushort2ushort_first2last;
+translate[IN_USHORT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_POLYGON] = translate_polygon_ushort2ushort_first2last;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_POINTS] = translate_points_ushort2ushort_last2first;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_LINES] = translate_lines_ushort2ushort_last2first;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_ushort2ushort_last2first;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_ushort2ushort_last2first;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLES] = translate_tris_ushort2ushort_last2first;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_ushort2ushort_last2first;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_ushort2ushort_last2first;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_QUADS] = translate_quads_ushort2ushort_last2first;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_ushort2ushort_last2first;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_POLYGON] = translate_polygon_ushort2ushort_last2first;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_POINTS] = translate_points_ushort2ushort_last2last;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_LINES] = translate_lines_ushort2ushort_last2last;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_ushort2ushort_last2last;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_ushort2ushort_last2last;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLES] = translate_tris_ushort2ushort_last2last;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_ushort2ushort_last2last;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_ushort2ushort_last2last;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_QUADS] = translate_quads_ushort2ushort_last2last;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_ushort2ushort_last2last;
+translate[IN_USHORT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_POLYGON] = translate_polygon_ushort2ushort_last2last;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_POINTS] = translate_points_ushort2uint_first2first;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINES] = translate_lines_ushort2uint_first2first;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_ushort2uint_first2first;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_ushort2uint_first2first;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLES] = translate_tris_ushort2uint_first2first;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_ushort2uint_first2first;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_ushort2uint_first2first;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_QUADS] = translate_quads_ushort2uint_first2first;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_ushort2uint_first2first;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_POLYGON] = translate_polygon_ushort2uint_first2first;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_POINTS] = translate_points_ushort2uint_first2last;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_LINES] = translate_lines_ushort2uint_first2last;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_ushort2uint_first2last;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_ushort2uint_first2last;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLES] = translate_tris_ushort2uint_first2last;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_ushort2uint_first2last;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_ushort2uint_first2last;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_QUADS] = translate_quads_ushort2uint_first2last;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_ushort2uint_first2last;
+translate[IN_USHORT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_POLYGON] = translate_polygon_ushort2uint_first2last;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_POINTS] = translate_points_ushort2uint_last2first;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_LINES] = translate_lines_ushort2uint_last2first;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_ushort2uint_last2first;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_ushort2uint_last2first;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLES] = translate_tris_ushort2uint_last2first;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_ushort2uint_last2first;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_ushort2uint_last2first;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_QUADS] = translate_quads_ushort2uint_last2first;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_ushort2uint_last2first;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_POLYGON] = translate_polygon_ushort2uint_last2first;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_POINTS] = translate_points_ushort2uint_last2last;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_LINES] = translate_lines_ushort2uint_last2last;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_ushort2uint_last2last;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_ushort2uint_last2last;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLES] = translate_tris_ushort2uint_last2last;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_ushort2uint_last2last;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_ushort2uint_last2last;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_QUADS] = translate_quads_ushort2uint_last2last;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_ushort2uint_last2last;
+translate[IN_USHORT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_POLYGON] = translate_polygon_ushort2uint_last2last;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_POINTS] = translate_points_uint2ushort_first2first;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINES] = translate_lines_uint2ushort_first2first;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_uint2ushort_first2first;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_uint2ushort_first2first;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLES] = translate_tris_uint2ushort_first2first;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_uint2ushort_first2first;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_uint2ushort_first2first;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_QUADS] = translate_quads_uint2ushort_first2first;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_uint2ushort_first2first;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_FIRST][PIPE_PRIM_POLYGON] = translate_polygon_uint2ushort_first2first;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_POINTS] = translate_points_uint2ushort_first2last;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_LINES] = translate_lines_uint2ushort_first2last;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_uint2ushort_first2last;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_uint2ushort_first2last;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLES] = translate_tris_uint2ushort_first2last;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_uint2ushort_first2last;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_uint2ushort_first2last;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_QUADS] = translate_quads_uint2ushort_first2last;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_uint2ushort_first2last;
+translate[IN_UINT][OUT_USHORT][PV_FIRST][PV_LAST][PIPE_PRIM_POLYGON] = translate_polygon_uint2ushort_first2last;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_POINTS] = translate_points_uint2ushort_last2first;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_LINES] = translate_lines_uint2ushort_last2first;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_uint2ushort_last2first;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_uint2ushort_last2first;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLES] = translate_tris_uint2ushort_last2first;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_uint2ushort_last2first;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_uint2ushort_last2first;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_QUADS] = translate_quads_uint2ushort_last2first;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_uint2ushort_last2first;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_FIRST][PIPE_PRIM_POLYGON] = translate_polygon_uint2ushort_last2first;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_POINTS] = translate_points_uint2ushort_last2last;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_LINES] = translate_lines_uint2ushort_last2last;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_uint2ushort_last2last;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_uint2ushort_last2last;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLES] = translate_tris_uint2ushort_last2last;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_uint2ushort_last2last;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_uint2ushort_last2last;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_QUADS] = translate_quads_uint2ushort_last2last;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_uint2ushort_last2last;
+translate[IN_UINT][OUT_USHORT][PV_LAST][PV_LAST][PIPE_PRIM_POLYGON] = translate_polygon_uint2ushort_last2last;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_POINTS] = translate_points_uint2uint_first2first;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINES] = translate_lines_uint2uint_first2first;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_uint2uint_first2first;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_uint2uint_first2first;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLES] = translate_tris_uint2uint_first2first;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_uint2uint_first2first;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_uint2uint_first2first;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_QUADS] = translate_quads_uint2uint_first2first;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_uint2uint_first2first;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_FIRST][PIPE_PRIM_POLYGON] = translate_polygon_uint2uint_first2first;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_POINTS] = translate_points_uint2uint_first2last;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_LINES] = translate_lines_uint2uint_first2last;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_uint2uint_first2last;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_uint2uint_first2last;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLES] = translate_tris_uint2uint_first2last;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_uint2uint_first2last;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_uint2uint_first2last;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_QUADS] = translate_quads_uint2uint_first2last;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_uint2uint_first2last;
+translate[IN_UINT][OUT_UINT][PV_FIRST][PV_LAST][PIPE_PRIM_POLYGON] = translate_polygon_uint2uint_first2last;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_POINTS] = translate_points_uint2uint_last2first;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_LINES] = translate_lines_uint2uint_last2first;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_uint2uint_last2first;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_uint2uint_last2first;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLES] = translate_tris_uint2uint_last2first;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_uint2uint_last2first;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_uint2uint_last2first;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_QUADS] = translate_quads_uint2uint_last2first;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_uint2uint_last2first;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_FIRST][PIPE_PRIM_POLYGON] = translate_polygon_uint2uint_last2first;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_POINTS] = translate_points_uint2uint_last2last;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_LINES] = translate_lines_uint2uint_last2last;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_LINE_STRIP] = translate_linestrip_uint2uint_last2last;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_LINE_LOOP] = translate_lineloop_uint2uint_last2last;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLES] = translate_tris_uint2uint_last2last;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLE_FAN] = translate_trifan_uint2uint_last2last;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_TRIANGLE_STRIP] = translate_tristrip_uint2uint_last2last;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_QUADS] = translate_quads_uint2uint_last2last;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_QUAD_STRIP] = translate_quadstrip_uint2uint_last2last;
+translate[IN_UINT][OUT_UINT][PV_LAST][PV_LAST][PIPE_PRIM_POLYGON] = translate_polygon_uint2uint_last2last;
+}
+#include "indices/u_indices.c"
diff --git a/src/gallium/auxiliary/indices/u_indices_gen.py b/src/gallium/auxiliary/indices/u_indices_gen.py
new file mode 100644
index 00000000000..0dc58d0cd0e
--- /dev/null
+++ b/src/gallium/auxiliary/indices/u_indices_gen.py
@@ -0,0 +1,319 @@
+#!/usr/bin/env python
+copyright = '''
+/*
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * VMWARE AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+'''
+
+GENERATE, UBYTE, USHORT, UINT = 'generate', 'ubyte', 'ushort', 'uint'
+FIRST, LAST = 'first', 'last'
+
+INTYPES = (GENERATE, UBYTE, USHORT, UINT)
+OUTTYPES = (USHORT, UINT)
+PVS=(FIRST, LAST)
+PRIMS=('points', 
+       'lines', 
+       'linestrip', 
+       'lineloop', 
+       'tris', 
+       'trifan', 
+       'tristrip', 
+       'quads', 
+       'quadstrip', 
+       'polygon')
+
+LONGPRIMS=('PIPE_PRIM_POINTS', 
+           'PIPE_PRIM_LINES', 
+           'PIPE_PRIM_LINE_STRIP', 
+           'PIPE_PRIM_LINE_LOOP', 
+           'PIPE_PRIM_TRIANGLES', 
+           'PIPE_PRIM_TRIANGLE_FAN', 
+           'PIPE_PRIM_TRIANGLE_STRIP', 
+           'PIPE_PRIM_QUADS', 
+           'PIPE_PRIM_QUAD_STRIP', 
+           'PIPE_PRIM_POLYGON')
+
+longprim = dict(zip(PRIMS, LONGPRIMS))
+intype_idx = dict(ubyte='IN_UBYTE', ushort='IN_USHORT', uint='IN_UINT')
+outtype_idx = dict(ushort='OUT_USHORT', uint='OUT_UINT')
+pv_idx = dict(first='PV_FIRST', last='PV_LAST')
+
+
+def prolog():
+    print '''/* File automatically generated by indices.py */'''
+    print copyright
+    print r'''
+
+/**
+ * @file
+ * Functions to translate and generate index lists
+ */
+
+#include "indices/u_indices.h"
+#include "indices/u_indices_priv.h"
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_defines.h"
+#include "util/u_memory.h"
+
+
+static unsigned out_size_idx( unsigned index_size )
+{
+   switch (index_size) {
+   case 4: return OUT_UINT;
+   case 2: return OUT_USHORT;
+   default: assert(0); return OUT_USHORT;
+   }
+}
+
+static unsigned in_size_idx( unsigned index_size )
+{
+   switch (index_size) {
+   case 4: return IN_UINT;
+   case 2: return IN_USHORT;
+   case 1: return IN_UBYTE;
+   default: assert(0); return IN_UBYTE;
+   }
+}
+
+
+static u_translate_func translate[IN_COUNT][OUT_COUNT][PV_COUNT][PV_COUNT][PRIM_COUNT];
+static u_generate_func  generate[OUT_COUNT][PV_COUNT][PV_COUNT][PRIM_COUNT];
+
+
+'''
+
+def vert( intype, outtype, v0 ):
+    if intype == GENERATE:
+        return '(' + outtype + ')(' + v0 + ')'
+    else:
+        return '(' + outtype + ')in[' + v0 + ']'
+
+def point( intype, outtype, ptr, v0 ):
+    print '      (' + ptr + ')[0] = ' + vert( intype, outtype, v0 ) + ';'
+
+def line( intype, outtype, ptr, v0, v1 ):
+    print '      (' + ptr + ')[0] = ' + vert( intype, outtype, v0 ) + ';'
+    print '      (' + ptr + ')[1] = ' + vert( intype, outtype, v1 ) + ';'
+
+def tri( intype, outtype, ptr, v0, v1, v2 ):
+    print '      (' + ptr + ')[0] = ' + vert( intype, outtype, v0 ) + ';'
+    print '      (' + ptr + ')[1] = ' + vert( intype, outtype, v1 ) + ';'
+    print '      (' + ptr + ')[2] = ' + vert( intype, outtype, v2 ) + ';'
+
+def do_point( intype, outtype, ptr, v0 ):
+    point( intype, outtype, ptr, v0 )
+
+def do_line( intype, outtype, ptr, v0, v1, inpv, outpv ):
+    if inpv == outpv:
+        line( intype, outtype, ptr, v0, v1 )
+    else:
+        line( intype, outtype, ptr, v1, v0 )
+
+def do_tri( intype, outtype, ptr, v0, v1, v2, inpv, outpv ):
+    if inpv == outpv:
+        tri( intype, outtype, ptr, v0, v1, v2 )
+    else: 
+        if inpv == FIRST:
+            tri( intype, outtype, ptr, v1, v2, v0 )
+        else:
+            tri( intype, outtype, ptr, v2, v0, v1 )
+
+def do_quad( intype, outtype, ptr, v0, v1, v2, v3, inpv, outpv ):
+    do_tri( intype, outtype, ptr+'+0',  v0, v1, v3, inpv, outpv );
+    do_tri( intype, outtype, ptr+'+3',  v1, v2, v3, inpv, outpv );
+
+def name(intype, outtype, inpv, outpv, prim):
+    if intype == GENERATE:
+        return 'generate_' + prim + '_' + outtype + '_' + inpv + '2' + outpv
+    else:
+        return 'translate_' + prim + '_' + intype + '2' + outtype + '_' + inpv + '2' + outpv
+
+def preamble(intype, outtype, inpv, outpv, prim):
+    print 'static void ' + name( intype, outtype, inpv, outpv, prim ) + '('
+    if intype != GENERATE:
+        print '    const void * _in,'
+    print '    unsigned nr,'
+    print '    void *_out )'
+    print '{'
+    if intype != GENERATE:
+        print '  const ' + intype + '*in = (const ' + intype + '*)_in;'
+    print '  ' + outtype + ' *out = (' + outtype + '*)_out;'
+    print '  unsigned i, j;'
+    print '  (void)j;'
+
+def postamble():
+    print '}'
+
+
+def points(intype, outtype, inpv, outpv):
+    preamble(intype, outtype, inpv, outpv, prim='points')
+    print '  for (i = 0; i < nr; i++) { '
+    do_point( intype, outtype, 'out+i',  'i' );
+    print '   }'
+    postamble()
+
+def lines(intype, outtype, inpv, outpv):
+    preamble(intype, outtype, inpv, outpv, prim='lines')
+    print '  for (i = 0; i < nr; i+=2) { '
+    do_line( intype, outtype, 'out+i',  'i', 'i+1', inpv, outpv );
+    print '   }'
+    postamble()
+
+def linestrip(intype, outtype, inpv, outpv):
+    preamble(intype, outtype, inpv, outpv, prim='linestrip')
+    print '  for (j = i = 0; j < nr; j+=2, i++) { '
+    do_line( intype, outtype, 'out+j',  'i', 'i+1', inpv, outpv );
+    print '   }'
+    postamble()
+
+def lineloop(intype, outtype, inpv, outpv):
+    preamble(intype, outtype, inpv, outpv, prim='lineloop')
+    print '  for (j = i = 0; j < nr - 2; j+=2, i++) { '
+    do_line( intype, outtype, 'out+j',  'i', 'i+1', inpv, outpv );
+    print '   }'
+    do_line( intype, outtype, 'out+j',  'i', '0', inpv, outpv );
+    postamble()
+
+def tris(intype, outtype, inpv, outpv):
+    preamble(intype, outtype, inpv, outpv, prim='tris')
+    print '  for (i = 0; i < nr; i+=3) { '
+    do_tri( intype, outtype, 'out+i',  'i', 'i+1', 'i+2', inpv, outpv );
+    print '   }'
+    postamble()
+
+
+def tristrip(intype, outtype, inpv, outpv):
+    preamble(intype, outtype, inpv, outpv, prim='tristrip')
+    print '  for (j = i = 0; j < nr; j+=3, i++) { '
+    if inpv == FIRST:
+        do_tri( intype, outtype, 'out+j',  'i', 'i+1+(i&1)', 'i+2-(i&1)', inpv, outpv );
+    else:
+        do_tri( intype, outtype, 'out+j',  'i+(i&1)', 'i+1-(i&1)', 'i+2', inpv, outpv );
+    print '   }'
+    postamble()
+
+
+def trifan(intype, outtype, inpv, outpv):
+    preamble(intype, outtype, inpv, outpv, prim='trifan')
+    print '  for (j = i = 0; j < nr; j+=3, i++) { '
+    do_tri( intype, outtype, 'out+j',  '0', 'i+1', 'i+2', inpv, outpv );
+    print '   }'
+    postamble()
+
+
+
+def polygon(intype, outtype, inpv, outpv):
+    preamble(intype, outtype, inpv, outpv, prim='polygon')
+    print '  for (j = i = 0; j < nr; j+=3, i++) { '
+    if inpv == FIRST:
+        do_tri( intype, outtype, 'out+j',  '0', 'i+1', 'i+2', inpv, outpv );
+    else:
+        do_tri( intype, outtype, 'out+j',  'i+1', 'i+2', '0', inpv, outpv );
+    print '   }'
+    postamble()
+
+
+def quads(intype, outtype, inpv, outpv):
+    preamble(intype, outtype, inpv, outpv, prim='quads')
+    print '  for (j = i = 0; j < nr; j+=6, i+=4) { '
+    do_quad( intype, outtype, 'out+j', 'i+0', 'i+1', 'i+2', 'i+3', inpv, outpv );
+    print '   }'
+    postamble()
+
+
+def quadstrip(intype, outtype, inpv, outpv):
+    preamble(intype, outtype, inpv, outpv, prim='quadstrip')
+    print '  for (j = i = 0; j < nr; j+=6, i+=2) { '
+    do_quad( intype, outtype, 'out+j', 'i+2', 'i+0', 'i+1', 'i+3', inpv, outpv );
+    print '   }'
+    postamble()
+
+
+def emit_funcs():
+    for intype in INTYPES:
+        for outtype in OUTTYPES:
+            for inpv in (FIRST, LAST):
+                for outpv in (FIRST, LAST):
+                    points(intype, outtype, inpv, outpv)
+                    lines(intype, outtype, inpv, outpv)
+                    linestrip(intype, outtype, inpv, outpv)
+                    lineloop(intype, outtype, inpv, outpv)
+                    tris(intype, outtype, inpv, outpv)
+                    tristrip(intype, outtype, inpv, outpv)
+                    trifan(intype, outtype, inpv, outpv)
+                    quads(intype, outtype, inpv, outpv)
+                    quadstrip(intype, outtype, inpv, outpv)
+                    polygon(intype, outtype, inpv, outpv)
+
+def init(intype, outtype, inpv, outpv, prim):
+    if intype == GENERATE:
+        print ('generate[' + 
+               outtype_idx[outtype] + 
+               '][' + pv_idx[inpv] + 
+               '][' + pv_idx[outpv] + 
+               '][' + longprim[prim] + 
+               '] = ' + name( intype, outtype, inpv, outpv, prim ) + ';')
+    else:
+        print ('translate[' + 
+               intype_idx[intype] + 
+               '][' + outtype_idx[outtype] + 
+               '][' + pv_idx[inpv] + 
+               '][' + pv_idx[outpv] + 
+               '][' + longprim[prim] + 
+               '] = ' + name( intype, outtype, inpv, outpv, prim ) + ';')
+
+
+def emit_all_inits():
+    for intype in INTYPES:
+        for outtype in OUTTYPES:
+            for inpv in PVS:
+                for outpv in PVS:
+                    for prim in PRIMS:
+                        init(intype, outtype, inpv, outpv, prim)
+
+def emit_init():
+    print 'void u_index_init( void )'
+    print '{'
+    print '  static int firsttime = 1;'
+    print '  if (!firsttime) return;'
+    print '  firsttime = 0;'
+    emit_all_inits()
+    print '}'
+
+
+    
+
+def epilog():
+    print '#include "indices/u_indices.c"'
+
+
+def main():
+    prolog()
+    emit_funcs()
+    emit_init()
+    epilog()
+
+
+if __name__ == '__main__':
+    main()
diff --git a/src/gallium/auxiliary/indices/u_indices_priv.h b/src/gallium/auxiliary/indices/u_indices_priv.h
new file mode 100644
index 00000000000..9c3298c24dc
--- /dev/null
+++ b/src/gallium/auxiliary/indices/u_indices_priv.h
@@ -0,0 +1,43 @@
+/*
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+#ifndef U_INDICES_PRIV_H
+#define U_INDICES_PRIV_H
+
+#include "pipe/p_compiler.h"
+#include "u_indices.h"
+
+#define IN_UBYTE      0
+#define IN_USHORT     1
+#define IN_UINT       2
+#define IN_COUNT      3
+
+#define OUT_USHORT    0
+#define OUT_UINT      1
+#define OUT_COUNT     2
+
+
+#define PRIM_COUNT   (PIPE_PRIM_POLYGON + 1)
+
+#endif
diff --git a/src/gallium/auxiliary/pipebuffer/Makefile b/src/gallium/auxiliary/pipebuffer/Makefile
new file mode 100644
index 00000000000..3b501c51dff
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/Makefile
@@ -0,0 +1,22 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = pipebuffer
+
+C_SOURCES = \
+	pb_buffer_fenced.c \
+	pb_buffer_malloc.c \
+	pb_bufmgr_alt.c \
+	pb_bufmgr_cache.c \
+	pb_bufmgr_debug.c \
+	pb_bufmgr_fenced.c \
+	pb_bufmgr_mm.c \
+	pb_bufmgr_ondemand.c \
+	pb_bufmgr_pool.c \
+	pb_bufmgr_slab.c \
+	pb_validate.c
+
+include ../../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/auxiliary/pipebuffer/SConscript b/src/gallium/auxiliary/pipebuffer/SConscript
new file mode 100644
index 00000000000..8e9f06abe45
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/SConscript
@@ -0,0 +1,19 @@
+Import('*')
+
+pipebuffer = env.ConvenienceLibrary(
+	target = 'pipebuffer',
+	source = [
+		'pb_buffer_fenced.c',
+		'pb_buffer_malloc.c',
+		'pb_bufmgr_alt.c',
+		'pb_bufmgr_cache.c',
+		'pb_bufmgr_debug.c',
+		'pb_bufmgr_fenced.c',
+		'pb_bufmgr_mm.c',
+		'pb_bufmgr_ondemand.c',
+		'pb_bufmgr_pool.c',
+		'pb_bufmgr_slab.c',
+		'pb_validate.c',
+	])
+
+auxiliaries.insert(0, pipebuffer)
diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer.h b/src/gallium/auxiliary/pipebuffer/pb_buffer.h
new file mode 100644
index 00000000000..d8f1f02d681
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/pb_buffer.h
@@ -0,0 +1,292 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Generic code for buffers.
+ * 
+ * Behind a pipe buffle handle there can be DMA buffers, client (or user) 
+ * buffers, regular malloced buffers, etc. This file provides an abstract base 
+ * buffer handle that allows the driver to cope with all those kinds of buffers 
+ * in a more flexible way.
+ * 
+ * There is no obligation of a winsys driver to use this library. And a pipe
+ * driver should be completly agnostic about it.
+ * 
+ * \author Jose Fonseca <[email protected]>
+ */
+
+#ifndef PB_BUFFER_H_
+#define PB_BUFFER_H_
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_error.h"
+#include "pipe/p_state.h"
+#include "pipe/p_inlines.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+struct pb_vtbl;
+struct pb_validate;
+
+
+/**
+ * Buffer description.
+ * 
+ * Used when allocating the buffer.
+ */
+struct pb_desc
+{
+   unsigned alignment;
+   unsigned usage;
+};
+
+
+/**
+ * Base class for all pb_* buffers.
+ */
+struct pb_buffer 
+{
+   struct pipe_buffer base;
+
+   /**
+    * Pointer to the virtual function table.
+    *
+    * Avoid accessing this table directly. Use the inline functions below 
+    * instead to avoid mistakes. 
+    */
+   const struct pb_vtbl *vtbl;
+};
+
+
+/**
+ * Virtual function table for the buffer storage operations.
+ * 
+ * Note that creation is not done through this table.
+ */
+struct pb_vtbl
+{
+   void (*destroy)( struct pb_buffer *buf );
+
+   /** 
+    * Map the entire data store of a buffer object into the client's address.
+    * flags is bitmask of PIPE_BUFFER_FLAG_READ/WRITE. 
+    */
+   void *(*map)( struct pb_buffer *buf, 
+                 unsigned flags );
+   
+   void (*unmap)( struct pb_buffer *buf );
+
+   enum pipe_error (*validate)( struct pb_buffer *buf, 
+                                struct pb_validate *vl,
+                                unsigned flags );
+
+   void (*fence)( struct pb_buffer *buf, 
+                  struct pipe_fence_handle *fence );
+
+   /**
+    * Get the base buffer and the offset.
+    * 
+    * A buffer can be subdivided in smaller buffers. This method should return
+    * the underlaying buffer, and the relative offset.
+    * 
+    * Buffers without an underlaying base buffer should return themselves, with 
+    * a zero offset.
+    * 
+    * Note that this will increase the reference count of the base buffer.
+    */
+   void (*get_base_buffer)( struct pb_buffer *buf,
+                            struct pb_buffer **base_buf,
+                            unsigned *offset );
+   
+};
+
+
+static INLINE struct pipe_buffer *
+pb_pipe_buffer( struct pb_buffer *pbuf )
+{
+   assert(pbuf);
+   return &pbuf->base;
+}
+
+
+static INLINE struct pb_buffer *
+pb_buffer( struct pipe_buffer *buf )
+{
+   assert(buf);
+   /* Could add a magic cookie check on debug builds.
+    */
+   return (struct pb_buffer *)buf;
+}
+
+
+/* Accessor functions for pb->vtbl:
+ */
+static INLINE void *
+pb_map(struct pb_buffer *buf, 
+       unsigned flags)
+{
+   assert(buf);
+   if(!buf)
+      return NULL;
+   assert(buf->base.refcount > 0);
+   return buf->vtbl->map(buf, flags);
+}
+
+
+static INLINE void 
+pb_unmap(struct pb_buffer *buf)
+{
+   assert(buf);
+   if(!buf)
+      return;
+   assert(buf->base.refcount > 0);
+   buf->vtbl->unmap(buf);
+}
+
+
+static INLINE void
+pb_get_base_buffer( struct pb_buffer *buf,
+		    struct pb_buffer **base_buf,
+		    unsigned *offset )
+{
+   assert(buf);
+   if(!buf) {
+      base_buf = NULL;
+      offset = 0;
+      return;
+   }
+   assert(buf->base.refcount > 0);
+   assert(buf->vtbl->get_base_buffer);
+   buf->vtbl->get_base_buffer(buf, base_buf, offset);
+   assert(*base_buf);
+   assert(*offset < (*base_buf)->base.size);
+}
+
+
+static INLINE enum pipe_error 
+pb_validate(struct pb_buffer *buf, struct pb_validate *vl, unsigned flags)
+{
+   assert(buf);
+   if(!buf)
+      return PIPE_ERROR;
+   assert(buf->vtbl->validate);
+   return buf->vtbl->validate(buf, vl, flags);
+}
+
+
+static INLINE void 
+pb_fence(struct pb_buffer *buf, struct pipe_fence_handle *fence)
+{
+   assert(buf);
+   if(!buf)
+      return;
+   assert(buf->vtbl->fence);
+   buf->vtbl->fence(buf, fence);
+}
+
+
+static INLINE void 
+pb_destroy(struct pb_buffer *buf)
+{
+   assert(buf);
+   if(!buf)
+      return;
+   assert(buf->base.refcount == 0);
+   buf->vtbl->destroy(buf);
+}
+
+
+/* XXX: thread safety issues!
+ */
+static INLINE void
+pb_reference(struct pb_buffer **dst,
+             struct pb_buffer *src)
+{
+   if (src) {
+      assert(src->base.refcount);
+      src->base.refcount++;
+   }
+
+   if (*dst) {
+      assert((*dst)->base.refcount);
+      if(--(*dst)->base.refcount == 0)
+         pb_destroy( *dst );
+   }
+
+   *dst = src;
+}
+
+
+/**
+ * Utility function to check whether the provided alignment is consistent with
+ * the requested or not.
+ */
+static INLINE boolean
+pb_check_alignment(size_t requested, size_t provided)
+{
+   if(!requested)
+      return TRUE;
+   if(requested > provided)
+      return FALSE;
+   if(provided % requested != 0)
+      return FALSE;
+   return TRUE;
+}
+
+
+/**
+ * Utility function to check whether the provided alignment is consistent with
+ * the requested or not.
+ */
+static INLINE boolean
+pb_check_usage(unsigned requested, unsigned provided)
+{
+   return (requested & provided) == requested ? TRUE : FALSE;
+}
+
+
+/**
+ * Malloc-based buffer to store data that can't be used by the graphics 
+ * hardware.
+ */
+struct pb_buffer *
+pb_malloc_buffer_create(size_t size, 
+                        const struct pb_desc *desc);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /*PB_BUFFER_H_*/
diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
new file mode 100644
index 00000000000..f9e62264368
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.c
@@ -0,0 +1,501 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Implementation of fenced buffers.
+ * 
+ * \author Jose Fonseca <jrfonseca-at-tungstengraphics-dot-com>
+ * \author Thomas Hellström <thomas-at-tungstengraphics-dot-com>
+ */
+
+
+#include "pipe/p_config.h"
+
+#if defined(PIPE_OS_LINUX)
+#include <unistd.h>
+#include <sched.h>
+#endif
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_error.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_thread.h"
+#include "util/u_memory.h"
+#include "util/u_double_list.h"
+
+#include "pb_buffer.h"
+#include "pb_buffer_fenced.h"
+
+
+
+/**
+ * Convenience macro (type safe).
+ */
+#define SUPER(__derived) (&(__derived)->base)
+
+
+struct fenced_buffer_list
+{
+   pipe_mutex mutex;
+   
+   struct pb_fence_ops *ops;
+   
+   size_t numDelayed;
+   
+   struct list_head delayed;
+};
+
+
+/**
+ * Wrapper around a pipe buffer which adds fencing and reference counting.
+ */
+struct fenced_buffer
+{
+   struct pb_buffer base;
+   
+   struct pb_buffer *buffer;
+
+   /* FIXME: protect access with mutex */
+
+   /**
+    * A bitmask of PIPE_BUFFER_USAGE_CPU/GPU_READ/WRITE describing the current
+    * buffer usage.
+    */
+   unsigned flags;
+
+   unsigned mapcount;
+   struct pb_validate *vl;
+   unsigned validation_flags;
+   struct pipe_fence_handle *fence;
+
+   struct list_head head;
+   struct fenced_buffer_list *list;
+};
+
+
+static INLINE struct fenced_buffer *
+fenced_buffer(struct pb_buffer *buf)
+{
+   assert(buf);
+   return (struct fenced_buffer *)buf;
+}
+
+
+static INLINE void
+_fenced_buffer_add(struct fenced_buffer *fenced_buf)
+{
+   struct fenced_buffer_list *fenced_list = fenced_buf->list;
+
+   assert(fenced_buf->base.base.refcount);
+   assert(fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE);
+   assert(fenced_buf->fence);
+
+   assert(!fenced_buf->head.prev);
+   assert(!fenced_buf->head.next);
+   LIST_ADDTAIL(&fenced_buf->head, &fenced_list->delayed);
+   ++fenced_list->numDelayed;
+}
+
+
+/**
+ * Actually destroy the buffer.
+ */
+static INLINE void
+_fenced_buffer_destroy(struct fenced_buffer *fenced_buf)
+{
+   assert(!fenced_buf->base.base.refcount);
+   assert(!fenced_buf->fence);
+   pb_reference(&fenced_buf->buffer, NULL);
+   FREE(fenced_buf);
+}
+
+
+static INLINE void
+_fenced_buffer_remove(struct fenced_buffer_list *fenced_list,
+                      struct fenced_buffer *fenced_buf)
+{
+   struct pb_fence_ops *ops = fenced_list->ops;
+
+   assert(fenced_buf->fence);
+   assert(fenced_buf->list == fenced_list);
+   
+   ops->fence_reference(ops, &fenced_buf->fence, NULL);
+   fenced_buf->flags &= ~PIPE_BUFFER_USAGE_GPU_READ_WRITE;
+   
+   assert(fenced_buf->head.prev);
+   assert(fenced_buf->head.next);
+   LIST_DEL(&fenced_buf->head);
+#ifdef DEBUG
+   fenced_buf->head.prev = NULL;
+   fenced_buf->head.next = NULL;
+#endif
+   
+   assert(fenced_list->numDelayed);
+   --fenced_list->numDelayed;
+   
+   if(!fenced_buf->base.base.refcount)
+      _fenced_buffer_destroy(fenced_buf);
+}
+
+
+static INLINE enum pipe_error
+_fenced_buffer_finish(struct fenced_buffer *fenced_buf)
+{
+   struct fenced_buffer_list *fenced_list = fenced_buf->list;
+   struct pb_fence_ops *ops = fenced_list->ops;
+
+#if 0
+   debug_warning("waiting for GPU");
+#endif
+
+   assert(fenced_buf->fence);
+   if(fenced_buf->fence) {
+      if(ops->fence_finish(ops, fenced_buf->fence, 0) != 0) {
+	 return PIPE_ERROR;
+      }
+      /* Remove from the fenced list */
+      /* TODO: remove consequents */
+      _fenced_buffer_remove(fenced_list, fenced_buf);
+   }
+
+   fenced_buf->flags &= ~PIPE_BUFFER_USAGE_GPU_READ_WRITE;
+   return PIPE_OK;
+}
+
+
+/**
+ * Free as many fenced buffers from the list head as possible. 
+ */
+static void
+_fenced_buffer_list_check_free(struct fenced_buffer_list *fenced_list, 
+                               int wait)
+{
+   struct pb_fence_ops *ops = fenced_list->ops;
+   struct list_head *curr, *next;
+   struct fenced_buffer *fenced_buf;
+   struct pipe_fence_handle *prev_fence = NULL;
+
+   curr = fenced_list->delayed.next;
+   next = curr->next;
+   while(curr != &fenced_list->delayed) {
+      fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head);
+
+      if(fenced_buf->fence != prev_fence) {
+	 int signaled;
+	 if (wait)
+	    signaled = ops->fence_finish(ops, fenced_buf->fence, 0);
+	 else
+	    signaled = ops->fence_signalled(ops, fenced_buf->fence, 0);
+	 if (signaled != 0)
+	    break;
+	 prev_fence = fenced_buf->fence;
+      }
+      else {
+	 assert(ops->fence_signalled(ops, fenced_buf->fence, 0) == 0);
+      }
+
+      _fenced_buffer_remove(fenced_list, fenced_buf);
+
+      curr = next; 
+      next = curr->next;
+   }
+}
+
+
+static void
+fenced_buffer_destroy(struct pb_buffer *buf)
+{
+   struct fenced_buffer *fenced_buf = fenced_buffer(buf);   
+   struct fenced_buffer_list *fenced_list = fenced_buf->list;
+
+   pipe_mutex_lock(fenced_list->mutex);
+   assert(fenced_buf->base.base.refcount == 0);
+   if (fenced_buf->fence) {
+      struct pb_fence_ops *ops = fenced_list->ops;
+      if(ops->fence_signalled(ops, fenced_buf->fence, 0) == 0) {
+	 struct list_head *curr, *prev;
+	 curr = &fenced_buf->head;
+	 prev = curr->prev;
+	 do {
+	    fenced_buf = LIST_ENTRY(struct fenced_buffer, curr, head);
+	    assert(ops->fence_signalled(ops, fenced_buf->fence, 0) == 0);
+	    _fenced_buffer_remove(fenced_list, fenced_buf);
+	    curr = prev;
+	    prev = curr->prev;
+	 } while (curr != &fenced_list->delayed);
+      }	  
+      else {
+	 /* delay destruction */
+      }
+   }
+   else {
+      _fenced_buffer_destroy(fenced_buf);
+   }
+   pipe_mutex_unlock(fenced_list->mutex);
+}
+
+
+static void *
+fenced_buffer_map(struct pb_buffer *buf, 
+                  unsigned flags)
+{
+   struct fenced_buffer *fenced_buf = fenced_buffer(buf);
+   void *map;
+
+   assert(flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE);
+   assert(!(flags & ~PIPE_BUFFER_USAGE_CPU_READ_WRITE));
+   flags &= PIPE_BUFFER_USAGE_CPU_READ_WRITE;
+   
+   /* Check for GPU read/write access */
+   if(fenced_buf->flags & PIPE_BUFFER_USAGE_GPU_WRITE) {
+      /* Wait for the GPU to finish writing */
+      _fenced_buffer_finish(fenced_buf);
+   }
+
+#if 0
+   /* Check for CPU write access (read is OK) */
+   if(fenced_buf->flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE) {
+      /* this is legal -- just for debugging */
+      debug_warning("concurrent CPU writes");
+   }
+#endif
+   
+   map = pb_map(fenced_buf->buffer, flags);
+   if(map) {
+      ++fenced_buf->mapcount;
+      fenced_buf->flags |= flags;
+   }
+
+   return map;
+}
+
+
+static void
+fenced_buffer_unmap(struct pb_buffer *buf)
+{
+   struct fenced_buffer *fenced_buf = fenced_buffer(buf);
+   assert(fenced_buf->mapcount);
+   if(fenced_buf->mapcount) {
+      pb_unmap(fenced_buf->buffer);
+      --fenced_buf->mapcount;
+      if(!fenced_buf->mapcount)
+	 fenced_buf->flags &= ~PIPE_BUFFER_USAGE_CPU_READ_WRITE;
+   }
+}
+
+
+static enum pipe_error
+fenced_buffer_validate(struct pb_buffer *buf,
+                       struct pb_validate *vl,
+                       unsigned flags)
+{
+   struct fenced_buffer *fenced_buf = fenced_buffer(buf);
+   enum pipe_error ret;
+   
+   if(!vl) {
+      /* invalidate */
+      fenced_buf->vl = NULL;
+      fenced_buf->validation_flags = 0;
+      return PIPE_OK;
+   }
+   
+   assert(flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE);
+   assert(!(flags & ~PIPE_BUFFER_USAGE_GPU_READ_WRITE));
+   flags &= PIPE_BUFFER_USAGE_GPU_READ_WRITE;
+
+   /* Buffer cannot be validated in two different lists */ 
+   if(fenced_buf->vl && fenced_buf->vl != vl)
+      return PIPE_ERROR_RETRY;
+   
+   /* Do not validate if buffer is still mapped */
+   if(fenced_buf->flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE) {
+      /* TODO: wait for the thread that mapped the buffer to unmap it */
+      return PIPE_ERROR_RETRY;
+   }
+
+   if(fenced_buf->vl == vl &&
+      (fenced_buf->validation_flags & flags) == flags) {
+      /* Nothing to do -- buffer already validated */
+      return PIPE_OK;
+   }
+
+   /* Final sanity checking */
+   assert(!(fenced_buf->flags & PIPE_BUFFER_USAGE_CPU_READ_WRITE));
+   assert(!fenced_buf->mapcount);
+   
+   ret = pb_validate(fenced_buf->buffer, vl, flags);
+   if (ret != PIPE_OK)
+      return ret;
+   
+   fenced_buf->vl = vl;
+   fenced_buf->validation_flags |= flags;
+   
+   return PIPE_OK;
+}
+
+
+static void
+fenced_buffer_fence(struct pb_buffer *buf,
+                    struct pipe_fence_handle *fence)
+{
+   struct fenced_buffer *fenced_buf;
+   struct fenced_buffer_list *fenced_list;
+   struct pb_fence_ops *ops;
+
+   fenced_buf = fenced_buffer(buf);
+   fenced_list = fenced_buf->list;
+   ops = fenced_list->ops;
+   
+   if(fence == fenced_buf->fence) {
+      /* Nothing to do */
+      return;
+   }
+
+   assert(fenced_buf->vl);
+   assert(fenced_buf->validation_flags);
+   
+   pipe_mutex_lock(fenced_list->mutex);
+   if (fenced_buf->fence)
+      _fenced_buffer_remove(fenced_list, fenced_buf);
+   if (fence) {
+      ops->fence_reference(ops, &fenced_buf->fence, fence);
+      fenced_buf->flags |= fenced_buf->validation_flags;
+      _fenced_buffer_add(fenced_buf);
+   }
+   pipe_mutex_unlock(fenced_list->mutex);
+   
+   pb_fence(fenced_buf->buffer, fence);
+
+   fenced_buf->vl = NULL;
+   fenced_buf->validation_flags = 0;
+}
+
+
+static void
+fenced_buffer_get_base_buffer(struct pb_buffer *buf,
+                              struct pb_buffer **base_buf,
+                              unsigned *offset)
+{
+   struct fenced_buffer *fenced_buf = fenced_buffer(buf);
+   pb_get_base_buffer(fenced_buf->buffer, base_buf, offset);
+}
+
+
+static const struct pb_vtbl 
+fenced_buffer_vtbl = {
+      fenced_buffer_destroy,
+      fenced_buffer_map,
+      fenced_buffer_unmap,
+      fenced_buffer_validate,
+      fenced_buffer_fence,
+      fenced_buffer_get_base_buffer
+};
+
+
+struct pb_buffer *
+fenced_buffer_create(struct fenced_buffer_list *fenced_list, 
+                     struct pb_buffer *buffer)
+{
+   struct fenced_buffer *buf;
+   
+   if(!buffer)
+      return NULL;
+   
+   buf = CALLOC_STRUCT(fenced_buffer);
+   if(!buf) {
+      pb_reference(&buffer, NULL);
+      return NULL;
+   }
+   
+   buf->base.base.refcount = 1;
+   buf->base.base.alignment = buffer->base.alignment;
+   buf->base.base.usage = buffer->base.usage;
+   buf->base.base.size = buffer->base.size;
+   
+   buf->base.vtbl = &fenced_buffer_vtbl;
+   buf->buffer = buffer;
+   buf->list = fenced_list;
+   
+   return &buf->base;
+}
+
+
+struct fenced_buffer_list *
+fenced_buffer_list_create(struct pb_fence_ops *ops) 
+{
+   struct fenced_buffer_list *fenced_list;
+
+   fenced_list = CALLOC_STRUCT(fenced_buffer_list);
+   if (!fenced_list)
+      return NULL;
+
+   fenced_list->ops = ops;
+
+   LIST_INITHEAD(&fenced_list->delayed);
+
+   fenced_list->numDelayed = 0;
+   
+   pipe_mutex_init(fenced_list->mutex);
+
+   return fenced_list;
+}
+
+
+void
+fenced_buffer_list_check_free(struct fenced_buffer_list *fenced_list, 
+                              int wait)
+{
+   pipe_mutex_lock(fenced_list->mutex);
+   _fenced_buffer_list_check_free(fenced_list, wait);
+   pipe_mutex_unlock(fenced_list->mutex);
+}
+
+
+void
+fenced_buffer_list_destroy(struct fenced_buffer_list *fenced_list)
+{
+   pipe_mutex_lock(fenced_list->mutex);
+
+   /* Wait on outstanding fences */
+   while (fenced_list->numDelayed) {
+      pipe_mutex_unlock(fenced_list->mutex);
+#if defined(PIPE_OS_LINUX)
+      sched_yield();
+#endif
+      _fenced_buffer_list_check_free(fenced_list, 1);
+      pipe_mutex_lock(fenced_list->mutex);
+   }
+
+   pipe_mutex_unlock(fenced_list->mutex);
+   
+   fenced_list->ops->destroy(fenced_list->ops);
+   
+   FREE(fenced_list);
+}
+
+
diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.h b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.h
new file mode 100644
index 00000000000..d1c9d4c17df
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/pb_buffer_fenced.h
@@ -0,0 +1,135 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Buffer fencing.
+ * 
+ * "Fenced buffers" is actually a misnomer. They should be referred as 
+ * "fenceable buffers", i.e, buffers that can be fenced, but I couldn't find
+ * the word "fenceable" in the dictionary.
+ * 
+ * A "fenced buffer" is a decorator around a normal buffer, which adds two 
+ * special properties:
+ * - the ability for the destruction to be delayed by a fence;
+ * - reference counting.
+ * 
+ * Usually DMA buffers have a life-time that will extend the life-time of its 
+ * handle. The end-of-life is dictated by the fence signalling. 
+ * 
+ * Between the handle's destruction, and the fence signalling, the buffer is 
+ * stored in a fenced buffer list.
+ * 
+ * \author Jose Fonseca <[email protected]>
+ */
+
+#ifndef PB_BUFFER_FENCED_H_
+#define PB_BUFFER_FENCED_H_
+
+
+#include "pipe/p_debug.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+struct pipe_buffer;
+struct pipe_fence_handle;
+
+
+/**
+ * List of buffers which are awaiting fence signalling.
+ */
+struct fenced_buffer_list;
+
+
+struct pb_fence_ops
+{
+   void (*destroy)( struct pb_fence_ops *ops );
+
+   /** Set ptr = fence, with reference counting */
+   void (*fence_reference)( struct pb_fence_ops *ops,
+                            struct pipe_fence_handle **ptr,
+                            struct pipe_fence_handle *fence );
+
+   /**
+    * Checks whether the fence has been signalled.
+    * \param flags  driver-specific meaning
+    * \return zero on success.
+    */
+   int (*fence_signalled)( struct pb_fence_ops *ops,
+                           struct pipe_fence_handle *fence,
+                           unsigned flag );
+
+   /**
+    * Wait for the fence to finish.
+    * \param flags  driver-specific meaning
+    * \return zero on success.
+    */
+   int (*fence_finish)( struct pb_fence_ops *ops,
+                        struct pipe_fence_handle *fence,
+                        unsigned flag );
+};
+
+
+/**
+ * Create a fenced buffer list.
+ * 
+ * See also fenced_bufmgr_create for a more convenient way to use this.
+ */
+struct fenced_buffer_list *
+fenced_buffer_list_create(struct pb_fence_ops *ops);
+
+
+/**
+ * Walk the fenced buffer list to check and free signalled buffers.
+ */ 
+void
+fenced_buffer_list_check_free(struct fenced_buffer_list *fenced_list, 
+                              int wait);
+
+void
+fenced_buffer_list_destroy(struct fenced_buffer_list *fenced_list);
+
+
+/**
+ * Wrap a buffer in a fenced buffer.
+ * 
+ * NOTE: this will not increase the buffer reference count.
+ */
+struct pb_buffer *
+fenced_buffer_create(struct fenced_buffer_list *fenced, 
+                     struct pb_buffer *buffer);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /*PB_BUFFER_FENCED_H_*/
diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer_malloc.c b/src/gallium/auxiliary/pipebuffer/pb_buffer_malloc.c
new file mode 100644
index 00000000000..53f497cfb04
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/pb_buffer_malloc.c
@@ -0,0 +1,186 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Implementation of malloc-based buffers to store data that can't be processed
+ * by the hardware. 
+ * 
+ * \author Jose Fonseca <[email protected]>
+ */
+
+
+#include "pipe/p_debug.h"
+#include "util/u_memory.h"
+#include "pb_buffer.h"
+#include "pb_bufmgr.h"
+
+
+struct malloc_buffer 
+{
+   struct pb_buffer base;
+   void *data;
+};
+
+
+extern const struct pb_vtbl malloc_buffer_vtbl;
+
+static INLINE struct malloc_buffer *
+malloc_buffer(struct pb_buffer *buf)
+{
+   assert(buf);
+   assert(buf->vtbl == &malloc_buffer_vtbl);
+   return (struct malloc_buffer *)buf;
+}
+
+
+static void
+malloc_buffer_destroy(struct pb_buffer *buf)
+{
+   align_free(malloc_buffer(buf)->data);
+   FREE(buf);
+}
+
+
+static void *
+malloc_buffer_map(struct pb_buffer *buf, 
+                  unsigned flags)
+{
+   return malloc_buffer(buf)->data;
+}
+
+
+static void
+malloc_buffer_unmap(struct pb_buffer *buf)
+{
+   /* No-op */
+}
+
+
+static enum pipe_error 
+malloc_buffer_validate(struct pb_buffer *buf, 
+                       struct pb_validate *vl,
+                       unsigned flags)
+{
+   assert(0);
+   return PIPE_ERROR;
+}
+
+
+static void
+malloc_buffer_fence(struct pb_buffer *buf, 
+                    struct pipe_fence_handle *fence)
+{
+   assert(0);
+}
+
+
+static void
+malloc_buffer_get_base_buffer(struct pb_buffer *buf,
+                              struct pb_buffer **base_buf,
+                              unsigned *offset)
+{
+   *base_buf = buf;
+   *offset = 0;
+}
+
+
+const struct pb_vtbl 
+malloc_buffer_vtbl = {
+      malloc_buffer_destroy,
+      malloc_buffer_map,
+      malloc_buffer_unmap,
+      malloc_buffer_validate,
+      malloc_buffer_fence,
+      malloc_buffer_get_base_buffer
+};
+
+
+struct pb_buffer *
+pb_malloc_buffer_create(size_t size,
+                   	const struct pb_desc *desc) 
+{
+   struct malloc_buffer *buf;
+   
+   /* TODO: do a single allocation */
+   
+   buf = CALLOC_STRUCT(malloc_buffer);
+   if(!buf)
+      return NULL;
+
+   buf->base.base.refcount = 1;
+   buf->base.base.alignment = desc->alignment;
+   buf->base.base.usage = desc->usage;
+   buf->base.base.size = size;
+   buf->base.vtbl = &malloc_buffer_vtbl;
+
+   buf->data = align_malloc(size, desc->alignment < sizeof(void*) ? sizeof(void*) : desc->alignment);
+   if(!buf->data) {
+      FREE(buf);
+      return NULL;
+   }
+
+   return &buf->base;
+}
+
+
+static struct pb_buffer *
+pb_malloc_bufmgr_create_buffer(struct pb_manager *mgr, 
+                               size_t size,
+                               const struct pb_desc *desc) 
+{
+   return pb_malloc_buffer_create(size, desc);
+}
+
+
+static void
+pb_malloc_bufmgr_flush(struct pb_manager *mgr) 
+{
+   /* No-op */
+}
+
+
+static void
+pb_malloc_bufmgr_destroy(struct pb_manager *mgr) 
+{
+   /* No-op */
+}
+
+
+static struct pb_manager 
+pb_malloc_bufmgr = {
+   pb_malloc_bufmgr_destroy,
+   pb_malloc_bufmgr_create_buffer,
+   pb_malloc_bufmgr_flush
+};
+
+
+struct pb_manager *
+pb_malloc_bufmgr_create(void) 
+{
+  return &pb_malloc_bufmgr;
+}
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h b/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h
new file mode 100644
index 00000000000..fec8db91c71
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h
@@ -0,0 +1,213 @@
+/**************************************************************************
+ *
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Buffer management.
+ * 
+ * A buffer manager does only one basic thing: it creates buffers. Actually,
+ * "buffer factory" would probably a more accurate description.
+ * 
+ * You can chain buffer managers so that you can have a finer grained memory
+ * management and pooling.
+ * 
+ * For example, for a simple batch buffer manager you would chain:
+ * - the native buffer manager, which provides DMA memory from the graphics
+ * memory space;
+ * - the pool buffer manager, which keep around a pool of equally sized buffers
+ * to avoid latency associated with the native buffer manager; 
+ * - the fenced buffer manager, which will delay buffer destruction until the 
+ * the moment the card finishing processing it. 
+ * 
+ * \author Jose Fonseca <[email protected]>
+ */
+
+#ifndef PB_BUFMGR_H_
+#define PB_BUFMGR_H_
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_error.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+struct pb_desc;
+struct pipe_buffer;
+
+
+/** 
+ * Abstract base class for all buffer managers.
+ */
+struct pb_manager
+{
+   void
+   (*destroy)( struct pb_manager *mgr );
+
+   struct pb_buffer *
+   (*create_buffer)( struct pb_manager *mgr, 
+	             size_t size,
+	             const struct pb_desc *desc);
+
+   /**
+    * Flush all temporary-held buffers.
+    * 
+    * Used mostly to aid debugging memory issues or to clean up resources when 
+    * the drivers are long lived.
+    */
+   void
+   (*flush)( struct pb_manager *mgr );
+};
+
+
+/**
+ * Malloc buffer provider.
+ * 
+ * Simple wrapper around pb_malloc_buffer_create for convenience.
+ */
+struct pb_manager *
+pb_malloc_bufmgr_create(void);
+
+
+/** 
+ * Static buffer pool sub-allocator.
+ * 
+ * Manages the allocation of equally sized buffers. It does so by allocating
+ * a single big buffer and divide it equally sized buffers. 
+ * 
+ * It is meant to manage the allocation of batch buffer pools.
+ */
+struct pb_manager *
+pool_bufmgr_create(struct pb_manager *provider, 
+                   size_t n, size_t size,
+                   const struct pb_desc *desc);
+
+
+/** 
+ * Static sub-allocator based the old memory manager.
+ * 
+ * It managers buffers of different sizes. It does so by allocating a buffer
+ * with the size of the heap, and then using the old mm memory manager to manage
+ * that heap. 
+ */
+struct pb_manager *
+mm_bufmgr_create(struct pb_manager *provider, 
+                 size_t size, size_t align2);
+
+/**
+ * Same as mm_bufmgr_create.
+ * 
+ * Buffer will be release when the manager is destroyed.
+ */
+struct pb_manager *
+mm_bufmgr_create_from_buffer(struct pb_buffer *buffer, 
+                             size_t size, size_t align2);
+
+
+/**
+ * Slab sub-allocator.
+ */
+struct pb_manager *
+pb_slab_manager_create(struct pb_manager *provider,
+                       size_t bufSize,
+                       size_t slabSize,
+                       const struct pb_desc *desc);
+
+/**
+ * Allow a range of buffer size, by aggregating multiple slabs sub-allocators 
+ * with different bucket sizes.
+ */
+struct pb_manager *
+pb_slab_range_manager_create(struct pb_manager *provider,
+                             size_t minBufSize,
+                             size_t maxBufSize,
+                             size_t slabSize,
+                             const struct pb_desc *desc);
+
+
+/** 
+ * Time-based buffer cache.
+ *
+ * This manager keeps a cache of destroyed buffers during a time interval. 
+ */
+struct pb_manager *
+pb_cache_manager_create(struct pb_manager *provider, 
+                     	unsigned usecs); 
+
+
+struct pb_fence_ops;
+
+/** 
+ * Fenced buffer manager.
+ *
+ * This manager is just meant for convenience. It wraps the buffers returned
+ * by another manager in fenced buffers, so that  
+ * 
+ * NOTE: the buffer manager that provides the buffers will be destroyed
+ * at the same time.
+ */
+struct pb_manager *
+fenced_bufmgr_create(struct pb_manager *provider,
+                     struct pb_fence_ops *ops);
+
+
+struct pb_manager *
+pb_alt_manager_create(struct pb_manager *provider1, 
+                      struct pb_manager *provider2);
+
+
+/** 
+ * Ondemand buffer manager.
+ * 
+ * Buffers are created in malloc'ed memory (fast and cached), and the constents
+ * is transfered to a buffer from the provider (typically in slow uncached 
+ * memory) when there is an attempt to validate the buffer.
+ * 
+ * Ideal for situations where one does not know before hand whether a given
+ * buffer will effectively be used by the hardware or not. 
+ */
+struct pb_manager *
+pb_ondemand_manager_create(struct pb_manager *provider); 
+
+
+/** 
+ * Debug buffer manager to detect buffer under- and overflows.
+ *
+ * Band size should be a multiple of the largest alignment
+ */
+struct pb_manager *
+pb_debug_manager_create(struct pb_manager *provider, size_t band_size); 
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /*PB_BUFMGR_H_*/
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_alt.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_alt.c
new file mode 100644
index 00000000000..c956924cc76
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_alt.c
@@ -0,0 +1,120 @@
+/**************************************************************************
+ *
+ * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Allocate buffers from two alternative buffer providers.
+ * 
+ * \author Jose Fonseca <[email protected]>
+ */
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+#include "util/u_memory.h"
+
+#include "pb_buffer.h"
+#include "pb_bufmgr.h"
+
+
+struct pb_alt_manager
+{
+   struct pb_manager base;
+
+   struct pb_manager *provider1;
+   struct pb_manager *provider2;
+};
+
+
+static INLINE struct pb_alt_manager *
+pb_alt_manager(struct pb_manager *mgr)
+{
+   assert(mgr);
+   return (struct pb_alt_manager *)mgr;
+}
+
+
+static struct pb_buffer *
+pb_alt_manager_create_buffer(struct pb_manager *_mgr, 
+                             size_t size,
+                             const struct pb_desc *desc)
+{
+   struct pb_alt_manager *mgr = pb_alt_manager(_mgr);
+   struct pb_buffer *buf;
+   
+   buf = mgr->provider1->create_buffer(mgr->provider1, size, desc);
+   if(buf)
+      return buf;
+   
+   buf = mgr->provider2->create_buffer(mgr->provider2, size, desc);
+   return buf;
+}
+
+
+static void
+pb_alt_manager_flush(struct pb_manager *_mgr)
+{
+   struct pb_alt_manager *mgr = pb_alt_manager(_mgr);
+   
+   assert(mgr->provider1->flush);
+   if(mgr->provider1->flush)
+      mgr->provider1->flush(mgr->provider1);
+   
+   assert(mgr->provider2->flush);
+   if(mgr->provider2->flush)
+      mgr->provider2->flush(mgr->provider2);
+}
+
+
+static void
+pb_alt_manager_destroy(struct pb_manager *mgr)
+{
+   FREE(mgr);
+}
+
+
+struct pb_manager *
+pb_alt_manager_create(struct pb_manager *provider1, 
+                      struct pb_manager *provider2)
+{
+   struct pb_alt_manager *mgr;
+
+   if(!provider1 || !provider2)
+      return NULL;
+   
+   mgr = CALLOC_STRUCT(pb_alt_manager);
+   if (!mgr)
+      return NULL;
+
+   mgr->base.destroy = pb_alt_manager_destroy;
+   mgr->base.create_buffer = pb_alt_manager_create_buffer;
+   mgr->base.flush = pb_alt_manager_flush;
+   mgr->provider1 = provider1;
+   mgr->provider2 = provider2;
+      
+   return &mgr->base;
+}
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_cache.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_cache.c
new file mode 100644
index 00000000000..a1688537134
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_cache.c
@@ -0,0 +1,384 @@
+/**************************************************************************
+ *
+ * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Buffer cache.
+ * 
+ * \author Jose Fonseca <jrfonseca-at-tungstengraphics-dot-com>
+ * \author Thomas Hellström <thomas-at-tungstengraphics-dot-com>
+ */
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_thread.h"
+#include "util/u_memory.h"
+#include "util/u_double_list.h"
+#include "util/u_time.h"
+
+#include "pb_buffer.h"
+#include "pb_bufmgr.h"
+
+
+/**
+ * Convenience macro (type safe).
+ */
+#define SUPER(__derived) (&(__derived)->base)
+
+
+struct pb_cache_manager;
+
+
+/**
+ * Wrapper around a pipe buffer which adds delayed destruction.
+ */
+struct pb_cache_buffer
+{
+   struct pb_buffer base;
+   
+   struct pb_buffer *buffer;
+   struct pb_cache_manager *mgr;
+
+   /** Caching time interval */
+   struct util_time start, end;
+
+   struct list_head head;
+};
+
+
+struct pb_cache_manager
+{
+   struct pb_manager base;
+
+   struct pb_manager *provider;
+   unsigned usecs;
+   
+   pipe_mutex mutex;
+   
+   struct list_head delayed;
+   size_t numDelayed;
+};
+
+
+static INLINE struct pb_cache_buffer *
+pb_cache_buffer(struct pb_buffer *buf)
+{
+   assert(buf);
+   return (struct pb_cache_buffer *)buf;
+}
+
+
+static INLINE struct pb_cache_manager *
+pb_cache_manager(struct pb_manager *mgr)
+{
+   assert(mgr);
+   return (struct pb_cache_manager *)mgr;
+}
+
+
+/**
+ * Actually destroy the buffer.
+ */
+static INLINE void
+_pb_cache_buffer_destroy(struct pb_cache_buffer *buf)
+{
+   struct pb_cache_manager *mgr = buf->mgr;
+
+   LIST_DEL(&buf->head);
+   assert(mgr->numDelayed);
+   --mgr->numDelayed;
+   assert(!buf->base.base.refcount);
+   pb_reference(&buf->buffer, NULL);
+   FREE(buf);
+}
+
+
+/**
+ * Free as many cache buffers from the list head as possible. 
+ */
+static void
+_pb_cache_buffer_list_check_free(struct pb_cache_manager *mgr)
+{
+   struct list_head *curr, *next;
+   struct pb_cache_buffer *buf;
+   struct util_time now;
+   
+   util_time_get(&now);
+   
+   curr = mgr->delayed.next;
+   next = curr->next;
+   while(curr != &mgr->delayed) {
+      buf = LIST_ENTRY(struct pb_cache_buffer, curr, head);
+
+      if(!util_time_timeout(&buf->start, &buf->end, &now))
+	 break;
+	 
+      _pb_cache_buffer_destroy(buf);
+
+      curr = next; 
+      next = curr->next;
+   }
+}
+
+
+static void
+pb_cache_buffer_destroy(struct pb_buffer *_buf)
+{
+   struct pb_cache_buffer *buf = pb_cache_buffer(_buf);   
+   struct pb_cache_manager *mgr = buf->mgr;
+
+   pipe_mutex_lock(mgr->mutex);
+   assert(buf->base.base.refcount == 0);
+   
+   _pb_cache_buffer_list_check_free(mgr);
+   
+   util_time_get(&buf->start);
+   util_time_add(&buf->start, mgr->usecs, &buf->end);
+   LIST_ADDTAIL(&buf->head, &mgr->delayed);
+   ++mgr->numDelayed;
+   pipe_mutex_unlock(mgr->mutex);
+}
+
+
+static void *
+pb_cache_buffer_map(struct pb_buffer *_buf, 
+                  unsigned flags)
+{
+   struct pb_cache_buffer *buf = pb_cache_buffer(_buf);   
+   return pb_map(buf->buffer, flags);
+}
+
+
+static void
+pb_cache_buffer_unmap(struct pb_buffer *_buf)
+{
+   struct pb_cache_buffer *buf = pb_cache_buffer(_buf);   
+   pb_unmap(buf->buffer);
+}
+
+
+static enum pipe_error 
+pb_cache_buffer_validate(struct pb_buffer *_buf, 
+                         struct pb_validate *vl,
+                         unsigned flags)
+{
+   struct pb_cache_buffer *buf = pb_cache_buffer(_buf);
+   return pb_validate(buf->buffer, vl, flags);
+}
+
+
+static void
+pb_cache_buffer_fence(struct pb_buffer *_buf, 
+                      struct pipe_fence_handle *fence)
+{
+   struct pb_cache_buffer *buf = pb_cache_buffer(_buf);
+   pb_fence(buf->buffer, fence);
+}
+
+
+static void
+pb_cache_buffer_get_base_buffer(struct pb_buffer *_buf,
+                              struct pb_buffer **base_buf,
+                              unsigned *offset)
+{
+   struct pb_cache_buffer *buf = pb_cache_buffer(_buf);
+   pb_get_base_buffer(buf->buffer, base_buf, offset);
+}
+
+
+const struct pb_vtbl 
+pb_cache_buffer_vtbl = {
+      pb_cache_buffer_destroy,
+      pb_cache_buffer_map,
+      pb_cache_buffer_unmap,
+      pb_cache_buffer_validate,
+      pb_cache_buffer_fence,
+      pb_cache_buffer_get_base_buffer
+};
+
+
+static INLINE boolean
+pb_cache_is_buffer_compat(struct pb_cache_buffer *buf,  
+                          size_t size,
+                          const struct pb_desc *desc)
+{
+   if(buf->base.base.size < size)
+      return FALSE;
+
+   /* be lenient with size */
+   if(buf->base.base.size >= 2*size)
+      return FALSE;
+   
+   if(!pb_check_alignment(desc->alignment, buf->base.base.alignment))
+      return FALSE;
+   
+   if(!pb_check_usage(desc->usage, buf->base.base.usage))
+      return FALSE;
+   
+   return TRUE;
+}
+
+
+static struct pb_buffer *
+pb_cache_manager_create_buffer(struct pb_manager *_mgr, 
+                               size_t size,
+                               const struct pb_desc *desc)
+{
+   struct pb_cache_manager *mgr = pb_cache_manager(_mgr);
+   struct pb_cache_buffer *buf;
+   struct pb_cache_buffer *curr_buf;
+   struct list_head *curr, *next;
+   struct util_time now;
+   
+   pipe_mutex_lock(mgr->mutex);
+
+   buf = NULL;
+   curr = mgr->delayed.next;
+   next = curr->next;
+   
+   /* search in the expired buffers, freeing them in the process */
+   util_time_get(&now);
+   while(curr != &mgr->delayed) {
+      curr_buf = LIST_ENTRY(struct pb_cache_buffer, curr, head);
+      if(!buf && pb_cache_is_buffer_compat(curr_buf, size, desc))
+	 buf = curr_buf;
+      else if(util_time_timeout(&curr_buf->start, &curr_buf->end, &now))
+	 _pb_cache_buffer_destroy(curr_buf);
+      else
+         /* This buffer (and all hereafter) are still hot in cache */
+         break;
+      curr = next; 
+      next = curr->next;
+   }
+
+   /* keep searching in the hot buffers */
+   if(!buf) {
+      while(curr != &mgr->delayed) {
+         curr_buf = LIST_ENTRY(struct pb_cache_buffer, curr, head);
+         if(pb_cache_is_buffer_compat(curr_buf, size, desc)) {
+            buf = curr_buf;
+            break;
+         }
+         /* no need to check the timeout here */
+         curr = next;
+         next = curr->next;
+      }
+   }
+   
+   if(buf) {
+      LIST_DEL(&buf->head);
+      pipe_mutex_unlock(mgr->mutex);
+      ++buf->base.base.refcount;
+      return &buf->base;
+   }
+   
+   pipe_mutex_unlock(mgr->mutex);
+
+   buf = CALLOC_STRUCT(pb_cache_buffer);
+   if(!buf)
+      return NULL;
+   
+   buf->buffer = mgr->provider->create_buffer(mgr->provider, size, desc);
+   if(!buf->buffer) {
+      FREE(buf);
+      return NULL;
+   }
+   
+   assert(buf->buffer->base.refcount >= 1);
+   assert(pb_check_alignment(desc->alignment, buf->buffer->base.alignment));
+   assert(pb_check_usage(desc->usage, buf->buffer->base.usage));
+   assert(buf->buffer->base.size >= size);
+   
+   buf->base.base.refcount = 1;
+   buf->base.base.alignment = buf->buffer->base.alignment;
+   buf->base.base.usage = buf->buffer->base.usage;
+   buf->base.base.size = buf->buffer->base.size;
+   
+   buf->base.vtbl = &pb_cache_buffer_vtbl;
+   buf->mgr = mgr;
+   
+   return &buf->base;
+}
+
+
+static void
+pb_cache_manager_flush(struct pb_manager *_mgr)
+{
+   struct pb_cache_manager *mgr = pb_cache_manager(_mgr);
+   struct list_head *curr, *next;
+   struct pb_cache_buffer *buf;
+
+   pipe_mutex_lock(mgr->mutex);
+   curr = mgr->delayed.next;
+   next = curr->next;
+   while(curr != &mgr->delayed) {
+      buf = LIST_ENTRY(struct pb_cache_buffer, curr, head);
+      _pb_cache_buffer_destroy(buf);
+      curr = next; 
+      next = curr->next;
+   }
+   pipe_mutex_unlock(mgr->mutex);
+   
+   assert(mgr->provider->flush);
+   if(mgr->provider->flush)
+      mgr->provider->flush(mgr->provider);
+}
+
+
+static void
+pb_cache_manager_destroy(struct pb_manager *mgr)
+{
+   pb_cache_manager_flush(mgr);
+   FREE(mgr);
+}
+
+
+struct pb_manager *
+pb_cache_manager_create(struct pb_manager *provider, 
+                     	unsigned usecs) 
+{
+   struct pb_cache_manager *mgr;
+
+   if(!provider)
+      return NULL;
+   
+   mgr = CALLOC_STRUCT(pb_cache_manager);
+   if (!mgr)
+      return NULL;
+
+   mgr->base.destroy = pb_cache_manager_destroy;
+   mgr->base.create_buffer = pb_cache_manager_create_buffer;
+   mgr->base.flush = pb_cache_manager_flush;
+   mgr->provider = provider;
+   mgr->usecs = usecs;
+   LIST_INITHEAD(&mgr->delayed);
+   mgr->numDelayed = 0;
+   pipe_mutex_init(mgr->mutex);
+      
+   return &mgr->base;
+}
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_debug.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_debug.c
new file mode 100644
index 00000000000..26d9c24aec2
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_debug.c
@@ -0,0 +1,390 @@
+/**************************************************************************
+ *
+ * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Debug buffer manager to detect buffer under- and overflows.
+ * 
+ * \author Jose Fonseca <[email protected]>
+ */
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_thread.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "util/u_double_list.h"
+#include "util/u_time.h"
+
+#include "pb_buffer.h"
+#include "pb_bufmgr.h"
+
+
+#ifdef DEBUG
+
+
+/**
+ * Convenience macro (type safe).
+ */
+#define SUPER(__derived) (&(__derived)->base)
+
+
+struct pb_debug_manager;
+
+
+/**
+ * Wrapper around a pipe buffer which adds delayed destruction.
+ */
+struct pb_debug_buffer
+{
+   struct pb_buffer base;
+   
+   struct pb_buffer *buffer;
+   struct pb_debug_manager *mgr;
+   
+   size_t underflow_size;
+   size_t overflow_size;
+};
+
+
+struct pb_debug_manager
+{
+   struct pb_manager base;
+
+   struct pb_manager *provider;
+
+   size_t band_size;
+};
+
+
+static INLINE struct pb_debug_buffer *
+pb_debug_buffer(struct pb_buffer *buf)
+{
+   assert(buf);
+   return (struct pb_debug_buffer *)buf;
+}
+
+
+static INLINE struct pb_debug_manager *
+pb_debug_manager(struct pb_manager *mgr)
+{
+   assert(mgr);
+   return (struct pb_debug_manager *)mgr;
+}
+
+
+static const uint8_t random_pattern[32] = {
+   0xaf, 0xcf, 0xa5, 0xa2, 0xc2, 0x63, 0x15, 0x1a, 
+   0x7e, 0xe2, 0x7e, 0x84, 0x15, 0x49, 0xa2, 0x1e,
+   0x49, 0x63, 0xf5, 0x52, 0x74, 0x66, 0x9e, 0xc4, 
+   0x6d, 0xcf, 0x2c, 0x4a, 0x74, 0xe6, 0xfd, 0x94
+};
+
+
+static INLINE void 
+fill_random_pattern(uint8_t *dst, size_t size)
+{
+   size_t i = 0;
+   while(size--) {
+      *dst++ = random_pattern[i++];
+      i &= sizeof(random_pattern) - 1;
+   }
+}
+
+
+static INLINE boolean 
+check_random_pattern(const uint8_t *dst, size_t size, 
+                     size_t *min_ofs, size_t *max_ofs) 
+{
+   boolean result = TRUE;
+   size_t i;
+   *min_ofs = size;
+   *max_ofs = 0;
+   for(i = 0; i < size; ++i) {
+      if(*dst++ != random_pattern[i % sizeof(random_pattern)]) {
+         *min_ofs = MIN2(*min_ofs, i);
+         *max_ofs = MAX2(*max_ofs, i);
+	 result = FALSE;
+      }
+   }
+   return result;
+}
+
+
+static void
+pb_debug_buffer_fill(struct pb_debug_buffer *buf)
+{
+   uint8_t *map;
+   
+   map = pb_map(buf->buffer, PIPE_BUFFER_USAGE_CPU_WRITE);
+   assert(map);
+   if(map) {
+      fill_random_pattern(map, buf->underflow_size);
+      fill_random_pattern(map + buf->underflow_size + buf->base.base.size, 
+                          buf->overflow_size);
+      pb_unmap(buf->buffer);
+   }
+}
+
+
+/**
+ * Check for under/over flows.
+ * 
+ * Should be called with the buffer unmaped.
+ */
+static void
+pb_debug_buffer_check(struct pb_debug_buffer *buf)
+{
+   uint8_t *map;
+   
+   map = pb_map(buf->buffer, PIPE_BUFFER_USAGE_CPU_READ);
+   assert(map);
+   if(map) {
+      boolean underflow, overflow;
+      size_t min_ofs, max_ofs;
+      
+      underflow = !check_random_pattern(map, buf->underflow_size, 
+                                        &min_ofs, &max_ofs);
+      if(underflow) {
+         debug_printf("buffer underflow (offset -%u%s to -%u bytes) detected\n",
+                      buf->underflow_size - min_ofs,
+                      min_ofs == 0 ? "+" : "",
+                      buf->underflow_size - max_ofs);
+      }
+      
+      overflow = !check_random_pattern(map + buf->underflow_size + buf->base.base.size, 
+                                       buf->overflow_size, 
+                                       &min_ofs, &max_ofs);
+      if(overflow) {
+         debug_printf("buffer overflow (size %u plus offset %u to %u%s bytes) detected\n",
+                      buf->base.base.size,
+                      min_ofs,
+                      max_ofs,
+                      max_ofs == buf->overflow_size - 1 ? "+" : "");
+      }
+      
+      debug_assert(!underflow && !overflow);
+
+      /* re-fill if not aborted */
+      if(underflow)
+         fill_random_pattern(map, buf->underflow_size);
+      if(overflow)
+         fill_random_pattern(map + buf->underflow_size + buf->base.base.size, 
+                             buf->overflow_size);
+
+      pb_unmap(buf->buffer);
+   }
+}
+
+
+static void
+pb_debug_buffer_destroy(struct pb_buffer *_buf)
+{
+   struct pb_debug_buffer *buf = pb_debug_buffer(_buf);  
+   
+   assert(!buf->base.base.refcount);
+   
+   pb_debug_buffer_check(buf);
+
+   pb_reference(&buf->buffer, NULL);
+   FREE(buf);
+}
+
+
+static void *
+pb_debug_buffer_map(struct pb_buffer *_buf, 
+                    unsigned flags)
+{
+   struct pb_debug_buffer *buf = pb_debug_buffer(_buf);
+   void *map;
+   
+   pb_debug_buffer_check(buf);
+
+   map = pb_map(buf->buffer, flags);
+   if(!map)
+      return NULL;
+   
+   return (uint8_t *)map + buf->underflow_size;
+}
+
+
+static void
+pb_debug_buffer_unmap(struct pb_buffer *_buf)
+{
+   struct pb_debug_buffer *buf = pb_debug_buffer(_buf);   
+   pb_unmap(buf->buffer);
+   
+   pb_debug_buffer_check(buf);
+}
+
+
+static void
+pb_debug_buffer_get_base_buffer(struct pb_buffer *_buf,
+                                struct pb_buffer **base_buf,
+                                unsigned *offset)
+{
+   struct pb_debug_buffer *buf = pb_debug_buffer(_buf);
+   pb_get_base_buffer(buf->buffer, base_buf, offset);
+   *offset += buf->underflow_size;
+}
+
+
+static enum pipe_error 
+pb_debug_buffer_validate(struct pb_buffer *_buf, 
+                         struct pb_validate *vl,
+                         unsigned flags)
+{
+   struct pb_debug_buffer *buf = pb_debug_buffer(_buf);
+   
+   pb_debug_buffer_check(buf);
+
+   return pb_validate(buf->buffer, vl, flags);
+}
+
+
+static void
+pb_debug_buffer_fence(struct pb_buffer *_buf, 
+                      struct pipe_fence_handle *fence)
+{
+   struct pb_debug_buffer *buf = pb_debug_buffer(_buf);
+   pb_fence(buf->buffer, fence);
+}
+
+
+const struct pb_vtbl 
+pb_debug_buffer_vtbl = {
+      pb_debug_buffer_destroy,
+      pb_debug_buffer_map,
+      pb_debug_buffer_unmap,
+      pb_debug_buffer_validate,
+      pb_debug_buffer_fence,
+      pb_debug_buffer_get_base_buffer
+};
+
+
+static struct pb_buffer *
+pb_debug_manager_create_buffer(struct pb_manager *_mgr, 
+                               size_t size,
+                               const struct pb_desc *desc)
+{
+   struct pb_debug_manager *mgr = pb_debug_manager(_mgr);
+   struct pb_debug_buffer *buf;
+   struct pb_desc real_desc;
+   size_t real_size;
+   
+   buf = CALLOC_STRUCT(pb_debug_buffer);
+   if(!buf)
+      return NULL;
+   
+   real_size = size + 2*mgr->band_size;
+   real_desc = *desc;
+   real_desc.usage |= PIPE_BUFFER_USAGE_CPU_WRITE;
+   real_desc.usage |= PIPE_BUFFER_USAGE_CPU_READ;
+
+   buf->buffer = mgr->provider->create_buffer(mgr->provider, 
+                                              real_size, 
+                                              &real_desc);
+   if(!buf->buffer) {
+      FREE(buf);
+      return NULL;
+   }
+   
+   assert(buf->buffer->base.refcount >= 1);
+   assert(pb_check_alignment(real_desc.alignment, buf->buffer->base.alignment));
+   assert(pb_check_usage(real_desc.usage, buf->buffer->base.usage));
+   assert(buf->buffer->base.size >= real_size);
+   
+   buf->base.base.refcount = 1;
+   buf->base.base.alignment = desc->alignment;
+   buf->base.base.usage = desc->usage;
+   buf->base.base.size = size;
+   
+   buf->base.vtbl = &pb_debug_buffer_vtbl;
+   buf->mgr = mgr;
+
+   buf->underflow_size = mgr->band_size;
+   buf->overflow_size = buf->buffer->base.size - buf->underflow_size - size;
+   
+   pb_debug_buffer_fill(buf);
+   
+   return &buf->base;
+}
+
+
+static void
+pb_debug_manager_flush(struct pb_manager *_mgr)
+{
+   struct pb_debug_manager *mgr = pb_debug_manager(_mgr);
+   assert(mgr->provider->flush);
+   if(mgr->provider->flush)
+      mgr->provider->flush(mgr->provider);
+}
+
+
+static void
+pb_debug_manager_destroy(struct pb_manager *_mgr)
+{
+   struct pb_debug_manager *mgr = pb_debug_manager(_mgr);
+   mgr->provider->destroy(mgr->provider);
+   FREE(mgr);
+}
+
+
+struct pb_manager *
+pb_debug_manager_create(struct pb_manager *provider, size_t band_size) 
+{
+   struct pb_debug_manager *mgr;
+
+   if(!provider)
+      return NULL;
+   
+   mgr = CALLOC_STRUCT(pb_debug_manager);
+   if (!mgr)
+      return NULL;
+
+   mgr->base.destroy = pb_debug_manager_destroy;
+   mgr->base.create_buffer = pb_debug_manager_create_buffer;
+   mgr->base.flush = pb_debug_manager_flush;
+   mgr->provider = provider;
+   mgr->band_size = band_size;
+      
+   return &mgr->base;
+}
+
+
+#else /* !DEBUG */
+
+
+struct pb_manager *
+pb_debug_manager_create(struct pb_manager *provider, size_t band_size) 
+{
+   return provider;
+}
+
+
+#endif /* !DEBUG */
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_fenced.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_fenced.c
new file mode 100644
index 00000000000..47e9fee5338
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_fenced.c
@@ -0,0 +1,148 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+
+/**
+ * \file
+ * A buffer manager that wraps buffers in fenced buffers.
+ * 
+ * \author Jose Fonseca <[email protected]>
+ */
+
+
+#include "pipe/p_debug.h"
+#include "util/u_memory.h"
+
+#include "pb_buffer.h"
+#include "pb_buffer_fenced.h"
+#include "pb_bufmgr.h"
+
+
+struct fenced_pb_manager
+{
+   struct pb_manager base;
+
+   struct pb_manager *provider;
+   
+   struct fenced_buffer_list *fenced_list;
+};
+
+
+static INLINE struct fenced_pb_manager *
+fenced_pb_manager(struct pb_manager *mgr)
+{
+   assert(mgr);
+   return (struct fenced_pb_manager *)mgr;
+}
+
+
+static struct pb_buffer *
+fenced_bufmgr_create_buffer(struct pb_manager *mgr, 
+                            size_t size,
+                            const struct pb_desc *desc)
+{
+   struct fenced_pb_manager *fenced_mgr = fenced_pb_manager(mgr);
+   struct pb_buffer *buf;
+   struct pb_buffer *fenced_buf;
+
+   /* check for free buffers before allocating new ones */
+   fenced_buffer_list_check_free(fenced_mgr->fenced_list, 0);
+   
+   buf = fenced_mgr->provider->create_buffer(fenced_mgr->provider, size, desc);
+   if(!buf) {
+      /* try harder to get a buffer */
+      fenced_buffer_list_check_free(fenced_mgr->fenced_list, 1);
+      
+      buf = fenced_mgr->provider->create_buffer(fenced_mgr->provider, size, desc);
+      if(!buf) {
+         /* give up */
+         return NULL;
+      }
+   }
+   
+   fenced_buf = fenced_buffer_create(fenced_mgr->fenced_list, buf);
+   if(!fenced_buf) {
+      pb_reference(&buf, NULL);
+   }
+   
+   return fenced_buf;
+}
+
+
+static void
+fenced_bufmgr_flush(struct pb_manager *mgr)
+{
+   struct fenced_pb_manager *fenced_mgr = fenced_pb_manager(mgr);
+
+   fenced_buffer_list_check_free(fenced_mgr->fenced_list, TRUE);
+
+   assert(fenced_mgr->provider->flush);
+   if(fenced_mgr->provider->flush)
+      fenced_mgr->provider->flush(fenced_mgr->provider);
+}
+
+
+static void
+fenced_bufmgr_destroy(struct pb_manager *mgr)
+{
+   struct fenced_pb_manager *fenced_mgr = fenced_pb_manager(mgr);
+
+   fenced_buffer_list_destroy(fenced_mgr->fenced_list);
+
+   if(fenced_mgr->provider)
+      fenced_mgr->provider->destroy(fenced_mgr->provider);
+   
+   FREE(fenced_mgr);
+}
+
+
+struct pb_manager *
+fenced_bufmgr_create(struct pb_manager *provider, 
+                     struct pb_fence_ops *ops) 
+{
+   struct fenced_pb_manager *fenced_mgr;
+
+   if(!provider)
+      return NULL;
+   
+   fenced_mgr = CALLOC_STRUCT(fenced_pb_manager);
+   if (!fenced_mgr)
+      return NULL;
+
+   fenced_mgr->base.destroy = fenced_bufmgr_destroy;
+   fenced_mgr->base.create_buffer = fenced_bufmgr_create_buffer;
+   fenced_mgr->base.flush = fenced_bufmgr_flush;
+
+   fenced_mgr->provider = provider;
+   fenced_mgr->fenced_list = fenced_buffer_list_create(ops);
+   if(!fenced_mgr->fenced_list) {
+      FREE(fenced_mgr);
+      return NULL;
+   }
+      
+   return &fenced_mgr->base;
+}
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_mm.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_mm.c
new file mode 100644
index 00000000000..2f5a5d8ea07
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_mm.c
@@ -0,0 +1,322 @@
+/**************************************************************************
+ *
+ * Copyright 2006 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file
+ * Buffer manager using the old texture memory manager.
+ * 
+ * \author Jose Fonseca <[email protected]>
+ */
+
+
+#include "pipe/p_defines.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_thread.h"
+#include "util/u_memory.h"
+#include "util/u_double_list.h"
+#include "util/u_mm.h"
+#include "pb_buffer.h"
+#include "pb_bufmgr.h"
+
+
+/**
+ * Convenience macro (type safe).
+ */
+#define SUPER(__derived) (&(__derived)->base)
+
+
+struct mm_pb_manager
+{
+   struct pb_manager base;
+   
+   pipe_mutex mutex;
+   
+   size_t size;
+   struct mem_block *heap;
+   
+   size_t align2;
+   
+   struct pb_buffer *buffer;
+   void *map;
+};
+
+
+static INLINE struct mm_pb_manager *
+mm_pb_manager(struct pb_manager *mgr)
+{
+   assert(mgr);
+   return (struct mm_pb_manager *)mgr;
+}
+
+
+struct mm_buffer
+{
+   struct pb_buffer base;
+   
+   struct mm_pb_manager *mgr;
+   
+   struct mem_block *block;
+};
+
+
+static INLINE struct mm_buffer *
+mm_buffer(struct pb_buffer *buf)
+{
+   assert(buf);
+   return (struct mm_buffer *)buf;
+}
+
+
+static void
+mm_buffer_destroy(struct pb_buffer *buf)
+{
+   struct mm_buffer *mm_buf = mm_buffer(buf);
+   struct mm_pb_manager *mm = mm_buf->mgr;
+   
+   assert(buf->base.refcount == 0);
+   
+   pipe_mutex_lock(mm->mutex);
+   u_mmFreeMem(mm_buf->block);
+   FREE(buf);
+   pipe_mutex_unlock(mm->mutex);
+}
+
+
+static void *
+mm_buffer_map(struct pb_buffer *buf,
+              unsigned flags)
+{
+   struct mm_buffer *mm_buf = mm_buffer(buf);
+   struct mm_pb_manager *mm = mm_buf->mgr;
+
+   return (unsigned char *) mm->map + mm_buf->block->ofs;
+}
+
+
+static void
+mm_buffer_unmap(struct pb_buffer *buf)
+{
+   /* No-op */
+}
+
+
+static enum pipe_error 
+mm_buffer_validate(struct pb_buffer *buf, 
+                   struct pb_validate *vl,
+                   unsigned flags)
+{
+   struct mm_buffer *mm_buf = mm_buffer(buf);
+   struct mm_pb_manager *mm = mm_buf->mgr;
+   return pb_validate(mm->buffer, vl, flags);
+}
+
+
+static void
+mm_buffer_fence(struct pb_buffer *buf, 
+                struct pipe_fence_handle *fence)
+{
+   struct mm_buffer *mm_buf = mm_buffer(buf);
+   struct mm_pb_manager *mm = mm_buf->mgr;
+   pb_fence(mm->buffer, fence);
+}
+
+
+static void
+mm_buffer_get_base_buffer(struct pb_buffer *buf,
+                          struct pb_buffer **base_buf,
+                          unsigned *offset)
+{
+   struct mm_buffer *mm_buf = mm_buffer(buf);
+   struct mm_pb_manager *mm = mm_buf->mgr;
+   pb_get_base_buffer(mm->buffer, base_buf, offset);
+   *offset += mm_buf->block->ofs;
+}
+
+
+static const struct pb_vtbl 
+mm_buffer_vtbl = {
+      mm_buffer_destroy,
+      mm_buffer_map,
+      mm_buffer_unmap,
+      mm_buffer_validate,
+      mm_buffer_fence,
+      mm_buffer_get_base_buffer
+};
+
+
+static struct pb_buffer *
+mm_bufmgr_create_buffer(struct pb_manager *mgr, 
+                        size_t size,
+                        const struct pb_desc *desc)
+{
+   struct mm_pb_manager *mm = mm_pb_manager(mgr);
+   struct mm_buffer *mm_buf;
+
+   /* We don't handle alignments larger then the one initially setup */
+   assert(pb_check_alignment(desc->alignment, 1 << mm->align2));
+   if(!pb_check_alignment(desc->alignment, 1 << mm->align2))
+      return NULL;
+   
+   pipe_mutex_lock(mm->mutex);
+
+   mm_buf = CALLOC_STRUCT(mm_buffer);
+   if (!mm_buf) {
+      pipe_mutex_unlock(mm->mutex);
+      return NULL;
+   }
+
+   mm_buf->base.base.refcount = 1;
+   mm_buf->base.base.alignment = desc->alignment;
+   mm_buf->base.base.usage = desc->usage;
+   mm_buf->base.base.size = size;
+   
+   mm_buf->base.vtbl = &mm_buffer_vtbl;
+   
+   mm_buf->mgr = mm;
+   
+   mm_buf->block = u_mmAllocMem(mm->heap, size, mm->align2, 0);
+   if(!mm_buf->block) {
+      debug_printf("warning: heap full\n");
+#if 0
+      mmDumpMemInfo(mm->heap);
+#endif
+      
+      mm_buf->block = u_mmAllocMem(mm->heap, size, mm->align2, 0);
+      if(!mm_buf->block) {
+         FREE(mm_buf);
+         pipe_mutex_unlock(mm->mutex);
+         return NULL;
+      }
+   }
+   
+   /* Some sanity checks */
+   assert(0 <= (unsigned)mm_buf->block->ofs && (unsigned)mm_buf->block->ofs < mm->size);
+   assert(size <= (unsigned)mm_buf->block->size && (unsigned)mm_buf->block->ofs + (unsigned)mm_buf->block->size <= mm->size);
+   
+   pipe_mutex_unlock(mm->mutex);
+   return SUPER(mm_buf);
+}
+
+
+static void
+mm_bufmgr_flush(struct pb_manager *mgr)
+{
+   /* No-op */
+}
+
+
+static void
+mm_bufmgr_destroy(struct pb_manager *mgr)
+{
+   struct mm_pb_manager *mm = mm_pb_manager(mgr);
+   
+   pipe_mutex_lock(mm->mutex);
+
+   u_mmDestroy(mm->heap);
+   
+   pb_unmap(mm->buffer);
+   pb_reference(&mm->buffer, NULL);
+   
+   pipe_mutex_unlock(mm->mutex);
+   
+   FREE(mgr);
+}
+
+
+struct pb_manager *
+mm_bufmgr_create_from_buffer(struct pb_buffer *buffer, 
+                             size_t size, size_t align2) 
+{
+   struct mm_pb_manager *mm;
+
+   if(!buffer)
+      return NULL;
+   
+   mm = CALLOC_STRUCT(mm_pb_manager);
+   if (!mm)
+      return NULL;
+
+   mm->base.destroy = mm_bufmgr_destroy;
+   mm->base.create_buffer = mm_bufmgr_create_buffer;
+   mm->base.flush = mm_bufmgr_flush;
+
+   mm->size = size;
+   mm->align2 = align2; /* 64-byte alignment */
+
+   pipe_mutex_init(mm->mutex);
+
+   mm->buffer = buffer; 
+
+   mm->map = pb_map(mm->buffer, 
+		    PIPE_BUFFER_USAGE_CPU_READ |
+		    PIPE_BUFFER_USAGE_CPU_WRITE);
+   if(!mm->map)
+      goto failure;
+
+   mm->heap = u_mmInit(0, size); 
+   if (!mm->heap)
+      goto failure;
+
+   return SUPER(mm);
+   
+failure:
+if(mm->heap)
+   u_mmDestroy(mm->heap);
+   if(mm->map)
+      pb_unmap(mm->buffer);
+   if(mm)
+      FREE(mm);
+   return NULL;
+}
+
+
+struct pb_manager *
+mm_bufmgr_create(struct pb_manager *provider, 
+                 size_t size, size_t align2) 
+{
+   struct pb_buffer *buffer;
+   struct pb_manager *mgr;
+   struct pb_desc desc;
+
+   if(!provider)
+      return NULL;
+   
+   memset(&desc, 0, sizeof(desc));
+   desc.alignment = 1 << align2;
+   
+   buffer = provider->create_buffer(provider, size, &desc); 
+   if (!buffer)
+      return NULL;
+   
+   mgr = mm_bufmgr_create_from_buffer(buffer, size, align2);
+   if (!mgr) {
+      pb_reference(&buffer, NULL);
+      return NULL;
+   }
+
+  return mgr;
+}
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_ondemand.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_ondemand.c
new file mode 100644
index 00000000000..ba02a84e62d
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_ondemand.c
@@ -0,0 +1,303 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * A variation of malloc buffers which get transferred to real graphics memory
+ * when there is an attempt to validate them. 
+ * 
+ * @author Jose Fonseca <[email protected]>
+ */
+
+
+#include "pipe/p_debug.h"
+#include "util/u_memory.h"
+#include "pb_buffer.h"
+#include "pb_bufmgr.h"
+
+
+struct pb_ondemand_manager;
+
+
+struct pb_ondemand_buffer 
+{
+   struct pb_buffer base;
+   
+   struct pb_ondemand_manager *mgr;
+   
+   /** Regular malloc'ed memory */
+   void *data;
+   unsigned mapcount;
+   
+   /** Real buffer */
+   struct pb_buffer *buffer;
+   size_t size;
+   struct pb_desc desc;
+};
+
+
+struct pb_ondemand_manager
+{
+   struct pb_manager base;
+   
+   struct pb_manager *provider;
+};
+
+
+extern const struct pb_vtbl pb_ondemand_buffer_vtbl;
+
+static INLINE struct pb_ondemand_buffer *
+pb_ondemand_buffer(struct pb_buffer *buf)
+{
+   assert(buf);
+   assert(buf->vtbl == &pb_ondemand_buffer_vtbl);
+   return (struct pb_ondemand_buffer *)buf;
+}
+
+static INLINE struct pb_ondemand_manager *
+pb_ondemand_manager(struct pb_manager *mgr)
+{
+   assert(mgr);
+   return (struct pb_ondemand_manager *)mgr;
+}
+
+
+static void
+pb_ondemand_buffer_destroy(struct pb_buffer *_buf)
+{
+   struct pb_ondemand_buffer *buf = pb_ondemand_buffer(_buf);
+   
+   pb_reference(&buf->buffer, NULL);
+   
+   align_free(buf->data);
+   
+   FREE(buf);
+}
+
+
+static void *
+pb_ondemand_buffer_map(struct pb_buffer *_buf, 
+                       unsigned flags)
+{
+   struct pb_ondemand_buffer *buf = pb_ondemand_buffer(_buf);
+
+   if(buf->buffer) {
+      assert(!buf->data);
+      return pb_map(buf->buffer, flags);
+   }
+   else {
+      assert(buf->data);
+      ++buf->mapcount;
+      return buf->data;
+   }
+}
+
+
+static void
+pb_ondemand_buffer_unmap(struct pb_buffer *_buf)
+{
+   struct pb_ondemand_buffer *buf = pb_ondemand_buffer(_buf);
+
+   if(buf->buffer) {
+      assert(!buf->data);
+      pb_unmap(buf->buffer);
+   }
+   else {
+      assert(buf->data);
+      assert(buf->mapcount);
+      if(buf->mapcount)
+         --buf->mapcount;
+   }
+}
+
+
+static enum pipe_error 
+pb_ondemand_buffer_instantiate(struct pb_ondemand_buffer *buf)
+{
+   if(!buf->buffer) {
+      struct pb_manager *provider = buf->mgr->provider;
+      uint8_t *map;
+      
+      assert(!buf->mapcount);
+      
+      buf->buffer = provider->create_buffer(provider, buf->size, &buf->desc);
+      if(!buf->buffer)
+         return PIPE_ERROR_OUT_OF_MEMORY;
+      
+      map = pb_map(buf->buffer, PIPE_BUFFER_USAGE_CPU_READ);
+      if(!map) {
+         pb_reference(&buf->buffer, NULL);
+         return PIPE_ERROR;
+      }
+      
+      memcpy(map, buf->data, buf->size);
+      
+      pb_unmap(buf->buffer);
+      
+      if(!buf->mapcount) {
+         FREE(buf->data);
+         buf->data = NULL;
+      }
+   }
+   
+   return PIPE_OK;
+}
+
+static enum pipe_error 
+pb_ondemand_buffer_validate(struct pb_buffer *_buf, 
+                            struct pb_validate *vl,
+                            unsigned flags)
+{
+   struct pb_ondemand_buffer *buf = pb_ondemand_buffer(_buf);
+   enum pipe_error ret; 
+
+   assert(!buf->mapcount);
+   if(buf->mapcount)
+      return PIPE_ERROR;
+
+   ret = pb_ondemand_buffer_instantiate(buf);
+   if(ret != PIPE_OK)
+      return ret;
+   
+   return pb_validate(buf->buffer, vl, flags);
+}
+
+
+static void
+pb_ondemand_buffer_fence(struct pb_buffer *_buf, 
+                         struct pipe_fence_handle *fence)
+{
+   struct pb_ondemand_buffer *buf = pb_ondemand_buffer(_buf);
+   
+   assert(buf->buffer);
+   if(!buf->buffer)
+      return;
+   
+   pb_fence(buf->buffer, fence);
+}
+
+
+static void
+pb_ondemand_buffer_get_base_buffer(struct pb_buffer *_buf,
+                                   struct pb_buffer **base_buf,
+                                   unsigned *offset)
+{
+   struct pb_ondemand_buffer *buf = pb_ondemand_buffer(_buf);
+
+   if(pb_ondemand_buffer_instantiate(buf) != PIPE_OK) {
+      assert(0);
+      *base_buf = &buf->base;
+      *offset = 0;
+      return;
+   }
+
+   pb_get_base_buffer(buf->buffer, base_buf, offset);
+}
+
+
+const struct pb_vtbl 
+pb_ondemand_buffer_vtbl = {
+      pb_ondemand_buffer_destroy,
+      pb_ondemand_buffer_map,
+      pb_ondemand_buffer_unmap,
+      pb_ondemand_buffer_validate,
+      pb_ondemand_buffer_fence,
+      pb_ondemand_buffer_get_base_buffer
+};
+
+
+static struct pb_buffer *
+pb_ondemand_manager_create_buffer(struct pb_manager *_mgr, 
+                                  size_t size,
+                                  const struct pb_desc *desc) 
+{
+   struct pb_ondemand_manager *mgr = pb_ondemand_manager(_mgr);
+   struct pb_ondemand_buffer *buf;
+   
+   buf = CALLOC_STRUCT(pb_ondemand_buffer);
+   if(!buf)
+      return NULL;
+
+   buf->base.base.refcount = 1;
+   buf->base.base.alignment = desc->alignment;
+   buf->base.base.usage = desc->usage;
+   buf->base.base.size = size;
+   buf->base.vtbl = &pb_ondemand_buffer_vtbl;
+   
+   buf->mgr = mgr;
+   
+   buf->data = align_malloc(size, desc->alignment < sizeof(void*) ? sizeof(void*) : desc->alignment);
+   if(!buf->data) {
+      FREE(buf);
+      return NULL;
+   }
+   
+   buf->size = size;
+   buf->desc = *desc;
+
+   return &buf->base;
+}
+
+
+static void
+pb_ondemand_manager_flush(struct pb_manager *_mgr) 
+{
+   struct pb_ondemand_manager *mgr = pb_ondemand_manager(_mgr);
+   
+   mgr->provider->flush(mgr->provider);
+}
+
+
+static void
+pb_ondemand_manager_destroy(struct pb_manager *_mgr) 
+{
+   struct pb_ondemand_manager *mgr = pb_ondemand_manager(_mgr);
+
+   FREE(mgr);
+}
+
+
+struct pb_manager *
+pb_ondemand_manager_create(struct pb_manager *provider) 
+{
+   struct pb_ondemand_manager *mgr;
+
+   if(!provider)
+      return NULL;
+   
+   mgr = CALLOC_STRUCT(pb_ondemand_manager);
+   if(!mgr)
+      return NULL;
+   
+   mgr->base.destroy = pb_ondemand_manager_destroy;
+   mgr->base.create_buffer = pb_ondemand_manager_create_buffer;
+   mgr->base.flush = pb_ondemand_manager_flush;
+   
+   mgr->provider = provider;
+
+   return &mgr->base;
+}
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_pool.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_pool.c
new file mode 100644
index 00000000000..a6ff37653e9
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_pool.c
@@ -0,0 +1,321 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+
+/**
+ * \file
+ * Batch buffer pool management.
+ * 
+ * \author Jose Fonseca <jrfonseca-at-tungstengraphics-dot-com>
+ * \author Thomas Hellström <thomas-at-tungstengraphics-dot-com>
+ */
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_thread.h"
+#include "pipe/p_defines.h"
+#include "util/u_memory.h"
+#include "util/u_double_list.h"
+
+#include "pb_buffer.h"
+#include "pb_bufmgr.h"
+
+
+/**
+ * Convenience macro (type safe).
+ */
+#define SUPER(__derived) (&(__derived)->base)
+
+
+struct pool_pb_manager
+{
+   struct pb_manager base;
+   
+   pipe_mutex mutex;
+   
+   size_t bufSize;
+   size_t bufAlign;
+   
+   size_t numFree;
+   size_t numTot;
+   
+   struct list_head free;
+   
+   struct pb_buffer *buffer;
+   void *map;
+   
+   struct pool_buffer *bufs;
+};
+
+
+static INLINE struct pool_pb_manager *
+pool_pb_manager(struct pb_manager *mgr)
+{
+   assert(mgr);
+   return (struct pool_pb_manager *)mgr;
+}
+
+
+struct pool_buffer
+{
+   struct pb_buffer base;
+   
+   struct pool_pb_manager *mgr;
+   
+   struct list_head head;
+   
+   size_t start;
+};
+
+
+static INLINE struct pool_buffer *
+pool_buffer(struct pb_buffer *buf)
+{
+   assert(buf);
+   return (struct pool_buffer *)buf;
+}
+
+
+
+static void
+pool_buffer_destroy(struct pb_buffer *buf)
+{
+   struct pool_buffer *pool_buf = pool_buffer(buf);
+   struct pool_pb_manager *pool = pool_buf->mgr;
+   
+   assert(pool_buf->base.base.refcount == 0);
+
+   pipe_mutex_lock(pool->mutex);
+   LIST_ADD(&pool_buf->head, &pool->free);
+   pool->numFree++;
+   pipe_mutex_unlock(pool->mutex);
+}
+
+
+static void *
+pool_buffer_map(struct pb_buffer *buf, unsigned flags)
+{
+   struct pool_buffer *pool_buf = pool_buffer(buf);
+   struct pool_pb_manager *pool = pool_buf->mgr;
+   void *map;
+
+   pipe_mutex_lock(pool->mutex);
+   map = (unsigned char *) pool->map + pool_buf->start;
+   pipe_mutex_unlock(pool->mutex);
+   return map;
+}
+
+
+static void
+pool_buffer_unmap(struct pb_buffer *buf)
+{
+   /* No-op */
+}
+
+
+static enum pipe_error 
+pool_buffer_validate(struct pb_buffer *buf, 
+                     struct pb_validate *vl,
+                     unsigned flags)
+{
+   struct pool_buffer *pool_buf = pool_buffer(buf);
+   struct pool_pb_manager *pool = pool_buf->mgr;
+   return pb_validate(pool->buffer, vl, flags);
+}
+
+
+static void
+pool_buffer_fence(struct pb_buffer *buf, 
+                  struct pipe_fence_handle *fence)
+{
+   struct pool_buffer *pool_buf = pool_buffer(buf);
+   struct pool_pb_manager *pool = pool_buf->mgr;
+   pb_fence(pool->buffer, fence);
+}
+
+
+static void
+pool_buffer_get_base_buffer(struct pb_buffer *buf,
+                            struct pb_buffer **base_buf,
+                            unsigned *offset)
+{
+   struct pool_buffer *pool_buf = pool_buffer(buf);
+   struct pool_pb_manager *pool = pool_buf->mgr;
+   pb_get_base_buffer(pool->buffer, base_buf, offset);
+   *offset += pool_buf->start;
+}
+
+
+static const struct pb_vtbl 
+pool_buffer_vtbl = {
+      pool_buffer_destroy,
+      pool_buffer_map,
+      pool_buffer_unmap,
+      pool_buffer_validate,
+      pool_buffer_fence,
+      pool_buffer_get_base_buffer
+};
+
+
+static struct pb_buffer *
+pool_bufmgr_create_buffer(struct pb_manager *mgr,
+                          size_t size,
+                          const struct pb_desc *desc)
+{
+   struct pool_pb_manager *pool = pool_pb_manager(mgr);
+   struct pool_buffer *pool_buf;
+   struct list_head *item;
+
+   assert(size == pool->bufSize);
+   assert(pool->bufAlign % desc->alignment == 0);
+   
+   pipe_mutex_lock(pool->mutex);
+
+   if (pool->numFree == 0) {
+      pipe_mutex_unlock(pool->mutex);
+      debug_printf("warning: out of fixed size buffer objects\n");
+      return NULL;
+   }
+
+   item = pool->free.next;
+
+   if (item == &pool->free) {
+      pipe_mutex_unlock(pool->mutex);
+      debug_printf("error: fixed size buffer pool corruption\n");
+      return NULL;
+   }
+
+   LIST_DEL(item);
+   --pool->numFree;
+
+   pipe_mutex_unlock(pool->mutex);
+   
+   pool_buf = LIST_ENTRY(struct pool_buffer, item, head);
+   assert(pool_buf->base.base.refcount == 0);
+   pool_buf->base.base.refcount = 1;
+   pool_buf->base.base.alignment = desc->alignment;
+   pool_buf->base.base.usage = desc->usage;
+   
+   return SUPER(pool_buf);
+}
+
+
+static void
+pool_bufmgr_flush(struct pb_manager *mgr)
+{
+   /* No-op */
+}
+
+
+static void
+pool_bufmgr_destroy(struct pb_manager *mgr)
+{
+   struct pool_pb_manager *pool = pool_pb_manager(mgr);
+   pipe_mutex_lock(pool->mutex);
+
+   FREE(pool->bufs);
+   
+   pb_unmap(pool->buffer);
+   pb_reference(&pool->buffer, NULL);
+   
+   pipe_mutex_unlock(pool->mutex);
+   
+   FREE(mgr);
+}
+
+
+struct pb_manager *
+pool_bufmgr_create(struct pb_manager *provider, 
+                   size_t numBufs, 
+                   size_t bufSize,
+                   const struct pb_desc *desc) 
+{
+   struct pool_pb_manager *pool;
+   struct pool_buffer *pool_buf;
+   size_t i;
+
+   if(!provider)
+      return NULL;
+   
+   pool = CALLOC_STRUCT(pool_pb_manager);
+   if (!pool)
+      return NULL;
+
+   pool->base.destroy = pool_bufmgr_destroy;
+   pool->base.create_buffer = pool_bufmgr_create_buffer;
+   pool->base.flush = pool_bufmgr_flush;
+
+   LIST_INITHEAD(&pool->free);
+
+   pool->numTot = numBufs;
+   pool->numFree = numBufs;
+   pool->bufSize = bufSize;
+   pool->bufAlign = desc->alignment; 
+   
+   pipe_mutex_init(pool->mutex);
+
+   pool->buffer = provider->create_buffer(provider, numBufs*bufSize, desc); 
+   if (!pool->buffer)
+      goto failure;
+
+   pool->map = pb_map(pool->buffer,
+                          PIPE_BUFFER_USAGE_CPU_READ |
+                          PIPE_BUFFER_USAGE_CPU_WRITE);
+   if(!pool->map)
+      goto failure;
+
+   pool->bufs = (struct pool_buffer *)CALLOC(numBufs, sizeof(*pool->bufs));
+   if (!pool->bufs)
+      goto failure;
+
+   pool_buf = pool->bufs;
+   for (i = 0; i < numBufs; ++i) {
+      pool_buf->base.base.refcount = 0;
+      pool_buf->base.base.alignment = 0;
+      pool_buf->base.base.usage = 0;
+      pool_buf->base.base.size = bufSize;
+      pool_buf->base.vtbl = &pool_buffer_vtbl;
+      pool_buf->mgr = pool;
+      pool_buf->start = i * bufSize;
+      LIST_ADDTAIL(&pool_buf->head, &pool->free);
+      pool_buf++;
+   }
+
+   return SUPER(pool);
+   
+failure:
+   if(pool->bufs)
+      FREE(pool->bufs);
+   if(pool->map)
+      pb_unmap(pool->buffer);
+   if(pool->buffer)
+      pb_reference(&pool->buffer, NULL);
+   if(pool)
+      FREE(pool);
+   return NULL;
+}
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c
new file mode 100644
index 00000000000..9b9fedccb4b
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c
@@ -0,0 +1,584 @@
+/**************************************************************************
+ *
+ * Copyright 2006-2008 Tungsten Graphics, Inc., Cedar Park, TX., USA
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, FREE of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * S-lab pool implementation.
+ * 
+ * @sa http://en.wikipedia.org/wiki/Slab_allocation
+ * 
+ * @author Thomas Hellstrom <thomas-at-tungstengraphics-dot-com>
+ * @author Jose Fonseca <[email protected]>
+ */
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_error.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_thread.h"
+#include "pipe/p_defines.h"
+#include "util/u_memory.h"
+#include "util/u_double_list.h"
+#include "util/u_time.h"
+
+#include "pb_buffer.h"
+#include "pb_bufmgr.h"
+
+
+struct pb_slab;
+
+
+/**
+ * Buffer in a slab.
+ * 
+ * Sub-allocation of a contiguous buffer.
+ */
+struct pb_slab_buffer
+{
+   struct pb_buffer base;
+   
+   struct pb_slab *slab;
+   
+   struct list_head head;
+   
+   unsigned mapCount;
+   
+   /** Offset relative to the start of the slab buffer. */
+   size_t start;
+   
+   /** Use when validating, to signal that all mappings are finished */
+   /* TODO: Actually validation does not reach this stage yet */
+   pipe_condvar event;
+};
+
+
+/**
+ * Slab -- a contiguous piece of memory. 
+ */
+struct pb_slab
+{
+   struct list_head head;
+   struct list_head freeBuffers;
+   size_t numBuffers;
+   size_t numFree;
+   
+   struct pb_slab_buffer *buffers;
+   struct pb_slab_manager *mgr;
+   
+   /** Buffer from the provider */
+   struct pb_buffer *bo;
+   
+   void *virtual;   
+};
+
+
+/**
+ * It adds/removes slabs as needed in order to meet the allocation/destruction 
+ * of individual buffers.
+ */
+struct pb_slab_manager 
+{
+   struct pb_manager base;
+   
+   /** From where we get our buffers */
+   struct pb_manager *provider;
+   
+   /** Size of the buffers we hand on downstream */
+   size_t bufSize;
+   
+   /** Size of the buffers we request upstream */
+   size_t slabSize;
+   
+   /** 
+    * Alignment, usage to be used to allocate the slab buffers.
+    * 
+    * We can only provide buffers which are consistent (in alignment, usage) 
+    * with this description.   
+    */
+   struct pb_desc desc;
+
+   /** 
+    * Partial slabs
+    * 
+    * Full slabs are not stored in any list. Empty slabs are destroyed 
+    * immediatly.
+    */
+   struct list_head slabs;
+   
+   pipe_mutex mutex;
+};
+
+
+/**
+ * Wrapper around several slabs, therefore capable of handling buffers of 
+ * multiple sizes. 
+ * 
+ * This buffer manager just dispatches buffer allocations to the appropriate slab
+ * manager, according to the requested buffer size, or by passes the slab 
+ * managers altogether for even greater sizes.
+ * 
+ * The data of this structure remains constant after
+ * initialization and thus needs no mutex protection.
+ */
+struct pb_slab_range_manager 
+{
+   struct pb_manager base;
+
+   struct pb_manager *provider;
+   
+   size_t minBufSize;
+   size_t maxBufSize;
+   
+   /** @sa pb_slab_manager::desc */ 
+   struct pb_desc desc;
+   
+   unsigned numBuckets;
+   size_t *bucketSizes;
+   
+   /** Array of pb_slab_manager, one for each bucket size */
+   struct pb_manager **buckets;
+};
+
+
+static INLINE struct pb_slab_buffer *
+pb_slab_buffer(struct pb_buffer *buf)
+{
+   assert(buf);
+   return (struct pb_slab_buffer *)buf;
+}
+
+
+static INLINE struct pb_slab_manager *
+pb_slab_manager(struct pb_manager *mgr)
+{
+   assert(mgr);
+   return (struct pb_slab_manager *)mgr;
+}
+
+
+static INLINE struct pb_slab_range_manager *
+pb_slab_range_manager(struct pb_manager *mgr)
+{
+   assert(mgr);
+   return (struct pb_slab_range_manager *)mgr;
+}
+
+
+/**
+ * Delete a buffer from the slab delayed list and put
+ * it on the slab FREE list.
+ */
+static void
+pb_slab_buffer_destroy(struct pb_buffer *_buf)
+{
+   struct pb_slab_buffer *buf = pb_slab_buffer(_buf);
+   struct pb_slab *slab = buf->slab;
+   struct pb_slab_manager *mgr = slab->mgr;
+   struct list_head *list = &buf->head;
+
+   pipe_mutex_lock(mgr->mutex);
+   
+   assert(buf->base.base.refcount == 0);
+   
+   buf->mapCount = 0;
+
+   LIST_DEL(list);
+   LIST_ADDTAIL(list, &slab->freeBuffers);
+   slab->numFree++;
+
+   if (slab->head.next == &slab->head)
+      LIST_ADDTAIL(&slab->head, &mgr->slabs);
+
+   /* If the slab becomes totally empty, free it */
+   if (slab->numFree == slab->numBuffers) {
+      list = &slab->head;
+      LIST_DELINIT(list);
+      pb_reference(&slab->bo, NULL);
+      FREE(slab->buffers);
+      FREE(slab);
+   }
+
+   pipe_mutex_unlock(mgr->mutex);
+}
+
+
+static void *
+pb_slab_buffer_map(struct pb_buffer *_buf, 
+                   unsigned flags)
+{
+   struct pb_slab_buffer *buf = pb_slab_buffer(_buf);
+
+   ++buf->mapCount;
+   return (void *) ((uint8_t *) buf->slab->virtual + buf->start);
+}
+
+
+static void
+pb_slab_buffer_unmap(struct pb_buffer *_buf)
+{
+   struct pb_slab_buffer *buf = pb_slab_buffer(_buf);
+
+   --buf->mapCount;
+   if (buf->mapCount == 0) 
+       pipe_condvar_broadcast(buf->event);
+}
+
+
+static enum pipe_error 
+pb_slab_buffer_validate(struct pb_buffer *_buf, 
+                         struct pb_validate *vl,
+                         unsigned flags)
+{
+   struct pb_slab_buffer *buf = pb_slab_buffer(_buf);
+   return pb_validate(buf->slab->bo, vl, flags);
+}
+
+
+static void
+pb_slab_buffer_fence(struct pb_buffer *_buf, 
+                      struct pipe_fence_handle *fence)
+{
+   struct pb_slab_buffer *buf = pb_slab_buffer(_buf);
+   pb_fence(buf->slab->bo, fence);
+}
+
+
+static void
+pb_slab_buffer_get_base_buffer(struct pb_buffer *_buf,
+                               struct pb_buffer **base_buf,
+                               unsigned *offset)
+{
+   struct pb_slab_buffer *buf = pb_slab_buffer(_buf);
+   pb_get_base_buffer(buf->slab->bo, base_buf, offset);
+   *offset += buf->start;
+}
+
+
+static const struct pb_vtbl 
+pb_slab_buffer_vtbl = {
+      pb_slab_buffer_destroy,
+      pb_slab_buffer_map,
+      pb_slab_buffer_unmap,
+      pb_slab_buffer_validate,
+      pb_slab_buffer_fence,
+      pb_slab_buffer_get_base_buffer
+};
+
+
+/**
+ * Create a new slab.
+ * 
+ * Called when we ran out of free slabs.
+ */
+static enum pipe_error
+pb_slab_create(struct pb_slab_manager *mgr)
+{
+   struct pb_slab *slab;
+   struct pb_slab_buffer *buf;
+   unsigned numBuffers;
+   unsigned i;
+   enum pipe_error ret;
+
+   slab = CALLOC_STRUCT(pb_slab);
+   if (!slab)
+      return PIPE_ERROR_OUT_OF_MEMORY;
+
+   slab->bo = mgr->provider->create_buffer(mgr->provider, mgr->slabSize, &mgr->desc);
+   if(!slab->bo) {
+      ret = PIPE_ERROR_OUT_OF_MEMORY;
+      goto out_err0;
+   }
+
+   /* Note down the slab virtual address. All mappings are accessed directly 
+    * through this address so it is required that the buffer is pinned. */
+   slab->virtual = pb_map(slab->bo, 
+                          PIPE_BUFFER_USAGE_CPU_READ |
+                          PIPE_BUFFER_USAGE_CPU_WRITE);
+   if(!slab->virtual) {
+      ret = PIPE_ERROR_OUT_OF_MEMORY;
+      goto out_err1;
+   }
+   pb_unmap(slab->bo);
+
+   numBuffers = slab->bo->base.size / mgr->bufSize;
+
+   slab->buffers = CALLOC(numBuffers, sizeof(*slab->buffers));
+   if (!slab->buffers) {
+      ret = PIPE_ERROR_OUT_OF_MEMORY;
+      goto out_err1;
+   }
+
+   LIST_INITHEAD(&slab->head);
+   LIST_INITHEAD(&slab->freeBuffers);
+   slab->numBuffers = numBuffers;
+   slab->numFree = 0;
+   slab->mgr = mgr;
+
+   buf = slab->buffers;
+   for (i=0; i < numBuffers; ++i) {
+      buf->base.base.refcount = 0;
+      buf->base.base.size = mgr->bufSize;
+      buf->base.base.alignment = 0;
+      buf->base.base.usage = 0;
+      buf->base.vtbl = &pb_slab_buffer_vtbl;
+      buf->slab = slab;
+      buf->start = i* mgr->bufSize;
+      buf->mapCount = 0;
+      pipe_condvar_init(buf->event);
+      LIST_ADDTAIL(&buf->head, &slab->freeBuffers);
+      slab->numFree++;
+      buf++;
+   }
+
+   /* Add this slab to the list of partial slabs */
+   LIST_ADDTAIL(&slab->head, &mgr->slabs);
+
+   return PIPE_OK;
+
+out_err1: 
+   pb_reference(&slab->bo, NULL);
+out_err0: 
+   FREE(slab);
+   return ret;
+}
+
+
+static struct pb_buffer *
+pb_slab_manager_create_buffer(struct pb_manager *_mgr,
+                              size_t size,
+                              const struct pb_desc *desc)
+{
+   struct pb_slab_manager *mgr = pb_slab_manager(_mgr);
+   static struct pb_slab_buffer *buf;
+   struct pb_slab *slab;
+   struct list_head *list;
+
+   /* check size */
+   assert(size <= mgr->bufSize);
+   if(size > mgr->bufSize)
+      return NULL;
+   
+   /* check if we can provide the requested alignment */
+   assert(pb_check_alignment(desc->alignment, mgr->desc.alignment));
+   if(!pb_check_alignment(desc->alignment, mgr->desc.alignment))
+      return NULL;
+   assert(pb_check_alignment(desc->alignment, mgr->bufSize));
+   if(!pb_check_alignment(desc->alignment, mgr->bufSize))
+      return NULL;
+
+   assert(pb_check_usage(desc->usage, mgr->desc.usage));
+   if(!pb_check_usage(desc->usage, mgr->desc.usage))
+      return NULL;
+
+   pipe_mutex_lock(mgr->mutex);
+   
+   /* Create a new slab, if we run out of partial slabs */
+   if (mgr->slabs.next == &mgr->slabs) {
+      (void) pb_slab_create(mgr);
+      if (mgr->slabs.next == &mgr->slabs) {
+	 pipe_mutex_unlock(mgr->mutex);
+	 return NULL;
+      }
+   }
+   
+   /* Allocate the buffer from a partial (or just created) slab */
+   list = mgr->slabs.next;
+   slab = LIST_ENTRY(struct pb_slab, list, head);
+   
+   /* If totally full remove from the partial slab list */
+   if (--slab->numFree == 0)
+      LIST_DELINIT(list);
+
+   list = slab->freeBuffers.next;
+   LIST_DELINIT(list);
+
+   pipe_mutex_unlock(mgr->mutex);
+   buf = LIST_ENTRY(struct pb_slab_buffer, list, head);
+   
+   ++buf->base.base.refcount;
+   buf->base.base.alignment = desc->alignment;
+   buf->base.base.usage = desc->usage;
+   
+   return &buf->base;
+}
+
+
+static void
+pb_slab_manager_flush(struct pb_manager *_mgr)
+{
+   struct pb_slab_manager *mgr = pb_slab_manager(_mgr);
+
+   assert(mgr->provider->flush);
+   if(mgr->provider->flush)
+      mgr->provider->flush(mgr->provider);
+}
+
+
+static void
+pb_slab_manager_destroy(struct pb_manager *_mgr)
+{
+   struct pb_slab_manager *mgr = pb_slab_manager(_mgr);
+
+   /* TODO: cleanup all allocated buffers */
+   FREE(mgr);
+}
+
+
+struct pb_manager *
+pb_slab_manager_create(struct pb_manager *provider,
+                       size_t bufSize,
+                       size_t slabSize,
+                       const struct pb_desc *desc)
+{
+   struct pb_slab_manager *mgr;
+
+   mgr = CALLOC_STRUCT(pb_slab_manager);
+   if (!mgr)
+      return NULL;
+
+   mgr->base.destroy = pb_slab_manager_destroy;
+   mgr->base.create_buffer = pb_slab_manager_create_buffer;
+   mgr->base.flush = pb_slab_manager_flush;
+
+   mgr->provider = provider;
+   mgr->bufSize = bufSize;
+   mgr->slabSize = slabSize;
+   mgr->desc = *desc;
+
+   LIST_INITHEAD(&mgr->slabs);
+   
+   pipe_mutex_init(mgr->mutex);
+
+   return &mgr->base;
+}
+
+
+static struct pb_buffer *
+pb_slab_range_manager_create_buffer(struct pb_manager *_mgr,
+                                    size_t size,
+                                    const struct pb_desc *desc)
+{
+   struct pb_slab_range_manager *mgr = pb_slab_range_manager(_mgr);
+   size_t bufSize;
+   unsigned i;
+
+   bufSize = mgr->minBufSize;
+   for (i = 0; i < mgr->numBuckets; ++i) {
+      if(bufSize >= size)
+	 return mgr->buckets[i]->create_buffer(mgr->buckets[i], size, desc);
+      bufSize *= 2;
+   }
+
+   /* Fall back to allocate a buffer object directly from the provider. */
+   return mgr->provider->create_buffer(mgr->provider, size, desc);
+}
+
+
+static void
+pb_slab_range_manager_flush(struct pb_manager *_mgr)
+{
+   struct pb_slab_range_manager *mgr = pb_slab_range_manager(_mgr);
+
+   /* Individual slabs don't hold any temporary buffers so no need to call them */
+   
+   assert(mgr->provider->flush);
+   if(mgr->provider->flush)
+      mgr->provider->flush(mgr->provider);
+}
+
+
+static void
+pb_slab_range_manager_destroy(struct pb_manager *_mgr)
+{
+   struct pb_slab_range_manager *mgr = pb_slab_range_manager(_mgr);
+   unsigned i;
+   
+   for (i = 0; i < mgr->numBuckets; ++i)
+      mgr->buckets[i]->destroy(mgr->buckets[i]);
+   FREE(mgr->buckets);
+   FREE(mgr->bucketSizes);
+   FREE(mgr);
+}
+
+
+struct pb_manager *
+pb_slab_range_manager_create(struct pb_manager *provider,
+                             size_t minBufSize,
+                             size_t maxBufSize,
+                             size_t slabSize,
+                             const struct pb_desc *desc)
+{
+   struct pb_slab_range_manager *mgr;
+   size_t bufSize;
+   unsigned i;
+
+   if(!provider)
+      return NULL;
+   
+   mgr = CALLOC_STRUCT(pb_slab_range_manager);
+   if (!mgr)
+      goto out_err0;
+
+   mgr->base.destroy = pb_slab_range_manager_destroy;
+   mgr->base.create_buffer = pb_slab_range_manager_create_buffer;
+   mgr->base.flush = pb_slab_range_manager_flush;
+
+   mgr->provider = provider;
+   mgr->minBufSize = minBufSize;
+   mgr->maxBufSize = maxBufSize;
+
+   mgr->numBuckets = 1;
+   bufSize = minBufSize;
+   while(bufSize < maxBufSize) {
+      bufSize *= 2;
+      ++mgr->numBuckets;
+   }
+   
+   mgr->buckets = CALLOC(mgr->numBuckets, sizeof(*mgr->buckets));
+   if (!mgr->buckets)
+      goto out_err1;
+
+   bufSize = minBufSize;
+   for (i = 0; i < mgr->numBuckets; ++i) {
+      mgr->buckets[i] = pb_slab_manager_create(provider, bufSize, slabSize, desc);
+      if(!mgr->buckets[i])
+	 goto out_err2;
+      bufSize *= 2;
+   }
+
+   return &mgr->base;
+
+out_err2: 
+   for (i = 0; i < mgr->numBuckets; ++i)
+      if(mgr->buckets[i])
+	    mgr->buckets[i]->destroy(mgr->buckets[i]);
+   FREE(mgr->buckets);
+out_err1: 
+   FREE(mgr);
+out_err0:
+   return NULL;
+}
diff --git a/src/gallium/auxiliary/pipebuffer/pb_validate.c b/src/gallium/auxiliary/pipebuffer/pb_validate.c
new file mode 100644
index 00000000000..94532bb4cec
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/pb_validate.c
@@ -0,0 +1,193 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Buffer validation.
+ * 
+ * @author Jose Fonseca <[email protected]>
+ */
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_error.h"
+#include "util/u_memory.h"
+#include "pipe/p_debug.h"
+
+#include "pb_buffer.h"
+#include "pb_buffer_fenced.h"
+#include "pb_validate.h"
+
+
+#define PB_VALIDATE_INITIAL_SIZE 1 /* 512 */ 
+
+
+struct pb_validate_entry
+{
+   struct pb_buffer *buf;
+   unsigned flags;
+};
+
+
+struct pb_validate
+{
+   struct pb_validate_entry *entries;
+   unsigned used;
+   unsigned size;
+};
+
+
+enum pipe_error
+pb_validate_add_buffer(struct pb_validate *vl,
+                       struct pb_buffer *buf,
+                       unsigned flags)
+{
+   assert(buf);
+   if(!buf)
+      return PIPE_ERROR;
+
+   assert(flags & PIPE_BUFFER_USAGE_GPU_READ_WRITE);
+   assert(!(flags & ~PIPE_BUFFER_USAGE_GPU_READ_WRITE));
+   flags &= PIPE_BUFFER_USAGE_GPU_READ_WRITE;
+
+   /* We only need to store one reference for each buffer, so avoid storing
+    * consecutive references for the same buffer. It might not be the most 
+    * common pattern, but it is easy to implement.
+    */
+   if(vl->used && vl->entries[vl->used - 1].buf == buf) {
+      vl->entries[vl->used - 1].flags |= flags;
+      return PIPE_OK;
+   }
+   
+   /* Grow the table */
+   if(vl->used == vl->size) {
+      unsigned new_size;
+      struct pb_validate_entry *new_entries;
+      
+      new_size = vl->size * 2;
+      if(!new_size)
+	 return PIPE_ERROR_OUT_OF_MEMORY;
+
+      new_entries = (struct pb_validate_entry *)REALLOC(vl->entries,
+                                                        vl->size*sizeof(struct pb_validate_entry),
+                                                        new_size*sizeof(struct pb_validate_entry));
+      if(!new_entries)
+         return PIPE_ERROR_OUT_OF_MEMORY;
+      
+      memset(new_entries + vl->size, 0, (new_size - vl->size)*sizeof(struct pb_validate_entry));
+      
+      vl->size = new_size;
+      vl->entries = new_entries;
+   }
+   
+   assert(!vl->entries[vl->used].buf);
+   pb_reference(&vl->entries[vl->used].buf, buf);
+   vl->entries[vl->used].flags = flags;
+   ++vl->used;
+   
+   return PIPE_OK;
+}
+
+
+enum pipe_error
+pb_validate_foreach(struct pb_validate *vl,
+                    enum pipe_error (*callback)(struct pb_buffer *buf, void *data),
+                    void *data)
+{
+   unsigned i;
+   for(i = 0; i < vl->used; ++i) {
+      enum pipe_error ret;
+      ret = callback(vl->entries[i].buf, data);
+      if(ret != PIPE_OK)
+         return ret;
+   }
+   return PIPE_OK;
+}
+
+
+enum pipe_error
+pb_validate_validate(struct pb_validate *vl) 
+{
+   unsigned i;
+   
+   for(i = 0; i < vl->used; ++i) {
+      enum pipe_error ret;
+      ret = pb_validate(vl->entries[i].buf, vl, vl->entries[i].flags);
+      if(ret != PIPE_OK) {
+         while(i--)
+            pb_validate(vl->entries[i].buf, NULL, 0);
+         return ret;
+      }
+   }
+
+   return PIPE_OK;
+}
+
+
+void
+pb_validate_fence(struct pb_validate *vl,
+                  struct pipe_fence_handle *fence)
+{
+   unsigned i;
+   for(i = 0; i < vl->used; ++i) {
+      pb_fence(vl->entries[i].buf, fence);
+      pb_reference(&vl->entries[i].buf, NULL);
+   }
+   vl->used = 0;
+}
+
+
+void
+pb_validate_destroy(struct pb_validate *vl)
+{
+   unsigned i;
+   for(i = 0; i < vl->used; ++i)
+      pb_reference(&vl->entries[i].buf, NULL);
+   FREE(vl->entries);
+   FREE(vl);
+}
+
+
+struct pb_validate *
+pb_validate_create()
+{
+   struct pb_validate *vl;
+   
+   vl = CALLOC_STRUCT(pb_validate);
+   if(!vl)
+      return NULL;
+   
+   vl->size = PB_VALIDATE_INITIAL_SIZE;
+   vl->entries = (struct pb_validate_entry *)CALLOC(vl->size, sizeof(struct pb_validate_entry));
+   if(!vl->entries) {
+      FREE(vl);
+      return NULL;
+   }
+
+   return vl;
+}
+
diff --git a/src/gallium/auxiliary/pipebuffer/pb_validate.h b/src/gallium/auxiliary/pipebuffer/pb_validate.h
new file mode 100644
index 00000000000..dfb84df1cef
--- /dev/null
+++ b/src/gallium/auxiliary/pipebuffer/pb_validate.h
@@ -0,0 +1,97 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Buffer validation.
+ * 
+ * @author Jose Fonseca <[email protected]>
+ */
+
+#ifndef PB_VALIDATE_H_
+#define PB_VALIDATE_H_
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_error.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+struct pb_buffer;
+struct pipe_fence_handle;
+
+
+/**
+ * Buffer validation list.
+ * 
+ * It holds a list of buffers to be validated and fenced when flushing.
+ */
+struct pb_validate;
+
+
+enum pipe_error
+pb_validate_add_buffer(struct pb_validate *vl,
+                       struct pb_buffer *buf,
+                       unsigned flags);
+
+enum pipe_error
+pb_validate_foreach(struct pb_validate *vl,
+                    enum pipe_error (*callback)(struct pb_buffer *buf, void *data),
+                    void *data);
+
+/**
+ * Validate all buffers for hardware access.
+ * 
+ * Should be called right before issuing commands to the hardware.
+ */
+enum pipe_error
+pb_validate_validate(struct pb_validate *vl);
+
+/**
+ * Fence all buffers and clear the list.
+ * 
+ * Should be called right after issuing commands to the hardware.
+ */
+void
+pb_validate_fence(struct pb_validate *vl,
+                  struct pipe_fence_handle *fence);
+
+struct pb_validate *
+pb_validate_create(void);
+
+void
+pb_validate_destroy(struct pb_validate *vl);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /*PB_VALIDATE_H_*/
diff --git a/src/gallium/auxiliary/rtasm/Makefile b/src/gallium/auxiliary/rtasm/Makefile
new file mode 100644
index 00000000000..252dc5274ab
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/Makefile
@@ -0,0 +1,16 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = rtasm
+
+C_SOURCES = \
+	rtasm_cpu.c \
+	rtasm_execmem.c \
+	rtasm_x86sse.c \
+	rtasm_ppc.c \
+	rtasm_ppc_spe.c
+
+include ../../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/auxiliary/rtasm/SConscript b/src/gallium/auxiliary/rtasm/SConscript
new file mode 100644
index 00000000000..eb48368accb
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/SConscript
@@ -0,0 +1,13 @@
+Import('*')
+
+rtasm = env.ConvenienceLibrary(
+	target = 'rtasm',
+	source = [
+		'rtasm_cpu.c',
+		'rtasm_execmem.c',
+		'rtasm_x86sse.c',
+		'rtasm_ppc.c',
+		'rtasm_ppc_spe.c',
+	])
+
+auxiliaries.insert(0, rtasm)
diff --git a/src/gallium/auxiliary/rtasm/rtasm_cpu.c b/src/gallium/auxiliary/rtasm/rtasm_cpu.c
new file mode 100644
index 00000000000..5499018b219
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/rtasm_cpu.c
@@ -0,0 +1,65 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+
+#include "pipe/p_debug.h"
+#include "rtasm_cpu.h"
+
+
+static boolean rtasm_sse_enabled(void)
+{
+   static boolean firsttime = 1;
+   static boolean enabled;
+   
+   /* This gets called quite often at the moment:
+    */
+   if (firsttime) {
+      enabled =  !debug_get_bool_option("GALLIUM_NOSSE", FALSE);
+      firsttime = FALSE;
+   }
+   return enabled;
+}
+
+int rtasm_cpu_has_sse(void)
+{
+   /* FIXME: actually detect this at run-time */
+#if defined(PIPE_ARCH_X86)
+   return rtasm_sse_enabled();
+#else
+   return 0;
+#endif
+}
+
+int rtasm_cpu_has_sse2(void) 
+{
+   /* FIXME: actually detect this at run-time */
+#if defined(PIPE_ARCH_X86)
+   return rtasm_sse_enabled();
+#else
+   return 0;
+#endif
+}
diff --git a/src/gallium/auxiliary/rtasm/rtasm_cpu.h b/src/gallium/auxiliary/rtasm/rtasm_cpu.h
new file mode 100644
index 00000000000..ebc71634fdf
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/rtasm_cpu.h
@@ -0,0 +1,42 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Runtime detection of CPU capabilities.
+ */
+
+#ifndef _RTASM_CPU_H_
+#define _RTASM_CPU_H_
+
+
+int rtasm_cpu_has_sse(void);
+
+int rtasm_cpu_has_sse2(void);
+
+
+#endif /* _RTASM_CPU_H_ */
diff --git a/src/gallium/auxiliary/rtasm/rtasm_execmem.c b/src/gallium/auxiliary/rtasm/rtasm_execmem.c
new file mode 100644
index 00000000000..be7433baf87
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/rtasm_execmem.c
@@ -0,0 +1,137 @@
+/**************************************************************************
+ *
+ * Copyright (C) 1999-2005  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+
+/**
+ * \file exemem.c
+ * Functions for allocating executable memory.
+ *
+ * \author Keith Whitwell
+ */
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_thread.h"
+#include "util/u_memory.h"
+
+#include "rtasm_execmem.h"
+
+
+#if defined(PIPE_OS_LINUX)
+
+
+/*
+ * Allocate a large block of memory which can hold code then dole it out
+ * in pieces by means of the generic memory manager code.
+ */
+
+#include <unistd.h>
+#include <sys/mman.h>
+#include "pipe/p_thread.h"
+#include "util/u_mm.h"
+
+#define EXEC_HEAP_SIZE (10*1024*1024)
+
+pipe_static_mutex(exec_mutex);
+
+static struct mem_block *exec_heap = NULL;
+static unsigned char *exec_mem = NULL;
+
+
+static void
+init_heap(void)
+{
+   if (!exec_heap)
+      exec_heap = u_mmInit( 0, EXEC_HEAP_SIZE );
+   
+   if (!exec_mem)
+      exec_mem = (unsigned char *) mmap(0, EXEC_HEAP_SIZE, 
+					PROT_EXEC | PROT_READ | PROT_WRITE, 
+					MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
+}
+
+
+void *
+rtasm_exec_malloc(size_t size)
+{
+   struct mem_block *block = NULL;
+   void *addr = NULL;
+
+   pipe_mutex_lock(exec_mutex);
+
+   init_heap();
+
+   if (exec_heap) {
+      size = (size + 31) & ~31;  /* next multiple of 32 bytes */
+      block = u_mmAllocMem( exec_heap, size, 5, 0 ); /* 5 -> 32-byte alignment */
+   }
+
+   if (block)
+      addr = exec_mem + block->ofs;
+   else 
+      debug_printf("rtasm_exec_malloc failed\n");
+   
+   pipe_mutex_unlock(exec_mutex);
+   
+   return addr;
+}
+
+ 
+void 
+rtasm_exec_free(void *addr)
+{
+   pipe_mutex_lock(exec_mutex);
+
+   if (exec_heap) {
+      struct mem_block *block = u_mmFindBlock(exec_heap, (unsigned char *)addr - exec_mem);
+   
+      if (block)
+	 u_mmFreeMem(block);
+   }
+
+   pipe_mutex_unlock(exec_mutex);
+}
+
+
+#else /* PIPE_OS_LINUX */
+
+/*
+ * Just use regular memory.
+ */
+
+void *
+rtasm_exec_malloc(size_t size)
+{
+   return MALLOC( size );
+}
+
+ 
+void 
+rtasm_exec_free(void *addr)
+{
+   FREE(addr);
+}
+
+
+#endif /* PIPE_OS_LINUX */
diff --git a/src/gallium/auxiliary/rtasm/rtasm_execmem.h b/src/gallium/auxiliary/rtasm/rtasm_execmem.h
new file mode 100644
index 00000000000..155c6d34e0a
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/rtasm_execmem.h
@@ -0,0 +1,45 @@
+/**************************************************************************
+ *
+ * Copyright (C) 1999-2005  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * \file exemem.c
+ * Functions for allocating executable memory.
+ *
+ * \author Keith Whitwell
+ */
+
+#ifndef _RTASM_EXECMEM_H_
+#define _RTASM_EXECMEM_H_
+
+#include "pipe/p_compiler.h"
+
+
+extern void *
+rtasm_exec_malloc( size_t size );
+
+
+extern void 
+rtasm_exec_free( void *addr );
+
+
+#endif
diff --git a/src/gallium/auxiliary/rtasm/rtasm_ppc.c b/src/gallium/auxiliary/rtasm/rtasm_ppc.c
new file mode 100644
index 00000000000..1bb90262051
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/rtasm_ppc.c
@@ -0,0 +1,1077 @@
+/**************************************************************************
+ *
+ * Copyright (C) 2008 Tungsten Graphics, Inc.   All Rights Reserved.
+ * Copyright (C) 2009 VMware, Inc.  All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * PPC code generation.
+ * For reference, see http://www.power.org/resources/reading/PowerISA_V2.05.pdf
+ * ABI info: http://www.cs.utsa.edu/~whaley/teach/cs6463FHPO/LEC/lec12_ho.pdf
+ *
+ * Other PPC refs:
+ * http://www-01.ibm.com/chips/techlib/techlib.nsf/techdocs/852569B20050FF778525699600719DF2
+ * http://www.ibm.com/developerworks/eserver/library/es-archguide-v2.html
+ * http://www.freescale.com/files/product/doc/MPCFPE32B.pdf
+ *
+ * \author Brian Paul
+ */
+
+
+#include <stdio.h>
+#include "util/u_memory.h"
+#include "pipe/p_debug.h"
+#include "rtasm_execmem.h"
+#include "rtasm_ppc.h"
+
+
+void
+ppc_init_func(struct ppc_function *p)
+{
+   uint i;
+
+   memset(p, 0, sizeof(*p));
+
+   p->num_inst = 0;
+   p->max_inst = 100; /* first guess at buffer size */
+   p->store = rtasm_exec_malloc(p->max_inst * PPC_INST_SIZE);
+   p->reg_used = 0x0;
+   p->fp_used = 0x0;
+   p->vec_used = 0x0;
+
+   p->print = FALSE;
+   p->indent = 0;
+
+   /* only allow using gp registers 3..12 for now */
+   for (i = 0; i < 3; i++)
+      ppc_reserve_register(p, i);
+   for (i = 12; i < PPC_NUM_REGS; i++)
+      ppc_reserve_register(p, i);
+}
+
+
+void
+ppc_release_func(struct ppc_function *p)
+{
+   assert(p->num_inst <= p->max_inst);
+   if (p->store != NULL) {
+      rtasm_exec_free(p->store);
+   }
+   p->store = NULL;
+}
+
+
+uint
+ppc_num_instructions(const struct ppc_function *p)
+{
+   return p->num_inst;
+}
+
+
+void (*ppc_get_func(struct ppc_function *p))(void)
+{
+#if 0
+   DUMP_END();
+   if (DISASSEM && p->store)
+      debug_printf("disassemble %p %p\n", p->store, p->csr);
+
+   if (p->store == p->error_overflow)
+      return (void (*)(void)) NULL;
+   else
+#endif
+      return (void (*)(void)) p->store;
+}
+
+
+void
+ppc_dump_func(const struct ppc_function *p)
+{
+   uint i;
+   for (i = 0; i < p->num_inst; i++) {
+      debug_printf("%3u: 0x%08x\n", i, p->store[i]);
+   }
+}
+
+
+void
+ppc_print_code(struct ppc_function *p, boolean enable)
+{
+   p->print = enable;
+}
+
+
+void
+ppc_indent(struct ppc_function *p, int spaces)
+{
+   p->indent += spaces;
+}
+
+
+static void
+indent(const struct ppc_function *p)
+{
+   int i;
+   for (i = 0; i < p->indent; i++) {
+      putchar(' ');
+   }
+}
+
+
+void
+ppc_comment(struct ppc_function *p, int rel_indent, const char *s)
+{
+   if (p->print) {
+      p->indent += rel_indent;
+      indent(p);
+      p->indent -= rel_indent;
+      printf("# %s\n", s);
+   }
+}
+
+
+/**
+ * Mark a register as being unavailable.
+ */
+int
+ppc_reserve_register(struct ppc_function *p, int reg)
+{
+   assert(reg < PPC_NUM_REGS);
+   p->reg_used |= (1 << reg);
+   return reg;
+}
+
+
+/**
+ * Allocate a general purpose register.
+ * \return register index or -1 if none left.
+ */
+int
+ppc_allocate_register(struct ppc_function *p)
+{
+   unsigned i;
+   for (i = 0; i < PPC_NUM_REGS; i++) {
+      const uint64_t mask = 1 << i;
+      if ((p->reg_used & mask) == 0) {
+         p->reg_used |= mask;
+         return i;
+      }
+   }
+   printf("OUT OF PPC registers!\n");
+   return -1;
+}
+
+
+/**
+ * Mark the given general purpose register as "unallocated".
+ */
+void
+ppc_release_register(struct ppc_function *p, int reg)
+{
+   assert(reg < PPC_NUM_REGS);
+   assert(p->reg_used & (1 << reg));
+   p->reg_used &= ~(1 << reg);
+}
+
+
+/**
+ * Allocate a floating point register.
+ * \return register index or -1 if none left.
+ */
+int
+ppc_allocate_fp_register(struct ppc_function *p)
+{
+   unsigned i;
+   for (i = 0; i < PPC_NUM_FP_REGS; i++) {
+      const uint64_t mask = 1 << i;
+      if ((p->fp_used & mask) == 0) {
+         p->fp_used |= mask;
+         return i;
+      }
+   }
+   printf("OUT OF PPC FP registers!\n");
+   return -1;
+}
+
+
+/**
+ * Mark the given floating point register as "unallocated".
+ */
+void
+ppc_release_fp_register(struct ppc_function *p, int reg)
+{
+   assert(reg < PPC_NUM_FP_REGS);
+   assert(p->fp_used & (1 << reg));
+   p->fp_used &= ~(1 << reg);
+}
+
+
+/**
+ * Allocate a vector register.
+ * \return register index or -1 if none left.
+ */
+int
+ppc_allocate_vec_register(struct ppc_function *p)
+{
+   unsigned i;
+   for (i = 0; i < PPC_NUM_VEC_REGS; i++) {
+      const uint64_t mask = 1 << i;
+      if ((p->vec_used & mask) == 0) {
+         p->vec_used |= mask;
+         return i;
+      }
+   }
+   printf("OUT OF PPC VEC registers!\n");
+   return -1;
+}
+
+
+/**
+ * Mark the given vector register as "unallocated".
+ */
+void
+ppc_release_vec_register(struct ppc_function *p, int reg)
+{
+   assert(reg < PPC_NUM_VEC_REGS);
+   assert(p->vec_used & (1 << reg));
+   p->vec_used &= ~(1 << reg);
+}
+
+
+/**
+ * Append instruction to instruction buffer.  Grow buffer if out of room.
+ */
+static void
+emit_instruction(struct ppc_function *p, uint32_t inst_bits)
+{
+   if (!p->store)
+      return;  /* out of memory, drop the instruction */
+
+   if (p->num_inst == p->max_inst) {
+      /* allocate larger buffer */
+      uint32_t *newbuf;
+      p->max_inst *= 2;  /* 2x larger */
+      newbuf = rtasm_exec_malloc(p->max_inst * PPC_INST_SIZE);
+      if (newbuf) {
+         memcpy(newbuf, p->store, p->num_inst * PPC_INST_SIZE);
+      }
+      rtasm_exec_free(p->store);
+      p->store = newbuf;
+      if (!p->store) {
+         /* out of memory */
+         p->num_inst = 0;
+         return;
+      }
+   }
+
+   p->store[p->num_inst++] = inst_bits;
+}
+
+
+union vx_inst {
+   uint32_t bits;
+   struct {
+      unsigned op:6;
+      unsigned vD:5;
+      unsigned vA:5;
+      unsigned vB:5;
+      unsigned op2:11;
+   } inst;
+};
+
+static INLINE void
+emit_vx(struct ppc_function *p, uint op2, uint vD, uint vA, uint vB,
+        const char *format, boolean transpose)
+{
+   union vx_inst inst;
+   inst.inst.op = 4;
+   inst.inst.vD = vD;
+   inst.inst.vA = vA;
+   inst.inst.vB = vB;
+   inst.inst.op2 = op2;
+   emit_instruction(p, inst.bits);
+   if (p->print) {
+      indent(p);
+      if (transpose)
+         printf(format, vD, vB, vA);
+      else
+         printf(format, vD, vA, vB);
+   }
+}
+
+
+union vxr_inst {
+   uint32_t bits;
+   struct {
+      unsigned op:6;
+      unsigned vD:5;
+      unsigned vA:5;
+      unsigned vB:5;
+      unsigned rC:1;
+      unsigned op2:10;
+   } inst;
+};
+
+static INLINE void
+emit_vxr(struct ppc_function *p, uint op2, uint vD, uint vA, uint vB,
+         const char *format)
+{
+   union vxr_inst inst;
+   inst.inst.op = 4;
+   inst.inst.vD = vD;
+   inst.inst.vA = vA;
+   inst.inst.vB = vB;
+   inst.inst.rC = 0;
+   inst.inst.op2 = op2;
+   emit_instruction(p, inst.bits);
+   if (p->print) {
+      indent(p);
+      printf(format, vD, vA, vB);
+   }
+}
+
+
+union va_inst {
+   uint32_t bits;
+   struct {
+      unsigned op:6;
+      unsigned vD:5;
+      unsigned vA:5;
+      unsigned vB:5;
+      unsigned vC:5;
+      unsigned op2:6;
+   } inst;
+};
+
+static INLINE void
+emit_va(struct ppc_function *p, uint op2, uint vD, uint vA, uint vB, uint vC,
+        const char *format)
+{
+   union va_inst inst;
+   inst.inst.op = 4;
+   inst.inst.vD = vD;
+   inst.inst.vA = vA;
+   inst.inst.vB = vB;
+   inst.inst.vC = vC;
+   inst.inst.op2 = op2;
+   emit_instruction(p, inst.bits);
+   if (p->print) {
+      indent(p);
+      printf(format, vD, vA, vB, vC);
+   }
+}
+
+
+union i_inst {
+   uint32_t bits;
+   struct {
+      unsigned op:6;
+      unsigned li:24;
+      unsigned aa:1;
+      unsigned lk:1;
+   } inst;
+};
+
+static INLINE void
+emit_i(struct ppc_function *p, uint op, uint li, uint aa, uint lk)
+{
+   union i_inst inst;
+   inst.inst.op = op;
+   inst.inst.li = li;
+   inst.inst.aa = aa;
+   inst.inst.lk = lk;
+   emit_instruction(p, inst.bits);
+}
+
+
+union xl_inst {
+   uint32_t bits;
+   struct {
+      unsigned op:6;
+      unsigned bo:5;
+      unsigned bi:5;
+      unsigned unused:3;
+      unsigned bh:2;
+      unsigned op2:10;
+      unsigned lk:1;
+   } inst;
+};
+
+static INLINE void
+emit_xl(struct ppc_function *p, uint op, uint bo, uint bi, uint bh,
+        uint op2, uint lk)
+{
+   union xl_inst inst;
+   inst.inst.op = op;
+   inst.inst.bo = bo;
+   inst.inst.bi = bi;
+   inst.inst.unused = 0x0;
+   inst.inst.bh = bh;
+   inst.inst.op2 = op2;
+   inst.inst.lk = lk;
+   emit_instruction(p, inst.bits);
+}
+
+static INLINE void
+dump_xl(const char *name, uint inst)
+{
+   union xl_inst i;
+
+   i.bits = inst;
+   debug_printf("%s = 0x%08x\n", name, inst);
+   debug_printf(" op: %d 0x%x\n", i.inst.op, i.inst.op);
+   debug_printf(" bo: %d 0x%x\n", i.inst.bo, i.inst.bo);
+   debug_printf(" bi: %d 0x%x\n", i.inst.bi, i.inst.bi);
+   debug_printf(" unused: %d 0x%x\n", i.inst.unused, i.inst.unused);
+   debug_printf(" bh: %d 0x%x\n", i.inst.bh, i.inst.bh);
+   debug_printf(" op2: %d 0x%x\n", i.inst.op2, i.inst.op2);
+   debug_printf(" lk: %d 0x%x\n", i.inst.lk, i.inst.lk);
+}
+
+
+union x_inst {
+   uint32_t bits;
+   struct {
+      unsigned op:6;
+      unsigned vrs:5;
+      unsigned ra:5;
+      unsigned rb:5;
+      unsigned op2:10;
+      unsigned unused:1;
+   } inst;
+};
+
+static INLINE void
+emit_x(struct ppc_function *p, uint op, uint vrs, uint ra, uint rb, uint op2,
+       const char *format)
+{
+   union x_inst inst;
+   inst.inst.op = op;
+   inst.inst.vrs = vrs;
+   inst.inst.ra = ra;
+   inst.inst.rb = rb;
+   inst.inst.op2 = op2;
+   inst.inst.unused = 0x0;
+   emit_instruction(p, inst.bits);
+   if (p->print) {
+      indent(p);
+      printf(format, vrs, ra, rb);
+   }
+}
+
+
+union d_inst {
+   uint32_t bits;
+   struct {
+      unsigned op:6;
+      unsigned rt:5;
+      unsigned ra:5;
+      unsigned si:16;
+   } inst;
+};
+
+static INLINE void
+emit_d(struct ppc_function *p, uint op, uint rt, uint ra, int si,
+       const char *format, boolean transpose)
+{
+   union d_inst inst;
+   assert(si >= -32768);
+   assert(si <= 32767);
+   inst.inst.op = op;
+   inst.inst.rt = rt;
+   inst.inst.ra = ra;
+   inst.inst.si = (unsigned) (si & 0xffff);
+   emit_instruction(p, inst.bits);
+   if (p->print) {
+      indent(p);
+      if (transpose)
+         printf(format, rt, si, ra);
+      else
+         printf(format, rt, ra, si);
+   }
+}
+
+
+union a_inst {
+   uint32_t bits;
+   struct {
+      unsigned op:6;
+      unsigned frt:5;
+      unsigned fra:5;
+      unsigned frb:5;
+      unsigned unused:5;
+      unsigned op2:5;
+      unsigned rc:1;
+   } inst;
+};
+
+static INLINE void
+emit_a(struct ppc_function *p, uint op, uint frt, uint fra, uint frb, uint op2,
+       uint rc, const char *format)
+{
+   union a_inst inst;
+   inst.inst.op = op;
+   inst.inst.frt = frt;
+   inst.inst.fra = fra;
+   inst.inst.frb = frb;
+   inst.inst.unused = 0x0;
+   inst.inst.op2 = op2;
+   inst.inst.rc = rc;
+   emit_instruction(p, inst.bits);
+   if (p->print) {
+      indent(p);
+      printf(format, frt, fra, frb);
+   }
+}
+
+
+union xo_inst {
+   uint32_t bits;
+   struct {
+      unsigned op:6;
+      unsigned rt:5;
+      unsigned ra:5;
+      unsigned rb:5;
+      unsigned oe:1;
+      unsigned op2:9;
+      unsigned rc:1;
+   } inst;
+};
+
+static INLINE void
+emit_xo(struct ppc_function *p, uint op, uint rt, uint ra, uint rb, uint oe,
+        uint op2, uint rc, const char *format)
+{
+   union xo_inst inst;
+   inst.inst.op = op;
+   inst.inst.rt = rt;
+   inst.inst.ra = ra;
+   inst.inst.rb = rb;
+   inst.inst.oe = oe;
+   inst.inst.op2 = op2;
+   inst.inst.rc = rc;
+   emit_instruction(p, inst.bits);
+   if (p->print) {
+      indent(p);
+      printf(format, rt, ra, rb);
+   }
+}
+
+
+
+
+
+/**
+ ** float vector arithmetic
+ **/
+
+/** vector float add */
+void
+ppc_vaddfp(struct ppc_function *p, uint vD, uint vA, uint vB)
+{
+   emit_vx(p, 10, vD, vA, vB, "vaddfp\t%u, v%u, v%u\n", FALSE);
+}
+
+/** vector float substract */
+void
+ppc_vsubfp(struct ppc_function *p, uint vD, uint vA, uint vB)
+{
+   emit_vx(p, 74, vD, vA, vB, "vsubfp\tv%u, v%u, v%u\n", FALSE);
+}
+
+/** vector float min */
+void
+ppc_vminfp(struct ppc_function *p, uint vD, uint vA, uint vB)
+{
+   emit_vx(p, 1098, vD, vA, vB, "vminfp\tv%u, v%u, v%u\n", FALSE);
+}
+
+/** vector float max */
+void
+ppc_vmaxfp(struct ppc_function *p, uint vD, uint vA, uint vB)
+{
+   emit_vx(p, 1034, vD, vA, vB, "vmaxfp\tv%u, v%u, v%u\n", FALSE);
+}
+
+/** vector float mult add: vD = vA * vB + vC */
+void
+ppc_vmaddfp(struct ppc_function *p, uint vD, uint vA, uint vB, uint vC)
+{
+   /* note arg order */
+   emit_va(p, 46, vD, vA, vC, vB, "vmaddfp\tv%u, v%u, v%u, v%u\n");
+}
+
+/** vector float negative mult subtract: vD = vA - vB * vC */
+void
+ppc_vnmsubfp(struct ppc_function *p, uint vD, uint vA, uint vB, uint vC)
+{
+   /* note arg order */
+   emit_va(p, 47, vD, vB, vA, vC, "vnmsubfp\tv%u, v%u, v%u, v%u\n");
+}
+
+/** vector float compare greater than */
+void
+ppc_vcmpgtfpx(struct ppc_function *p, uint vD, uint vA, uint vB)
+{
+   emit_vxr(p, 710, vD, vA, vB, "vcmpgtfpx\tv%u, v%u, v%u");
+}
+
+/** vector float compare greater than or equal to */
+void
+ppc_vcmpgefpx(struct ppc_function *p, uint vD, uint vA, uint vB)
+{
+   emit_vxr(p, 454, vD, vA, vB, "vcmpgefpx\tv%u, v%u, v%u");
+}
+
+/** vector float compare equal */
+void
+ppc_vcmpeqfpx(struct ppc_function *p, uint vD, uint vA, uint vB)
+{
+   emit_vxr(p, 198, vD, vA, vB, "vcmpeqfpx\tv%u, v%u, v%u");
+}
+
+/** vector float 2^x */
+void
+ppc_vexptefp(struct ppc_function *p, uint vD, uint vB)
+{
+   emit_vx(p, 394, vD, 0, vB, "vexptefp\tv%u, 0%u, v%u\n", FALSE);
+}
+
+/** vector float log2(x) */
+void
+ppc_vlogefp(struct ppc_function *p, uint vD, uint vB)
+{
+   emit_vx(p, 458, vD, 0, vB, "vlogefp\tv%u, 0%u, v%u\n", FALSE);
+}
+
+/** vector float reciprocol */
+void
+ppc_vrefp(struct ppc_function *p, uint vD, uint vB)
+{
+   emit_vx(p, 266, vD, 0, vB, "vrefp\tv%u, 0%u, v%u\n", FALSE);
+}
+
+/** vector float reciprocol sqrt estimate */
+void
+ppc_vrsqrtefp(struct ppc_function *p, uint vD, uint vB)
+{
+   emit_vx(p, 330, vD, 0, vB, "vrsqrtefp\tv%u, 0%u, v%u\n", FALSE);
+}
+
+/** vector float round to negative infinity */
+void
+ppc_vrfim(struct ppc_function *p, uint vD, uint vB)
+{
+   emit_vx(p, 714, vD, 0, vB, "vrfim\tv%u, 0%u, v%u\n", FALSE);
+}
+
+/** vector float round to positive infinity */
+void
+ppc_vrfip(struct ppc_function *p, uint vD, uint vB)
+{
+   emit_vx(p, 650, vD, 0, vB, "vrfip\tv%u, 0%u, v%u\n", FALSE);
+}
+
+/** vector float round to nearest int */
+void
+ppc_vrfin(struct ppc_function *p, uint vD, uint vB)
+{
+   emit_vx(p, 522, vD, 0, vB, "vrfin\tv%u, 0%u, v%u\n", FALSE);
+}
+
+/** vector float round to int toward zero */
+void
+ppc_vrfiz(struct ppc_function *p, uint vD, uint vB)
+{
+   emit_vx(p, 586, vD, 0, vB, "vrfiz\tv%u, 0%u, v%u\n", FALSE);
+}
+
+/** vector store: store vR at mem[rA+rB] */
+void
+ppc_stvx(struct ppc_function *p, uint vR, uint rA, uint rB)
+{
+   emit_x(p, 31, vR, rA, rB, 231, "stvx\tv%u, r%u, r%u\n");
+}
+
+/** vector load: vR = mem[rA+rB] */
+void
+ppc_lvx(struct ppc_function *p, uint vR, uint rA, uint rB)
+{
+   emit_x(p, 31, vR, rA, rB, 103, "lvx\tv%u, r%u, r%u\n");
+}
+
+/** load vector element word: vR = mem_word[ra+rb] */
+void
+ppc_lvewx(struct ppc_function *p, uint vR, uint rA, uint rB)
+{
+   emit_x(p, 31, vR, rA, rB, 71, "lvewx\tv%u, r%u, r%u\n");
+}
+
+
+
+
+/**
+ ** vector bitwise operations
+ **/
+
+/** vector and */
+void
+ppc_vand(struct ppc_function *p, uint vD, uint vA, uint vB)
+{
+   emit_vx(p, 1028, vD, vA, vB, "vand\tv%u, v%u, v%u\n", FALSE);
+}
+
+/** vector and complement */
+void
+ppc_vandc(struct ppc_function *p, uint vD, uint vA, uint vB)
+{
+   emit_vx(p, 1092, vD, vA, vB, "vandc\tv%u, v%u, v%u\n", FALSE);
+}
+
+/** vector or */
+void
+ppc_vor(struct ppc_function *p, uint vD, uint vA, uint vB)
+{
+   emit_vx(p, 1156, vD, vA, vB, "vor\tv%u, v%u, v%u\n", FALSE);
+}
+
+/** vector nor */
+void
+ppc_vnor(struct ppc_function *p, uint vD, uint vA, uint vB)
+{
+   emit_vx(p, 1284, vD, vA, vB, "vnor\tv%u, v%u, v%u\n", FALSE);
+}
+
+/** vector xor */
+void
+ppc_vxor(struct ppc_function *p, uint vD, uint vA, uint vB)
+{
+   emit_vx(p, 1220, vD, vA, vB, "vxor\tv%u, v%u, v%u\n", FALSE);
+}
+
+/** Pseudo-instruction: vector move */
+void
+ppc_vmove(struct ppc_function *p, uint vD, uint vA)
+{
+   boolean print = p->print;
+   p->print = FALSE;
+   ppc_vor(p, vD, vA, vA);
+   if (print) {
+      indent(p);
+      printf("vor\tv%u, v%u, v%u \t# v%u = v%u\n", vD, vA, vA, vD, vA);
+   }
+   p->print = print;
+}
+
+/** Set vector register to {0,0,0,0} */
+void
+ppc_vzero(struct ppc_function *p, uint vr)
+{
+   boolean print = p->print;
+   p->print = FALSE;
+   ppc_vxor(p, vr, vr, vr);
+   if (print) {
+      indent(p);
+      printf("vxor\tv%u, v%u, v%u \t# v%u = {0,0,0,0}\n", vr, vr, vr, vr);
+   }
+   p->print = print;
+}
+
+
+
+
+/**
+ ** Vector shuffle / select / splat / etc
+ **/
+
+/** vector permute */
+void
+ppc_vperm(struct ppc_function *p, uint vD, uint vA, uint vB, uint vC)
+{
+   emit_va(p, 43, vD, vA, vB, vC, "vperm\tr%u, r%u, r%u, r%u");
+}
+
+/** vector select */
+void
+ppc_vsel(struct ppc_function *p, uint vD, uint vA, uint vB, uint vC)
+{
+   emit_va(p, 42, vD, vA, vB, vC, "vsel\tr%u, r%u, r%u, r%u");
+}
+
+/** vector splat byte */
+void
+ppc_vspltb(struct ppc_function *p, uint vD, uint vB, uint imm)
+{
+   emit_vx(p, 42, vD, imm, vB, "vspltb\tv%u, v%u, %u\n", TRUE);
+}
+
+/** vector splat half word */
+void
+ppc_vsplthw(struct ppc_function *p, uint vD, uint vB, uint imm)
+{
+   emit_vx(p, 588, vD, imm, vB, "vsplthw\tv%u, v%u, %u\n", TRUE);
+}
+
+/** vector splat word */
+void
+ppc_vspltw(struct ppc_function *p, uint vD, uint vB, uint imm)
+{
+   emit_vx(p, 652, vD, imm, vB, "vspltw\tv%u, v%u, %u\n", TRUE);
+}
+
+/** vector splat signed immediate word */
+void
+ppc_vspltisw(struct ppc_function *p, uint vD, int imm)
+{
+   assert(imm >= -16);
+   assert(imm < 15);
+   emit_vx(p, 908, vD, imm, 0, "vspltisw\tv%u, %d, %u\n", FALSE);
+}
+
+/** vector shift left word: vD[word] = vA[word] << (vB[word] & 0x1f) */
+void
+ppc_vslw(struct ppc_function *p, uint vD, uint vA, uint vB)
+{
+   emit_vx(p, 388, vD, vA, vB, "vslw\tv%u, v%u, v%u\n", FALSE);
+}
+
+
+
+
+/**
+ ** integer arithmetic
+ **/
+
+/** rt = ra + imm */
+void
+ppc_addi(struct ppc_function *p, uint rt, uint ra, int imm)
+{
+   emit_d(p, 14, rt, ra, imm, "addi\tr%u, r%u, %d\n", FALSE);
+}
+
+/** rt = ra + (imm << 16) */
+void
+ppc_addis(struct ppc_function *p, uint rt, uint ra, int imm)
+{
+   emit_d(p, 15, rt, ra, imm, "addis\tr%u, r%u, %d\n", FALSE);
+}
+
+/** rt = ra + rb */
+void
+ppc_add(struct ppc_function *p, uint rt, uint ra, uint rb)
+{
+   emit_xo(p, 31, rt, ra, rb, 0, 266, 0, "add\tr%u, r%u, r%u\n");
+}
+
+/** rt = ra AND ra */
+void
+ppc_and(struct ppc_function *p, uint rt, uint ra, uint rb)
+{
+   emit_x(p, 31, ra, rt, rb, 28, "and\tr%u, r%u, r%u\n");  /* note argument order */
+}
+
+/** rt = ra AND imm */
+void
+ppc_andi(struct ppc_function *p, uint rt, uint ra, int imm)
+{
+   /* note argument order */
+   emit_d(p, 28, ra, rt, imm, "andi\tr%u, r%u, %d\n", FALSE);
+}
+
+/** rt = ra OR ra */
+void
+ppc_or(struct ppc_function *p, uint rt, uint ra, uint rb)
+{
+   emit_x(p, 31, ra, rt, rb, 444, "or\tr%u, r%u, r%u\n");  /* note argument order */
+}
+
+/** rt = ra OR imm */
+void
+ppc_ori(struct ppc_function *p, uint rt, uint ra, int imm)
+{
+   /* note argument order */
+   emit_d(p, 24, ra, rt, imm, "ori\tr%u, r%u, %d\n", FALSE);
+}
+
+/** rt = ra XOR ra */
+void
+ppc_xor(struct ppc_function *p, uint rt, uint ra, uint rb)
+{
+   emit_x(p, 31, ra, rt, rb, 316, "xor\tr%u, r%u, r%u\n");  /* note argument order */
+}
+
+/** rt = ra XOR imm */
+void
+ppc_xori(struct ppc_function *p, uint rt, uint ra, int imm)
+{
+   /* note argument order */
+   emit_d(p, 26, ra, rt, imm, "xori\tr%u, r%u, %d\n", FALSE);
+}
+
+/** pseudo instruction: move: rt = ra */
+void
+ppc_mr(struct ppc_function *p, uint rt, uint ra)
+{
+   ppc_or(p, rt, ra, ra);
+}
+
+/** pseudo instruction: load immediate: rt = imm */
+void
+ppc_li(struct ppc_function *p, uint rt, int imm)
+{
+   boolean print = p->print;
+   p->print = FALSE;
+   ppc_addi(p, rt, 0, imm);
+   if (print) {
+      indent(p);
+      printf("addi\tr%u, r0, %d \t# r%u = %d\n", rt, imm, rt, imm);
+   }
+   p->print = print;
+}
+
+/** rt = imm << 16 */
+void
+ppc_lis(struct ppc_function *p, uint rt, int imm)
+{
+   ppc_addis(p, rt, 0, imm);
+}
+
+/** rt = imm */
+void
+ppc_load_int(struct ppc_function *p, uint rt, int imm)
+{
+   ppc_lis(p, rt, (imm >> 16));          /* rt = imm >> 16 */
+   ppc_ori(p, rt, rt, (imm & 0xffff));   /* rt = rt | (imm & 0xffff) */
+}
+
+
+
+
+/**
+ ** integer load/store
+ **/
+
+/** store rs at memory[(ra)+d],
+ * then update ra = (ra)+d
+ */
+void
+ppc_stwu(struct ppc_function *p, uint rs, uint ra, int d)
+{
+   emit_d(p, 37, rs, ra, d, "stwu\tr%u, %d(r%u)\n", TRUE);
+}
+
+/** store rs at memory[(ra)+d] */
+void
+ppc_stw(struct ppc_function *p, uint rs, uint ra, int d)
+{
+   emit_d(p, 36, rs, ra, d, "stw\tr%u, %d(r%u)\n", TRUE);
+}
+
+/** Load rt = mem[(ra)+d];  then zero set high 32 bits to zero. */
+void
+ppc_lwz(struct ppc_function *p, uint rt, uint ra, int d)
+{
+   emit_d(p, 32, rt, ra, d, "lwz\tr%u, %d(r%u)\n", TRUE);
+}
+
+
+
+/**
+ ** Float (non-vector) arithmetic
+ **/
+
+/** add: frt = fra + frb */
+void
+ppc_fadd(struct ppc_function *p, uint frt, uint fra, uint frb)
+{
+   emit_a(p, 63, frt, fra, frb, 21, 0, "fadd\tf%u, f%u, f%u\n");
+}
+
+/** sub: frt = fra - frb */
+void
+ppc_fsub(struct ppc_function *p, uint frt, uint fra, uint frb)
+{
+   emit_a(p, 63, frt, fra, frb, 20, 0, "fsub\tf%u, f%u, f%u\n");
+}
+
+/** convert to int: rt = (int) ra */
+void
+ppc_fctiwz(struct ppc_function *p, uint rt, uint fra)
+{
+   emit_x(p, 63, rt, 0, fra, 15, "fctiwz\tr%u, r%u, r%u\n");
+}
+
+/** store frs at mem[(ra)+offset] */
+void
+ppc_stfs(struct ppc_function *p, uint frs, uint ra, int offset)
+{
+   emit_d(p, 52, frs, ra, offset, "stfs\tr%u, %d(r%u)\n", TRUE);
+}
+
+/** store frs at mem[(ra)+(rb)] */
+void
+ppc_stfiwx(struct ppc_function *p, uint frs, uint ra, uint rb)
+{
+   emit_x(p, 31, frs, ra, rb, 983, "stfiwx\tr%u, r%u, r%u\n");
+}
+
+/** load frt = mem[(ra)+offset] */
+void
+ppc_lfs(struct ppc_function *p, uint frt, uint ra, int offset)
+{
+   emit_d(p, 48, frt, ra, offset, "stfs\tr%u, %d(r%u)\n", TRUE);
+}
+
+
+
+
+
+/**
+ ** branch instructions
+ **/
+
+/** BLR: Branch to link register (p. 35) */
+void
+ppc_blr(struct ppc_function *p)
+{
+   emit_i(p, 18, 0, 0, 1);
+   if (p->print) {
+      indent(p);
+      printf("blr\n");
+   }
+}
+
+/** Branch Conditional to Link Register (p. 36) */
+void
+ppc_bclr(struct ppc_function *p, uint condOp, uint branchHint, uint condReg)
+{
+   emit_xl(p, 19, condOp, condReg, branchHint, 16, 0);
+   if (p->print) {
+      indent(p);
+      printf("bclr\t%u %u %u\n", condOp, branchHint, condReg);
+   }
+}
+
+/** Pseudo instruction: return from subroutine */
+void
+ppc_return(struct ppc_function *p)
+{
+   ppc_bclr(p, BRANCH_COND_ALWAYS, BRANCH_HINT_SUB_RETURN, 0);
+}
diff --git a/src/gallium/auxiliary/rtasm/rtasm_ppc.h b/src/gallium/auxiliary/rtasm/rtasm_ppc.h
new file mode 100644
index 00000000000..93e5f5187de
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/rtasm_ppc.h
@@ -0,0 +1,342 @@
+/**************************************************************************
+ *
+ * Copyright (C) 2008 Tungsten Graphics, Inc.   All Rights Reserved.
+ * Copyright (C) 2009 VMware, Inc.  All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * PPC code generation.
+ * \author Brian Paul
+ */
+
+
+#ifndef RTASM_PPC_H
+#define RTASM_PPC_H
+
+
+#include "pipe/p_compiler.h"
+
+
+#define PPC_INST_SIZE 4  /**< 4 bytes / instruction */
+
+#define PPC_NUM_REGS 32
+#define PPC_NUM_FP_REGS 32
+#define PPC_NUM_VEC_REGS 32
+
+/** Stack pointer register */
+#define PPC_REG_SP 1
+
+/** Branch conditions */
+#define BRANCH_COND_ALWAYS       0x14  /* binary 1z1zz (z=ignored) */
+
+/** Branch hints */
+#define BRANCH_HINT_SUB_RETURN   0x0   /* binary 00 */
+
+
+struct ppc_function
+{
+   uint32_t *store;  /**< instruction buffer */
+   uint num_inst;
+   uint max_inst;
+   uint32_t reg_used;   /** used/free general-purpose registers bitmask */
+   uint32_t fp_used;   /** used/free floating point registers bitmask */
+   uint32_t vec_used;   /** used/free vector registers bitmask */
+   int indent;
+   boolean print;
+};
+
+
+
+extern void ppc_init_func(struct ppc_function *p);
+extern void ppc_release_func(struct ppc_function *p);
+extern uint ppc_num_instructions(const struct ppc_function *p);
+extern void (*ppc_get_func( struct ppc_function *p ))( void );
+extern void ppc_dump_func(const struct ppc_function *p);
+
+extern void ppc_print_code(struct ppc_function *p, boolean enable);
+extern void ppc_indent(struct ppc_function *p, int spaces);
+extern void ppc_comment(struct ppc_function *p, int rel_indent, const char *s);
+
+extern int ppc_reserve_register(struct ppc_function *p, int reg);
+extern int ppc_allocate_register(struct ppc_function *p);
+extern void ppc_release_register(struct ppc_function *p, int reg);
+extern int ppc_allocate_fp_register(struct ppc_function *p);
+extern void ppc_release_fp_register(struct ppc_function *p, int reg);
+extern int ppc_allocate_vec_register(struct ppc_function *p);
+extern void ppc_release_vec_register(struct ppc_function *p, int reg);
+
+
+
+/**
+ ** float vector arithmetic
+ **/
+
+/** vector float add */
+extern void
+ppc_vaddfp(struct ppc_function *p,uint vD, uint vA, uint vB);
+
+/** vector float substract */
+extern void
+ppc_vsubfp(struct ppc_function *p, uint vD, uint vA, uint vB);
+
+/** vector float min */
+extern void
+ppc_vminfp(struct ppc_function *p, uint vD, uint vA, uint vB);
+
+/** vector float max */
+extern void
+ppc_vmaxfp(struct ppc_function *p, uint vD, uint vA, uint vB);
+
+/** vector float mult add: vD = vA * vB + vC */
+extern void
+ppc_vmaddfp(struct ppc_function *p, uint vD, uint vA, uint vB, uint vC);
+
+/** vector float negative mult subtract: vD = vA - vB * vC */
+extern void
+ppc_vnmsubfp(struct ppc_function *p, uint vD, uint vA, uint vB, uint vC);
+
+/** vector float compare greater than */
+extern void
+ppc_vcmpgtfpx(struct ppc_function *p, uint vD, uint vA, uint vB);
+
+/** vector float compare greater than or equal to */
+extern void
+ppc_vcmpgefpx(struct ppc_function *p, uint vD, uint vA, uint vB);
+
+/** vector float compare equal */
+extern void
+ppc_vcmpeqfpx(struct ppc_function *p, uint vD, uint vA, uint vB);
+
+/** vector float 2^x */
+extern void
+ppc_vexptefp(struct ppc_function *p, uint vD, uint vB);
+
+/** vector float log2(x) */
+extern void
+ppc_vlogefp(struct ppc_function *p, uint vD, uint vB);
+
+/** vector float reciprocol */
+extern void
+ppc_vrefp(struct ppc_function *p, uint vD, uint vB);
+
+/** vector float reciprocol sqrt estimate */
+extern void
+ppc_vrsqrtefp(struct ppc_function *p, uint vD, uint vB);
+
+/** vector float round to negative infinity */
+extern void
+ppc_vrfim(struct ppc_function *p, uint vD, uint vB);
+
+/** vector float round to positive infinity */
+extern void
+ppc_vrfip(struct ppc_function *p, uint vD, uint vB);
+
+/** vector float round to nearest int */
+extern void
+ppc_vrfin(struct ppc_function *p, uint vD, uint vB);
+
+/** vector float round to int toward zero */
+extern void
+ppc_vrfiz(struct ppc_function *p, uint vD, uint vB);
+
+
+/** vector store: store vR at mem[vA+vB] */
+extern void
+ppc_stvx(struct ppc_function *p, uint vR, uint vA, uint vB);
+
+/** vector load: vR = mem[vA+vB] */
+extern void
+ppc_lvx(struct ppc_function *p, uint vR, uint vA, uint vB);
+
+/** load vector element word: vR = mem_word[vA+vB] */
+extern void
+ppc_lvewx(struct ppc_function *p, uint vR, uint vA, uint vB);
+
+
+
+/**
+ ** vector bitwise operations
+ **/
+
+
+/** vector and */
+extern void
+ppc_vand(struct ppc_function *p, uint vD, uint vA, uint vB);
+
+/** vector and complement */
+extern void
+ppc_vandc(struct ppc_function *p, uint vD, uint vA, uint vB);
+
+/** vector or */
+extern void
+ppc_vor(struct ppc_function *p, uint vD, uint vA, uint vB);
+
+/** vector nor */
+extern void
+ppc_vnor(struct ppc_function *p, uint vD, uint vA, uint vB);
+
+/** vector xor */
+extern void
+ppc_vxor(struct ppc_function *p, uint vD, uint vA, uint vB);
+
+/** Pseudo-instruction: vector move */
+extern void
+ppc_vmove(struct ppc_function *p, uint vD, uint vA);
+
+/** Set vector register to {0,0,0,0} */
+extern void
+ppc_vzero(struct ppc_function *p, uint vr);
+
+
+
+/**
+ ** Vector shuffle / select / splat / etc
+ **/
+
+/** vector permute */
+extern void
+ppc_vperm(struct ppc_function *p, uint vD, uint vA, uint vB, uint vC);
+
+/** vector select */
+extern void
+ppc_vsel(struct ppc_function *p, uint vD, uint vA, uint vB, uint vC);
+
+/** vector splat byte */
+extern void
+ppc_vspltb(struct ppc_function *p, uint vD, uint vB, uint imm);
+
+/** vector splat half word */
+extern void
+ppc_vsplthw(struct ppc_function *p, uint vD, uint vB, uint imm);
+
+/** vector splat word */
+extern void
+ppc_vspltw(struct ppc_function *p, uint vD, uint vB, uint imm);
+
+/** vector splat signed immediate word */
+extern void
+ppc_vspltisw(struct ppc_function *p, uint vD, int imm);
+
+/** vector shift left word: vD[word] = vA[word] << (vB[word] & 0x1f) */
+extern void
+ppc_vslw(struct ppc_function *p, uint vD, uint vA, uint vB);
+
+
+
+/**
+ ** scalar arithmetic
+ **/
+
+extern void
+ppc_add(struct ppc_function *p, uint rt, uint ra, uint rb);
+
+extern void
+ppc_addi(struct ppc_function *p, uint rt, uint ra, int imm);
+
+extern void
+ppc_addis(struct ppc_function *p, uint rt, uint ra, int imm);
+
+extern void
+ppc_and(struct ppc_function *p, uint rt, uint ra, uint rb);
+
+extern void
+ppc_andi(struct ppc_function *p, uint rt, uint ra, int imm);
+
+extern void
+ppc_or(struct ppc_function *p, uint rt, uint ra, uint rb);
+
+extern void
+ppc_ori(struct ppc_function *p, uint rt, uint ra, int imm);
+
+extern void
+ppc_xor(struct ppc_function *p, uint rt, uint ra, uint rb);
+
+extern void
+ppc_xori(struct ppc_function *p, uint rt, uint ra, int imm);
+
+extern void
+ppc_mr(struct ppc_function *p, uint rt, uint ra);
+
+extern void
+ppc_li(struct ppc_function *p, uint rt, int imm);
+
+extern void
+ppc_lis(struct ppc_function *p, uint rt, int imm);
+
+extern void
+ppc_load_int(struct ppc_function *p, uint rt, int imm);
+
+
+
+/**
+ ** scalar load/store
+ **/
+
+extern void
+ppc_stwu(struct ppc_function *p, uint rs, uint ra, int d);
+
+extern void
+ppc_stw(struct ppc_function *p, uint rs, uint ra, int d);
+
+extern void
+ppc_lwz(struct ppc_function *p, uint rs, uint ra, int d);
+
+
+
+/**
+ ** Float (non-vector) arithmetic
+ **/
+
+extern void
+ppc_fadd(struct ppc_function *p, uint frt, uint fra, uint frb);
+
+extern void
+ppc_fsub(struct ppc_function *p, uint frt, uint fra, uint frb);
+
+extern void
+ppc_fctiwz(struct ppc_function *p, uint rt, uint ra);
+
+extern void
+ppc_stfs(struct ppc_function *p, uint frs, uint ra, int offset);
+
+extern void
+ppc_stfiwx(struct ppc_function *p, uint frs, uint ra, uint rb);
+
+extern void
+ppc_lfs(struct ppc_function *p, uint frt, uint ra, int offset);
+
+
+
+/**
+ ** branch instructions
+ **/
+
+extern void
+ppc_blr(struct ppc_function *p);
+
+void
+ppc_bclr(struct ppc_function *p, uint condOp, uint branchHint, uint condReg);
+
+extern void
+ppc_return(struct ppc_function *p);
+
+
+#endif /* RTASM_PPC_H */
diff --git a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
new file mode 100644
index 00000000000..53a0e722cff
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.c
@@ -0,0 +1,1067 @@
+/*
+ * (C) Copyright IBM Corporation 2008
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * AUTHORS, COPYRIGHT HOLDERS, AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/**
+ * \file
+ * Real-time assembly generation interface for Cell B.E. SPEs.
+ *
+ * \author Ian Romanick <[email protected]>
+ * \author Brian Paul
+ */
+
+
+#include <stdio.h>
+#include "pipe/p_compiler.h"
+#include "util/u_memory.h"
+#include "rtasm_ppc_spe.h"
+
+
+#ifdef GALLIUM_CELL
+/**
+ * SPE instruction types
+ *
+ * There are 6 primary instruction encodings used on the Cell's SPEs.  Each of
+ * the following unions encodes one type.
+ *
+ * \bug
+ * If, at some point, we start generating SPE code from a little-endian host
+ * these unions will not work.
+ */
+/*@{*/
+/**
+ * Encode one output register with two input registers
+ */
+union spe_inst_RR {
+    uint32_t bits;
+    struct {
+	unsigned op:11;
+	unsigned rB:7;
+	unsigned rA:7;
+	unsigned rT:7;
+    } inst;
+};
+
+
+/**
+ * Encode one output register with three input registers
+ */
+union spe_inst_RRR {
+    uint32_t bits;
+    struct {
+	unsigned op:4;
+	unsigned rT:7;
+	unsigned rB:7;
+	unsigned rA:7;
+	unsigned rC:7;
+    } inst;
+};
+
+
+/**
+ * Encode one output register with one input reg. and a 7-bit signed immed
+ */
+union spe_inst_RI7 {
+    uint32_t bits;
+    struct {
+	unsigned op:11;
+	unsigned i7:7;
+	unsigned rA:7;
+	unsigned rT:7;
+    } inst;
+};
+
+
+/**
+ * Encode one output register with one input reg. and an 8-bit signed immed
+ */
+union spe_inst_RI8 {
+    uint32_t bits;
+    struct {
+	unsigned op:10;
+	unsigned i8:8;
+	unsigned rA:7;
+	unsigned rT:7;
+    } inst;
+};
+
+
+/**
+ * Encode one output register with one input reg. and a 10-bit signed immed
+ */
+union spe_inst_RI10 {
+    uint32_t bits;
+    struct {
+	unsigned op:8;
+	unsigned i10:10;
+	unsigned rA:7;
+	unsigned rT:7;
+    } inst;
+};
+
+
+/**
+ * Encode one output register with a 16-bit signed immediate
+ */
+union spe_inst_RI16 {
+    uint32_t bits;
+    struct {
+	unsigned op:9;
+	unsigned i16:16;
+	unsigned rT:7;
+    } inst;
+};
+
+
+/**
+ * Encode one output register with a 18-bit signed immediate
+ */
+union spe_inst_RI18 {
+    uint32_t bits;
+    struct {
+	unsigned op:7;
+	unsigned i18:18;
+	unsigned rT:7;
+    } inst;
+};
+/*@}*/
+
+
+static void
+indent(const struct spe_function *p)
+{
+   int i;
+   for (i = 0; i < p->indent; i++) {
+      putchar(' ');
+   }
+}
+
+
+static const char *
+rem_prefix(const char *longname)
+{
+   return longname + 4;
+}
+
+
+static const char *
+reg_name(int reg)
+{
+   switch (reg) {
+   case SPE_REG_SP:
+      return "$sp";
+   case SPE_REG_RA:
+      return "$lr";
+   default:
+      {
+         /* cycle through four buffers to handle multiple calls per printf */
+         static char buf[4][10];
+         static int b = 0;
+         b = (b + 1) % 4;
+         sprintf(buf[b], "$%d", reg);
+         return buf[b];
+      }
+   }
+}
+
+
+static void
+emit_instruction(struct spe_function *p, uint32_t inst_bits)
+{
+   if (!p->store)
+      return;  /* out of memory, drop the instruction */
+
+   if (p->num_inst == p->max_inst) {
+      /* allocate larger buffer */
+      uint32_t *newbuf;
+      p->max_inst *= 2;  /* 2x larger */
+      newbuf = align_malloc(p->max_inst * SPE_INST_SIZE, 16);
+      if (newbuf) {
+         memcpy(newbuf, p->store, p->num_inst * SPE_INST_SIZE);
+      }
+      align_free(p->store);
+      p->store = newbuf;
+      if (!p->store) {
+         /* out of memory */
+         p->num_inst = 0;
+         return;
+      }
+   }
+
+   p->store[p->num_inst++] = inst_bits;
+}
+
+
+
+static void emit_RR(struct spe_function *p, unsigned op, int rT,
+		    int rA, int rB, const char *name)
+{
+    union spe_inst_RR inst;
+    inst.inst.op = op;
+    inst.inst.rB = rB;
+    inst.inst.rA = rA;
+    inst.inst.rT = rT;
+    emit_instruction(p, inst.bits);
+    if (p->print) {
+       indent(p);
+       printf("%s\t%s, %s, %s\n",
+              rem_prefix(name), reg_name(rT), reg_name(rA), reg_name(rB));
+    }
+}
+
+
+static void emit_RRR(struct spe_function *p, unsigned op, int rT,
+                     int rA, int rB, int rC, const char *name)
+{
+    union spe_inst_RRR inst;
+    inst.inst.op = op;
+    inst.inst.rT = rT;
+    inst.inst.rB = rB;
+    inst.inst.rA = rA;
+    inst.inst.rC = rC;
+    emit_instruction(p, inst.bits);
+    if (p->print) {
+       indent(p);
+       printf("%s\t%s, %s, %s, %s\n", rem_prefix(name), reg_name(rT),
+              reg_name(rA), reg_name(rB), reg_name(rC));
+    }
+}
+
+
+static void emit_RI7(struct spe_function *p, unsigned op, int rT,
+		     int rA, int imm, const char *name)
+{
+    union spe_inst_RI7 inst;
+    inst.inst.op = op;
+    inst.inst.i7 = imm;
+    inst.inst.rA = rA;
+    inst.inst.rT = rT;
+    emit_instruction(p, inst.bits);
+    if (p->print) {
+       indent(p);
+       printf("%s\t%s, %s, 0x%x\n",
+              rem_prefix(name), reg_name(rT), reg_name(rA), imm);
+    }
+}
+
+
+
+static void emit_RI8(struct spe_function *p, unsigned op, int rT,
+		     int rA, int imm, const char *name)
+{
+    union spe_inst_RI8 inst;
+    inst.inst.op = op;
+    inst.inst.i8 = imm;
+    inst.inst.rA = rA;
+    inst.inst.rT = rT;
+    emit_instruction(p, inst.bits);
+    if (p->print) {
+       indent(p);
+       printf("%s\t%s, %s, 0x%x\n",
+              rem_prefix(name), reg_name(rT), reg_name(rA), imm);
+    }
+}
+
+
+
+static void emit_RI10(struct spe_function *p, unsigned op, int rT,
+		      int rA, int imm, const char *name)
+{
+    union spe_inst_RI10 inst;
+    inst.inst.op = op;
+    inst.inst.i10 = imm;
+    inst.inst.rA = rA;
+    inst.inst.rT = rT;
+    emit_instruction(p, inst.bits);
+    if (p->print) {
+       indent(p);
+       printf("%s\t%s, %s, 0x%x\n",
+              rem_prefix(name), reg_name(rT), reg_name(rA), imm);
+    }
+}
+
+
+/** As above, but do range checking on signed immediate value */
+static void emit_RI10s(struct spe_function *p, unsigned op, int rT,
+                       int rA, int imm, const char *name)
+{
+    assert(imm <= 511);
+    assert(imm >= -512);
+    emit_RI10(p, op, rT, rA, imm, name);
+}
+
+
+static void emit_RI16(struct spe_function *p, unsigned op, int rT,
+		      int imm, const char *name)
+{
+    union spe_inst_RI16 inst;
+    inst.inst.op = op;
+    inst.inst.i16 = imm;
+    inst.inst.rT = rT;
+    emit_instruction(p, inst.bits);
+    if (p->print) {
+       indent(p);
+       printf("%s\t%s, 0x%x\n", rem_prefix(name), reg_name(rT), imm);
+    }
+}
+
+
+static void emit_RI18(struct spe_function *p, unsigned op, int rT,
+		      int imm, const char *name)
+{
+    union spe_inst_RI18 inst;
+    inst.inst.op = op;
+    inst.inst.i18 = imm;
+    inst.inst.rT = rT;
+    emit_instruction(p, inst.bits);
+    if (p->print) {
+       indent(p);
+       printf("%s\t%s, 0x%x\n", rem_prefix(name), reg_name(rT), imm);
+    }
+}
+
+
+#define EMIT(_name, _op) \
+void _name (struct spe_function *p) \
+{ \
+   emit_RR(p, _op, 0, 0, 0, __FUNCTION__); \
+}
+
+#define EMIT_(_name, _op) \
+void _name (struct spe_function *p, int rT) \
+{ \
+   emit_RR(p, _op, rT, 0, 0, __FUNCTION__); \
+}
+
+#define EMIT_R(_name, _op) \
+void _name (struct spe_function *p, int rT, int rA) \
+{ \
+   emit_RR(p, _op, rT, rA, 0, __FUNCTION__);                 \
+}
+
+#define EMIT_RR(_name, _op) \
+void _name (struct spe_function *p, int rT, int rA, int rB) \
+{ \
+   emit_RR(p, _op, rT, rA, rB, __FUNCTION__);                \
+}
+
+#define EMIT_RRR(_name, _op) \
+void _name (struct spe_function *p, int rT, int rA, int rB, int rC) \
+{ \
+   emit_RRR(p, _op, rT, rA, rB, rC, __FUNCTION__);           \
+}
+
+#define EMIT_RI7(_name, _op) \
+void _name (struct spe_function *p, int rT, int rA, int imm) \
+{ \
+   emit_RI7(p, _op, rT, rA, imm, __FUNCTION__);              \
+}
+
+#define EMIT_RI8(_name, _op, bias) \
+void _name (struct spe_function *p, int rT, int rA, int imm) \
+{ \
+   emit_RI8(p, _op, rT, rA, bias - imm, __FUNCTION__);       \
+}
+
+#define EMIT_RI10(_name, _op) \
+void _name (struct spe_function *p, int rT, int rA, int imm) \
+{ \
+   emit_RI10(p, _op, rT, rA, imm, __FUNCTION__);             \
+}
+
+#define EMIT_RI10s(_name, _op) \
+void _name (struct spe_function *p, int rT, int rA, int imm) \
+{ \
+   emit_RI10s(p, _op, rT, rA, imm, __FUNCTION__);             \
+}
+
+#define EMIT_RI16(_name, _op) \
+void _name (struct spe_function *p, int rT, int imm) \
+{ \
+   emit_RI16(p, _op, rT, imm, __FUNCTION__);                 \
+}
+
+#define EMIT_RI18(_name, _op) \
+void _name (struct spe_function *p, int rT, int imm) \
+{ \
+   emit_RI18(p, _op, rT, imm, __FUNCTION__);                 \
+}
+
+#define EMIT_I16(_name, _op) \
+void _name (struct spe_function *p, int imm) \
+{ \
+   emit_RI16(p, _op, 0, imm, __FUNCTION__);                  \
+}
+
+#include "rtasm_ppc_spe.h"
+
+
+
+/**
+ * Initialize an spe_function.
+ * \param code_size  initial size of instruction buffer to allocate, in bytes.
+ *                   If zero, use a default.
+ */
+void spe_init_func(struct spe_function *p, unsigned code_size)
+{
+    uint i;
+
+    if (!code_size)
+       code_size = 64;
+
+    p->num_inst = 0;
+    p->max_inst = code_size / SPE_INST_SIZE;
+    p->store = align_malloc(code_size, 16);
+
+    p->set_count = 0;
+    memset(p->regs, 0, SPE_NUM_REGS * sizeof(p->regs[0]));
+
+    /* Conservatively treat R0 - R2 and R80 - R127 as non-volatile.
+     */
+    p->regs[0] = p->regs[1] = p->regs[2] = 1;
+    for (i = 80; i <= 127; i++) {
+      p->regs[i] = 1;
+    }
+
+    p->print = FALSE;
+    p->indent = 0;
+}
+
+
+void spe_release_func(struct spe_function *p)
+{
+    assert(p->num_inst <= p->max_inst);
+    if (p->store != NULL) {
+        align_free(p->store);
+    }
+    p->store = NULL;
+}
+
+
+/** Return current code size in bytes. */
+unsigned spe_code_size(const struct spe_function *p)
+{
+   return p->num_inst * SPE_INST_SIZE;
+}
+
+
+/**
+ * Allocate a SPE register.
+ * \return register index or -1 if none left.
+ */
+int spe_allocate_available_register(struct spe_function *p)
+{
+   unsigned i;
+   for (i = 0; i < SPE_NUM_REGS; i++) {
+      if (p->regs[i] == 0) {
+         p->regs[i] = 1;
+         return i;
+      }
+   }
+
+   return -1;
+}
+
+
+/**
+ * Mark the given SPE register as "allocated".
+ */
+int spe_allocate_register(struct spe_function *p, int reg)
+{
+   assert(reg < SPE_NUM_REGS);
+   assert(p->regs[reg] == 0);
+   p->regs[reg] = 1;
+   return reg;
+}
+
+
+/**
+ * Mark the given SPE register as "unallocated".  Note that this should
+ * only be used on registers allocated in the current register set; an
+ * assertion will fail if an attempt is made to deallocate a register
+ * allocated in an earlier register set.
+ */
+void spe_release_register(struct spe_function *p, int reg)
+{
+   assert(reg >= 0);
+   assert(reg < SPE_NUM_REGS);
+   assert(p->regs[reg] == 1);
+
+   p->regs[reg] = 0;
+}
+
+/**
+ * Start a new set of registers.  This can be called if
+ * it will be difficult later to determine exactly what
+ * registers were actually allocated during a code generation
+ * sequence, and you really just want to deallocate all of them.
+ */
+void spe_allocate_register_set(struct spe_function *p)
+{
+   uint i;
+
+   /* Keep track of the set count.  If it ever wraps around to 0, 
+    * we're in trouble.
+    */
+   p->set_count++;
+   assert(p->set_count > 0);
+
+   /* Increment the allocation count of all registers currently
+    * allocated.  Then any registers that are allocated in this set
+    * will be the only ones with a count of 1; they'll all be released
+    * when the register set is released.
+    */
+   for (i = 0; i < SPE_NUM_REGS; i++) {
+      if (p->regs[i] > 0)
+         p->regs[i]++;
+   }
+}
+
+void spe_release_register_set(struct spe_function *p)
+{
+   uint i;
+
+   /* If the set count drops below zero, we're in trouble. */
+   assert(p->set_count > 0);
+   p->set_count--;
+
+   /* Drop the allocation level of all registers.  Any allocated
+    * during this register set will drop to 0 and then become
+    * available.
+    */
+   for (i = 0; i < SPE_NUM_REGS; i++) {
+      if (p->regs[i] > 0)
+         p->regs[i]--;
+   }
+}
+
+
+unsigned
+spe_get_registers_used(const struct spe_function *p, ubyte used[])
+{
+   unsigned i, num = 0;
+   /* only count registers in the range available to callers */
+   for (i = 2; i < 80; i++) {
+      if (p->regs[i]) {
+         used[num++] = i;
+      }
+   }
+   return num;
+}
+
+
+void
+spe_print_code(struct spe_function *p, boolean enable)
+{
+   p->print = enable;
+}
+
+
+void
+spe_indent(struct spe_function *p, int spaces)
+{
+   p->indent += spaces;
+}
+
+
+void
+spe_comment(struct spe_function *p, int rel_indent, const char *s)
+{
+   if (p->print) {
+      p->indent += rel_indent;
+      indent(p);
+      p->indent -= rel_indent;
+      printf("# %s\n", s);
+   }
+}
+
+
+/**
+ * Load quad word.
+ * NOTE: offset is in bytes and the least significant 4 bits must be zero!
+ */
+void spe_lqd(struct spe_function *p, int rT, int rA, int offset)
+{
+   const boolean pSave = p->print;
+
+   /* offset must be a multiple of 16 */
+   assert(offset % 16 == 0);
+   /* offset must fit in 10-bit signed int field, after shifting */
+   assert((offset >> 4) <= 511);
+   assert((offset >> 4) >= -512);
+
+   p->print = FALSE;
+   emit_RI10(p, 0x034, rT, rA, offset >> 4, "spe_lqd");
+   p->print = pSave;
+
+   if (p->print) {
+      indent(p);
+      printf("lqd\t%s, %d(%s)\n", reg_name(rT), offset, reg_name(rA));
+   }
+}
+
+
+/**
+ * Store quad word.
+ * NOTE: offset is in bytes and the least significant 4 bits must be zero!
+ */
+void spe_stqd(struct spe_function *p, int rT, int rA, int offset)
+{
+   const boolean pSave = p->print;
+
+   /* offset must be a multiple of 16 */
+   assert(offset % 16 == 0);
+   /* offset must fit in 10-bit signed int field, after shifting */
+   assert((offset >> 4) <= 511);
+   assert((offset >> 4) >= -512);
+
+   p->print = FALSE;
+   emit_RI10(p, 0x024, rT, rA, offset >> 4, "spe_stqd");
+   p->print = pSave;
+
+   if (p->print) {
+      indent(p);
+      printf("stqd\t%s, %d(%s)\n", reg_name(rT), offset, reg_name(rA));
+   }
+}
+
+
+/**
+ * For branch instructions:
+ * \param d  if 1, disable interupts if branch is taken
+ * \param e  if 1, enable interupts if branch is taken
+ * If d and e are both zero, don't change interupt status (right?)
+ */
+
+/** Branch Indirect to address in rA */
+void spe_bi(struct spe_function *p, int rA, int d, int e)
+{
+   emit_RI7(p, 0x1a8, 0, rA, (d << 5) | (e << 4), __FUNCTION__);
+}
+
+/** Interupt Return */
+void spe_iret(struct spe_function *p, int rA, int d, int e)
+{
+   emit_RI7(p, 0x1aa, 0, rA, (d << 5) | (e << 4), __FUNCTION__);
+}
+
+/** Branch indirect and set link on external data */
+void spe_bisled(struct spe_function *p, int rT, int rA, int d,
+		int e)
+{
+   emit_RI7(p, 0x1ab, rT, rA, (d << 5) | (e << 4), __FUNCTION__);
+}
+
+/** Branch indirect and set link.  Save PC in rT, jump to rA. */
+void spe_bisl(struct spe_function *p, int rT, int rA, int d,
+		int e)
+{
+   emit_RI7(p, 0x1a9, rT, rA, (d << 5) | (e << 4), __FUNCTION__);
+}
+
+/** Branch indirect if zero word.  If rT.word[0]==0, jump to rA. */
+void spe_biz(struct spe_function *p, int rT, int rA, int d, int e)
+{
+   emit_RI7(p, 0x128, rT, rA, (d << 5) | (e << 4), __FUNCTION__);
+}
+
+/** Branch indirect if non-zero word.  If rT.word[0]!=0, jump to rA. */
+void spe_binz(struct spe_function *p, int rT, int rA, int d, int e)
+{
+   emit_RI7(p, 0x129, rT, rA, (d << 5) | (e << 4), __FUNCTION__);
+}
+
+/** Branch indirect if zero halfword.  If rT.halfword[1]==0, jump to rA. */
+void spe_bihz(struct spe_function *p, int rT, int rA, int d, int e)
+{
+   emit_RI7(p, 0x12a, rT, rA, (d << 5) | (e << 4), __FUNCTION__);
+}
+
+/** Branch indirect if non-zero halfword.  If rT.halfword[1]!=0, jump to rA. */
+void spe_bihnz(struct spe_function *p, int rT, int rA, int d, int e)
+{
+   emit_RI7(p, 0x12b, rT, rA, (d << 5) | (e << 4), __FUNCTION__);
+}
+
+
+/* Hint-for-branch instructions
+ */
+#if 0
+hbr;
+hbra;
+hbrr;
+#endif
+
+
+/* Control instructions
+ */
+#if 0
+stop;
+EMIT_RR  (spe_stopd, 0x140);
+EMIT_    (spe_nop,   0x201);
+sync;
+EMIT_    (spe_dsync, 0x003);
+EMIT_R   (spe_mfspr, 0x00c);
+EMIT_R   (spe_mtspr, 0x10c);
+#endif
+
+
+/**
+ ** Helper / "macro" instructions.
+ ** Use somewhat verbose names as a reminder that these aren't native
+ ** SPE instructions.
+ **/
+
+
+void
+spe_load_float(struct spe_function *p, int rT, float x)
+{
+   if (x == 0.0f) {
+      spe_il(p, rT, 0x0);
+   }
+   else if (x == 0.5f) {
+      spe_ilhu(p, rT, 0x3f00);
+   }
+   else if (x == 1.0f) {
+      spe_ilhu(p, rT, 0x3f80);
+   }
+   else if (x == -1.0f) {
+      spe_ilhu(p, rT, 0xbf80);
+   }
+   else {
+      union {
+         float f;
+         unsigned u;
+      } bits;
+      bits.f = x;
+      spe_ilhu(p, rT, bits.u >> 16);
+      spe_iohl(p, rT, bits.u & 0xffff);
+   }
+}
+
+
+void
+spe_load_int(struct spe_function *p, int rT, int i)
+{
+   if (-32768 <= i && i <= 32767) {
+      spe_il(p, rT, i);
+   }
+   else {
+      spe_ilhu(p, rT, i >> 16);
+      if (i & 0xffff)
+         spe_iohl(p, rT, i & 0xffff);
+   }
+}
+
+void spe_load_uint(struct spe_function *p, int rT, uint ui)
+{
+   /* If the whole value is in the lower 18 bits, use ila, which
+    * doesn't sign-extend.  Otherwise, if the two halfwords of
+    * the constant are identical, use ilh.  Otherwise, if every byte of
+    * the desired value is 0x00 or 0xff, we can use Form Select Mask for
+    * Bytes Immediate (fsmbi) to load the value in a single instruction.
+    * Otherwise, in the general case, we have to use ilhu followed by iohl.
+    */
+   if ((ui & 0x0003ffff) == ui) {
+      spe_ila(p, rT, ui);
+   }
+   else if ((ui >> 16) == (ui & 0xffff)) {
+      spe_ilh(p, rT, ui & 0xffff);
+   }
+   else if (
+      ((ui & 0x000000ff) == 0 || (ui & 0x000000ff) == 0x000000ff) &&
+      ((ui & 0x0000ff00) == 0 || (ui & 0x0000ff00) == 0x0000ff00) &&
+      ((ui & 0x00ff0000) == 0 || (ui & 0x00ff0000) == 0x00ff0000) &&
+      ((ui & 0xff000000) == 0 || (ui & 0xff000000) == 0xff000000)
+   ) {
+      uint mask = 0;
+      /* fsmbi duplicates each bit in the given mask eight times,
+       * using a 16-bit value to initialize a 16-byte quadword.
+       * Each 4-bit nybble of the mask corresponds to a full word
+       * of the result; look at the value and figure out the mask
+       * (replicated for each word in the quadword), and then
+       * form the "select mask" to get the value.
+       */
+      if ((ui & 0x000000ff) == 0x000000ff) mask |= 0x1111;
+      if ((ui & 0x0000ff00) == 0x0000ff00) mask |= 0x2222;
+      if ((ui & 0x00ff0000) == 0x00ff0000) mask |= 0x4444;
+      if ((ui & 0xff000000) == 0xff000000) mask |= 0x8888;
+      spe_fsmbi(p, rT, mask);
+   }
+   else {
+      /* The general case: this usually uses two instructions, but
+       * may use only one if the low-order 16 bits of each word are 0.
+       */
+      spe_ilhu(p, rT, ui >> 16);
+      if (ui & 0xffff)
+         spe_iohl(p, rT, ui & 0xffff);
+   }
+}
+
+/**
+ * This function is constructed identically to spe_xor_uint() below.
+ * Changes to one should be made in the other.
+ */
+void
+spe_and_uint(struct spe_function *p, int rT, int rA, uint ui)
+{
+   /* If we can, emit a single instruction, either And Byte Immediate
+    * (which uses the same constant across each byte), And Halfword Immediate
+    * (which sign-extends a 10-bit immediate to 16 bits and uses that
+    * across each halfword), or And Word Immediate (which sign-extends
+    * a 10-bit immediate to 32 bits).
+    *
+    * Otherwise, we'll need to use a temporary register.
+    */
+   uint tmp;
+
+   /* If the upper 23 bits are all 0s or all 1s, sign extension
+    * will work and we can use And Word Immediate
+    */
+   tmp = ui & 0xfffffe00;
+   if (tmp == 0xfffffe00 || tmp  == 0) {
+      spe_andi(p, rT, rA, ui & 0x000003ff);
+      return;
+   }
+   
+   /* If the ui field is symmetric along halfword boundaries and
+    * the upper 7 bits of each halfword are all 0s or 1s, we
+    * can use And Halfword Immediate
+    */
+   tmp = ui & 0xfe00fe00;
+   if ((tmp == 0xfe00fe00 || tmp == 0) && ((ui >> 16) == (ui & 0x0000ffff))) {
+      spe_andhi(p, rT, rA, ui & 0x000003ff);
+      return;
+   }
+
+   /* If the ui field is symmetric in each byte, then we can use
+    * the And Byte Immediate instruction.
+    */
+   tmp = ui & 0x000000ff;
+   if ((ui >> 24) == tmp && ((ui >> 16) & 0xff) == tmp && ((ui >> 8) & 0xff) == tmp) {
+      spe_andbi(p, rT, rA, tmp);
+      return;
+   }
+
+   /* Otherwise, we'll have to use a temporary register. */
+   int tmp_reg = spe_allocate_available_register(p);
+   spe_load_uint(p, tmp_reg, ui);
+   spe_and(p, rT, rA, tmp_reg);
+   spe_release_register(p, tmp_reg);
+}
+
+
+/**
+ * This function is constructed identically to spe_and_uint() above.
+ * Changes to one should be made in the other.
+ */
+void
+spe_xor_uint(struct spe_function *p, int rT, int rA, uint ui)
+{
+   /* If we can, emit a single instruction, either Exclusive Or Byte 
+    * Immediate (which uses the same constant across each byte), Exclusive 
+    * Or Halfword Immediate (which sign-extends a 10-bit immediate to 
+    * 16 bits and uses that across each halfword), or Exclusive Or Word 
+    * Immediate (which sign-extends a 10-bit immediate to 32 bits).
+    *
+    * Otherwise, we'll need to use a temporary register.
+    */
+   uint tmp;
+
+   /* If the upper 23 bits are all 0s or all 1s, sign extension
+    * will work and we can use Exclusive Or Word Immediate
+    */
+   tmp = ui & 0xfffffe00;
+   if (tmp == 0xfffffe00 || tmp  == 0) {
+      spe_xori(p, rT, rA, ui & 0x000003ff);
+      return;
+   }
+   
+   /* If the ui field is symmetric along halfword boundaries and
+    * the upper 7 bits of each halfword are all 0s or 1s, we
+    * can use Exclusive Or Halfword Immediate
+    */
+   tmp = ui & 0xfe00fe00;
+   if ((tmp == 0xfe00fe00 || tmp == 0) && ((ui >> 16) == (ui & 0x0000ffff))) {
+      spe_xorhi(p, rT, rA, ui & 0x000003ff);
+      return;
+   }
+
+   /* If the ui field is symmetric in each byte, then we can use
+    * the Exclusive Or Byte Immediate instruction.
+    */
+   tmp = ui & 0x000000ff;
+   if ((ui >> 24) == tmp && ((ui >> 16) & 0xff) == tmp && ((ui >> 8) & 0xff) == tmp) {
+      spe_xorbi(p, rT, rA, tmp);
+      return;
+   }
+
+   /* Otherwise, we'll have to use a temporary register. */
+   int tmp_reg = spe_allocate_available_register(p);
+   spe_load_uint(p, tmp_reg, ui);
+   spe_xor(p, rT, rA, tmp_reg);
+   spe_release_register(p, tmp_reg);
+}
+
+void
+spe_compare_equal_uint(struct spe_function *p, int rT, int rA, uint ui)
+{
+   /* If the comparison value is 9 bits or less, it fits inside a
+    * Compare Equal Word Immediate instruction.
+    */
+   if ((ui & 0x000001ff) == ui) {
+      spe_ceqi(p, rT, rA, ui);
+   }
+   /* Otherwise, we're going to have to load a word first. */
+   else {
+      int tmp_reg = spe_allocate_available_register(p);
+      spe_load_uint(p, tmp_reg, ui);
+      spe_ceq(p, rT, rA, tmp_reg);
+      spe_release_register(p, tmp_reg);
+   }
+}
+
+void
+spe_compare_greater_uint(struct spe_function *p, int rT, int rA, uint ui)
+{
+   /* If the comparison value is 10 bits or less, it fits inside a
+    * Compare Logical Greater Than Word Immediate instruction.
+    */
+   if ((ui & 0x000003ff) == ui) {
+      spe_clgti(p, rT, rA, ui);
+   }
+   /* Otherwise, we're going to have to load a word first. */
+   else {
+      int tmp_reg = spe_allocate_available_register(p);
+      spe_load_uint(p, tmp_reg, ui);
+      spe_clgt(p, rT, rA, tmp_reg);
+      spe_release_register(p, tmp_reg);
+   }
+}
+
+void
+spe_splat(struct spe_function *p, int rT, int rA)
+{
+   /* Use a temporary, just in case rT == rA */
+   int tmp_reg = spe_allocate_available_register(p);
+   /* Duplicate bytes 0, 1, 2, and 3 across the whole register */
+   spe_ila(p, tmp_reg, 0x00010203);
+   spe_shufb(p, rT, rA, rA, tmp_reg);
+   spe_release_register(p, tmp_reg);
+}
+
+
+void
+spe_complement(struct spe_function *p, int rT, int rA)
+{
+   spe_nor(p, rT, rA, rA);
+}
+
+
+void
+spe_move(struct spe_function *p, int rT, int rA)
+{
+   /* Use different instructions depending on the instruction address
+    * to take advantage of the dual pipelines.
+    */
+   if (p->num_inst & 1)
+      spe_shlqbyi(p, rT, rA, 0);  /* odd pipe */
+   else
+      spe_ori(p, rT, rA, 0);  /* even pipe */
+}
+
+
+void
+spe_zero(struct spe_function *p, int rT)
+{
+   spe_xor(p, rT, rT, rT);
+}
+
+
+void
+spe_splat_word(struct spe_function *p, int rT, int rA, int word)
+{
+   assert(word >= 0);
+   assert(word <= 3);
+
+   if (word == 0) {
+      int tmp1 = rT;
+      spe_ila(p, tmp1, 66051);
+      spe_shufb(p, rT, rA, rA, tmp1);
+   }
+   else {
+      /* XXX review this, we may not need the rotqbyi instruction */
+      int tmp1 = rT;
+      int tmp2 = spe_allocate_available_register(p);
+
+      spe_ila(p, tmp1, 66051);
+      spe_rotqbyi(p, tmp2, rA, 4 * word);
+      spe_shufb(p, rT, tmp2, tmp2, tmp1);
+
+      spe_release_register(p, tmp2);
+   }
+}
+
+/**
+ * For each 32-bit float element of rA and rB, choose the smaller of the
+ * two, compositing them into the rT register.
+ * 
+ * The Float Compare Greater Than (fcgt) instruction will put 1s into
+ * compare_reg where rA > rB, and 0s where rA <= rB.
+ *
+ * Then the Select Bits (selb) instruction will take bits from rA where
+ * compare_reg is 0, and from rB where compare_reg is 1; i.e., from rA
+ * where rA <= rB and from rB where rB > rA, which is exactly the
+ * "min" operation.
+ *
+ * The compare_reg could in many cases be the same as rT, unless
+ * rT == rA || rt == rB.  But since this is common in constructions
+ * like "x = min(x, a)", we always allocate a new register to be safe.
+ */
+void 
+spe_float_min(struct spe_function *p, int rT, int rA, int rB)
+{
+   int compare_reg = spe_allocate_available_register(p);
+   spe_fcgt(p, compare_reg, rA, rB);
+   spe_selb(p, rT, rA, rB, compare_reg);
+   spe_release_register(p, compare_reg);
+}
+
+/**
+ * For each 32-bit float element of rA and rB, choose the greater of the
+ * two, compositing them into the rT register.
+ * 
+ * The logic is similar to that of spe_float_min() above; the only
+ * difference is that the registers on spe_selb() have been reversed,
+ * so that the larger of the two is selected instead of the smaller.
+ */
+void 
+spe_float_max(struct spe_function *p, int rT, int rA, int rB)
+{
+   int compare_reg = spe_allocate_available_register(p);
+   spe_fcgt(p, compare_reg, rA, rB);
+   spe_selb(p, rT, rB, rA, compare_reg);
+   spe_release_register(p, compare_reg);
+}
+
+#endif /* GALLIUM_CELL */
diff --git a/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h
new file mode 100644
index 00000000000..65d9c774154
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/rtasm_ppc_spe.h
@@ -0,0 +1,433 @@
+/*
+ * (C) Copyright IBM Corporation 2008
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * AUTHORS, COPYRIGHT HOLDERS, AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/**
+ * \file
+ * Real-time assembly generation interface for Cell B.E. SPEs.
+ * For details, see /opt/cell/sdk/docs/arch/SPU_ISA_v1.2_27Jan2007_pub.pdf
+ *
+ * \author Ian Romanick <[email protected]>
+ * \author Brian Paul
+ */
+
+#ifndef RTASM_PPC_SPE_H
+#define RTASM_PPC_SPE_H
+
+/** 4 bytes per instruction */
+#define SPE_INST_SIZE 4
+
+/** number of general-purpose SIMD registers */
+#define SPE_NUM_REGS  128
+
+/** Return Address register (aka $lr / Link Register) */
+#define SPE_REG_RA  0
+
+/** Stack Pointer register (aka $sp) */
+#define SPE_REG_SP  1
+
+
+struct spe_function
+{
+   uint32_t *store;  /**< instruction buffer */
+   uint num_inst;
+   uint max_inst;
+
+   /**
+    * The "set count" reflects the number of nested register sets
+    * are allowed.  In the unlikely case that we exceed the set count,
+    * register allocation will start to be confused, which is critical
+    * enough that we check for it.
+    */
+   unsigned char set_count;
+
+   /** 
+    * Flags for used and unused registers.  Each byte corresponds to a
+    * register; a 0 in that byte means that the register is available.
+    * A value of 1 means that the register was allocated in the current
+    * register set.  Any other value N means that the register was allocated
+    * N register sets ago.
+     *
+     * \sa
+     * spe_allocate_register, spe_allocate_available_register,
+     * spe_allocate_register_set, spe_release_register_set, spe_release_register, 
+     */
+    unsigned char regs[SPE_NUM_REGS];
+
+    boolean print; /**< print/dump instructions as they're emitted? */
+    int indent;    /**< number of spaces to indent */
+};
+
+
+extern void spe_init_func(struct spe_function *p, uint code_size);
+extern void spe_release_func(struct spe_function *p);
+extern uint spe_code_size(const struct spe_function *p);
+
+extern int spe_allocate_available_register(struct spe_function *p);
+extern int spe_allocate_register(struct spe_function *p, int reg);
+extern void spe_release_register(struct spe_function *p, int reg);
+extern void spe_allocate_register_set(struct spe_function *p);
+extern void spe_release_register_set(struct spe_function *p);
+
+extern uint spe_get_registers_used(const struct spe_function *p, ubyte used[]);
+
+extern void spe_print_code(struct spe_function *p, boolean enable);
+extern void spe_indent(struct spe_function *p, int spaces);
+extern void spe_comment(struct spe_function *p, int rel_indent, const char *s);
+
+
+#endif /* RTASM_PPC_SPE_H */
+
+#ifndef EMIT
+#define EMIT(_name, _op) \
+    extern void _name (struct spe_function *p);
+#define EMIT_(_name, _op) \
+    extern void _name (struct spe_function *p, int rT);
+#define EMIT_R(_name, _op) \
+    extern void _name (struct spe_function *p, int rT, int rA);
+#define EMIT_RR(_name, _op) \
+    extern void _name (struct spe_function *p, int rT, int rA, int rB);
+#define EMIT_RRR(_name, _op) \
+    extern void _name (struct spe_function *p, int rT, int rA, int rB, int rC);
+#define EMIT_RI7(_name, _op) \
+    extern void _name (struct spe_function *p, int rT, int rA, int imm);
+#define EMIT_RI8(_name, _op, bias) \
+    extern void _name (struct spe_function *p, int rT, int rA, int imm);
+#define EMIT_RI10(_name, _op) \
+    extern void _name (struct spe_function *p, int rT, int rA, int imm);
+#define EMIT_RI10s(_name, _op) \
+    extern void _name (struct spe_function *p, int rT, int rA, int imm);
+#define EMIT_RI16(_name, _op) \
+    extern void _name (struct spe_function *p, int rT, int imm);
+#define EMIT_RI18(_name, _op) \
+    extern void _name (struct spe_function *p, int rT, int imm);
+#define EMIT_I16(_name, _op) \
+    extern void _name (struct spe_function *p, int imm);
+#define UNDEF_EMIT_MACROS
+#endif /* EMIT */
+
+
+/* Memory load / store instructions
+ */
+EMIT_RR  (spe_lqx,  0x1c4)
+EMIT_RI16(spe_lqa,  0x061)
+EMIT_RI16(spe_lqr,  0x067)
+EMIT_RR  (spe_stqx, 0x144)
+EMIT_RI16(spe_stqa, 0x041)
+EMIT_RI16(spe_stqr, 0x047)
+EMIT_RI7 (spe_cbd,  0x1f4)
+EMIT_RR  (spe_cbx,  0x1d4)
+EMIT_RI7 (spe_chd,  0x1f5)
+EMIT_RI7 (spe_chx,  0x1d5)
+EMIT_RI7 (spe_cwd,  0x1f6)
+EMIT_RI7 (spe_cwx,  0x1d6)
+EMIT_RI7 (spe_cdd,  0x1f7)
+EMIT_RI7 (spe_cdx,  0x1d7)
+
+
+/* Constant formation instructions
+ */
+EMIT_RI16(spe_ilh,   0x083)
+EMIT_RI16(spe_ilhu,  0x082)
+EMIT_RI16(spe_il,    0x081)
+EMIT_RI18(spe_ila,   0x021)
+EMIT_RI16(spe_iohl,  0x0c1)
+EMIT_RI16(spe_fsmbi, 0x065)
+
+
+
+/* Integer and logical instructions
+ */
+EMIT_RR  (spe_ah,      0x0c8)
+EMIT_RI10(spe_ahi,     0x01d)
+EMIT_RR  (spe_a,       0x0c0)
+EMIT_RI10s(spe_ai,      0x01c)
+EMIT_RR  (spe_sfh,     0x048)
+EMIT_RI10(spe_sfhi,    0x00d)
+EMIT_RR  (spe_sf,      0x040)
+EMIT_RI10(spe_sfi,     0x00c)
+EMIT_RR  (spe_addx,    0x340)
+EMIT_RR  (spe_cg,      0x0c2)
+EMIT_RR  (spe_cgx,     0x342)
+EMIT_RR  (spe_sfx,     0x341)
+EMIT_RR  (spe_bg,      0x042)
+EMIT_RR  (spe_bgx,     0x343)
+EMIT_RR  (spe_mpy,     0x3c4)
+EMIT_RR  (spe_mpyu,    0x3cc)
+EMIT_RI10(spe_mpyi,    0x074)
+EMIT_RI10(spe_mpyui,   0x075)
+EMIT_RRR (spe_mpya,    0x00c)
+EMIT_RR  (spe_mpyh,    0x3c5)
+EMIT_RR  (spe_mpys,    0x3c7)
+EMIT_RR  (spe_mpyhh,   0x3c6)
+EMIT_RR  (spe_mpyhha,  0x346)
+EMIT_RR  (spe_mpyhhu,  0x3ce)
+EMIT_RR  (spe_mpyhhau, 0x34e)
+EMIT_R   (spe_clz,     0x2a5)
+EMIT_R   (spe_cntb,    0x2b4)
+EMIT_R   (spe_fsmb,    0x1b6)
+EMIT_R   (spe_fsmh,    0x1b5)
+EMIT_R   (spe_fsm,     0x1b4)
+EMIT_R   (spe_gbb,     0x1b2)
+EMIT_R   (spe_gbh,     0x1b1)
+EMIT_R   (spe_gb,      0x1b0)
+EMIT_RR  (spe_avgb,    0x0d3)
+EMIT_RR  (spe_absdb,   0x053)
+EMIT_RR  (spe_sumb,    0x253)
+EMIT_R   (spe_xsbh,    0x2b6)
+EMIT_R   (spe_xshw,    0x2ae)
+EMIT_R   (spe_xswd,    0x2a6)
+EMIT_RR  (spe_and,     0x0c1)
+EMIT_RR  (spe_andc,    0x2c1)
+EMIT_RI10s(spe_andbi,   0x016)
+EMIT_RI10s(spe_andhi,   0x015)
+EMIT_RI10s(spe_andi,    0x014)
+EMIT_RR  (spe_or,      0x041)
+EMIT_RR  (spe_orc,     0x2c9)
+EMIT_RI10s(spe_orbi,    0x006)
+EMIT_RI10s(spe_orhi,    0x005)
+EMIT_RI10s(spe_ori,     0x004)
+EMIT_R   (spe_orx,     0x1f0)
+EMIT_RR  (spe_xor,     0x241)
+EMIT_RI10s(spe_xorbi,   0x046)
+EMIT_RI10s(spe_xorhi,   0x045)
+EMIT_RI10s(spe_xori,    0x044)
+EMIT_RR  (spe_nand,    0x0c9)
+EMIT_RR  (spe_nor,     0x049)
+EMIT_RR  (spe_eqv,     0x249)
+EMIT_RRR (spe_selb,    0x008)
+EMIT_RRR (spe_shufb,   0x00b)
+
+
+/* Shift and rotate instructions
+ */
+EMIT_RR  (spe_shlh,      0x05f)
+EMIT_RI7 (spe_shlhi,     0x07f)
+EMIT_RR  (spe_shl,       0x05b)
+EMIT_RI7 (spe_shli,      0x07b)
+EMIT_RR  (spe_shlqbi,    0x1db)
+EMIT_RI7 (spe_shlqbii,   0x1fb)
+EMIT_RR  (spe_shlqby,    0x1df)
+EMIT_RI7 (spe_shlqbyi,   0x1ff)
+EMIT_RR  (spe_shlqbybi,  0x1cf)
+EMIT_RR  (spe_roth,      0x05c)
+EMIT_RI7 (spe_rothi,     0x07c)
+EMIT_RR  (spe_rot,       0x058)
+EMIT_RI7 (spe_roti,      0x078)
+EMIT_RR  (spe_rotqby,    0x1dc)
+EMIT_RI7 (spe_rotqbyi,   0x1fc)
+EMIT_RR  (spe_rotqbybi,  0x1cc)
+EMIT_RR  (spe_rotqbi,    0x1d8)
+EMIT_RI7 (spe_rotqbii,   0x1f8)
+EMIT_RR  (spe_rothm,     0x05d)
+EMIT_RI7 (spe_rothmi,    0x07d)
+EMIT_RR  (spe_rotm,      0x059)
+EMIT_RI7 (spe_rotmi,     0x079)
+EMIT_RR  (spe_rotqmby,   0x1dd)
+EMIT_RI7 (spe_rotqmbyi,  0x1fd)
+EMIT_RR  (spe_rotqmbybi, 0x1cd)
+EMIT_RR  (spe_rotqmbi,   0x1c9)
+EMIT_RI7 (spe_rotqmbii,  0x1f9)
+EMIT_RR  (spe_rotmah,    0x05e)
+EMIT_RI7 (spe_rotmahi,   0x07e)
+EMIT_RR  (spe_rotma,     0x05a)
+EMIT_RI7 (spe_rotmai,    0x07a)
+
+
+/* Compare, branch, and halt instructions
+ */
+EMIT_RR  (spe_heq,       0x3d8)
+EMIT_RI10(spe_heqi,      0x07f)
+EMIT_RR  (spe_hgt,       0x258)
+EMIT_RI10(spe_hgti,      0x04f)
+EMIT_RR  (spe_hlgt,      0x2d8)
+EMIT_RI10(spe_hlgti,     0x05f)
+EMIT_RR  (spe_ceqb,      0x3d0)
+EMIT_RI10(spe_ceqbi,     0x07e)
+EMIT_RR  (spe_ceqh,      0x3c8)
+EMIT_RI10(spe_ceqhi,     0x07d)
+EMIT_RR  (spe_ceq,       0x3c0)
+EMIT_RI10(spe_ceqi,      0x07c)
+EMIT_RR  (spe_cgtb,      0x250)
+EMIT_RI10(spe_cgtbi,     0x04e)
+EMIT_RR  (spe_cgth,      0x248)
+EMIT_RI10(spe_cgthi,     0x04d)
+EMIT_RR  (spe_cgt,       0x240)
+EMIT_RI10(spe_cgti,      0x04c)
+EMIT_RR  (spe_clgtb,     0x2d0)
+EMIT_RI10(spe_clgtbi,    0x05e)
+EMIT_RR  (spe_clgth,     0x2c8)
+EMIT_RI10(spe_clgthi,    0x05d)
+EMIT_RR  (spe_clgt,      0x2c0)
+EMIT_RI10(spe_clgti,     0x05c)
+EMIT_I16 (spe_br,        0x064)
+EMIT_I16 (spe_bra,       0x060)
+EMIT_RI16(spe_brsl,      0x066)
+EMIT_RI16(spe_brasl,     0x062)
+EMIT_RI16(spe_brnz,      0x042)
+EMIT_RI16(spe_brz,       0x040)
+EMIT_RI16(spe_brhnz,     0x046)
+EMIT_RI16(spe_brhz,      0x044)
+
+/* Control instructions
+ */
+EMIT     (spe_lnop,      0x001)
+
+extern void
+spe_lqd(struct spe_function *p, int rT, int rA, int offset);
+
+extern void
+spe_stqd(struct spe_function *p, int rT, int rA, int offset);
+
+extern void spe_bi(struct spe_function *p, int rA, int d, int e);
+extern void spe_iret(struct spe_function *p, int rA, int d, int e);
+extern void spe_bisled(struct spe_function *p, int rT, int rA,
+    int d, int e);
+extern void spe_bisl(struct spe_function *p, int rT, int rA,
+    int d, int e);
+extern void spe_biz(struct spe_function *p, int rT, int rA,
+    int d, int e);
+extern void spe_binz(struct spe_function *p, int rT, int rA,
+    int d, int e);
+extern void spe_bihz(struct spe_function *p, int rT, int rA,
+    int d, int e);
+extern void spe_bihnz(struct spe_function *p, int rT, int rA,
+    int d, int e);
+
+
+/** Load/splat immediate float into rT. */
+extern void
+spe_load_float(struct spe_function *p, int rT, float x);
+
+/** Load/splat immediate int into rT. */
+extern void
+spe_load_int(struct spe_function *p, int rT, int i);
+
+/** Load/splat immediate unsigned int into rT. */
+extern void
+spe_load_uint(struct spe_function *p, int rT, uint ui);
+
+/** And immediate value into rT. */
+extern void
+spe_and_uint(struct spe_function *p, int rT, int rA, uint ui);
+
+/** Xor immediate value into rT. */
+extern void
+spe_xor_uint(struct spe_function *p, int rT, int rA, uint ui);
+
+/** Compare equal with immediate value. */
+extern void
+spe_compare_equal_uint(struct spe_function *p, int rT, int rA, uint ui);
+
+/** Compare greater with immediate value. */
+extern void
+spe_compare_greater_uint(struct spe_function *p, int rT, int rA, uint ui);
+
+/** Replicate word 0 of rA across rT. */
+extern void
+spe_splat(struct spe_function *p, int rT, int rA);
+
+/** rT = complement_all_bits(rA). */
+extern void
+spe_complement(struct spe_function *p, int rT, int rA);
+
+/** rT = rA. */
+extern void
+spe_move(struct spe_function *p, int rT, int rA);
+
+/** rT = {0,0,0,0}. */
+extern void
+spe_zero(struct spe_function *p, int rT);
+
+/** rT = splat(rA, word) */
+extern void
+spe_splat_word(struct spe_function *p, int rT, int rA, int word);
+
+/** rT = float min(rA, rB) */
+extern void
+spe_float_min(struct spe_function *p, int rT, int rA, int rB);
+
+/** rT = float max(rA, rB) */
+extern void
+spe_float_max(struct spe_function *p, int rT, int rA, int rB);
+
+
+/* Floating-point instructions
+ */
+EMIT_RR  (spe_fa,         0x2c4)
+EMIT_RR  (spe_dfa,        0x2cc)
+EMIT_RR  (spe_fs,         0x2c5)
+EMIT_RR  (spe_dfs,        0x2cd)
+EMIT_RR  (spe_fm,         0x2c6)
+EMIT_RR  (spe_dfm,        0x2ce)
+EMIT_RRR (spe_fma,        0x00e)
+EMIT_RR  (spe_dfma,       0x35c)
+EMIT_RRR (spe_fnms,       0x00d)
+EMIT_RR  (spe_dfnms,      0x35e)
+EMIT_RRR (spe_fms,        0x00f)
+EMIT_RR  (spe_dfms,       0x35d)
+EMIT_RR  (spe_dfnma,      0x35f)
+EMIT_R   (spe_frest,      0x1b8)
+EMIT_R   (spe_frsqest,    0x1b9)
+EMIT_RR  (spe_fi,         0x3d4)
+EMIT_RI8 (spe_csflt,      0x1da, 155)
+EMIT_RI8 (spe_cflts,      0x1d8, 173)
+EMIT_RI8 (spe_cuflt,      0x1db, 155)
+EMIT_RI8 (spe_cfltu,      0x1d9, 173)
+EMIT_R   (spe_frds,       0x3b9)
+EMIT_R   (spe_fesd,       0x3b8)
+EMIT_RR  (spe_dfceq,      0x3c3)
+EMIT_RR  (spe_dfcmeq,     0x3cb)
+EMIT_RR  (spe_dfcgt,      0x2c3)
+EMIT_RR  (spe_dfcmgt,     0x2cb)
+EMIT_RI7 (spe_dftsv,      0x3bf)
+EMIT_RR  (spe_fceq,       0x3c2)
+EMIT_RR  (spe_fcmeq,      0x3ca)
+EMIT_RR  (spe_fcgt,       0x2c2)
+EMIT_RR  (spe_fcmgt,      0x2ca)
+EMIT_R   (spe_fscrwr,     0x3ba)
+EMIT_    (spe_fscrrd,     0x398)
+
+
+/* Channel instructions
+ */
+EMIT_R   (spe_rdch,       0x00d)
+EMIT_R   (spe_rdchcnt,    0x00f)
+EMIT_R   (spe_wrch,       0x10d)
+
+
+#ifdef UNDEF_EMIT_MACROS
+#undef EMIT
+#undef EMIT_
+#undef EMIT_R
+#undef EMIT_RR
+#undef EMIT_RRR
+#undef EMIT_RI7
+#undef EMIT_RI8
+#undef EMIT_RI10
+#undef EMIT_RI10s
+#undef EMIT_RI16
+#undef EMIT_RI18
+#undef EMIT_I16
+#undef UNDEF_EMIT_MACROS
+#endif /* EMIT_ */
diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
new file mode 100644
index 00000000000..99ee74cf14b
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
@@ -0,0 +1,1748 @@
+/**************************************************************************
+ *
+ * Copyright (C) 1999-2005  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include "pipe/p_config.h"
+
+#if defined(PIPE_ARCH_X86)
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+#include "util/u_pointer.h"
+
+#include "rtasm_execmem.h"
+#include "rtasm_x86sse.h"
+
+#define DISASSEM 0
+#define X86_TWOB 0x0f
+
+
+#define DUMP_SSE  0
+
+
+void x86_print_reg( struct x86_reg reg )
+{
+   if (reg.mod != mod_REG) 
+      debug_printf( "[" );
+      
+   switch( reg.file ) {
+   case file_REG32:
+      switch( reg.idx ) {
+      case reg_AX: debug_printf( "EAX" ); break;
+      case reg_CX: debug_printf( "ECX" ); break;
+      case reg_DX: debug_printf( "EDX" ); break;
+      case reg_BX: debug_printf( "EBX" ); break;
+      case reg_SP: debug_printf( "ESP" ); break;
+      case reg_BP: debug_printf( "EBP" ); break;
+      case reg_SI: debug_printf( "ESI" ); break;
+      case reg_DI: debug_printf( "EDI" ); break;
+      }
+      break;
+   case file_MMX:
+      debug_printf( "MMX%u", reg.idx );
+      break;
+   case file_XMM:
+      debug_printf( "XMM%u", reg.idx );
+      break;
+   case file_x87:
+      debug_printf( "fp%u", reg.idx );
+      break;
+   }
+
+   if (reg.mod == mod_DISP8 ||
+       reg.mod == mod_DISP32)
+      debug_printf("+%d", reg.disp);
+
+   if (reg.mod != mod_REG) 
+      debug_printf( "]" );
+}
+
+#if DUMP_SSE
+
+#define DUMP_START() debug_printf( "\n" )
+#define DUMP_END() debug_printf( "\n" )
+
+#define DUMP() do {                             \
+   const char *foo = __FUNCTION__;              \
+   while (*foo && *foo != '_')                  \
+      foo++;                                    \
+   if  (*foo)                                   \
+      foo++;                                    \
+   debug_printf( "\n% 4x% 15s ", p->csr - p->store, foo );             \
+} while (0)
+
+#define DUMP_I( I ) do {                        \
+   DUMP();                                      \
+   debug_printf( "%u", I );                     \
+} while( 0 )
+
+#define DUMP_R( R0 ) do {                       \
+   DUMP();                                      \
+   x86_print_reg( R0 );                            \
+} while( 0 )
+
+#define DUMP_RR( R0, R1 ) do {                  \
+   DUMP();                                      \
+   x86_print_reg( R0 );                            \
+   debug_printf( ", " );                        \
+   x86_print_reg( R1 );                            \
+} while( 0 )
+
+#define DUMP_RI( R0, I ) do {                   \
+   DUMP();                                      \
+   x86_print_reg( R0 );                            \
+   debug_printf( ", %u", I );                   \
+} while( 0 )
+
+#define DUMP_RRI( R0, R1, I ) do {              \
+   DUMP();                                      \
+   x86_print_reg( R0 );                            \
+   debug_printf( ", " );                        \
+   x86_print_reg( R1 );                            \
+   debug_printf( ", %u", I );                   \
+} while( 0 )
+
+#else
+
+#define DUMP_START()
+#define DUMP_END()
+#define DUMP( )
+#define DUMP_I( I )
+#define DUMP_R( R0 )
+#define DUMP_RR( R0, R1 )
+#define DUMP_RI( R0, I )
+#define DUMP_RRI( R0, R1, I )
+
+#endif
+
+
+static void do_realloc( struct x86_function *p )
+{
+   if (p->store == p->error_overflow) {
+      p->csr = p->store;
+   }
+   else if (p->size == 0) {
+      p->size = 1024;
+      p->store = rtasm_exec_malloc(p->size);
+      p->csr = p->store;
+   }
+   else {
+      uintptr_t used = pointer_to_uintptr( p->csr ) - pointer_to_uintptr( p->store );
+      unsigned char *tmp = p->store;
+      p->size *= 2;
+      p->store = rtasm_exec_malloc(p->size);
+
+      if (p->store) {
+         memcpy(p->store, tmp, used);
+         p->csr = p->store + used;
+      }
+      else {
+         p->csr = p->store;
+      }
+
+      rtasm_exec_free(tmp);
+   }
+
+   if (p->store == NULL) {
+      p->store = p->csr = p->error_overflow;
+      p->size = sizeof(p->error_overflow);
+   }
+}
+
+/* Emit bytes to the instruction stream:
+ */
+static unsigned char *reserve( struct x86_function *p, int bytes )
+{
+   if (p->csr + bytes - p->store > (int) p->size)
+      do_realloc(p);
+
+   {
+      unsigned char *csr = p->csr;
+      p->csr += bytes;
+      return csr;
+   }
+}
+
+
+
+static void emit_1b( struct x86_function *p, char b0 )
+{
+   char *csr = (char *)reserve(p, 1);
+   *csr = b0;
+}
+
+static void emit_1i( struct x86_function *p, int i0 )
+{
+   int *icsr = (int *)reserve(p, sizeof(i0));
+   *icsr = i0;
+}
+
+static void emit_1ub( struct x86_function *p, unsigned char b0 )
+{
+   unsigned char *csr = reserve(p, 1);
+   *csr++ = b0;
+}
+
+static void emit_2ub( struct x86_function *p, unsigned char b0, unsigned char b1 )
+{
+   unsigned char *csr = reserve(p, 2);
+   *csr++ = b0;
+   *csr++ = b1;
+}
+
+static void emit_3ub( struct x86_function *p, unsigned char b0, unsigned char b1, unsigned char b2 )
+{
+   unsigned char *csr = reserve(p, 3);
+   *csr++ = b0;
+   *csr++ = b1;
+   *csr++ = b2;
+}
+
+
+/* Build a modRM byte + possible displacement.  No treatment of SIB
+ * indexing.  BZZT - no way to encode an absolute address.
+ *
+ * This is the "/r" field in the x86 manuals...
+ */
+static void emit_modrm( struct x86_function *p, 
+			struct x86_reg reg, 
+			struct x86_reg regmem )
+{
+   unsigned char val = 0;
+   
+   assert(reg.mod == mod_REG);
+   
+   val |= regmem.mod << 6;     	/* mod field */
+   val |= reg.idx << 3;		/* reg field */
+   val |= regmem.idx;		/* r/m field */
+   
+   emit_1ub(p, val);
+
+   /* Oh-oh we've stumbled into the SIB thing.
+    */
+   if (regmem.file == file_REG32 &&
+       regmem.idx == reg_SP &&
+       regmem.mod != mod_REG) {
+      emit_1ub(p, 0x24);		/* simplistic! */
+   }
+
+   switch (regmem.mod) {
+   case mod_REG:
+   case mod_INDIRECT:
+      break;
+   case mod_DISP8:
+      emit_1b(p, (char) regmem.disp);
+      break;
+   case mod_DISP32:
+      emit_1i(p, regmem.disp);
+      break;
+   default:
+      assert(0);
+      break;
+   }
+}
+
+/* Emits the "/0".."/7" specialized versions of the modrm ("/r") bytes.
+ */
+static void emit_modrm_noreg( struct x86_function *p,
+			      unsigned op,
+			      struct x86_reg regmem )
+{
+   struct x86_reg dummy = x86_make_reg(file_REG32, op);
+   emit_modrm(p, dummy, regmem);
+}
+
+/* Many x86 instructions have two opcodes to cope with the situations
+ * where the destination is a register or memory reference
+ * respectively.  This function selects the correct opcode based on
+ * the arguments presented.
+ */
+static void emit_op_modrm( struct x86_function *p,
+			   unsigned char op_dst_is_reg, 
+			   unsigned char op_dst_is_mem,
+			   struct x86_reg dst,
+			   struct x86_reg src )
+{  
+   switch (dst.mod) {
+   case mod_REG:
+      emit_1ub(p, op_dst_is_reg);
+      emit_modrm(p, dst, src);
+      break;
+   case mod_INDIRECT:
+   case mod_DISP32:
+   case mod_DISP8:
+      assert(src.mod == mod_REG);
+      emit_1ub(p, op_dst_is_mem);
+      emit_modrm(p, src, dst);
+      break;
+   default:
+      assert(0);
+      break;
+   }
+}
+
+
+
+
+
+
+
+/* Create and manipulate registers and regmem values:
+ */
+struct x86_reg x86_make_reg( enum x86_reg_file file,
+			     enum x86_reg_name idx )
+{
+   struct x86_reg reg;
+
+   reg.file = file;
+   reg.idx = idx;
+   reg.mod = mod_REG;
+   reg.disp = 0;
+
+   return reg;
+}
+
+struct x86_reg x86_make_disp( struct x86_reg reg,
+			      int disp )
+{
+   assert(reg.file == file_REG32);
+
+   if (reg.mod == mod_REG)
+      reg.disp = disp;
+   else
+      reg.disp += disp;
+
+   if (reg.disp == 0 && reg.idx != reg_BP)
+      reg.mod = mod_INDIRECT;
+   else if (reg.disp <= 127 && reg.disp >= -128)
+      reg.mod = mod_DISP8;
+   else
+      reg.mod = mod_DISP32;
+
+   return reg;
+}
+
+struct x86_reg x86_deref( struct x86_reg reg )
+{
+   return x86_make_disp(reg, 0);
+}
+
+struct x86_reg x86_get_base_reg( struct x86_reg reg )
+{
+   return x86_make_reg( reg.file, reg.idx );
+}
+
+int x86_get_label( struct x86_function *p )
+{
+   return p->csr - p->store;
+}
+
+
+
+/***********************************************************************
+ * x86 instructions
+ */
+
+
+void x86_jcc( struct x86_function *p,
+	      enum x86_cc cc,
+	      int label )
+{
+   int offset = label - (x86_get_label(p) + 2);
+   DUMP_I(cc);
+   
+   if (offset < 0) {
+      /*assert(p->csr - p->store > -offset);*/
+      if (p->csr - p->store <= -offset) {
+         /* probably out of memory (using the error_overflow buffer) */
+         return;
+      }
+   }
+
+   if (offset <= 127 && offset >= -128) {
+      emit_1ub(p, 0x70 + cc);
+      emit_1b(p, (char) offset);
+   }
+   else {
+      offset = label - (x86_get_label(p) + 6);
+      emit_2ub(p, 0x0f, 0x80 + cc);
+      emit_1i(p, offset);
+   }
+}
+
+/* Always use a 32bit offset for forward jumps:
+ */
+int x86_jcc_forward( struct x86_function *p,
+                     enum x86_cc cc )
+{
+   DUMP_I(cc);
+   emit_2ub(p, 0x0f, 0x80 + cc);
+   emit_1i(p, 0);
+   return x86_get_label(p);
+}
+
+int x86_jmp_forward( struct x86_function *p)
+{
+   DUMP();
+   emit_1ub(p, 0xe9);
+   emit_1i(p, 0);
+   return x86_get_label(p);
+}
+
+int x86_call_forward( struct x86_function *p)
+{
+   DUMP();
+
+   emit_1ub(p, 0xe8);
+   emit_1i(p, 0);
+   return x86_get_label(p);
+}
+
+/* Fixup offset from forward jump:
+ */
+void x86_fixup_fwd_jump( struct x86_function *p,
+			 int fixup )
+{
+   *(int *)(p->store + fixup - 4) = x86_get_label(p) - fixup;
+}
+
+void x86_jmp( struct x86_function *p, int label)
+{
+   DUMP_I( label );
+   emit_1ub(p, 0xe9);
+   emit_1i(p, label - x86_get_label(p) - 4);
+}
+
+void x86_call( struct x86_function *p, struct x86_reg reg)
+{
+   DUMP_R( reg );
+   emit_1ub(p, 0xff);
+   emit_modrm_noreg(p, 2, reg);
+}
+
+
+void x86_mov_reg_imm( struct x86_function *p, struct x86_reg dst, int imm )
+{
+   DUMP_RI( dst, imm );
+   assert(dst.file == file_REG32);
+   assert(dst.mod == mod_REG);
+   emit_1ub(p, 0xb8 + dst.idx);
+   emit_1i(p, imm);
+}
+
+/**
+ * Immediate group 1 instructions.
+ */
+static INLINE void 
+x86_group1_imm( struct x86_function *p, 
+                unsigned op, struct x86_reg dst, int imm )
+{
+   assert(dst.file == file_REG32);
+   assert(dst.mod == mod_REG);
+   if(-0x80 <= imm && imm < 0x80) {
+      emit_1ub(p, 0x83);
+      emit_modrm_noreg(p, op, dst);
+      emit_1b(p, (char)imm);
+   }
+   else {
+      emit_1ub(p, 0x81);
+      emit_modrm_noreg(p, op, dst);
+      emit_1i(p, imm);
+   }
+}
+
+void x86_add_imm( struct x86_function *p, struct x86_reg dst, int imm )
+{
+   DUMP_RI( dst, imm );
+   x86_group1_imm(p, 0, dst, imm);
+}
+
+void x86_or_imm( struct x86_function *p, struct x86_reg dst, int imm )
+{
+   DUMP_RI( dst, imm );
+   x86_group1_imm(p, 1, dst, imm);
+}
+
+void x86_and_imm( struct x86_function *p, struct x86_reg dst, int imm )
+{
+   DUMP_RI( dst, imm );
+   x86_group1_imm(p, 4, dst, imm);
+}
+
+void x86_sub_imm( struct x86_function *p, struct x86_reg dst, int imm )
+{
+   DUMP_RI( dst, imm );
+   x86_group1_imm(p, 5, dst, imm);
+}
+
+void x86_xor_imm( struct x86_function *p, struct x86_reg dst, int imm )
+{
+   DUMP_RI( dst, imm );
+   x86_group1_imm(p, 6, dst, imm);
+}
+
+void x86_cmp_imm( struct x86_function *p, struct x86_reg dst, int imm )
+{
+   DUMP_RI( dst, imm );
+   x86_group1_imm(p, 7, dst, imm);
+}
+
+
+void x86_push( struct x86_function *p,
+	       struct x86_reg reg )
+{
+   DUMP_R( reg );
+   if (reg.mod == mod_REG)
+      emit_1ub(p, 0x50 + reg.idx);
+   else 
+   {
+      emit_1ub(p, 0xff);
+      emit_modrm_noreg(p, 6, reg);
+   }
+
+
+   p->stack_offset += 4;
+}
+
+void x86_push_imm32( struct x86_function *p,
+                     int imm32 )
+{
+   DUMP_I( imm32 );
+   emit_1ub(p, 0x68);
+   emit_1i(p,  imm32);
+
+   p->stack_offset += 4;
+}
+
+
+void x86_pop( struct x86_function *p,
+	      struct x86_reg reg )
+{
+   DUMP_R( reg );
+   assert(reg.mod == mod_REG);
+   emit_1ub(p, 0x58 + reg.idx);
+   p->stack_offset -= 4;
+}
+
+void x86_inc( struct x86_function *p,
+	      struct x86_reg reg )
+{
+   DUMP_R( reg );
+   assert(reg.mod == mod_REG);
+   emit_1ub(p, 0x40 + reg.idx);
+}
+
+void x86_dec( struct x86_function *p,
+	      struct x86_reg reg )
+{
+   DUMP_R( reg );
+   assert(reg.mod == mod_REG);
+   emit_1ub(p, 0x48 + reg.idx);
+}
+
+void x86_ret( struct x86_function *p )
+{
+   DUMP();
+   assert(p->stack_offset == 0);
+   emit_1ub(p, 0xc3);
+}
+
+void x86_retw( struct x86_function *p, unsigned short imm )
+{
+   DUMP();
+   emit_3ub(p, 0xc2, imm & 0xff, (imm >> 8) & 0xff);
+}
+
+void x86_sahf( struct x86_function *p )
+{
+   DUMP();
+   emit_1ub(p, 0x9e);
+}
+
+void x86_mov( struct x86_function *p,
+	      struct x86_reg dst,
+	      struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_op_modrm( p, 0x8b, 0x89, dst, src );
+}
+
+void x86_xor( struct x86_function *p,
+	      struct x86_reg dst,
+	      struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_op_modrm( p, 0x33, 0x31, dst, src );
+}
+
+void x86_cmp( struct x86_function *p,
+	      struct x86_reg dst,
+	      struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_op_modrm( p, 0x3b, 0x39, dst, src );
+}
+
+void x86_lea( struct x86_function *p,
+	      struct x86_reg dst,
+	      struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_1ub(p, 0x8d);
+   emit_modrm( p, dst, src );
+}
+
+void x86_test( struct x86_function *p,
+	       struct x86_reg dst,
+	       struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_1ub(p, 0x85);
+   emit_modrm( p, dst, src );
+}
+
+void x86_add( struct x86_function *p,
+	       struct x86_reg dst,
+	       struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_op_modrm(p, 0x03, 0x01, dst, src );
+}
+
+/* Calculate EAX * src, results in EDX:EAX.
+ */
+void x86_mul( struct x86_function *p,
+	       struct x86_reg src )
+{
+   DUMP_R(  src );
+   emit_1ub(p, 0xf7);
+   emit_modrm_noreg(p, 4, src );
+}
+
+
+void x86_imul( struct x86_function *p,
+	       struct x86_reg dst,
+	       struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, X86_TWOB, 0xAF);
+   emit_modrm(p, dst, src);
+}
+
+
+void x86_sub( struct x86_function *p,
+	       struct x86_reg dst,
+	       struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_op_modrm(p, 0x2b, 0x29, dst, src );
+}
+
+void x86_or( struct x86_function *p,
+             struct x86_reg dst,
+             struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_op_modrm( p, 0x0b, 0x09, dst, src );
+}
+
+void x86_and( struct x86_function *p,
+              struct x86_reg dst,
+              struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_op_modrm( p, 0x23, 0x21, dst, src );
+}
+
+
+
+/***********************************************************************
+ * SSE instructions
+ */
+
+void sse_prefetchnta( struct x86_function *p, struct x86_reg ptr)
+{
+   DUMP_R( ptr );
+   assert(ptr.mod != mod_REG);
+   emit_2ub(p, 0x0f, 0x18);
+   emit_modrm_noreg(p, 0, ptr);
+}
+
+void sse_prefetch0( struct x86_function *p, struct x86_reg ptr)
+{
+   DUMP_R( ptr );
+   assert(ptr.mod != mod_REG);
+   emit_2ub(p, 0x0f, 0x18);
+   emit_modrm_noreg(p, 1, ptr);
+}
+
+void sse_prefetch1( struct x86_function *p, struct x86_reg ptr)
+{
+   DUMP_R( ptr );
+   assert(ptr.mod != mod_REG);
+   emit_2ub(p, 0x0f, 0x18);
+   emit_modrm_noreg(p, 2, ptr);
+}
+
+void sse_movntps( struct x86_function *p, 
+                  struct x86_reg dst,
+                  struct x86_reg src)
+{
+   DUMP_RR( dst, src );
+
+   assert(dst.mod != mod_REG);
+   assert(src.mod == mod_REG);
+   emit_2ub(p, 0x0f, 0x2b);
+   emit_modrm(p, src, dst);
+}
+
+
+
+
+void sse_movss( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, 0xF3, X86_TWOB);
+   emit_op_modrm( p, 0x10, 0x11, dst, src );
+}
+
+void sse_movaps( struct x86_function *p,
+		 struct x86_reg dst,
+		 struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_1ub(p, X86_TWOB);
+   emit_op_modrm( p, 0x28, 0x29, dst, src );
+}
+
+void sse_movups( struct x86_function *p,
+		 struct x86_reg dst,
+		 struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_1ub(p, X86_TWOB);
+   emit_op_modrm( p, 0x10, 0x11, dst, src );
+}
+
+void sse_movhps( struct x86_function *p,
+		 struct x86_reg dst,
+		 struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   assert(dst.mod != mod_REG || src.mod != mod_REG);
+   emit_1ub(p, X86_TWOB);
+   emit_op_modrm( p, 0x16, 0x17, dst, src ); /* cf movlhps */
+}
+
+void sse_movlps( struct x86_function *p,
+		 struct x86_reg dst,
+		 struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   assert(dst.mod != mod_REG || src.mod != mod_REG);
+   emit_1ub(p, X86_TWOB);
+   emit_op_modrm( p, 0x12, 0x13, dst, src ); /* cf movhlps */
+}
+
+void sse_maxps( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, X86_TWOB, 0x5F);
+   emit_modrm( p, dst, src );
+}
+
+void sse_maxss( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_3ub(p, 0xF3, X86_TWOB, 0x5F);
+   emit_modrm( p, dst, src );
+}
+
+void sse_divss( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_3ub(p, 0xF3, X86_TWOB, 0x5E);
+   emit_modrm( p, dst, src );
+}
+
+void sse_minps( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, X86_TWOB, 0x5D);
+   emit_modrm( p, dst, src );
+}
+
+void sse_subps( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, X86_TWOB, 0x5C);
+   emit_modrm( p, dst, src );
+}
+
+void sse_mulps( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, X86_TWOB, 0x59);
+   emit_modrm( p, dst, src );
+}
+
+void sse_mulss( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_3ub(p, 0xF3, X86_TWOB, 0x59);
+   emit_modrm( p, dst, src );
+}
+
+void sse_addps( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, X86_TWOB, 0x58);
+   emit_modrm( p, dst, src );
+}
+
+void sse_addss( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_3ub(p, 0xF3, X86_TWOB, 0x58);
+   emit_modrm( p, dst, src );
+}
+
+void sse_andnps( struct x86_function *p,
+                 struct x86_reg dst,
+                 struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, X86_TWOB, 0x55);
+   emit_modrm( p, dst, src );
+}
+
+void sse_andps( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, X86_TWOB, 0x54);
+   emit_modrm( p, dst, src );
+}
+
+void sse_rsqrtps( struct x86_function *p,
+                  struct x86_reg dst,
+                  struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, X86_TWOB, 0x52);
+   emit_modrm( p, dst, src );
+}
+
+void sse_rsqrtss( struct x86_function *p,
+		  struct x86_reg dst,
+		  struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_3ub(p, 0xF3, X86_TWOB, 0x52);
+   emit_modrm( p, dst, src );
+
+}
+
+void sse_movhlps( struct x86_function *p,
+		  struct x86_reg dst,
+		  struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   assert(dst.mod == mod_REG && src.mod == mod_REG);
+   emit_2ub(p, X86_TWOB, 0x12);
+   emit_modrm( p, dst, src );
+}
+
+void sse_movlhps( struct x86_function *p,
+		  struct x86_reg dst,
+		  struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   assert(dst.mod == mod_REG && src.mod == mod_REG);
+   emit_2ub(p, X86_TWOB, 0x16);
+   emit_modrm( p, dst, src );
+}
+
+void sse_orps( struct x86_function *p,
+               struct x86_reg dst,
+               struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, X86_TWOB, 0x56);
+   emit_modrm( p, dst, src );
+}
+
+void sse_xorps( struct x86_function *p,
+                struct x86_reg dst,
+                struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, X86_TWOB, 0x57);
+   emit_modrm( p, dst, src );
+}
+
+void sse_cvtps2pi( struct x86_function *p,
+		   struct x86_reg dst,
+		   struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   assert(dst.file == file_MMX && 
+	  (src.file == file_XMM || src.mod != mod_REG));
+
+   p->need_emms = 1;
+
+   emit_2ub(p, X86_TWOB, 0x2d);
+   emit_modrm( p, dst, src );
+}
+
+void sse2_cvtdq2ps( struct x86_function *p,
+		   struct x86_reg dst,
+		   struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, X86_TWOB, 0x5b);
+   emit_modrm( p, dst, src );
+}
+
+
+/* Shufps can also be used to implement a reduced swizzle when dest ==
+ * arg0.
+ */
+void sse_shufps( struct x86_function *p,
+		 struct x86_reg dst,
+		 struct x86_reg src,
+		 unsigned char shuf) 
+{
+   DUMP_RRI( dst, src, shuf );
+   emit_2ub(p, X86_TWOB, 0xC6);
+   emit_modrm(p, dst, src);
+   emit_1ub(p, shuf); 
+}
+
+void sse_unpckhps( struct x86_function *p, struct x86_reg dst, struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_2ub( p, X86_TWOB, 0x15 );
+   emit_modrm( p, dst, src );
+}
+
+void sse_unpcklps( struct x86_function *p, struct x86_reg dst, struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_2ub( p, X86_TWOB, 0x14 );
+   emit_modrm( p, dst, src );
+}
+
+void sse_cmpps( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src,
+		enum sse_cc cc) 
+{
+   DUMP_RRI( dst, src, cc );
+   emit_2ub(p, X86_TWOB, 0xC2);
+   emit_modrm(p, dst, src);
+   emit_1ub(p, cc); 
+}
+
+void sse_pmovmskb( struct x86_function *p,
+                   struct x86_reg dst,
+                   struct x86_reg src)
+{
+   DUMP_RR( dst, src );
+   emit_3ub(p, 0x66, X86_TWOB, 0xD7);
+   emit_modrm(p, dst, src);
+}
+
+/***********************************************************************
+ * SSE2 instructions
+ */
+
+/**
+ * Perform a reduced swizzle:
+ */
+void sse2_pshufd( struct x86_function *p,
+		  struct x86_reg dst,
+		  struct x86_reg src,
+		  unsigned char shuf) 
+{
+   DUMP_RRI( dst, src, shuf );
+   emit_3ub(p, 0x66, X86_TWOB, 0x70);
+   emit_modrm(p, dst, src);
+   emit_1ub(p, shuf); 
+}
+
+void sse2_cvttps2dq( struct x86_function *p,
+                     struct x86_reg dst,
+                     struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_3ub( p, 0xF3, X86_TWOB, 0x5B );
+   emit_modrm( p, dst, src );
+}
+
+void sse2_cvtps2dq( struct x86_function *p,
+		    struct x86_reg dst,
+		    struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_3ub(p, 0x66, X86_TWOB, 0x5B);
+   emit_modrm( p, dst, src );
+}
+
+void sse2_packssdw( struct x86_function *p,
+		    struct x86_reg dst,
+		    struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_3ub(p, 0x66, X86_TWOB, 0x6B);
+   emit_modrm( p, dst, src );
+}
+
+void sse2_packsswb( struct x86_function *p,
+		    struct x86_reg dst,
+		    struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_3ub(p, 0x66, X86_TWOB, 0x63);
+   emit_modrm( p, dst, src );
+}
+
+void sse2_packuswb( struct x86_function *p,
+		    struct x86_reg dst,
+		    struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_3ub(p, 0x66, X86_TWOB, 0x67);
+   emit_modrm( p, dst, src );
+}
+
+void sse2_punpcklbw( struct x86_function *p,
+		    struct x86_reg dst,
+		    struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_3ub(p, 0x66, X86_TWOB, 0x60);
+   emit_modrm( p, dst, src );
+}
+
+
+void sse2_rcpps( struct x86_function *p,
+                 struct x86_reg dst,
+                 struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, X86_TWOB, 0x53);
+   emit_modrm( p, dst, src );
+}
+
+void sse2_rcpss( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_3ub(p, 0xF3, X86_TWOB, 0x53);
+   emit_modrm( p, dst, src );
+}
+
+void sse2_movd( struct x86_function *p,
+		struct x86_reg dst,
+		struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   emit_2ub(p, 0x66, X86_TWOB);
+   emit_op_modrm( p, 0x6e, 0x7e, dst, src );
+}
+
+
+
+
+/***********************************************************************
+ * x87 instructions
+ */
+static void note_x87_pop( struct x86_function *p )
+{
+   p->x87_stack--;
+   assert(p->x87_stack >= 0);
+}
+
+static void note_x87_push( struct x86_function *p )
+{
+   p->x87_stack++;
+   assert(p->x87_stack <= 7);
+}
+
+void x87_assert_stack_empty( struct x86_function *p )
+{
+   assert (p->x87_stack == 0);
+}
+
+
+void x87_fist( struct x86_function *p, struct x86_reg dst )
+{
+   DUMP_R( dst );
+   emit_1ub(p, 0xdb);
+   emit_modrm_noreg(p, 2, dst);
+}
+
+void x87_fistp( struct x86_function *p, struct x86_reg dst )
+{
+   DUMP_R( dst );
+   emit_1ub(p, 0xdb);
+   emit_modrm_noreg(p, 3, dst);
+   note_x87_pop(p);
+}
+
+void x87_fild( struct x86_function *p, struct x86_reg arg )
+{
+   DUMP_R( arg );
+   emit_1ub(p, 0xdf);
+   emit_modrm_noreg(p, 0, arg);
+   note_x87_push(p);
+}
+
+void x87_fldz( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xd9, 0xee);
+   note_x87_push(p);
+}
+
+
+void x87_fldcw( struct x86_function *p, struct x86_reg arg )
+{
+   DUMP_R( arg );
+   assert(arg.file == file_REG32);
+   assert(arg.mod != mod_REG);
+   emit_1ub(p, 0xd9);
+   emit_modrm_noreg(p, 5, arg);
+}
+
+void x87_fld1( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xd9, 0xe8);
+   note_x87_push(p);
+}
+
+void x87_fldl2e( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xd9, 0xea);
+   note_x87_push(p);
+}
+
+void x87_fldln2( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xd9, 0xed);
+   note_x87_push(p);
+}
+
+void x87_fwait( struct x86_function *p )
+{
+   DUMP();
+   emit_1ub(p, 0x9b);
+}
+
+void x87_fnclex( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xdb, 0xe2);
+}
+
+void x87_fclex( struct x86_function *p )
+{
+   x87_fwait(p);
+   x87_fnclex(p);
+}
+
+void x87_fcmovb( struct x86_function *p, struct x86_reg arg )
+{
+   DUMP_R( arg );
+   assert(arg.file == file_x87);
+   emit_2ub(p, 0xda, 0xc0+arg.idx);
+}
+
+void x87_fcmove( struct x86_function *p, struct x86_reg arg )
+{
+   DUMP_R( arg );
+   assert(arg.file == file_x87);
+   emit_2ub(p, 0xda, 0xc8+arg.idx);
+}
+
+void x87_fcmovbe( struct x86_function *p, struct x86_reg arg )
+{
+   DUMP_R( arg );
+   assert(arg.file == file_x87);
+   emit_2ub(p, 0xda, 0xd0+arg.idx);
+}
+
+void x87_fcmovnb( struct x86_function *p, struct x86_reg arg )
+{
+   DUMP_R( arg );
+   assert(arg.file == file_x87);
+   emit_2ub(p, 0xdb, 0xc0+arg.idx);
+}
+
+void x87_fcmovne( struct x86_function *p, struct x86_reg arg )
+{
+   DUMP_R( arg );
+   assert(arg.file == file_x87);
+   emit_2ub(p, 0xdb, 0xc8+arg.idx);
+}
+
+void x87_fcmovnbe( struct x86_function *p, struct x86_reg arg )
+{
+   DUMP_R( arg );
+   assert(arg.file == file_x87);
+   emit_2ub(p, 0xdb, 0xd0+arg.idx);
+}
+
+
+
+static void x87_arith_op( struct x86_function *p, struct x86_reg dst, struct x86_reg arg,
+			  unsigned char dst0ub0,
+			  unsigned char dst0ub1,
+			  unsigned char arg0ub0,
+			  unsigned char arg0ub1,
+			  unsigned char argmem_noreg)
+{
+   assert(dst.file == file_x87);
+
+   if (arg.file == file_x87) {
+      if (dst.idx == 0) 
+	 emit_2ub(p, dst0ub0, dst0ub1+arg.idx);
+      else if (arg.idx == 0) 
+	 emit_2ub(p, arg0ub0, arg0ub1+arg.idx);
+      else
+	 assert(0);
+   }
+   else if (dst.idx == 0) {
+      assert(arg.file == file_REG32);
+      emit_1ub(p, 0xd8);
+      emit_modrm_noreg(p, argmem_noreg, arg);
+   }
+   else
+      assert(0);
+}
+
+void x87_fmul( struct x86_function *p, struct x86_reg dst, struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   x87_arith_op(p, dst, src, 
+		0xd8, 0xc8,
+		0xdc, 0xc8,
+		4);
+}
+
+void x87_fsub( struct x86_function *p, struct x86_reg dst, struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   x87_arith_op(p, dst, src, 
+		0xd8, 0xe0,
+		0xdc, 0xe8,
+		4);
+}
+
+void x87_fsubr( struct x86_function *p, struct x86_reg dst, struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   x87_arith_op(p, dst, src, 
+		0xd8, 0xe8,
+		0xdc, 0xe0,
+		5);
+}
+
+void x87_fadd( struct x86_function *p, struct x86_reg dst, struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   x87_arith_op(p, dst, src, 
+		0xd8, 0xc0,
+		0xdc, 0xc0,
+		0);
+}
+
+void x87_fdiv( struct x86_function *p, struct x86_reg dst, struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   x87_arith_op(p, dst, src, 
+		0xd8, 0xf0,
+		0xdc, 0xf8,
+		6);
+}
+
+void x87_fdivr( struct x86_function *p, struct x86_reg dst, struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   x87_arith_op(p, dst, src, 
+		0xd8, 0xf8,
+		0xdc, 0xf0,
+		7);
+}
+
+void x87_fmulp( struct x86_function *p, struct x86_reg dst )
+{
+   DUMP_R( dst );
+   assert(dst.file == file_x87);
+   assert(dst.idx >= 1);
+   emit_2ub(p, 0xde, 0xc8+dst.idx);
+   note_x87_pop(p);
+}
+
+void x87_fsubp( struct x86_function *p, struct x86_reg dst )
+{
+   DUMP_R( dst );
+   assert(dst.file == file_x87);
+   assert(dst.idx >= 1);
+   emit_2ub(p, 0xde, 0xe8+dst.idx);
+   note_x87_pop(p);
+}
+
+void x87_fsubrp( struct x86_function *p, struct x86_reg dst )
+{
+   DUMP_R( dst );
+   assert(dst.file == file_x87);
+   assert(dst.idx >= 1);
+   emit_2ub(p, 0xde, 0xe0+dst.idx);
+   note_x87_pop(p);
+}
+
+void x87_faddp( struct x86_function *p, struct x86_reg dst )
+{
+   DUMP_R( dst );
+   assert(dst.file == file_x87);
+   assert(dst.idx >= 1);
+   emit_2ub(p, 0xde, 0xc0+dst.idx);
+   note_x87_pop(p);
+}
+
+void x87_fdivp( struct x86_function *p, struct x86_reg dst )
+{
+   DUMP_R( dst );
+   assert(dst.file == file_x87);
+   assert(dst.idx >= 1);
+   emit_2ub(p, 0xde, 0xf8+dst.idx);
+   note_x87_pop(p);
+}
+
+void x87_fdivrp( struct x86_function *p, struct x86_reg dst )
+{
+   DUMP_R( dst );
+   assert(dst.file == file_x87);
+   assert(dst.idx >= 1);
+   emit_2ub(p, 0xde, 0xf0+dst.idx);
+   note_x87_pop(p);
+}
+
+void x87_ftst( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xd9, 0xe4);
+}
+
+void x87_fucom( struct x86_function *p, struct x86_reg arg )
+{
+   DUMP_R( arg );
+   assert(arg.file == file_x87);
+   emit_2ub(p, 0xdd, 0xe0+arg.idx);
+}
+
+void x87_fucomp( struct x86_function *p, struct x86_reg arg )
+{
+   DUMP_R( arg );
+   assert(arg.file == file_x87);
+   emit_2ub(p, 0xdd, 0xe8+arg.idx);
+   note_x87_pop(p);
+}
+
+void x87_fucompp( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xda, 0xe9);
+   note_x87_pop(p);             /* pop twice */
+   note_x87_pop(p);             /* pop twice */
+}
+
+void x87_fxch( struct x86_function *p, struct x86_reg arg )
+{
+   DUMP_R( arg );
+   assert(arg.file == file_x87);
+   emit_2ub(p, 0xd9, 0xc8+arg.idx);
+}
+
+void x87_fabs( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xd9, 0xe1);
+}
+
+void x87_fchs( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xd9, 0xe0);
+}
+
+void x87_fcos( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xd9, 0xff);
+}
+
+
+void x87_fprndint( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xd9, 0xfc);
+}
+
+void x87_fscale( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xd9, 0xfd);
+}
+
+void x87_fsin( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xd9, 0xfe);
+}
+
+void x87_fsincos( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xd9, 0xfb);
+}
+
+void x87_fsqrt( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xd9, 0xfa);
+}
+
+void x87_fxtract( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xd9, 0xf4);
+}
+
+/* st0 = (2^st0)-1
+ *
+ * Restrictions: -1.0 <= st0 <= 1.0
+ */
+void x87_f2xm1( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xd9, 0xf0);
+}
+
+/* st1 = st1 * log2(st0);
+ * pop_stack;
+ */
+void x87_fyl2x( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xd9, 0xf1);
+   note_x87_pop(p);
+}
+
+/* st1 = st1 * log2(st0 + 1.0);
+ * pop_stack;
+ *
+ * A fast operation, with restrictions: -.29 < st0 < .29 
+ */
+void x87_fyl2xp1( struct x86_function *p )
+{
+   DUMP();
+   emit_2ub(p, 0xd9, 0xf9);
+   note_x87_pop(p);
+}
+
+
+void x87_fld( struct x86_function *p, struct x86_reg arg )
+{
+   DUMP_R( arg );
+   if (arg.file == file_x87) 
+      emit_2ub(p, 0xd9, 0xc0 + arg.idx);
+   else {
+      emit_1ub(p, 0xd9);
+      emit_modrm_noreg(p, 0, arg);
+   }
+   note_x87_push(p);
+}
+
+void x87_fst( struct x86_function *p, struct x86_reg dst )
+{
+   DUMP_R( dst );
+   if (dst.file == file_x87) 
+      emit_2ub(p, 0xdd, 0xd0 + dst.idx);
+   else {
+      emit_1ub(p, 0xd9);
+      emit_modrm_noreg(p, 2, dst);
+   }
+}
+
+void x87_fstp( struct x86_function *p, struct x86_reg dst )
+{
+   DUMP_R( dst );
+   if (dst.file == file_x87) 
+      emit_2ub(p, 0xdd, 0xd8 + dst.idx);
+   else {
+      emit_1ub(p, 0xd9);
+      emit_modrm_noreg(p, 3, dst);
+   }
+   note_x87_pop(p);
+}
+
+void x87_fpop( struct x86_function *p )
+{
+   x87_fstp( p, x86_make_reg( file_x87, 0 ));
+}
+
+
+void x87_fcom( struct x86_function *p, struct x86_reg dst )
+{
+   DUMP_R( dst );
+   if (dst.file == file_x87) 
+      emit_2ub(p, 0xd8, 0xd0 + dst.idx);
+   else {
+      emit_1ub(p, 0xd8);
+      emit_modrm_noreg(p, 2, dst);
+   }
+}
+
+
+void x87_fcomp( struct x86_function *p, struct x86_reg dst )
+{
+   DUMP_R( dst );
+   if (dst.file == file_x87) 
+      emit_2ub(p, 0xd8, 0xd8 + dst.idx);
+   else {
+      emit_1ub(p, 0xd8);
+      emit_modrm_noreg(p, 3, dst);
+   }
+   note_x87_pop(p);
+}
+
+void x87_fcomi( struct x86_function *p, struct x86_reg arg )
+{
+   DUMP_R( arg );
+   emit_2ub(p, 0xdb, 0xf0+arg.idx);
+}
+
+void x87_fcomip( struct x86_function *p, struct x86_reg arg )
+{
+   DUMP_R( arg );
+   emit_2ub(p, 0xdb, 0xf0+arg.idx);
+   note_x87_pop(p);
+}
+
+
+void x87_fnstsw( struct x86_function *p, struct x86_reg dst )
+{
+   DUMP_R( dst );
+   assert(dst.file == file_REG32);
+
+   if (dst.idx == reg_AX &&
+       dst.mod == mod_REG) 
+      emit_2ub(p, 0xdf, 0xe0);
+   else {
+      emit_1ub(p, 0xdd);
+      emit_modrm_noreg(p, 7, dst);
+   }
+}
+
+
+void x87_fnstcw( struct x86_function *p, struct x86_reg dst )
+{
+   DUMP_R( dst );
+   assert(dst.file == file_REG32);
+
+   emit_1ub(p, 0x9b);           /* WAIT -- needed? */
+   emit_1ub(p, 0xd9);
+   emit_modrm_noreg(p, 7, dst);
+}
+
+
+
+
+/***********************************************************************
+ * MMX instructions
+ */
+
+void mmx_emms( struct x86_function *p )
+{
+   DUMP();
+   assert(p->need_emms);
+   emit_2ub(p, 0x0f, 0x77);
+   p->need_emms = 0;
+}
+
+void mmx_packssdw( struct x86_function *p,
+		   struct x86_reg dst,
+		   struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   assert(dst.file == file_MMX && 
+	  (src.file == file_MMX || src.mod != mod_REG));
+
+   p->need_emms = 1;
+
+   emit_2ub(p, X86_TWOB, 0x6b);
+   emit_modrm( p, dst, src );
+}
+
+void mmx_packuswb( struct x86_function *p,
+		   struct x86_reg dst,
+		   struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   assert(dst.file == file_MMX && 
+	  (src.file == file_MMX || src.mod != mod_REG));
+
+   p->need_emms = 1;
+
+   emit_2ub(p, X86_TWOB, 0x67);
+   emit_modrm( p, dst, src );
+}
+
+void mmx_movd( struct x86_function *p,
+	       struct x86_reg dst,
+	       struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   p->need_emms = 1;
+   emit_1ub(p, X86_TWOB);
+   emit_op_modrm( p, 0x6e, 0x7e, dst, src );
+}
+
+void mmx_movq( struct x86_function *p,
+	       struct x86_reg dst,
+	       struct x86_reg src )
+{
+   DUMP_RR( dst, src );
+   p->need_emms = 1;
+   emit_1ub(p, X86_TWOB);
+   emit_op_modrm( p, 0x6f, 0x7f, dst, src );
+}
+
+
+/***********************************************************************
+ * Helper functions
+ */
+
+
+void x86_cdecl_caller_push_regs( struct x86_function *p )
+{
+   x86_push(p, x86_make_reg(file_REG32, reg_AX));
+   x86_push(p, x86_make_reg(file_REG32, reg_CX));
+   x86_push(p, x86_make_reg(file_REG32, reg_DX));
+}
+
+void x86_cdecl_caller_pop_regs( struct x86_function *p )
+{
+   x86_pop(p, x86_make_reg(file_REG32, reg_DX));
+   x86_pop(p, x86_make_reg(file_REG32, reg_CX));
+   x86_pop(p, x86_make_reg(file_REG32, reg_AX));
+}
+
+
+/* Retreive a reference to one of the function arguments, taking into
+ * account any push/pop activity:
+ */
+struct x86_reg x86_fn_arg( struct x86_function *p,
+			   unsigned arg )
+{
+   return x86_make_disp(x86_make_reg(file_REG32, reg_SP), 
+			p->stack_offset + arg * 4);	/* ??? */
+}
+
+
+void x86_init_func( struct x86_function *p )
+{
+   p->size = 0;
+   p->store = NULL;
+   p->csr = p->store;
+   DUMP_START();
+}
+
+void x86_init_func_size( struct x86_function *p, unsigned code_size )
+{
+   p->size = code_size;
+   p->store = rtasm_exec_malloc(code_size);
+   if (p->store == NULL) {
+      p->store = p->error_overflow;
+   }
+   p->csr = p->store;
+   DUMP_START();
+}
+
+void x86_release_func( struct x86_function *p )
+{
+   if (p->store && p->store != p->error_overflow)
+      rtasm_exec_free(p->store);
+
+   p->store = NULL;
+   p->csr = NULL;
+   p->size = 0;
+}
+
+
+void (*x86_get_func( struct x86_function *p ))(void)
+{
+   DUMP_END();
+   if (DISASSEM && p->store)
+      debug_printf("disassemble %p %p\n", p->store, p->csr);
+
+   if (p->store == p->error_overflow)
+      return (void (*)(void)) NULL;
+   else
+      return (void (*)(void)) p->store;
+}
+
+#else
+
+void x86sse_dummy( void )
+{
+}
+
+#endif
diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.h b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h
new file mode 100644
index 00000000000..1b5eaaca850
--- /dev/null
+++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.h
@@ -0,0 +1,319 @@
+/**************************************************************************
+ *
+ * Copyright (C) 1999-2005  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#ifndef _RTASM_X86SSE_H_
+#define _RTASM_X86SSE_H_
+
+#include "pipe/p_config.h"
+
+#if defined(PIPE_ARCH_X86)
+
+/* It is up to the caller to ensure that instructions issued are
+ * suitable for the host cpu.  There are no checks made in this module
+ * for mmx/sse/sse2 support on the cpu.
+ */
+struct x86_reg {
+   unsigned file:3;
+   unsigned idx:3;
+   unsigned mod:2;		/* mod_REG if this is just a register */
+   int      disp:24;		/* only +/- 23bits of offset - should be enough... */
+};
+
+struct x86_function {
+   unsigned size;
+   unsigned char *store;
+   unsigned char *csr;
+
+   unsigned stack_offset:16;
+   unsigned need_emms:8;
+   int x87_stack:8;
+
+   unsigned char error_overflow[4];
+};
+
+enum x86_reg_file {
+   file_REG32,
+   file_MMX,
+   file_XMM,
+   file_x87
+};
+
+/* Values for mod field of modr/m byte
+ */
+enum x86_reg_mod {
+   mod_INDIRECT,
+   mod_DISP8,
+   mod_DISP32,
+   mod_REG
+};
+
+enum x86_reg_name {
+   reg_AX,
+   reg_CX,
+   reg_DX,
+   reg_BX,
+   reg_SP,
+   reg_BP,
+   reg_SI,
+   reg_DI
+};
+
+
+enum x86_cc {
+   cc_O,			/* overflow */
+   cc_NO,			/* not overflow */
+   cc_NAE,			/* not above or equal / carry */
+   cc_AE,			/* above or equal / not carry */
+   cc_E,			/* equal / zero */
+   cc_NE			/* not equal / not zero */
+};
+
+enum sse_cc {
+   cc_Equal,
+   cc_LessThan,
+   cc_LessThanEqual,
+   cc_Unordered,
+   cc_NotEqual,
+   cc_NotLessThan,
+   cc_NotLessThanEqual,
+   cc_Ordered
+};
+
+#define cc_Z  cc_E
+#define cc_NZ cc_NE
+
+/* Begin/end/retreive function creation:
+ */
+
+
+void x86_init_func( struct x86_function *p );
+void x86_init_func_size( struct x86_function *p, unsigned code_size );
+void x86_release_func( struct x86_function *p );
+void (*x86_get_func( struct x86_function *p ))( void );
+
+/* Debugging:
+ */
+void x86_print_reg( struct x86_reg reg );
+
+
+/* Create and manipulate registers and regmem values:
+ */
+struct x86_reg x86_make_reg( enum x86_reg_file file,
+			     enum x86_reg_name idx );
+
+struct x86_reg x86_make_disp( struct x86_reg reg,
+			      int disp );
+
+struct x86_reg x86_deref( struct x86_reg reg );
+
+struct x86_reg x86_get_base_reg( struct x86_reg reg );
+
+
+/* Labels, jumps and fixup:
+ */
+int x86_get_label( struct x86_function *p );
+
+void x86_jcc( struct x86_function *p,
+	      enum x86_cc cc,
+	      int label );
+
+int x86_jcc_forward( struct x86_function *p,
+			  enum x86_cc cc );
+
+int x86_jmp_forward( struct x86_function *p);
+
+int x86_call_forward( struct x86_function *p);
+
+void x86_fixup_fwd_jump( struct x86_function *p,
+			 int fixup );
+
+void x86_jmp( struct x86_function *p, int label );
+
+/* void x86_call( struct x86_function *p, void (*label)() ); */
+void x86_call( struct x86_function *p, struct x86_reg reg);
+
+void x86_mov_reg_imm( struct x86_function *p, struct x86_reg dst, int imm );
+void x86_add_imm( struct x86_function *p, struct x86_reg dst, int imm );
+void x86_or_imm( struct x86_function *p, struct x86_reg dst, int imm );
+void x86_and_imm( struct x86_function *p, struct x86_reg dst, int imm );
+void x86_sub_imm( struct x86_function *p, struct x86_reg dst, int imm );
+void x86_xor_imm( struct x86_function *p, struct x86_reg dst, int imm );
+void x86_cmp_imm( struct x86_function *p, struct x86_reg dst, int imm );
+
+
+/* Macro for sse_shufps() and sse2_pshufd():
+ */
+#define SHUF(_x,_y,_z,_w)       (((_x)<<0) | ((_y)<<2) | ((_z)<<4) | ((_w)<<6))
+#define SHUF_NOOP               RSW(0,1,2,3)
+#define GET_SHUF(swz, idx)      (((swz) >> ((idx)*2)) & 0x3)
+
+void mmx_emms( struct x86_function *p );
+void mmx_movd( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void mmx_movq( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void mmx_packssdw( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void mmx_packuswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+
+void sse2_cvtps2dq( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_cvttps2dq( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_cvtdq2ps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_movd( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_packssdw( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_packsswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_packuswb( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_pshufd( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0,
+                  unsigned char shuf );
+void sse2_rcpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse2_rcpss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+
+
+void sse_prefetchnta( struct x86_function *p, struct x86_reg ptr);
+void sse_prefetch0( struct x86_function *p, struct x86_reg ptr);
+void sse_prefetch1( struct x86_function *p, struct x86_reg ptr);
+
+void sse_movntps( struct x86_function *p, struct x86_reg dst, struct x86_reg src);
+
+void sse_addps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_addss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_cvtps2pi( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_divss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_andnps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_andps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_cmpps( struct x86_function *p, struct x86_reg dst, struct x86_reg src,
+                enum sse_cc cc );
+void sse_maxps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_maxss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_minps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movaps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movhlps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movhps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movlhps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movlps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_movups( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_mulps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_mulss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_orps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_xorps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_subps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_rsqrtps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_rsqrtss( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_shufps( struct x86_function *p, struct x86_reg dest, struct x86_reg arg0,
+                 unsigned char shuf );
+void sse_unpckhps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_unpcklps( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void sse_pmovmskb( struct x86_function *p, struct x86_reg dest, struct x86_reg src );
+void sse2_punpcklbw( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+
+void x86_add( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_and( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_cmp( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_dec( struct x86_function *p, struct x86_reg reg );
+void x86_inc( struct x86_function *p, struct x86_reg reg );
+void x86_lea( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_mov( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_mul( struct x86_function *p, struct x86_reg src );
+void x86_imul( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_or( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_pop( struct x86_function *p, struct x86_reg reg );
+void x86_push( struct x86_function *p, struct x86_reg reg );
+void x86_push_imm32( struct x86_function *p, int imm );
+void x86_ret( struct x86_function *p );
+void x86_retw( struct x86_function *p, unsigned short imm );
+void x86_sub( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_test( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_xor( struct x86_function *p, struct x86_reg dst, struct x86_reg src );
+void x86_sahf( struct x86_function *p );
+
+
+void x86_cdecl_caller_push_regs( struct x86_function *p );
+void x86_cdecl_caller_pop_regs( struct x86_function *p );
+
+void x87_assert_stack_empty( struct x86_function *p );
+
+void x87_f2xm1( struct x86_function *p );
+void x87_fabs( struct x86_function *p );
+void x87_fadd( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_faddp( struct x86_function *p, struct x86_reg dst );
+void x87_fchs( struct x86_function *p );
+void x87_fclex( struct x86_function *p );
+void x87_fcmovb( struct x86_function *p, struct x86_reg src );
+void x87_fcmovbe( struct x86_function *p, struct x86_reg src );
+void x87_fcmove( struct x86_function *p, struct x86_reg src );
+void x87_fcmovnb( struct x86_function *p, struct x86_reg src );
+void x87_fcmovnbe( struct x86_function *p, struct x86_reg src );
+void x87_fcmovne( struct x86_function *p, struct x86_reg src );
+void x87_fcom( struct x86_function *p, struct x86_reg dst );
+void x87_fcomi( struct x86_function *p, struct x86_reg dst );
+void x87_fcomip( struct x86_function *p, struct x86_reg dst );
+void x87_fcomp( struct x86_function *p, struct x86_reg dst );
+void x87_fcos( struct x86_function *p );
+void x87_fdiv( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_fdivp( struct x86_function *p, struct x86_reg dst );
+void x87_fdivr( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_fdivrp( struct x86_function *p, struct x86_reg dst );
+void x87_fild( struct x86_function *p, struct x86_reg arg );
+void x87_fist( struct x86_function *p, struct x86_reg dst );
+void x87_fistp( struct x86_function *p, struct x86_reg dst );
+void x87_fld( struct x86_function *p, struct x86_reg arg );
+void x87_fld1( struct x86_function *p );
+void x87_fldcw( struct x86_function *p, struct x86_reg arg );
+void x87_fldl2e( struct x86_function *p );
+void x87_fldln2( struct x86_function *p );
+void x87_fldz( struct x86_function *p );
+void x87_fmul( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_fmulp( struct x86_function *p, struct x86_reg dst );
+void x87_fnclex( struct x86_function *p );
+void x87_fprndint( struct x86_function *p );
+void x87_fpop( struct x86_function *p );
+void x87_fscale( struct x86_function *p );
+void x87_fsin( struct x86_function *p );
+void x87_fsincos( struct x86_function *p );
+void x87_fsqrt( struct x86_function *p );
+void x87_fst( struct x86_function *p, struct x86_reg dst );
+void x87_fstp( struct x86_function *p, struct x86_reg dst );
+void x87_fsub( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_fsubp( struct x86_function *p, struct x86_reg dst );
+void x87_fsubr( struct x86_function *p, struct x86_reg dst, struct x86_reg arg );
+void x87_fsubrp( struct x86_function *p, struct x86_reg dst );
+void x87_ftst( struct x86_function *p );
+void x87_fxch( struct x86_function *p, struct x86_reg dst );
+void x87_fxtract( struct x86_function *p );
+void x87_fyl2x( struct x86_function *p );
+void x87_fyl2xp1( struct x86_function *p );
+void x87_fwait( struct x86_function *p );
+void x87_fnstcw( struct x86_function *p, struct x86_reg dst );
+void x87_fnstsw( struct x86_function *p, struct x86_reg dst );
+void x87_fucompp( struct x86_function *p );
+void x87_fucomp( struct x86_function *p, struct x86_reg arg );
+void x87_fucom( struct x86_function *p, struct x86_reg arg );
+
+
+
+/* Retreive a reference to one of the function arguments, taking into
+ * account any push/pop activity.  Note - doesn't track explict
+ * manipulation of ESP by other instructions.
+ */
+struct x86_reg x86_fn_arg( struct x86_function *p, unsigned arg );
+
+#endif
+#endif
diff --git a/src/gallium/auxiliary/sct/Makefile b/src/gallium/auxiliary/sct/Makefile
new file mode 100644
index 00000000000..516d1756cf9
--- /dev/null
+++ b/src/gallium/auxiliary/sct/Makefile
@@ -0,0 +1,12 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = sct
+
+C_SOURCES = \
+	sct.c
+
+include ../../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/auxiliary/sct/SConscript b/src/gallium/auxiliary/sct/SConscript
new file mode 100644
index 00000000000..76927d973f8
--- /dev/null
+++ b/src/gallium/auxiliary/sct/SConscript
@@ -0,0 +1,9 @@
+Import('*')
+
+sct = env.ConvenienceLibrary(
+	target = 'sct',
+	source = [
+		'sct.c'
+	])
+
+auxiliaries.insert(0, sct)
diff --git a/src/gallium/auxiliary/sct/sct.c b/src/gallium/auxiliary/sct/sct.c
new file mode 100644
index 00000000000..49bb7ea92e2
--- /dev/null
+++ b/src/gallium/auxiliary/sct/sct.c
@@ -0,0 +1,454 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "util/u_memory.h"
+#include "pipe/p_state.h"
+#include "pipe/p_inlines.h"
+#include "sct.h"
+
+
+struct texture_list
+{
+   struct pipe_texture *texture;
+   struct texture_list *next;
+};
+
+
+
+#define MAX_SURFACES  ((PIPE_MAX_COLOR_BUFS) + 1)
+
+struct sct_context
+{
+   const struct pipe_context *context;
+
+   /** surfaces the context is drawing into */
+   struct pipe_surface *surfaces[MAX_SURFACES];
+
+   /** currently bound textures */
+   struct pipe_texture *textures[PIPE_MAX_SAMPLERS];
+
+   /** previously bound textures, used but not flushed */
+   struct texture_list *textures_used;
+
+   boolean needs_flush;
+
+   struct sct_context *next;
+};
+
+
+
+struct sct_surface
+{
+   const struct pipe_surface *surface;
+
+   /** list of contexts drawing to this surface */
+   struct sct_context_list *contexts;
+
+   struct sct_surface *next;
+};
+
+
+
+/**
+ * Find the surface_info for the given pipe_surface
+ */
+static struct sct_surface *
+find_surface_info(struct surface_context_tracker *sct,
+                  const struct pipe_surface *surface)
+{
+   struct sct_surface *si;
+   for (si = sct->surfaces; si; si = si->next)
+      if (si->surface == surface)
+         return si;
+   return NULL;
+}
+
+
+/**
+ * As above, but create new surface_info if surface is new.
+ */
+static struct sct_surface *
+find_create_surface_info(struct surface_context_tracker *sct,
+                         const struct pipe_surface *surface)
+{
+   struct sct_surface *si = find_surface_info(sct, surface);
+   if (si)
+      return si;
+
+   /* alloc new */
+   si = CALLOC_STRUCT(sct_surface);
+   if (si) {
+      si->surface = surface;
+
+      /* insert at head */
+      si->next = sct->surfaces;
+      sct->surfaces = si;
+   }
+
+   return si;
+}
+
+
+/**
+ * Find a context_info for the given context.
+ */
+static struct sct_context *
+find_context_info(struct surface_context_tracker *sct,
+                  const struct pipe_context *context)
+{
+   struct sct_context *ci;
+   for (ci = sct->contexts; ci; ci = ci->next)
+      if (ci->context == context)
+         return ci;
+   return NULL;
+}
+
+
+/**
+ * As above, but create new context_info if context is new.
+ */
+static struct sct_context *
+find_create_context_info(struct surface_context_tracker *sct,
+                         const struct pipe_context *context)
+{
+   struct sct_context *ci = find_context_info(sct, context);
+   if (ci)
+      return ci;
+
+   /* alloc new */
+   ci = CALLOC_STRUCT(sct_context);
+   if (ci) {
+      ci->context = context;
+
+      /* insert at head */
+      ci->next = sct->contexts;
+      sct->contexts = ci;
+   }
+
+   return ci;
+}
+
+
+/**
+ * Is the context already bound to the surface?
+ */
+static boolean
+find_surface_context(const struct sct_surface *si,
+                     const struct pipe_context *context)
+{
+   const struct sct_context_list *cl;
+   for (cl = si->contexts; cl; cl = cl->next) {
+      if (cl->context == context) {
+         return TRUE;
+      }
+   }
+   return FALSE;
+}
+
+
+/**
+ * Add a context to the list of contexts associated with a surface.
+ */
+static void
+add_context_to_surface(struct sct_surface *si,
+                       const struct pipe_context *context)
+{
+   struct sct_context_list *cl = CALLOC_STRUCT(sct_context_list);
+   if (cl) {
+      cl->context = context;
+      /* insert at head of list of contexts */
+      cl->next = si->contexts;
+      si->contexts = cl;
+   }
+}
+
+
+/**
+ * Remove a context from the list of contexts associated with a surface.
+ */
+static void
+remove_context_from_surface(struct sct_surface *si,
+                            const struct pipe_context *context)
+{
+   struct sct_context_list *prev = NULL, *curr, *next;
+
+   for (curr = si->contexts; curr; curr = next) {
+      if (curr->context == context) {
+         /* remove */
+         if (prev)
+            prev->next = curr->next;
+         else
+            si->contexts = curr->next;
+         next = curr->next;
+         FREE(curr);
+      }
+      else {
+         prev = curr;
+         next = curr->next;
+      }
+   }
+}
+
+
+/**
+ * Unbind context from surface.
+ */
+static void
+unbind_context_surface(struct surface_context_tracker *sct,
+                       struct pipe_context *context,
+                       struct pipe_surface *surface)
+{
+   struct sct_surface *si = find_surface_info(sct, surface);
+   if (si) {
+      remove_context_from_surface(si, context);
+   }
+}
+
+
+/**
+ * Bind context to a set of surfaces (color + Z).
+ * Like MakeCurrent().
+ */
+void
+sct_bind_surfaces(struct surface_context_tracker *sct,
+                  struct pipe_context *context,
+                  uint num_surf,
+                  struct pipe_surface **surfaces)
+{
+   struct sct_context *ci = find_create_context_info(sct, context);
+   uint i;
+
+   if (!ci) {
+      return; /* out of memory */
+   }
+
+   /* unbind currently bound surfaces */
+   for (i = 0; i < MAX_SURFACES; i++) {
+      if (ci->surfaces[i]) {
+         unbind_context_surface(sct, context, ci->surfaces[i]);
+      }
+   }
+
+   /* bind new surfaces */
+   for (i = 0; i < num_surf; i++) {
+      struct sct_surface *si = find_create_surface_info(sct, surfaces[i]);
+      if (!find_surface_context(si, context)) {
+         add_context_to_surface(si, context);
+      }
+   }
+}
+
+
+/**
+ * Return list of contexts bound to a surface.
+ */
+const struct sct_context_list *
+sct_get_surface_contexts(struct surface_context_tracker *sct,
+                         const struct pipe_surface *surface)
+{
+   const struct sct_surface *si = find_surface_info(sct, surface);
+   return si->contexts;
+}
+
+
+
+static boolean
+find_texture(const struct sct_context *ci,
+             const struct pipe_texture *texture)
+{
+   const struct texture_list *tl;
+
+   for (tl = ci->textures_used; tl; tl = tl->next) {
+      if (tl->texture == texture) {
+         return TRUE;
+      }
+   }
+   return FALSE;
+}
+
+
+/**
+ * Add the given texture to the context's list of used textures.
+ */
+static void
+add_texture_used(struct sct_context *ci,
+                 struct pipe_texture *texture)
+{
+   if (!find_texture(ci, texture)) {
+      /* add to list */
+      struct texture_list *tl = CALLOC_STRUCT(texture_list);
+      if (tl) {
+         pipe_texture_reference(&tl->texture, texture);
+         /* insert at head */
+         tl->next = ci->textures_used;
+         ci->textures_used = tl;
+      }
+   }
+}
+
+
+/**
+ * Bind a texture to a rendering context.
+ */
+void
+sct_bind_texture(struct surface_context_tracker *sct,
+                 struct pipe_context *context,
+                 uint unit,
+                 struct pipe_texture *tex)
+{
+   struct sct_context *ci = find_context_info(sct, context);
+
+   if (ci->textures[unit] != tex) {
+      /* put texture on the 'used' list */
+      add_texture_used(ci, tex);
+      /* bind new */
+      pipe_texture_reference(&ci->textures[unit], tex);
+   }
+}
+
+
+/**
+ * Check if the given texture has been used by the rendering context
+ * since the last call to sct_flush_textures().
+ */
+boolean
+sct_is_texture_used(struct surface_context_tracker *sct,
+                    const struct pipe_context *context,
+                    const struct pipe_texture *texture)
+{
+   const struct sct_context *ci = find_context_info(sct, context);
+   return find_texture(ci, texture);
+}
+
+
+/**
+ * To be called when the image contents of a texture are changed, such
+ * as for gl[Copy]TexSubImage().
+ * XXX this may not be needed
+ */
+void
+sct_update_texture(struct pipe_texture *tex)
+{
+
+}
+
+
+/**
+ * When a scene is flushed/rendered we can release the list of
+ * used textures.
+ */
+void
+sct_flush_textures(struct surface_context_tracker *sct,
+                   struct pipe_context *context)
+{
+   struct sct_context *ci = find_context_info(sct, context);
+   struct texture_list *tl, *next;
+   uint i;
+
+   for (tl = ci->textures_used; tl; tl = next) {
+      next = tl->next;
+      pipe_texture_release(&tl->texture);
+      FREE(tl);
+   }
+   ci->textures_used = NULL;
+
+   /* put the currently bound textures on the 'used' list */
+   for (i = 0; i < PIPE_MAX_SAMPLERS; i++) {
+      add_texture_used(ci, ci->textures[i]);
+   }
+}
+
+
+
+void
+sct_destroy_context(struct surface_context_tracker *sct,
+                    struct pipe_context *context)
+{
+   /* XXX should we require an unbinding first? */
+   {
+      struct sct_surface *si;
+      for (si = sct->surfaces; si; si = si->next) {
+         remove_context_from_surface(si, context);
+      }
+   }
+
+   /* remove context from context_info list */
+   {
+      struct sct_context *ci, *next, *prev = NULL;
+      for (ci = sct->contexts; ci; ci = next) {
+         next = ci->next;
+         if (ci->context == context) {
+            if (prev)
+               prev->next = ci->next;
+            else
+               sct->contexts = ci->next;
+            FREE(ci);
+         }
+         else {
+            prev = ci;
+         }
+      }
+   }
+
+}
+
+
+void
+sct_destroy_surface(struct surface_context_tracker *sct,
+                    struct pipe_surface *surface)
+{
+   if (1) {
+      /* debug/sanity: no context should be bound to surface */
+      struct sct_context *ci;
+      uint i;
+      for (ci = sct->contexts; ci; ci = ci->next) {
+         for (i = 0; i < MAX_SURFACES; i++) {
+            assert(ci->surfaces[i] != surface);
+         }
+      }
+   }
+
+   /* remove surface from sct_surface list */
+   {
+      struct sct_surface *si, *next, *prev = NULL;
+      for (si = sct->surfaces; si; si = next) {
+         next = si->next;
+         if (si->surface == surface) {
+            /* unlink */
+            if (prev)
+               prev->next = si->next;
+            else
+               sct->surfaces = si->next;
+            FREE(si);
+         }
+         else {
+            prev = si;
+         }
+      }
+   }
+}
diff --git a/src/gallium/auxiliary/sct/sct.h b/src/gallium/auxiliary/sct/sct.h
new file mode 100644
index 00000000000..cf7c4d3bdfd
--- /dev/null
+++ b/src/gallium/auxiliary/sct/sct.h
@@ -0,0 +1,123 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * Surface/Context Tracking
+ *
+ * For some drivers, we need to monitor the binding between contexts and
+ * surfaces/textures.
+ * This code may evolve quite a bit...
+ */
+
+
+#ifndef SCT_H
+#define SCT_H
+
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+
+struct pipe_context;
+struct pipe_surface;
+
+struct sct_context;
+struct sct_surface;
+
+
+/**
+ * Per-device info, basically
+ */
+struct surface_context_tracker
+{
+   struct sct_context *contexts;
+   struct sct_surface *surfaces;
+};
+
+
+
+/**
+ * Simple linked list of contexts
+ */
+struct sct_context_list
+{
+   const struct pipe_context *context;
+   struct sct_context_list *next;
+};
+
+
+
+extern void
+sct_bind_surfaces(struct surface_context_tracker *sct,
+                  struct pipe_context *context,
+                  uint num_surf,
+                  struct pipe_surface **surfaces);
+
+
+extern void
+sct_bind_texture(struct surface_context_tracker *sct,
+                 struct pipe_context *context,
+                 uint unit,
+                 struct pipe_texture *texture);
+
+
+extern void
+sct_update_texture(struct pipe_texture *tex);
+
+
+extern boolean
+sct_is_texture_used(struct surface_context_tracker *sct,
+                    const struct pipe_context *context,
+                    const struct pipe_texture *texture);
+
+extern void
+sct_flush_textures(struct surface_context_tracker *sct,
+                   struct pipe_context *context);
+
+
+extern const struct sct_context_list *
+sct_get_surface_contexts(struct surface_context_tracker *sct,
+                         const struct pipe_surface *surf);
+
+
+extern void
+sct_destroy_context(struct surface_context_tracker *sct,
+                    struct pipe_context *context);
+
+
+extern void
+sct_destroy_surface(struct surface_context_tracker *sct,
+                    struct pipe_surface *surface);
+
+
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif /* SCT_H */
diff --git a/src/gallium/auxiliary/sct/usage.c b/src/gallium/auxiliary/sct/usage.c
new file mode 100644
index 00000000000..6227f199628
--- /dev/null
+++ b/src/gallium/auxiliary/sct/usage.c
@@ -0,0 +1,61 @@
+/* surface / context tracking */
+
+
+/*
+
+context A:
+  render to texture T
+
+context B:
+  texture from T
+
+-----------------------
+
+flush surface:
+  which contexts are bound to the surface?
+
+-----------------------
+
+glTexSubImage():
+  which contexts need to be flushed?
+
+ */
+
+
+/*
+
+in MakeCurrent():
+
+  call sct_bind_surfaces(context, list of surfaces) to update the
+  dependencies between context and surfaces
+
+
+in SurfaceFlush(), or whatever it is in D3D:
+
+  call sct_get_surface_contexts(surface) to get a list of contexts
+  which are currently bound to the surface.
+
+
+
+in BindTexture():
+
+  call sct_bind_texture(context, texture) to indicate that the texture
+  is used in the scene.
+
+
+in glTexSubImage() or RenderToTexture():
+
+  call sct_is_texture_used(context, texture) to determine if the texture
+  has been used in the scene, but the scene's not flushed.  If TRUE is
+  returned it means the scene has to be rendered/flushed before the contents
+  of the texture can be changed.
+
+
+in psb_scene_flush/terminate():
+
+  call sct_flush_textures(context) to tell the SCT that the textures which
+  were used in the scene can be released.
+
+
+
+*/
diff --git a/src/gallium/auxiliary/tgsi/Makefile b/src/gallium/auxiliary/tgsi/Makefile
new file mode 100644
index 00000000000..d7df9490cfa
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/Makefile
@@ -0,0 +1,24 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = tgsi
+
+C_SOURCES = \
+	tgsi_sanity.c \
+	tgsi_build.c \
+	tgsi_dump.c \
+	tgsi_exec.c \
+	tgsi_info.c \
+	tgsi_iterate.c \
+	tgsi_parse.c \
+	tgsi_ppc.c \
+	tgsi_scan.c \
+	tgsi_sse2.c \
+	tgsi_text.c \
+	tgsi_transform.c \
+	tgsi_util.c
+
+include ../../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/auxiliary/tgsi/SConscript b/src/gallium/auxiliary/tgsi/SConscript
new file mode 100644
index 00000000000..8200cce42f5
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/SConscript
@@ -0,0 +1,22 @@
+Import('*')
+
+tgsi = env.ConvenienceLibrary(
+	target = 'tgsi',
+	source = [
+		'tgsi_build.c',
+		'tgsi_dump.c',
+		'tgsi_dump_c.c',
+		'tgsi_exec.c',
+		'tgsi_info.c',
+		'tgsi_iterate.c',
+		'tgsi_parse.c',
+		'tgsi_sanity.c',
+		'tgsi_scan.c',
+		'tgsi_ppc.c',
+		'tgsi_sse2.c',
+		'tgsi_text.c',
+		'tgsi_transform.c',
+		'tgsi_util.c',
+	])
+
+auxiliaries.insert(0, tgsi)
diff --git a/src/gallium/auxiliary/tgsi/tgsi_build.c b/src/gallium/auxiliary/tgsi/tgsi_build.c
new file mode 100644
index 00000000000..17886540cf7
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_build.c
@@ -0,0 +1,1335 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_debug.h"
+#include "pipe/p_shader_tokens.h"
+#include "tgsi_build.h"
+#include "tgsi_parse.h"
+
+/*
+ * version
+ */
+
+struct tgsi_version
+tgsi_build_version( void )
+{
+   struct tgsi_version  version;
+
+   version.MajorVersion = 1;
+   version.MinorVersion = 1;
+   version.Padding = 0;
+
+   return version;
+}
+
+/*
+ * header
+ */
+
+struct tgsi_header
+tgsi_build_header( void )
+{
+   struct tgsi_header header;
+
+   header.HeaderSize = 1;
+   header.BodySize = 0;
+
+   return header;
+}
+
+static void
+header_headersize_grow( struct tgsi_header *header )
+{
+   assert( header->HeaderSize < 0xFF );
+   assert( header->BodySize == 0 );
+
+   header->HeaderSize++;
+}
+
+static void
+header_bodysize_grow( struct tgsi_header *header )
+{
+   assert( header->BodySize < 0xFFFFFF );
+
+   header->BodySize++;
+}
+
+struct tgsi_processor
+tgsi_default_processor( void )
+{
+   struct tgsi_processor processor;
+
+   processor.Processor = TGSI_PROCESSOR_FRAGMENT;
+   processor.Padding = 0;
+
+   return processor;
+}
+
+struct tgsi_processor
+tgsi_build_processor(
+   unsigned type,
+   struct tgsi_header *header )
+{
+   struct tgsi_processor processor;
+
+   processor = tgsi_default_processor();
+   processor.Processor = type;
+
+   header_headersize_grow( header );
+
+   return processor;
+}
+
+/*
+ * declaration
+ */
+
+struct tgsi_declaration
+tgsi_default_declaration( void )
+{
+   struct tgsi_declaration declaration;
+
+   declaration.Type = TGSI_TOKEN_TYPE_DECLARATION;
+   declaration.NrTokens = 1;
+   declaration.File = TGSI_FILE_NULL;
+   declaration.UsageMask = TGSI_WRITEMASK_XYZW;
+   declaration.Interpolate = TGSI_INTERPOLATE_CONSTANT;
+   declaration.Semantic = 0;
+   declaration.Centroid = 0;
+   declaration.Invariant = 0;
+   declaration.Padding = 0;
+   declaration.Extended = 0;
+
+   return declaration;
+}
+
+struct tgsi_declaration
+tgsi_build_declaration(
+   unsigned file,
+   unsigned usage_mask,
+   unsigned interpolate,
+   unsigned semantic,
+   unsigned centroid,
+   unsigned invariant,
+   struct tgsi_header *header )
+{
+   struct tgsi_declaration declaration;
+
+   assert( file <= TGSI_FILE_IMMEDIATE );
+   assert( interpolate <= TGSI_INTERPOLATE_PERSPECTIVE );
+
+   declaration = tgsi_default_declaration();
+   declaration.File = file;
+   declaration.UsageMask = usage_mask;
+   declaration.Interpolate = interpolate;
+   declaration.Semantic = semantic;
+   declaration.Centroid = centroid;
+   declaration.Invariant = invariant;
+
+   header_bodysize_grow( header );
+
+   return declaration;
+}
+
+static void
+declaration_grow(
+   struct tgsi_declaration *declaration,
+   struct tgsi_header *header )
+{
+   assert( declaration->NrTokens < 0xFF );
+
+   declaration->NrTokens++;
+
+   header_bodysize_grow( header );
+}
+
+struct tgsi_full_declaration
+tgsi_default_full_declaration( void )
+{
+   struct tgsi_full_declaration  full_declaration;
+
+   full_declaration.Declaration  = tgsi_default_declaration();
+   full_declaration.DeclarationRange = tgsi_default_declaration_range();
+   full_declaration.Semantic = tgsi_default_declaration_semantic();
+
+   return full_declaration;
+}
+
+unsigned
+tgsi_build_full_declaration(
+   const struct tgsi_full_declaration *full_decl,
+   struct tgsi_token *tokens,
+   struct tgsi_header *header,
+   unsigned maxsize )
+{
+   unsigned size = 0;
+   struct tgsi_declaration *declaration;
+   struct tgsi_declaration_range *dr;
+
+   if( maxsize <= size )
+     return 0;
+   declaration = (struct tgsi_declaration *) &tokens[size];
+   size++;
+
+   *declaration = tgsi_build_declaration(
+      full_decl->Declaration.File,
+      full_decl->Declaration.UsageMask,
+      full_decl->Declaration.Interpolate,
+      full_decl->Declaration.Semantic,
+      full_decl->Declaration.Centroid,
+      full_decl->Declaration.Invariant,
+      header );
+
+   if (maxsize <= size)
+      return 0;
+   dr = (struct tgsi_declaration_range *) &tokens[size];
+   size++;
+
+   *dr = tgsi_build_declaration_range(
+      full_decl->DeclarationRange.First,
+      full_decl->DeclarationRange.Last,
+      declaration,
+      header );
+
+   if( full_decl->Declaration.Semantic ) {
+      struct tgsi_declaration_semantic *ds;
+
+      if( maxsize <= size )
+         return  0;
+      ds = (struct tgsi_declaration_semantic *) &tokens[size];
+      size++;
+
+      *ds = tgsi_build_declaration_semantic(
+         full_decl->Semantic.SemanticName,
+         full_decl->Semantic.SemanticIndex,
+         declaration,
+         header );
+   }
+
+   return size;
+}
+
+struct tgsi_declaration_range
+tgsi_default_declaration_range( void )
+{
+   struct tgsi_declaration_range dr;
+
+   dr.First = 0;
+   dr.Last = 0;
+
+   return dr;
+}
+
+struct tgsi_declaration_range
+tgsi_build_declaration_range(
+   unsigned first,
+   unsigned last,
+   struct tgsi_declaration *declaration,
+   struct tgsi_header *header )
+{
+   struct tgsi_declaration_range declaration_range;
+
+   assert( last >= first );
+   assert( last <= 0xFFFF );
+
+   declaration_range = tgsi_default_declaration_range();
+   declaration_range.First = first;
+   declaration_range.Last = last;
+
+   declaration_grow( declaration, header );
+
+   return declaration_range;
+}
+
+struct tgsi_declaration_semantic
+tgsi_default_declaration_semantic( void )
+{
+   struct tgsi_declaration_semantic ds;
+
+   ds.SemanticName = TGSI_SEMANTIC_POSITION;
+   ds.SemanticIndex = 0;
+   ds.Padding = 0;
+
+   return ds;
+}
+
+struct tgsi_declaration_semantic
+tgsi_build_declaration_semantic(
+   unsigned semantic_name,
+   unsigned semantic_index,
+   struct tgsi_declaration *declaration,
+   struct tgsi_header *header )
+{
+   struct tgsi_declaration_semantic ds;
+
+   assert( semantic_name <= TGSI_SEMANTIC_COUNT );
+   assert( semantic_index <= 0xFFFF );
+
+   ds = tgsi_default_declaration_semantic();
+   ds.SemanticName = semantic_name;
+   ds.SemanticIndex = semantic_index;
+
+   declaration_grow( declaration, header );
+
+   return ds;
+}
+
+/*
+ * immediate
+ */
+
+struct tgsi_immediate
+tgsi_default_immediate( void )
+{
+   struct tgsi_immediate immediate;
+
+   immediate.Type = TGSI_TOKEN_TYPE_IMMEDIATE;
+   immediate.NrTokens = 1;
+   immediate.DataType = TGSI_IMM_FLOAT32;
+   immediate.Padding = 0;
+   immediate.Extended = 0;
+
+   return immediate;
+}
+
+struct tgsi_immediate
+tgsi_build_immediate(
+   struct tgsi_header *header )
+{
+   struct tgsi_immediate immediate;
+
+   immediate = tgsi_default_immediate();
+
+   header_bodysize_grow( header );
+
+   return immediate;
+}
+
+struct tgsi_full_immediate
+tgsi_default_full_immediate( void )
+{
+   struct tgsi_full_immediate fullimm;
+
+   fullimm.Immediate = tgsi_default_immediate();
+   fullimm.u.Pointer = (void *) 0;
+
+   return fullimm;
+}
+
+static void
+immediate_grow(
+   struct tgsi_immediate *immediate,
+   struct tgsi_header *header )
+{
+   assert( immediate->NrTokens < 0xFF );
+
+   immediate->NrTokens++;
+
+   header_bodysize_grow( header );
+}
+
+struct tgsi_immediate_float32
+tgsi_build_immediate_float32(
+   float value,
+   struct tgsi_immediate *immediate,
+   struct tgsi_header *header )
+{
+   struct tgsi_immediate_float32 immediate_float32;
+
+   immediate_float32.Float = value;
+
+   immediate_grow( immediate, header );
+
+   return immediate_float32;
+}
+
+unsigned
+tgsi_build_full_immediate(
+   const struct tgsi_full_immediate *full_imm,
+   struct tgsi_token *tokens,
+   struct tgsi_header *header,
+   unsigned maxsize )
+{
+   unsigned size = 0, i;
+   struct tgsi_immediate *immediate;
+
+   if( maxsize <= size )
+      return 0;
+   immediate = (struct tgsi_immediate *) &tokens[size];
+   size++;
+
+   *immediate = tgsi_build_immediate( header );
+
+   for( i = 0; i < full_imm->Immediate.NrTokens - 1; i++ ) {
+      struct tgsi_immediate_float32 *if32;
+
+      if( maxsize <= size )
+         return  0;
+      if32 = (struct tgsi_immediate_float32 *) &tokens[size];
+      size++;
+
+      *if32 = tgsi_build_immediate_float32(
+         full_imm->u.ImmediateFloat32[i].Float,
+         immediate,
+         header );
+   }
+
+   return size;
+}
+
+/*
+ * instruction
+ */
+
+struct tgsi_instruction
+tgsi_default_instruction( void )
+{
+   struct tgsi_instruction instruction;
+
+   instruction.Type = TGSI_TOKEN_TYPE_INSTRUCTION;
+   instruction.NrTokens = 1;
+   instruction.Opcode = TGSI_OPCODE_MOV;
+   instruction.Saturate = TGSI_SAT_NONE;
+   instruction.NumDstRegs = 1;
+   instruction.NumSrcRegs = 1;
+   instruction.Padding  = 0;
+   instruction.Extended = 0;
+
+   return instruction;
+}
+
+struct tgsi_instruction
+tgsi_build_instruction(
+   unsigned opcode,
+   unsigned saturate,
+   unsigned num_dst_regs,
+   unsigned num_src_regs,
+   struct tgsi_header *header )
+{
+   struct tgsi_instruction instruction;
+
+   assert (opcode <= TGSI_OPCODE_LAST);
+   assert (saturate <= TGSI_SAT_MINUS_PLUS_ONE);
+   assert (num_dst_regs <= 3);
+   assert (num_src_regs <= 15);
+
+   instruction = tgsi_default_instruction();
+   instruction.Opcode = opcode;
+   instruction.Saturate = saturate;
+   instruction.NumDstRegs = num_dst_regs;
+   instruction.NumSrcRegs = num_src_regs;
+
+   header_bodysize_grow( header );
+
+   return instruction;
+}
+
+static void
+instruction_grow(
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   assert (instruction->NrTokens <   0xFF);
+
+   instruction->NrTokens++;
+
+   header_bodysize_grow( header );
+}
+
+struct tgsi_full_instruction
+tgsi_default_full_instruction( void )
+{
+   struct tgsi_full_instruction full_instruction;
+   unsigned i;
+
+   full_instruction.Instruction = tgsi_default_instruction();
+   full_instruction.InstructionExtNv = tgsi_default_instruction_ext_nv();
+   full_instruction.InstructionExtLabel = tgsi_default_instruction_ext_label();
+   full_instruction.InstructionExtTexture = tgsi_default_instruction_ext_texture();
+   for( i = 0;  i < TGSI_FULL_MAX_DST_REGISTERS; i++ ) {
+      full_instruction.FullDstRegisters[i] = tgsi_default_full_dst_register();
+   }
+   for( i = 0;  i < TGSI_FULL_MAX_SRC_REGISTERS; i++ ) {
+      full_instruction.FullSrcRegisters[i] = tgsi_default_full_src_register();
+   }
+
+   return full_instruction;
+}
+
+unsigned
+tgsi_build_full_instruction(
+   const struct tgsi_full_instruction *full_inst,
+   struct  tgsi_token *tokens,
+   struct  tgsi_header *header,
+   unsigned  maxsize )
+{
+   unsigned size = 0;
+   unsigned i;
+   struct tgsi_instruction *instruction;
+   struct tgsi_token *prev_token;
+
+   if( maxsize <= size )
+      return 0;
+   instruction = (struct tgsi_instruction *) &tokens[size];
+   size++;
+
+   *instruction = tgsi_build_instruction(
+      full_inst->Instruction.Opcode,
+      full_inst->Instruction.Saturate,
+      full_inst->Instruction.NumDstRegs,
+      full_inst->Instruction.NumSrcRegs,
+      header );
+   prev_token = (struct tgsi_token  *) instruction;
+
+   if( tgsi_compare_instruction_ext_nv(
+         full_inst->InstructionExtNv,
+         tgsi_default_instruction_ext_nv() ) ) {
+      struct tgsi_instruction_ext_nv *instruction_ext_nv;
+
+      if( maxsize <= size )
+         return 0;
+      instruction_ext_nv =
+         (struct  tgsi_instruction_ext_nv *) &tokens[size];
+      size++;
+
+      *instruction_ext_nv  = tgsi_build_instruction_ext_nv(
+         full_inst->InstructionExtNv.Precision,
+         full_inst->InstructionExtNv.CondDstIndex,
+         full_inst->InstructionExtNv.CondFlowIndex,
+         full_inst->InstructionExtNv.CondMask,
+         full_inst->InstructionExtNv.CondSwizzleX,
+         full_inst->InstructionExtNv.CondSwizzleY,
+         full_inst->InstructionExtNv.CondSwizzleZ,
+         full_inst->InstructionExtNv.CondSwizzleW,
+         full_inst->InstructionExtNv.CondDstUpdate,
+         full_inst->InstructionExtNv.CondFlowEnable,
+         prev_token,
+         instruction,
+         header );
+      prev_token = (struct tgsi_token  *) instruction_ext_nv;
+   }
+
+   if( tgsi_compare_instruction_ext_label(
+         full_inst->InstructionExtLabel,
+         tgsi_default_instruction_ext_label() ) ) {
+      struct tgsi_instruction_ext_label *instruction_ext_label;
+
+      if( maxsize <= size )
+         return 0;
+      instruction_ext_label =
+         (struct  tgsi_instruction_ext_label *) &tokens[size];
+      size++;
+
+      *instruction_ext_label = tgsi_build_instruction_ext_label(
+         full_inst->InstructionExtLabel.Label,
+         prev_token,
+         instruction,
+         header );
+      prev_token = (struct tgsi_token  *) instruction_ext_label;
+   }
+
+   if( tgsi_compare_instruction_ext_texture(
+         full_inst->InstructionExtTexture,
+         tgsi_default_instruction_ext_texture() ) ) {
+      struct tgsi_instruction_ext_texture *instruction_ext_texture;
+
+      if( maxsize <= size )
+         return 0;
+      instruction_ext_texture =
+         (struct  tgsi_instruction_ext_texture *) &tokens[size];
+      size++;
+
+      *instruction_ext_texture = tgsi_build_instruction_ext_texture(
+         full_inst->InstructionExtTexture.Texture,
+         prev_token,
+         instruction,
+         header   );
+      prev_token = (struct tgsi_token  *) instruction_ext_texture;
+   }
+
+   for( i = 0;  i <   full_inst->Instruction.NumDstRegs; i++ ) {
+      const struct tgsi_full_dst_register *reg = &full_inst->FullDstRegisters[i];
+      struct tgsi_dst_register *dst_register;
+      struct tgsi_token *prev_token;
+
+      if( maxsize <= size )
+         return 0;
+      dst_register = (struct tgsi_dst_register *) &tokens[size];
+      size++;
+
+      *dst_register = tgsi_build_dst_register(
+         reg->DstRegister.File,
+         reg->DstRegister.WriteMask,
+         reg->DstRegister.Index,
+         instruction,
+         header );
+      prev_token = (struct tgsi_token  *) dst_register;
+
+      if( tgsi_compare_dst_register_ext_concode(
+            reg->DstRegisterExtConcode,
+            tgsi_default_dst_register_ext_concode() ) ) {
+         struct tgsi_dst_register_ext_concode *dst_register_ext_concode;
+
+         if( maxsize <= size )
+            return 0;
+         dst_register_ext_concode =
+            (struct  tgsi_dst_register_ext_concode *) &tokens[size];
+         size++;
+
+         *dst_register_ext_concode =   tgsi_build_dst_register_ext_concode(
+            reg->DstRegisterExtConcode.CondMask,
+            reg->DstRegisterExtConcode.CondSwizzleX,
+            reg->DstRegisterExtConcode.CondSwizzleY,
+            reg->DstRegisterExtConcode.CondSwizzleZ,
+            reg->DstRegisterExtConcode.CondSwizzleW,
+            reg->DstRegisterExtConcode.CondSrcIndex,
+            prev_token,
+            instruction,
+            header );
+         prev_token = (struct tgsi_token  *) dst_register_ext_concode;
+      }
+
+      if( tgsi_compare_dst_register_ext_modulate(
+            reg->DstRegisterExtModulate,
+            tgsi_default_dst_register_ext_modulate() ) ) {
+         struct tgsi_dst_register_ext_modulate *dst_register_ext_modulate;
+
+         if( maxsize <= size )
+            return 0;
+         dst_register_ext_modulate =
+            (struct  tgsi_dst_register_ext_modulate *) &tokens[size];
+         size++;
+
+         *dst_register_ext_modulate = tgsi_build_dst_register_ext_modulate(
+            reg->DstRegisterExtModulate.Modulate,
+            prev_token,
+            instruction,
+            header );
+         prev_token = (struct tgsi_token  *) dst_register_ext_modulate;
+      }
+   }
+
+   for( i = 0;  i < full_inst->Instruction.NumSrcRegs; i++ ) {
+      const struct tgsi_full_src_register *reg = &full_inst->FullSrcRegisters[i];
+      struct tgsi_src_register *src_register;
+      struct tgsi_token *prev_token;
+
+      if( maxsize <= size )
+         return 0;
+      src_register = (struct tgsi_src_register *)  &tokens[size];
+      size++;
+
+      *src_register = tgsi_build_src_register(
+         reg->SrcRegister.File,
+         reg->SrcRegister.SwizzleX,
+         reg->SrcRegister.SwizzleY,
+         reg->SrcRegister.SwizzleZ,
+         reg->SrcRegister.SwizzleW,
+         reg->SrcRegister.Negate,
+         reg->SrcRegister.Indirect,
+         reg->SrcRegister.Dimension,
+         reg->SrcRegister.Index,
+         instruction,
+         header );
+      prev_token = (struct tgsi_token  *) src_register;
+
+      if( tgsi_compare_src_register_ext_swz(
+            reg->SrcRegisterExtSwz,
+            tgsi_default_src_register_ext_swz() ) ) {
+         struct tgsi_src_register_ext_swz *src_register_ext_swz;
+
+         /* Use of the extended swizzle requires the simple swizzle to be identity.
+          */
+         assert( reg->SrcRegister.SwizzleX == TGSI_SWIZZLE_X );
+         assert( reg->SrcRegister.SwizzleY == TGSI_SWIZZLE_Y );
+         assert( reg->SrcRegister.SwizzleZ == TGSI_SWIZZLE_Z );
+         assert( reg->SrcRegister.SwizzleW == TGSI_SWIZZLE_W );
+         assert( reg->SrcRegister.Negate == FALSE );
+
+         if( maxsize <= size )
+            return 0;
+         src_register_ext_swz =
+            (struct  tgsi_src_register_ext_swz *) &tokens[size];
+         size++;
+
+         *src_register_ext_swz = tgsi_build_src_register_ext_swz(
+            reg->SrcRegisterExtSwz.ExtSwizzleX,
+            reg->SrcRegisterExtSwz.ExtSwizzleY,
+            reg->SrcRegisterExtSwz.ExtSwizzleZ,
+            reg->SrcRegisterExtSwz.ExtSwizzleW,
+            reg->SrcRegisterExtSwz.NegateX,
+            reg->SrcRegisterExtSwz.NegateY,
+            reg->SrcRegisterExtSwz.NegateZ,
+            reg->SrcRegisterExtSwz.NegateW,
+            prev_token,
+            instruction,
+            header );
+         prev_token = (struct tgsi_token  *) src_register_ext_swz;
+      }
+
+      if( tgsi_compare_src_register_ext_mod(
+            reg->SrcRegisterExtMod,
+            tgsi_default_src_register_ext_mod() ) ) {
+         struct tgsi_src_register_ext_mod *src_register_ext_mod;
+
+         if( maxsize <= size )
+            return 0;
+         src_register_ext_mod =
+            (struct  tgsi_src_register_ext_mod *) &tokens[size];
+         size++;
+
+         *src_register_ext_mod = tgsi_build_src_register_ext_mod(
+            reg->SrcRegisterExtMod.Complement,
+            reg->SrcRegisterExtMod.Bias,
+            reg->SrcRegisterExtMod.Scale2X,
+            reg->SrcRegisterExtMod.Absolute,
+            reg->SrcRegisterExtMod.Negate,
+            prev_token,
+            instruction,
+            header );
+         prev_token = (struct tgsi_token  *) src_register_ext_mod;
+      }
+
+      if( reg->SrcRegister.Indirect ) {
+         struct  tgsi_src_register *ind;
+
+         if( maxsize <= size )
+            return 0;
+         ind = (struct tgsi_src_register *) &tokens[size];
+         size++;
+
+         *ind = tgsi_build_src_register(
+            reg->SrcRegisterInd.File,
+            reg->SrcRegisterInd.SwizzleX,
+            reg->SrcRegisterInd.SwizzleY,
+            reg->SrcRegisterInd.SwizzleZ,
+            reg->SrcRegisterInd.SwizzleW,
+            reg->SrcRegisterInd.Negate,
+            reg->SrcRegisterInd.Indirect,
+            reg->SrcRegisterInd.Dimension,
+            reg->SrcRegisterInd.Index,
+            instruction,
+            header );
+      }
+
+      if( reg->SrcRegister.Dimension ) {
+         struct  tgsi_dimension *dim;
+
+         assert( !reg->SrcRegisterDim.Dimension );
+
+         if( maxsize <= size )
+            return 0;
+         dim = (struct tgsi_dimension *) &tokens[size];
+         size++;
+
+         *dim = tgsi_build_dimension(
+            reg->SrcRegisterDim.Indirect,
+            reg->SrcRegisterDim.Index,
+            instruction,
+            header );
+
+         if( reg->SrcRegisterDim.Indirect ) {
+            struct tgsi_src_register *ind;
+
+            if( maxsize <= size )
+               return 0;
+            ind = (struct tgsi_src_register *) &tokens[size];
+            size++;
+
+            *ind = tgsi_build_src_register(
+               reg->SrcRegisterDimInd.File,
+               reg->SrcRegisterDimInd.SwizzleX,
+               reg->SrcRegisterDimInd.SwizzleY,
+               reg->SrcRegisterDimInd.SwizzleZ,
+               reg->SrcRegisterDimInd.SwizzleW,
+               reg->SrcRegisterDimInd.Negate,
+               reg->SrcRegisterDimInd.Indirect,
+               reg->SrcRegisterDimInd.Dimension,
+               reg->SrcRegisterDimInd.Index,
+               instruction,
+               header );
+         }
+      }
+   }
+
+   return size;
+}
+
+struct tgsi_instruction_ext_nv
+tgsi_default_instruction_ext_nv( void )
+{
+   struct tgsi_instruction_ext_nv instruction_ext_nv;
+
+   instruction_ext_nv.Type = TGSI_INSTRUCTION_EXT_TYPE_NV;
+   instruction_ext_nv.Precision = TGSI_PRECISION_DEFAULT;
+   instruction_ext_nv.CondDstIndex = 0;
+   instruction_ext_nv.CondFlowIndex = 0;
+   instruction_ext_nv.CondMask = TGSI_CC_TR;
+   instruction_ext_nv.CondSwizzleX = TGSI_SWIZZLE_X;
+   instruction_ext_nv.CondSwizzleY = TGSI_SWIZZLE_Y;
+   instruction_ext_nv.CondSwizzleZ = TGSI_SWIZZLE_Z;
+   instruction_ext_nv.CondSwizzleW = TGSI_SWIZZLE_W;
+   instruction_ext_nv.CondDstUpdate = 0;
+   instruction_ext_nv.CondFlowEnable = 0;
+   instruction_ext_nv.Padding = 0;
+   instruction_ext_nv.Extended = 0;
+
+   return instruction_ext_nv;
+}
+
+
+/** test for inequality of 32-bit values pointed to by a and b */
+static INLINE boolean
+compare32(const void *a, const void *b)
+{
+   return *((uint32_t *) a) != *((uint32_t *) b);
+}
+
+
+unsigned
+tgsi_compare_instruction_ext_nv(
+   struct tgsi_instruction_ext_nv a,
+   struct tgsi_instruction_ext_nv b )
+{
+   a.Padding = b.Padding = 0;
+   a.Extended = b.Extended = 0;
+   return compare32(&a, &b);
+}
+
+struct tgsi_instruction_ext_nv
+tgsi_build_instruction_ext_nv(
+   unsigned precision,
+   unsigned cond_dst_index,
+   unsigned cond_flow_index,
+   unsigned cond_mask,
+   unsigned cond_swizzle_x,
+   unsigned cond_swizzle_y,
+   unsigned cond_swizzle_z,
+   unsigned cond_swizzle_w,
+   unsigned cond_dst_update,
+   unsigned cond_flow_enable,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_instruction_ext_nv instruction_ext_nv;
+
+   instruction_ext_nv = tgsi_default_instruction_ext_nv();
+   instruction_ext_nv.Precision = precision;
+   instruction_ext_nv.CondDstIndex = cond_dst_index;
+   instruction_ext_nv.CondFlowIndex = cond_flow_index;
+   instruction_ext_nv.CondMask = cond_mask;
+   instruction_ext_nv.CondSwizzleX = cond_swizzle_x;
+   instruction_ext_nv.CondSwizzleY = cond_swizzle_y;
+   instruction_ext_nv.CondSwizzleZ = cond_swizzle_z;
+   instruction_ext_nv.CondSwizzleW = cond_swizzle_w;
+   instruction_ext_nv.CondDstUpdate = cond_dst_update;
+   instruction_ext_nv.CondFlowEnable = cond_flow_enable;
+
+   prev_token->Extended = 1;
+   instruction_grow( instruction, header );
+
+   return instruction_ext_nv;
+}
+
+struct tgsi_instruction_ext_label
+tgsi_default_instruction_ext_label( void )
+{
+   struct tgsi_instruction_ext_label instruction_ext_label;
+
+   instruction_ext_label.Type = TGSI_INSTRUCTION_EXT_TYPE_LABEL;
+   instruction_ext_label.Label = 0;
+   instruction_ext_label.Padding = 0;
+   instruction_ext_label.Extended = 0;
+
+   return instruction_ext_label;
+}
+
+unsigned
+tgsi_compare_instruction_ext_label(
+   struct tgsi_instruction_ext_label a,
+   struct tgsi_instruction_ext_label b )
+{
+   a.Padding = b.Padding = 0;
+   a.Extended = b.Extended = 0;
+   return compare32(&a, &b);
+}
+
+struct tgsi_instruction_ext_label
+tgsi_build_instruction_ext_label(
+   unsigned label,
+   struct tgsi_token  *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_instruction_ext_label instruction_ext_label;
+
+   instruction_ext_label = tgsi_default_instruction_ext_label();
+   instruction_ext_label.Label = label;
+
+   prev_token->Extended = 1;
+   instruction_grow( instruction, header );
+
+   return instruction_ext_label;
+}
+
+struct tgsi_instruction_ext_texture
+tgsi_default_instruction_ext_texture( void )
+{
+   struct tgsi_instruction_ext_texture instruction_ext_texture;
+
+   instruction_ext_texture.Type = TGSI_INSTRUCTION_EXT_TYPE_TEXTURE;
+   instruction_ext_texture.Texture = TGSI_TEXTURE_UNKNOWN;
+   instruction_ext_texture.Padding = 0;
+   instruction_ext_texture.Extended = 0;
+
+   return instruction_ext_texture;
+}
+
+unsigned
+tgsi_compare_instruction_ext_texture(
+   struct tgsi_instruction_ext_texture a,
+   struct tgsi_instruction_ext_texture b )
+{
+   a.Padding = b.Padding = 0;
+   a.Extended = b.Extended = 0;
+   return compare32(&a, &b);
+}
+
+struct tgsi_instruction_ext_texture
+tgsi_build_instruction_ext_texture(
+   unsigned texture,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_instruction_ext_texture instruction_ext_texture;
+
+   instruction_ext_texture = tgsi_default_instruction_ext_texture();
+   instruction_ext_texture.Texture = texture;
+
+   prev_token->Extended = 1;
+   instruction_grow( instruction, header );
+
+   return instruction_ext_texture;
+}
+
+struct tgsi_src_register
+tgsi_default_src_register( void )
+{
+   struct tgsi_src_register src_register;
+
+   src_register.File = TGSI_FILE_NULL;
+   src_register.SwizzleX = TGSI_SWIZZLE_X;
+   src_register.SwizzleY = TGSI_SWIZZLE_Y;
+   src_register.SwizzleZ = TGSI_SWIZZLE_Z;
+   src_register.SwizzleW = TGSI_SWIZZLE_W;
+   src_register.Negate = 0;
+   src_register.Indirect = 0;
+   src_register.Dimension = 0;
+   src_register.Index = 0;
+   src_register.Extended = 0;
+
+   return src_register;
+}
+
+struct tgsi_src_register
+tgsi_build_src_register(
+   unsigned file,
+   unsigned swizzle_x,
+   unsigned swizzle_y,
+   unsigned swizzle_z,
+   unsigned swizzle_w,
+   unsigned negate,
+   unsigned indirect,
+   unsigned dimension,
+   int index,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_src_register   src_register;
+
+   assert( file <= TGSI_FILE_IMMEDIATE );
+   assert( swizzle_x <= TGSI_SWIZZLE_W );
+   assert( swizzle_y <= TGSI_SWIZZLE_W );
+   assert( swizzle_z <= TGSI_SWIZZLE_W );
+   assert( swizzle_w <= TGSI_SWIZZLE_W );
+   assert( negate <= 1 );
+   assert( index >= -0x8000 && index <= 0x7FFF );
+
+   src_register = tgsi_default_src_register();
+   src_register.File = file;
+   src_register.SwizzleX = swizzle_x;
+   src_register.SwizzleY = swizzle_y;
+   src_register.SwizzleZ = swizzle_z;
+   src_register.SwizzleW = swizzle_w;
+   src_register.Negate = negate;
+   src_register.Indirect = indirect;
+   src_register.Dimension = dimension;
+   src_register.Index = index;
+
+   instruction_grow( instruction, header );
+
+   return src_register;
+}
+
+struct tgsi_full_src_register
+tgsi_default_full_src_register( void )
+{
+   struct tgsi_full_src_register full_src_register;
+
+   full_src_register.SrcRegister = tgsi_default_src_register();
+   full_src_register.SrcRegisterExtSwz = tgsi_default_src_register_ext_swz();
+   full_src_register.SrcRegisterExtMod = tgsi_default_src_register_ext_mod();
+   full_src_register.SrcRegisterInd = tgsi_default_src_register();
+   full_src_register.SrcRegisterDim = tgsi_default_dimension();
+   full_src_register.SrcRegisterDimInd = tgsi_default_src_register();
+
+   return full_src_register;
+}
+
+struct tgsi_src_register_ext_swz
+tgsi_default_src_register_ext_swz( void )
+{
+   struct tgsi_src_register_ext_swz src_register_ext_swz;
+
+   src_register_ext_swz.Type = TGSI_SRC_REGISTER_EXT_TYPE_SWZ;
+   src_register_ext_swz.ExtSwizzleX = TGSI_EXTSWIZZLE_X;
+   src_register_ext_swz.ExtSwizzleY = TGSI_EXTSWIZZLE_Y;
+   src_register_ext_swz.ExtSwizzleZ = TGSI_EXTSWIZZLE_Z;
+   src_register_ext_swz.ExtSwizzleW = TGSI_EXTSWIZZLE_W;
+   src_register_ext_swz.NegateX = 0;
+   src_register_ext_swz.NegateY = 0;
+   src_register_ext_swz.NegateZ = 0;
+   src_register_ext_swz.NegateW = 0;
+   src_register_ext_swz.Padding = 0;
+   src_register_ext_swz.Extended = 0;
+
+   return src_register_ext_swz;
+}
+
+unsigned
+tgsi_compare_src_register_ext_swz(
+   struct tgsi_src_register_ext_swz a,
+   struct tgsi_src_register_ext_swz b )
+{
+   a.Padding = b.Padding = 0;
+   a.Extended = b.Extended = 0;
+   return compare32(&a, &b);
+}
+
+struct tgsi_src_register_ext_swz
+tgsi_build_src_register_ext_swz(
+   unsigned ext_swizzle_x,
+   unsigned ext_swizzle_y,
+   unsigned ext_swizzle_z,
+   unsigned ext_swizzle_w,
+   unsigned negate_x,
+   unsigned negate_y,
+   unsigned negate_z,
+   unsigned negate_w,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_src_register_ext_swz src_register_ext_swz;
+
+   assert( ext_swizzle_x <= TGSI_EXTSWIZZLE_ONE );
+   assert( ext_swizzle_y <= TGSI_EXTSWIZZLE_ONE );
+   assert( ext_swizzle_z <= TGSI_EXTSWIZZLE_ONE );
+   assert( ext_swizzle_w <= TGSI_EXTSWIZZLE_ONE );
+   assert( negate_x <= 1 );
+   assert( negate_y <= 1 );
+   assert( negate_z <= 1 );
+   assert( negate_w <= 1 );
+
+   src_register_ext_swz = tgsi_default_src_register_ext_swz();
+   src_register_ext_swz.ExtSwizzleX = ext_swizzle_x;
+   src_register_ext_swz.ExtSwizzleY = ext_swizzle_y;
+   src_register_ext_swz.ExtSwizzleZ = ext_swizzle_z;
+   src_register_ext_swz.ExtSwizzleW = ext_swizzle_w;
+   src_register_ext_swz.NegateX = negate_x;
+   src_register_ext_swz.NegateY = negate_y;
+   src_register_ext_swz.NegateZ = negate_z;
+   src_register_ext_swz.NegateW = negate_w;
+
+   prev_token->Extended = 1;
+   instruction_grow( instruction, header );
+
+   return src_register_ext_swz;
+}
+
+struct tgsi_src_register_ext_mod
+tgsi_default_src_register_ext_mod( void )
+{
+   struct tgsi_src_register_ext_mod src_register_ext_mod;
+
+   src_register_ext_mod.Type = TGSI_SRC_REGISTER_EXT_TYPE_MOD;
+   src_register_ext_mod.Complement = 0;
+   src_register_ext_mod.Bias = 0;
+   src_register_ext_mod.Scale2X = 0;
+   src_register_ext_mod.Absolute = 0;
+   src_register_ext_mod.Negate = 0;
+   src_register_ext_mod.Padding = 0;
+   src_register_ext_mod.Extended = 0;
+
+   return src_register_ext_mod;
+}
+
+unsigned
+tgsi_compare_src_register_ext_mod(
+   struct tgsi_src_register_ext_mod a,
+   struct tgsi_src_register_ext_mod b )
+{
+   a.Padding = b.Padding = 0;
+   a.Extended = b.Extended = 0;
+   return compare32(&a, &b);
+}
+
+struct tgsi_src_register_ext_mod
+tgsi_build_src_register_ext_mod(
+   unsigned complement,
+   unsigned bias,
+   unsigned scale_2x,
+   unsigned absolute,
+   unsigned negate,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_src_register_ext_mod src_register_ext_mod;
+
+   assert( complement <= 1 );
+   assert( bias <= 1 );
+   assert( scale_2x <= 1 );
+   assert( absolute <= 1 );
+   assert( negate <= 1 );
+
+   src_register_ext_mod = tgsi_default_src_register_ext_mod();
+   src_register_ext_mod.Complement = complement;
+   src_register_ext_mod.Bias = bias;
+   src_register_ext_mod.Scale2X = scale_2x;
+   src_register_ext_mod.Absolute = absolute;
+   src_register_ext_mod.Negate = negate;
+
+   prev_token->Extended = 1;
+   instruction_grow( instruction, header );
+
+   return src_register_ext_mod;
+}
+
+struct tgsi_dimension
+tgsi_default_dimension( void )
+{
+   struct tgsi_dimension dimension;
+
+   dimension.Indirect = 0;
+   dimension.Dimension = 0;
+   dimension.Padding = 0;
+   dimension.Index = 0;
+   dimension.Extended = 0;
+
+   return dimension;
+}
+
+struct tgsi_dimension
+tgsi_build_dimension(
+   unsigned indirect,
+   unsigned index,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_dimension dimension;
+
+   dimension = tgsi_default_dimension();
+   dimension.Indirect = indirect;
+   dimension.Index = index;
+
+   instruction_grow( instruction, header );
+
+   return dimension;
+}
+
+struct tgsi_dst_register
+tgsi_default_dst_register( void )
+{
+   struct tgsi_dst_register dst_register;
+
+   dst_register.File = TGSI_FILE_NULL;
+   dst_register.WriteMask = TGSI_WRITEMASK_XYZW;
+   dst_register.Indirect = 0;
+   dst_register.Dimension = 0;
+   dst_register.Index = 0;
+   dst_register.Padding = 0;
+   dst_register.Extended = 0;
+
+   return dst_register;
+}
+
+struct tgsi_dst_register
+tgsi_build_dst_register(
+   unsigned file,
+   unsigned mask,
+   int index,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_dst_register dst_register;
+
+   assert( file <= TGSI_FILE_IMMEDIATE );
+   assert( mask <= TGSI_WRITEMASK_XYZW );
+   assert( index >= -32768 && index <= 32767 );
+
+   dst_register = tgsi_default_dst_register();
+   dst_register.File = file;
+   dst_register.WriteMask = mask;
+   dst_register.Index = index;
+
+   instruction_grow( instruction, header );
+
+   return dst_register;
+}
+
+struct tgsi_full_dst_register
+tgsi_default_full_dst_register( void )
+{
+   struct tgsi_full_dst_register full_dst_register;
+
+   full_dst_register.DstRegister = tgsi_default_dst_register();
+   full_dst_register.DstRegisterExtConcode =
+      tgsi_default_dst_register_ext_concode();
+   full_dst_register.DstRegisterExtModulate =
+      tgsi_default_dst_register_ext_modulate();
+
+   return full_dst_register;
+}
+
+struct tgsi_dst_register_ext_concode
+tgsi_default_dst_register_ext_concode( void )
+{
+   struct tgsi_dst_register_ext_concode dst_register_ext_concode;
+
+   dst_register_ext_concode.Type = TGSI_DST_REGISTER_EXT_TYPE_CONDCODE;
+   dst_register_ext_concode.CondMask = TGSI_CC_TR;
+   dst_register_ext_concode.CondSwizzleX = TGSI_SWIZZLE_X;
+   dst_register_ext_concode.CondSwizzleY = TGSI_SWIZZLE_Y;
+   dst_register_ext_concode.CondSwizzleZ = TGSI_SWIZZLE_Z;
+   dst_register_ext_concode.CondSwizzleW = TGSI_SWIZZLE_W;
+   dst_register_ext_concode.CondSrcIndex = 0;
+   dst_register_ext_concode.Padding = 0;
+   dst_register_ext_concode.Extended = 0;
+
+   return dst_register_ext_concode;
+}
+
+unsigned
+tgsi_compare_dst_register_ext_concode(
+   struct tgsi_dst_register_ext_concode a,
+   struct tgsi_dst_register_ext_concode b )
+{
+   a.Padding = b.Padding = 0;
+   a.Extended = b.Extended = 0;
+   return compare32(&a, &b);
+}
+
+struct tgsi_dst_register_ext_concode
+tgsi_build_dst_register_ext_concode(
+   unsigned cc,
+   unsigned swizzle_x,
+   unsigned swizzle_y,
+   unsigned swizzle_z,
+   unsigned swizzle_w,
+   int index,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_dst_register_ext_concode dst_register_ext_concode;
+
+   assert( cc <= TGSI_CC_FL );
+   assert( swizzle_x <= TGSI_SWIZZLE_W );
+   assert( swizzle_y <= TGSI_SWIZZLE_W );
+   assert( swizzle_z <= TGSI_SWIZZLE_W );
+   assert( swizzle_w <= TGSI_SWIZZLE_W );
+   assert( index >= -32768 && index <= 32767 );
+
+   dst_register_ext_concode = tgsi_default_dst_register_ext_concode();
+   dst_register_ext_concode.CondMask = cc;
+   dst_register_ext_concode.CondSwizzleX = swizzle_x;
+   dst_register_ext_concode.CondSwizzleY = swizzle_y;
+   dst_register_ext_concode.CondSwizzleZ = swizzle_z;
+   dst_register_ext_concode.CondSwizzleW = swizzle_w;
+   dst_register_ext_concode.CondSrcIndex = index;
+
+   prev_token->Extended = 1;
+   instruction_grow( instruction, header );
+
+   return dst_register_ext_concode;
+}
+
+struct tgsi_dst_register_ext_modulate
+tgsi_default_dst_register_ext_modulate( void )
+{
+   struct tgsi_dst_register_ext_modulate dst_register_ext_modulate;
+
+   dst_register_ext_modulate.Type = TGSI_DST_REGISTER_EXT_TYPE_MODULATE;
+   dst_register_ext_modulate.Modulate = TGSI_MODULATE_1X;
+   dst_register_ext_modulate.Padding = 0;
+   dst_register_ext_modulate.Extended = 0;
+
+   return dst_register_ext_modulate;
+}
+
+unsigned
+tgsi_compare_dst_register_ext_modulate(
+   struct tgsi_dst_register_ext_modulate a,
+   struct tgsi_dst_register_ext_modulate b )
+{
+   a.Padding = b.Padding = 0;
+   a.Extended = b.Extended = 0;
+   return compare32(&a, &b);
+}
+
+struct tgsi_dst_register_ext_modulate
+tgsi_build_dst_register_ext_modulate(
+   unsigned modulate,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header )
+{
+   struct tgsi_dst_register_ext_modulate dst_register_ext_modulate;
+
+   assert( modulate <= TGSI_MODULATE_EIGHTH );
+
+   dst_register_ext_modulate = tgsi_default_dst_register_ext_modulate();
+   dst_register_ext_modulate.Modulate = modulate;
+
+   prev_token->Extended = 1;
+   instruction_grow( instruction, header );
+
+   return dst_register_ext_modulate;
+}
diff --git a/src/gallium/auxiliary/tgsi/tgsi_build.h b/src/gallium/auxiliary/tgsi/tgsi_build.h
new file mode 100644
index 00000000000..0fd6fabd83d
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_build.h
@@ -0,0 +1,338 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef TGSI_BUILD_H
+#define TGSI_BUILD_H
+
+
+struct tgsi_token;
+
+
+#if defined __cplusplus
+extern "C" {
+#endif
+
+/*
+ * version
+ */
+
+struct tgsi_version
+tgsi_build_version( void );
+
+/*
+ * header
+ */
+
+struct tgsi_header
+tgsi_build_header( void );
+
+struct tgsi_processor
+tgsi_default_processor( void );
+
+struct tgsi_processor
+tgsi_build_processor(
+   unsigned processor,
+   struct tgsi_header *header );
+
+/*
+ * declaration
+ */
+
+struct tgsi_declaration
+tgsi_default_declaration( void );
+
+struct tgsi_declaration
+tgsi_build_declaration(
+   unsigned file,
+   unsigned usage_mask,
+   unsigned interpolate,
+   unsigned semantic,
+   unsigned centroid,
+   unsigned invariant,
+   struct tgsi_header *header );
+
+struct tgsi_full_declaration
+tgsi_default_full_declaration( void );
+
+unsigned
+tgsi_build_full_declaration(
+   const struct tgsi_full_declaration *full_decl,
+   struct tgsi_token *tokens,
+   struct tgsi_header *header,
+   unsigned maxsize );
+
+struct tgsi_declaration_range
+tgsi_default_declaration_range( void );
+
+struct tgsi_declaration_range
+tgsi_build_declaration_range(
+   unsigned first,
+   unsigned last,
+   struct tgsi_declaration *declaration,
+   struct tgsi_header *header );
+
+struct tgsi_declaration_semantic
+tgsi_default_declaration_semantic( void );
+
+struct tgsi_declaration_semantic
+tgsi_build_declaration_semantic(
+   unsigned semantic_name,
+   unsigned semantic_index,
+   struct tgsi_declaration *declaration,
+   struct tgsi_header *header );
+
+/*
+ * immediate
+ */
+
+struct tgsi_immediate
+tgsi_default_immediate( void );
+
+struct tgsi_immediate
+tgsi_build_immediate(
+   struct tgsi_header *header );
+
+struct tgsi_full_immediate
+tgsi_default_full_immediate( void );
+
+struct tgsi_immediate_float32
+tgsi_build_immediate_float32(
+   float value,
+   struct tgsi_immediate *immediate,
+   struct tgsi_header *header );
+
+unsigned
+tgsi_build_full_immediate(
+   const struct tgsi_full_immediate *full_imm,
+   struct tgsi_token *tokens,
+   struct tgsi_header *header,
+   unsigned maxsize );
+
+/*
+ * instruction
+ */
+
+struct tgsi_instruction
+tgsi_default_instruction( void );
+
+struct tgsi_instruction
+tgsi_build_instruction(
+   unsigned opcode,
+   unsigned saturate,
+   unsigned num_dst_regs,
+   unsigned num_src_regs,
+   struct tgsi_header *header );
+
+struct tgsi_full_instruction
+tgsi_default_full_instruction( void );
+
+unsigned
+tgsi_build_full_instruction(
+   const struct tgsi_full_instruction *full_inst,
+   struct tgsi_token *tokens,
+   struct tgsi_header *header,
+   unsigned maxsize );
+
+struct tgsi_instruction_ext_nv
+tgsi_default_instruction_ext_nv( void );
+
+unsigned
+tgsi_compare_instruction_ext_nv(
+   struct tgsi_instruction_ext_nv a,
+   struct tgsi_instruction_ext_nv b );
+
+struct tgsi_instruction_ext_nv
+tgsi_build_instruction_ext_nv(
+   unsigned precision,
+   unsigned cond_dst_index,
+   unsigned cond_flow_index,
+   unsigned cond_mask,
+   unsigned cond_swizzle_x,
+   unsigned cond_swizzle_y,
+   unsigned cond_swizzle_z,
+   unsigned cond_swizzle_w,
+   unsigned cond_dst_update,
+   unsigned cond_flow_enable,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_instruction_ext_label
+tgsi_default_instruction_ext_label( void );
+
+unsigned
+tgsi_compare_instruction_ext_label(
+   struct tgsi_instruction_ext_label a,
+   struct tgsi_instruction_ext_label b );
+
+struct tgsi_instruction_ext_label
+tgsi_build_instruction_ext_label(
+   unsigned label,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_instruction_ext_texture
+tgsi_default_instruction_ext_texture( void );
+
+unsigned
+tgsi_compare_instruction_ext_texture(
+   struct tgsi_instruction_ext_texture a,
+   struct tgsi_instruction_ext_texture b );
+
+struct tgsi_instruction_ext_texture
+tgsi_build_instruction_ext_texture(
+   unsigned texture,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_src_register
+tgsi_default_src_register( void );
+
+struct tgsi_src_register
+tgsi_build_src_register(
+   unsigned file,
+   unsigned swizzle_x,
+   unsigned swizzle_y,
+   unsigned swizzle_z,
+   unsigned swizzle_w,
+   unsigned negate,
+   unsigned indirect,
+   unsigned dimension,
+   int index,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_full_src_register
+tgsi_default_full_src_register( void );
+
+struct tgsi_src_register_ext_swz
+tgsi_default_src_register_ext_swz( void );
+
+unsigned
+tgsi_compare_src_register_ext_swz(
+   struct tgsi_src_register_ext_swz a,
+   struct tgsi_src_register_ext_swz b );
+
+struct tgsi_src_register_ext_swz
+tgsi_build_src_register_ext_swz(
+   unsigned ext_swizzle_x,
+   unsigned ext_swizzle_y,
+   unsigned ext_swizzle_z,
+   unsigned ext_swizzle_w,
+   unsigned negate_x,
+   unsigned negate_y,
+   unsigned negate_z,
+   unsigned negate_w,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_src_register_ext_mod
+tgsi_default_src_register_ext_mod( void );
+
+unsigned
+tgsi_compare_src_register_ext_mod(
+   struct tgsi_src_register_ext_mod a,
+   struct tgsi_src_register_ext_mod b );
+
+struct tgsi_src_register_ext_mod
+tgsi_build_src_register_ext_mod(
+   unsigned complement,
+   unsigned bias,
+   unsigned scale_2x,
+   unsigned absolute,
+   unsigned negate,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_dimension
+tgsi_default_dimension( void );
+
+struct tgsi_dimension
+tgsi_build_dimension(
+   unsigned indirect,
+   unsigned index,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_dst_register
+tgsi_default_dst_register( void );
+
+struct tgsi_dst_register
+tgsi_build_dst_register(
+   unsigned file,
+   unsigned mask,
+   int index,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_full_dst_register
+tgsi_default_full_dst_register( void );
+
+struct tgsi_dst_register_ext_concode
+tgsi_default_dst_register_ext_concode( void );
+
+unsigned
+tgsi_compare_dst_register_ext_concode(
+   struct tgsi_dst_register_ext_concode a,
+   struct tgsi_dst_register_ext_concode b );
+
+struct tgsi_dst_register_ext_concode
+tgsi_build_dst_register_ext_concode(
+   unsigned cc,
+   unsigned swizzle_x,
+   unsigned swizzle_y,
+   unsigned swizzle_z,
+   unsigned swizzle_w,
+   int index,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+struct tgsi_dst_register_ext_modulate
+tgsi_default_dst_register_ext_modulate( void );
+
+unsigned
+tgsi_compare_dst_register_ext_modulate(
+   struct tgsi_dst_register_ext_modulate a,
+   struct tgsi_dst_register_ext_modulate b );
+
+struct tgsi_dst_register_ext_modulate
+tgsi_build_dst_register_ext_modulate(
+   unsigned modulate,
+   struct tgsi_token *prev_token,
+   struct tgsi_instruction *instruction,
+   struct tgsi_header *header );
+
+#if defined __cplusplus
+}
+#endif
+
+#endif /* TGSI_BUILD_H */
diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump.c b/src/gallium/auxiliary/tgsi/tgsi_dump.c
new file mode 100644
index 00000000000..ab2b1f2c58a
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_dump.c
@@ -0,0 +1,558 @@
+/**************************************************************************
+ * 
+ * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_debug.h"
+#include "util/u_string.h"
+#include "tgsi_dump.h"
+#include "tgsi_info.h"
+#include "tgsi_iterate.h"
+
+struct dump_ctx
+{
+   struct tgsi_iterate_context iter;
+
+   uint instno;
+   
+   void (*printf)(struct dump_ctx *ctx, const char *format, ...);
+};
+
+static void 
+dump_ctx_printf(struct dump_ctx *ctx, const char *format, ...)
+{
+   va_list ap;
+   (void)ctx;
+   va_start(ap, format);
+   debug_vprintf(format, ap);
+   va_end(ap);
+}
+
+static void
+dump_enum(
+   struct dump_ctx *ctx,
+   uint e,
+   const char **enums,
+   uint enum_count )
+{
+   if (e >= enum_count)
+      ctx->printf( ctx, "%u", e );
+   else
+      ctx->printf( ctx, "%s", enums[e] );
+}
+
+#define EOL()           ctx->printf( ctx, "\n" )
+#define TXT(S)          ctx->printf( ctx, "%s", S )
+#define CHR(C)          ctx->printf( ctx, "%c", C )
+#define UIX(I)          ctx->printf( ctx, "0x%x", I )
+#define UID(I)          ctx->printf( ctx, "%u", I )
+#define INSTID(I)          ctx->printf( ctx, "% 3u", I )
+#define SID(I)          ctx->printf( ctx, "%d", I )
+#define FLT(F)          ctx->printf( ctx, "%10.4f", F )
+#define ENM(E,ENUMS)    dump_enum( ctx, E, ENUMS, sizeof( ENUMS ) / sizeof( *ENUMS ) )
+
+static const char *processor_type_names[] =
+{
+   "FRAG",
+   "VERT",
+   "GEOM"
+};
+
+static const char *file_names[] =
+{
+   "NULL",
+   "CONST",
+   "IN",
+   "OUT",
+   "TEMP",
+   "SAMP",
+   "ADDR",
+   "IMM"
+};
+
+static const char *interpolate_names[] =
+{
+   "CONSTANT",
+   "LINEAR",
+   "PERSPECTIVE"
+};
+
+static const char *semantic_names[] =
+{
+   "POSITION",
+   "COLOR",
+   "BCOLOR",
+   "FOG",
+   "PSIZE",
+   "GENERIC",
+   "NORMAL"
+};
+
+static const char *immediate_type_names[] =
+{
+   "FLT32"
+};
+
+static const char *swizzle_names[] =
+{
+   "x",
+   "y",
+   "z",
+   "w"
+};
+
+static const char *texture_names[] =
+{
+   "UNKNOWN",
+   "1D",
+   "2D",
+   "3D",
+   "CUBE",
+   "RECT",
+   "SHADOW1D",
+   "SHADOW2D",
+   "SHADOWRECT"
+};
+
+static const char *extswizzle_names[] =
+{
+   "x",
+   "y",
+   "z",
+   "w",
+   "0",
+   "1"
+};
+
+static const char *modulate_names[TGSI_MODULATE_COUNT] =
+{
+   "",
+   "_2X",
+   "_4X",
+   "_8X",
+   "_D2",
+   "_D4",
+   "_D8"
+};
+
+static void
+_dump_register(
+   struct dump_ctx *ctx,
+   uint file,
+   int first,
+   int last )
+{
+   ENM( file, file_names );
+   CHR( '[' );
+   SID( first );
+   if (first != last) {
+      TXT( ".." );
+      SID( last );
+   }
+   CHR( ']' );
+}
+
+static void
+_dump_register_ind(
+   struct dump_ctx *ctx,
+   uint file,
+   int index,
+   uint ind_file,
+   int ind_index,
+   uint ind_swizzle )
+{
+   ENM( file, file_names );
+   CHR( '[' );
+   ENM( ind_file, file_names );
+   CHR( '[' );
+   SID( ind_index );
+   TXT( "]." );
+   ENM( ind_swizzle, swizzle_names );
+   if (index != 0) {
+      if (index > 0)
+         CHR( '+' );
+      SID( index );
+   }
+   CHR( ']' );
+}
+
+static void
+_dump_writemask(
+   struct dump_ctx *ctx,
+   uint writemask )
+{
+   if (writemask != TGSI_WRITEMASK_XYZW) {
+      CHR( '.' );
+      if (writemask & TGSI_WRITEMASK_X)
+         CHR( 'x' );
+      if (writemask & TGSI_WRITEMASK_Y)
+         CHR( 'y' );
+      if (writemask & TGSI_WRITEMASK_Z)
+         CHR( 'z' );
+      if (writemask & TGSI_WRITEMASK_W)
+         CHR( 'w' );
+   }
+}
+
+static boolean
+iter_declaration(
+   struct tgsi_iterate_context *iter,
+   struct tgsi_full_declaration *decl )
+{
+   struct dump_ctx *ctx = (struct dump_ctx *)iter;
+
+   TXT( "DCL " );
+
+   _dump_register(
+      ctx,
+      decl->Declaration.File,
+      decl->DeclarationRange.First,
+      decl->DeclarationRange.Last );
+   _dump_writemask(
+      ctx,
+      decl->Declaration.UsageMask );
+
+   if (decl->Declaration.Semantic) {
+      TXT( ", " );
+      ENM( decl->Semantic.SemanticName, semantic_names );
+      if (decl->Semantic.SemanticIndex != 0 ||
+          decl->Semantic.SemanticName == TGSI_SEMANTIC_GENERIC) {
+         CHR( '[' );
+         UID( decl->Semantic.SemanticIndex );
+         CHR( ']' );
+      }
+   }
+
+   TXT( ", " );
+   ENM( decl->Declaration.Interpolate, interpolate_names );
+
+   if (decl->Declaration.Centroid) {
+      TXT( ", CENTROID" );
+   }
+
+   if (decl->Declaration.Invariant) {
+      TXT( ", INVARIANT" );
+   }
+
+   EOL();
+
+   return TRUE;
+}
+
+void
+tgsi_dump_declaration(
+   const struct tgsi_full_declaration *decl )
+{
+   struct dump_ctx ctx;
+
+   ctx.printf = dump_ctx_printf;
+
+   iter_declaration( &ctx.iter, (struct tgsi_full_declaration *)decl );
+}
+
+static boolean
+iter_immediate(
+   struct tgsi_iterate_context *iter,
+   struct tgsi_full_immediate *imm )
+{
+   struct dump_ctx *ctx = (struct dump_ctx *) iter;
+
+   uint i;
+
+   TXT( "IMM " );
+   ENM( imm->Immediate.DataType, immediate_type_names );
+
+   TXT( " { " );
+   for (i = 0; i < imm->Immediate.NrTokens - 1; i++) {
+      switch (imm->Immediate.DataType) {
+      case TGSI_IMM_FLOAT32:
+         FLT( imm->u.ImmediateFloat32[i].Float );
+         break;
+      default:
+         assert( 0 );
+      }
+
+      if (i < imm->Immediate.NrTokens - 2)
+         TXT( ", " );
+   }
+   TXT( " }" );
+
+   EOL();
+
+   return TRUE;
+}
+
+void
+tgsi_dump_immediate(
+   const struct tgsi_full_immediate *imm )
+{
+   struct dump_ctx ctx;
+
+   ctx.printf = dump_ctx_printf;
+
+   iter_immediate( &ctx.iter, (struct tgsi_full_immediate *)imm );
+}
+
+static boolean
+iter_instruction(
+   struct tgsi_iterate_context *iter,
+   struct tgsi_full_instruction *inst )
+{
+   struct dump_ctx *ctx = (struct dump_ctx *) iter;
+   uint instno = ctx->instno++;
+   
+   uint i;
+   boolean first_reg = TRUE;
+
+   INSTID( instno );
+   TXT( ": " );
+   TXT( tgsi_get_opcode_info( inst->Instruction.Opcode )->mnemonic );
+
+   switch (inst->Instruction.Saturate) {
+   case TGSI_SAT_NONE:
+      break;
+   case TGSI_SAT_ZERO_ONE:
+      TXT( "_SAT" );
+      break;
+   case TGSI_SAT_MINUS_PLUS_ONE:
+      TXT( "_SATNV" );
+      break;
+   default:
+      assert( 0 );
+   }
+
+   for (i = 0; i < inst->Instruction.NumDstRegs; i++) {
+      const struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i];
+
+      if (!first_reg)
+         CHR( ',' );
+      CHR( ' ' );
+
+      _dump_register(
+         ctx,
+         dst->DstRegister.File,
+         dst->DstRegister.Index,
+         dst->DstRegister.Index );
+      ENM( dst->DstRegisterExtModulate.Modulate, modulate_names );
+      _dump_writemask( ctx, dst->DstRegister.WriteMask );
+
+      first_reg = FALSE;
+   }
+
+   for (i = 0; i < inst->Instruction.NumSrcRegs; i++) {
+      const struct tgsi_full_src_register *src = &inst->FullSrcRegisters[i];
+
+      if (!first_reg)
+         CHR( ',' );
+      CHR( ' ' );
+
+      if (src->SrcRegisterExtMod.Negate)
+         TXT( "-(" );
+      if (src->SrcRegisterExtMod.Absolute)
+         CHR( '|' );
+      if (src->SrcRegisterExtMod.Scale2X)
+         TXT( "2*(" );
+      if (src->SrcRegisterExtMod.Bias)
+         CHR( '(' );
+      if (src->SrcRegisterExtMod.Complement)
+         TXT( "1-(" );
+      if (src->SrcRegister.Negate)
+         CHR( '-' );
+
+      if (src->SrcRegister.Indirect) {
+         _dump_register_ind(
+            ctx,
+            src->SrcRegister.File,
+            src->SrcRegister.Index,
+            src->SrcRegisterInd.File,
+            src->SrcRegisterInd.Index,
+            src->SrcRegisterInd.SwizzleX );
+      }
+      else {
+         _dump_register(
+            ctx,
+            src->SrcRegister.File,
+            src->SrcRegister.Index,
+            src->SrcRegister.Index );
+      }
+
+      if (src->SrcRegister.SwizzleX != TGSI_SWIZZLE_X ||
+          src->SrcRegister.SwizzleY != TGSI_SWIZZLE_Y ||
+          src->SrcRegister.SwizzleZ != TGSI_SWIZZLE_Z ||
+          src->SrcRegister.SwizzleW != TGSI_SWIZZLE_W) {
+         CHR( '.' );
+         ENM( src->SrcRegister.SwizzleX, swizzle_names );
+         ENM( src->SrcRegister.SwizzleY, swizzle_names );
+         ENM( src->SrcRegister.SwizzleZ, swizzle_names );
+         ENM( src->SrcRegister.SwizzleW, swizzle_names );
+      }
+      if (src->SrcRegisterExtSwz.ExtSwizzleX != TGSI_EXTSWIZZLE_X ||
+          src->SrcRegisterExtSwz.ExtSwizzleY != TGSI_EXTSWIZZLE_Y ||
+          src->SrcRegisterExtSwz.ExtSwizzleZ != TGSI_EXTSWIZZLE_Z ||
+          src->SrcRegisterExtSwz.ExtSwizzleW != TGSI_EXTSWIZZLE_W) {
+         CHR( '.' );
+         if (src->SrcRegisterExtSwz.NegateX)
+            TXT("-");
+         ENM( src->SrcRegisterExtSwz.ExtSwizzleX, extswizzle_names );
+         if (src->SrcRegisterExtSwz.NegateY)
+            TXT("-");
+         ENM( src->SrcRegisterExtSwz.ExtSwizzleY, extswizzle_names );
+         if (src->SrcRegisterExtSwz.NegateZ)
+            TXT("-");
+         ENM( src->SrcRegisterExtSwz.ExtSwizzleZ, extswizzle_names );
+         if (src->SrcRegisterExtSwz.NegateW)
+            TXT("-");
+         ENM( src->SrcRegisterExtSwz.ExtSwizzleW, extswizzle_names );
+      }
+
+      if (src->SrcRegisterExtMod.Complement)
+         CHR( ')' );
+      if (src->SrcRegisterExtMod.Bias)
+         TXT( ")-.5" );
+      if (src->SrcRegisterExtMod.Scale2X)
+         CHR( ')' );
+      if (src->SrcRegisterExtMod.Absolute)
+         CHR( '|' );
+      if (src->SrcRegisterExtMod.Negate)
+         CHR( ')' );
+
+      first_reg = FALSE;
+   }
+
+   if (inst->InstructionExtTexture.Texture != TGSI_TEXTURE_UNKNOWN) {
+      TXT( ", " );
+      ENM( inst->InstructionExtTexture.Texture, texture_names );
+   }
+
+   switch (inst->Instruction.Opcode) {
+   case TGSI_OPCODE_IF:
+   case TGSI_OPCODE_ELSE:
+   case TGSI_OPCODE_BGNLOOP2:
+   case TGSI_OPCODE_ENDLOOP2:
+   case TGSI_OPCODE_CAL:
+      TXT( " :" );
+      UID( inst->InstructionExtLabel.Label );
+      break;
+   }
+
+   EOL();
+
+   return TRUE;
+}
+
+void
+tgsi_dump_instruction(
+   const struct tgsi_full_instruction *inst,
+   uint instno )
+{
+   struct dump_ctx ctx;
+
+   ctx.instno = instno;
+   ctx.printf = dump_ctx_printf;
+
+   iter_instruction( &ctx.iter, (struct tgsi_full_instruction *)inst );
+}
+
+static boolean
+prolog(
+   struct tgsi_iterate_context *iter )
+{
+   struct dump_ctx *ctx = (struct dump_ctx *) iter;
+   ENM( iter->processor.Processor, processor_type_names );
+   UID( iter->version.MajorVersion );
+   CHR( '.' );
+   UID( iter->version.MinorVersion );
+   EOL();
+   return TRUE;
+}
+
+void
+tgsi_dump(
+   const struct tgsi_token *tokens,
+   uint flags )
+{
+   struct dump_ctx ctx;
+
+   ctx.iter.prolog = prolog;
+   ctx.iter.iterate_instruction = iter_instruction;
+   ctx.iter.iterate_declaration = iter_declaration;
+   ctx.iter.iterate_immediate = iter_immediate;
+   ctx.iter.epilog = NULL;
+
+   ctx.instno = 0;
+   ctx.printf = dump_ctx_printf;
+
+   tgsi_iterate_shader( tokens, &ctx.iter );
+}
+
+struct str_dump_ctx
+{
+   struct dump_ctx base;
+   char *str;
+   char *ptr;
+   size_t left;
+};
+
+static void
+str_dump_ctx_printf(struct dump_ctx *ctx, const char *format, ...)
+{
+   struct str_dump_ctx *sctx = (struct str_dump_ctx *)ctx;
+   
+   if(sctx->left > 1) {
+      size_t written;
+      va_list ap;
+      va_start(ap, format);
+      written = util_vsnprintf(sctx->ptr, sctx->left, format, ap);
+      va_end(ap);
+      sctx->ptr += written;
+      sctx->left -= written;
+   }
+}
+
+void
+tgsi_dump_str(
+   const struct tgsi_token *tokens,
+   uint flags,
+   char *str,
+   size_t size)
+{
+   struct str_dump_ctx ctx;
+
+   ctx.base.iter.prolog = prolog;
+   ctx.base.iter.iterate_instruction = iter_instruction;
+   ctx.base.iter.iterate_declaration = iter_declaration;
+   ctx.base.iter.iterate_immediate = iter_immediate;
+   ctx.base.iter.epilog = NULL;
+
+   ctx.base.instno = 0;
+   ctx.base.printf = &str_dump_ctx_printf;
+
+   ctx.str = str;
+   ctx.str[0] = 0;
+   ctx.ptr = str;
+   ctx.left = size;
+
+   tgsi_iterate_shader( tokens, &ctx.base.iter );
+}
diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump.h b/src/gallium/auxiliary/tgsi/tgsi_dump.h
new file mode 100644
index 00000000000..ad1e647ec90
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_dump.h
@@ -0,0 +1,70 @@
+/**************************************************************************
+ * 
+ * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef TGSI_DUMP_H
+#define TGSI_DUMP_H
+
+#include "pipe/p_shader_tokens.h"
+
+#if defined __cplusplus
+extern "C" {
+#endif
+
+void
+tgsi_dump_str(
+   const struct tgsi_token *tokens,
+   uint flags,
+   char *str,
+   size_t size);
+
+void
+tgsi_dump(
+   const struct tgsi_token *tokens,
+   uint flags );
+
+struct tgsi_full_immediate;
+struct tgsi_full_instruction;
+struct tgsi_full_declaration;
+
+void
+tgsi_dump_immediate(
+   const struct tgsi_full_immediate *imm );
+
+void
+tgsi_dump_instruction(
+   const struct tgsi_full_instruction *inst,
+   uint instno );
+
+void
+tgsi_dump_declaration(
+   const struct tgsi_full_declaration *decl );
+
+#if defined __cplusplus
+}
+#endif
+
+#endif /* TGSI_DUMP_H */
diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump_c.c b/src/gallium/auxiliary/tgsi/tgsi_dump_c.c
new file mode 100644
index 00000000000..2ecf1e2f14b
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_dump_c.c
@@ -0,0 +1,718 @@
+/**************************************************************************
+ * 
+ * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_debug.h"
+#include "util/u_string.h"
+#include "tgsi_dump_c.h"
+#include "tgsi_build.h"
+#include "tgsi_info.h"
+#include "tgsi_parse.h"
+
+static void
+dump_enum(
+   const unsigned    e,
+   const char        **enums,
+   const unsigned    enums_count )
+{
+   if (e >= enums_count) {
+      debug_printf( "%u", e );
+   }
+   else {
+      debug_printf( "%s", enums[e] );
+   }
+}
+
+#define EOL()           debug_printf( "\n" )
+#define TXT(S)          debug_printf( "%s", S )
+#define CHR(C)          debug_printf( "%c", C )
+#define UIX(I)          debug_printf( "0x%x", I )
+#define UID(I)          debug_printf( "%u", I )
+#define SID(I)          debug_printf( "%d", I )
+#define FLT(F)          debug_printf( "%10.4f", F )
+#define ENM(E,ENUMS)    dump_enum( E, ENUMS, sizeof( ENUMS ) / sizeof( *ENUMS ) )
+
+static const char *TGSI_PROCESSOR_TYPES[] =
+{
+   "PROCESSOR_FRAGMENT",
+   "PROCESSOR_VERTEX",
+   "PROCESSOR_GEOMETRY"
+};
+
+static const char *TGSI_TOKEN_TYPES[] =
+{
+   "TOKEN_TYPE_DECLARATION",
+   "TOKEN_TYPE_IMMEDIATE",
+   "TOKEN_TYPE_INSTRUCTION"
+};
+
+static const char *TGSI_FILES[] =
+{
+   "FILE_NULL",
+   "FILE_CONSTANT",
+   "FILE_INPUT",
+   "FILE_OUTPUT",
+   "FILE_TEMPORARY",
+   "FILE_SAMPLER",
+   "FILE_ADDRESS",
+   "FILE_IMMEDIATE"
+};
+
+static const char *TGSI_INTERPOLATES[] =
+{
+   "INTERPOLATE_CONSTANT",
+   "INTERPOLATE_LINEAR",
+   "INTERPOLATE_PERSPECTIVE"
+};
+
+static const char *TGSI_SEMANTICS[] =
+{
+   "SEMANTIC_POSITION",
+   "SEMANTIC_COLOR",
+   "SEMANTIC_BCOLOR",
+   "SEMANTIC_FOG",
+   "SEMANTIC_PSIZE",
+   "SEMANTIC_GENERIC",
+   "SEMANTIC_NORMAL"
+};
+
+static const char *TGSI_IMMS[] =
+{
+   "IMM_FLOAT32"
+};
+
+static const char *TGSI_SATS[] =
+{
+   "SAT_NONE",
+   "SAT_ZERO_ONE",
+   "SAT_MINUS_PLUS_ONE"
+};
+
+static const char *TGSI_INSTRUCTION_EXTS[] =
+{
+   "INSTRUCTION_EXT_TYPE_NV",
+   "INSTRUCTION_EXT_TYPE_LABEL",
+   "INSTRUCTION_EXT_TYPE_TEXTURE"
+};
+
+static const char *TGSI_PRECISIONS[] =
+{
+   "PRECISION_DEFAULT",
+   "PRECISION_FLOAT32",
+   "PRECISION_FLOAT16",
+   "PRECISION_FIXED12"
+};
+
+static const char *TGSI_CCS[] =
+{
+   "CC_GT",
+   "CC_EQ",
+   "CC_LT",
+   "CC_UN",
+   "CC_GE",
+   "CC_LE",
+   "CC_NE",
+   "CC_TR",
+   "CC_FL"
+};
+
+static const char *TGSI_SWIZZLES[] =
+{
+   "SWIZZLE_X",
+   "SWIZZLE_Y",
+   "SWIZZLE_Z",
+   "SWIZZLE_W"
+};
+
+static const char *TGSI_TEXTURES[] =
+{
+   "TEXTURE_UNKNOWN",
+   "TEXTURE_1D",
+   "TEXTURE_2D",
+   "TEXTURE_3D",
+   "TEXTURE_CUBE",
+   "TEXTURE_RECT",
+   "TEXTURE_SHADOW1D",
+   "TEXTURE_SHADOW2D",
+   "TEXTURE_SHADOWRECT"
+};
+
+static const char *TGSI_SRC_REGISTER_EXTS[] =
+{
+   "SRC_REGISTER_EXT_TYPE_SWZ",
+   "SRC_REGISTER_EXT_TYPE_MOD"
+};
+
+static const char *TGSI_EXTSWIZZLES[] =
+{
+   "EXTSWIZZLE_X",
+   "EXTSWIZZLE_Y",
+   "EXTSWIZZLE_Z",
+   "EXTSWIZZLE_W",
+   "EXTSWIZZLE_ZERO",
+   "EXTSWIZZLE_ONE"
+};
+
+static const char *TGSI_WRITEMASKS[] =
+{
+   "0",
+   "WRITEMASK_X",
+   "WRITEMASK_Y",
+   "WRITEMASK_XY",
+   "WRITEMASK_Z",
+   "WRITEMASK_XZ",
+   "WRITEMASK_YZ",
+   "WRITEMASK_XYZ",
+   "WRITEMASK_W",
+   "WRITEMASK_XW",
+   "WRITEMASK_YW",
+   "WRITEMASK_XYW",
+   "WRITEMASK_ZW",
+   "WRITEMASK_XZW",
+   "WRITEMASK_YZW",
+   "WRITEMASK_XYZW"
+};
+
+static const char *TGSI_DST_REGISTER_EXTS[] =
+{
+   "DST_REGISTER_EXT_TYPE_CONDCODE",
+   "DST_REGISTER_EXT_TYPE_MODULATE"
+};
+
+static const char *TGSI_MODULATES[] =
+{
+   "MODULATE_1X",
+   "MODULATE_2X",
+   "MODULATE_4X",
+   "MODULATE_8X",
+   "MODULATE_HALF",
+   "MODULATE_QUARTER",
+   "MODULATE_EIGHTH"
+};
+
+static void
+dump_declaration_verbose(
+   struct tgsi_full_declaration  *decl,
+   unsigned                      ignored,
+   unsigned                      deflt,
+   struct tgsi_full_declaration  *fd )
+{
+   TXT( "\nFile       : " );
+   ENM( decl->Declaration.File, TGSI_FILES );
+   if( deflt || fd->Declaration.UsageMask != decl->Declaration.UsageMask ) {
+      TXT( "\nUsageMask  : " );
+      if( decl->Declaration.UsageMask & TGSI_WRITEMASK_X ) {
+         CHR( 'X' );
+      }
+      if( decl->Declaration.UsageMask & TGSI_WRITEMASK_Y ) {
+         CHR( 'Y' );
+      }
+      if( decl->Declaration.UsageMask & TGSI_WRITEMASK_Z ) {
+         CHR( 'Z' );
+      }
+      if( decl->Declaration.UsageMask & TGSI_WRITEMASK_W ) {
+         CHR( 'W' );
+      }
+   }
+   if( deflt || fd->Declaration.Interpolate != decl->Declaration.Interpolate ) {
+      TXT( "\nInterpolate: " );
+      ENM( decl->Declaration.Interpolate, TGSI_INTERPOLATES );
+   }
+   if( deflt || fd->Declaration.Semantic != decl->Declaration.Semantic ) {
+      TXT( "\nSemantic   : " );
+      UID( decl->Declaration.Semantic );
+   }
+   if( ignored ) {
+      TXT( "\nPadding    : " );
+      UIX( decl->Declaration.Padding );
+   }
+
+   EOL();
+   TXT( "\nFirst: " );
+   UID( decl->DeclarationRange.First );
+   TXT( "\nLast : " );
+   UID( decl->DeclarationRange.Last );
+
+   if( decl->Declaration.Semantic ) {
+      EOL();
+      TXT( "\nSemanticName : " );
+      ENM( decl->Semantic.SemanticName, TGSI_SEMANTICS );
+      TXT( "\nSemanticIndex: " );
+      UID( decl->Semantic.SemanticIndex );
+      if( ignored ) {
+         TXT( "\nPadding      : " );
+         UIX( decl->Semantic.Padding );
+      }
+   }
+}
+
+static void
+dump_immediate_verbose(
+   struct tgsi_full_immediate *imm,
+   unsigned                   ignored )
+{
+   unsigned i;
+
+   TXT( "\nDataType   : " );
+   ENM( imm->Immediate.DataType, TGSI_IMMS );
+   if( ignored ) {
+      TXT( "\nPadding    : " );
+      UIX( imm->Immediate.Padding );
+   }
+
+   for( i = 0; i < imm->Immediate.NrTokens - 1; i++ ) {
+      EOL();
+      switch( imm->Immediate.DataType ) {
+      case TGSI_IMM_FLOAT32:
+         TXT( "\nFloat: " );
+         FLT( imm->u.ImmediateFloat32[i].Float );
+         break;
+
+      default:
+         assert( 0 );
+      }
+   }
+}
+
+static void
+dump_instruction_verbose(
+   struct tgsi_full_instruction  *inst,
+   unsigned                      ignored,
+   unsigned                      deflt,
+   struct tgsi_full_instruction  *fi )
+{
+   unsigned i;
+
+   TXT( "\nOpcode     : OPCODE_" );
+   TXT( tgsi_get_opcode_info( inst->Instruction.Opcode )->mnemonic );
+   if( deflt || fi->Instruction.Saturate != inst->Instruction.Saturate ) {
+      TXT( "\nSaturate   : " );
+      ENM( inst->Instruction.Saturate, TGSI_SATS );
+   }
+   if( deflt || fi->Instruction.NumDstRegs != inst->Instruction.NumDstRegs ) {
+      TXT( "\nNumDstRegs : " );
+      UID( inst->Instruction.NumDstRegs );
+   }
+   if( deflt || fi->Instruction.NumSrcRegs != inst->Instruction.NumSrcRegs ) {
+      TXT( "\nNumSrcRegs : " );
+      UID( inst->Instruction.NumSrcRegs );
+   }
+   if( ignored ) {
+      TXT( "\nPadding    : " );
+      UIX( inst->Instruction.Padding );
+   }
+
+   if( deflt || tgsi_compare_instruction_ext_nv( inst->InstructionExtNv, fi->InstructionExtNv ) ) {
+      EOL();
+      TXT( "\nType          : " );
+      ENM( inst->InstructionExtNv.Type, TGSI_INSTRUCTION_EXTS );
+      if( deflt || fi->InstructionExtNv.Precision != inst->InstructionExtNv.Precision ) {
+         TXT( "\nPrecision     : " );
+         ENM( inst->InstructionExtNv.Precision, TGSI_PRECISIONS );
+      }
+      if( deflt || fi->InstructionExtNv.CondDstIndex != inst->InstructionExtNv.CondDstIndex ) {
+         TXT( "\nCondDstIndex  : " );
+         UID( inst->InstructionExtNv.CondDstIndex );
+      }
+      if( deflt || fi->InstructionExtNv.CondFlowIndex != inst->InstructionExtNv.CondFlowIndex ) {
+         TXT( "\nCondFlowIndex : " );
+         UID( inst->InstructionExtNv.CondFlowIndex );
+      }
+      if( deflt || fi->InstructionExtNv.CondMask != inst->InstructionExtNv.CondMask ) {
+         TXT( "\nCondMask      : " );
+         ENM( inst->InstructionExtNv.CondMask, TGSI_CCS );
+      }
+      if( deflt || fi->InstructionExtNv.CondSwizzleX != inst->InstructionExtNv.CondSwizzleX ) {
+         TXT( "\nCondSwizzleX  : " );
+         ENM( inst->InstructionExtNv.CondSwizzleX, TGSI_SWIZZLES );
+      }
+      if( deflt || fi->InstructionExtNv.CondSwizzleY != inst->InstructionExtNv.CondSwizzleY ) {
+         TXT( "\nCondSwizzleY  : " );
+         ENM( inst->InstructionExtNv.CondSwizzleY, TGSI_SWIZZLES );
+      }
+      if( deflt || fi->InstructionExtNv.CondSwizzleZ != inst->InstructionExtNv.CondSwizzleZ ) {
+         TXT( "\nCondSwizzleZ  : " );
+         ENM( inst->InstructionExtNv.CondSwizzleZ, TGSI_SWIZZLES );
+      }
+      if( deflt || fi->InstructionExtNv.CondSwizzleW != inst->InstructionExtNv.CondSwizzleW ) {
+         TXT( "\nCondSwizzleW  : " );
+         ENM( inst->InstructionExtNv.CondSwizzleW, TGSI_SWIZZLES );
+      }
+      if( deflt || fi->InstructionExtNv.CondDstUpdate != inst->InstructionExtNv.CondDstUpdate ) {
+         TXT( "\nCondDstUpdate : " );
+         UID( inst->InstructionExtNv.CondDstUpdate );
+      }
+      if( deflt || fi->InstructionExtNv.CondFlowEnable != inst->InstructionExtNv.CondFlowEnable ) {
+         TXT( "\nCondFlowEnable: " );
+         UID( inst->InstructionExtNv.CondFlowEnable );
+      }
+      if( ignored ) {
+         TXT( "\nPadding       : " );
+         UIX( inst->InstructionExtNv.Padding );
+         if( deflt || fi->InstructionExtNv.Extended != inst->InstructionExtNv.Extended ) {
+            TXT( "\nExtended      : " );
+            UID( inst->InstructionExtNv.Extended );
+         }
+      }
+   }
+
+   if( deflt || tgsi_compare_instruction_ext_label( inst->InstructionExtLabel, fi->InstructionExtLabel ) ) {
+      EOL();
+      TXT( "\nType    : " );
+      ENM( inst->InstructionExtLabel.Type, TGSI_INSTRUCTION_EXTS );
+      if( deflt || fi->InstructionExtLabel.Label != inst->InstructionExtLabel.Label ) {
+         TXT( "\nLabel   : " );
+         UID( inst->InstructionExtLabel.Label );
+      }
+      if( ignored ) {
+         TXT( "\nPadding : " );
+         UIX( inst->InstructionExtLabel.Padding );
+         if( deflt || fi->InstructionExtLabel.Extended != inst->InstructionExtLabel.Extended ) {
+            TXT( "\nExtended: " );
+            UID( inst->InstructionExtLabel.Extended );
+         }
+      }
+   }
+
+   if( deflt || tgsi_compare_instruction_ext_texture( inst->InstructionExtTexture, fi->InstructionExtTexture ) ) {
+      EOL();
+      TXT( "\nType    : " );
+      ENM( inst->InstructionExtTexture.Type, TGSI_INSTRUCTION_EXTS );
+      if( deflt || fi->InstructionExtTexture.Texture != inst->InstructionExtTexture.Texture ) {
+         TXT( "\nTexture : " );
+         ENM( inst->InstructionExtTexture.Texture, TGSI_TEXTURES );
+      }
+      if( ignored ) {
+         TXT( "\nPadding : " );
+         UIX( inst->InstructionExtTexture.Padding );
+         if( deflt || fi->InstructionExtTexture.Extended != inst->InstructionExtTexture.Extended ) {
+            TXT( "\nExtended: " );
+            UID( inst->InstructionExtTexture.Extended );
+         }
+      }
+   }
+
+   for( i = 0; i < inst->Instruction.NumDstRegs; i++ ) {
+      struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i];
+      struct tgsi_full_dst_register *fd = &fi->FullDstRegisters[i];
+
+      EOL();
+      TXT( "\nFile     : " );
+      ENM( dst->DstRegister.File, TGSI_FILES );
+      if( deflt || fd->DstRegister.WriteMask != dst->DstRegister.WriteMask ) {
+         TXT( "\nWriteMask: " );
+         ENM( dst->DstRegister.WriteMask, TGSI_WRITEMASKS );
+      }
+      if( ignored ) {
+         if( deflt || fd->DstRegister.Indirect != dst->DstRegister.Indirect ) {
+            TXT( "\nIndirect : " );
+            UID( dst->DstRegister.Indirect );
+         }
+         if( deflt || fd->DstRegister.Dimension != dst->DstRegister.Dimension ) {
+            TXT( "\nDimension: " );
+            UID( dst->DstRegister.Dimension );
+         }
+      }
+      if( deflt || fd->DstRegister.Index != dst->DstRegister.Index ) {
+         TXT( "\nIndex    : " );
+         SID( dst->DstRegister.Index );
+      }
+      if( ignored ) {
+         TXT( "\nPadding  : " );
+         UIX( dst->DstRegister.Padding );
+         if( deflt || fd->DstRegister.Extended != dst->DstRegister.Extended ) {
+            TXT( "\nExtended : " );
+            UID( dst->DstRegister.Extended );
+         }
+      }
+
+      if( deflt || tgsi_compare_dst_register_ext_concode( dst->DstRegisterExtConcode, fd->DstRegisterExtConcode ) ) {
+         EOL();
+         TXT( "\nType        : " );
+         ENM( dst->DstRegisterExtConcode.Type, TGSI_DST_REGISTER_EXTS );
+         if( deflt || fd->DstRegisterExtConcode.CondMask != dst->DstRegisterExtConcode.CondMask ) {
+            TXT( "\nCondMask    : " );
+            ENM( dst->DstRegisterExtConcode.CondMask, TGSI_CCS );
+         }
+         if( deflt || fd->DstRegisterExtConcode.CondSwizzleX != dst->DstRegisterExtConcode.CondSwizzleX ) {
+            TXT( "\nCondSwizzleX: " );
+            ENM( dst->DstRegisterExtConcode.CondSwizzleX, TGSI_SWIZZLES );
+         }
+         if( deflt || fd->DstRegisterExtConcode.CondSwizzleY != dst->DstRegisterExtConcode.CondSwizzleY ) {
+            TXT( "\nCondSwizzleY: " );
+            ENM( dst->DstRegisterExtConcode.CondSwizzleY, TGSI_SWIZZLES );
+         }
+         if( deflt || fd->DstRegisterExtConcode.CondSwizzleZ != dst->DstRegisterExtConcode.CondSwizzleZ ) {
+            TXT( "\nCondSwizzleZ: " );
+            ENM( dst->DstRegisterExtConcode.CondSwizzleZ, TGSI_SWIZZLES );
+         }
+         if( deflt || fd->DstRegisterExtConcode.CondSwizzleW != dst->DstRegisterExtConcode.CondSwizzleW ) {
+            TXT( "\nCondSwizzleW: " );
+            ENM( dst->DstRegisterExtConcode.CondSwizzleW, TGSI_SWIZZLES );
+         }
+         if( deflt || fd->DstRegisterExtConcode.CondSrcIndex != dst->DstRegisterExtConcode.CondSrcIndex ) {
+            TXT( "\nCondSrcIndex: " );
+            UID( dst->DstRegisterExtConcode.CondSrcIndex );
+         }
+         if( ignored ) {
+            TXT( "\nPadding     : " );
+            UIX( dst->DstRegisterExtConcode.Padding );
+            if( deflt || fd->DstRegisterExtConcode.Extended != dst->DstRegisterExtConcode.Extended ) {
+               TXT( "\nExtended    : " );
+               UID( dst->DstRegisterExtConcode.Extended );
+            }
+         }
+      }
+
+      if( deflt || tgsi_compare_dst_register_ext_modulate( dst->DstRegisterExtModulate, fd->DstRegisterExtModulate ) ) {
+         EOL();
+         TXT( "\nType    : " );
+         ENM( dst->DstRegisterExtModulate.Type, TGSI_DST_REGISTER_EXTS );
+         if( deflt || fd->DstRegisterExtModulate.Modulate != dst->DstRegisterExtModulate.Modulate ) {
+            TXT( "\nModulate: " );
+            ENM( dst->DstRegisterExtModulate.Modulate, TGSI_MODULATES );
+         }
+         if( ignored ) {
+            TXT( "\nPadding : " );
+            UIX( dst->DstRegisterExtModulate.Padding );
+            if( deflt || fd->DstRegisterExtModulate.Extended != dst->DstRegisterExtModulate.Extended ) {
+               TXT( "\nExtended: " );
+               UID( dst->DstRegisterExtModulate.Extended );
+            }
+         }
+      }
+   }
+
+   for( i = 0; i < inst->Instruction.NumSrcRegs; i++ ) {
+      struct tgsi_full_src_register *src = &inst->FullSrcRegisters[i];
+      struct tgsi_full_src_register *fs = &fi->FullSrcRegisters[i];
+
+      EOL();
+      TXT( "\nFile     : ");
+      ENM( src->SrcRegister.File, TGSI_FILES );
+      if( deflt || fs->SrcRegister.SwizzleX != src->SrcRegister.SwizzleX ) {
+         TXT( "\nSwizzleX : " );
+         ENM( src->SrcRegister.SwizzleX, TGSI_SWIZZLES );
+      }
+      if( deflt || fs->SrcRegister.SwizzleY != src->SrcRegister.SwizzleY ) {
+         TXT( "\nSwizzleY : " );
+         ENM( src->SrcRegister.SwizzleY, TGSI_SWIZZLES );
+      }
+      if( deflt || fs->SrcRegister.SwizzleZ != src->SrcRegister.SwizzleZ ) {
+         TXT( "\nSwizzleZ : " );
+         ENM( src->SrcRegister.SwizzleZ, TGSI_SWIZZLES );
+      }
+      if( deflt || fs->SrcRegister.SwizzleW != src->SrcRegister.SwizzleW ) {
+         TXT( "\nSwizzleW : " );
+         ENM( src->SrcRegister.SwizzleW, TGSI_SWIZZLES );
+      }
+      if( deflt || fs->SrcRegister.Negate != src->SrcRegister.Negate ) {
+         TXT( "\nNegate   : " );
+         UID( src->SrcRegister.Negate );
+      }
+      if( ignored ) {
+         if( deflt || fs->SrcRegister.Indirect != src->SrcRegister.Indirect ) {
+            TXT( "\nIndirect : " );
+            UID( src->SrcRegister.Indirect );
+         }
+         if( deflt || fs->SrcRegister.Dimension != src->SrcRegister.Dimension ) {
+            TXT( "\nDimension: " );
+            UID( src->SrcRegister.Dimension );
+         }
+      }
+      if( deflt || fs->SrcRegister.Index != src->SrcRegister.Index ) {
+         TXT( "\nIndex    : " );
+         SID( src->SrcRegister.Index );
+      }
+      if( ignored ) {
+         if( deflt || fs->SrcRegister.Extended != src->SrcRegister.Extended ) {
+            TXT( "\nExtended : " );
+            UID( src->SrcRegister.Extended );
+         }
+      }
+
+      if( deflt || tgsi_compare_src_register_ext_swz( src->SrcRegisterExtSwz, fs->SrcRegisterExtSwz ) ) {
+         EOL();
+         TXT( "\nType       : " );
+         ENM( src->SrcRegisterExtSwz.Type, TGSI_SRC_REGISTER_EXTS );
+         if( deflt || fs->SrcRegisterExtSwz.ExtSwizzleX != src->SrcRegisterExtSwz.ExtSwizzleX ) {
+            TXT( "\nExtSwizzleX: " );
+            ENM( src->SrcRegisterExtSwz.ExtSwizzleX, TGSI_EXTSWIZZLES );
+         }
+         if( deflt || fs->SrcRegisterExtSwz.ExtSwizzleY != src->SrcRegisterExtSwz.ExtSwizzleY ) {
+            TXT( "\nExtSwizzleY: " );
+            ENM( src->SrcRegisterExtSwz.ExtSwizzleY, TGSI_EXTSWIZZLES );
+         }
+         if( deflt || fs->SrcRegisterExtSwz.ExtSwizzleZ != src->SrcRegisterExtSwz.ExtSwizzleZ ) {
+            TXT( "\nExtSwizzleZ: " );
+            ENM( src->SrcRegisterExtSwz.ExtSwizzleZ, TGSI_EXTSWIZZLES );
+         }
+         if( deflt || fs->SrcRegisterExtSwz.ExtSwizzleW != src->SrcRegisterExtSwz.ExtSwizzleW ) {
+            TXT( "\nExtSwizzleW: " );
+            ENM( src->SrcRegisterExtSwz.ExtSwizzleW, TGSI_EXTSWIZZLES );
+         }
+         if( deflt || fs->SrcRegisterExtSwz.NegateX != src->SrcRegisterExtSwz.NegateX ) {
+            TXT( "\nNegateX   : " );
+            UID( src->SrcRegisterExtSwz.NegateX );
+         }
+         if( deflt || fs->SrcRegisterExtSwz.NegateY != src->SrcRegisterExtSwz.NegateY ) {
+            TXT( "\nNegateY   : " );
+            UID( src->SrcRegisterExtSwz.NegateY );
+         }
+         if( deflt || fs->SrcRegisterExtSwz.NegateZ != src->SrcRegisterExtSwz.NegateZ ) {
+            TXT( "\nNegateZ   : " );
+            UID( src->SrcRegisterExtSwz.NegateZ );
+         }
+         if( deflt || fs->SrcRegisterExtSwz.NegateW != src->SrcRegisterExtSwz.NegateW ) {
+            TXT( "\nNegateW   : " );
+            UID( src->SrcRegisterExtSwz.NegateW );
+         }
+         if( ignored ) {
+            TXT( "\nPadding   : " );
+            UIX( src->SrcRegisterExtSwz.Padding );
+            if( deflt || fs->SrcRegisterExtSwz.Extended != src->SrcRegisterExtSwz.Extended ) {
+               TXT( "\nExtended   : " );
+               UID( src->SrcRegisterExtSwz.Extended );
+            }
+         }
+      }
+
+      if( deflt || tgsi_compare_src_register_ext_mod( src->SrcRegisterExtMod, fs->SrcRegisterExtMod ) ) {
+         EOL();
+         TXT( "\nType     : " );
+         ENM( src->SrcRegisterExtMod.Type, TGSI_SRC_REGISTER_EXTS );
+         if( deflt || fs->SrcRegisterExtMod.Complement != src->SrcRegisterExtMod.Complement ) {
+            TXT( "\nComplement: " );
+            UID( src->SrcRegisterExtMod.Complement );
+         }
+         if( deflt || fs->SrcRegisterExtMod.Bias != src->SrcRegisterExtMod.Bias ) {
+            TXT( "\nBias     : " );
+            UID( src->SrcRegisterExtMod.Bias );
+         }
+         if( deflt || fs->SrcRegisterExtMod.Scale2X != src->SrcRegisterExtMod.Scale2X ) {
+            TXT( "\nScale2X   : " );
+            UID( src->SrcRegisterExtMod.Scale2X );
+         }
+         if( deflt || fs->SrcRegisterExtMod.Absolute != src->SrcRegisterExtMod.Absolute ) {
+            TXT( "\nAbsolute  : " );
+            UID( src->SrcRegisterExtMod.Absolute );
+         }
+         if( deflt || fs->SrcRegisterExtMod.Negate != src->SrcRegisterExtMod.Negate ) {
+            TXT( "\nNegate   : " );
+            UID( src->SrcRegisterExtMod.Negate );
+         }
+         if( ignored ) {
+            TXT( "\nPadding   : " );
+            UIX( src->SrcRegisterExtMod.Padding );
+            if( deflt || fs->SrcRegisterExtMod.Extended != src->SrcRegisterExtMod.Extended ) {
+               TXT( "\nExtended  : " );
+               UID( src->SrcRegisterExtMod.Extended );
+            }
+         }
+      }
+   }
+}
+
+void
+tgsi_dump_c(
+   const struct tgsi_token *tokens,
+   uint flags )
+{
+   struct tgsi_parse_context parse;
+   struct tgsi_full_instruction fi;
+   struct tgsi_full_declaration fd;
+   uint ignored = flags & TGSI_DUMP_C_IGNORED;
+   uint deflt = flags & TGSI_DUMP_C_DEFAULT;
+
+   tgsi_parse_init( &parse, tokens );
+
+   TXT( "tgsi-dump begin -----------------" );
+
+   TXT( "\nMajorVersion: " );
+   UID( parse.FullVersion.Version.MajorVersion );
+   TXT( "\nMinorVersion: " );
+   UID( parse.FullVersion.Version.MinorVersion );
+   EOL();
+
+   TXT( "\nHeaderSize: " );
+   UID( parse.FullHeader.Header.HeaderSize );
+   TXT( "\nBodySize  : " );
+   UID( parse.FullHeader.Header.BodySize );
+   TXT( "\nProcessor : " );
+   ENM( parse.FullHeader.Processor.Processor, TGSI_PROCESSOR_TYPES );
+   EOL();
+
+   fi = tgsi_default_full_instruction();
+   fd = tgsi_default_full_declaration();
+
+   while( !tgsi_parse_end_of_tokens( &parse ) ) {
+      tgsi_parse_token( &parse );
+
+      TXT( "\nType       : " );
+      ENM( parse.FullToken.Token.Type, TGSI_TOKEN_TYPES );
+      if( ignored ) {
+         TXT( "\nSize       : " );
+         UID( parse.FullToken.Token.NrTokens );
+         if( deflt || parse.FullToken.Token.Extended ) {
+            TXT( "\nExtended   : " );
+            UID( parse.FullToken.Token.Extended );
+         }
+      }
+
+      switch( parse.FullToken.Token.Type ) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         dump_declaration_verbose(
+            &parse.FullToken.FullDeclaration,
+            ignored,
+            deflt,
+            &fd );
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         dump_immediate_verbose(
+            &parse.FullToken.FullImmediate,
+            ignored );
+         break;
+
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         dump_instruction_verbose(
+            &parse.FullToken.FullInstruction,
+            ignored,
+            deflt,
+            &fi );
+         break;
+
+      default:
+         assert( 0 );
+      }
+
+      EOL();
+   }
+
+   TXT( "\ntgsi-dump end -------------------\n" );
+
+   tgsi_parse_free( &parse );
+}
diff --git a/src/gallium/auxiliary/tgsi/tgsi_dump_c.h b/src/gallium/auxiliary/tgsi/tgsi_dump_c.h
new file mode 100644
index 00000000000..d91cd35b3b7
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_dump_c.h
@@ -0,0 +1,49 @@
+/**************************************************************************
+ * 
+ * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef TGSI_DUMP_C_H
+#define TGSI_DUMP_C_H
+
+#include "pipe/p_shader_tokens.h"
+
+#if defined __cplusplus
+extern "C" {
+#endif
+
+#define TGSI_DUMP_C_IGNORED 1
+#define TGSI_DUMP_C_DEFAULT 2
+
+void
+tgsi_dump_c(
+   const struct tgsi_token *tokens,
+   uint flags );
+
+#if defined __cplusplus
+}
+#endif
+
+#endif /* TGSI_DUMP_C_H */
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.c b/src/gallium/auxiliary/tgsi/tgsi_exec.c
new file mode 100644
index 00000000000..ab641efb603
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.c
@@ -0,0 +1,2960 @@
+/**************************************************************************
+ * 
+ * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * TGSI interpreter/executor.
+ *
+ * Flow control information:
+ *
+ * Since we operate on 'quads' (4 pixels or 4 vertices in parallel)
+ * flow control statements (IF/ELSE/ENDIF, LOOP/ENDLOOP) require special
+ * care since a condition may be true for some quad components but false
+ * for other components.
+ *
+ * We basically execute all statements (even if they're in the part of
+ * an IF/ELSE clause that's "not taken") and use a special mask to
+ * control writing to destination registers.  This is the ExecMask.
+ * See store_dest().
+ *
+ * The ExecMask is computed from three other masks (CondMask, LoopMask and
+ * ContMask) which are controlled by the flow control instructions (namely:
+ * (IF/ELSE/ENDIF, LOOP/ENDLOOP and CONT).
+ *
+ *
+ * Authors:
+ *   Michal Krol
+ *   Brian Paul
+ */
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_state.h"
+#include "pipe/p_shader_tokens.h"
+#include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_util.h"
+#include "tgsi_exec.h"
+#include "util/u_memory.h"
+#include "util/u_math.h"
+
+#define FAST_MATH 1
+
+#define TILE_TOP_LEFT     0
+#define TILE_TOP_RIGHT    1
+#define TILE_BOTTOM_LEFT  2
+#define TILE_BOTTOM_RIGHT 3
+
+#define CHAN_X  0
+#define CHAN_Y  1
+#define CHAN_Z  2
+#define CHAN_W  3
+
+/*
+ * Shorthand locations of various utility registers (_I = Index, _C = Channel)
+ */
+#define TEMP_0_I           TGSI_EXEC_TEMP_00000000_I
+#define TEMP_0_C           TGSI_EXEC_TEMP_00000000_C
+#define TEMP_7F_I          TGSI_EXEC_TEMP_7FFFFFFF_I
+#define TEMP_7F_C          TGSI_EXEC_TEMP_7FFFFFFF_C
+#define TEMP_80_I          TGSI_EXEC_TEMP_80000000_I
+#define TEMP_80_C          TGSI_EXEC_TEMP_80000000_C
+#define TEMP_FF_I          TGSI_EXEC_TEMP_FFFFFFFF_I
+#define TEMP_FF_C          TGSI_EXEC_TEMP_FFFFFFFF_C
+#define TEMP_1_I           TGSI_EXEC_TEMP_ONE_I
+#define TEMP_1_C           TGSI_EXEC_TEMP_ONE_C
+#define TEMP_2_I           TGSI_EXEC_TEMP_TWO_I
+#define TEMP_2_C           TGSI_EXEC_TEMP_TWO_C
+#define TEMP_128_I         TGSI_EXEC_TEMP_128_I
+#define TEMP_128_C         TGSI_EXEC_TEMP_128_C
+#define TEMP_M128_I        TGSI_EXEC_TEMP_MINUS_128_I
+#define TEMP_M128_C        TGSI_EXEC_TEMP_MINUS_128_C
+#define TEMP_KILMASK_I     TGSI_EXEC_TEMP_KILMASK_I
+#define TEMP_KILMASK_C     TGSI_EXEC_TEMP_KILMASK_C
+#define TEMP_OUTPUT_I      TGSI_EXEC_TEMP_OUTPUT_I
+#define TEMP_OUTPUT_C      TGSI_EXEC_TEMP_OUTPUT_C
+#define TEMP_PRIMITIVE_I   TGSI_EXEC_TEMP_PRIMITIVE_I
+#define TEMP_PRIMITIVE_C   TGSI_EXEC_TEMP_PRIMITIVE_C
+#define TEMP_CC_I          TGSI_EXEC_TEMP_CC_I
+#define TEMP_CC_C          TGSI_EXEC_TEMP_CC_C
+#define TEMP_3_I           TGSI_EXEC_TEMP_THREE_I
+#define TEMP_3_C           TGSI_EXEC_TEMP_THREE_C
+#define TEMP_HALF_I        TGSI_EXEC_TEMP_HALF_I
+#define TEMP_HALF_C        TGSI_EXEC_TEMP_HALF_C
+#define TEMP_R0            TGSI_EXEC_TEMP_R0
+
+#define IS_CHANNEL_ENABLED(INST, CHAN)\
+   ((INST).FullDstRegisters[0].DstRegister.WriteMask & (1 << (CHAN)))
+
+#define IS_CHANNEL_ENABLED2(INST, CHAN)\
+   ((INST).FullDstRegisters[1].DstRegister.WriteMask & (1 << (CHAN)))
+
+#define FOR_EACH_ENABLED_CHANNEL(INST, CHAN)\
+   for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++)\
+      if (IS_CHANNEL_ENABLED( INST, CHAN ))
+
+#define FOR_EACH_ENABLED_CHANNEL2(INST, CHAN)\
+   for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++)\
+      if (IS_CHANNEL_ENABLED2( INST, CHAN ))
+
+
+/** The execution mask depends on the conditional mask and the loop mask */
+#define UPDATE_EXEC_MASK(MACH) \
+      MACH->ExecMask = MACH->CondMask & MACH->LoopMask & MACH->ContMask & MACH->FuncMask
+
+/**
+ * Initialize machine state by expanding tokens to full instructions,
+ * allocating temporary storage, setting up constants, etc.
+ * After this, we can call tgsi_exec_machine_run() many times.
+ */
+void 
+tgsi_exec_machine_bind_shader(
+   struct tgsi_exec_machine *mach,
+   const struct tgsi_token *tokens,
+   uint numSamplers,
+   struct tgsi_sampler **samplers)
+{
+   uint k;
+   struct tgsi_parse_context parse;
+   struct tgsi_exec_labels *labels = &mach->Labels;
+   struct tgsi_full_instruction *instructions;
+   struct tgsi_full_declaration *declarations;
+   uint maxInstructions = 10, numInstructions = 0;
+   uint maxDeclarations = 10, numDeclarations = 0;
+   uint instno = 0;
+
+#if 0
+   tgsi_dump(tokens, 0);
+#endif
+
+   util_init_math();
+
+   mach->Tokens = tokens;
+   mach->Samplers = samplers;
+
+   k = tgsi_parse_init (&parse, mach->Tokens);
+   if (k != TGSI_PARSE_OK) {
+      debug_printf( "Problem parsing!\n" );
+      return;
+   }
+
+   mach->Processor = parse.FullHeader.Processor.Processor;
+   mach->ImmLimit = 0;
+   labels->count = 0;
+
+   declarations = (struct tgsi_full_declaration *)
+      MALLOC( maxDeclarations * sizeof(struct tgsi_full_declaration) );
+
+   if (!declarations) {
+      return;
+   }
+
+   instructions = (struct tgsi_full_instruction *)
+      MALLOC( maxInstructions * sizeof(struct tgsi_full_instruction) );
+
+   if (!instructions) {
+      FREE( declarations );
+      return;
+   }
+
+   while( !tgsi_parse_end_of_tokens( &parse ) ) {
+      uint pointer = parse.Position;
+      uint i;
+
+      tgsi_parse_token( &parse );
+      switch( parse.FullToken.Token.Type ) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         /* save expanded declaration */
+         if (numDeclarations == maxDeclarations) {
+            declarations = REALLOC(declarations,
+                                   maxDeclarations
+                                   * sizeof(struct tgsi_full_declaration),
+                                   (maxDeclarations + 10)
+                                   * sizeof(struct tgsi_full_declaration));
+            maxDeclarations += 10;
+         }
+         memcpy(declarations + numDeclarations,
+                &parse.FullToken.FullDeclaration,
+                sizeof(declarations[0]));
+         numDeclarations++;
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         {
+            uint size = parse.FullToken.FullImmediate.Immediate.NrTokens - 1;
+            assert( size % 4 == 0 );
+            assert( mach->ImmLimit + size / 4 <= TGSI_EXEC_NUM_IMMEDIATES );
+
+            for( i = 0; i < size; i++ ) {
+               mach->Imms[mach->ImmLimit + i / 4][i % 4] = 
+		  parse.FullToken.FullImmediate.u.ImmediateFloat32[i].Float;
+            }
+            mach->ImmLimit += size / 4;
+         }
+         break;
+
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         assert( labels->count < MAX_LABELS );
+
+         labels->labels[labels->count][0] = instno;
+         labels->labels[labels->count][1] = pointer;
+         labels->count++;
+
+         /* save expanded instruction */
+         if (numInstructions == maxInstructions) {
+            instructions = REALLOC(instructions,
+                                   maxInstructions
+                                   * sizeof(struct tgsi_full_instruction),
+                                   (maxInstructions + 10)
+                                   * sizeof(struct tgsi_full_instruction));
+            maxInstructions += 10;
+         }
+         memcpy(instructions + numInstructions,
+                &parse.FullToken.FullInstruction,
+                sizeof(instructions[0]));
+         numInstructions++;
+         break;
+
+      default:
+         assert( 0 );
+      }
+   }
+   tgsi_parse_free (&parse);
+
+   if (mach->Declarations) {
+      FREE( mach->Declarations );
+   }
+   mach->Declarations = declarations;
+   mach->NumDeclarations = numDeclarations;
+
+   if (mach->Instructions) {
+      FREE( mach->Instructions );
+   }
+   mach->Instructions = instructions;
+   mach->NumInstructions = numInstructions;
+}
+
+
+void
+tgsi_exec_machine_init(
+   struct tgsi_exec_machine *mach )
+{
+   uint i;
+
+   mach->Temps = (struct tgsi_exec_vector *) tgsi_align_128bit( mach->_Temps);
+   mach->Addrs = &mach->Temps[TGSI_EXEC_TEMP_ADDR];
+
+   /* Setup constants. */
+   for( i = 0; i < 4; i++ ) {
+      mach->Temps[TEMP_0_I].xyzw[TEMP_0_C].u[i] = 0x00000000;
+      mach->Temps[TEMP_7F_I].xyzw[TEMP_7F_C].u[i] = 0x7FFFFFFF;
+      mach->Temps[TEMP_80_I].xyzw[TEMP_80_C].u[i] = 0x80000000;
+      mach->Temps[TEMP_FF_I].xyzw[TEMP_FF_C].u[i] = 0xFFFFFFFF;
+      mach->Temps[TEMP_1_I].xyzw[TEMP_1_C].f[i] = 1.0f;
+      mach->Temps[TEMP_2_I].xyzw[TEMP_2_C].f[i] = 2.0f;
+      mach->Temps[TEMP_128_I].xyzw[TEMP_128_C].f[i] = 128.0f;
+      mach->Temps[TEMP_M128_I].xyzw[TEMP_M128_C].f[i] = -128.0f;
+      mach->Temps[TEMP_3_I].xyzw[TEMP_3_C].f[i] = 3.0f;
+      mach->Temps[TEMP_HALF_I].xyzw[TEMP_HALF_C].f[i] = 0.5f;
+   }
+}
+
+
+void
+tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach)
+{
+   if (mach->Instructions) {
+      FREE(mach->Instructions);
+      mach->Instructions = NULL;
+      mach->NumInstructions = 0;
+   }
+   if (mach->Declarations) {
+      FREE(mach->Declarations);
+      mach->Declarations = NULL;
+      mach->NumDeclarations = 0;
+   }
+}
+
+
+static void
+micro_abs(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = fabsf( src->f[0] );
+   dst->f[1] = fabsf( src->f[1] );
+   dst->f[2] = fabsf( src->f[2] );
+   dst->f[3] = fabsf( src->f[3] );
+}
+
+static void
+micro_add(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->f[0] = src0->f[0] + src1->f[0];
+   dst->f[1] = src0->f[1] + src1->f[1];
+   dst->f[2] = src0->f[2] + src1->f[2];
+   dst->f[3] = src0->f[3] + src1->f[3];
+}
+
+#if 0
+static void
+micro_iadd(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->i[0] = src0->i[0] + src1->i[0];
+   dst->i[1] = src0->i[1] + src1->i[1];
+   dst->i[2] = src0->i[2] + src1->i[2];
+   dst->i[3] = src0->i[3] + src1->i[3];
+}
+#endif
+
+static void
+micro_and(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->u[0] = src0->u[0] & src1->u[0];
+   dst->u[1] = src0->u[1] & src1->u[1];
+   dst->u[2] = src0->u[2] & src1->u[2];
+   dst->u[3] = src0->u[3] & src1->u[3];
+}
+
+static void
+micro_ceil(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = ceilf( src->f[0] );
+   dst->f[1] = ceilf( src->f[1] );
+   dst->f[2] = ceilf( src->f[2] );
+   dst->f[3] = ceilf( src->f[3] );
+}
+
+static void
+micro_cos(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = cosf( src->f[0] );
+   dst->f[1] = cosf( src->f[1] );
+   dst->f[2] = cosf( src->f[2] );
+   dst->f[3] = cosf( src->f[3] );
+}
+
+static void
+micro_ddx(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] =
+   dst->f[1] =
+   dst->f[2] =
+   dst->f[3] = src->f[TILE_BOTTOM_RIGHT] - src->f[TILE_BOTTOM_LEFT];
+}
+
+static void
+micro_ddy(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] =
+   dst->f[1] =
+   dst->f[2] =
+   dst->f[3] = src->f[TILE_TOP_LEFT] - src->f[TILE_BOTTOM_LEFT];
+}
+
+static void
+micro_div(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   if (src1->f[0] != 0) {
+      dst->f[0] = src0->f[0] / src1->f[0];
+   }
+   if (src1->f[1] != 0) {
+      dst->f[1] = src0->f[1] / src1->f[1];
+   }
+   if (src1->f[2] != 0) {
+      dst->f[2] = src0->f[2] / src1->f[2];
+   }
+   if (src1->f[3] != 0) {
+      dst->f[3] = src0->f[3] / src1->f[3];
+   }
+}
+
+#if 0
+static void
+micro_udiv(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->u[0] = src0->u[0] / src1->u[0];
+   dst->u[1] = src0->u[1] / src1->u[1];
+   dst->u[2] = src0->u[2] / src1->u[2];
+   dst->u[3] = src0->u[3] / src1->u[3];
+}
+#endif
+
+static void
+micro_eq(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1,
+   const union tgsi_exec_channel *src2,
+   const union tgsi_exec_channel *src3 )
+{
+   dst->f[0] = src0->f[0] == src1->f[0] ? src2->f[0] : src3->f[0];
+   dst->f[1] = src0->f[1] == src1->f[1] ? src2->f[1] : src3->f[1];
+   dst->f[2] = src0->f[2] == src1->f[2] ? src2->f[2] : src3->f[2];
+   dst->f[3] = src0->f[3] == src1->f[3] ? src2->f[3] : src3->f[3];
+}
+
+#if 0
+static void
+micro_ieq(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1,
+   const union tgsi_exec_channel *src2,
+   const union tgsi_exec_channel *src3 )
+{
+   dst->i[0] = src0->i[0] == src1->i[0] ? src2->i[0] : src3->i[0];
+   dst->i[1] = src0->i[1] == src1->i[1] ? src2->i[1] : src3->i[1];
+   dst->i[2] = src0->i[2] == src1->i[2] ? src2->i[2] : src3->i[2];
+   dst->i[3] = src0->i[3] == src1->i[3] ? src2->i[3] : src3->i[3];
+}
+#endif
+
+static void
+micro_exp2(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src)
+{
+#if FAST_MATH
+   dst->f[0] = util_fast_exp2( src->f[0] );
+   dst->f[1] = util_fast_exp2( src->f[1] );
+   dst->f[2] = util_fast_exp2( src->f[2] );
+   dst->f[3] = util_fast_exp2( src->f[3] );
+#else
+   dst->f[0] = powf( 2.0f, src->f[0] );
+   dst->f[1] = powf( 2.0f, src->f[1] );
+   dst->f[2] = powf( 2.0f, src->f[2] );
+   dst->f[3] = powf( 2.0f, src->f[3] );
+#endif
+}
+
+#if 0
+static void
+micro_f2ut(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->u[0] = (uint) src->f[0];
+   dst->u[1] = (uint) src->f[1];
+   dst->u[2] = (uint) src->f[2];
+   dst->u[3] = (uint) src->f[3];
+}
+#endif
+
+static void
+micro_flr(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = floorf( src->f[0] );
+   dst->f[1] = floorf( src->f[1] );
+   dst->f[2] = floorf( src->f[2] );
+   dst->f[3] = floorf( src->f[3] );
+}
+
+static void
+micro_frc(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = src->f[0] - floorf( src->f[0] );
+   dst->f[1] = src->f[1] - floorf( src->f[1] );
+   dst->f[2] = src->f[2] - floorf( src->f[2] );
+   dst->f[3] = src->f[3] - floorf( src->f[3] );
+}
+
+static void
+micro_ge(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1,
+   const union tgsi_exec_channel *src2,
+   const union tgsi_exec_channel *src3 )
+{
+   dst->f[0] = src0->f[0] >= src1->f[0] ? src2->f[0] : src3->f[0];
+   dst->f[1] = src0->f[1] >= src1->f[1] ? src2->f[1] : src3->f[1];
+   dst->f[2] = src0->f[2] >= src1->f[2] ? src2->f[2] : src3->f[2];
+   dst->f[3] = src0->f[3] >= src1->f[3] ? src2->f[3] : src3->f[3];
+}
+
+static void
+micro_i2f(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = (float) src->i[0];
+   dst->f[1] = (float) src->i[1];
+   dst->f[2] = (float) src->i[2];
+   dst->f[3] = (float) src->i[3];
+}
+
+static void
+micro_lg2(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+#if FAST_MATH
+   dst->f[0] = util_fast_log2( src->f[0] );
+   dst->f[1] = util_fast_log2( src->f[1] );
+   dst->f[2] = util_fast_log2( src->f[2] );
+   dst->f[3] = util_fast_log2( src->f[3] );
+#else
+   dst->f[0] = logf( src->f[0] ) * 1.442695f;
+   dst->f[1] = logf( src->f[1] ) * 1.442695f;
+   dst->f[2] = logf( src->f[2] ) * 1.442695f;
+   dst->f[3] = logf( src->f[3] ) * 1.442695f;
+#endif
+}
+
+static void
+micro_le(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1,
+   const union tgsi_exec_channel *src2,
+   const union tgsi_exec_channel *src3 )
+{
+   dst->f[0] = src0->f[0] <= src1->f[0] ? src2->f[0] : src3->f[0];
+   dst->f[1] = src0->f[1] <= src1->f[1] ? src2->f[1] : src3->f[1];
+   dst->f[2] = src0->f[2] <= src1->f[2] ? src2->f[2] : src3->f[2];
+   dst->f[3] = src0->f[3] <= src1->f[3] ? src2->f[3] : src3->f[3];
+}
+
+static void
+micro_lt(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1,
+   const union tgsi_exec_channel *src2,
+   const union tgsi_exec_channel *src3 )
+{
+   dst->f[0] = src0->f[0] < src1->f[0] ? src2->f[0] : src3->f[0];
+   dst->f[1] = src0->f[1] < src1->f[1] ? src2->f[1] : src3->f[1];
+   dst->f[2] = src0->f[2] < src1->f[2] ? src2->f[2] : src3->f[2];
+   dst->f[3] = src0->f[3] < src1->f[3] ? src2->f[3] : src3->f[3];
+}
+
+#if 0
+static void
+micro_ilt(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1,
+   const union tgsi_exec_channel *src2,
+   const union tgsi_exec_channel *src3 )
+{
+   dst->i[0] = src0->i[0] < src1->i[0] ? src2->i[0] : src3->i[0];
+   dst->i[1] = src0->i[1] < src1->i[1] ? src2->i[1] : src3->i[1];
+   dst->i[2] = src0->i[2] < src1->i[2] ? src2->i[2] : src3->i[2];
+   dst->i[3] = src0->i[3] < src1->i[3] ? src2->i[3] : src3->i[3];
+}
+#endif
+
+#if 0
+static void
+micro_ult(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1,
+   const union tgsi_exec_channel *src2,
+   const union tgsi_exec_channel *src3 )
+{
+   dst->u[0] = src0->u[0] < src1->u[0] ? src2->u[0] : src3->u[0];
+   dst->u[1] = src0->u[1] < src1->u[1] ? src2->u[1] : src3->u[1];
+   dst->u[2] = src0->u[2] < src1->u[2] ? src2->u[2] : src3->u[2];
+   dst->u[3] = src0->u[3] < src1->u[3] ? src2->u[3] : src3->u[3];
+}
+#endif
+
+static void
+micro_max(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->f[0] = src0->f[0] > src1->f[0] ? src0->f[0] : src1->f[0];
+   dst->f[1] = src0->f[1] > src1->f[1] ? src0->f[1] : src1->f[1];
+   dst->f[2] = src0->f[2] > src1->f[2] ? src0->f[2] : src1->f[2];
+   dst->f[3] = src0->f[3] > src1->f[3] ? src0->f[3] : src1->f[3];
+}
+
+#if 0
+static void
+micro_imax(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->i[0] = src0->i[0] > src1->i[0] ? src0->i[0] : src1->i[0];
+   dst->i[1] = src0->i[1] > src1->i[1] ? src0->i[1] : src1->i[1];
+   dst->i[2] = src0->i[2] > src1->i[2] ? src0->i[2] : src1->i[2];
+   dst->i[3] = src0->i[3] > src1->i[3] ? src0->i[3] : src1->i[3];
+}
+#endif
+
+#if 0
+static void
+micro_umax(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->u[0] = src0->u[0] > src1->u[0] ? src0->u[0] : src1->u[0];
+   dst->u[1] = src0->u[1] > src1->u[1] ? src0->u[1] : src1->u[1];
+   dst->u[2] = src0->u[2] > src1->u[2] ? src0->u[2] : src1->u[2];
+   dst->u[3] = src0->u[3] > src1->u[3] ? src0->u[3] : src1->u[3];
+}
+#endif
+
+static void
+micro_min(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->f[0] = src0->f[0] < src1->f[0] ? src0->f[0] : src1->f[0];
+   dst->f[1] = src0->f[1] < src1->f[1] ? src0->f[1] : src1->f[1];
+   dst->f[2] = src0->f[2] < src1->f[2] ? src0->f[2] : src1->f[2];
+   dst->f[3] = src0->f[3] < src1->f[3] ? src0->f[3] : src1->f[3];
+}
+
+#if 0
+static void
+micro_imin(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->i[0] = src0->i[0] < src1->i[0] ? src0->i[0] : src1->i[0];
+   dst->i[1] = src0->i[1] < src1->i[1] ? src0->i[1] : src1->i[1];
+   dst->i[2] = src0->i[2] < src1->i[2] ? src0->i[2] : src1->i[2];
+   dst->i[3] = src0->i[3] < src1->i[3] ? src0->i[3] : src1->i[3];
+}
+#endif
+
+#if 0
+static void
+micro_umin(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->u[0] = src0->u[0] < src1->u[0] ? src0->u[0] : src1->u[0];
+   dst->u[1] = src0->u[1] < src1->u[1] ? src0->u[1] : src1->u[1];
+   dst->u[2] = src0->u[2] < src1->u[2] ? src0->u[2] : src1->u[2];
+   dst->u[3] = src0->u[3] < src1->u[3] ? src0->u[3] : src1->u[3];
+}
+#endif
+
+#if 0
+static void
+micro_umod(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->u[0] = src0->u[0] % src1->u[0];
+   dst->u[1] = src0->u[1] % src1->u[1];
+   dst->u[2] = src0->u[2] % src1->u[2];
+   dst->u[3] = src0->u[3] % src1->u[3];
+}
+#endif
+
+static void
+micro_mul(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->f[0] = src0->f[0] * src1->f[0];
+   dst->f[1] = src0->f[1] * src1->f[1];
+   dst->f[2] = src0->f[2] * src1->f[2];
+   dst->f[3] = src0->f[3] * src1->f[3];
+}
+
+#if 0
+static void
+micro_imul(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->i[0] = src0->i[0] * src1->i[0];
+   dst->i[1] = src0->i[1] * src1->i[1];
+   dst->i[2] = src0->i[2] * src1->i[2];
+   dst->i[3] = src0->i[3] * src1->i[3];
+}
+#endif
+
+#if 0
+static void
+micro_imul64(
+   union tgsi_exec_channel *dst0,
+   union tgsi_exec_channel *dst1,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst1->i[0] = src0->i[0] * src1->i[0];
+   dst1->i[1] = src0->i[1] * src1->i[1];
+   dst1->i[2] = src0->i[2] * src1->i[2];
+   dst1->i[3] = src0->i[3] * src1->i[3];
+   dst0->i[0] = 0;
+   dst0->i[1] = 0;
+   dst0->i[2] = 0;
+   dst0->i[3] = 0;
+}
+#endif
+
+#if 0
+static void
+micro_umul64(
+   union tgsi_exec_channel *dst0,
+   union tgsi_exec_channel *dst1,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst1->u[0] = src0->u[0] * src1->u[0];
+   dst1->u[1] = src0->u[1] * src1->u[1];
+   dst1->u[2] = src0->u[2] * src1->u[2];
+   dst1->u[3] = src0->u[3] * src1->u[3];
+   dst0->u[0] = 0;
+   dst0->u[1] = 0;
+   dst0->u[2] = 0;
+   dst0->u[3] = 0;
+}
+#endif
+
+
+#if 0
+static void
+micro_movc(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1,
+   const union tgsi_exec_channel *src2 )
+{
+   dst->u[0] = src0->u[0] ? src1->u[0] : src2->u[0];
+   dst->u[1] = src0->u[1] ? src1->u[1] : src2->u[1];
+   dst->u[2] = src0->u[2] ? src1->u[2] : src2->u[2];
+   dst->u[3] = src0->u[3] ? src1->u[3] : src2->u[3];
+}
+#endif
+
+static void
+micro_neg(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = -src->f[0];
+   dst->f[1] = -src->f[1];
+   dst->f[2] = -src->f[2];
+   dst->f[3] = -src->f[3];
+}
+
+#if 0
+static void
+micro_ineg(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->i[0] = -src->i[0];
+   dst->i[1] = -src->i[1];
+   dst->i[2] = -src->i[2];
+   dst->i[3] = -src->i[3];
+}
+#endif
+
+static void
+micro_not(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->u[0] = ~src->u[0];
+   dst->u[1] = ~src->u[1];
+   dst->u[2] = ~src->u[2];
+   dst->u[3] = ~src->u[3];
+}
+
+static void
+micro_or(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->u[0] = src0->u[0] | src1->u[0];
+   dst->u[1] = src0->u[1] | src1->u[1];
+   dst->u[2] = src0->u[2] | src1->u[2];
+   dst->u[3] = src0->u[3] | src1->u[3];
+}
+
+static void
+micro_pow(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+#if FAST_MATH
+   dst->f[0] = util_fast_pow( src0->f[0], src1->f[0] );
+   dst->f[1] = util_fast_pow( src0->f[1], src1->f[1] );
+   dst->f[2] = util_fast_pow( src0->f[2], src1->f[2] );
+   dst->f[3] = util_fast_pow( src0->f[3], src1->f[3] );
+#else
+   dst->f[0] = powf( src0->f[0], src1->f[0] );
+   dst->f[1] = powf( src0->f[1], src1->f[1] );
+   dst->f[2] = powf( src0->f[2], src1->f[2] );
+   dst->f[3] = powf( src0->f[3], src1->f[3] );
+#endif
+}
+
+static void
+micro_rnd(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = floorf( src->f[0] + 0.5f );
+   dst->f[1] = floorf( src->f[1] + 0.5f );
+   dst->f[2] = floorf( src->f[2] + 0.5f );
+   dst->f[3] = floorf( src->f[3] + 0.5f );
+}
+
+static void
+micro_sgn(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = src->f[0] < 0.0f ? -1.0f : src->f[0] > 0.0f ? 1.0f : 0.0f;
+   dst->f[1] = src->f[1] < 0.0f ? -1.0f : src->f[1] > 0.0f ? 1.0f : 0.0f;
+   dst->f[2] = src->f[2] < 0.0f ? -1.0f : src->f[2] > 0.0f ? 1.0f : 0.0f;
+   dst->f[3] = src->f[3] < 0.0f ? -1.0f : src->f[3] > 0.0f ? 1.0f : 0.0f;
+}
+
+static void
+micro_shl(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->i[0] = src0->i[0] << src1->i[0];
+   dst->i[1] = src0->i[1] << src1->i[1];
+   dst->i[2] = src0->i[2] << src1->i[2];
+   dst->i[3] = src0->i[3] << src1->i[3];
+}
+
+static void
+micro_ishr(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->i[0] = src0->i[0] >> src1->i[0];
+   dst->i[1] = src0->i[1] >> src1->i[1];
+   dst->i[2] = src0->i[2] >> src1->i[2];
+   dst->i[3] = src0->i[3] >> src1->i[3];
+}
+
+static void
+micro_trunc(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0 )
+{
+   dst->f[0] = (float) (int) src0->f[0];
+   dst->f[1] = (float) (int) src0->f[1];
+   dst->f[2] = (float) (int) src0->f[2];
+   dst->f[3] = (float) (int) src0->f[3];
+}
+
+#if 0
+static void
+micro_ushr(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->u[0] = src0->u[0] >> src1->u[0];
+   dst->u[1] = src0->u[1] >> src1->u[1];
+   dst->u[2] = src0->u[2] >> src1->u[2];
+   dst->u[3] = src0->u[3] >> src1->u[3];
+}
+#endif
+
+static void
+micro_sin(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = sinf( src->f[0] );
+   dst->f[1] = sinf( src->f[1] );
+   dst->f[2] = sinf( src->f[2] );
+   dst->f[3] = sinf( src->f[3] );
+}
+
+static void
+micro_sqrt( union tgsi_exec_channel *dst,
+            const union tgsi_exec_channel *src )
+{
+   dst->f[0] = sqrtf( src->f[0] );
+   dst->f[1] = sqrtf( src->f[1] );
+   dst->f[2] = sqrtf( src->f[2] );
+   dst->f[3] = sqrtf( src->f[3] );
+}
+
+static void
+micro_sub(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->f[0] = src0->f[0] - src1->f[0];
+   dst->f[1] = src0->f[1] - src1->f[1];
+   dst->f[2] = src0->f[2] - src1->f[2];
+   dst->f[3] = src0->f[3] - src1->f[3];
+}
+
+#if 0
+static void
+micro_u2f(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src )
+{
+   dst->f[0] = (float) src->u[0];
+   dst->f[1] = (float) src->u[1];
+   dst->f[2] = (float) src->u[2];
+   dst->f[3] = (float) src->u[3];
+}
+#endif
+
+static void
+micro_xor(
+   union tgsi_exec_channel *dst,
+   const union tgsi_exec_channel *src0,
+   const union tgsi_exec_channel *src1 )
+{
+   dst->u[0] = src0->u[0] ^ src1->u[0];
+   dst->u[1] = src0->u[1] ^ src1->u[1];
+   dst->u[2] = src0->u[2] ^ src1->u[2];
+   dst->u[3] = src0->u[3] ^ src1->u[3];
+}
+
+static void
+fetch_src_file_channel(
+   const struct tgsi_exec_machine *mach,
+   const uint file,
+   const uint swizzle,
+   const union tgsi_exec_channel *index,
+   union tgsi_exec_channel *chan )
+{
+   switch( swizzle ) {
+   case TGSI_EXTSWIZZLE_X:
+   case TGSI_EXTSWIZZLE_Y:
+   case TGSI_EXTSWIZZLE_Z:
+   case TGSI_EXTSWIZZLE_W:
+      switch( file ) {
+      case TGSI_FILE_CONSTANT:
+         assert(mach->Consts);
+         if (index->i[0] < 0)
+            chan->f[0] = 0.0f;
+         else
+            chan->f[0] = mach->Consts[index->i[0]][swizzle];
+         if (index->i[1] < 0)
+            chan->f[1] = 0.0f;
+         else
+            chan->f[1] = mach->Consts[index->i[1]][swizzle];
+         if (index->i[2] < 0)
+            chan->f[2] = 0.0f;
+         else
+            chan->f[2] = mach->Consts[index->i[2]][swizzle];
+         if (index->i[3] < 0)
+            chan->f[3] = 0.0f;
+         else
+            chan->f[3] = mach->Consts[index->i[3]][swizzle];
+         break;
+
+      case TGSI_FILE_INPUT:
+         chan->u[0] = mach->Inputs[index->i[0]].xyzw[swizzle].u[0];
+         chan->u[1] = mach->Inputs[index->i[1]].xyzw[swizzle].u[1];
+         chan->u[2] = mach->Inputs[index->i[2]].xyzw[swizzle].u[2];
+         chan->u[3] = mach->Inputs[index->i[3]].xyzw[swizzle].u[3];
+         break;
+
+      case TGSI_FILE_TEMPORARY:
+         assert(index->i[0] < TGSI_EXEC_NUM_TEMPS);
+         chan->u[0] = mach->Temps[index->i[0]].xyzw[swizzle].u[0];
+         chan->u[1] = mach->Temps[index->i[1]].xyzw[swizzle].u[1];
+         chan->u[2] = mach->Temps[index->i[2]].xyzw[swizzle].u[2];
+         chan->u[3] = mach->Temps[index->i[3]].xyzw[swizzle].u[3];
+         break;
+
+      case TGSI_FILE_IMMEDIATE:
+         assert( index->i[0] < (int) mach->ImmLimit );
+         chan->f[0] = mach->Imms[index->i[0]][swizzle];
+         assert( index->i[1] < (int) mach->ImmLimit );
+         chan->f[1] = mach->Imms[index->i[1]][swizzle];
+         assert( index->i[2] < (int) mach->ImmLimit );
+         chan->f[2] = mach->Imms[index->i[2]][swizzle];
+         assert( index->i[3] < (int) mach->ImmLimit );
+         chan->f[3] = mach->Imms[index->i[3]][swizzle];
+         break;
+
+      case TGSI_FILE_ADDRESS:
+         chan->u[0] = mach->Addrs[index->i[0]].xyzw[swizzle].u[0];
+         chan->u[1] = mach->Addrs[index->i[1]].xyzw[swizzle].u[1];
+         chan->u[2] = mach->Addrs[index->i[2]].xyzw[swizzle].u[2];
+         chan->u[3] = mach->Addrs[index->i[3]].xyzw[swizzle].u[3];
+         break;
+
+      case TGSI_FILE_OUTPUT:
+         /* vertex/fragment output vars can be read too */
+         chan->u[0] = mach->Outputs[index->i[0]].xyzw[swizzle].u[0];
+         chan->u[1] = mach->Outputs[index->i[1]].xyzw[swizzle].u[1];
+         chan->u[2] = mach->Outputs[index->i[2]].xyzw[swizzle].u[2];
+         chan->u[3] = mach->Outputs[index->i[3]].xyzw[swizzle].u[3];
+         break;
+
+      default:
+         assert( 0 );
+      }
+      break;
+
+   case TGSI_EXTSWIZZLE_ZERO:
+      *chan = mach->Temps[TEMP_0_I].xyzw[TEMP_0_C];
+      break;
+
+   case TGSI_EXTSWIZZLE_ONE:
+      *chan = mach->Temps[TEMP_1_I].xyzw[TEMP_1_C];
+      break;
+
+   default:
+      assert( 0 );
+   }
+}
+
+static void
+fetch_source(
+   const struct tgsi_exec_machine *mach,
+   union tgsi_exec_channel *chan,
+   const struct tgsi_full_src_register *reg,
+   const uint chan_index )
+{
+   union tgsi_exec_channel index;
+   uint swizzle;
+
+   /* We start with a direct index into a register file.
+    *
+    *    file[1],
+    *    where:
+    *       file = SrcRegister.File
+    *       [1] = SrcRegister.Index
+    */
+   index.i[0] =
+   index.i[1] =
+   index.i[2] =
+   index.i[3] = reg->SrcRegister.Index;
+
+   /* There is an extra source register that indirectly subscripts
+    * a register file. The direct index now becomes an offset
+    * that is being added to the indirect register.
+    *
+    *    file[ind[2].x+1],
+    *    where:
+    *       ind = SrcRegisterInd.File
+    *       [2] = SrcRegisterInd.Index
+    *       .x = SrcRegisterInd.SwizzleX
+    */
+   if (reg->SrcRegister.Indirect) {
+      union tgsi_exec_channel index2;
+      union tgsi_exec_channel indir_index;
+      const uint execmask = mach->ExecMask;
+      uint i;
+
+      /* which address register (always zero now) */
+      index2.i[0] =
+      index2.i[1] =
+      index2.i[2] =
+      index2.i[3] = reg->SrcRegisterInd.Index;
+
+      /* get current value of address register[swizzle] */
+      swizzle = tgsi_util_get_src_register_swizzle( &reg->SrcRegisterInd, CHAN_X );
+      fetch_src_file_channel(
+         mach,
+         reg->SrcRegisterInd.File,
+         swizzle,
+         &index2,
+         &indir_index );
+
+      /* add value of address register to the offset */
+      index.i[0] += (int) indir_index.f[0];
+      index.i[1] += (int) indir_index.f[1];
+      index.i[2] += (int) indir_index.f[2];
+      index.i[3] += (int) indir_index.f[3];
+
+      /* for disabled execution channels, zero-out the index to
+       * avoid using a potential garbage value.
+       */
+      for (i = 0; i < QUAD_SIZE; i++) {
+         if ((execmask & (1 << i)) == 0)
+            index.i[i] = 0;
+      }
+   }
+
+   /* There is an extra source register that is a second
+    * subscript to a register file. Effectively it means that
+    * the register file is actually a 2D array of registers.
+    *
+    *    file[1][3] == file[1*sizeof(file[1])+3],
+    *    where:
+    *       [3] = SrcRegisterDim.Index
+    */
+   if (reg->SrcRegister.Dimension) {
+      /* The size of the first-order array depends on the register file type.
+       * We need to multiply the index to the first array to get an effective,
+       * "flat" index that points to the beginning of the second-order array.
+       */
+      switch (reg->SrcRegister.File) {
+      case TGSI_FILE_INPUT:
+         index.i[0] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
+         index.i[1] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
+         index.i[2] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
+         index.i[3] *= TGSI_EXEC_MAX_INPUT_ATTRIBS;
+         break;
+      case TGSI_FILE_CONSTANT:
+         index.i[0] *= TGSI_EXEC_MAX_CONST_BUFFER;
+         index.i[1] *= TGSI_EXEC_MAX_CONST_BUFFER;
+         index.i[2] *= TGSI_EXEC_MAX_CONST_BUFFER;
+         index.i[3] *= TGSI_EXEC_MAX_CONST_BUFFER;
+         break;
+      default:
+         assert( 0 );
+      }
+
+      index.i[0] += reg->SrcRegisterDim.Index;
+      index.i[1] += reg->SrcRegisterDim.Index;
+      index.i[2] += reg->SrcRegisterDim.Index;
+      index.i[3] += reg->SrcRegisterDim.Index;
+
+      /* Again, the second subscript index can be addressed indirectly
+       * identically to the first one.
+       * Nothing stops us from indirectly addressing the indirect register,
+       * but there is no need for that, so we won't exercise it.
+       *
+       *    file[1][ind[4].y+3],
+       *    where:
+       *       ind = SrcRegisterDimInd.File
+       *       [4] = SrcRegisterDimInd.Index
+       *       .y = SrcRegisterDimInd.SwizzleX
+       */
+      if (reg->SrcRegisterDim.Indirect) {
+         union tgsi_exec_channel index2;
+         union tgsi_exec_channel indir_index;
+         const uint execmask = mach->ExecMask;
+         uint i;
+
+         index2.i[0] =
+         index2.i[1] =
+         index2.i[2] =
+         index2.i[3] = reg->SrcRegisterDimInd.Index;
+
+         swizzle = tgsi_util_get_src_register_swizzle( &reg->SrcRegisterDimInd, CHAN_X );
+         fetch_src_file_channel(
+            mach,
+            reg->SrcRegisterDimInd.File,
+            swizzle,
+            &index2,
+            &indir_index );
+
+         index.i[0] += (int) indir_index.f[0];
+         index.i[1] += (int) indir_index.f[1];
+         index.i[2] += (int) indir_index.f[2];
+         index.i[3] += (int) indir_index.f[3];
+
+         /* for disabled execution channels, zero-out the index to
+          * avoid using a potential garbage value.
+          */
+         for (i = 0; i < QUAD_SIZE; i++) {
+            if ((execmask & (1 << i)) == 0)
+               index.i[i] = 0;
+         }
+      }
+
+      /* If by any chance there was a need for a 3D array of register
+       * files, we would have to check whether SrcRegisterDim is followed
+       * by a dimension register and continue the saga.
+       */
+   }
+
+   swizzle = tgsi_util_get_full_src_register_extswizzle( reg, chan_index );
+   fetch_src_file_channel(
+      mach,
+      reg->SrcRegister.File,
+      swizzle,
+      &index,
+      chan );
+
+   switch (tgsi_util_get_full_src_register_sign_mode( reg, chan_index )) {
+   case TGSI_UTIL_SIGN_CLEAR:
+      micro_abs( chan, chan );
+      break;
+
+   case TGSI_UTIL_SIGN_SET:
+      micro_abs( chan, chan );
+      micro_neg( chan, chan );
+      break;
+
+   case TGSI_UTIL_SIGN_TOGGLE:
+      micro_neg( chan, chan );
+      break;
+
+   case TGSI_UTIL_SIGN_KEEP:
+      break;
+   }
+
+   if (reg->SrcRegisterExtMod.Complement) {
+      micro_sub( chan, &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], chan );
+   }
+}
+
+static void
+store_dest(
+   struct tgsi_exec_machine *mach,
+   const union tgsi_exec_channel *chan,
+   const struct tgsi_full_dst_register *reg,
+   const struct tgsi_full_instruction *inst,
+   uint chan_index )
+{
+   uint i;
+   union tgsi_exec_channel null;
+   union tgsi_exec_channel *dst;
+   uint execmask = mach->ExecMask;
+
+   switch (reg->DstRegister.File) {
+   case TGSI_FILE_NULL:
+      dst = &null;
+      break;
+
+   case TGSI_FILE_OUTPUT:
+      dst = &mach->Outputs[mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0]
+                           + reg->DstRegister.Index].xyzw[chan_index];
+      break;
+
+   case TGSI_FILE_TEMPORARY:
+      assert( reg->DstRegister.Index < TGSI_EXEC_NUM_TEMPS );
+      dst = &mach->Temps[reg->DstRegister.Index].xyzw[chan_index];
+      break;
+
+   case TGSI_FILE_ADDRESS:
+      dst = &mach->Addrs[reg->DstRegister.Index].xyzw[chan_index];
+      break;
+
+   default:
+      assert( 0 );
+      return;
+   }
+
+   if (inst->InstructionExtNv.CondFlowEnable) {
+      union tgsi_exec_channel *cc = &mach->Temps[TEMP_CC_I].xyzw[TEMP_CC_C];
+      uint swizzle;
+      uint shift;
+      uint mask;
+      uint test;
+
+      /* Only CC0 supported.
+       */
+      assert( inst->InstructionExtNv.CondFlowIndex < 1 );
+
+      switch (chan_index) {
+      case CHAN_X:
+         swizzle = inst->InstructionExtNv.CondSwizzleX;
+         break;
+      case CHAN_Y:
+         swizzle = inst->InstructionExtNv.CondSwizzleY;
+         break;
+      case CHAN_Z:
+         swizzle = inst->InstructionExtNv.CondSwizzleZ;
+         break;
+      case CHAN_W:
+         swizzle = inst->InstructionExtNv.CondSwizzleW;
+         break;
+      default:
+         assert( 0 );
+         return;
+      }
+
+      switch (swizzle) {
+      case TGSI_SWIZZLE_X:
+         shift = TGSI_EXEC_CC_X_SHIFT;
+         mask = TGSI_EXEC_CC_X_MASK;
+         break;
+      case TGSI_SWIZZLE_Y:
+         shift = TGSI_EXEC_CC_Y_SHIFT;
+         mask = TGSI_EXEC_CC_Y_MASK;
+         break;
+      case TGSI_SWIZZLE_Z:
+         shift = TGSI_EXEC_CC_Z_SHIFT;
+         mask = TGSI_EXEC_CC_Z_MASK;
+         break;
+      case TGSI_SWIZZLE_W:
+         shift = TGSI_EXEC_CC_W_SHIFT;
+         mask = TGSI_EXEC_CC_W_MASK;
+         break;
+      default:
+         assert( 0 );
+         return;
+      }
+
+      switch (inst->InstructionExtNv.CondMask) {
+      case TGSI_CC_GT:
+         test = ~(TGSI_EXEC_CC_GT << shift) & mask;
+         for (i = 0; i < QUAD_SIZE; i++)
+            if (cc->u[i] & test)
+               execmask &= ~(1 << i);
+         break;
+
+      case TGSI_CC_EQ:
+         test = ~(TGSI_EXEC_CC_EQ << shift) & mask;
+         for (i = 0; i < QUAD_SIZE; i++)
+            if (cc->u[i] & test)
+               execmask &= ~(1 << i);
+         break;
+
+      case TGSI_CC_LT:
+         test = ~(TGSI_EXEC_CC_LT << shift) & mask;
+         for (i = 0; i < QUAD_SIZE; i++)
+            if (cc->u[i] & test)
+               execmask &= ~(1 << i);
+         break;
+
+      case TGSI_CC_GE:
+         test = ~((TGSI_EXEC_CC_GT | TGSI_EXEC_CC_EQ) << shift) & mask;
+         for (i = 0; i < QUAD_SIZE; i++)
+            if (cc->u[i] & test)
+               execmask &= ~(1 << i);
+         break;
+
+      case TGSI_CC_LE:
+         test = ~((TGSI_EXEC_CC_LT | TGSI_EXEC_CC_EQ) << shift) & mask;
+         for (i = 0; i < QUAD_SIZE; i++)
+            if (cc->u[i] & test)
+               execmask &= ~(1 << i);
+         break;
+
+      case TGSI_CC_NE:
+         test = ~((TGSI_EXEC_CC_GT | TGSI_EXEC_CC_LT | TGSI_EXEC_CC_UN) << shift) & mask;
+         for (i = 0; i < QUAD_SIZE; i++)
+            if (cc->u[i] & test)
+               execmask &= ~(1 << i);
+         break;
+
+      case TGSI_CC_TR:
+         break;
+
+      case TGSI_CC_FL:
+         for (i = 0; i < QUAD_SIZE; i++)
+            execmask &= ~(1 << i);
+         break;
+
+      default:
+         assert( 0 );
+         return;
+      }
+   }
+
+   switch (inst->Instruction.Saturate) {
+   case TGSI_SAT_NONE:
+      for (i = 0; i < QUAD_SIZE; i++)
+         if (execmask & (1 << i))
+            dst->i[i] = chan->i[i];
+      break;
+
+   case TGSI_SAT_ZERO_ONE:
+      for (i = 0; i < QUAD_SIZE; i++)
+         if (execmask & (1 << i)) {
+            if (chan->f[i] < 0.0f)
+               dst->f[i] = 0.0f;
+            else if (chan->f[i] > 1.0f)
+               dst->f[i] = 1.0f;
+            else
+               dst->i[i] = chan->i[i];
+         }
+      break;
+
+   case TGSI_SAT_MINUS_PLUS_ONE:
+      for (i = 0; i < QUAD_SIZE; i++)
+         if (execmask & (1 << i)) {
+            if (chan->f[i] < -1.0f)
+               dst->f[i] = -1.0f;
+            else if (chan->f[i] > 1.0f)
+               dst->f[i] = 1.0f;
+            else
+               dst->i[i] = chan->i[i];
+         }
+      break;
+
+   default:
+      assert( 0 );
+   }
+
+   if (inst->InstructionExtNv.CondDstUpdate) {
+      union tgsi_exec_channel *cc = &mach->Temps[TEMP_CC_I].xyzw[TEMP_CC_C];
+      uint shift;
+      uint mask;
+
+      /* Only CC0 supported.
+       */
+      assert( inst->InstructionExtNv.CondDstIndex < 1 );
+
+      switch (chan_index) {
+      case CHAN_X:
+         shift = TGSI_EXEC_CC_X_SHIFT;
+         mask = ~TGSI_EXEC_CC_X_MASK;
+         break;
+      case CHAN_Y:
+         shift = TGSI_EXEC_CC_Y_SHIFT;
+         mask = ~TGSI_EXEC_CC_Y_MASK;
+         break;
+      case CHAN_Z:
+         shift = TGSI_EXEC_CC_Z_SHIFT;
+         mask = ~TGSI_EXEC_CC_Z_MASK;
+         break;
+      case CHAN_W:
+         shift = TGSI_EXEC_CC_W_SHIFT;
+         mask = ~TGSI_EXEC_CC_W_MASK;
+         break;
+      default:
+         assert( 0 );
+         return;
+      }
+
+      for (i = 0; i < QUAD_SIZE; i++)
+         if (execmask & (1 << i)) {
+            cc->u[i] &= mask;
+            if (dst->f[i] < 0.0f)
+               cc->u[i] |= TGSI_EXEC_CC_LT << shift;
+            else if (dst->f[i] > 0.0f)
+               cc->u[i] |= TGSI_EXEC_CC_GT << shift;
+            else if (dst->f[i] == 0.0f)
+               cc->u[i] |= TGSI_EXEC_CC_EQ << shift;
+            else
+               cc->u[i] |= TGSI_EXEC_CC_UN << shift;
+         }
+   }
+}
+
+#define FETCH(VAL,INDEX,CHAN)\
+    fetch_source (mach, VAL, &inst->FullSrcRegisters[INDEX], CHAN)
+
+#define STORE(VAL,INDEX,CHAN)\
+    store_dest (mach, VAL, &inst->FullDstRegisters[INDEX], inst, CHAN )
+
+
+/**
+ * Execute ARB-style KIL which is predicated by a src register.
+ * Kill fragment if any of the four values is less than zero.
+ */
+static void
+exec_kil(struct tgsi_exec_machine *mach,
+         const struct tgsi_full_instruction *inst)
+{
+   uint uniquemask;
+   uint chan_index;
+   uint kilmask = 0; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
+   union tgsi_exec_channel r[1];
+
+   /* This mask stores component bits that were already tested. Note that
+    * we test if the value is less than zero, so 1.0 and 0.0 need not to be
+    * tested. */
+   uniquemask = (1 << TGSI_EXTSWIZZLE_ZERO) | (1 << TGSI_EXTSWIZZLE_ONE);
+
+   for (chan_index = 0; chan_index < 4; chan_index++)
+   {
+      uint swizzle;
+      uint i;
+
+      /* unswizzle channel */
+      swizzle = tgsi_util_get_full_src_register_extswizzle (
+                        &inst->FullSrcRegisters[0],
+                        chan_index);
+
+      /* check if the component has not been already tested */
+      if (uniquemask & (1 << swizzle))
+         continue;
+      uniquemask |= 1 << swizzle;
+
+      FETCH(&r[0], 0, chan_index);
+      for (i = 0; i < 4; i++)
+         if (r[0].f[i] < 0.0f)
+            kilmask |= 1 << i;
+   }
+
+   mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask;
+}
+
+/**
+ * Execute NVIDIA-style KIL which is predicated by a condition code.
+ * Kill fragment if the condition code is TRUE.
+ */
+static void
+exec_kilp(struct tgsi_exec_machine *mach,
+          const struct tgsi_full_instruction *inst)
+{
+   uint kilmask; /* bit 0 = pixel 0, bit 1 = pixel 1, etc */
+
+   if (inst->InstructionExtNv.CondFlowEnable) {
+      uint swizzle[4];
+      uint chan_index;
+
+      kilmask = 0x0;
+
+      swizzle[0] = inst->InstructionExtNv.CondSwizzleX;
+      swizzle[1] = inst->InstructionExtNv.CondSwizzleY;
+      swizzle[2] = inst->InstructionExtNv.CondSwizzleZ;
+      swizzle[3] = inst->InstructionExtNv.CondSwizzleW;
+
+      for (chan_index = 0; chan_index < 4; chan_index++)
+      {
+         uint i;
+
+         for (i = 0; i < 4; i++) {
+            /* TODO: evaluate the condition code */
+            if (0)
+               kilmask |= 1 << i;
+         }
+      }
+   }
+   else {
+      /* "unconditional" kil */
+      kilmask = mach->ExecMask;
+   }
+   mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] |= kilmask;
+}
+
+
+/*
+ * Fetch a four texture samples using STR texture coordinates.
+ */
+static void
+fetch_texel( struct tgsi_sampler *sampler,
+             const union tgsi_exec_channel *s,
+             const union tgsi_exec_channel *t,
+             const union tgsi_exec_channel *p,
+             float lodbias,  /* XXX should be float[4] */
+             union tgsi_exec_channel *r,
+             union tgsi_exec_channel *g,
+             union tgsi_exec_channel *b,
+             union tgsi_exec_channel *a )
+{
+   uint j;
+   float rgba[NUM_CHANNELS][QUAD_SIZE];
+
+   sampler->get_samples(sampler, s->f, t->f, p->f, lodbias, rgba);
+
+   for (j = 0; j < 4; j++) {
+      r->f[j] = rgba[0][j];
+      g->f[j] = rgba[1][j];
+      b->f[j] = rgba[2][j];
+      a->f[j] = rgba[3][j];
+   }
+}
+
+
+static void
+exec_tex(struct tgsi_exec_machine *mach,
+         const struct tgsi_full_instruction *inst,
+         boolean biasLod,
+         boolean projected)
+{
+   const uint unit = inst->FullSrcRegisters[1].SrcRegister.Index;
+   union tgsi_exec_channel r[4];
+   uint chan_index;
+   float lodBias;
+
+   /*   debug_printf("Sampler %u unit %u\n", sampler, unit); */
+
+   switch (inst->InstructionExtTexture.Texture) {
+   case TGSI_TEXTURE_1D:
+
+      FETCH(&r[0], 0, CHAN_X);
+
+      if (projected) {
+         FETCH(&r[1], 0, CHAN_W);
+         micro_div( &r[0], &r[0], &r[1] );
+      }
+
+      if (biasLod) {
+         FETCH(&r[1], 0, CHAN_W);
+         lodBias = r[2].f[0];
+      }
+      else
+         lodBias = 0.0;
+
+      fetch_texel(mach->Samplers[unit],
+                  &r[0], NULL, NULL, lodBias,  /* S, T, P, BIAS */
+                  &r[0], &r[1], &r[2], &r[3]); /* R, G, B, A */
+      break;
+
+   case TGSI_TEXTURE_2D:
+   case TGSI_TEXTURE_RECT:
+
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 0, CHAN_Y);
+      FETCH(&r[2], 0, CHAN_Z);
+
+      if (projected) {
+         FETCH(&r[3], 0, CHAN_W);
+         micro_div( &r[0], &r[0], &r[3] );
+         micro_div( &r[1], &r[1], &r[3] );
+         micro_div( &r[2], &r[2], &r[3] );
+      }
+
+      if (biasLod) {
+         FETCH(&r[3], 0, CHAN_W);
+         lodBias = r[3].f[0];
+      }
+      else
+         lodBias = 0.0;
+
+      fetch_texel(mach->Samplers[unit],
+                  &r[0], &r[1], &r[2], lodBias,  /* inputs */
+                  &r[0], &r[1], &r[2], &r[3]);  /* outputs */
+      break;
+
+   case TGSI_TEXTURE_3D:
+   case TGSI_TEXTURE_CUBE:
+
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 0, CHAN_Y);
+      FETCH(&r[2], 0, CHAN_Z);
+
+      if (projected) {
+         FETCH(&r[3], 0, CHAN_W);
+         micro_div( &r[0], &r[0], &r[3] );
+         micro_div( &r[1], &r[1], &r[3] );
+         micro_div( &r[2], &r[2], &r[3] );
+      }
+
+      if (biasLod) {
+         FETCH(&r[3], 0, CHAN_W);
+         lodBias = r[3].f[0];
+      }
+      else
+         lodBias = 0.0;
+
+      fetch_texel(mach->Samplers[unit],
+                  &r[0], &r[1], &r[2], lodBias,
+                  &r[0], &r[1], &r[2], &r[3]);
+      break;
+
+   default:
+      assert (0);
+   }
+
+   FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+      STORE( &r[chan_index], 0, chan_index );
+   }
+}
+
+
+/**
+ * Evaluate a constant-valued coefficient at the position of the
+ * current quad.
+ */
+static void
+eval_constant_coef(
+   struct tgsi_exec_machine *mach,
+   unsigned attrib,
+   unsigned chan )
+{
+   unsigned i;
+
+   for( i = 0; i < QUAD_SIZE; i++ ) {
+      mach->Inputs[attrib].xyzw[chan].f[i] = mach->InterpCoefs[attrib].a0[chan];
+   }
+}
+
+/**
+ * Evaluate a linear-valued coefficient at the position of the
+ * current quad.
+ */
+static void
+eval_linear_coef(
+   struct tgsi_exec_machine *mach,
+   unsigned attrib,
+   unsigned chan )
+{
+   const float x = mach->QuadPos.xyzw[0].f[0];
+   const float y = mach->QuadPos.xyzw[1].f[0];
+   const float dadx = mach->InterpCoefs[attrib].dadx[chan];
+   const float dady = mach->InterpCoefs[attrib].dady[chan];
+   const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y;
+   mach->Inputs[attrib].xyzw[chan].f[0] = a0;
+   mach->Inputs[attrib].xyzw[chan].f[1] = a0 + dadx;
+   mach->Inputs[attrib].xyzw[chan].f[2] = a0 + dady;
+   mach->Inputs[attrib].xyzw[chan].f[3] = a0 + dadx + dady;
+}
+
+/**
+ * Evaluate a perspective-valued coefficient at the position of the
+ * current quad.
+ */
+static void
+eval_perspective_coef(
+   struct tgsi_exec_machine *mach,
+   unsigned attrib,
+   unsigned chan )
+{
+   const float x = mach->QuadPos.xyzw[0].f[0];
+   const float y = mach->QuadPos.xyzw[1].f[0];
+   const float dadx = mach->InterpCoefs[attrib].dadx[chan];
+   const float dady = mach->InterpCoefs[attrib].dady[chan];
+   const float a0 = mach->InterpCoefs[attrib].a0[chan] + dadx * x + dady * y;
+   const float *w = mach->QuadPos.xyzw[3].f;
+   /* divide by W here */
+   mach->Inputs[attrib].xyzw[chan].f[0] = a0 / w[0];
+   mach->Inputs[attrib].xyzw[chan].f[1] = (a0 + dadx) / w[1];
+   mach->Inputs[attrib].xyzw[chan].f[2] = (a0 + dady) / w[2];
+   mach->Inputs[attrib].xyzw[chan].f[3] = (a0 + dadx + dady) / w[3];
+}
+
+
+typedef void (* eval_coef_func)(
+   struct tgsi_exec_machine *mach,
+   unsigned attrib,
+   unsigned chan );
+
+static void
+exec_declaration(
+   struct tgsi_exec_machine *mach,
+   const struct tgsi_full_declaration *decl )
+{
+   if( mach->Processor == TGSI_PROCESSOR_FRAGMENT ) {
+      if( decl->Declaration.File == TGSI_FILE_INPUT ) {
+         unsigned first, last, mask;
+         eval_coef_func eval;
+
+         first = decl->DeclarationRange.First;
+         last = decl->DeclarationRange.Last;
+         mask = decl->Declaration.UsageMask;
+
+         switch( decl->Declaration.Interpolate ) {
+         case TGSI_INTERPOLATE_CONSTANT:
+            eval = eval_constant_coef;
+            break;
+
+         case TGSI_INTERPOLATE_LINEAR:
+            eval = eval_linear_coef;
+            break;
+
+         case TGSI_INTERPOLATE_PERSPECTIVE:
+            eval = eval_perspective_coef;
+            break;
+
+         default:
+            eval = NULL;
+            assert( 0 );
+         }
+
+         if( mask == TGSI_WRITEMASK_XYZW ) {
+            unsigned i, j;
+
+            for( i = first; i <= last; i++ ) {
+               for( j = 0; j < NUM_CHANNELS; j++ ) {
+                  eval( mach, i, j );
+               }
+            }
+         }
+         else {
+            unsigned i, j;
+
+            for( j = 0; j < NUM_CHANNELS; j++ ) {
+               if( mask & (1 << j) ) {
+                  for( i = first; i <= last; i++ ) {
+                     eval( mach, i, j );
+                  }
+               }
+            }
+         }
+      }
+   }
+}
+
+static void
+exec_instruction(
+   struct tgsi_exec_machine *mach,
+   const struct tgsi_full_instruction *inst,
+   int *pc )
+{
+   uint chan_index;
+   union tgsi_exec_channel r[8];
+
+   (*pc)++;
+
+   switch (inst->Instruction.Opcode) {
+   case TGSI_OPCODE_ARL:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_flr( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_MOV:
+   case TGSI_OPCODE_SWZ:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_LIT:
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) {
+	 STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_X );
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Y ) || IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+	 FETCH( &r[0], 0, CHAN_X );
+	 if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
+	    micro_max( &r[0], &r[0], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] );
+	    STORE( &r[0], 0, CHAN_Y );
+	 }
+
+	 if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+	    FETCH( &r[1], 0, CHAN_Y );
+	    micro_max( &r[1], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] );
+
+	    FETCH( &r[2], 0, CHAN_W );
+	    micro_min( &r[2], &r[2], &mach->Temps[TEMP_128_I].xyzw[TEMP_128_C] );
+	    micro_max( &r[2], &r[2], &mach->Temps[TEMP_M128_I].xyzw[TEMP_M128_C] );
+	    micro_pow( &r[1], &r[1], &r[2] );
+	    micro_lt( &r[0], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &r[0], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] );
+	    STORE( &r[0], 0, CHAN_Z );
+	 }
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
+	 STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_RCP:
+   /* TGSI_OPCODE_RECIP */
+      FETCH( &r[0], 0, CHAN_X );
+      micro_div( &r[0], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &r[0] );
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_RSQ:
+   /* TGSI_OPCODE_RECIPSQRT */
+      FETCH( &r[0], 0, CHAN_X );
+      micro_sqrt( &r[0], &r[0] );
+      micro_div( &r[0], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &r[0] );
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_EXP:
+      FETCH( &r[0], 0, CHAN_X );
+      micro_flr( &r[1], &r[0] );  /* r1 = floor(r0) */
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) {
+         micro_exp2( &r[2], &r[1] );       /* r2 = 2 ^ r1 */
+         STORE( &r[2], 0, CHAN_X );        /* store r2 */
+      }
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
+         micro_sub( &r[2], &r[0], &r[1] ); /* r2 = r0 - r1 */
+         STORE( &r[2], 0, CHAN_Y );        /* store r2 */
+      }
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+         micro_exp2( &r[2], &r[0] );       /* r2 = 2 ^ r0 */
+         STORE( &r[2], 0, CHAN_Z );        /* store r2 */
+      }
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
+         STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_LOG:
+      FETCH( &r[0], 0, CHAN_X );
+      micro_abs( &r[2], &r[0] );  /* r2 = abs(r0) */
+      micro_lg2( &r[1], &r[2] );  /* r1 = lg2(r2) */
+      micro_flr( &r[0], &r[1] );  /* r0 = floor(r1) */
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) {
+         STORE( &r[0], 0, CHAN_X );
+      }
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
+         micro_exp2( &r[0], &r[0] );       /* r0 = 2 ^ r0 */
+         micro_div( &r[0], &r[2], &r[0] ); /* r0 = r2 / r0 */
+         STORE( &r[0], 0, CHAN_Y );
+      }
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+         STORE( &r[1], 0, CHAN_Z );
+      }
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
+         STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_MUL:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index )
+      {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+
+         micro_mul( &r[0], &r[0], &r[1] );
+
+         STORE(&r[0], 0, chan_index);
+      }
+      break;
+
+   case TGSI_OPCODE_ADD:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_add( &r[0], &r[0], &r[1] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DP3:
+   /* TGSI_OPCODE_DOT3 */
+      FETCH( &r[0], 0, CHAN_X );
+      FETCH( &r[1], 1, CHAN_X );
+      micro_mul( &r[0], &r[0], &r[1] );
+
+      FETCH( &r[1], 0, CHAN_Y );
+      FETCH( &r[2], 1, CHAN_Y );
+      micro_mul( &r[1], &r[1], &r[2] );
+      micro_add( &r[0], &r[0], &r[1] );
+
+      FETCH( &r[1], 0, CHAN_Z );
+      FETCH( &r[2], 1, CHAN_Z );
+      micro_mul( &r[1], &r[1], &r[2] );
+      micro_add( &r[0], &r[0], &r[1] );
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+    case TGSI_OPCODE_DP4:
+    /* TGSI_OPCODE_DOT4 */
+       FETCH(&r[0], 0, CHAN_X);
+       FETCH(&r[1], 1, CHAN_X);
+
+       micro_mul( &r[0], &r[0], &r[1] );
+
+       FETCH(&r[1], 0, CHAN_Y);
+       FETCH(&r[2], 1, CHAN_Y);
+
+       micro_mul( &r[1], &r[1], &r[2] );
+       micro_add( &r[0], &r[0], &r[1] );
+
+       FETCH(&r[1], 0, CHAN_Z);
+       FETCH(&r[2], 1, CHAN_Z);
+
+       micro_mul( &r[1], &r[1], &r[2] );
+       micro_add( &r[0], &r[0], &r[1] );
+
+       FETCH(&r[1], 0, CHAN_W);
+       FETCH(&r[2], 1, CHAN_W);
+
+       micro_mul( &r[1], &r[1], &r[2] );
+       micro_add( &r[0], &r[0], &r[1] );
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DST:
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) {
+	 STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_X );
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
+	 FETCH( &r[0], 0, CHAN_Y );
+	 FETCH( &r[1], 1, CHAN_Y);
+	 micro_mul( &r[0], &r[0], &r[1] );
+	 STORE( &r[0], 0, CHAN_Y );
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+	 FETCH( &r[0], 0, CHAN_Z );
+	 STORE( &r[0], 0, CHAN_Z );
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
+	 FETCH( &r[0], 1, CHAN_W );
+	 STORE( &r[0], 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_MIN:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+
+         /* XXX use micro_min()?? */
+         micro_lt( &r[0], &r[0], &r[1], &r[0], &r[1] );
+
+         STORE(&r[0], 0, chan_index);
+      }
+      break;
+
+   case TGSI_OPCODE_MAX:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+
+         /* XXX use micro_max()?? */
+         micro_lt( &r[0], &r[0], &r[1], &r[1], &r[0] );
+
+         STORE(&r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SLT:
+   /* TGSI_OPCODE_SETLT */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_lt( &r[0], &r[0], &r[1], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SGE:
+   /* TGSI_OPCODE_SETGE */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_ge( &r[0], &r[0], &r[1], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_MAD:
+   /* TGSI_OPCODE_MADD */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_mul( &r[0], &r[0], &r[1] );
+         FETCH( &r[1], 2, chan_index );
+         micro_add( &r[0], &r[0], &r[1] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SUB:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+
+         micro_sub( &r[0], &r[0], &r[1] );
+
+         STORE(&r[0], 0, chan_index);
+      }
+      break;
+
+   case TGSI_OPCODE_LERP:
+   /* TGSI_OPCODE_LRP */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+         FETCH(&r[2], 2, chan_index);
+
+         micro_sub( &r[1], &r[1], &r[2] );
+         micro_mul( &r[0], &r[0], &r[1] );
+         micro_add( &r[0], &r[0], &r[2] );
+
+         STORE(&r[0], 0, chan_index);
+      }
+      break;
+
+   case TGSI_OPCODE_CND:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_CND0:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_DOT2ADD:
+      /* TGSI_OPCODE_DP2A */
+      FETCH( &r[0], 0, CHAN_X );
+      FETCH( &r[1], 1, CHAN_X );
+      micro_mul( &r[0], &r[0], &r[1] );
+
+      FETCH( &r[1], 0, CHAN_Y );
+      FETCH( &r[2], 1, CHAN_Y );
+      micro_mul( &r[1], &r[1], &r[2] );
+      micro_add( &r[0], &r[0], &r[1] );
+
+      FETCH( &r[2], 2, CHAN_X );
+      micro_add( &r[0], &r[0], &r[2] );
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_INDEX:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_NEGATE:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_FRAC:
+   /* TGSI_OPCODE_FRC */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_frc( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_CLAMP:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_FLOOR:
+   /* TGSI_OPCODE_FLR */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_flr( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_ROUND:
+   case TGSI_OPCODE_ARR:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_rnd( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_EXPBASE2:
+    /* TGSI_OPCODE_EX2 */
+      FETCH(&r[0], 0, CHAN_X);
+
+#if FAST_MATH
+      micro_exp2( &r[0], &r[0] );
+#else
+      micro_pow( &r[0], &mach->Temps[TEMP_2_I].xyzw[TEMP_2_C], &r[0] );
+#endif
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_LOGBASE2:
+   /* TGSI_OPCODE_LG2 */
+      FETCH( &r[0], 0, CHAN_X );
+      micro_lg2( &r[0], &r[0] );
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_POWER:
+      /* TGSI_OPCODE_POW */
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 1, CHAN_X);
+
+      micro_pow( &r[0], &r[0], &r[1] );
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_CROSSPRODUCT:
+      /* TGSI_OPCODE_XPD */
+      FETCH(&r[0], 0, CHAN_Y);
+      FETCH(&r[1], 1, CHAN_Z);
+
+      micro_mul( &r[2], &r[0], &r[1] );
+
+      FETCH(&r[3], 0, CHAN_Z);
+      FETCH(&r[4], 1, CHAN_Y);
+
+      micro_mul( &r[5], &r[3], &r[4] );
+      micro_sub( &r[2], &r[2], &r[5] );
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_X )) {
+         STORE( &r[2], 0, CHAN_X );
+      }
+
+      FETCH(&r[2], 1, CHAN_X);
+
+      micro_mul( &r[3], &r[3], &r[2] );
+
+      FETCH(&r[5], 0, CHAN_X);
+
+      micro_mul( &r[1], &r[1], &r[5] );
+      micro_sub( &r[3], &r[3], &r[1] );
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Y )) {
+         STORE( &r[3], 0, CHAN_Y );
+      }
+
+      micro_mul( &r[5], &r[5], &r[4] );
+      micro_mul( &r[0], &r[0], &r[2] );
+      micro_sub( &r[5], &r[5], &r[0] );
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+         STORE( &r[5], 0, CHAN_Z );
+      }
+
+      if (IS_CHANNEL_ENABLED( *inst, CHAN_W )) {
+         STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
+      }
+      break;
+
+    case TGSI_OPCODE_MULTIPLYMATRIX:
+       assert (0);
+       break;
+
+    case TGSI_OPCODE_ABS:
+       FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+          FETCH(&r[0], 0, chan_index);
+
+          micro_abs( &r[0], &r[0] );
+
+          STORE(&r[0], 0, chan_index);
+       }
+       break;
+
+   case TGSI_OPCODE_RCC:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_DPH:
+      FETCH(&r[0], 0, CHAN_X);
+      FETCH(&r[1], 1, CHAN_X);
+
+      micro_mul( &r[0], &r[0], &r[1] );
+
+      FETCH(&r[1], 0, CHAN_Y);
+      FETCH(&r[2], 1, CHAN_Y);
+
+      micro_mul( &r[1], &r[1], &r[2] );
+      micro_add( &r[0], &r[0], &r[1] );
+
+      FETCH(&r[1], 0, CHAN_Z);
+      FETCH(&r[2], 1, CHAN_Z);
+
+      micro_mul( &r[1], &r[1], &r[2] );
+      micro_add( &r[0], &r[0], &r[1] );
+
+      FETCH(&r[1], 1, CHAN_W);
+
+      micro_add( &r[0], &r[0], &r[1] );
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_COS:
+      FETCH(&r[0], 0, CHAN_X);
+
+      micro_cos( &r[0], &r[0] );
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+	 STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DDX:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_ddx( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DDY:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_ddy( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_KILP:
+      exec_kilp (mach, inst);
+      break;
+
+   case TGSI_OPCODE_KIL:
+      exec_kil (mach, inst);
+      break;
+
+   case TGSI_OPCODE_PK2H:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_PK2US:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_PK4B:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_PK4UB:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_RFL:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_SEQ:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_eq( &r[0], &r[0], &r[1],
+                   &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C],
+                   &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SFL:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_SGT:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_le( &r[0], &r[0], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SIN:
+      FETCH( &r[0], 0, CHAN_X );
+      micro_sin( &r[0], &r[0] );
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SLE:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_le( &r[0], &r[0], &r[1], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SNE:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_eq( &r[0], &r[0], &r[1], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_STR:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_TEX:
+      /* simple texture lookup */
+      /* src[0] = texcoord */
+      /* src[1] = sampler unit */
+      exec_tex(mach, inst, FALSE, FALSE);
+      break;
+
+   case TGSI_OPCODE_TXB:
+      /* Texture lookup with lod bias */
+      /* src[0] = texcoord (src[0].w = LOD bias) */
+      /* src[1] = sampler unit */
+      exec_tex(mach, inst, TRUE, FALSE);
+      break;
+
+   case TGSI_OPCODE_TXD:
+      /* Texture lookup with explict partial derivatives */
+      /* src[0] = texcoord */
+      /* src[1] = d[strq]/dx */
+      /* src[2] = d[strq]/dy */
+      /* src[3] = sampler unit */
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_TXL:
+      /* Texture lookup with explit LOD */
+      /* src[0] = texcoord (src[0].w = LOD) */
+      /* src[1] = sampler unit */
+      exec_tex(mach, inst, TRUE, FALSE);
+      break;
+
+   case TGSI_OPCODE_TXP:
+      /* Texture lookup with projection */
+      /* src[0] = texcoord (src[0].w = projection) */
+      /* src[1] = sampler unit */
+      exec_tex(mach, inst, FALSE, TRUE);
+      break;
+
+   case TGSI_OPCODE_UP2H:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_UP2US:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_UP4B:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_UP4UB:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_X2D:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_ARA:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_BRA:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_CAL:
+      /* skip the call if no execution channels are enabled */
+      if (mach->ExecMask) {
+         /* do the call */
+
+         /* push the Cond, Loop, Cont stacks */
+         assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
+         mach->CondStack[mach->CondStackTop++] = mach->CondMask;
+         assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+         mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
+         assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+         mach->ContStack[mach->ContStackTop++] = mach->ContMask;
+
+         assert(mach->FuncStackTop < TGSI_EXEC_MAX_CALL_NESTING);
+         mach->FuncStack[mach->FuncStackTop++] = mach->FuncMask;
+
+         /* note that PC was already incremented above */
+         mach->CallStack[mach->CallStackTop++] = *pc;
+         *pc = inst->InstructionExtLabel.Label;
+      }
+      break;
+
+   case TGSI_OPCODE_RET:
+      mach->FuncMask &= ~mach->ExecMask;
+      UPDATE_EXEC_MASK(mach);
+
+      if (mach->FuncMask == 0x0) {
+         /* really return now (otherwise, keep executing */
+
+         if (mach->CallStackTop == 0) {
+            /* returning from main() */
+            *pc = -1;
+            return;
+         }
+         *pc = mach->CallStack[--mach->CallStackTop];
+
+         /* pop the Cond, Loop, Cont stacks */
+         assert(mach->CondStackTop > 0);
+         mach->CondMask = mach->CondStack[--mach->CondStackTop];
+         assert(mach->LoopStackTop > 0);
+         mach->LoopMask = mach->LoopStack[--mach->LoopStackTop];
+         assert(mach->ContStackTop > 0);
+         mach->ContMask = mach->ContStack[--mach->ContStackTop];
+         assert(mach->FuncStackTop > 0);
+         mach->FuncMask = mach->FuncStack[--mach->FuncStackTop];
+
+         UPDATE_EXEC_MASK(mach);
+      }
+      break;
+
+   case TGSI_OPCODE_SSG:
+   /* TGSI_OPCODE_SGN */
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_sgn( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_CMP:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH(&r[0], 0, chan_index);
+         FETCH(&r[1], 1, chan_index);
+         FETCH(&r[2], 2, chan_index);
+
+         micro_lt( &r[0], &r[0], &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], &r[1], &r[2] );
+
+         STORE(&r[0], 0, chan_index);
+      }
+      break;
+
+   case TGSI_OPCODE_SCS:
+      if( IS_CHANNEL_ENABLED( *inst, CHAN_X ) || IS_CHANNEL_ENABLED( *inst, CHAN_Y ) ) {
+         FETCH( &r[0], 0, CHAN_X );
+      }
+      if( IS_CHANNEL_ENABLED( *inst, CHAN_X ) ) {
+         micro_cos( &r[1], &r[0] );
+         STORE( &r[1], 0, CHAN_X );
+      }
+      if( IS_CHANNEL_ENABLED( *inst, CHAN_Y ) ) {
+         micro_sin( &r[1], &r[0] );
+         STORE( &r[1], 0, CHAN_Y );
+      }
+      if( IS_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
+         STORE( &mach->Temps[TEMP_0_I].xyzw[TEMP_0_C], 0, CHAN_Z );
+      }
+      if( IS_CHANNEL_ENABLED( *inst, CHAN_W ) ) {
+         STORE( &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_NRM:
+      /* 3-component vector normalize */
+      {
+         union tgsi_exec_channel tmp, dot;
+
+         /* tmp = dp3(src0, src0): */
+         FETCH( &r[0], 0, CHAN_X );
+         micro_mul( &tmp, &r[0], &r[0] );
+
+         FETCH( &r[1], 0, CHAN_Y );
+         micro_mul( &dot, &r[1], &r[1] );
+         micro_add( &tmp, &tmp, &dot );
+
+         FETCH( &r[2], 0, CHAN_Z );
+         micro_mul( &dot, &r[2], &r[2] );
+         micro_add( &tmp, &tmp, &dot );
+
+         /* tmp = 1 / sqrt(tmp) */
+         micro_sqrt( &tmp, &tmp );
+         micro_div( &tmp, &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &tmp );
+
+         /* note: w channel is undefined */
+         FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+            /* chan = chan * tmp */
+            micro_mul( &r[chan_index], &tmp, &r[chan_index] );
+            STORE( &r[chan_index], 0, chan_index );
+         }
+      }
+      break;
+
+   case TGSI_OPCODE_NRM4:
+      /* 4-component vector normalize */
+      {
+         union tgsi_exec_channel tmp, dot;
+
+         /* tmp = dp4(src0, src0): */
+         FETCH( &r[0], 0, CHAN_X );
+         micro_mul( &tmp, &r[0], &r[0] );
+
+         FETCH( &r[1], 0, CHAN_Y );
+         micro_mul( &dot, &r[1], &r[1] );
+         micro_add( &tmp, &tmp, &dot );
+
+         FETCH( &r[2], 0, CHAN_Z );
+         micro_mul( &dot, &r[2], &r[2] );
+         micro_add( &tmp, &tmp, &dot );
+
+         FETCH( &r[3], 0, CHAN_W );
+         micro_mul( &dot, &r[3], &r[3] );
+         micro_add( &tmp, &tmp, &dot );
+
+         /* tmp = 1 / sqrt(tmp) */
+         micro_sqrt( &tmp, &tmp );
+         micro_div( &tmp, &mach->Temps[TEMP_1_I].xyzw[TEMP_1_C], &tmp );
+
+         FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+            /* chan = chan * tmp */
+            micro_mul( &r[chan_index], &tmp, &r[chan_index] );
+            STORE( &r[chan_index], 0, chan_index );
+         }
+      }
+      break;
+
+   case TGSI_OPCODE_DIV:
+      assert( 0 );
+      break;
+
+   case TGSI_OPCODE_DP2:
+      FETCH( &r[0], 0, CHAN_X );
+      FETCH( &r[1], 1, CHAN_X );
+      micro_mul( &r[0], &r[0], &r[1] );
+
+      FETCH( &r[1], 0, CHAN_Y );
+      FETCH( &r[2], 1, CHAN_Y );
+      micro_mul( &r[1], &r[1], &r[2] );
+      micro_add( &r[0], &r[0], &r[1] );
+
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_IF:
+      /* push CondMask */
+      assert(mach->CondStackTop < TGSI_EXEC_MAX_COND_NESTING);
+      mach->CondStack[mach->CondStackTop++] = mach->CondMask;
+      FETCH( &r[0], 0, CHAN_X );
+      /* update CondMask */
+      if( ! r[0].u[0] ) {
+         mach->CondMask &= ~0x1;
+      }
+      if( ! r[0].u[1] ) {
+         mach->CondMask &= ~0x2;
+      }
+      if( ! r[0].u[2] ) {
+         mach->CondMask &= ~0x4;
+      }
+      if( ! r[0].u[3] ) {
+         mach->CondMask &= ~0x8;
+      }
+      UPDATE_EXEC_MASK(mach);
+      /* Todo: If CondMask==0, jump to ELSE */
+      break;
+
+   case TGSI_OPCODE_ELSE:
+      /* invert CondMask wrt previous mask */
+      {
+         uint prevMask;
+         assert(mach->CondStackTop > 0);
+         prevMask = mach->CondStack[mach->CondStackTop - 1];
+         mach->CondMask = ~mach->CondMask & prevMask;
+         UPDATE_EXEC_MASK(mach);
+         /* Todo: If CondMask==0, jump to ENDIF */
+      }
+      break;
+
+   case TGSI_OPCODE_ENDIF:
+      /* pop CondMask */
+      assert(mach->CondStackTop > 0);
+      mach->CondMask = mach->CondStack[--mach->CondStackTop];
+      UPDATE_EXEC_MASK(mach);
+      break;
+
+   case TGSI_OPCODE_END:
+      /* halt execution */
+      *pc = -1;
+      break;
+
+   case TGSI_OPCODE_REP:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_ENDREP:
+       assert (0);
+       break;
+
+   case TGSI_OPCODE_PUSHA:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_POPA:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_CEIL:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_ceil( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_I2F:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_i2f( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_NOT:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_not( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_TRUNC:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         micro_trunc( &r[0], &r[0] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SHL:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_shl( &r[0], &r[0], &r[1] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SHR:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_ishr( &r[0], &r[0], &r[1] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_AND:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_and( &r[0], &r[0], &r[1] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_OR:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_or( &r[0], &r[0], &r[1] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_MOD:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_XOR:
+      FOR_EACH_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( &r[0], 0, chan_index );
+         FETCH( &r[1], 1, chan_index );
+         micro_xor( &r[0], &r[0], &r[1] );
+         STORE( &r[0], 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SAD:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_TXF:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_TXQ:
+      assert (0);
+      break;
+
+   case TGSI_OPCODE_EMIT:
+      mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] += 16;
+      mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]]++;
+      break;
+
+   case TGSI_OPCODE_ENDPRIM:
+      mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]++;
+      mach->Primitives[mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0]] = 0;
+      break;
+
+   case TGSI_OPCODE_LOOP:
+      /* fall-through (for now) */
+   case TGSI_OPCODE_BGNLOOP2:
+      /* push LoopMask and ContMasks */
+      assert(mach->LoopStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+      mach->LoopStack[mach->LoopStackTop++] = mach->LoopMask;
+      assert(mach->ContStackTop < TGSI_EXEC_MAX_LOOP_NESTING);
+      mach->ContStack[mach->ContStackTop++] = mach->ContMask;
+      break;
+
+   case TGSI_OPCODE_ENDLOOP:
+      /* fall-through (for now at least) */
+   case TGSI_OPCODE_ENDLOOP2:
+      /* Restore ContMask, but don't pop */
+      assert(mach->ContStackTop > 0);
+      mach->ContMask = mach->ContStack[mach->ContStackTop - 1];
+      UPDATE_EXEC_MASK(mach);
+      if (mach->ExecMask) {
+         /* repeat loop: jump to instruction just past BGNLOOP */
+         *pc = inst->InstructionExtLabel.Label + 1;
+      }
+      else {
+         /* exit loop: pop LoopMask */
+         assert(mach->LoopStackTop > 0);
+         mach->LoopMask = mach->LoopStack[--mach->LoopStackTop];
+         /* pop ContMask */
+         assert(mach->ContStackTop > 0);
+         mach->ContMask = mach->ContStack[--mach->ContStackTop];
+      }
+      UPDATE_EXEC_MASK(mach);
+      break;
+
+   case TGSI_OPCODE_BRK:
+      /* turn off loop channels for each enabled exec channel */
+      mach->LoopMask &= ~mach->ExecMask;
+      /* Todo: if mach->LoopMask == 0, jump to end of loop */
+      UPDATE_EXEC_MASK(mach);
+      break;
+
+   case TGSI_OPCODE_CONT:
+      /* turn off cont channels for each enabled exec channel */
+      mach->ContMask &= ~mach->ExecMask;
+      /* Todo: if mach->LoopMask == 0, jump to end of loop */
+      UPDATE_EXEC_MASK(mach);
+      break;
+
+   case TGSI_OPCODE_BGNSUB:
+      /* no-op */
+      break;
+
+   case TGSI_OPCODE_ENDSUB:
+      /* no-op */
+      break;
+
+   case TGSI_OPCODE_NOISE1:
+      assert( 0 );
+      break;
+
+   case TGSI_OPCODE_NOISE2:
+      assert( 0 );
+      break;
+
+   case TGSI_OPCODE_NOISE3:
+      assert( 0 );
+      break;
+
+   case TGSI_OPCODE_NOISE4:
+      assert( 0 );
+      break;
+
+   case TGSI_OPCODE_NOP:
+      break;
+
+   default:
+      assert( 0 );
+   }
+}
+
+
+/**
+ * Run TGSI interpreter.
+ * \return bitmask of "alive" quad components
+ */
+uint
+tgsi_exec_machine_run( struct tgsi_exec_machine *mach )
+{
+   uint i;
+   int pc = 0;
+
+   mach->CondMask = 0xf;
+   mach->LoopMask = 0xf;
+   mach->ContMask = 0xf;
+   mach->FuncMask = 0xf;
+   mach->ExecMask = 0xf;
+
+   mach->CondStackTop = 0; /* temporarily subvert this assertion */
+   assert(mach->CondStackTop == 0);
+   assert(mach->LoopStackTop == 0);
+   assert(mach->ContStackTop == 0);
+   assert(mach->CallStackTop == 0);
+
+   mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0] = 0;
+   mach->Temps[TEMP_OUTPUT_I].xyzw[TEMP_OUTPUT_C].u[0] = 0;
+
+   if( mach->Processor == TGSI_PROCESSOR_GEOMETRY ) {
+      mach->Temps[TEMP_PRIMITIVE_I].xyzw[TEMP_PRIMITIVE_C].u[0] = 0;
+      mach->Primitives[0] = 0;
+   }
+
+   for (i = 0; i < QUAD_SIZE; i++) {
+      mach->Temps[TEMP_CC_I].xyzw[TEMP_CC_C].u[i] =
+         (TGSI_EXEC_CC_EQ << TGSI_EXEC_CC_X_SHIFT) |
+         (TGSI_EXEC_CC_EQ << TGSI_EXEC_CC_Y_SHIFT) |
+         (TGSI_EXEC_CC_EQ << TGSI_EXEC_CC_Z_SHIFT) |
+         (TGSI_EXEC_CC_EQ << TGSI_EXEC_CC_W_SHIFT);
+   }
+
+   /* execute declarations (interpolants) */
+   for (i = 0; i < mach->NumDeclarations; i++) {
+      exec_declaration( mach, mach->Declarations+i );
+   }
+
+   /* execute instructions, until pc is set to -1 */
+   while (pc != -1) {
+      assert(pc < (int) mach->NumInstructions);
+      exec_instruction( mach, mach->Instructions + pc, &pc );
+   }
+
+#if 0
+   /* we scale from floats in [0,1] to Zbuffer ints in sp_quad_depth_test.c */
+   if (mach->Processor == TGSI_PROCESSOR_FRAGMENT) {
+      /*
+       * Scale back depth component.
+       */
+      for (i = 0; i < 4; i++)
+         mach->Outputs[0].xyzw[2].f[i] *= ctx->DrawBuffer->_DepthMaxF;
+   }
+#endif
+
+   return ~mach->Temps[TEMP_KILMASK_I].xyzw[TEMP_KILMASK_C].u[0];
+}
+
+
diff --git a/src/gallium/auxiliary/tgsi/tgsi_exec.h b/src/gallium/auxiliary/tgsi/tgsi_exec.h
new file mode 100644
index 00000000000..4ffd4efbffa
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_exec.h
@@ -0,0 +1,300 @@
+/**************************************************************************
+ * 
+ * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef TGSI_EXEC_H
+#define TGSI_EXEC_H
+
+#include "pipe/p_compiler.h"
+
+#if defined __cplusplus
+extern "C" {
+#endif
+
+#define MAX_LABELS 1024
+
+#define NUM_CHANNELS 4  /* R,G,B,A */
+#define QUAD_SIZE    4  /* 4 pixel/quad */
+
+/**
+  * Registers may be treated as float, signed int or unsigned int.
+  */
+union tgsi_exec_channel
+{
+   float    f[QUAD_SIZE];
+   int      i[QUAD_SIZE];
+   unsigned u[QUAD_SIZE];
+};
+
+/**
+  * A vector[RGBA] of channels[4 pixels]
+  */
+struct tgsi_exec_vector
+{
+   union tgsi_exec_channel xyzw[NUM_CHANNELS];
+};
+
+/**
+ * For fragment programs, information for computing fragment input
+ * values from plane equation of the triangle/line.
+ */
+struct tgsi_interp_coef
+{
+   float a0[NUM_CHANNELS];	/* in an xyzw layout */
+   float dadx[NUM_CHANNELS];
+   float dady[NUM_CHANNELS];
+};
+
+/**
+ * Information for sampling textures, which must be implemented
+ * by code outside the TGSI executor.
+ */
+struct tgsi_sampler
+{
+   /** Get samples for four fragments in a quad */
+   void (*get_samples)(struct tgsi_sampler *sampler,
+                       const float s[QUAD_SIZE],
+                       const float t[QUAD_SIZE],
+                       const float p[QUAD_SIZE],
+                       float lodbias,
+                       float rgba[NUM_CHANNELS][QUAD_SIZE]);
+};
+
+/**
+ * For branching/calling subroutines.
+ */
+struct tgsi_exec_labels
+{
+   unsigned labels[MAX_LABELS][2];
+   unsigned count;
+};
+
+
+#define TGSI_EXEC_NUM_TEMPS       128
+#define TGSI_EXEC_NUM_TEMP_EXTRAS   6
+#define TGSI_EXEC_NUM_IMMEDIATES  256
+
+/*
+ * Locations of various utility registers (_I = Index, _C = Channel)
+ */
+#define TGSI_EXEC_TEMP_00000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
+#define TGSI_EXEC_TEMP_00000000_C   0
+
+#define TGSI_EXEC_TEMP_7FFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
+#define TGSI_EXEC_TEMP_7FFFFFFF_C   1
+
+#define TGSI_EXEC_TEMP_80000000_I   (TGSI_EXEC_NUM_TEMPS + 0)
+#define TGSI_EXEC_TEMP_80000000_C   2
+
+#define TGSI_EXEC_TEMP_FFFFFFFF_I   (TGSI_EXEC_NUM_TEMPS + 0)
+#define TGSI_EXEC_TEMP_FFFFFFFF_C   3
+
+#define TGSI_EXEC_TEMP_ONE_I        (TGSI_EXEC_NUM_TEMPS + 1)
+#define TGSI_EXEC_TEMP_ONE_C        0
+
+#define TGSI_EXEC_TEMP_TWO_I        (TGSI_EXEC_NUM_TEMPS + 1)
+#define TGSI_EXEC_TEMP_TWO_C        1
+
+#define TGSI_EXEC_TEMP_128_I        (TGSI_EXEC_NUM_TEMPS + 1)
+#define TGSI_EXEC_TEMP_128_C        2
+
+#define TGSI_EXEC_TEMP_MINUS_128_I  (TGSI_EXEC_NUM_TEMPS + 1)
+#define TGSI_EXEC_TEMP_MINUS_128_C  3
+
+#define TGSI_EXEC_TEMP_KILMASK_I    (TGSI_EXEC_NUM_TEMPS + 2)
+#define TGSI_EXEC_TEMP_KILMASK_C    0
+
+#define TGSI_EXEC_TEMP_OUTPUT_I     (TGSI_EXEC_NUM_TEMPS + 2)
+#define TGSI_EXEC_TEMP_OUTPUT_C     1
+
+#define TGSI_EXEC_TEMP_PRIMITIVE_I  (TGSI_EXEC_NUM_TEMPS + 2)
+#define TGSI_EXEC_TEMP_PRIMITIVE_C  2
+
+/* NVIDIA condition code (CC) vector
+ */
+#define TGSI_EXEC_CC_GT       0x01
+#define TGSI_EXEC_CC_EQ       0x02
+#define TGSI_EXEC_CC_LT       0x04
+#define TGSI_EXEC_CC_UN       0x08
+
+#define TGSI_EXEC_CC_X_MASK   0x000000ff
+#define TGSI_EXEC_CC_X_SHIFT  0
+#define TGSI_EXEC_CC_Y_MASK   0x0000ff00
+#define TGSI_EXEC_CC_Y_SHIFT  8
+#define TGSI_EXEC_CC_Z_MASK   0x00ff0000
+#define TGSI_EXEC_CC_Z_SHIFT  16
+#define TGSI_EXEC_CC_W_MASK   0xff000000
+#define TGSI_EXEC_CC_W_SHIFT  24
+
+#define TGSI_EXEC_TEMP_CC_I         (TGSI_EXEC_NUM_TEMPS + 2)
+#define TGSI_EXEC_TEMP_CC_C         3
+
+#define TGSI_EXEC_TEMP_THREE_I      (TGSI_EXEC_NUM_TEMPS + 3)
+#define TGSI_EXEC_TEMP_THREE_C      0
+
+#define TGSI_EXEC_TEMP_HALF_I       (TGSI_EXEC_NUM_TEMPS + 3)
+#define TGSI_EXEC_TEMP_HALF_C       1
+
+/* execution mask, each value is either 0 or ~0 */
+#define TGSI_EXEC_MASK_I            (TGSI_EXEC_NUM_TEMPS + 3)
+#define TGSI_EXEC_MASK_C            2
+
+#define TGSI_EXEC_TEMP_R0           (TGSI_EXEC_NUM_TEMPS + 4)
+
+#define TGSI_EXEC_TEMP_ADDR         (TGSI_EXEC_NUM_TEMPS + 5)
+
+
+#define TGSI_EXEC_MAX_COND_NESTING  20
+#define TGSI_EXEC_MAX_LOOP_NESTING  20
+#define TGSI_EXEC_MAX_CALL_NESTING  20
+
+/* The maximum number of input attributes per vertex. For 2D
+ * input register files, this is the stride between two 1D
+ * arrays.
+ */
+#define TGSI_EXEC_MAX_INPUT_ATTRIBS 17
+
+/* The maximum number of constant vectors per constant buffer.
+ */
+#define TGSI_EXEC_MAX_CONST_BUFFER  4096
+
+/**
+ * Run-time virtual machine state for executing TGSI shader.
+ */
+struct tgsi_exec_machine
+{
+   /* Total = program temporaries + internal temporaries
+    *         + 1 padding to align to 16 bytes
+    */
+   struct tgsi_exec_vector       _Temps[TGSI_EXEC_NUM_TEMPS +
+                                        TGSI_EXEC_NUM_TEMP_EXTRAS + 1];
+
+   /*
+    * This will point to _Temps after aligning to 16B boundary.
+    */
+   struct tgsi_exec_vector       *Temps;
+   struct tgsi_exec_vector       *Addrs;
+
+   struct tgsi_sampler           **Samplers;
+
+   float                         Imms[TGSI_EXEC_NUM_IMMEDIATES][4];
+   unsigned                      ImmLimit;
+   const float                   (*Consts)[4];
+   struct tgsi_exec_vector       *Inputs;
+   struct tgsi_exec_vector       *Outputs;
+   const struct tgsi_token       *Tokens;
+   unsigned                      Processor;
+
+   /* GEOMETRY processor only. */
+   unsigned                      *Primitives;
+
+   /* FRAGMENT processor only. */
+   const struct tgsi_interp_coef *InterpCoefs;
+   struct tgsi_exec_vector       QuadPos;
+
+   /* Conditional execution masks */
+   uint CondMask;  /**< For IF/ELSE/ENDIF */
+   uint LoopMask;  /**< For BGNLOOP/ENDLOOP */
+   uint ContMask;  /**< For loop CONT statements */
+   uint FuncMask;  /**< For function calls */
+   uint ExecMask;  /**< = CondMask & LoopMask */
+
+   /** Condition mask stack (for nested conditionals) */
+   uint CondStack[TGSI_EXEC_MAX_COND_NESTING];
+   int CondStackTop;
+
+   /** Loop mask stack (for nested loops) */
+   uint LoopStack[TGSI_EXEC_MAX_LOOP_NESTING];
+   int LoopStackTop;
+
+   /** Loop continue mask stack (see comments in tgsi_exec.c) */
+   uint ContStack[TGSI_EXEC_MAX_LOOP_NESTING];
+   int ContStackTop;
+
+   /** Function execution mask stack (for executing subroutine code) */
+   uint FuncStack[TGSI_EXEC_MAX_CALL_NESTING];
+   int FuncStackTop;
+
+   /** Function call stack for saving/restoring the program counter */
+   uint CallStack[TGSI_EXEC_MAX_CALL_NESTING];
+   int CallStackTop;
+
+   struct tgsi_full_instruction *Instructions;
+   uint NumInstructions;
+
+   struct tgsi_full_declaration *Declarations;
+   uint NumDeclarations;
+
+   struct tgsi_exec_labels Labels;
+};
+
+void
+tgsi_exec_machine_init(
+   struct tgsi_exec_machine *mach );
+
+
+void 
+tgsi_exec_machine_bind_shader(
+   struct tgsi_exec_machine *mach,
+   const struct tgsi_token *tokens,
+   uint numSamplers,
+   struct tgsi_sampler **samplers);
+
+uint
+tgsi_exec_machine_run(
+   struct tgsi_exec_machine *mach );
+
+
+void
+tgsi_exec_machine_free_data(struct tgsi_exec_machine *mach);
+
+
+static INLINE void
+tgsi_set_kill_mask(struct tgsi_exec_machine *mach, unsigned mask)
+{
+   mach->Temps[TGSI_EXEC_TEMP_KILMASK_I].xyzw[TGSI_EXEC_TEMP_KILMASK_C].u[0] =
+      mask;
+}
+
+
+/** Set execution mask values prior to executing the shader */
+static INLINE void
+tgsi_set_exec_mask(struct tgsi_exec_machine *mach,
+                   boolean ch0, boolean ch1, boolean ch2, boolean ch3)
+{
+   int *mask = mach->Temps[TGSI_EXEC_MASK_I].xyzw[TGSI_EXEC_MASK_C].i;
+   mask[0] = ch0 ? ~0 : 0;
+   mask[1] = ch1 ? ~0 : 0;
+   mask[2] = ch2 ? ~0 : 0;
+   mask[3] = ch3 ? ~0 : 0;
+}
+
+
+#if defined __cplusplus
+} /* extern "C" */
+#endif
+
+#endif /* TGSI_EXEC_H */
diff --git a/src/gallium/auxiliary/tgsi/tgsi_info.c b/src/gallium/auxiliary/tgsi/tgsi_info.c
new file mode 100644
index 00000000000..68c7a6b7f58
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_info.c
@@ -0,0 +1,161 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_debug.h"
+#include "tgsi_info.h"
+
+static const struct tgsi_opcode_info opcode_info[TGSI_OPCODE_LAST] =
+{
+   { 1, 1, 0, 0, "ARL" },
+   { 1, 1, 0, 0, "MOV" },
+   { 1, 1, 0, 0, "LIT" },
+   { 1, 1, 0, 0, "RCP" },
+   { 1, 1, 0, 0, "RSQ" },
+   { 1, 1, 0, 0, "EXP" },
+   { 1, 1, 0, 0, "LOG" },
+   { 1, 2, 0, 0, "MUL" },
+   { 1, 2, 0, 0, "ADD" },
+   { 1, 2, 0, 0, "DP3" },
+   { 1, 2, 0, 0, "DP4" },
+   { 1, 2, 0, 0, "DST" },
+   { 1, 2, 0, 0, "MIN" },
+   { 1, 2, 0, 0, "MAX" },
+   { 1, 2, 0, 0, "SLT" },
+   { 1, 2, 0, 0, "SGE" },
+   { 1, 3, 0, 0, "MAD" },
+   { 1, 2, 0, 0, "SUB" },
+   { 1, 3, 0, 0, "LERP" },
+   { 1, 3, 0, 0, "CND" },
+   { 1, 3, 0, 0, "CND0" },
+   { 1, 3, 0, 0, "DOT2ADD" },
+   { 1, 2, 0, 0, "INDEX" },
+   { 1, 1, 0, 0, "NEGATE" },
+   { 1, 1, 0, 0, "FRAC" },
+   { 1, 3, 0, 0, "CLAMP" },
+   { 1, 1, 0, 0, "FLOOR" },
+   { 1, 1, 0, 0, "ROUND" },
+   { 1, 1, 0, 0, "EXPBASE2" },
+   { 1, 1, 0, 0, "LOGBASE2" },
+   { 1, 2, 0, 0, "POWER" },
+   { 1, 2, 0, 0, "CROSSPRODUCT" },
+   { 1, 2, 0, 0, "MULTIPLYMATRIX" },
+   { 1, 1, 0, 0, "ABS" },
+   { 1, 1, 0, 0, "RCC" },
+   { 1, 2, 0, 0, "DPH" },
+   { 1, 1, 0, 0, "COS" },
+   { 1, 1, 0, 0, "DDX" },
+   { 1, 1, 0, 0, "DDY" },
+   { 0, 0, 0, 0, "KILP" },
+   { 1, 1, 0, 0, "PK2H" },
+   { 1, 1, 0, 0, "PK2US" },
+   { 1, 1, 0, 0, "PK4B" },
+   { 1, 1, 0, 0, "PK4UB" },
+   { 1, 2, 0, 0, "RFL" },
+   { 1, 2, 0, 0, "SEQ" },
+   { 1, 2, 0, 0, "SFL" },
+   { 1, 2, 0, 0, "SGT" },
+   { 1, 1, 0, 0, "SIN" },
+   { 1, 2, 0, 0, "SLE" },
+   { 1, 2, 0, 0, "SNE" },
+   { 1, 2, 0, 0, "STR" },
+   { 1, 2, 1, 0, "TEX" },
+   { 1, 4, 1, 0, "TXD" },
+   { 1, 2, 1, 0, "TXP" },
+   { 1, 1, 0, 0, "UP2H" },
+   { 1, 1, 0, 0, "UP2US" },
+   { 1, 1, 0, 0, "UP4B" },
+   { 1, 1, 0, 0, "UP4UB" },
+   { 1, 3, 0, 0, "X2D" },
+   { 1, 1, 0, 0, "ARA" },
+   { 1, 1, 0, 0, "ARR" },
+   { 0, 1, 0, 0, "BRA" },
+   { 0, 0, 0, 1, "CAL" },
+   { 0, 0, 0, 0, "RET" },
+   { 1, 1, 0, 0, "SSG" },
+   { 1, 3, 0, 0, "CMP" },
+   { 1, 1, 0, 0, "SCS" },
+   { 1, 2, 1, 0, "TXB" },
+   { 1, 1, 0, 0, "NRM" },
+   { 1, 2, 0, 0, "DIV" },
+   { 1, 2, 0, 0, "DP2" },
+   { 1, 2, 1, 0, "TXL" },
+   { 0, 0, 0, 0, "BRK" },
+   { 0, 1, 0, 1, "IF" },
+   { 0, 0, 0, 0, "LOOP" },
+   { 0, 1, 0, 0, "REP" },
+   { 0, 0, 0, 1, "ELSE" },
+   { 0, 0, 0, 0, "ENDIF" },
+   { 0, 0, 0, 0, "ENDLOOP" },
+   { 0, 0, 0, 0, "ENDREP" },
+   { 0, 1, 0, 0, "PUSHA" },
+   { 1, 0, 0, 0, "POPA" },
+   { 1, 1, 0, 0, "CEIL" },
+   { 1, 1, 0, 0, "I2F" },
+   { 1, 1, 0, 0, "NOT" },
+   { 1, 1, 0, 0, "TRUNC" },
+   { 1, 2, 0, 0, "SHL" },
+   { 1, 2, 0, 0, "SHR" },
+   { 1, 2, 0, 0, "AND" },
+   { 1, 2, 0, 0, "OR" },
+   { 1, 2, 0, 0, "MOD" },
+   { 1, 2, 0, 0, "XOR" },
+   { 1, 3, 0, 0, "SAD" },
+   { 1, 2, 1, 0, "TXF" },
+   { 1, 2, 1, 0, "TXQ" },
+   { 0, 0, 0, 0, "CONT" },
+   { 0, 0, 0, 0, "EMIT" },
+   { 0, 0, 0, 0, "ENDPRIM" },
+   { 0, 0, 0, 1, "BGNLOOP2" },
+   { 0, 0, 0, 0, "BGNSUB" },
+   { 0, 0, 0, 1, "ENDLOOP2" },
+   { 0, 0, 0, 0, "ENDSUB" },
+   { 1, 1, 0, 0, "NOISE1" },
+   { 1, 1, 0, 0, "NOISE2" },
+   { 1, 1, 0, 0, "NOISE3" },
+   { 1, 1, 0, 0, "NOISE4" },
+   { 0, 0, 0, 0, "NOP" },
+   { 1, 2, 0, 0, "M4X3" },
+   { 1, 2, 0, 0, "M3X4" },
+   { 1, 2, 0, 0, "M3X3" },
+   { 1, 2, 0, 0, "M3X2" },
+   { 1, 1, 0, 0, "NRM4" },
+   { 0, 1, 0, 0, "CALLNZ" },
+   { 0, 1, 0, 0, "IFC" },
+   { 0, 1, 0, 0, "BREAKC" },
+   { 0, 1, 0, 0, "KIL" },
+   { 0, 0, 0, 0, "END" },
+   { 1, 1, 0, 0, "SWZ" }
+};
+
+const struct tgsi_opcode_info *
+tgsi_get_opcode_info( uint opcode )
+{
+   if (opcode < TGSI_OPCODE_LAST)
+      return &opcode_info[opcode];
+   assert( 0 );
+   return NULL;
+}
diff --git a/src/gallium/auxiliary/tgsi/tgsi_info.h b/src/gallium/auxiliary/tgsi/tgsi_info.h
new file mode 100644
index 00000000000..7230bdaae33
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_info.h
@@ -0,0 +1,53 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef TGSI_INFO_H
+#define TGSI_INFO_H
+
+#include "pipe/p_shader_tokens.h"
+
+#if defined __cplusplus
+extern "C" {
+#endif
+
+struct tgsi_opcode_info
+{
+   uint num_dst;
+   uint num_src;
+   boolean is_tex;
+   boolean is_branch;
+   const char *mnemonic;
+};
+
+const struct tgsi_opcode_info *
+tgsi_get_opcode_info( uint opcode );
+
+#if defined __cplusplus
+}
+#endif
+
+#endif /* TGSI_INFO_H */
diff --git a/src/gallium/auxiliary/tgsi/tgsi_iterate.c b/src/gallium/auxiliary/tgsi/tgsi_iterate.c
new file mode 100644
index 00000000000..5371a88b964
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_iterate.c
@@ -0,0 +1,85 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_debug.h"
+#include "tgsi_iterate.h"
+
+boolean
+tgsi_iterate_shader(
+   const struct tgsi_token *tokens,
+   struct tgsi_iterate_context *ctx )
+{
+   struct tgsi_parse_context parse;
+
+   if (tgsi_parse_init( &parse, tokens ) != TGSI_PARSE_OK)
+      return FALSE;
+
+   ctx->processor = parse.FullHeader.Processor;
+   ctx->version = parse.FullVersion.Version;
+
+   if (ctx->prolog)
+      if (!ctx->prolog( ctx ))
+         goto fail;
+
+   while (!tgsi_parse_end_of_tokens( &parse )) {
+      tgsi_parse_token( &parse );
+
+      switch (parse.FullToken.Token.Type) {
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         if (ctx->iterate_instruction)
+            if (!ctx->iterate_instruction( ctx, &parse.FullToken.FullInstruction ))
+               goto fail;
+         break;
+
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         if (ctx->iterate_declaration)
+            if (!ctx->iterate_declaration( ctx, &parse.FullToken.FullDeclaration ))
+               goto fail;
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         if (ctx->iterate_immediate)
+            if (!ctx->iterate_immediate( ctx, &parse.FullToken.FullImmediate ))
+               goto fail;
+         break;
+
+      default:
+         assert( 0 );
+      }
+   }
+
+   if (ctx->epilog)
+      if (!ctx->epilog( ctx ))
+         goto fail;
+
+   tgsi_parse_free( &parse );
+   return TRUE;
+
+fail:
+   tgsi_parse_free( &parse );
+   return FALSE;
+}
diff --git a/src/gallium/auxiliary/tgsi/tgsi_iterate.h b/src/gallium/auxiliary/tgsi/tgsi_iterate.h
new file mode 100644
index 00000000000..ec7b85bf63d
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_iterate.h
@@ -0,0 +1,76 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef TGSI_ITERATE_H
+#define TGSI_ITERATE_H
+
+#include "pipe/p_shader_tokens.h"
+#include "tgsi/tgsi_parse.h"
+
+#if defined __cplusplus
+extern "C" {
+#endif
+
+struct tgsi_iterate_context
+{
+   boolean
+   (* prolog)(
+      struct tgsi_iterate_context *ctx );
+
+   boolean
+   (* iterate_instruction)(
+      struct tgsi_iterate_context *ctx,
+      struct tgsi_full_instruction *inst );
+
+   boolean
+   (* iterate_declaration)(
+      struct tgsi_iterate_context *ctx,
+      struct tgsi_full_declaration *decl );
+
+   boolean
+   (* iterate_immediate)(
+      struct tgsi_iterate_context *ctx,
+      struct tgsi_full_immediate *imm );
+
+   boolean
+   (* epilog)(
+      struct tgsi_iterate_context *ctx );
+
+   struct tgsi_processor processor;
+   struct tgsi_version version;
+};
+
+boolean
+tgsi_iterate_shader(
+   const struct tgsi_token *tokens,
+   struct tgsi_iterate_context *ctx );
+
+#if defined __cplusplus
+}
+#endif
+
+#endif /* TGSI_ITERATE_H */
diff --git a/src/gallium/auxiliary/tgsi/tgsi_parse.c b/src/gallium/auxiliary/tgsi/tgsi_parse.c
new file mode 100644
index 00000000000..d374b16f9a1
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_parse.c
@@ -0,0 +1,344 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_debug.h"
+#include "pipe/p_shader_tokens.h"
+#include "tgsi_parse.h"
+#include "tgsi_build.h"
+#include "util/u_memory.h"
+
+void
+tgsi_full_token_init(
+   union tgsi_full_token *full_token )
+{
+   full_token->Token.Type = TGSI_TOKEN_TYPE_DECLARATION;
+}
+
+void
+tgsi_full_token_free(
+   union tgsi_full_token *full_token )
+{
+   if( full_token->Token.Type == TGSI_TOKEN_TYPE_IMMEDIATE ) {
+      FREE( (void *) full_token->FullImmediate.u.Pointer );
+   }
+}
+
+unsigned
+tgsi_parse_init(
+   struct tgsi_parse_context *ctx,
+   const struct tgsi_token *tokens )
+{
+   ctx->FullVersion.Version = *(struct tgsi_version *) &tokens[0];
+   if( ctx->FullVersion.Version.MajorVersion > 1 ) {
+      return TGSI_PARSE_ERROR;
+   }
+
+   ctx->FullHeader.Header = *(struct tgsi_header *) &tokens[1];
+   if( ctx->FullHeader.Header.HeaderSize >= 2 ) {
+      ctx->FullHeader.Processor = *(struct tgsi_processor *) &tokens[2];
+   }
+   else {
+      ctx->FullHeader.Processor = tgsi_default_processor();
+   }
+
+   ctx->Tokens = tokens;
+   ctx->Position = 1 + ctx->FullHeader.Header.HeaderSize;
+
+   tgsi_full_token_init( &ctx->FullToken );
+
+   return TGSI_PARSE_OK;
+}
+
+void
+tgsi_parse_free(
+   struct tgsi_parse_context *ctx )
+{
+   tgsi_full_token_free( &ctx->FullToken );
+}
+
+boolean
+tgsi_parse_end_of_tokens(
+   struct tgsi_parse_context *ctx )
+{
+   return ctx->Position >=
+      1 + ctx->FullHeader.Header.HeaderSize + ctx->FullHeader.Header.BodySize;
+}
+
+
+/**
+ * This function is used to avoid and work-around type punning/aliasing
+ * warnings.  The warnings seem harmless on x86 but on PPC they cause
+ * real failures.
+ */
+static INLINE void
+copy_token(void *dst, const void *src)
+{
+   memcpy(dst, src, 4);
+}
+
+
+/**
+ * Get next 4-byte token, return it at address specified by 'token'
+ */
+static void
+next_token(
+   struct tgsi_parse_context *ctx,
+   void *token )
+{
+   assert( !tgsi_parse_end_of_tokens( ctx ) );
+   copy_token(token, &ctx->Tokens[ctx->Position]);
+   ctx->Position++;
+}
+
+
+void
+tgsi_parse_token(
+   struct tgsi_parse_context *ctx )
+{
+   struct tgsi_token token;
+   unsigned i;
+
+   tgsi_full_token_free( &ctx->FullToken );
+   tgsi_full_token_init( &ctx->FullToken );
+
+   next_token( ctx, &token );
+
+   switch( token.Type ) {
+   case TGSI_TOKEN_TYPE_DECLARATION:
+   {
+      struct tgsi_full_declaration *decl = &ctx->FullToken.FullDeclaration;
+
+      *decl = tgsi_default_full_declaration();
+      copy_token(&decl->Declaration, &token);
+
+      next_token( ctx, &decl->DeclarationRange );
+
+      if( decl->Declaration.Semantic ) {
+         next_token( ctx, &decl->Semantic );
+      }
+
+      break;
+   }
+
+   case TGSI_TOKEN_TYPE_IMMEDIATE:
+   {
+      struct tgsi_full_immediate *imm = &ctx->FullToken.FullImmediate;
+
+      *imm = tgsi_default_full_immediate();
+      copy_token(&imm->Immediate, &token);
+      assert( !imm->Immediate.Extended );
+
+      switch (imm->Immediate.DataType) {
+      case TGSI_IMM_FLOAT32:
+         imm->u.Pointer = MALLOC(
+            sizeof( struct tgsi_immediate_float32 ) * (imm->Immediate.NrTokens - 1) );
+         for( i = 0; i < imm->Immediate.NrTokens - 1; i++ ) {
+            next_token( ctx, (struct tgsi_immediate_float32 *) &imm->u.ImmediateFloat32[i] );
+         }
+         break;
+
+      default:
+         assert( 0 );
+      }
+
+      break;
+   }
+
+   case TGSI_TOKEN_TYPE_INSTRUCTION:
+   {
+      struct tgsi_full_instruction *inst = &ctx->FullToken.FullInstruction;
+      unsigned extended;
+
+      *inst = tgsi_default_full_instruction();
+      copy_token(&inst->Instruction, &token);
+      extended = inst->Instruction.Extended;
+
+      while( extended ) {
+         struct tgsi_src_register_ext token;
+
+         next_token( ctx, &token );
+
+         switch( token.Type ) {
+         case TGSI_INSTRUCTION_EXT_TYPE_NV:
+            copy_token(&inst->InstructionExtNv, &token);
+            break;
+
+         case TGSI_INSTRUCTION_EXT_TYPE_LABEL:
+            copy_token(&inst->InstructionExtLabel, &token);
+            break;
+
+         case TGSI_INSTRUCTION_EXT_TYPE_TEXTURE:
+            copy_token(&inst->InstructionExtTexture, &token);
+            break;
+
+         default:
+            assert( 0 );
+         }
+
+         extended = token.Extended;
+      }
+
+      assert( inst->Instruction.NumDstRegs <= TGSI_FULL_MAX_DST_REGISTERS );
+
+      for(  i = 0; i < inst->Instruction.NumDstRegs; i++ ) {
+         unsigned extended;
+
+         next_token( ctx, &inst->FullDstRegisters[i].DstRegister );
+
+         /*
+          * No support for indirect or multi-dimensional addressing.
+          */
+         assert( !inst->FullDstRegisters[i].DstRegister.Indirect );
+         assert( !inst->FullDstRegisters[i].DstRegister.Dimension );
+
+         extended = inst->FullDstRegisters[i].DstRegister.Extended;
+
+         while( extended ) {
+            struct tgsi_src_register_ext token;
+
+            next_token( ctx, &token );
+
+            switch( token.Type ) {
+            case TGSI_DST_REGISTER_EXT_TYPE_CONDCODE:
+               copy_token(&inst->FullDstRegisters[i].DstRegisterExtConcode,
+                          &token);
+               break;
+
+            case TGSI_DST_REGISTER_EXT_TYPE_MODULATE:
+               copy_token(&inst->FullDstRegisters[i].DstRegisterExtModulate,
+                          &token);
+               break;
+
+            default:
+               assert( 0 );
+            }
+
+            extended = token.Extended;
+         }
+      }
+
+      assert( inst->Instruction.NumSrcRegs <= TGSI_FULL_MAX_SRC_REGISTERS );
+
+      for( i = 0; i < inst->Instruction.NumSrcRegs; i++ ) {
+         unsigned extended;
+
+         next_token( ctx, &inst->FullSrcRegisters[i].SrcRegister );
+
+         extended = inst->FullSrcRegisters[i].SrcRegister.Extended;
+
+         while( extended ) {
+            struct tgsi_src_register_ext token;
+
+            next_token( ctx, &token );
+
+            switch( token.Type ) {
+            case TGSI_SRC_REGISTER_EXT_TYPE_SWZ:
+               copy_token(&inst->FullSrcRegisters[i].SrcRegisterExtSwz,
+                          &token);
+               break;
+
+            case TGSI_SRC_REGISTER_EXT_TYPE_MOD:
+               copy_token(&inst->FullSrcRegisters[i].SrcRegisterExtMod,
+                          &token);
+               break;
+
+            default:
+               assert( 0 );
+            }
+
+            extended = token.Extended;
+         }
+
+         if( inst->FullSrcRegisters[i].SrcRegister.Indirect ) {
+            next_token( ctx, &inst->FullSrcRegisters[i].SrcRegisterInd );
+
+            /*
+             * No support for indirect or multi-dimensional addressing.
+             */
+            assert( !inst->FullSrcRegisters[i].SrcRegisterInd.Indirect );
+            assert( !inst->FullSrcRegisters[i].SrcRegisterInd.Dimension );
+            assert( !inst->FullSrcRegisters[i].SrcRegisterInd.Extended );
+         }
+
+         if( inst->FullSrcRegisters[i].SrcRegister.Dimension ) {
+            next_token( ctx, &inst->FullSrcRegisters[i].SrcRegisterDim );
+
+            /*
+             * No support for multi-dimensional addressing.
+             */
+            assert( !inst->FullSrcRegisters[i].SrcRegisterDim.Dimension );
+            assert( !inst->FullSrcRegisters[i].SrcRegisterDim.Extended );
+
+            if( inst->FullSrcRegisters[i].SrcRegisterDim.Indirect ) {
+               next_token( ctx, &inst->FullSrcRegisters[i].SrcRegisterDimInd );
+
+               /*
+               * No support for indirect or multi-dimensional addressing.
+               */
+               assert( !inst->FullSrcRegisters[i].SrcRegisterInd.Indirect );
+               assert( !inst->FullSrcRegisters[i].SrcRegisterInd.Dimension );
+               assert( !inst->FullSrcRegisters[i].SrcRegisterInd.Extended );
+            }
+         }
+      }
+
+      break;
+   }
+
+   default:
+      assert( 0 );
+   }
+}
+
+
+unsigned
+tgsi_num_tokens(const struct tgsi_token *tokens)
+{
+   struct tgsi_parse_context ctx;
+   if (tgsi_parse_init(&ctx, tokens) == TGSI_PARSE_OK) {
+      unsigned len = (ctx.FullHeader.Header.HeaderSize +
+                      ctx.FullHeader.Header.BodySize +
+                      1);
+      return len;
+   }
+   return 0;
+}
+
+
+/**
+ * Make a new copy of a token array.
+ */
+struct tgsi_token *
+tgsi_dup_tokens(const struct tgsi_token *tokens)
+{
+   unsigned n = tgsi_num_tokens(tokens);
+   unsigned bytes = n * sizeof(struct tgsi_token);
+   struct tgsi_token *new_tokens = (struct tgsi_token *) MALLOC(bytes);
+   if (new_tokens)
+      memcpy(new_tokens, tokens, bytes);
+   return new_tokens;
+}
diff --git a/src/gallium/auxiliary/tgsi/tgsi_parse.h b/src/gallium/auxiliary/tgsi/tgsi_parse.h
new file mode 100644
index 00000000000..054350712d8
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_parse.h
@@ -0,0 +1,151 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef TGSI_PARSE_H
+#define TGSI_PARSE_H
+
+#include "pipe/p_shader_tokens.h"
+
+#if defined __cplusplus
+extern "C" {
+#endif
+
+struct tgsi_full_version
+{
+   struct tgsi_version  Version;
+};
+
+struct tgsi_full_header
+{
+   struct tgsi_header      Header;
+   struct tgsi_processor   Processor;
+};
+
+struct tgsi_full_dst_register
+{
+   struct tgsi_dst_register               DstRegister;
+   struct tgsi_dst_register_ext_concode   DstRegisterExtConcode;
+   struct tgsi_dst_register_ext_modulate  DstRegisterExtModulate;
+};
+
+struct tgsi_full_src_register
+{
+   struct tgsi_src_register         SrcRegister;
+   struct tgsi_src_register_ext_swz SrcRegisterExtSwz;
+   struct tgsi_src_register_ext_mod SrcRegisterExtMod;
+   struct tgsi_src_register         SrcRegisterInd;
+   struct tgsi_dimension            SrcRegisterDim;
+   struct tgsi_src_register         SrcRegisterDimInd;
+};
+
+struct tgsi_full_declaration
+{
+   struct tgsi_declaration Declaration;
+   struct tgsi_declaration_range DeclarationRange;
+   struct tgsi_declaration_semantic Semantic;
+};
+
+struct tgsi_full_immediate
+{
+   struct tgsi_immediate   Immediate;
+   union
+   {
+      const void                          *Pointer;
+      const struct tgsi_immediate_float32 *ImmediateFloat32;
+   } u;
+};
+
+#define TGSI_FULL_MAX_DST_REGISTERS 2
+#define TGSI_FULL_MAX_SRC_REGISTERS 4 /* TXD has 4 */
+
+struct tgsi_full_instruction
+{
+   struct tgsi_instruction             Instruction;
+   struct tgsi_instruction_ext_nv      InstructionExtNv;
+   struct tgsi_instruction_ext_label   InstructionExtLabel;
+   struct tgsi_instruction_ext_texture InstructionExtTexture;
+   struct tgsi_full_dst_register       FullDstRegisters[TGSI_FULL_MAX_DST_REGISTERS];
+   struct tgsi_full_src_register       FullSrcRegisters[TGSI_FULL_MAX_SRC_REGISTERS];
+};
+
+union tgsi_full_token
+{
+   struct tgsi_token             Token;
+   struct tgsi_full_declaration  FullDeclaration;
+   struct tgsi_full_immediate    FullImmediate;
+   struct tgsi_full_instruction  FullInstruction;
+};
+
+void
+tgsi_full_token_init(
+   union tgsi_full_token *full_token );
+
+void
+tgsi_full_token_free(
+   union tgsi_full_token *full_token );
+
+struct tgsi_parse_context
+{
+   const struct tgsi_token    *Tokens;
+   unsigned                   Position;
+   struct tgsi_full_version   FullVersion;
+   struct tgsi_full_header    FullHeader;
+   union tgsi_full_token      FullToken;
+};
+
+#define TGSI_PARSE_OK      0
+#define TGSI_PARSE_ERROR   1
+
+unsigned
+tgsi_parse_init(
+   struct tgsi_parse_context *ctx,
+   const struct tgsi_token *tokens );
+
+void
+tgsi_parse_free(
+   struct tgsi_parse_context *ctx );
+
+boolean
+tgsi_parse_end_of_tokens(
+   struct tgsi_parse_context *ctx );
+
+void
+tgsi_parse_token(
+   struct tgsi_parse_context *ctx );
+
+unsigned
+tgsi_num_tokens(const struct tgsi_token *tokens);
+
+struct tgsi_token *
+tgsi_dup_tokens(const struct tgsi_token *tokens);
+
+#if defined __cplusplus
+}
+#endif
+
+#endif /* TGSI_PARSE_H */
+
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ppc.c b/src/gallium/auxiliary/tgsi/tgsi_ppc.c
new file mode 100644
index 00000000000..f365030e526
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_ppc.c
@@ -0,0 +1,1363 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * TGSI to PowerPC code generation.
+ */
+
+#include "pipe/p_config.h"
+
+#if defined(PIPE_ARCH_PPC)
+
+#include "pipe/p_debug.h"
+#include "pipe/p_shader_tokens.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "util/u_sse.h"
+#include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_util.h"
+#include "tgsi_dump.h"
+#include "tgsi_exec.h"
+#include "tgsi_ppc.h"
+#include "rtasm/rtasm_ppc.h"
+
+
+/**
+ * Since it's pretty much impossible to form PPC vector immediates, load
+ * them from memory here:
+ */
+const float ppc_builtin_constants[] ALIGN16_ATTRIB = {
+   1.0f, -128.0f, 128.0, 0.0
+};
+
+
+#define FOR_EACH_CHANNEL( CHAN )\
+   for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++)
+
+#define IS_DST0_CHANNEL_ENABLED( INST, CHAN )\
+   ((INST).FullDstRegisters[0].DstRegister.WriteMask & (1 << (CHAN)))
+
+#define IF_IS_DST0_CHANNEL_ENABLED( INST, CHAN )\
+   if (IS_DST0_CHANNEL_ENABLED( INST, CHAN ))
+
+#define FOR_EACH_DST0_ENABLED_CHANNEL( INST, CHAN )\
+   FOR_EACH_CHANNEL( CHAN )\
+      IF_IS_DST0_CHANNEL_ENABLED( INST, CHAN )
+
+#define CHAN_X 0
+#define CHAN_Y 1
+#define CHAN_Z 2
+#define CHAN_W 3
+
+
+/**
+ * How many TGSI temps should be implemented with real PPC vector registers
+ * rather than memory.
+ */
+#define MAX_PPC_TEMPS 3
+
+
+/**
+ * Context/state used during code gen.
+ */
+struct gen_context
+{
+   struct ppc_function *f;
+   int inputs_reg;    /**< GP register pointing to input params */
+   int outputs_reg;   /**< GP register pointing to output params */
+   int temps_reg;     /**< GP register pointing to temporary "registers" */
+   int immed_reg;     /**< GP register pointing to immediates buffer */
+   int const_reg;     /**< GP register pointing to constants buffer */
+   int builtins_reg;  /**< GP register pointint to built-in constants */
+
+   int offset_reg;    /**< used to reduce redundant li instructions */
+   int offset_value;
+
+   int one_vec;       /**< vector register with {1.0, 1.0, 1.0, 1.0} */
+   int bit31_vec;     /**< vector register with {1<<31, 1<<31, 1<<31, 1<<31} */
+
+   /**
+    * Map TGSI temps to PPC vector temps.
+    * We have 32 PPC vector regs.  Use 16 of them for storing 4 TGSI temps.
+    * XXX currently only do this for TGSI temps [0..MAX_PPC_TEMPS-1].
+    */
+   int temps_map[MAX_PPC_TEMPS][4];
+
+   /**
+    * Cache of src registers.
+    * This is used to avoid redundant load instructions.
+    */
+   struct {
+      struct tgsi_full_src_register src;
+      uint chan;
+      uint vec;
+   } regs[12];  /* 3 src regs, 4 channels */
+   uint num_regs;
+};
+
+
+/**
+ * Initialize code generation context.
+ */
+static void
+init_gen_context(struct gen_context *gen, struct ppc_function *func)
+{
+   uint i;
+
+   memset(gen, 0, sizeof(*gen));
+   gen->f = func;
+   gen->inputs_reg = ppc_reserve_register(func, 3);   /* first function param */
+   gen->outputs_reg = ppc_reserve_register(func, 4);  /* second function param */
+   gen->temps_reg = ppc_reserve_register(func, 5);    /* ... */
+   gen->immed_reg = ppc_reserve_register(func, 6);
+   gen->const_reg = ppc_reserve_register(func, 7);
+   gen->builtins_reg = ppc_reserve_register(func, 8);
+   gen->one_vec = -1;
+   gen->bit31_vec = -1;
+   gen->offset_reg = -1;
+   gen->offset_value = -9999999;
+   for (i = 0; i < MAX_PPC_TEMPS; i++) {
+      gen->temps_map[i][0] = ppc_allocate_vec_register(gen->f);
+      gen->temps_map[i][1] = ppc_allocate_vec_register(gen->f);
+      gen->temps_map[i][2] = ppc_allocate_vec_register(gen->f);
+      gen->temps_map[i][3] = ppc_allocate_vec_register(gen->f);
+   }
+}
+
+
+/**
+ * Is the given TGSI register stored as a real PPC vector register?
+ */
+static boolean
+is_ppc_vec_temporary(const struct tgsi_full_src_register *reg)
+{
+   return (reg->SrcRegister.File == TGSI_FILE_TEMPORARY &&
+           reg->SrcRegister.Index < MAX_PPC_TEMPS);
+}
+
+
+/**
+ * Is the given TGSI register stored as a real PPC vector register?
+ */
+static boolean
+is_ppc_vec_temporary_dst(const struct tgsi_full_dst_register *reg)
+{
+   return (reg->DstRegister.File == TGSI_FILE_TEMPORARY &&
+           reg->DstRegister.Index < MAX_PPC_TEMPS);
+}
+
+
+
+/**
+ * All PPC vector load/store instructions form an effective address
+ * by adding the contents of two registers.  For example:
+ *    lvx v2,r8,r9   # v2 = memory[r8 + r9]
+ *    stvx v2,r8,r9  # memory[r8 + r9] = v2;
+ * So our lvx/stvx instructions are typically preceded by an 'li' instruction
+ * to load r9 (above) with an immediate (an offset).
+ * This code emits that 'li' instruction, but only if the offset value is
+ * different than the previous 'li'.
+ * This optimization seems to save about 10% in the instruction count.
+ * Note that we need to unconditionally emit an 'li' inside basic blocks
+ * (such as inside loops).
+ */
+static int
+emit_li_offset(struct gen_context *gen, int offset)
+{
+   if (gen->offset_reg <= 0) {
+      /* allocate a GP register for storing load/store offset */
+      gen->offset_reg = ppc_allocate_register(gen->f);
+   }
+
+   /* emit new 'li' if offset is changing */
+   if (gen->offset_value < 0 || gen->offset_value != offset) {
+      gen->offset_value = offset;
+      ppc_li(gen->f, gen->offset_reg, offset);
+   }
+
+   return gen->offset_reg;
+}
+
+
+/**
+ * Forces subsequent emit_li_offset() calls to emit an 'li'.
+ * To be called at the top of basic blocks.
+ */
+static void
+reset_li_offset(struct gen_context *gen)
+{
+   gen->offset_value = -9999999;
+}
+
+
+
+/**
+ * Load the given vector register with {value, value, value, value}.
+ * The value must be in the ppu_builtin_constants[] array.
+ * We wouldn't need this if there was a simple way to load PPC vector
+ * registers with immediate values!
+ */
+static void
+load_constant_vec(struct gen_context *gen, int dst_vec, float value)
+{
+   uint pos;
+   for (pos = 0; pos < Elements(ppc_builtin_constants); pos++) {
+      if (ppc_builtin_constants[pos] == value) {
+         int offset = pos * 4;
+         int offset_reg = emit_li_offset(gen, offset);
+
+         /* Load 4-byte word into vector register.
+          * The vector slot depends on the effective address we load from.
+          * We know that our builtins start at a 16-byte boundary so we
+          * know that 'swizzle' tells us which vector slot will have the
+          * loaded word.  The other vector slots will be undefined.
+          */
+         ppc_lvewx(gen->f, dst_vec, gen->builtins_reg, offset_reg);
+         /* splat word[pos % 4] across the vector reg */
+         ppc_vspltw(gen->f, dst_vec, dst_vec, pos % 4);
+         return;
+      }
+   }
+   assert(0 && "Need to add new constant to ppc_builtin_constants array");
+}
+
+
+/**
+ * Return index of vector register containing {1.0, 1.0, 1.0, 1.0}.
+ */
+static int
+gen_one_vec(struct gen_context *gen)
+{
+   if (gen->one_vec < 0) {
+      gen->one_vec = ppc_allocate_vec_register(gen->f);
+      load_constant_vec(gen, gen->one_vec, 1.0f);
+   }
+   return gen->one_vec;
+}
+
+/**
+ * Return index of vector register containing {1<<31, 1<<31, 1<<31, 1<<31}.
+ */
+static int
+gen_get_bit31_vec(struct gen_context *gen)
+{
+   if (gen->bit31_vec < 0) {
+      gen->bit31_vec = ppc_allocate_vec_register(gen->f);
+      ppc_vspltisw(gen->f, gen->bit31_vec, -1);
+      ppc_vslw(gen->f, gen->bit31_vec, gen->bit31_vec, gen->bit31_vec);
+   }
+   return gen->bit31_vec;
+}
+
+
+/**
+ * Register fetch.  Return PPC vector register with result.
+ */
+static int
+emit_fetch(struct gen_context *gen,
+           const struct tgsi_full_src_register *reg,
+           const unsigned chan_index)
+{
+   uint swizzle = tgsi_util_get_full_src_register_extswizzle(reg, chan_index);
+   int dst_vec = -1;
+
+   switch (swizzle) {
+   case TGSI_EXTSWIZZLE_X:
+   case TGSI_EXTSWIZZLE_Y:
+   case TGSI_EXTSWIZZLE_Z:
+   case TGSI_EXTSWIZZLE_W:
+      switch (reg->SrcRegister.File) {
+      case TGSI_FILE_INPUT:
+         {
+            int offset = (reg->SrcRegister.Index * 4 + swizzle) * 16;
+            int offset_reg = emit_li_offset(gen, offset);
+            dst_vec = ppc_allocate_vec_register(gen->f);
+            ppc_lvx(gen->f, dst_vec, gen->inputs_reg, offset_reg);
+         }
+         break;
+      case TGSI_FILE_TEMPORARY:
+         if (is_ppc_vec_temporary(reg)) {
+            /* use PPC vec register */
+            dst_vec = gen->temps_map[reg->SrcRegister.Index][swizzle];
+         }
+         else {
+            /* use memory-based temp register "file" */
+            int offset = (reg->SrcRegister.Index * 4 + swizzle) * 16;
+            int offset_reg = emit_li_offset(gen, offset);
+            dst_vec = ppc_allocate_vec_register(gen->f);
+            ppc_lvx(gen->f, dst_vec, gen->temps_reg, offset_reg);
+         }
+         break;
+      case TGSI_FILE_IMMEDIATE:
+         {
+            int offset = (reg->SrcRegister.Index * 4 + swizzle) * 4;
+            int offset_reg = emit_li_offset(gen, offset);
+            dst_vec = ppc_allocate_vec_register(gen->f);
+            /* Load 4-byte word into vector register.
+             * The vector slot depends on the effective address we load from.
+             * We know that our immediates start at a 16-byte boundary so we
+             * know that 'swizzle' tells us which vector slot will have the
+             * loaded word.  The other vector slots will be undefined.
+             */
+            ppc_lvewx(gen->f, dst_vec, gen->immed_reg, offset_reg);
+            /* splat word[swizzle] across the vector reg */
+            ppc_vspltw(gen->f, dst_vec, dst_vec, swizzle);
+         }
+         break;
+      case TGSI_FILE_CONSTANT:
+         {
+            int offset = (reg->SrcRegister.Index * 4 + swizzle) * 4;
+            int offset_reg = emit_li_offset(gen, offset);
+            dst_vec = ppc_allocate_vec_register(gen->f);
+            /* Load 4-byte word into vector register.
+             * The vector slot depends on the effective address we load from.
+             * We know that our constants start at a 16-byte boundary so we
+             * know that 'swizzle' tells us which vector slot will have the
+             * loaded word.  The other vector slots will be undefined.
+             */
+            ppc_lvewx(gen->f, dst_vec, gen->const_reg, offset_reg);
+            /* splat word[swizzle] across the vector reg */
+            ppc_vspltw(gen->f, dst_vec, dst_vec, swizzle);
+         }
+         break;
+      default:
+         assert( 0 );
+      }
+      break;
+   case TGSI_EXTSWIZZLE_ZERO:
+      ppc_vzero(gen->f, dst_vec);
+      break;
+   case TGSI_EXTSWIZZLE_ONE:
+      {
+         int one_vec = gen_one_vec(gen);
+         dst_vec = ppc_allocate_vec_register(gen->f);
+         ppc_vmove(gen->f, dst_vec, one_vec);
+      }
+      break;
+   default:
+      assert( 0 );
+   }
+
+   assert(dst_vec >= 0);
+
+   {
+      uint sign_op = tgsi_util_get_full_src_register_sign_mode(reg, chan_index);
+      if (sign_op != TGSI_UTIL_SIGN_KEEP) {
+         int bit31_vec = gen_get_bit31_vec(gen);
+         int dst_vec2;
+
+         if (is_ppc_vec_temporary(reg)) {
+            /* need to use a new temp */
+            dst_vec2 = ppc_allocate_vec_register(gen->f);
+         }
+         else {
+            dst_vec2 = dst_vec;
+         }
+
+         switch (sign_op) {
+         case TGSI_UTIL_SIGN_CLEAR:
+            /* vec = vec & ~bit31 */
+            ppc_vandc(gen->f, dst_vec2, dst_vec, bit31_vec);
+            break;
+         case TGSI_UTIL_SIGN_SET:
+            /* vec = vec | bit31 */
+            ppc_vor(gen->f, dst_vec2, dst_vec, bit31_vec);
+            break;
+         case TGSI_UTIL_SIGN_TOGGLE:
+            /* vec = vec ^ bit31 */
+            ppc_vxor(gen->f, dst_vec2, dst_vec, bit31_vec);
+            break;
+         default:
+            assert(0);
+         }
+         return dst_vec2;
+      }
+   }
+
+   return dst_vec;
+}
+
+
+
+/**
+ * Test if two TGSI src registers refer to the same memory location.
+ * We use this to avoid redundant register loads.
+ */
+static boolean
+equal_src_locs(const struct tgsi_full_src_register *a, uint chan_a,
+               const struct tgsi_full_src_register *b, uint chan_b)
+{
+   int swz_a, swz_b;
+   int sign_a, sign_b;
+   if (a->SrcRegister.File != b->SrcRegister.File)
+      return FALSE;
+   if (a->SrcRegister.Index != b->SrcRegister.Index)
+      return FALSE;
+   swz_a = tgsi_util_get_full_src_register_extswizzle(a, chan_a);
+   swz_b = tgsi_util_get_full_src_register_extswizzle(b, chan_b);
+   if (swz_a != swz_b)
+      return FALSE;
+   sign_a = tgsi_util_get_full_src_register_sign_mode(a, chan_a);
+   sign_b = tgsi_util_get_full_src_register_sign_mode(b, chan_b);
+   if (sign_a != sign_b)
+      return FALSE;
+   return TRUE;
+}
+
+
+/**
+ * Given a TGSI src register and channel index, return the PPC vector
+ * register containing the value.  We use a cache to prevent re-loading
+ * the same register multiple times.
+ * \return index of PPC vector register with the desired src operand
+ */
+static int
+get_src_vec(struct gen_context *gen,
+            struct tgsi_full_instruction *inst, int src_reg, uint chan)
+{
+   const const struct tgsi_full_src_register *src = 
+      &inst->FullSrcRegisters[src_reg];
+   int vec;
+   uint i;
+
+   /* check the cache */
+   for (i = 0; i < gen->num_regs; i++) {
+      if (equal_src_locs(&gen->regs[i].src, gen->regs[i].chan, src, chan)) {
+         /* cache hit */
+         assert(gen->regs[i].vec >= 0);
+         return gen->regs[i].vec;
+      }
+   }
+
+   /* cache miss: allocate new vec reg and emit fetch/load code */
+   vec = emit_fetch(gen, src, chan);
+   gen->regs[gen->num_regs].src = *src;
+   gen->regs[gen->num_regs].chan = chan;
+   gen->regs[gen->num_regs].vec = vec;
+   gen->num_regs++;
+
+   assert(gen->num_regs <= Elements(gen->regs));
+
+   assert(vec >= 0);
+
+   return vec;
+}
+
+
+/**
+ * Clear the src operand cache.  To be called at the end of each emit function.
+ */
+static void
+release_src_vecs(struct gen_context *gen)
+{
+   uint i;
+   for (i = 0; i < gen->num_regs; i++) {
+      const const struct tgsi_full_src_register src = gen->regs[i].src;
+      if (!is_ppc_vec_temporary(&src)) {
+         ppc_release_vec_register(gen->f, gen->regs[i].vec);
+      }
+   }
+   gen->num_regs = 0;
+}
+
+
+
+static int
+get_dst_vec(struct gen_context *gen, 
+            const struct tgsi_full_instruction *inst,
+            unsigned chan_index)
+{
+   const struct tgsi_full_dst_register *reg = &inst->FullDstRegisters[0];
+
+   if (is_ppc_vec_temporary_dst(reg)) {
+      int vec = gen->temps_map[reg->DstRegister.Index][chan_index];
+      return vec;
+   }
+   else {
+      return ppc_allocate_vec_register(gen->f);
+   }
+}
+
+
+/**
+ * Register store.  Store 'src_vec' at location indicated by 'reg'.
+ * \param free_vec  Should the src_vec be released when done?
+ */
+static void
+emit_store(struct gen_context *gen,
+           int src_vec,
+           const struct tgsi_full_instruction *inst,
+           unsigned chan_index,
+           boolean free_vec)
+{
+   const struct tgsi_full_dst_register *reg = &inst->FullDstRegisters[0];
+
+   switch (reg->DstRegister.File) {
+   case TGSI_FILE_OUTPUT:
+      {
+         int offset = (reg->DstRegister.Index * 4 + chan_index) * 16;
+         int offset_reg = emit_li_offset(gen, offset);
+         ppc_stvx(gen->f, src_vec, gen->outputs_reg, offset_reg);
+      }
+      break;
+   case TGSI_FILE_TEMPORARY:
+      if (is_ppc_vec_temporary_dst(reg)) {
+         if (!free_vec) {
+            int dst_vec = gen->temps_map[reg->DstRegister.Index][chan_index];
+            if (dst_vec != src_vec)
+               ppc_vmove(gen->f, dst_vec, src_vec);
+         }
+         free_vec = FALSE;
+      }
+      else {
+         int offset = (reg->DstRegister.Index * 4 + chan_index) * 16;
+         int offset_reg = emit_li_offset(gen, offset);
+         ppc_stvx(gen->f, src_vec, gen->temps_reg, offset_reg);
+      }
+      break;
+#if 0
+   case TGSI_FILE_ADDRESS:
+      emit_addrs(
+         func,
+         xmm,
+         reg->DstRegister.Index,
+         chan_index );
+      break;
+#endif
+   default:
+      assert( 0 );
+   }
+
+#if 0
+   switch( inst->Instruction.Saturate ) {
+   case TGSI_SAT_NONE:
+      break;
+
+   case TGSI_SAT_ZERO_ONE:
+      /* assert( 0 ); */
+      break;
+
+   case TGSI_SAT_MINUS_PLUS_ONE:
+      assert( 0 );
+      break;
+   }
+#endif
+
+   if (free_vec)
+      ppc_release_vec_register(gen->f, src_vec);
+}
+
+
+static void
+emit_scalar_unaryop(struct gen_context *gen, struct tgsi_full_instruction *inst)
+{
+   int v0, v1;
+   uint chan_index;
+
+   v0 = get_src_vec(gen, inst, 0, CHAN_X);
+   v1 = ppc_allocate_vec_register(gen->f);
+
+   switch (inst->Instruction.Opcode) {
+   case TGSI_OPCODE_RSQ:
+      /* v1 = 1.0 / sqrt(v0) */
+      ppc_vrsqrtefp(gen->f, v1, v0);
+      break;
+   case TGSI_OPCODE_RCP:
+      /* v1 = 1.0 / v0 */
+      ppc_vrefp(gen->f, v1, v0);
+      break;
+   default:
+      assert(0);
+   }
+
+   FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+      emit_store(gen, v1, inst, chan_index, FALSE);
+   }
+
+   release_src_vecs(gen);
+   ppc_release_vec_register(gen->f, v1);
+}
+
+
+static void
+emit_unaryop(struct gen_context *gen, struct tgsi_full_instruction *inst)
+{
+   uint chan_index;
+
+   FOR_EACH_DST0_ENABLED_CHANNEL(*inst, chan_index) {
+      int v0 = get_src_vec(gen, inst, 0, chan_index);   /* v0 = srcreg[0] */
+      int v1 = get_dst_vec(gen, inst, chan_index);
+      switch (inst->Instruction.Opcode) {
+      case TGSI_OPCODE_ABS:
+         /* turn off the most significant bit of each vector float word */
+         {
+            int bit31_vec = gen_get_bit31_vec(gen);
+            ppc_vandc(gen->f, v1, v0, bit31_vec); /* v1 = v0 & ~bit31 */
+         }
+         break;
+      case TGSI_OPCODE_FLOOR:
+         ppc_vrfim(gen->f, v1, v0);         /* v1 = floor(v0) */
+         break;
+      case TGSI_OPCODE_FRAC:
+         ppc_vrfim(gen->f, v1, v0);      /* tmp = floor(v0) */
+         ppc_vsubfp(gen->f, v1, v0, v1); /* v1 = v0 - v1 */
+         break;
+      case TGSI_OPCODE_EXPBASE2:
+         ppc_vexptefp(gen->f, v1, v0);     /* v1 = 2^v0 */
+         break;
+      case TGSI_OPCODE_LOGBASE2:
+         /* XXX this may be broken! */
+         ppc_vlogefp(gen->f, v1, v0);      /* v1 = log2(v0) */
+         break;
+      case TGSI_OPCODE_MOV:
+      case TGSI_OPCODE_SWZ:
+         if (v0 != v1)
+            ppc_vmove(gen->f, v1, v0);
+         break;
+      default:
+         assert(0);
+      }
+      emit_store(gen, v1, inst, chan_index, TRUE);  /* store v0 */
+   }
+
+   release_src_vecs(gen);
+}
+
+
+static void
+emit_binop(struct gen_context *gen, struct tgsi_full_instruction *inst)
+{
+   int zero_vec = -1;
+   uint chan;
+
+   if (inst->Instruction.Opcode == TGSI_OPCODE_MUL) {
+      zero_vec = ppc_allocate_vec_register(gen->f);
+      ppc_vzero(gen->f, zero_vec);
+   }
+
+   FOR_EACH_DST0_ENABLED_CHANNEL(*inst, chan) {
+      /* fetch src operands */
+      int v0 = get_src_vec(gen, inst, 0, chan);
+      int v1 = get_src_vec(gen, inst, 1, chan);
+      int v2 = get_dst_vec(gen, inst, chan);
+
+      /* emit binop */
+      switch (inst->Instruction.Opcode) {
+      case TGSI_OPCODE_ADD:
+         ppc_vaddfp(gen->f, v2, v0, v1);
+         break;
+      case TGSI_OPCODE_SUB:
+         ppc_vsubfp(gen->f, v2, v0, v1);
+         break;
+      case TGSI_OPCODE_MUL:
+         ppc_vmaddfp(gen->f, v2, v0, v1, zero_vec);
+         break;
+      case TGSI_OPCODE_MIN:
+         ppc_vminfp(gen->f, v2, v0, v1);
+         break;
+      case TGSI_OPCODE_MAX:
+         ppc_vmaxfp(gen->f, v2, v0, v1);
+         break;
+      default:
+         assert(0);
+      }
+
+      /* store v2 */
+      emit_store(gen, v2, inst, chan, TRUE);
+   }
+
+   if (inst->Instruction.Opcode == TGSI_OPCODE_MUL)
+      ppc_release_vec_register(gen->f, zero_vec);
+
+   release_src_vecs(gen);
+}
+
+
+static void
+emit_triop(struct gen_context *gen, struct tgsi_full_instruction *inst)
+{
+   uint chan;
+
+   FOR_EACH_DST0_ENABLED_CHANNEL(*inst, chan) {
+      /* fetch src operands */
+      int v0 = get_src_vec(gen, inst, 0, chan);
+      int v1 = get_src_vec(gen, inst, 1, chan);
+      int v2 = get_src_vec(gen, inst, 2, chan);
+      int v3 = get_dst_vec(gen, inst, chan);
+
+      /* emit ALU */
+      switch (inst->Instruction.Opcode) {
+      case TGSI_OPCODE_MAD:
+         ppc_vmaddfp(gen->f, v3, v0, v1, v2);   /* v3 = v0 * v1 + v2 */
+         break;
+      case TGSI_OPCODE_LRP:
+         ppc_vsubfp(gen->f, v3, v1, v2);        /* v3 = v1 - v2 */
+         ppc_vmaddfp(gen->f, v3, v0, v3, v2);   /* v3 = v0 * v3 + v2 */
+         break;
+      default:
+         assert(0);
+      }
+
+      /* store v3 */
+      emit_store(gen, v3, inst, chan, TRUE);
+   }
+
+   release_src_vecs(gen);
+}
+
+
+/**
+ * Vector comparisons, resulting in 1.0 or 0.0 values.
+ */
+static void
+emit_inequality(struct gen_context *gen, struct tgsi_full_instruction *inst)
+{
+   uint chan;
+   int one_vec = gen_one_vec(gen);
+
+   FOR_EACH_DST0_ENABLED_CHANNEL(*inst, chan) {
+      /* fetch src operands */
+      int v0 = get_src_vec(gen, inst, 0, chan);
+      int v1 = get_src_vec(gen, inst, 1, chan);
+      int v2 = get_dst_vec(gen, inst, chan);
+      boolean complement = FALSE;
+
+      switch (inst->Instruction.Opcode) {
+      case TGSI_OPCODE_SNE:
+         complement = TRUE;
+         /* fall-through */
+      case TGSI_OPCODE_SEQ:
+         ppc_vcmpeqfpx(gen->f, v2, v0, v1); /* v2 = v0 == v1 ? ~0 : 0 */
+         break;
+
+      case TGSI_OPCODE_SGE:
+         complement = TRUE;
+         /* fall-through */
+      case TGSI_OPCODE_SLT:
+         ppc_vcmpgtfpx(gen->f, v2, v1, v0); /* v2 = v1 > v0 ? ~0 : 0 */
+         break;
+
+      case TGSI_OPCODE_SLE:
+         complement = TRUE;
+         /* fall-through */
+      case TGSI_OPCODE_SGT:
+         ppc_vcmpgtfpx(gen->f, v2, v0, v1); /* v2 = v0 > v1 ? ~0 : 0 */
+         break;
+      default:
+         assert(0);
+      }
+
+      /* v2 is now {0,0,0,0} or {~0,~0,~0,~0} */
+
+      if (complement)
+         ppc_vandc(gen->f, v2, one_vec, v2);    /* v2 = one_vec & ~v2 */
+      else
+         ppc_vand(gen->f, v2, one_vec, v2);     /* v2 = one_vec & v2 */
+
+      /* store v2 */
+      emit_store(gen, v2, inst, chan, TRUE);
+   }
+
+   release_src_vecs(gen);
+}
+
+
+static void
+emit_dotprod(struct gen_context *gen, struct tgsi_full_instruction *inst)
+{
+   int v0, v1, v2;
+   uint chan_index;
+
+   v2 = ppc_allocate_vec_register(gen->f);
+
+   ppc_vzero(gen->f, v2);                  /* v2 = {0, 0, 0, 0} */
+
+   v0 = get_src_vec(gen, inst, 0, CHAN_X); /* v0 = src0.XXXX */
+   v1 = get_src_vec(gen, inst, 1, CHAN_X); /* v1 = src1.XXXX */
+   ppc_vmaddfp(gen->f, v2, v0, v1, v2);    /* v2 = v0 * v1 + v2 */
+
+   v0 = get_src_vec(gen, inst, 0, CHAN_Y); /* v0 = src0.YYYY */
+   v1 = get_src_vec(gen, inst, 1, CHAN_Y); /* v1 = src1.YYYY */
+   ppc_vmaddfp(gen->f, v2, v0, v1, v2);    /* v2 = v0 * v1 + v2 */
+
+   v0 = get_src_vec(gen, inst, 0, CHAN_Z); /* v0 = src0.ZZZZ */
+   v1 = get_src_vec(gen, inst, 1, CHAN_Z); /* v1 = src1.ZZZZ */
+   ppc_vmaddfp(gen->f, v2, v0, v1, v2);    /* v2 = v0 * v1 + v2 */
+
+   if (inst->Instruction.Opcode == TGSI_OPCODE_DP4) {
+      v0 = get_src_vec(gen, inst, 0, CHAN_W); /* v0 = src0.WWWW */
+      v1 = get_src_vec(gen, inst, 1, CHAN_W); /* v1 = src1.WWWW */
+      ppc_vmaddfp(gen->f, v2, v0, v1, v2);    /* v2 = v0 * v1 + v2 */
+   }
+   else if (inst->Instruction.Opcode == TGSI_OPCODE_DPH) {
+      v1 = get_src_vec(gen, inst, 1, CHAN_W); /* v1 = src1.WWWW */
+      ppc_vaddfp(gen->f, v2, v2, v1);         /* v2 = v2 + v1 */
+   }
+
+   FOR_EACH_DST0_ENABLED_CHANNEL(*inst, chan_index) {
+      emit_store(gen, v2, inst, chan_index, FALSE);  /* store v2, free v2 later */
+   }
+
+   release_src_vecs(gen);
+
+   ppc_release_vec_register(gen->f, v2);
+}
+
+
+/** Approximation for vr = pow(va, vb) */
+static void
+ppc_vec_pow(struct ppc_function *f, int vr, int va, int vb)
+{
+   /* pow(a,b) ~= exp2(log2(a) * b) */
+   int t_vec = ppc_allocate_vec_register(f);
+   int zero_vec = ppc_allocate_vec_register(f);
+
+   ppc_vzero(f, zero_vec);
+
+   ppc_vlogefp(f, t_vec, va);                   /* t = log2(va) */
+   ppc_vmaddfp(f, t_vec, t_vec, vb, zero_vec);  /* t = t * vb + zero */
+   ppc_vexptefp(f, vr, t_vec);                  /* vr = 2^t */
+
+   ppc_release_vec_register(f, t_vec);
+   ppc_release_vec_register(f, zero_vec);
+}
+
+
+static void
+emit_lit(struct gen_context *gen, struct tgsi_full_instruction *inst)
+{
+   int one_vec = gen_one_vec(gen);
+
+   /* Compute X */
+   if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_X)) {
+      emit_store(gen, one_vec, inst, CHAN_X, FALSE);
+   }
+
+   /* Compute Y, Z */
+   if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Y) ||
+       IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Z)) {
+      int x_vec;
+      int zero_vec = ppc_allocate_vec_register(gen->f);
+
+      x_vec = get_src_vec(gen, inst, 0, CHAN_X);  /* x_vec = src[0].x */
+
+      ppc_vzero(gen->f, zero_vec);                /* zero = {0,0,0,0} */
+      ppc_vmaxfp(gen->f, x_vec, x_vec, zero_vec); /* x_vec = max(x_vec, 0) */
+
+      if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Y)) {
+         emit_store(gen, x_vec, inst, CHAN_Y, FALSE);
+      }
+
+      if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Z)) {
+         int y_vec, w_vec;
+         int z_vec = ppc_allocate_vec_register(gen->f);
+         int pow_vec = ppc_allocate_vec_register(gen->f);
+         int pos_vec = ppc_allocate_vec_register(gen->f);
+         int p128_vec = ppc_allocate_vec_register(gen->f);
+         int n128_vec = ppc_allocate_vec_register(gen->f);
+
+         y_vec = get_src_vec(gen, inst, 0, CHAN_Y);  /* y_vec = src[0].y */
+         ppc_vmaxfp(gen->f, y_vec, y_vec, zero_vec); /* y_vec = max(y_vec, 0) */
+
+         w_vec = get_src_vec(gen, inst, 0, CHAN_W);  /* w_vec = src[0].w */
+
+         /* clamp W to [-128, 128] */
+         load_constant_vec(gen, p128_vec, 128.0f);
+         load_constant_vec(gen, n128_vec, -128.0f);
+         ppc_vmaxfp(gen->f, w_vec, w_vec, n128_vec); /* w = max(w, -128) */
+         ppc_vminfp(gen->f, w_vec, w_vec, p128_vec); /* w = min(w, 128) */
+
+         /* if temp.x > 0
+          *    z = pow(tmp.y, tmp.w)
+          * else
+          *    z = 0.0
+          */
+         ppc_vec_pow(gen->f, pow_vec, y_vec, w_vec);      /* pow = pow(y, w) */
+         ppc_vcmpgtfpx(gen->f, pos_vec, x_vec, zero_vec); /* pos = x > 0 */
+         ppc_vand(gen->f, z_vec, pow_vec, pos_vec);       /* z = pow & pos */
+
+         emit_store(gen, z_vec, inst, CHAN_Z, FALSE);
+
+         ppc_release_vec_register(gen->f, z_vec);
+         ppc_release_vec_register(gen->f, pow_vec);
+         ppc_release_vec_register(gen->f, pos_vec);
+         ppc_release_vec_register(gen->f, p128_vec);
+         ppc_release_vec_register(gen->f, n128_vec);
+      }
+
+      ppc_release_vec_register(gen->f, zero_vec);
+   }
+
+   /* Compute W */
+   if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_W)) {
+      emit_store(gen, one_vec, inst, CHAN_W, FALSE);
+   }
+
+   release_src_vecs(gen);
+}
+
+
+static void
+emit_exp(struct gen_context *gen, struct tgsi_full_instruction *inst)
+{
+   const int one_vec = gen_one_vec(gen);
+   int src_vec;
+
+   /* get src arg */
+   src_vec = get_src_vec(gen, inst, 0, CHAN_X);
+
+   /* Compute X = 2^floor(src) */
+   if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_X)) {
+      int dst_vec = get_dst_vec(gen, inst, CHAN_X);
+      int tmp_vec = ppc_allocate_vec_register(gen->f);
+      ppc_vrfim(gen->f, tmp_vec, src_vec);             /* tmp = floor(src); */
+      ppc_vexptefp(gen->f, dst_vec, tmp_vec);          /* dst = 2 ^ tmp */
+      emit_store(gen, dst_vec, inst, CHAN_X, TRUE);
+      ppc_release_vec_register(gen->f, tmp_vec);
+   }
+
+   /* Compute Y = src - floor(src) */
+   if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Y)) {
+      int dst_vec = get_dst_vec(gen, inst, CHAN_Y);
+      int tmp_vec = ppc_allocate_vec_register(gen->f);
+      ppc_vrfim(gen->f, tmp_vec, src_vec);             /* tmp = floor(src); */
+      ppc_vsubfp(gen->f, dst_vec, src_vec, tmp_vec);   /* dst = src - tmp */
+      emit_store(gen, dst_vec, inst, CHAN_Y, TRUE);
+      ppc_release_vec_register(gen->f, tmp_vec);
+   }
+
+   /* Compute Z = RoughApprox2ToX(src) */
+   if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Z)) {
+      int dst_vec = get_dst_vec(gen, inst, CHAN_Z);
+      ppc_vexptefp(gen->f, dst_vec, src_vec);          /* dst = 2 ^ src */
+      emit_store(gen, dst_vec, inst, CHAN_Z, TRUE);
+   }
+
+   /* Compute W = 1.0 */
+   if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_W)) {
+      emit_store(gen, one_vec, inst, CHAN_W, FALSE);
+   }
+
+   release_src_vecs(gen);
+}
+
+
+static void
+emit_log(struct gen_context *gen, struct tgsi_full_instruction *inst)
+{
+   const int bit31_vec = gen_get_bit31_vec(gen);
+   const int one_vec = gen_one_vec(gen);
+   int src_vec, abs_vec;
+
+   /* get src arg */
+   src_vec = get_src_vec(gen, inst, 0, CHAN_X);
+
+   /* compute abs(src) */
+   abs_vec = ppc_allocate_vec_register(gen->f);
+   ppc_vandc(gen->f, abs_vec, src_vec, bit31_vec);     /* abs = src & ~bit31 */
+
+   if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_X) &&
+       IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Y)) {
+
+      /* compute tmp = floor(log2(abs)) */
+      int tmp_vec = ppc_allocate_vec_register(gen->f);
+      ppc_vlogefp(gen->f, tmp_vec, abs_vec);           /* tmp = log2(abs) */
+      ppc_vrfim(gen->f, tmp_vec, tmp_vec);             /* tmp = floor(tmp); */
+
+      /* Compute X = tmp */
+      if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_X)) {
+         emit_store(gen, tmp_vec, inst, CHAN_X, FALSE);
+      }
+      
+      /* Compute Y = abs / 2^tmp */
+      if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Y)) {
+         const int zero_vec = ppc_allocate_vec_register(gen->f);
+         ppc_vzero(gen->f, zero_vec);
+         ppc_vexptefp(gen->f, tmp_vec, tmp_vec);       /* tmp = 2 ^ tmp */
+         ppc_vrefp(gen->f, tmp_vec, tmp_vec);          /* tmp = 1 / tmp */
+         /* tmp = abs * tmp + zero */
+         ppc_vmaddfp(gen->f, tmp_vec, abs_vec, tmp_vec, zero_vec);
+         emit_store(gen, tmp_vec, inst, CHAN_Y, FALSE);
+         ppc_release_vec_register(gen->f, zero_vec);
+      }
+
+      ppc_release_vec_register(gen->f, tmp_vec);
+   }
+
+   /* Compute Z = RoughApproxLog2(abs) */
+   if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Z)) {
+      int dst_vec = get_dst_vec(gen, inst, CHAN_Z);
+      ppc_vlogefp(gen->f, dst_vec, abs_vec);           /* dst = log2(abs) */
+      emit_store(gen, dst_vec, inst, CHAN_Z, TRUE);
+   }
+
+   /* Compute W = 1.0 */
+   if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_W)) {
+      emit_store(gen, one_vec, inst, CHAN_W, FALSE);
+   }
+
+   ppc_release_vec_register(gen->f, abs_vec);
+   release_src_vecs(gen);
+}
+
+
+static void
+emit_pow(struct gen_context *gen, struct tgsi_full_instruction *inst)
+{
+   int s0_vec = get_src_vec(gen, inst, 0, CHAN_X);
+   int s1_vec = get_src_vec(gen, inst, 1, CHAN_X);
+   int pow_vec = ppc_allocate_vec_register(gen->f);
+   int chan;
+
+   ppc_vec_pow(gen->f, pow_vec, s0_vec, s1_vec);
+
+   FOR_EACH_DST0_ENABLED_CHANNEL(*inst, chan) {
+      emit_store(gen, pow_vec, inst, chan, FALSE);
+   }
+
+   ppc_release_vec_register(gen->f, pow_vec);
+
+   release_src_vecs(gen);
+}
+
+
+static void
+emit_xpd(struct gen_context *gen, struct tgsi_full_instruction *inst)
+{
+   int x0_vec, y0_vec, z0_vec;
+   int x1_vec, y1_vec, z1_vec;
+   int zero_vec, tmp_vec;
+   int tmp2_vec;
+
+   zero_vec = ppc_allocate_vec_register(gen->f);
+   ppc_vzero(gen->f, zero_vec);
+
+   tmp_vec = ppc_allocate_vec_register(gen->f);
+   tmp2_vec = ppc_allocate_vec_register(gen->f);
+
+   if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Y) ||
+       IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Z)) {
+      x0_vec = get_src_vec(gen, inst, 0, CHAN_X);
+      x1_vec = get_src_vec(gen, inst, 1, CHAN_X);
+   }
+   if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_X) ||
+       IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Z)) {
+      y0_vec = get_src_vec(gen, inst, 0, CHAN_Y);
+      y1_vec = get_src_vec(gen, inst, 1, CHAN_Y);
+   }
+   if (IS_DST0_CHANNEL_ENABLED(*inst, CHAN_X) ||
+       IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Y)) {
+      z0_vec = get_src_vec(gen, inst, 0, CHAN_Z);
+      z1_vec = get_src_vec(gen, inst, 1, CHAN_Z);
+   }
+
+   IF_IS_DST0_CHANNEL_ENABLED(*inst, CHAN_X) {
+      /* tmp = y0 * z1 */
+      ppc_vmaddfp(gen->f, tmp_vec, y0_vec, z1_vec, zero_vec);
+      /* tmp = tmp - z0 * y1*/
+      ppc_vnmsubfp(gen->f, tmp_vec, tmp_vec, z0_vec, y1_vec);
+      emit_store(gen, tmp_vec, inst, CHAN_X, FALSE);
+   }
+   IF_IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Y) {
+      /* tmp = z0 * x1 */
+      ppc_vmaddfp(gen->f, tmp_vec, z0_vec, x1_vec, zero_vec);
+      /* tmp = tmp - x0 * z1 */
+      ppc_vnmsubfp(gen->f, tmp_vec, tmp_vec, x0_vec, z1_vec);
+      emit_store(gen, tmp_vec, inst, CHAN_Y, FALSE);
+   }
+   IF_IS_DST0_CHANNEL_ENABLED(*inst, CHAN_Z) {
+      /* tmp = x0 * y1 */
+      ppc_vmaddfp(gen->f, tmp_vec, x0_vec, y1_vec, zero_vec);
+      /* tmp = tmp - y0 * x1 */
+      ppc_vnmsubfp(gen->f, tmp_vec, tmp_vec, y0_vec, x1_vec);
+      emit_store(gen, tmp_vec, inst, CHAN_Z, FALSE);
+   }
+   /* W is undefined */
+
+   ppc_release_vec_register(gen->f, tmp_vec);
+   ppc_release_vec_register(gen->f, zero_vec);
+   release_src_vecs(gen);
+}
+
+static int
+emit_instruction(struct gen_context *gen,
+                 struct tgsi_full_instruction *inst)
+{
+   switch (inst->Instruction.Opcode) {
+   case TGSI_OPCODE_MOV:
+   case TGSI_OPCODE_SWZ:
+   case TGSI_OPCODE_ABS:
+   case TGSI_OPCODE_FLOOR:
+   case TGSI_OPCODE_FRAC:
+   case TGSI_OPCODE_EXPBASE2:
+   case TGSI_OPCODE_LOGBASE2:
+      emit_unaryop(gen, inst);
+      break;
+   case TGSI_OPCODE_RSQ:
+   case TGSI_OPCODE_RCP:
+      emit_scalar_unaryop(gen, inst);
+      break;
+   case TGSI_OPCODE_ADD:
+   case TGSI_OPCODE_SUB:
+   case TGSI_OPCODE_MUL:
+   case TGSI_OPCODE_MIN:
+   case TGSI_OPCODE_MAX:
+      emit_binop(gen, inst);
+      break;
+   case TGSI_OPCODE_SEQ:
+   case TGSI_OPCODE_SNE:
+   case TGSI_OPCODE_SLT:
+   case TGSI_OPCODE_SGT:
+   case TGSI_OPCODE_SLE:
+   case TGSI_OPCODE_SGE:
+      emit_inequality(gen, inst);
+      break;
+   case TGSI_OPCODE_MAD:
+   case TGSI_OPCODE_LRP:
+      emit_triop(gen, inst);
+      break;
+   case TGSI_OPCODE_DP3:
+   case TGSI_OPCODE_DP4:
+   case TGSI_OPCODE_DPH:
+      emit_dotprod(gen, inst);
+      break;
+   case TGSI_OPCODE_LIT:
+      emit_lit(gen, inst);
+      break;
+   case TGSI_OPCODE_LOG:
+      emit_log(gen, inst);
+      break;
+   case TGSI_OPCODE_EXP:
+      emit_exp(gen, inst);
+      break;
+   case TGSI_OPCODE_POW:
+      emit_pow(gen, inst);
+      break;
+   case TGSI_OPCODE_XPD:
+      emit_xpd(gen, inst);
+      break;
+   case TGSI_OPCODE_END:
+      /* normal end */
+      return 1;
+   default:
+      return 0;
+   }
+   return 1;
+}
+
+
+static void
+emit_declaration(
+   struct ppc_function *func,
+   struct tgsi_full_declaration *decl )
+{
+   if( decl->Declaration.File == TGSI_FILE_INPUT ) {
+#if 0
+      unsigned first, last, mask;
+      unsigned i, j;
+
+      first = decl->DeclarationRange.First;
+      last = decl->DeclarationRange.Last;
+      mask = decl->Declaration.UsageMask;
+
+      for( i = first; i <= last; i++ ) {
+         for( j = 0; j < NUM_CHANNELS; j++ ) {
+            if( mask & (1 << j) ) {
+               switch( decl->Declaration.Interpolate ) {
+               case TGSI_INTERPOLATE_CONSTANT:
+                  emit_coef_a0( func, 0, i, j );
+                  emit_inputs( func, 0, i, j );
+                  break;
+
+               case TGSI_INTERPOLATE_LINEAR:
+                  emit_tempf( func, 0, 0, TGSI_SWIZZLE_X );
+                  emit_coef_dadx( func, 1, i, j );
+                  emit_tempf( func, 2, 0, TGSI_SWIZZLE_Y );
+                  emit_coef_dady( func, 3, i, j );
+                  emit_mul( func, 0, 1 );    /* x * dadx */
+                  emit_coef_a0( func, 4, i, j );
+                  emit_mul( func, 2, 3 );    /* y * dady */
+                  emit_add( func, 0, 4 );    /* x * dadx + a0 */
+                  emit_add( func, 0, 2 );    /* x * dadx + y * dady + a0 */
+                  emit_inputs( func, 0, i, j );
+                  break;
+
+               case TGSI_INTERPOLATE_PERSPECTIVE:
+                  emit_tempf( func, 0, 0, TGSI_SWIZZLE_X );
+                  emit_coef_dadx( func, 1, i, j );
+                  emit_tempf( func, 2, 0, TGSI_SWIZZLE_Y );
+                  emit_coef_dady( func, 3, i, j );
+                  emit_mul( func, 0, 1 );    /* x * dadx */
+                  emit_tempf( func, 4, 0, TGSI_SWIZZLE_W );
+                  emit_coef_a0( func, 5, i, j );
+                  emit_rcp( func, 4, 4 );    /* 1.0 / w */
+                  emit_mul( func, 2, 3 );    /* y * dady */
+                  emit_add( func, 0, 5 );    /* x * dadx + a0 */
+                  emit_add( func, 0, 2 );    /* x * dadx + y * dady + a0 */
+                  emit_mul( func, 0, 4 );    /* (x * dadx + y * dady + a0) / w */
+                  emit_inputs( func, 0, i, j );
+                  break;
+
+               default:
+                  assert( 0 );
+		  break;
+               }
+            }
+         }
+      }
+#endif
+   }
+}
+
+
+
+static void
+emit_prologue(struct ppc_function *func)
+{
+   /* XXX set up stack frame */
+}
+
+
+static void
+emit_epilogue(struct ppc_function *func)
+{
+   ppc_comment(func, -4, "Epilogue:");
+   ppc_return(func);
+   /* XXX restore prev stack frame */
+#if 0
+   debug_printf("PPC: Emitted %u instructions\n", func->num_inst);
+#endif
+}
+
+
+
+/**
+ * Translate a TGSI vertex/fragment shader to PPC code.
+ *
+ * \param tokens  the TGSI input shader
+ * \param func  the output PPC code/function
+ * \param immediates  buffer to place immediates, later passed to PPC func
+ * \return TRUE for success, FALSE if translation failed
+ */
+boolean
+tgsi_emit_ppc(const struct tgsi_token *tokens,
+              struct ppc_function *func,
+              float (*immediates)[4],
+              boolean do_swizzles )
+{
+   static int use_ppc_asm = -1;
+   struct tgsi_parse_context parse;
+   /*boolean instruction_phase = FALSE;*/
+   unsigned ok = 1;
+   uint num_immediates = 0;
+   struct gen_context gen;
+   uint ic = 0;
+
+   if (use_ppc_asm < 0) {
+      /* If GALLIUM_NOPPC is set, don't use PPC codegen */
+      use_ppc_asm = !debug_get_bool_option("GALLIUM_NOPPC", FALSE);
+   }
+   if (!use_ppc_asm)
+      return FALSE;
+
+   if (0) {
+      debug_printf("\n********* TGSI->PPC ********\n");
+      tgsi_dump(tokens, 0);
+   }
+
+   util_init_math();
+
+   init_gen_context(&gen, func);
+
+   emit_prologue(func);
+
+   tgsi_parse_init( &parse, tokens );
+
+   while (!tgsi_parse_end_of_tokens(&parse) && ok) {
+      tgsi_parse_token(&parse);
+
+      switch (parse.FullToken.Token.Type) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         if (parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_FRAGMENT) {
+            emit_declaration(func, &parse.FullToken.FullDeclaration );
+         }
+         break;
+
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         if (func->print) {
+            _debug_printf("# ");
+            ic++;
+            tgsi_dump_instruction(&parse.FullToken.FullInstruction, ic);
+         }
+
+         ok = emit_instruction(&gen, &parse.FullToken.FullInstruction);
+
+	 if (!ok) {
+	    debug_printf("failed to translate tgsi opcode %d to PPC (%s)\n", 
+			 parse.FullToken.FullInstruction.Instruction.Opcode,
+                         parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_VERTEX ?
+                         "vertex shader" : "fragment shader");
+	 }
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         /* splat each immediate component into a float[4] vector for SoA */
+         {
+            const uint size = parse.FullToken.FullImmediate.Immediate.NrTokens - 1;
+            uint i;
+            assert(size <= 4);
+            assert(num_immediates < TGSI_EXEC_NUM_IMMEDIATES);
+            for (i = 0; i < size; i++) {
+               immediates[num_immediates][i] =
+		  parse.FullToken.FullImmediate.u.ImmediateFloat32[i].Float;
+            }
+            num_immediates++;
+         }
+         break;
+
+      default:
+	 ok = 0;
+         assert( 0 );
+      }
+   }
+
+   emit_epilogue(func);
+
+   tgsi_parse_free( &parse );
+
+   if (ppc_num_instructions(func) == 0) {
+      /* ran out of memory for instructions */
+      ok = FALSE;
+   }
+
+   if (!ok)
+      debug_printf("TGSI->PPC translation failed\n");
+
+   return ok;
+}
+
+#endif /* PIPE_ARCH_PPC */
diff --git a/src/gallium/auxiliary/tgsi/tgsi_ppc.h b/src/gallium/auxiliary/tgsi/tgsi_ppc.h
new file mode 100644
index 00000000000..829ec075e7f
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_ppc.h
@@ -0,0 +1,51 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef TGSI_PPC_H
+#define TGSI_PPC_H
+
+#if defined __cplusplus
+extern "C" {
+#endif
+
+struct tgsi_token;
+struct ppc_function;
+
+extern const float ppc_builtin_constants[];
+
+
+boolean
+tgsi_emit_ppc(const struct tgsi_token *tokens,
+              struct ppc_function *function,
+              float (*immediates)[4],
+              boolean do_swizzles);
+
+#if defined __cplusplus
+}
+#endif
+
+#endif /* TGSI_PPC_H */
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sanity.c b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
new file mode 100644
index 00000000000..bc7b941b785
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_sanity.c
@@ -0,0 +1,364 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_debug.h"
+#include "tgsi_sanity.h"
+#include "tgsi_info.h"
+#include "tgsi_iterate.h"
+
+#define MAX_REGISTERS 256
+
+typedef uint reg_flag;
+
+#define BITS_IN_REG_FLAG (sizeof( reg_flag ) * 8)
+
+struct sanity_check_ctx
+{
+   struct tgsi_iterate_context iter;
+
+   reg_flag regs_decl[TGSI_FILE_COUNT][MAX_REGISTERS / BITS_IN_REG_FLAG];
+   reg_flag regs_used[TGSI_FILE_COUNT][MAX_REGISTERS / BITS_IN_REG_FLAG];
+   boolean regs_ind_used[TGSI_FILE_COUNT];
+   uint num_imms;
+   uint num_instructions;
+   uint index_of_END;
+
+   uint errors;
+   uint warnings;
+};
+
+static void
+report_error(
+   struct sanity_check_ctx *ctx,
+   const char *format,
+   ... )
+{
+   va_list args;
+
+   debug_printf( "Error  : " );
+   va_start( args, format );
+   _debug_vprintf( format, args );
+   va_end( args );
+   debug_printf( "\n" );
+   ctx->errors++;
+}
+
+static void
+report_warning(
+   struct sanity_check_ctx *ctx,
+   const char *format,
+   ... )
+{
+   va_list args;
+
+   debug_printf( "Warning: " );
+   va_start( args, format );
+   _debug_vprintf( format, args );
+   va_end( args );
+   debug_printf( "\n" );
+   ctx->warnings++;
+}
+
+static boolean
+check_file_name(
+   struct sanity_check_ctx *ctx,
+   uint file )
+{
+   if (file <= TGSI_FILE_NULL || file >= TGSI_FILE_COUNT) {
+      report_error( ctx, "Invalid register file name" );
+      return FALSE;
+   }
+   return TRUE;
+}
+
+static boolean
+is_register_declared(
+   struct sanity_check_ctx *ctx,
+   uint file,
+   int index )
+{
+   assert( index >= 0 && index < MAX_REGISTERS );
+
+   return (ctx->regs_decl[file][index / BITS_IN_REG_FLAG] & (1 << (index % BITS_IN_REG_FLAG))) ? TRUE : FALSE;
+}
+
+static boolean
+is_any_register_declared(
+   struct sanity_check_ctx *ctx,
+   uint file )
+{
+   uint i;
+
+   for (i = 0; i < MAX_REGISTERS / BITS_IN_REG_FLAG; i++)
+      if (ctx->regs_decl[file][i])
+         return TRUE;
+   return FALSE;
+}
+
+static boolean
+is_register_used(
+   struct sanity_check_ctx *ctx,
+   uint file,
+   int index )
+{
+   assert( index < MAX_REGISTERS );
+
+   return (ctx->regs_used[file][index / BITS_IN_REG_FLAG] & (1 << (index % BITS_IN_REG_FLAG))) ? TRUE : FALSE;
+}
+
+static const char *file_names[] =
+{
+   "NULL",
+   "CONST",
+   "IN",
+   "OUT",
+   "TEMP",
+   "SAMP",
+   "ADDR",
+   "IMM"
+};
+
+static boolean
+check_register_usage(
+   struct sanity_check_ctx *ctx,
+   uint file,
+   int index,
+   const char *name,
+   boolean indirect_access )
+{
+   if (!check_file_name( ctx, file ))
+      return FALSE;
+
+   if (indirect_access) {
+      /* Note that 'index' is an offset relative to the value of the
+       * address register.  No range checking done here.
+       */
+      if (!is_any_register_declared( ctx, file ))
+         report_error( ctx, "%s: Undeclared %s register", file_names[file], name );
+      ctx->regs_ind_used[file] = TRUE;
+   }
+   else {
+      if (index < 0 || index > MAX_REGISTERS) {
+         report_error( ctx, "%s[%i]: Invalid index %s",
+                       file_names[file], index, name );
+         return FALSE;
+      }
+
+      if (!is_register_declared( ctx, file, index ))
+         report_error( ctx, "%s[%d]: Undeclared %s register", file_names[file], index, name );
+      ctx->regs_used[file][index / BITS_IN_REG_FLAG] |= (1 << (index % BITS_IN_REG_FLAG));
+   }
+   return TRUE;
+}
+
+static boolean
+iter_instruction(
+   struct tgsi_iterate_context *iter,
+   struct tgsi_full_instruction *inst )
+{
+   struct sanity_check_ctx *ctx = (struct sanity_check_ctx *) iter;
+   const struct tgsi_opcode_info *info;
+   uint i;
+
+   if (inst->Instruction.Opcode == TGSI_OPCODE_END) {
+      if (ctx->index_of_END != ~0) {
+         report_error( ctx, "Too many END instructions" );
+      }
+      ctx->index_of_END = ctx->num_instructions;
+   }
+
+   info = tgsi_get_opcode_info( inst->Instruction.Opcode );
+   if (info == NULL) {
+      report_error( ctx, "Invalid instruction opcode" );
+      return TRUE;
+   }
+
+   if (info->num_dst != inst->Instruction.NumDstRegs) {
+      report_error( ctx, "Invalid number of destination operands" );
+   }
+   if (info->num_src != inst->Instruction.NumSrcRegs) {
+      report_error( ctx, "Invalid number of source operands" );
+   }
+
+   /* Check destination and source registers' validity.
+    * Mark the registers as used.
+    */
+   for (i = 0; i < inst->Instruction.NumDstRegs; i++) {
+      check_register_usage(
+         ctx,
+         inst->FullDstRegisters[i].DstRegister.File,
+         inst->FullDstRegisters[i].DstRegister.Index,
+         "destination",
+         FALSE );
+   }
+   for (i = 0; i < inst->Instruction.NumSrcRegs; i++) {
+      check_register_usage(
+         ctx,
+         inst->FullSrcRegisters[i].SrcRegister.File,
+         inst->FullSrcRegisters[i].SrcRegister.Index,
+         "source",
+         (boolean)inst->FullSrcRegisters[i].SrcRegister.Indirect );
+      if (inst->FullSrcRegisters[i].SrcRegister.Indirect) {
+         uint file;
+         int index;
+
+         file = inst->FullSrcRegisters[i].SrcRegisterInd.File;
+         index = inst->FullSrcRegisters[i].SrcRegisterInd.Index;
+         check_register_usage(
+            ctx,
+            file,
+            index,
+            "indirect",
+            FALSE );
+         if (file != TGSI_FILE_ADDRESS || index != 0)
+            report_warning( ctx, "Indirect register not ADDR[0]" );
+      }
+   }
+
+   ctx->num_instructions++;
+
+   return TRUE;
+}
+
+static boolean
+iter_declaration(
+   struct tgsi_iterate_context *iter,
+   struct tgsi_full_declaration *decl )
+{
+   struct sanity_check_ctx *ctx = (struct sanity_check_ctx *) iter;
+   uint file;
+   uint i;
+
+   /* No declarations allowed after the first instruction.
+    */
+   if (ctx->num_instructions > 0)
+      report_error( ctx, "Instruction expected but declaration found" );
+
+   /* Check registers' validity.
+    * Mark the registers as declared.
+    */
+   file = decl->Declaration.File;
+   if (!check_file_name( ctx, file ))
+      return TRUE;
+   for (i = decl->DeclarationRange.First; i <= decl->DeclarationRange.Last; i++) {
+      if (is_register_declared( ctx, file, i ))
+         report_error( ctx, "The same register declared twice" );
+      ctx->regs_decl[file][i / BITS_IN_REG_FLAG] |= (1 << (i % BITS_IN_REG_FLAG));
+   }
+
+   return TRUE;
+}
+
+static boolean
+iter_immediate(
+   struct tgsi_iterate_context *iter,
+   struct tgsi_full_immediate *imm )
+{
+   struct sanity_check_ctx *ctx = (struct sanity_check_ctx *) iter;
+
+   assert( ctx->num_imms < MAX_REGISTERS );
+
+   /* No immediates allowed after the first instruction.
+    */
+   if (ctx->num_instructions > 0)
+      report_error( ctx, "Instruction expected but immediate found" );
+
+   /* Mark the register as declared.
+    */
+   ctx->regs_decl[TGSI_FILE_IMMEDIATE][ctx->num_imms / BITS_IN_REG_FLAG] |= (1 << (ctx->num_imms % BITS_IN_REG_FLAG));
+   ctx->num_imms++;
+
+   /* Check data type validity.
+    */
+   if (imm->Immediate.DataType != TGSI_IMM_FLOAT32) {
+      report_error( ctx, "Invalid immediate data type" );
+      return TRUE;
+   }
+
+   return TRUE;
+}
+
+static boolean
+epilog(
+   struct tgsi_iterate_context *iter )
+{
+   struct sanity_check_ctx *ctx = (struct sanity_check_ctx *) iter;
+   uint file;
+
+   /* There must be an END instruction somewhere.
+    */
+   if (ctx->index_of_END == ~0) {
+      report_error( ctx, "Missing END instruction" );
+   }
+
+   /* Check if all declared registers were used.
+    */
+   for (file = TGSI_FILE_NULL; file < TGSI_FILE_COUNT; file++) {
+      uint i;
+
+      for (i = 0; i < MAX_REGISTERS; i++) {
+         if (is_register_declared( ctx, file, i ) && !is_register_used( ctx, file, i ) && !ctx->regs_ind_used[file]) {
+            report_warning( ctx, "Register never used" );
+         }
+      }
+   }
+
+   /* Print totals, if any.
+    */
+   if (ctx->errors || ctx->warnings)
+      debug_printf( "%u errors, %u warnings\n", ctx->errors, ctx->warnings );
+
+   return TRUE;
+}
+
+boolean
+tgsi_sanity_check(
+   struct tgsi_token *tokens )
+{
+   struct sanity_check_ctx ctx;
+
+   ctx.iter.prolog = NULL;
+   ctx.iter.iterate_instruction = iter_instruction;
+   ctx.iter.iterate_declaration = iter_declaration;
+   ctx.iter.iterate_immediate = iter_immediate;
+   ctx.iter.epilog = epilog;
+
+   memset( ctx.regs_decl, 0, sizeof( ctx.regs_decl ) );
+   memset( ctx.regs_used, 0, sizeof( ctx.regs_used ) );
+   memset( ctx.regs_ind_used, 0, sizeof( ctx.regs_ind_used ) );
+   ctx.num_imms = 0;
+   ctx.num_instructions = 0;
+   ctx.index_of_END = ~0;
+
+   ctx.errors = 0;
+   ctx.warnings = 0;
+
+   if (!tgsi_iterate_shader( tokens, &ctx.iter ))
+      return FALSE;
+
+   return ctx.errors == 0;
+}
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sanity.h b/src/gallium/auxiliary/tgsi/tgsi_sanity.h
new file mode 100644
index 00000000000..ca45e94c7ad
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_sanity.h
@@ -0,0 +1,49 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef TGSI_SANITY_H
+#define TGSI_SANITY_H
+
+#include "pipe/p_shader_tokens.h"
+
+#if defined __cplusplus
+extern "C" {
+#endif
+
+/* Check the given token stream for errors and common mistakes.
+ * Diagnostic messages are printed out to the debug output.
+ * Returns TRUE if there are no errors, even though there could be some warnings.
+ */
+boolean
+tgsi_sanity_check(
+   struct tgsi_token *tokens );
+
+#if defined __cplusplus
+}
+#endif
+
+#endif /* TGSI_SANITY_H */
diff --git a/src/gallium/auxiliary/tgsi/tgsi_scan.c b/src/gallium/auxiliary/tgsi/tgsi_scan.c
new file mode 100644
index 00000000000..d02205a63ef
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_scan.c
@@ -0,0 +1,249 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * Copyright 2008 VMware, Inc.  All rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * TGSI program scan utility.
+ * Used to determine which registers and instructions are used by a shader.
+ *
+ * Authors:  Brian Paul
+ */
+
+
+#include "util/u_math.h"
+#include "tgsi/tgsi_build.h"
+#include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_scan.h"
+
+
+
+
+/**
+ * Scan the given TGSI shader to collect information such as number of
+ * registers used, special instructions used, etc.
+ * \return info  the result of the scan
+ */
+void
+tgsi_scan_shader(const struct tgsi_token *tokens,
+                 struct tgsi_shader_info *info)
+{
+   uint procType, i;
+   struct tgsi_parse_context parse;
+
+   memset(info, 0, sizeof(*info));
+   for (i = 0; i < TGSI_FILE_COUNT; i++)
+      info->file_max[i] = -1;
+
+   /**
+    ** Setup to begin parsing input shader
+    **/
+   if (tgsi_parse_init( &parse, tokens ) != TGSI_PARSE_OK) {
+      debug_printf("tgsi_parse_init() failed in tgsi_scan_shader()!\n");
+      return;
+   }
+   procType = parse.FullHeader.Processor.Processor;
+   assert(procType == TGSI_PROCESSOR_FRAGMENT ||
+          procType == TGSI_PROCESSOR_VERTEX ||
+          procType == TGSI_PROCESSOR_GEOMETRY);
+
+
+   /**
+    ** Loop over incoming program tokens/instructions
+    */
+   while( !tgsi_parse_end_of_tokens( &parse ) ) {
+
+      info->num_tokens++;
+
+      tgsi_parse_token( &parse );
+
+      switch( parse.FullToken.Token.Type ) {
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         {
+            const struct tgsi_full_instruction *fullinst
+               = &parse.FullToken.FullInstruction;
+
+            assert(fullinst->Instruction.Opcode < TGSI_OPCODE_LAST);
+            info->opcode_count[fullinst->Instruction.Opcode]++;
+
+            /* special case: scan fragment shaders for use of the fog
+             * input/attribute.  The X component is fog, the Y component
+             * is the front/back-face flag.
+             */
+            if (procType == TGSI_PROCESSOR_FRAGMENT) {
+               uint i;
+               for (i = 0; i < fullinst->Instruction.NumSrcRegs; i++) {
+                  const struct tgsi_full_src_register *src =
+                     &fullinst->FullSrcRegisters[i];
+                  if (src->SrcRegister.File == TGSI_FILE_INPUT) {
+                     const int ind = src->SrcRegister.Index;
+                     if (info->input_semantic_name[ind] == TGSI_SEMANTIC_FOG) {
+                        if (src->SrcRegister.SwizzleX == TGSI_SWIZZLE_X) {
+                           info->uses_fogcoord = TRUE;
+                        }
+                        else if (src->SrcRegister.SwizzleX == TGSI_SWIZZLE_Y) {
+                           info->uses_frontfacing = TRUE;
+                        }
+                     }
+                  }
+               }
+            }
+         }
+         break;
+
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         {
+            const struct tgsi_full_declaration *fulldecl
+               = &parse.FullToken.FullDeclaration;
+            const uint file = fulldecl->Declaration.File;
+            uint reg;
+            for (reg = fulldecl->DeclarationRange.First;
+                 reg <= fulldecl->DeclarationRange.Last;
+                 reg++) {
+
+               /* only first 32 regs will appear in this bitfield */
+               info->file_mask[file] |= (1 << reg);
+               info->file_count[file]++;
+               info->file_max[file] = MAX2(info->file_max[file], (int)reg);
+
+               if (file == TGSI_FILE_INPUT) {
+                  info->input_semantic_name[reg] = (ubyte)fulldecl->Semantic.SemanticName;
+                  info->input_semantic_index[reg] = (ubyte)fulldecl->Semantic.SemanticIndex;
+                  info->num_inputs++;
+               }
+               else if (file == TGSI_FILE_OUTPUT) {
+                  info->output_semantic_name[reg] = (ubyte)fulldecl->Semantic.SemanticName;
+                  info->output_semantic_index[reg] = (ubyte)fulldecl->Semantic.SemanticIndex;
+                  info->num_outputs++;
+               }
+
+               /* special case */
+               if (procType == TGSI_PROCESSOR_FRAGMENT &&
+                   file == TGSI_FILE_OUTPUT &&
+                   fulldecl->Semantic.SemanticName == TGSI_SEMANTIC_POSITION) {
+                  info->writes_z = TRUE;
+               }
+            }
+         }
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         info->immediate_count++;
+         break;
+
+      default:
+         assert( 0 );
+      }
+   }
+
+   info->uses_kill = (info->opcode_count[TGSI_OPCODE_KIL] ||
+                      info->opcode_count[TGSI_OPCODE_KILP]);
+
+   tgsi_parse_free (&parse);
+}
+
+
+
+/**
+ * Check if the given shader is a "passthrough" shader consisting of only
+ * MOV instructions of the form:  MOV OUT[n], IN[n]
+ *  
+ */
+boolean
+tgsi_is_passthrough_shader(const struct tgsi_token *tokens)
+{
+   struct tgsi_parse_context parse;
+
+   /**
+    ** Setup to begin parsing input shader
+    **/
+   if (tgsi_parse_init(&parse, tokens) != TGSI_PARSE_OK) {
+      debug_printf("tgsi_parse_init() failed in tgsi_is_passthrough_shader()!\n");
+      return FALSE;
+   }
+
+   /**
+    ** Loop over incoming program tokens/instructions
+    */
+   while (!tgsi_parse_end_of_tokens(&parse)) {
+
+      tgsi_parse_token(&parse);
+
+      switch (parse.FullToken.Token.Type) {
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         {
+            struct tgsi_full_instruction *fullinst =
+               &parse.FullToken.FullInstruction;
+            const struct tgsi_full_src_register *src =
+               &fullinst->FullSrcRegisters[0];
+            const struct tgsi_full_dst_register *dst =
+               &fullinst->FullDstRegisters[0];
+
+            /* Do a whole bunch of checks for a simple move */
+            if (fullinst->Instruction.Opcode != TGSI_OPCODE_MOV ||
+                src->SrcRegister.File != TGSI_FILE_INPUT ||
+                dst->DstRegister.File != TGSI_FILE_OUTPUT ||
+                src->SrcRegister.Index != dst->DstRegister.Index ||
+
+                src->SrcRegister.Negate ||
+                src->SrcRegisterExtMod.Negate ||
+                src->SrcRegisterExtMod.Absolute ||
+                src->SrcRegisterExtMod.Scale2X ||
+                src->SrcRegisterExtMod.Bias ||
+                src->SrcRegisterExtMod.Complement ||
+
+                src->SrcRegister.SwizzleX != TGSI_SWIZZLE_X ||
+                src->SrcRegister.SwizzleY != TGSI_SWIZZLE_Y ||
+                src->SrcRegister.SwizzleZ != TGSI_SWIZZLE_Z ||
+                src->SrcRegister.SwizzleW != TGSI_SWIZZLE_W ||
+
+                src->SrcRegisterExtSwz.ExtSwizzleX != TGSI_EXTSWIZZLE_X ||
+                src->SrcRegisterExtSwz.ExtSwizzleY != TGSI_EXTSWIZZLE_Y ||
+                src->SrcRegisterExtSwz.ExtSwizzleZ != TGSI_EXTSWIZZLE_Z ||
+                src->SrcRegisterExtSwz.ExtSwizzleW != TGSI_EXTSWIZZLE_W ||
+
+                dst->DstRegister.WriteMask != TGSI_WRITEMASK_XYZW)
+            {
+               tgsi_parse_free(&parse);
+               return FALSE;
+            }
+         }
+         break;
+
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         /* fall-through */
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         /* fall-through */
+      default:
+         ; /* no-op */
+      }
+   }
+
+   tgsi_parse_free(&parse);
+
+   /* if we get here, it's a pass-through shader */
+   return TRUE;
+}
diff --git a/src/gallium/auxiliary/tgsi/tgsi_scan.h b/src/gallium/auxiliary/tgsi/tgsi_scan.h
new file mode 100644
index 00000000000..2c1a75bc812
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_scan.h
@@ -0,0 +1,75 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef TGSI_SCAN_H
+#define TGSI_SCAN_H
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_state.h"
+#include "pipe/p_shader_tokens.h"
+
+
+/**
+ * Shader summary info
+ */
+struct tgsi_shader_info
+{
+   uint num_tokens;
+
+   ubyte num_inputs;
+   ubyte num_outputs;
+   ubyte input_semantic_name[PIPE_MAX_SHADER_INPUTS]; /**< TGSI_SEMANTIC_x */
+   ubyte input_semantic_index[PIPE_MAX_SHADER_INPUTS];
+   ubyte output_semantic_name[PIPE_MAX_SHADER_OUTPUTS]; /**< TGSI_SEMANTIC_x */
+   ubyte output_semantic_index[PIPE_MAX_SHADER_OUTPUTS];
+
+   uint file_mask[TGSI_FILE_COUNT];  /**< bitmask of declared registers */
+   uint file_count[TGSI_FILE_COUNT];  /**< number of declared registers */
+   int file_max[TGSI_FILE_COUNT];  /**< highest index of declared registers */
+
+   uint immediate_count; /**< number of immediates declared */
+
+   uint opcode_count[TGSI_OPCODE_LAST];  /**< opcode histogram */
+
+   boolean writes_z;  /**< does fragment shader write Z value? */
+   boolean uses_kill;  /**< KIL or KILP instruction used? */
+   boolean uses_fogcoord; /**< fragment shader uses fog coord? */
+   boolean uses_frontfacing; /**< fragment shader uses front/back-face flag? */
+};
+
+
+extern void
+tgsi_scan_shader(const struct tgsi_token *tokens,
+                 struct tgsi_shader_info *info);
+
+
+extern boolean
+tgsi_is_passthrough_shader(const struct tgsi_token *tokens);
+
+
+#endif /* TGSI_SCAN_H */
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
new file mode 100644
index 00000000000..481ba89c5e7
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.c
@@ -0,0 +1,2723 @@
+/**************************************************************************
+ * 
+ * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_config.h"
+
+#if defined(PIPE_ARCH_X86)
+
+#include "pipe/p_debug.h"
+#include "pipe/p_shader_tokens.h"
+#include "util/u_math.h"
+#if defined(PIPE_ARCH_SSE)
+#include "util/u_sse.h"
+#endif
+#include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_util.h"
+#include "tgsi_exec.h"
+#include "tgsi_sse2.h"
+
+#include "rtasm/rtasm_x86sse.h"
+
+/* for 1/sqrt()
+ *
+ * This costs about 100fps (close to 10%) in gears:
+ */
+#define HIGH_PRECISION 1
+
+#define FAST_MATH 1
+
+
+#define FOR_EACH_CHANNEL( CHAN )\
+   for (CHAN = 0; CHAN < NUM_CHANNELS; CHAN++)
+
+#define IS_DST0_CHANNEL_ENABLED( INST, CHAN )\
+   ((INST).FullDstRegisters[0].DstRegister.WriteMask & (1 << (CHAN)))
+
+#define IF_IS_DST0_CHANNEL_ENABLED( INST, CHAN )\
+   if (IS_DST0_CHANNEL_ENABLED( INST, CHAN ))
+
+#define FOR_EACH_DST0_ENABLED_CHANNEL( INST, CHAN )\
+   FOR_EACH_CHANNEL( CHAN )\
+      IF_IS_DST0_CHANNEL_ENABLED( INST, CHAN )
+
+#define CHAN_X 0
+#define CHAN_Y 1
+#define CHAN_Z 2
+#define CHAN_W 3
+
+#define TEMP_ONE_I   TGSI_EXEC_TEMP_ONE_I
+#define TEMP_ONE_C   TGSI_EXEC_TEMP_ONE_C
+
+#define TEMP_R0   TGSI_EXEC_TEMP_R0
+#define TEMP_ADDR TGSI_EXEC_TEMP_ADDR
+#define TEMP_EXEC_MASK_I TGSI_EXEC_MASK_I
+#define TEMP_EXEC_MASK_C TGSI_EXEC_MASK_C
+
+
+/**
+ * X86 utility functions.
+ */
+
+static struct x86_reg
+make_xmm(
+   unsigned xmm )
+{
+   return x86_make_reg(
+      file_XMM,
+      (enum x86_reg_name) xmm );
+}
+
+/**
+ * X86 register mapping helpers.
+ */
+
+static struct x86_reg
+get_const_base( void )
+{
+   return x86_make_reg(
+      file_REG32,
+      reg_CX );
+}
+
+static struct x86_reg
+get_input_base( void )
+{
+   return x86_make_reg(
+      file_REG32,
+      reg_AX );
+}
+
+static struct x86_reg
+get_output_base( void )
+{
+   return x86_make_reg(
+      file_REG32,
+      reg_DX );
+}
+
+static struct x86_reg
+get_temp_base( void )
+{
+   return x86_make_reg(
+      file_REG32,
+      reg_BX );
+}
+
+static struct x86_reg
+get_coef_base( void )
+{
+   return get_output_base();
+}
+
+static struct x86_reg
+get_immediate_base( void )
+{
+   return x86_make_reg(
+      file_REG32,
+      reg_DI );
+}
+
+
+/**
+ * Data access helpers.
+ */
+
+
+static struct x86_reg
+get_immediate(
+   unsigned vec,
+   unsigned chan )
+{
+   return x86_make_disp(
+      get_immediate_base(),
+      (vec * 4 + chan) * 4 );
+}
+
+static struct x86_reg
+get_const(
+   unsigned vec,
+   unsigned chan )
+{
+   return x86_make_disp(
+      get_const_base(),
+      (vec * 4 + chan) * 4 );
+}
+
+static struct x86_reg
+get_input(
+   unsigned vec,
+   unsigned chan )
+{
+   return x86_make_disp(
+      get_input_base(),
+      (vec * 4 + chan) * 16 );
+}
+
+static struct x86_reg
+get_output(
+   unsigned vec,
+   unsigned chan )
+{
+   return x86_make_disp(
+      get_output_base(),
+      (vec * 4 + chan) * 16 );
+}
+
+static struct x86_reg
+get_temp(
+   unsigned vec,
+   unsigned chan )
+{
+   return x86_make_disp(
+      get_temp_base(),
+      (vec * 4 + chan) * 16 );
+}
+
+static struct x86_reg
+get_coef(
+   unsigned vec,
+   unsigned chan,
+   unsigned member )
+{
+   return x86_make_disp(
+      get_coef_base(),
+      ((vec * 3 + member) * 4 + chan) * 4 );
+}
+
+
+static void
+emit_ret(
+   struct x86_function  *func )
+{
+   x86_ret( func );
+}
+
+
+/**
+ * Data fetch helpers.
+ */
+
+/**
+ * Copy a shader constant to xmm register
+ * \param xmm  the destination xmm register
+ * \param vec  the src const buffer index
+ * \param chan  src channel to fetch (X, Y, Z or W)
+ */
+static void
+emit_const(
+   struct x86_function *func,
+   uint xmm,
+   int vec,
+   uint chan,
+   uint indirect,
+   uint indirectFile,
+   int indirectIndex )
+{
+   if (indirect) {
+      /* 'vec' is the offset from the address register's value.
+       * We're loading CONST[ADDR+vec] into an xmm register.
+       */
+      struct x86_reg r0 = get_input_base();
+      struct x86_reg r1 = get_output_base();
+      uint i;
+
+      assert( indirectFile == TGSI_FILE_ADDRESS );
+      assert( indirectIndex == 0 );
+
+      x86_push( func, r0 );
+      x86_push( func, r1 );
+
+      /*
+       * Loop over the four pixels or vertices in the quad.
+       * Get the value of the address (offset) register for pixel/vertex[i],
+       * add it to the src offset and index into the constant buffer.
+       * Note that we're working on SOA data.
+       * If any of the pixel/vertex execution channels are unused their
+       * values will be garbage.  It's very important that we don't use
+       * those garbage values as indexes into the constant buffer since
+       * that'll cause segfaults.
+       * The solution is to bitwise-AND the offset with the execution mask
+       * register whose values are either 0 or ~0.
+       * The caller must setup the execution mask register to indicate
+       * which channels are valid/alive before running the shader.
+       * The execution mask will also figure into loops and conditionals
+       * someday.
+       */
+      for (i = 0; i < QUAD_SIZE; i++) {
+         /* r1 = address register[i] */
+         x86_mov( func, r1, x86_make_disp( get_temp( TEMP_ADDR, CHAN_X ), i * 4 ) );
+         /* r0 = execution mask[i] */
+         x86_mov( func, r0, x86_make_disp( get_temp( TEMP_EXEC_MASK_I, TEMP_EXEC_MASK_C ), i * 4 ) );
+         /* r1 = r1 & r0 */
+         x86_and( func, r1, r0 );
+         /* r0 = 'vec', the offset */
+         x86_lea( func, r0, get_const( vec, chan ) );
+
+         /* Quick hack to multiply r1 by 16 -- need to add SHL to rtasm.
+          */
+         x86_add( func, r1, r1 );
+         x86_add( func, r1, r1 );
+         x86_add( func, r1, r1 );
+         x86_add( func, r1, r1 );
+
+         x86_add( func, r0, r1 );  /* r0 = r0 + r1 */
+         x86_mov( func, r1, x86_deref( r0 ) );
+         x86_mov( func, x86_make_disp( get_temp( TEMP_R0, CHAN_X ), i * 4 ), r1 );
+      }
+
+      x86_pop( func, r1 );
+      x86_pop( func, r0 );
+
+      sse_movaps(
+         func,
+         make_xmm( xmm ),
+         get_temp( TEMP_R0, CHAN_X ) );
+   }
+   else {
+      /* 'vec' is the index into the src register file, such as TEMP[vec] */
+      assert( vec >= 0 );
+
+      sse_movss(
+         func,
+         make_xmm( xmm ),
+         get_const( vec, chan ) );
+      sse_shufps(
+         func,
+         make_xmm( xmm ),
+         make_xmm( xmm ),
+         SHUF( 0, 0, 0, 0 ) );
+   }
+}
+
+static void
+emit_immediate(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   sse_movss(
+      func,
+      make_xmm( xmm ),
+      get_immediate( vec, chan ) );
+   sse_shufps(
+      func,
+      make_xmm( xmm ),
+      make_xmm( xmm ),
+      SHUF( 0, 0, 0, 0 ) );
+}
+
+
+/**
+ * Copy a shader input to xmm register
+ * \param xmm  the destination xmm register
+ * \param vec  the src input attrib
+ * \param chan  src channel to fetch (X, Y, Z or W)
+ */
+static void
+emit_inputf(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   sse_movups(
+      func,
+      make_xmm( xmm ),
+      get_input( vec, chan ) );
+}
+
+/**
+ * Store an xmm register to a shader output
+ * \param xmm  the source xmm register
+ * \param vec  the dest output attrib
+ * \param chan  src dest channel to store (X, Y, Z or W)
+ */
+static void
+emit_output(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   sse_movups(
+      func,
+      get_output( vec, chan ),
+      make_xmm( xmm ) );
+}
+
+/**
+ * Copy a shader temporary to xmm register
+ * \param xmm  the destination xmm register
+ * \param vec  the src temp register
+ * \param chan  src channel to fetch (X, Y, Z or W)
+ */
+static void
+emit_tempf(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   sse_movaps(
+      func,
+      make_xmm( xmm ),
+      get_temp( vec, chan ) );
+}
+
+/**
+ * Load an xmm register with an input attrib coefficient (a0, dadx or dady)
+ * \param xmm  the destination xmm register
+ * \param vec  the src input/attribute coefficient index
+ * \param chan  src channel to fetch (X, Y, Z or W)
+ * \param member  0=a0, 1=dadx, 2=dady
+ */
+static void
+emit_coef(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan,
+   unsigned member )
+{
+   sse_movss(
+      func,
+      make_xmm( xmm ),
+      get_coef( vec, chan, member ) );
+   sse_shufps(
+      func,
+      make_xmm( xmm ),
+      make_xmm( xmm ),
+      SHUF( 0, 0, 0, 0 ) );
+}
+
+/**
+ * Data store helpers.
+ */
+
+static void
+emit_inputs(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   sse_movups(
+      func,
+      get_input( vec, chan ),
+      make_xmm( xmm ) );
+}
+
+static void
+emit_temps(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   sse_movaps(
+      func,
+      get_temp( vec, chan ),
+      make_xmm( xmm ) );
+}
+
+static void
+emit_addrs(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   assert( vec == 0 );
+
+   emit_temps(
+      func,
+      xmm,
+      vec + TGSI_EXEC_TEMP_ADDR,
+      chan );
+}
+
+/**
+ * Coefficent fetch helpers.
+ */
+
+static void
+emit_coef_a0(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   emit_coef(
+      func,
+      xmm,
+      vec,
+      chan,
+      0 );
+}
+
+static void
+emit_coef_dadx(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   emit_coef(
+      func,
+      xmm,
+      vec,
+      chan,
+      1 );
+}
+
+static void
+emit_coef_dady(
+   struct x86_function *func,
+   unsigned xmm,
+   unsigned vec,
+   unsigned chan )
+{
+   emit_coef(
+      func,
+      xmm,
+      vec,
+      chan,
+      2 );
+}
+
+/**
+ * Function call helpers.
+ */
+
+/**
+ * NOTE: In gcc, if the destination uses the SSE intrinsics, then it must be 
+ * defined with __attribute__((force_align_arg_pointer)), as we do not guarantee
+ * that the stack pointer is 16 byte aligned, as expected.
+ */
+static void
+emit_func_call_dst(
+   struct x86_function *func,
+   unsigned xmm_save,
+   unsigned xmm_dst,
+   void (PIPE_CDECL *code)() )
+{
+   struct x86_reg ecx = x86_make_reg( file_REG32, reg_CX );
+   unsigned i, n;
+   unsigned xmm_mask;
+   
+   /* Bitmask of the xmm registers to save */
+   xmm_mask = (1 << xmm_save) - 1;
+   xmm_mask &= ~(1 << xmm_dst);
+
+   sse_movaps(
+      func,
+      get_temp( TEMP_R0, 0 ),
+      make_xmm( xmm_dst ) );
+
+   x86_push(
+      func,
+      x86_make_reg( file_REG32, reg_AX) );
+   x86_push(
+      func,
+      x86_make_reg( file_REG32, reg_CX) );
+   x86_push(
+      func,
+      x86_make_reg( file_REG32, reg_DX) );
+   
+   for(i = 0, n = 0; i < 8; ++i)
+      if(xmm_mask & (1 << i))
+         ++n;
+   
+   x86_sub_imm(
+      func, 
+      x86_make_reg( file_REG32, reg_SP ),
+      n*16);
+
+   for(i = 0, n = 0; i < 8; ++i)
+      if(xmm_mask & (1 << i)) {
+         sse_movups(
+            func,
+            x86_make_disp( x86_make_reg( file_REG32, reg_SP ), n*16 ),
+            make_xmm( i ) );
+         ++n;
+      }
+   
+   x86_lea(
+      func,
+      ecx,
+      get_temp( TEMP_R0, 0 ) );
+   
+   x86_push( func, ecx );
+   x86_mov_reg_imm( func, ecx, (unsigned long) code );
+   x86_call( func, ecx );
+   x86_pop(func, ecx );
+   
+   for(i = 0, n = 0; i < 8; ++i)
+      if(xmm_mask & (1 << i)) {
+         sse_movups(
+            func,
+            make_xmm( i ),
+            x86_make_disp( x86_make_reg( file_REG32, reg_SP ), n*16 ) );
+         ++n;
+      }
+   
+   x86_add_imm(
+      func, 
+      x86_make_reg( file_REG32, reg_SP ),
+      n*16);
+
+   /* Restore GP registers in a reverse order.
+    */
+   x86_pop(
+      func,
+      x86_make_reg( file_REG32, reg_DX) );
+   x86_pop(
+      func,
+      x86_make_reg( file_REG32, reg_CX) );
+   x86_pop(
+      func,
+      x86_make_reg( file_REG32, reg_AX) );
+
+   sse_movaps(
+      func,
+      make_xmm( xmm_dst ),
+      get_temp( TEMP_R0, 0 ) );
+}
+
+static void
+emit_func_call_dst_src(
+   struct x86_function *func,
+   unsigned xmm_save, 
+   unsigned xmm_dst,
+   unsigned xmm_src,
+   void (PIPE_CDECL *code)() )
+{
+   sse_movaps(
+      func,
+      get_temp( TEMP_R0, 1 ),
+      make_xmm( xmm_src ) );
+
+   emit_func_call_dst(
+      func,
+      xmm_save,
+      xmm_dst,
+      code );
+}
+
+
+#if defined(PIPE_ARCH_SSE)
+
+/*
+ * Fast SSE2 implementation of special math functions.
+ */
+
+#define POLY0(x, c0) _mm_set1_ps(c0)
+#define POLY1(x, c0, c1) _mm_add_ps(_mm_mul_ps(POLY0(x, c1), x), _mm_set1_ps(c0))
+#define POLY2(x, c0, c1, c2) _mm_add_ps(_mm_mul_ps(POLY1(x, c1, c2), x), _mm_set1_ps(c0))
+#define POLY3(x, c0, c1, c2, c3) _mm_add_ps(_mm_mul_ps(POLY2(x, c1, c2, c3), x), _mm_set1_ps(c0))
+#define POLY4(x, c0, c1, c2, c3, c4) _mm_add_ps(_mm_mul_ps(POLY3(x, c1, c2, c3, c4), x), _mm_set1_ps(c0))
+#define POLY5(x, c0, c1, c2, c3, c4, c5) _mm_add_ps(_mm_mul_ps(POLY4(x, c1, c2, c3, c4, c5), x), _mm_set1_ps(c0))
+
+#define EXP_POLY_DEGREE 3
+#define LOG_POLY_DEGREE 5
+
+/**
+ * See http://www.devmaster.net/forums/showthread.php?p=43580
+ */
+static INLINE __m128 
+exp2f4(__m128 x)
+{
+   __m128i ipart;
+   __m128 fpart, expipart, expfpart;
+
+   x = _mm_min_ps(x, _mm_set1_ps( 129.00000f));
+   x = _mm_max_ps(x, _mm_set1_ps(-126.99999f));
+
+   /* ipart = int(x - 0.5) */
+   ipart = _mm_cvtps_epi32(_mm_sub_ps(x, _mm_set1_ps(0.5f)));
+
+   /* fpart = x - ipart */
+   fpart = _mm_sub_ps(x, _mm_cvtepi32_ps(ipart));
+
+   /* expipart = (float) (1 << ipart) */
+   expipart = _mm_castsi128_ps(_mm_slli_epi32(_mm_add_epi32(ipart, _mm_set1_epi32(127)), 23));
+
+   /* minimax polynomial fit of 2**x, in range [-0.5, 0.5[ */
+#if EXP_POLY_DEGREE == 5
+   expfpart = POLY5(fpart, 9.9999994e-1f, 6.9315308e-1f, 2.4015361e-1f, 5.5826318e-2f, 8.9893397e-3f, 1.8775767e-3f);
+#elif EXP_POLY_DEGREE == 4
+   expfpart = POLY4(fpart, 1.0000026f, 6.9300383e-1f, 2.4144275e-1f, 5.2011464e-2f, 1.3534167e-2f);
+#elif EXP_POLY_DEGREE == 3
+   expfpart = POLY3(fpart, 9.9992520e-1f, 6.9583356e-1f, 2.2606716e-1f, 7.8024521e-2f);
+#elif EXP_POLY_DEGREE == 2
+   expfpart = POLY2(fpart, 1.0017247f, 6.5763628e-1f, 3.3718944e-1f);
+#else
+#error
+#endif
+
+   return _mm_mul_ps(expipart, expfpart);
+}
+
+
+/**
+ * See http://www.devmaster.net/forums/showthread.php?p=43580
+ */
+static INLINE __m128 
+log2f4(__m128 x)
+{
+   __m128i expmask = _mm_set1_epi32(0x7f800000);
+   __m128i mantmask = _mm_set1_epi32(0x007fffff);
+   __m128 one = _mm_set1_ps(1.0f);
+
+   __m128i i = _mm_castps_si128(x);
+
+   /* exp = (float) exponent(x) */
+   __m128 exp = _mm_cvtepi32_ps(_mm_sub_epi32(_mm_srli_epi32(_mm_and_si128(i, expmask), 23), _mm_set1_epi32(127)));
+
+   /* mant = (float) mantissa(x) */
+   __m128 mant = _mm_or_ps(_mm_castsi128_ps(_mm_and_si128(i, mantmask)), one);
+
+   __m128 logmant;
+
+   /* Minimax polynomial fit of log2(x)/(x - 1), for x in range [1, 2[ 
+    * These coefficients can be generate with 
+    * http://www.boost.org/doc/libs/1_36_0/libs/math/doc/sf_and_dist/html/math_toolkit/toolkit/internals2/minimax.html
+    */
+#if LOG_POLY_DEGREE == 6
+   logmant = POLY5(mant, 3.11578814719469302614f, -3.32419399085241980044f, 2.59883907202499966007f, -1.23152682416275988241f, 0.318212422185251071475f, -0.0344359067839062357313f);
+#elif LOG_POLY_DEGREE == 5
+   logmant = POLY4(mant, 2.8882704548164776201f, -2.52074962577807006663f, 1.48116647521213171641f, -0.465725644288844778798f, 0.0596515482674574969533f);
+#elif LOG_POLY_DEGREE == 4
+   logmant = POLY3(mant, 2.61761038894603480148f, -1.75647175389045657003f, 0.688243882994381274313f, -0.107254423828329604454f);
+#elif LOG_POLY_DEGREE == 3
+   logmant = POLY2(mant, 2.28330284476918490682f, -1.04913055217340124191f, 0.204446009836232697516f);
+#else
+#error
+#endif
+
+   /* This effectively increases the polynomial degree by one, but ensures that log2(1) == 0*/
+   logmant = _mm_mul_ps(logmant, _mm_sub_ps(mant, one));
+
+   return _mm_add_ps(logmant, exp);
+}
+
+
+static INLINE __m128
+powf4(__m128 x, __m128 y)
+{
+   return exp2f4(_mm_mul_ps(log2f4(x), y));
+}
+
+#endif /* PIPE_ARCH_SSE */
+
+
+
+/**
+ * Low-level instruction translators.
+ */
+
+static void
+emit_abs(
+   struct x86_function *func,
+   unsigned xmm )
+{
+   sse_andps(
+      func,
+      make_xmm( xmm ),
+      get_temp(
+         TGSI_EXEC_TEMP_7FFFFFFF_I,
+         TGSI_EXEC_TEMP_7FFFFFFF_C ) );
+}
+
+static void
+emit_add(
+   struct x86_function *func,
+   unsigned xmm_dst,
+   unsigned xmm_src )
+{
+   sse_addps(
+      func,
+      make_xmm( xmm_dst ),
+      make_xmm( xmm_src ) );
+}
+
+static void PIPE_CDECL
+cos4f(
+   float *store )
+{
+   store[0] = cosf( store[0] );
+   store[1] = cosf( store[1] );
+   store[2] = cosf( store[2] );
+   store[3] = cosf( store[3] );
+}
+
+static void
+emit_cos(
+   struct x86_function *func,
+   unsigned xmm_save, 
+   unsigned xmm_dst )
+{
+   emit_func_call_dst(
+      func,
+      xmm_save, 
+      xmm_dst,
+      cos4f );
+}
+
+static void PIPE_CDECL
+#if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_SSE)
+__attribute__((force_align_arg_pointer))
+#endif
+ex24f(
+   float *store )
+{
+#if defined(PIPE_ARCH_SSE)
+   _mm_store_ps(&store[0], exp2f4( _mm_load_ps(&store[0]) ));
+#else
+   store[0] = util_fast_exp2( store[0] );
+   store[1] = util_fast_exp2( store[1] );
+   store[2] = util_fast_exp2( store[2] );
+   store[3] = util_fast_exp2( store[3] );
+#endif
+}
+
+static void
+emit_ex2(
+   struct x86_function *func,
+   unsigned xmm_save, 
+   unsigned xmm_dst )
+{
+   emit_func_call_dst(
+      func,
+      xmm_save,
+      xmm_dst,
+      ex24f );
+}
+
+static void
+emit_f2it(
+   struct x86_function *func,
+   unsigned xmm )
+{
+   sse2_cvttps2dq(
+      func,
+      make_xmm( xmm ),
+      make_xmm( xmm ) );
+}
+
+static void
+emit_i2f(
+   struct x86_function *func,
+   unsigned xmm )
+{
+   sse2_cvtdq2ps(
+      func,
+      make_xmm( xmm ),
+      make_xmm( xmm ) );
+}
+
+static void PIPE_CDECL
+flr4f(
+   float *store )
+{
+   store[0] = floorf( store[0] );
+   store[1] = floorf( store[1] );
+   store[2] = floorf( store[2] );
+   store[3] = floorf( store[3] );
+}
+
+static void
+emit_flr(
+   struct x86_function *func,
+   unsigned xmm_save, 
+   unsigned xmm_dst )
+{
+   emit_func_call_dst(
+      func,
+      xmm_save,
+      xmm_dst,
+      flr4f );
+}
+
+static void PIPE_CDECL
+frc4f(
+   float *store )
+{
+   store[0] -= floorf( store[0] );
+   store[1] -= floorf( store[1] );
+   store[2] -= floorf( store[2] );
+   store[3] -= floorf( store[3] );
+}
+
+static void
+emit_frc(
+   struct x86_function *func,
+   unsigned xmm_save, 
+   unsigned xmm_dst )
+{
+   emit_func_call_dst(
+      func,
+      xmm_save,
+      xmm_dst,
+      frc4f );
+}
+
+static void PIPE_CDECL
+#if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_SSE)
+__attribute__((force_align_arg_pointer))
+#endif
+lg24f(
+   float *store )
+{
+#if defined(PIPE_ARCH_SSE)
+   _mm_store_ps(&store[0], log2f4( _mm_load_ps(&store[0]) ));
+#else
+   store[0] = util_fast_log2( store[0] );
+   store[1] = util_fast_log2( store[1] );
+   store[2] = util_fast_log2( store[2] );
+   store[3] = util_fast_log2( store[3] );
+#endif
+}
+
+static void
+emit_lg2(
+   struct x86_function *func,
+   unsigned xmm_save, 
+   unsigned xmm_dst )
+{
+   emit_func_call_dst(
+      func,
+      xmm_save,
+      xmm_dst,
+      lg24f );
+}
+
+static void
+emit_MOV(
+   struct x86_function *func,
+   unsigned xmm_dst,
+   unsigned xmm_src )
+{
+   sse_movups(
+      func,
+      make_xmm( xmm_dst ),
+      make_xmm( xmm_src ) );
+}
+
+static void
+emit_mul (struct x86_function *func,
+          unsigned xmm_dst,
+          unsigned xmm_src)
+{
+   sse_mulps(
+      func,
+      make_xmm( xmm_dst ),
+      make_xmm( xmm_src ) );
+}
+
+static void
+emit_neg(
+   struct x86_function *func,
+   unsigned xmm )
+{
+   sse_xorps(
+      func,
+      make_xmm( xmm ),
+      get_temp(
+         TGSI_EXEC_TEMP_80000000_I,
+         TGSI_EXEC_TEMP_80000000_C ) );
+}
+
+static void PIPE_CDECL
+#if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_SSE)
+__attribute__((force_align_arg_pointer))
+#endif
+pow4f(
+   float *store )
+{
+#if defined(PIPE_ARCH_SSE)
+   _mm_store_ps(&store[0], powf4( _mm_load_ps(&store[0]), _mm_load_ps(&store[4]) ));
+#else
+   store[0] = util_fast_pow( store[0], store[4] );
+   store[1] = util_fast_pow( store[1], store[5] );
+   store[2] = util_fast_pow( store[2], store[6] );
+   store[3] = util_fast_pow( store[3], store[7] );
+#endif
+}
+
+static void
+emit_pow(
+   struct x86_function *func,
+   unsigned xmm_save, 
+   unsigned xmm_dst,
+   unsigned xmm_src )
+{
+   emit_func_call_dst_src(
+      func,
+      xmm_save,
+      xmm_dst,
+      xmm_src,
+      pow4f );
+}
+
+static void
+emit_rcp (
+   struct x86_function *func,
+   unsigned xmm_dst,
+   unsigned xmm_src )
+{
+   /* On Intel CPUs at least, this is only accurate to 12 bits -- not
+    * good enough.  Need to either emit a proper divide or use the
+    * iterative technique described below in emit_rsqrt().
+    */
+   sse2_rcpps(
+      func,
+      make_xmm( xmm_dst ),
+      make_xmm( xmm_src ) );
+}
+
+static void PIPE_CDECL
+rnd4f(
+   float *store )
+{
+   store[0] = floorf( store[0] + 0.5f );
+   store[1] = floorf( store[1] + 0.5f );
+   store[2] = floorf( store[2] + 0.5f );
+   store[3] = floorf( store[3] + 0.5f );
+}
+
+static void
+emit_rnd(
+   struct x86_function *func,
+   unsigned xmm_save, 
+   unsigned xmm_dst )
+{
+   emit_func_call_dst(
+      func,
+      xmm_save,
+      xmm_dst,
+      rnd4f );
+}
+
+static void
+emit_rsqrt(
+   struct x86_function *func,
+   unsigned xmm_dst,
+   unsigned xmm_src )
+{
+#if HIGH_PRECISION
+   /* Although rsqrtps() and rcpps() are low precision on some/all SSE
+    * implementations, it is possible to improve its precision at
+    * fairly low cost, using a newton/raphson step, as below:
+    * 
+    * x1 = 2 * rcpps(a) - a * rcpps(a) * rcpps(a)
+    * x1 = 0.5 * rsqrtps(a) * [3.0 - (a * rsqrtps(a))* rsqrtps(a)]
+    *
+    * See: http://softwarecommunity.intel.com/articles/eng/1818.htm
+    */
+   {
+      struct x86_reg dst = make_xmm( xmm_dst );
+      struct x86_reg src = make_xmm( xmm_src );
+      struct x86_reg tmp0 = make_xmm( 2 );
+      struct x86_reg tmp1 = make_xmm( 3 );
+
+      assert( xmm_dst != xmm_src );
+      assert( xmm_dst != 2 && xmm_dst != 3 );
+      assert( xmm_src != 2 && xmm_src != 3 );
+
+      sse_movaps(  func, dst,  get_temp( TGSI_EXEC_TEMP_HALF_I, TGSI_EXEC_TEMP_HALF_C ) );
+      sse_movaps(  func, tmp0, get_temp( TGSI_EXEC_TEMP_THREE_I, TGSI_EXEC_TEMP_THREE_C ) );
+      sse_rsqrtps( func, tmp1, src  );
+      sse_mulps(   func, src,  tmp1 );
+      sse_mulps(   func, dst,  tmp1 );
+      sse_mulps(   func, src,  tmp1 );
+      sse_subps(   func, tmp0, src  );
+      sse_mulps(   func, dst,  tmp0 );
+   }
+#else
+   /* On Intel CPUs at least, this is only accurate to 12 bits -- not
+    * good enough.
+    */
+   sse_rsqrtps(
+      func,
+      make_xmm( xmm_dst ),
+      make_xmm( xmm_src ) );
+#endif
+}
+
+static void
+emit_setsign(
+   struct x86_function *func,
+   unsigned xmm )
+{
+   sse_orps(
+      func,
+      make_xmm( xmm ),
+      get_temp(
+         TGSI_EXEC_TEMP_80000000_I,
+         TGSI_EXEC_TEMP_80000000_C ) );
+}
+
+static void PIPE_CDECL
+sgn4f(
+   float *store )
+{
+   store[0] = store[0] < 0.0f ? -1.0f : store[0] > 0.0f ? 1.0f : 0.0f;
+   store[1] = store[1] < 0.0f ? -1.0f : store[1] > 0.0f ? 1.0f : 0.0f;
+   store[2] = store[2] < 0.0f ? -1.0f : store[2] > 0.0f ? 1.0f : 0.0f;
+   store[3] = store[3] < 0.0f ? -1.0f : store[3] > 0.0f ? 1.0f : 0.0f;
+}
+
+static void
+emit_sgn(
+   struct x86_function *func,
+   unsigned xmm_save, 
+   unsigned xmm_dst )
+{
+   emit_func_call_dst(
+      func,
+      xmm_save,
+      xmm_dst,
+      sgn4f );
+}
+
+static void PIPE_CDECL
+sin4f(
+   float *store )
+{
+   store[0] = sinf( store[0] );
+   store[1] = sinf( store[1] );
+   store[2] = sinf( store[2] );
+   store[3] = sinf( store[3] );
+}
+
+static void
+emit_sin (struct x86_function *func,
+          unsigned xmm_save, 
+          unsigned xmm_dst)
+{
+   emit_func_call_dst(
+      func,
+      xmm_save,
+      xmm_dst,
+      sin4f );
+}
+
+static void
+emit_sub(
+   struct x86_function *func,
+   unsigned xmm_dst,
+   unsigned xmm_src )
+{
+   sse_subps(
+      func,
+      make_xmm( xmm_dst ),
+      make_xmm( xmm_src ) );
+}
+
+/**
+ * Register fetch.
+ */
+
+static void
+emit_fetch(
+   struct x86_function *func,
+   unsigned xmm,
+   const struct tgsi_full_src_register *reg,
+   const unsigned chan_index )
+{
+   unsigned swizzle = tgsi_util_get_full_src_register_extswizzle( reg, chan_index );
+
+   switch (swizzle) {
+   case TGSI_EXTSWIZZLE_X:
+   case TGSI_EXTSWIZZLE_Y:
+   case TGSI_EXTSWIZZLE_Z:
+   case TGSI_EXTSWIZZLE_W:
+      switch (reg->SrcRegister.File) {
+      case TGSI_FILE_CONSTANT:
+         emit_const(
+            func,
+            xmm,
+            reg->SrcRegister.Index,
+            swizzle,
+            reg->SrcRegister.Indirect,
+            reg->SrcRegisterInd.File,
+            reg->SrcRegisterInd.Index );
+         break;
+
+      case TGSI_FILE_IMMEDIATE:
+         emit_immediate(
+            func,
+            xmm,
+            reg->SrcRegister.Index,
+            swizzle );
+         break;
+
+      case TGSI_FILE_INPUT:
+         emit_inputf(
+            func,
+            xmm,
+            reg->SrcRegister.Index,
+            swizzle );
+         break;
+
+      case TGSI_FILE_TEMPORARY:
+         emit_tempf(
+            func,
+            xmm,
+            reg->SrcRegister.Index,
+            swizzle );
+         break;
+
+      default:
+         assert( 0 );
+      }
+      break;
+
+   case TGSI_EXTSWIZZLE_ZERO:
+      emit_tempf(
+         func,
+         xmm,
+         TGSI_EXEC_TEMP_00000000_I,
+         TGSI_EXEC_TEMP_00000000_C );
+      break;
+
+   case TGSI_EXTSWIZZLE_ONE:
+      emit_tempf(
+         func,
+         xmm,
+         TEMP_ONE_I,
+         TEMP_ONE_C );
+      break;
+
+   default:
+      assert( 0 );
+   }
+
+   switch( tgsi_util_get_full_src_register_sign_mode( reg, chan_index ) ) {
+   case TGSI_UTIL_SIGN_CLEAR:
+      emit_abs( func, xmm );
+      break;
+
+   case TGSI_UTIL_SIGN_SET:
+      emit_setsign( func, xmm );
+      break;
+
+   case TGSI_UTIL_SIGN_TOGGLE:
+      emit_neg( func, xmm );
+      break;
+
+   case TGSI_UTIL_SIGN_KEEP:
+      break;
+   }
+}
+
+#define FETCH( FUNC, INST, XMM, INDEX, CHAN )\
+   emit_fetch( FUNC, XMM, &(INST).FullSrcRegisters[INDEX], CHAN )
+
+/**
+ * Register store.
+ */
+
+static void
+emit_store(
+   struct x86_function *func,
+   unsigned xmm,
+   const struct tgsi_full_dst_register *reg,
+   const struct tgsi_full_instruction *inst,
+   unsigned chan_index )
+{
+   switch( reg->DstRegister.File ) {
+   case TGSI_FILE_OUTPUT:
+      emit_output(
+         func,
+         xmm,
+         reg->DstRegister.Index,
+         chan_index );
+      break;
+
+   case TGSI_FILE_TEMPORARY:
+      emit_temps(
+         func,
+         xmm,
+         reg->DstRegister.Index,
+         chan_index );
+      break;
+
+   case TGSI_FILE_ADDRESS:
+      emit_addrs(
+         func,
+         xmm,
+         reg->DstRegister.Index,
+         chan_index );
+      break;
+
+   default:
+      assert( 0 );
+   }
+
+   switch( inst->Instruction.Saturate ) {
+   case TGSI_SAT_NONE:
+      break;
+
+   case TGSI_SAT_ZERO_ONE:
+      /* assert( 0 ); */
+      break;
+
+   case TGSI_SAT_MINUS_PLUS_ONE:
+      assert( 0 );
+      break;
+   }
+}
+
+#define STORE( FUNC, INST, XMM, INDEX, CHAN )\
+   emit_store( FUNC, XMM, &(INST).FullDstRegisters[INDEX], &(INST), CHAN )
+
+/**
+ * High-level instruction translators.
+ */
+
+static void
+emit_kil(
+   struct x86_function *func,
+   const struct tgsi_full_src_register *reg )
+{
+   unsigned uniquemask;
+   unsigned registers[4];
+   unsigned nextregister = 0;
+   unsigned firstchan = ~0;
+   unsigned chan_index;
+
+   /* This mask stores component bits that were already tested. Note that
+    * we test if the value is less than zero, so 1.0 and 0.0 need not to be
+    * tested. */
+   uniquemask = (1 << TGSI_EXTSWIZZLE_ZERO) | (1 << TGSI_EXTSWIZZLE_ONE);
+
+   FOR_EACH_CHANNEL( chan_index ) {
+      unsigned swizzle;
+
+      /* unswizzle channel */
+      swizzle = tgsi_util_get_full_src_register_extswizzle(
+         reg,
+         chan_index );
+
+      /* check if the component has not been already tested */
+      if( !(uniquemask & (1 << swizzle)) ) {
+         uniquemask |= 1 << swizzle;
+
+         /* allocate register */
+         registers[chan_index] = nextregister;
+         emit_fetch(
+            func,
+            nextregister,
+            reg,
+            chan_index );
+         nextregister++;
+
+         /* mark the first channel used */
+         if( firstchan == ~0 ) {
+            firstchan = chan_index;
+         }
+      }
+   }
+
+   x86_push(
+      func,
+      x86_make_reg( file_REG32, reg_AX ) );
+   x86_push(
+      func,
+      x86_make_reg( file_REG32, reg_DX ) );
+
+   FOR_EACH_CHANNEL( chan_index ) {
+      if( uniquemask & (1 << chan_index) ) {
+         sse_cmpps(
+            func,
+            make_xmm( registers[chan_index] ),
+            get_temp(
+               TGSI_EXEC_TEMP_00000000_I,
+               TGSI_EXEC_TEMP_00000000_C ),
+            cc_LessThan );
+
+         if( chan_index == firstchan ) {
+            sse_pmovmskb(
+               func,
+               x86_make_reg( file_REG32, reg_AX ),
+               make_xmm( registers[chan_index] ) );
+         }
+         else {
+            sse_pmovmskb(
+               func,
+               x86_make_reg( file_REG32, reg_DX ),
+               make_xmm( registers[chan_index] ) );
+            x86_or(
+               func,
+               x86_make_reg( file_REG32, reg_AX ),
+               x86_make_reg( file_REG32, reg_DX ) );
+         }
+      }
+   }
+
+   x86_or(
+      func,
+      get_temp(
+         TGSI_EXEC_TEMP_KILMASK_I,
+         TGSI_EXEC_TEMP_KILMASK_C ),
+      x86_make_reg( file_REG32, reg_AX ) );
+
+   x86_pop(
+      func,
+      x86_make_reg( file_REG32, reg_DX ) );
+   x86_pop(
+      func,
+      x86_make_reg( file_REG32, reg_AX ) );
+}
+
+
+static void
+emit_kilp(
+   struct x86_function *func )
+{
+   /* XXX todo / fix me */
+}
+
+
+static void
+emit_setcc(
+   struct x86_function *func,
+   struct tgsi_full_instruction *inst,
+   enum sse_cc cc )
+{
+   unsigned chan_index;
+
+   FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+      FETCH( func, *inst, 0, 0, chan_index );
+      FETCH( func, *inst, 1, 1, chan_index );
+      sse_cmpps(
+         func,
+         make_xmm( 0 ),
+         make_xmm( 1 ),
+         cc );
+      sse_andps(
+         func,
+         make_xmm( 0 ),
+         get_temp(
+            TEMP_ONE_I,
+            TEMP_ONE_C ) );
+      STORE( func, *inst, 0, 0, chan_index );
+   }
+}
+
+static void
+emit_cmp(
+   struct x86_function *func,
+   struct tgsi_full_instruction *inst )
+{
+   unsigned chan_index;
+
+   FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+      FETCH( func, *inst, 0, 0, chan_index );
+      FETCH( func, *inst, 1, 1, chan_index );
+      FETCH( func, *inst, 2, 2, chan_index );
+      sse_cmpps(
+         func,
+         make_xmm( 0 ),
+         get_temp(
+            TGSI_EXEC_TEMP_00000000_I,
+            TGSI_EXEC_TEMP_00000000_C ),
+         cc_LessThan );
+      sse_andps(
+         func,
+         make_xmm( 1 ),
+         make_xmm( 0 ) );
+      sse_andnps(
+         func,
+         make_xmm( 0 ),
+         make_xmm( 2 ) );
+      sse_orps(
+         func,
+         make_xmm( 0 ),
+         make_xmm( 1 ) );
+      STORE( func, *inst, 0, 0, chan_index );
+   }
+}
+
+static int
+emit_instruction(
+   struct x86_function *func,
+   struct tgsi_full_instruction *inst )
+{
+   unsigned chan_index;
+
+   switch (inst->Instruction.Opcode) {
+   case TGSI_OPCODE_ARL:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         emit_f2it( func, 0 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_MOV:
+   case TGSI_OPCODE_SWZ:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_LIT:
+      if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
+          IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W ) ) {
+         emit_tempf(
+            func,
+            0,
+            TEMP_ONE_I,
+            TEMP_ONE_C);
+         if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ) {
+            STORE( func, *inst, 0, 0, CHAN_X );
+         }
+         if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W ) ) {
+            STORE( func, *inst, 0, 0, CHAN_W );
+         }
+      }
+      if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ||
+          IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
+         if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ) {
+            FETCH( func, *inst, 0, 0, CHAN_X );
+            sse_maxps(
+               func,
+               make_xmm( 0 ),
+               get_temp(
+                  TGSI_EXEC_TEMP_00000000_I,
+                  TGSI_EXEC_TEMP_00000000_C ) );
+            STORE( func, *inst, 0, 0, CHAN_Y );
+         }
+         if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
+            /* XMM[1] = SrcReg[0].yyyy */
+            FETCH( func, *inst, 1, 0, CHAN_Y );
+            /* XMM[1] = max(XMM[1], 0) */
+            sse_maxps(
+               func,
+               make_xmm( 1 ),
+               get_temp(
+                  TGSI_EXEC_TEMP_00000000_I,
+                  TGSI_EXEC_TEMP_00000000_C ) );
+            /* XMM[2] = SrcReg[0].wwww */
+            FETCH( func, *inst, 2, 0, CHAN_W );
+            /* XMM[2] = min(XMM[2], 128.0) */
+            sse_minps(
+               func,
+               make_xmm( 2 ),
+               get_temp(
+                  TGSI_EXEC_TEMP_128_I,
+                  TGSI_EXEC_TEMP_128_C ) );
+            /* XMM[2] = max(XMM[2], -128.0) */
+            sse_maxps(
+               func,
+               make_xmm( 2 ),
+               get_temp(
+                  TGSI_EXEC_TEMP_MINUS_128_I,
+                  TGSI_EXEC_TEMP_MINUS_128_C ) );
+            emit_pow( func, 3, 1, 2 );
+            FETCH( func, *inst, 0, 0, CHAN_X );
+            sse_xorps(
+               func,
+               make_xmm( 2 ),
+               make_xmm( 2 ) );
+            sse_cmpps(
+               func,
+               make_xmm( 2 ),
+               make_xmm( 0 ),
+               cc_LessThanEqual );
+            sse_andps(
+               func,
+               make_xmm( 2 ),
+               make_xmm( 1 ) );
+            STORE( func, *inst, 2, 0, CHAN_Z );
+         }
+      }
+      break;
+
+   case TGSI_OPCODE_RCP:
+   /* TGSI_OPCODE_RECIP */
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      emit_rcp( func, 0, 0 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_RSQ:
+   /* TGSI_OPCODE_RECIPSQRT */
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      emit_rsqrt( func, 1, 0 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 1, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_EXP:
+      if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
+          IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ||
+          IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+         FETCH( func, *inst, 0, 0, CHAN_X );
+         if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
+             IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y )) {
+            emit_MOV( func, 1, 0 );
+            emit_flr( func, 2, 1 );
+            /* dst.x = ex2(floor(src.x)) */
+            if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X )) {
+               emit_MOV( func, 2, 1 );
+               emit_ex2( func, 3, 2 );
+               STORE( func, *inst, 2, 0, CHAN_X );
+            }
+            /* dst.y = src.x - floor(src.x) */
+            if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y )) {
+               emit_MOV( func, 2, 0 );
+               emit_sub( func, 2, 1 );
+               STORE( func, *inst, 2, 0, CHAN_Y );
+            }
+         }
+         /* dst.z = ex2(src.x) */
+         if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+            emit_ex2( func, 3, 0 );
+            STORE( func, *inst, 0, 0, CHAN_Z );
+         }
+      }
+      /* dst.w = 1.0 */
+      if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W )) {
+         emit_tempf( func, 0, TEMP_ONE_I, TEMP_ONE_C );
+         STORE( func, *inst, 0, 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_LOG:
+      if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
+          IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ||
+          IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+         FETCH( func, *inst, 0, 0, CHAN_X );
+         emit_abs( func, 0 );
+         emit_MOV( func, 1, 0 );
+         emit_lg2( func, 2, 1 );
+         /* dst.z = lg2(abs(src.x)) */
+         if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z )) {
+            STORE( func, *inst, 1, 0, CHAN_Z );
+         }
+         if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
+             IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y )) {
+            emit_flr( func, 2, 1 );
+            /* dst.x = floor(lg2(abs(src.x))) */
+            if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X )) {
+               STORE( func, *inst, 1, 0, CHAN_X );
+            }
+            /* dst.x = abs(src)/ex2(floor(lg2(abs(src.x)))) */
+            if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y )) {
+               emit_ex2( func, 2, 1 );
+               emit_rcp( func, 1, 1 );
+               emit_mul( func, 0, 1 );
+               STORE( func, *inst, 0, 0, CHAN_Y );
+            }
+         }
+      }
+      /* dst.w = 1.0 */
+      if (IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W )) {
+         emit_tempf( func, 0, TEMP_ONE_I, TEMP_ONE_C );
+         STORE( func, *inst, 0, 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_MUL:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         FETCH( func, *inst, 1, 1, chan_index );
+         emit_mul( func, 0, 1 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_ADD:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         FETCH( func, *inst, 1, 1, chan_index );
+         emit_add( func, 0, 1 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DP3:
+   /* TGSI_OPCODE_DOT3 */
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      FETCH( func, *inst, 1, 1, CHAN_X );
+      emit_mul( func, 0, 1 );
+      FETCH( func, *inst, 1, 0, CHAN_Y );
+      FETCH( func, *inst, 2, 1, CHAN_Y );
+      emit_mul( func, 1, 2 );
+      emit_add( func, 0, 1 );
+      FETCH( func, *inst, 1, 0, CHAN_Z );
+      FETCH( func, *inst, 2, 1, CHAN_Z );
+      emit_mul( func, 1, 2 );
+      emit_add( func, 0, 1 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DP4:
+   /* TGSI_OPCODE_DOT4 */
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      FETCH( func, *inst, 1, 1, CHAN_X );
+      emit_mul( func, 0, 1 );
+      FETCH( func, *inst, 1, 0, CHAN_Y );
+      FETCH( func, *inst, 2, 1, CHAN_Y );
+      emit_mul( func, 1, 2 );
+      emit_add( func, 0, 1 );
+      FETCH( func, *inst, 1, 0, CHAN_Z );
+      FETCH( func, *inst, 2, 1, CHAN_Z );
+      emit_mul(func, 1, 2 );
+      emit_add(func, 0, 1 );
+      FETCH( func, *inst, 1, 0, CHAN_W );
+      FETCH( func, *inst, 2, 1, CHAN_W );
+      emit_mul( func, 1, 2 );
+      emit_add( func, 0, 1 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DST:
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) {
+         emit_tempf(
+            func,
+            0,
+            TEMP_ONE_I,
+            TEMP_ONE_C );
+         STORE( func, *inst, 0, 0, CHAN_X );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) {
+         FETCH( func, *inst, 0, 0, CHAN_Y );
+         FETCH( func, *inst, 1, 1, CHAN_Y );
+         emit_mul( func, 0, 1 );
+         STORE( func, *inst, 0, 0, CHAN_Y );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) {
+         FETCH( func, *inst, 0, 0, CHAN_Z );
+         STORE( func, *inst, 0, 0, CHAN_Z );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W ) {
+         FETCH( func, *inst, 0, 1, CHAN_W );
+         STORE( func, *inst, 0, 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_MIN:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         FETCH( func, *inst, 1, 1, chan_index );
+         sse_minps(
+            func,
+            make_xmm( 0 ),
+            make_xmm( 1 ) );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_MAX:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         FETCH( func, *inst, 1, 1, chan_index );
+         sse_maxps(
+            func,
+            make_xmm( 0 ),
+            make_xmm( 1 ) );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SLT:
+   /* TGSI_OPCODE_SETLT */
+      emit_setcc( func, inst, cc_LessThan );
+      break;
+
+   case TGSI_OPCODE_SGE:
+   /* TGSI_OPCODE_SETGE */
+      emit_setcc( func, inst, cc_NotLessThan );
+      break;
+
+   case TGSI_OPCODE_MAD:
+   /* TGSI_OPCODE_MADD */
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         FETCH( func, *inst, 1, 1, chan_index );
+         FETCH( func, *inst, 2, 2, chan_index );
+         emit_mul( func, 0, 1 );
+         emit_add( func, 0, 2 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SUB:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         FETCH( func, *inst, 1, 1, chan_index );
+         emit_sub( func, 0, 1 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_LERP:
+   /* TGSI_OPCODE_LRP */
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         FETCH( func, *inst, 1, 1, chan_index );
+         FETCH( func, *inst, 2, 2, chan_index );
+         emit_sub( func, 1, 2 );
+         emit_mul( func, 0, 1 );
+         emit_add( func, 0, 2 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_CND:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_CND0:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_DOT2ADD:
+   /* TGSI_OPCODE_DP2A */
+      FETCH( func, *inst, 0, 0, CHAN_X );  /* xmm0 = src[0].x */
+      FETCH( func, *inst, 1, 1, CHAN_X );  /* xmm1 = src[1].x */
+      emit_mul( func, 0, 1 );              /* xmm0 = xmm0 * xmm1 */
+      FETCH( func, *inst, 1, 0, CHAN_Y );  /* xmm1 = src[0].y */
+      FETCH( func, *inst, 2, 1, CHAN_Y );  /* xmm2 = src[1].y */
+      emit_mul( func, 1, 2 );              /* xmm1 = xmm1 * xmm2 */
+      emit_add( func, 0, 1 );              /* xmm0 = xmm0 + xmm1 */
+      FETCH( func, *inst, 1, 2, CHAN_X );  /* xmm1 = src[2].x */
+      emit_add( func, 0, 1 );              /* xmm0 = xmm0 + xmm1 */
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );  /* dest[ch] = xmm0 */
+      }
+      break;
+
+   case TGSI_OPCODE_INDEX:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_NEGATE:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_FRAC:
+   /* TGSI_OPCODE_FRC */
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         emit_frc( func, 0, 0 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_CLAMP:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_FLOOR:
+   /* TGSI_OPCODE_FLR */
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         emit_flr( func, 0, 0 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_ROUND:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         emit_rnd( func, 0, 0 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_EXPBASE2:
+   /* TGSI_OPCODE_EX2 */
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      emit_ex2( func, 0, 0 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_LOGBASE2:
+   /* TGSI_OPCODE_LG2 */
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      emit_lg2( func, 0, 0 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_POWER:
+   /* TGSI_OPCODE_POW */
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      FETCH( func, *inst, 1, 1, CHAN_X );
+      emit_pow( func, 0, 0, 1 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_CROSSPRODUCT:
+   /* TGSI_OPCODE_XPD */
+      if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
+          IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ) {
+         FETCH( func, *inst, 1, 1, CHAN_Z );
+         FETCH( func, *inst, 3, 0, CHAN_Z );
+      }
+      if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) ||
+          IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
+         FETCH( func, *inst, 0, 0, CHAN_Y );
+         FETCH( func, *inst, 4, 1, CHAN_Y );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) {
+         emit_MOV( func, 2, 0 );
+         emit_mul( func, 2, 1 );
+         emit_MOV( func, 5, 3 );
+         emit_mul( func, 5, 4 );
+         emit_sub( func, 2, 5 );
+         STORE( func, *inst, 2, 0, CHAN_X );
+      }
+      if( IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) ||
+          IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) ) {
+         FETCH( func, *inst, 2, 1, CHAN_X );
+         FETCH( func, *inst, 5, 0, CHAN_X );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) {
+         emit_mul( func, 3, 2 );
+         emit_mul( func, 1, 5 );
+         emit_sub( func, 3, 1 );
+         STORE( func, *inst, 3, 0, CHAN_Y );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) {
+         emit_mul( func, 5, 4 );
+         emit_mul( func, 0, 2 );
+         emit_sub( func, 5, 0 );
+         STORE( func, *inst, 5, 0, CHAN_Z );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W ) {
+	 emit_tempf(
+	    func,
+	    0,
+	    TEMP_ONE_I,
+	    TEMP_ONE_C );
+         STORE( func, *inst, 0, 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_MULTIPLYMATRIX:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_ABS:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         emit_abs( func, 0) ;
+
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_RCC:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_DPH:
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      FETCH( func, *inst, 1, 1, CHAN_X );
+      emit_mul( func, 0, 1 );
+      FETCH( func, *inst, 1, 0, CHAN_Y );
+      FETCH( func, *inst, 2, 1, CHAN_Y );
+      emit_mul( func, 1, 2 );
+      emit_add( func, 0, 1 );
+      FETCH( func, *inst, 1, 0, CHAN_Z );
+      FETCH( func, *inst, 2, 1, CHAN_Z );
+      emit_mul( func, 1, 2 );
+      emit_add( func, 0, 1 );
+      FETCH( func, *inst, 1, 1, CHAN_W );
+      emit_add( func, 0, 1 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_COS:
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      emit_cos( func, 0, 0 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_DDX:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_DDY:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_KILP:
+      /* predicated kill */
+      emit_kilp( func );
+      return 0; /* XXX fix me */
+      break;
+
+   case TGSI_OPCODE_KIL:
+      /* conditional kill */
+      emit_kil( func, &inst->FullSrcRegisters[0] );
+      break;
+
+   case TGSI_OPCODE_PK2H:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_PK2US:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_PK4B:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_PK4UB:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_RFL:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_SEQ:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_SFL:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_SGT:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_SIN:
+      FETCH( func, *inst, 0, 0, CHAN_X );
+      emit_sin( func, 0, 0 );
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SLE:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_SNE:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_STR:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_TEX:
+      if (0) {
+	 /* Disable dummy texture code: 
+	  */
+	 emit_tempf(
+	    func,
+	    0,
+	    TEMP_ONE_I,
+	    TEMP_ONE_C );
+	 FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+	    STORE( func, *inst, 0, 0, chan_index );
+	 }
+      }
+      else {
+	 return 0;
+      }
+      break;
+
+   case TGSI_OPCODE_TXD:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_UP2H:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_UP2US:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_UP4B:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_UP4UB:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_X2D:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_ARA:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_ARR:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         emit_rnd( func, 0, 0 );
+         emit_f2it( func, 0 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_BRA:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_CAL:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_RET:
+      emit_ret( func );
+      break;
+
+   case TGSI_OPCODE_END:
+      break;
+
+   case TGSI_OPCODE_SSG:
+   /* TGSI_OPCODE_SGN */
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         emit_sgn( func, 0, 0 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_CMP:
+      emit_cmp (func, inst);
+      break;
+
+   case TGSI_OPCODE_SCS:
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_X ) {
+         FETCH( func, *inst, 0, 0, CHAN_X );
+         emit_cos( func, 0, 0 );
+         STORE( func, *inst, 0, 0, CHAN_X );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Y ) {
+         FETCH( func, *inst, 0, 0, CHAN_X );
+         emit_sin( func, 0, 0 );
+         STORE( func, *inst, 0, 0, CHAN_Y );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_Z ) {
+	 emit_tempf(
+	    func,
+	    0,
+	    TGSI_EXEC_TEMP_00000000_I,
+	    TGSI_EXEC_TEMP_00000000_C );
+         STORE( func, *inst, 0, 0, CHAN_Z );
+      }
+      IF_IS_DST0_CHANNEL_ENABLED( *inst, CHAN_W ) {
+	 emit_tempf(
+	    func,
+	    0,
+	    TEMP_ONE_I,
+	    TEMP_ONE_C );
+         STORE( func, *inst, 0, 0, CHAN_W );
+      }
+      break;
+
+   case TGSI_OPCODE_TXB:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_NRM:
+      /* fall-through */
+   case TGSI_OPCODE_NRM4:
+      /* 3 or 4-component normalization */
+      {
+         uint dims = (inst->Instruction.Opcode == TGSI_OPCODE_NRM) ? 3 : 4;
+         /* note: cannot use xmm regs 2/3 here (see emit_rsqrt() above) */
+         FETCH( func, *inst, 4, 0, CHAN_X );    /* xmm4 = src[0].x */
+         FETCH( func, *inst, 5, 0, CHAN_Y );    /* xmm5 = src[0].y */
+         FETCH( func, *inst, 6, 0, CHAN_Z );    /* xmm6 = src[0].z */
+         if (dims == 4) {
+            FETCH( func, *inst, 7, 0, CHAN_W ); /* xmm7 = src[0].w */
+         }
+         emit_MOV( func, 0, 4 );                /* xmm0 = xmm3 */
+         emit_mul( func, 0, 4 );                /* xmm0 *= xmm3 */
+         emit_MOV( func, 1, 5 );                /* xmm1 = xmm4 */
+         emit_mul( func, 1, 5 );                /* xmm1 *= xmm4 */
+         emit_add( func, 0, 1 );                /* xmm0 += xmm1 */
+         emit_MOV( func, 1, 6 );                /* xmm1 = xmm5 */
+         emit_mul( func, 1, 6 );                /* xmm1 *= xmm5 */
+         emit_add( func, 0, 1 );                /* xmm0 += xmm1 */
+         if (dims == 4) {
+            emit_MOV( func, 1, 7 );             /* xmm1 = xmm7 */
+            emit_mul( func, 1, 7 );             /* xmm1 *= xmm7 */
+            emit_add( func, 0, 0 );             /* xmm0 += xmm1 */
+         }
+         emit_rsqrt( func, 1, 0 );              /* xmm1 = 1/sqrt(xmm0) */
+         FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+            if (chan_index < dims) {
+               emit_mul( func, 4+chan_index, 1); /* xmm[4+ch] *= xmm1 */
+               STORE( func, *inst, 4+chan_index, 0, chan_index );
+            }
+         }
+      }
+      break;
+
+   case TGSI_OPCODE_DIV:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_DP2:
+      FETCH( func, *inst, 0, 0, CHAN_X );  /* xmm0 = src[0].x */
+      FETCH( func, *inst, 1, 1, CHAN_X );  /* xmm1 = src[1].x */
+      emit_mul( func, 0, 1 );              /* xmm0 = xmm0 * xmm1 */
+      FETCH( func, *inst, 1, 0, CHAN_Y );  /* xmm1 = src[0].y */
+      FETCH( func, *inst, 2, 1, CHAN_Y );  /* xmm2 = src[1].y */
+      emit_mul( func, 1, 2 );              /* xmm1 = xmm1 * xmm2 */
+      emit_add( func, 0, 1 );              /* xmm0 = xmm0 + xmm1 */
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         STORE( func, *inst, 0, 0, chan_index );  /* dest[ch] = xmm0 */
+      }
+      break;
+
+   case TGSI_OPCODE_TXL:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_BRK:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_IF:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_LOOP:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_REP:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_ELSE:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_ENDIF:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_ENDLOOP:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_ENDREP:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_PUSHA:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_POPA:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_CEIL:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_I2F:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_NOT:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_TRUNC:
+      FOR_EACH_DST0_ENABLED_CHANNEL( *inst, chan_index ) {
+         FETCH( func, *inst, 0, 0, chan_index );
+         emit_f2it( func, 0 );
+         emit_i2f( func, 0 );
+         STORE( func, *inst, 0, 0, chan_index );
+      }
+      break;
+
+   case TGSI_OPCODE_SHL:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_SHR:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_AND:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_OR:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_MOD:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_XOR:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_SAD:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_TXF:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_TXQ:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_CONT:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_EMIT:
+      return 0;
+      break;
+
+   case TGSI_OPCODE_ENDPRIM:
+      return 0;
+      break;
+
+   default:
+      return 0;
+   }
+   
+   return 1;
+}
+
+static void
+emit_declaration(
+   struct x86_function *func,
+   struct tgsi_full_declaration *decl )
+{
+   if( decl->Declaration.File == TGSI_FILE_INPUT ) {
+      unsigned first, last, mask;
+      unsigned i, j;
+
+      first = decl->DeclarationRange.First;
+      last = decl->DeclarationRange.Last;
+      mask = decl->Declaration.UsageMask;
+
+      for( i = first; i <= last; i++ ) {
+         for( j = 0; j < NUM_CHANNELS; j++ ) {
+            if( mask & (1 << j) ) {
+               switch( decl->Declaration.Interpolate ) {
+               case TGSI_INTERPOLATE_CONSTANT:
+                  emit_coef_a0( func, 0, i, j );
+                  emit_inputs( func, 0, i, j );
+                  break;
+
+               case TGSI_INTERPOLATE_LINEAR:
+                  emit_tempf( func, 0, 0, TGSI_SWIZZLE_X );
+                  emit_coef_dadx( func, 1, i, j );
+                  emit_tempf( func, 2, 0, TGSI_SWIZZLE_Y );
+                  emit_coef_dady( func, 3, i, j );
+                  emit_mul( func, 0, 1 );    /* x * dadx */
+                  emit_coef_a0( func, 4, i, j );
+                  emit_mul( func, 2, 3 );    /* y * dady */
+                  emit_add( func, 0, 4 );    /* x * dadx + a0 */
+                  emit_add( func, 0, 2 );    /* x * dadx + y * dady + a0 */
+                  emit_inputs( func, 0, i, j );
+                  break;
+
+               case TGSI_INTERPOLATE_PERSPECTIVE:
+                  emit_tempf( func, 0, 0, TGSI_SWIZZLE_X );
+                  emit_coef_dadx( func, 1, i, j );
+                  emit_tempf( func, 2, 0, TGSI_SWIZZLE_Y );
+                  emit_coef_dady( func, 3, i, j );
+                  emit_mul( func, 0, 1 );    /* x * dadx */
+                  emit_tempf( func, 4, 0, TGSI_SWIZZLE_W );
+                  emit_coef_a0( func, 5, i, j );
+                  emit_rcp( func, 4, 4 );    /* 1.0 / w */
+                  emit_mul( func, 2, 3 );    /* y * dady */
+                  emit_add( func, 0, 5 );    /* x * dadx + a0 */
+                  emit_add( func, 0, 2 );    /* x * dadx + y * dady + a0 */
+                  emit_mul( func, 0, 4 );    /* (x * dadx + y * dady + a0) / w */
+                  emit_inputs( func, 0, i, j );
+                  break;
+
+               default:
+                  assert( 0 );
+		  break;
+               }
+            }
+         }
+      }
+   }
+}
+
+static void aos_to_soa( struct x86_function *func, 
+                        uint arg_aos,
+                        uint arg_soa, 
+                        uint arg_num, 
+                        uint arg_stride )
+{
+   struct x86_reg soa_input = x86_make_reg( file_REG32, reg_AX );
+   struct x86_reg aos_input = x86_make_reg( file_REG32, reg_BX );
+   struct x86_reg num_inputs = x86_make_reg( file_REG32, reg_CX );
+   struct x86_reg stride = x86_make_reg( file_REG32, reg_DX );
+   int inner_loop;
+
+
+   /* Save EBX */
+   x86_push( func, x86_make_reg( file_REG32, reg_BX ) );
+
+   x86_mov( func, aos_input,  x86_fn_arg( func, arg_aos ) );
+   x86_mov( func, soa_input,  x86_fn_arg( func, arg_soa ) );
+   x86_mov( func, num_inputs, x86_fn_arg( func, arg_num ) );
+   x86_mov( func, stride,     x86_fn_arg( func, arg_stride ) );
+
+   /* do */
+   inner_loop = x86_get_label( func );
+   {
+      x86_push( func, aos_input );
+      sse_movlps( func, make_xmm( 0 ), x86_make_disp( aos_input, 0 ) );
+      sse_movlps( func, make_xmm( 3 ), x86_make_disp( aos_input, 8 ) );
+      x86_add( func, aos_input, stride );
+      sse_movhps( func, make_xmm( 0 ), x86_make_disp( aos_input, 0 ) );
+      sse_movhps( func, make_xmm( 3 ), x86_make_disp( aos_input, 8 ) );
+      x86_add( func, aos_input, stride );
+      sse_movlps( func, make_xmm( 1 ), x86_make_disp( aos_input, 0 ) );
+      sse_movlps( func, make_xmm( 4 ), x86_make_disp( aos_input, 8 ) );
+      x86_add( func, aos_input, stride );
+      sse_movhps( func, make_xmm( 1 ), x86_make_disp( aos_input, 0 ) );
+      sse_movhps( func, make_xmm( 4 ), x86_make_disp( aos_input, 8 ) );
+      x86_pop( func, aos_input );
+
+      sse_movaps( func, make_xmm( 2 ), make_xmm( 0 ) );
+      sse_movaps( func, make_xmm( 5 ), make_xmm( 3 ) );
+      sse_shufps( func, make_xmm( 0 ), make_xmm( 1 ), 0x88 );
+      sse_shufps( func, make_xmm( 2 ), make_xmm( 1 ), 0xdd );
+      sse_shufps( func, make_xmm( 3 ), make_xmm( 4 ), 0x88 );
+      sse_shufps( func, make_xmm( 5 ), make_xmm( 4 ), 0xdd );
+
+      sse_movups( func, x86_make_disp( soa_input, 0 ), make_xmm( 0 ) );
+      sse_movups( func, x86_make_disp( soa_input, 16 ), make_xmm( 2 ) );
+      sse_movups( func, x86_make_disp( soa_input, 32 ), make_xmm( 3 ) );
+      sse_movups( func, x86_make_disp( soa_input, 48 ), make_xmm( 5 ) );
+
+      /* Advance to next input */
+      x86_lea( func, aos_input, x86_make_disp(aos_input, 16) );
+      x86_lea( func, soa_input, x86_make_disp(soa_input, 64) );
+   }
+   /* while --num_inputs */
+   x86_dec( func, num_inputs );
+   x86_jcc( func, cc_NE, inner_loop );
+
+   /* Restore EBX */
+   x86_pop( func, aos_input );
+}
+
+static void soa_to_aos( struct x86_function *func, uint aos, uint soa, uint num, uint stride )
+{
+   struct x86_reg soa_output;
+   struct x86_reg aos_output;
+   struct x86_reg num_outputs;
+   struct x86_reg temp;
+   int inner_loop;
+
+   soa_output = x86_make_reg( file_REG32, reg_AX );
+   aos_output = x86_make_reg( file_REG32, reg_BX );
+   num_outputs = x86_make_reg( file_REG32, reg_CX );
+   temp = x86_make_reg( file_REG32, reg_DX );
+
+   /* Save EBX */
+   x86_push( func, aos_output );
+
+   x86_mov( func, soa_output, x86_fn_arg( func, soa ) );
+   x86_mov( func, aos_output, x86_fn_arg( func, aos ) );
+   x86_mov( func, num_outputs, x86_fn_arg( func, num ) );
+
+   /* do */
+   inner_loop = x86_get_label( func );
+   {
+      sse_movups( func, make_xmm( 0 ), x86_make_disp( soa_output, 0 ) );
+      sse_movups( func, make_xmm( 1 ), x86_make_disp( soa_output, 16 ) );
+      sse_movups( func, make_xmm( 3 ), x86_make_disp( soa_output, 32 ) );
+      sse_movups( func, make_xmm( 4 ), x86_make_disp( soa_output, 48 ) );
+
+      sse_movaps( func, make_xmm( 2 ), make_xmm( 0 ) );
+      sse_movaps( func, make_xmm( 5 ), make_xmm( 3 ) );
+      sse_unpcklps( func, make_xmm( 0 ), make_xmm( 1 ) );
+      sse_unpckhps( func, make_xmm( 2 ), make_xmm( 1 ) );
+      sse_unpcklps( func, make_xmm( 3 ), make_xmm( 4 ) );
+      sse_unpckhps( func, make_xmm( 5 ), make_xmm( 4 ) );
+
+      x86_mov( func, temp, x86_fn_arg( func, stride ) );
+      x86_push( func, aos_output );
+      sse_movlps( func, x86_make_disp( aos_output, 0 ), make_xmm( 0 ) );
+      sse_movlps( func, x86_make_disp( aos_output, 8 ), make_xmm( 3 ) );
+      x86_add( func, aos_output, temp );
+      sse_movhps( func, x86_make_disp( aos_output, 0 ), make_xmm( 0 ) );
+      sse_movhps( func, x86_make_disp( aos_output, 8 ), make_xmm( 3 ) );
+      x86_add( func, aos_output, temp );
+      sse_movlps( func, x86_make_disp( aos_output, 0 ), make_xmm( 2 ) );
+      sse_movlps( func, x86_make_disp( aos_output, 8 ), make_xmm( 5 ) );
+      x86_add( func, aos_output, temp );
+      sse_movhps( func, x86_make_disp( aos_output, 0 ), make_xmm( 2 ) );
+      sse_movhps( func, x86_make_disp( aos_output, 8 ), make_xmm( 5 ) );
+      x86_pop( func, aos_output );
+
+      /* Advance to next output */
+      x86_lea( func, aos_output, x86_make_disp(aos_output, 16) );
+      x86_lea( func, soa_output, x86_make_disp(soa_output, 64) );
+   }
+   /* while --num_outputs */
+   x86_dec( func, num_outputs );
+   x86_jcc( func, cc_NE, inner_loop );
+
+   /* Restore EBX */
+   x86_pop( func, aos_output );
+}
+
+/**
+ * Translate a TGSI vertex/fragment shader to SSE2 code.
+ * Slightly different things are done for vertex vs. fragment shaders.
+ *
+ * Note that fragment shaders are responsible for interpolating shader
+ * inputs. Because on x86 we have only 4 GP registers, and here we
+ * have 5 shader arguments (input, output, const, temp and coef), the
+ * code is split into two phases -- DECLARATION and INSTRUCTION phase.
+ * GP register holding the output argument is aliased with the coeff
+ * argument, as outputs are not needed in the DECLARATION phase.
+ *
+ * \param tokens  the TGSI input shader
+ * \param func  the output SSE code/function
+ * \param immediates  buffer to place immediates, later passed to SSE func
+ * \param return  1 for success, 0 if translation failed
+ */
+unsigned
+tgsi_emit_sse2(
+   const struct tgsi_token *tokens,
+   struct x86_function *func,
+   float (*immediates)[4],
+   boolean do_swizzles )
+{
+   struct tgsi_parse_context parse;
+   boolean instruction_phase = FALSE;
+   unsigned ok = 1;
+   uint num_immediates = 0;
+
+   util_init_math();
+
+   func->csr = func->store;
+
+   tgsi_parse_init( &parse, tokens );
+
+   /* Can't just use EDI, EBX without save/restoring them:
+    */
+   x86_push(
+      func,
+      get_immediate_base() );
+
+   x86_push(
+      func,
+      get_temp_base() );
+
+
+   /*
+    * Different function args for vertex/fragment shaders:
+    */
+   if (parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_FRAGMENT) {
+      /* DECLARATION phase, do not load output argument. */
+      x86_mov(
+         func,
+         get_input_base(),
+         x86_fn_arg( func, 1 ) );
+      /* skipping outputs argument here */
+      x86_mov(
+         func,
+         get_const_base(),
+         x86_fn_arg( func, 3 ) );
+      x86_mov(
+         func,
+         get_temp_base(),
+         x86_fn_arg( func, 4 ) );
+      x86_mov(
+         func,
+         get_coef_base(),
+         x86_fn_arg( func, 5 ) );
+      x86_mov(
+         func,
+         get_immediate_base(),
+         x86_fn_arg( func, 6 ) );
+   }
+   else {
+      assert(parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_VERTEX);
+
+      if (do_swizzles)
+         aos_to_soa( func, 
+                     6,         /* aos_input */
+                     1,         /* machine->input */
+                     7,         /* num_inputs */
+                     8 );       /* input_stride */
+
+      x86_mov(
+         func,
+         get_input_base(),
+         x86_fn_arg( func, 1 ) );
+      x86_mov(
+         func,
+         get_output_base(),
+         x86_fn_arg( func, 2 ) );
+      x86_mov(
+         func,
+         get_const_base(),
+         x86_fn_arg( func, 3 ) );
+      x86_mov(
+         func,
+         get_temp_base(),
+         x86_fn_arg( func, 4 ) );
+      x86_mov(
+         func,
+         get_immediate_base(),
+         x86_fn_arg( func, 5 ) );
+   }
+
+   while( !tgsi_parse_end_of_tokens( &parse ) && ok ) {
+      tgsi_parse_token( &parse );
+
+      switch( parse.FullToken.Token.Type ) {
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         if (parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_FRAGMENT) {
+            emit_declaration(
+               func,
+               &parse.FullToken.FullDeclaration );
+         }
+         break;
+
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         if (parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_FRAGMENT) {
+            if( !instruction_phase ) {
+               /* INSTRUCTION phase, overwrite coeff with output. */
+               instruction_phase = TRUE;
+               x86_mov(
+                  func,
+                  get_output_base(),
+                  x86_fn_arg( func, 2 ) );
+            }
+         }
+
+         ok = emit_instruction(
+            func,
+            &parse.FullToken.FullInstruction );
+
+	 if (!ok) {
+	    debug_printf("failed to translate tgsi opcode %d to SSE (%s)\n", 
+			 parse.FullToken.FullInstruction.Instruction.Opcode,
+                         parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_VERTEX ?
+                         "vertex shader" : "fragment shader");
+	 }
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         /* simply copy the immediate values into the next immediates[] slot */
+         {
+            const uint size = parse.FullToken.FullImmediate.Immediate.NrTokens - 1;
+            uint i;
+            assert(size <= 4);
+            assert(num_immediates < TGSI_EXEC_NUM_IMMEDIATES);
+            for( i = 0; i < size; i++ ) {
+               immediates[num_immediates][i] =
+		  parse.FullToken.FullImmediate.u.ImmediateFloat32[i].Float;
+            }
+#if 0
+            debug_printf("SSE FS immediate[%d] = %f %f %f %f\n",
+                   num_immediates,
+                   immediates[num_immediates][0],
+                   immediates[num_immediates][1],
+                   immediates[num_immediates][2],
+                   immediates[num_immediates][3]);
+#endif
+            num_immediates++;
+         }
+         break;
+
+      default:
+	 ok = 0;
+         assert( 0 );
+      }
+   }
+
+   if (parse.FullHeader.Processor.Processor == TGSI_PROCESSOR_VERTEX) {
+      if (do_swizzles)
+         soa_to_aos( func, 9, 2, 10, 11 );
+   }
+
+   /* Can't just use EBX, EDI without save/restoring them:
+    */
+   x86_pop(
+      func,
+      get_temp_base() );
+
+   x86_pop(
+      func,
+      get_immediate_base() );
+
+   emit_ret( func );
+
+   tgsi_parse_free( &parse );
+
+   return ok;
+}
+
+#endif /* PIPE_ARCH_X86 */
+
diff --git a/src/gallium/auxiliary/tgsi/tgsi_sse2.h b/src/gallium/auxiliary/tgsi/tgsi_sse2.h
new file mode 100644
index 00000000000..af838b2a25b
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_sse2.h
@@ -0,0 +1,49 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef TGSI_SSE2_H
+#define TGSI_SSE2_H
+
+#if defined __cplusplus
+extern "C" {
+#endif
+
+struct tgsi_token;
+struct x86_function;
+
+unsigned
+tgsi_emit_sse2(
+   const struct tgsi_token *tokens,
+   struct x86_function *function,
+   float (*immediates)[4],
+   boolean do_swizzles );
+
+#if defined __cplusplus
+}
+#endif
+
+#endif /* TGSI_SSE2_H */
diff --git a/src/gallium/auxiliary/tgsi/tgsi_text.c b/src/gallium/auxiliary/tgsi/tgsi_text.c
new file mode 100644
index 00000000000..1e822fbbea1
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_text.c
@@ -0,0 +1,1097 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_debug.h"
+#include "tgsi_text.h"
+#include "tgsi_build.h"
+#include "tgsi_info.h"
+#include "tgsi_parse.h"
+#include "tgsi_sanity.h"
+#include "tgsi_util.h"
+
+static boolean is_alpha_underscore( const char *cur )
+{
+   return
+      (*cur >= 'a' && *cur <= 'z') ||
+      (*cur >= 'A' && *cur <= 'Z') ||
+      *cur == '_';
+}
+
+static boolean is_digit( const char *cur )
+{
+   return *cur >= '0' && *cur <= '9';
+}
+
+static boolean is_digit_alpha_underscore( const char *cur )
+{
+   return is_digit( cur ) || is_alpha_underscore( cur );
+}
+
+static boolean uprcase( char c )
+{
+   if (c >= 'a' && c <= 'z')
+      return c += 'A' - 'a';
+   return c;
+}
+
+static boolean str_match_no_case( const char **pcur, const char *str )
+{
+   const char *cur = *pcur;
+
+   while (*str != '\0' && *str == uprcase( *cur )) {
+      str++;
+      cur++;
+   }
+   if (*str == '\0') {
+      *pcur = cur;
+      return TRUE;
+   }
+   return FALSE;
+}
+
+/* Eat zero or more whitespaces.
+ */
+static void eat_opt_white( const char **pcur )
+{
+   while (**pcur == ' ' || **pcur == '\t' || **pcur == '\n')
+      (*pcur)++;
+}
+
+/* Eat one or more whitespaces.
+ * Return TRUE if at least one whitespace eaten.
+ */
+static boolean eat_white( const char **pcur )
+{
+   const char *cur = *pcur;
+
+   eat_opt_white( pcur );
+   return *pcur > cur;
+}
+
+/* Parse unsigned integer.
+ * No checks for overflow.
+ */
+static boolean parse_uint( const char **pcur, uint *val )
+{
+   const char *cur = *pcur;
+
+   if (is_digit( cur )) {
+      *val = *cur++ - '0';
+      while (is_digit( cur ))
+         *val = *val * 10 + *cur++ - '0';
+      *pcur = cur;
+      return TRUE;
+   }
+   return FALSE;
+}
+
+/* Parse floating point.
+ */
+static boolean parse_float( const char **pcur, float *val )
+{
+   const char *cur = *pcur;
+   boolean integral_part = FALSE;
+   boolean fractional_part = FALSE;
+
+   *val = (float) atof( cur );
+
+   if (*cur == '-' || *cur == '+')
+      cur++;
+   if (is_digit( cur )) {
+      cur++;
+      integral_part = TRUE;
+      while (is_digit( cur ))
+         cur++;
+   }
+   if (*cur == '.') {
+      cur++;
+      if (is_digit( cur )) {
+         cur++;
+         fractional_part = TRUE;
+         while (is_digit( cur ))
+            cur++;
+      }
+   }
+   if (!integral_part && !fractional_part)
+      return FALSE;
+   if (uprcase( *cur ) == 'E') {
+      cur++;
+      if (*cur == '-' || *cur == '+')
+         cur++;
+      if (is_digit( cur )) {
+         cur++;
+         while (is_digit( cur ))
+            cur++;
+      }
+      else
+         return FALSE;
+   }
+   *pcur = cur;
+   return TRUE;
+}
+
+struct translate_ctx
+{
+   const char *text;
+   const char *cur;
+   struct tgsi_token *tokens;
+   struct tgsi_token *tokens_cur;
+   struct tgsi_token *tokens_end;
+   struct tgsi_header *header;
+};
+
+static void report_error( struct translate_ctx *ctx, const char *msg )
+{
+   debug_printf( "\nError: %s", msg );
+}
+
+/* Parse shader header.
+ * Return TRUE for one of the following headers.
+ *    FRAG1.1
+ *    GEOM1.1
+ *    VERT1.1
+ */
+static boolean parse_header( struct translate_ctx *ctx )
+{
+   uint processor;
+
+   if (str_match_no_case( &ctx->cur, "FRAG1.1" ))
+      processor = TGSI_PROCESSOR_FRAGMENT;
+   else if (str_match_no_case( &ctx->cur, "VERT1.1" ))
+      processor = TGSI_PROCESSOR_VERTEX;
+   else if (str_match_no_case( &ctx->cur, "GEOM1.1" ))
+      processor = TGSI_PROCESSOR_GEOMETRY;
+   else {
+      report_error( ctx, "Unknown header" );
+      return FALSE;
+   }
+
+   if (ctx->tokens_cur >= ctx->tokens_end)
+      return FALSE;
+   *(struct tgsi_version *) ctx->tokens_cur++ = tgsi_build_version();
+
+   if (ctx->tokens_cur >= ctx->tokens_end)
+      return FALSE;
+   ctx->header = (struct tgsi_header *) ctx->tokens_cur++;
+   *ctx->header = tgsi_build_header();
+
+   if (ctx->tokens_cur >= ctx->tokens_end)
+      return FALSE;
+   *(struct tgsi_processor *) ctx->tokens_cur++ = tgsi_build_processor( processor, ctx->header );
+
+   return TRUE;
+}
+
+static boolean parse_label( struct translate_ctx *ctx, uint *val )
+{
+   const char *cur = ctx->cur;
+
+   if (parse_uint( &cur, val )) {
+      eat_opt_white( &cur );
+      if (*cur == ':') {
+         cur++;
+         ctx->cur = cur;
+         return TRUE;
+      }
+   }
+   return FALSE;
+}
+
+static const char *file_names[TGSI_FILE_COUNT] =
+{
+   "NULL",
+   "CONST",
+   "IN",
+   "OUT",
+   "TEMP",
+   "SAMP",
+   "ADDR",
+   "IMM"
+};
+
+static boolean
+parse_file( const char **pcur, uint *file )
+{
+   uint i;
+
+   for (i = 0; i < TGSI_FILE_COUNT; i++) {
+      const char *cur = *pcur;
+
+      if (str_match_no_case( &cur, file_names[i] )) {
+         if (!is_digit_alpha_underscore( cur )) {
+            *pcur = cur;
+            *file = i;
+            return TRUE;
+         }
+      }
+   }
+   return FALSE;
+}
+
+static boolean
+parse_opt_writemask(
+   struct translate_ctx *ctx,
+   uint *writemask )
+{
+   const char *cur;
+
+   cur = ctx->cur;
+   eat_opt_white( &cur );
+   if (*cur == '.') {
+      cur++;
+      *writemask = TGSI_WRITEMASK_NONE;
+      eat_opt_white( &cur );
+      if (uprcase( *cur ) == 'X') {
+         cur++;
+         *writemask |= TGSI_WRITEMASK_X;
+      }
+      if (uprcase( *cur ) == 'Y') {
+         cur++;
+         *writemask |= TGSI_WRITEMASK_Y;
+      }
+      if (uprcase( *cur ) == 'Z') {
+         cur++;
+         *writemask |= TGSI_WRITEMASK_Z;
+      }
+      if (uprcase( *cur ) == 'W') {
+         cur++;
+         *writemask |= TGSI_WRITEMASK_W;
+      }
+
+      if (*writemask == TGSI_WRITEMASK_NONE) {
+         report_error( ctx, "Writemask expected" );
+         return FALSE;
+      }
+
+      ctx->cur = cur;
+   }
+   else {
+      *writemask = TGSI_WRITEMASK_XYZW;
+   }
+   return TRUE;
+}
+
+/* <register_file_bracket> ::= <file> `['
+ */
+static boolean
+parse_register_file_bracket(
+   struct translate_ctx *ctx,
+   uint *file )
+{
+   if (!parse_file( &ctx->cur, file )) {
+      report_error( ctx, "Unknown register file" );
+      return FALSE;
+   }
+   eat_opt_white( &ctx->cur );
+   if (*ctx->cur != '[') {
+      report_error( ctx, "Expected `['" );
+      return FALSE;
+   }
+   ctx->cur++;
+   return TRUE;
+}
+
+/* <register_file_bracket_index> ::= <register_file_bracket> <uint>
+ */
+static boolean
+parse_register_file_bracket_index(
+   struct translate_ctx *ctx,
+   uint *file,
+   int *index )
+{
+   uint uindex;
+
+   if (!parse_register_file_bracket( ctx, file ))
+      return FALSE;
+   eat_opt_white( &ctx->cur );
+   if (!parse_uint( &ctx->cur, &uindex )) {
+      report_error( ctx, "Expected literal unsigned integer" );
+      return FALSE;
+   }
+   *index = (int) uindex;
+   return TRUE;
+}
+
+/* Parse destination register operand.
+ *    <register_dst> ::= <register_file_bracket_index> `]'
+ */
+static boolean
+parse_register_dst(
+   struct translate_ctx *ctx,
+   uint *file,
+   int *index )
+{
+   if (!parse_register_file_bracket_index( ctx, file, index ))
+      return FALSE;
+   eat_opt_white( &ctx->cur );
+   if (*ctx->cur != ']') {
+      report_error( ctx, "Expected `]'" );
+      return FALSE;
+   }
+   ctx->cur++;
+   return TRUE;
+}
+
+/* Parse source register operand.
+ *    <register_src> ::= <register_file_bracket_index> `]' |
+ *                       <register_file_bracket> <register_dst> `]' |
+ *                       <register_file_bracket> <register_dst> `+' <uint> `]' |
+ *                       <register_file_bracket> <register_dst> `-' <uint> `]'
+ */
+static boolean
+parse_register_src(
+   struct translate_ctx *ctx,
+   uint *file,
+   int *index,
+   uint *ind_file,
+   int *ind_index )
+{
+   const char *cur;
+   uint uindex;
+
+   if (!parse_register_file_bracket( ctx, file ))
+      return FALSE;
+   eat_opt_white( &ctx->cur );
+   cur = ctx->cur;
+   if (parse_file( &cur, ind_file )) {
+      if (!parse_register_dst( ctx, ind_file, ind_index ))
+         return FALSE;
+      eat_opt_white( &ctx->cur );
+      if (*ctx->cur == '+' || *ctx->cur == '-') {
+         boolean negate;
+
+         negate = *ctx->cur == '-';
+         ctx->cur++;
+         eat_opt_white( &ctx->cur );
+         if (!parse_uint( &ctx->cur, &uindex )) {
+            report_error( ctx, "Expected literal unsigned integer" );
+            return FALSE;
+         }
+         if (negate)
+            *index = -(int) uindex;
+         else
+            *index = (int) uindex;
+      }
+      else {
+         *index = 0;
+      }
+   }
+   else {
+      if (!parse_uint( &ctx->cur, &uindex )) {
+         report_error( ctx, "Expected literal unsigned integer" );
+         return FALSE;
+      }
+      *index = (int) uindex;
+      *ind_file = TGSI_FILE_NULL;
+      *ind_index = 0;
+   }
+   eat_opt_white( &ctx->cur );
+   if (*ctx->cur != ']') {
+      report_error( ctx, "Expected `]'" );
+      return FALSE;
+   }
+   ctx->cur++;
+   return TRUE;
+}
+
+/* Parse register declaration.
+ *    <register_dcl> ::= <register_file_bracket_index> `]' |
+ *                       <register_file_bracket_index> `..' <index> `]'
+ */
+static boolean
+parse_register_dcl(
+   struct translate_ctx *ctx,
+   uint *file,
+   int *first,
+   int *last )
+{
+   if (!parse_register_file_bracket_index( ctx, file, first ))
+      return FALSE;
+   eat_opt_white( &ctx->cur );
+   if (ctx->cur[0] == '.' && ctx->cur[1] == '.') {
+      uint uindex;
+
+      ctx->cur += 2;
+      eat_opt_white( &ctx->cur );
+      if (!parse_uint( &ctx->cur, &uindex )) {
+         report_error( ctx, "Expected literal integer" );
+         return FALSE;
+      }
+      *last = (int) uindex;
+      eat_opt_white( &ctx->cur );
+   }
+   else {
+      *last = *first;
+   }
+   if (*ctx->cur != ']') {
+      report_error( ctx, "Expected `]' or `..'" );
+      return FALSE;
+   }
+   ctx->cur++;
+   return TRUE;
+}
+
+static const char *modulate_names[TGSI_MODULATE_COUNT] =
+{
+   "_1X",
+   "_2X",
+   "_4X",
+   "_8X",
+   "_D2",
+   "_D4",
+   "_D8"
+};
+
+static boolean
+parse_dst_operand(
+   struct translate_ctx *ctx,
+   struct tgsi_full_dst_register *dst )
+{
+   uint file;
+   int index;
+   uint writemask;
+   const char *cur;
+
+   if (!parse_register_dst( ctx, &file, &index ))
+      return FALSE;
+
+   cur = ctx->cur;
+   eat_opt_white( &cur );
+   if (*cur == '_') {
+      uint i;
+
+      for (i = 0; i < TGSI_MODULATE_COUNT; i++) {
+         if (str_match_no_case( &cur, modulate_names[i] )) {
+            if (!is_digit_alpha_underscore( cur )) {
+               dst->DstRegisterExtModulate.Modulate = i;
+               ctx->cur = cur;
+               break;
+            }
+         }
+      }
+   }
+
+   if (!parse_opt_writemask( ctx, &writemask ))
+      return FALSE;
+
+   dst->DstRegister.File = file;
+   dst->DstRegister.Index = index;
+   dst->DstRegister.WriteMask = writemask;
+   return TRUE;
+}
+
+static boolean
+parse_optional_swizzle(
+   struct translate_ctx *ctx,
+   uint swizzle[4],
+   boolean *parsed_swizzle,
+   boolean *parsed_extswizzle )
+{
+   const char *cur = ctx->cur;
+
+   *parsed_swizzle = FALSE;
+   *parsed_extswizzle = FALSE;
+
+   eat_opt_white( &cur );
+   if (*cur == '.') {
+      uint i;
+
+      cur++;
+      eat_opt_white( &cur );
+      for (i = 0; i < 4; i++) {
+         if (uprcase( *cur ) == 'X')
+            swizzle[i] = TGSI_SWIZZLE_X;
+         else if (uprcase( *cur ) == 'Y')
+            swizzle[i] = TGSI_SWIZZLE_Y;
+         else if (uprcase( *cur ) == 'Z')
+            swizzle[i] = TGSI_SWIZZLE_Z;
+         else if (uprcase( *cur ) == 'W')
+            swizzle[i] = TGSI_SWIZZLE_W;
+         else {
+            if (*cur == '0')
+               swizzle[i] = TGSI_EXTSWIZZLE_ZERO;
+            else if (*cur == '1')
+               swizzle[i] = TGSI_EXTSWIZZLE_ONE;
+            else {
+               report_error( ctx, "Expected register swizzle component `x', `y', `z', `w', `0' or `1'" );
+               return FALSE;
+            }
+            *parsed_extswizzle = TRUE;
+         }
+         cur++;
+      }
+      *parsed_swizzle = TRUE;
+      ctx->cur = cur;
+   }
+   return TRUE;
+}
+
+static boolean
+parse_src_operand(
+   struct translate_ctx *ctx,
+   struct tgsi_full_src_register *src )
+{
+   const char *cur;
+   float value;
+   uint file;
+   int index;
+   uint ind_file;
+   int ind_index;
+   uint swizzle[4];
+   boolean parsed_swizzle;
+   boolean parsed_extswizzle;
+
+   if (*ctx->cur == '-') {
+      cur = ctx->cur;
+      cur++;
+      eat_opt_white( &cur );
+      if (*cur == '(') {
+         cur++;
+         src->SrcRegisterExtMod.Negate = 1;
+         eat_opt_white( &cur );
+         ctx->cur = cur;
+      }
+   }
+
+   if (*ctx->cur == '|') {
+      ctx->cur++;
+      eat_opt_white( &ctx->cur );
+      src->SrcRegisterExtMod.Absolute = 1;
+   }
+
+   if (*ctx->cur == '-') {
+      ctx->cur++;
+      eat_opt_white( &ctx->cur );
+      src->SrcRegister.Negate = 1;
+   }
+
+   cur = ctx->cur;
+   if (parse_float( &cur, &value )) {
+      if (value == 2.0f) {
+         eat_opt_white( &cur );
+         if (*cur != '*') {
+            report_error( ctx, "Expected `*'" );
+            return FALSE;
+         }
+         cur++;
+         if (*cur != '(') {
+            report_error( ctx, "Expected `('" );
+            return FALSE;
+         }
+         cur++;
+         src->SrcRegisterExtMod.Scale2X = 1;
+         eat_opt_white( &cur );
+         ctx->cur = cur;
+      }
+   }
+
+   if (*ctx->cur == '(') {
+      ctx->cur++;
+      eat_opt_white( &ctx->cur );
+      src->SrcRegisterExtMod.Bias = 1;
+   }
+
+   cur = ctx->cur;
+   if (parse_float( &cur, &value )) {
+      if (value == 1.0f) {
+         eat_opt_white( &cur );
+         if (*cur != '-') {
+            report_error( ctx, "Expected `-'" );
+            return FALSE;
+         }
+         cur++;
+         if (*cur != '(') {
+            report_error( ctx, "Expected `('" );
+            return FALSE;
+         }
+         cur++;
+         src->SrcRegisterExtMod.Complement = 1;
+         eat_opt_white( &cur );
+         ctx->cur = cur;
+      }
+   }
+
+   if (!parse_register_src( ctx, &file, &index, &ind_file, &ind_index ))
+      return FALSE;
+   src->SrcRegister.File = file;
+   src->SrcRegister.Index = index;
+   if (ind_file != TGSI_FILE_NULL) {
+      src->SrcRegister.Indirect = 1;
+      src->SrcRegisterInd.File = ind_file;
+      src->SrcRegisterInd.Index = ind_index;
+   }
+
+   /* Parse optional swizzle.
+    */
+   if (parse_optional_swizzle( ctx, swizzle, &parsed_swizzle, &parsed_extswizzle )) {
+      if (parsed_extswizzle) {
+         assert( parsed_swizzle );
+
+         src->SrcRegisterExtSwz.ExtSwizzleX = swizzle[0];
+         src->SrcRegisterExtSwz.ExtSwizzleY = swizzle[1];
+         src->SrcRegisterExtSwz.ExtSwizzleZ = swizzle[2];
+         src->SrcRegisterExtSwz.ExtSwizzleW = swizzle[3];
+      }
+      else if (parsed_swizzle) {
+         src->SrcRegister.SwizzleX = swizzle[0];
+         src->SrcRegister.SwizzleY = swizzle[1];
+         src->SrcRegister.SwizzleZ = swizzle[2];
+         src->SrcRegister.SwizzleW = swizzle[3];
+      }
+   }
+
+   if (src->SrcRegisterExtMod.Complement) {
+      eat_opt_white( &ctx->cur );
+      if (*ctx->cur != ')') {
+         report_error( ctx, "Expected `)'" );
+         return FALSE;
+      }
+      ctx->cur++;
+   }
+
+   if (src->SrcRegisterExtMod.Bias) {
+      eat_opt_white( &ctx->cur );
+      if (*ctx->cur != ')') {
+         report_error( ctx, "Expected `)'" );
+         return FALSE;
+      }
+      ctx->cur++;
+      eat_opt_white( &ctx->cur );
+      if (*ctx->cur != '-') {
+         report_error( ctx, "Expected `-'" );
+         return FALSE;
+      }
+      ctx->cur++;
+      eat_opt_white( &ctx->cur );
+      if (!parse_float( &ctx->cur, &value )) {
+         report_error( ctx, "Expected literal floating point" );
+         return FALSE;
+      }
+      if (value != 0.5f) {
+         report_error( ctx, "Expected 0.5" );
+         return FALSE;
+      }
+   }
+
+   if (src->SrcRegisterExtMod.Scale2X) {
+      eat_opt_white( &ctx->cur );
+      if (*ctx->cur != ')') {
+         report_error( ctx, "Expected `)'" );
+         return FALSE;
+      }
+      ctx->cur++;
+   }
+
+   if (src->SrcRegisterExtMod.Absolute) {
+      eat_opt_white( &ctx->cur );
+      if (*ctx->cur != '|') {
+         report_error( ctx, "Expected `|'" );
+         return FALSE;
+      }
+      ctx->cur++;
+   }
+
+   if (src->SrcRegisterExtMod.Negate) {
+      eat_opt_white( &ctx->cur );
+      if (*ctx->cur != ')') {
+         report_error( ctx, "Expected `)'" );
+         return FALSE;
+      }
+      ctx->cur++;
+   }
+
+   return TRUE;
+}
+
+static const char *texture_names[TGSI_TEXTURE_COUNT] =
+{
+   "UNKNOWN",
+   "1D",
+   "2D",
+   "3D",
+   "CUBE",
+   "RECT",
+   "SHADOW1D",
+   "SHADOW2D",
+   "SHADOWRECT"
+};
+
+static boolean
+parse_instruction(
+   struct translate_ctx *ctx,
+   boolean has_label )
+{
+   uint i;
+   uint saturate = TGSI_SAT_NONE;
+   const struct tgsi_opcode_info *info;
+   struct tgsi_full_instruction inst;
+   uint advance;
+
+   /* Parse instruction name.
+    */
+   eat_opt_white( &ctx->cur );
+   for (i = 0; i < TGSI_OPCODE_LAST; i++) {
+      const char *cur = ctx->cur;
+
+      info = tgsi_get_opcode_info( i );
+      if (str_match_no_case( &cur, info->mnemonic )) {
+         if (str_match_no_case( &cur, "_SATNV" ))
+            saturate = TGSI_SAT_MINUS_PLUS_ONE;
+         else if (str_match_no_case( &cur, "_SAT" ))
+            saturate = TGSI_SAT_ZERO_ONE;
+
+         if (info->num_dst + info->num_src + info->is_tex == 0) {
+            if (!is_digit_alpha_underscore( cur )) {
+               ctx->cur = cur;
+               break;
+            }
+         }
+         else if (*cur == '\0' || eat_white( &cur )) {
+            ctx->cur = cur;
+            break;
+         }
+      }
+   }
+   if (i == TGSI_OPCODE_LAST) {
+      if (has_label)
+         report_error( ctx, "Unknown opcode" );
+      else
+         report_error( ctx, "Expected `DCL', `IMM' or a label" );
+      return FALSE;
+   }
+
+   inst = tgsi_default_full_instruction();
+   inst.Instruction.Opcode = i;
+   inst.Instruction.Saturate = saturate;
+   inst.Instruction.NumDstRegs = info->num_dst;
+   inst.Instruction.NumSrcRegs = info->num_src;
+
+   /* Parse instruction operands.
+    */
+   for (i = 0; i < info->num_dst + info->num_src + info->is_tex; i++) {
+      if (i > 0) {
+         eat_opt_white( &ctx->cur );
+         if (*ctx->cur != ',') {
+            report_error( ctx, "Expected `,'" );
+            return FALSE;
+         }
+         ctx->cur++;
+         eat_opt_white( &ctx->cur );
+      }
+
+      if (i < info->num_dst) {
+         if (!parse_dst_operand( ctx, &inst.FullDstRegisters[i] ))
+            return FALSE;
+      }
+      else if (i < info->num_dst + info->num_src) {
+         if (!parse_src_operand( ctx, &inst.FullSrcRegisters[i - info->num_dst] ))
+            return FALSE;
+      }
+      else {
+         uint j;
+
+         for (j = 0; j < TGSI_TEXTURE_COUNT; j++) {
+            if (str_match_no_case( &ctx->cur, texture_names[j] )) {
+               if (!is_digit_alpha_underscore( ctx->cur )) {
+                  inst.InstructionExtTexture.Texture = j;
+                  break;
+               }
+            }
+         }
+         if (j == TGSI_TEXTURE_COUNT) {
+            report_error( ctx, "Expected texture target" );
+            return FALSE;
+         }
+      }
+   }
+
+   if (info->is_branch) {
+      uint target;
+
+      eat_opt_white( &ctx->cur );
+      if (*ctx->cur != ':') {
+         report_error( ctx, "Expected `:'" );
+         return FALSE;
+      }
+      ctx->cur++;
+      eat_opt_white( &ctx->cur );
+      if (!parse_uint( &ctx->cur, &target )) {
+         report_error( ctx, "Expected a label" );
+         return FALSE;
+      }
+      inst.InstructionExtLabel.Label = target;
+   }
+
+   advance = tgsi_build_full_instruction(
+      &inst,
+      ctx->tokens_cur,
+      ctx->header,
+      (uint) (ctx->tokens_end - ctx->tokens_cur) );
+   if (advance == 0)
+      return FALSE;
+   ctx->tokens_cur += advance;
+
+   return TRUE;
+}
+
+static const char *semantic_names[TGSI_SEMANTIC_COUNT] =
+{
+   "POSITION",
+   "COLOR",
+   "BCOLOR",
+   "FOG",
+   "PSIZE",
+   "GENERIC",
+   "NORMAL"
+};
+
+static const char *interpolate_names[TGSI_INTERPOLATE_COUNT] =
+{
+   "CONSTANT",
+   "LINEAR",
+   "PERSPECTIVE"
+};
+
+static boolean parse_declaration( struct translate_ctx *ctx )
+{
+   struct tgsi_full_declaration decl;
+   uint file;
+   int first;
+   int last;
+   uint writemask;
+   const char *cur;
+   uint advance;
+
+   if (!eat_white( &ctx->cur )) {
+      report_error( ctx, "Syntax error" );
+      return FALSE;
+   }
+   if (!parse_register_dcl( ctx, &file, &first, &last ))
+      return FALSE;
+   if (!parse_opt_writemask( ctx, &writemask ))
+      return FALSE;
+
+   decl = tgsi_default_full_declaration();
+   decl.Declaration.File = file;
+   decl.Declaration.UsageMask = writemask;
+   decl.DeclarationRange.First = first;
+   decl.DeclarationRange.Last = last;
+
+   cur = ctx->cur;
+   eat_opt_white( &cur );
+   if (*cur == ',') {
+      uint i;
+
+      cur++;
+      eat_opt_white( &cur );
+      for (i = 0; i < TGSI_SEMANTIC_COUNT; i++) {
+         if (str_match_no_case( &cur, semantic_names[i] )) {
+            const char *cur2 = cur;
+            uint index;
+
+            if (is_digit_alpha_underscore( cur ))
+               continue;
+            eat_opt_white( &cur2 );
+            if (*cur2 == '[') {
+               cur2++;
+               eat_opt_white( &cur2 );
+               if (!parse_uint( &cur2, &index )) {
+                  report_error( ctx, "Expected literal integer" );
+                  return FALSE;
+               }
+               eat_opt_white( &cur2 );
+               if (*cur2 != ']') {
+                  report_error( ctx, "Expected `]'" );
+                  return FALSE;
+               }
+               cur2++;
+
+               decl.Semantic.SemanticIndex = index;
+
+               cur = cur2;
+            }
+
+            decl.Declaration.Semantic = 1;
+            decl.Semantic.SemanticName = i;
+
+            ctx->cur = cur;
+            break;
+         }
+      }
+   }
+
+   cur = ctx->cur;
+   eat_opt_white( &cur );
+   if (*cur == ',') {
+      uint i;
+
+      cur++;
+      eat_opt_white( &cur );
+      for (i = 0; i < TGSI_INTERPOLATE_COUNT; i++) {
+         if (str_match_no_case( &cur, interpolate_names[i] )) {
+            if (is_digit_alpha_underscore( cur ))
+               continue;
+            decl.Declaration.Interpolate = i;
+
+            ctx->cur = cur;
+            break;
+         }
+      }
+      if (i == TGSI_INTERPOLATE_COUNT) {
+         report_error( ctx, "Expected semantic or interpolate attribute" );
+         return FALSE;
+      }
+   }
+
+   advance = tgsi_build_full_declaration(
+      &decl,
+      ctx->tokens_cur,
+      ctx->header,
+      (uint) (ctx->tokens_end - ctx->tokens_cur) );
+   if (advance == 0)
+      return FALSE;
+   ctx->tokens_cur += advance;
+
+   return TRUE;
+}
+
+static boolean parse_immediate( struct translate_ctx *ctx )
+{
+   struct tgsi_full_immediate imm;
+   uint i;
+   float values[4];
+   uint advance;
+
+   if (!eat_white( &ctx->cur )) {
+      report_error( ctx, "Syntax error" );
+      return FALSE;
+   }
+   if (!str_match_no_case( &ctx->cur, "FLT32" ) || is_digit_alpha_underscore( ctx->cur )) {
+      report_error( ctx, "Expected `FLT32'" );
+      return FALSE;
+   }
+   eat_opt_white( &ctx->cur );
+   if (*ctx->cur != '{') {
+      report_error( ctx, "Expected `{'" );
+      return FALSE;
+   }
+   ctx->cur++;
+   for (i = 0; i < 4; i++) {
+      eat_opt_white( &ctx->cur );
+      if (i > 0) {
+         if (*ctx->cur != ',') {
+            report_error( ctx, "Expected `,'" );
+            return FALSE;
+         }
+         ctx->cur++;
+         eat_opt_white( &ctx->cur );
+      }
+      if (!parse_float( &ctx->cur, &values[i] )) {
+         report_error( ctx, "Expected literal floating point" );
+         return FALSE;
+      }
+   }
+   eat_opt_white( &ctx->cur );
+   if (*ctx->cur != '}') {
+      report_error( ctx, "Expected `}'" );
+      return FALSE;
+   }
+   ctx->cur++;
+
+   imm = tgsi_default_full_immediate();
+   imm.Immediate.NrTokens += 4;
+   imm.Immediate.DataType = TGSI_IMM_FLOAT32;
+   imm.u.Pointer = values;
+
+   advance = tgsi_build_full_immediate(
+      &imm,
+      ctx->tokens_cur,
+      ctx->header,
+      (uint) (ctx->tokens_end - ctx->tokens_cur) );
+   if (advance == 0)
+      return FALSE;
+   ctx->tokens_cur += advance;
+
+   return TRUE;
+}
+
+static boolean translate( struct translate_ctx *ctx )
+{
+   eat_opt_white( &ctx->cur );
+   if (!parse_header( ctx ))
+      return FALSE;
+
+   while (*ctx->cur != '\0') {
+      uint label_val = 0;
+
+      if (!eat_white( &ctx->cur )) {
+         report_error( ctx, "Syntax error" );
+         return FALSE;
+      }
+
+      if (*ctx->cur == '\0')
+         break;
+
+      if (parse_label( ctx, &label_val )) {
+         if (!parse_instruction( ctx, TRUE ))
+            return FALSE;
+      }
+      else if (str_match_no_case( &ctx->cur, "DCL" )) {
+         if (!parse_declaration( ctx ))
+            return FALSE;
+      }
+      else if (str_match_no_case( &ctx->cur, "IMM" )) {
+         if (!parse_immediate( ctx ))
+            return FALSE;
+      }
+      else if (!parse_instruction( ctx, FALSE )) {
+         return FALSE;
+      }
+   }
+
+   return TRUE;
+}
+
+boolean
+tgsi_text_translate(
+   const char *text,
+   struct tgsi_token *tokens,
+   uint num_tokens )
+{
+   struct translate_ctx ctx;
+
+   ctx.text = text;
+   ctx.cur = text;
+   ctx.tokens = tokens;
+   ctx.tokens_cur = tokens;
+   ctx.tokens_end = tokens + num_tokens;
+
+   if (!translate( &ctx ))
+      return FALSE;
+
+   return tgsi_sanity_check( tokens );
+}
diff --git a/src/gallium/auxiliary/tgsi/tgsi_text.h b/src/gallium/auxiliary/tgsi/tgsi_text.h
new file mode 100644
index 00000000000..8eeeeef1402
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_text.h
@@ -0,0 +1,47 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef TGSI_TEXT_H
+#define TGSI_TEXT_H
+
+#include "pipe/p_shader_tokens.h"
+
+#if defined __cplusplus
+extern "C" {
+#endif
+
+boolean
+tgsi_text_translate(
+   const char *text,
+   struct tgsi_token *tokens,
+   uint num_tokens );
+
+#if defined __cplusplus
+}
+#endif
+
+#endif /* TGSI_TEXT_H */
diff --git a/src/gallium/auxiliary/tgsi/tgsi_transform.c b/src/gallium/auxiliary/tgsi/tgsi_transform.c
new file mode 100644
index 00000000000..ea87da31e50
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_transform.c
@@ -0,0 +1,200 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * TGSI program transformation utility.
+ *
+ * Authors:  Brian Paul
+ */
+
+#include "pipe/p_debug.h"
+
+#include "tgsi_transform.h"
+
+
+
+static void
+emit_instruction(struct tgsi_transform_context *ctx,
+                 const struct tgsi_full_instruction *inst)
+{
+   uint ti = ctx->ti;
+
+   ti += tgsi_build_full_instruction(inst,
+                                     ctx->tokens_out + ti,
+                                     ctx->header,
+                                     ctx->max_tokens_out - ti);
+   ctx->ti = ti;
+}
+
+
+static void
+emit_declaration(struct tgsi_transform_context *ctx,
+                 const struct tgsi_full_declaration *decl)
+{
+   uint ti = ctx->ti;
+
+   ti += tgsi_build_full_declaration(decl,
+                                     ctx->tokens_out + ti,
+                                     ctx->header,
+                                     ctx->max_tokens_out - ti);
+   ctx->ti = ti;
+}
+
+
+static void
+emit_immediate(struct tgsi_transform_context *ctx,
+               const struct tgsi_full_immediate *imm)
+{
+   uint ti = ctx->ti;
+
+   ti += tgsi_build_full_immediate(imm,
+                                   ctx->tokens_out + ti,
+                                   ctx->header,
+                                   ctx->max_tokens_out - ti);
+   ctx->ti = ti;
+}
+
+
+
+/**
+ * Apply user-defined transformations to the input shader to produce
+ * the output shader.
+ * For example, a register search-and-replace operation could be applied
+ * by defining a transform_instruction() callback that examined and changed
+ * the instruction src/dest regs.
+ *
+ * \return number of tokens emitted
+ */
+int
+tgsi_transform_shader(const struct tgsi_token *tokens_in,
+                      struct tgsi_token *tokens_out,
+                      uint max_tokens_out,
+                      struct tgsi_transform_context *ctx)
+{
+   uint procType;
+
+   /* input shader */
+   struct tgsi_parse_context parse;
+
+   /* output shader */
+   struct tgsi_processor *processor;
+
+
+   /**
+    ** callback context init
+    **/
+   ctx->emit_instruction = emit_instruction;
+   ctx->emit_declaration = emit_declaration;
+   ctx->emit_immediate = emit_immediate;
+   ctx->tokens_out = tokens_out;
+   ctx->max_tokens_out = max_tokens_out;
+
+
+   /**
+    ** Setup to begin parsing input shader
+    **/
+   if (tgsi_parse_init( &parse, tokens_in ) != TGSI_PARSE_OK) {
+      debug_printf("tgsi_parse_init() failed in tgsi_transform_shader()!\n");
+      return -1;
+   }
+   procType = parse.FullHeader.Processor.Processor;
+   assert(procType == TGSI_PROCESSOR_FRAGMENT ||
+          procType == TGSI_PROCESSOR_VERTEX ||
+          procType == TGSI_PROCESSOR_GEOMETRY);
+
+
+   /**
+    **  Setup output shader
+    **/
+   *(struct tgsi_version *) &tokens_out[0] = tgsi_build_version();
+
+   ctx->header = (struct tgsi_header *) (tokens_out + 1);
+   *ctx->header = tgsi_build_header();
+
+   processor = (struct tgsi_processor *) (tokens_out + 2);
+   *processor = tgsi_build_processor( procType, ctx->header );
+
+   ctx->ti = 3;
+
+
+   /**
+    ** Loop over incoming program tokens/instructions
+    */
+   while( !tgsi_parse_end_of_tokens( &parse ) ) {
+
+      tgsi_parse_token( &parse );
+
+      switch( parse.FullToken.Token.Type ) {
+      case TGSI_TOKEN_TYPE_INSTRUCTION:
+         {
+            struct tgsi_full_instruction *fullinst
+               = &parse.FullToken.FullInstruction;
+
+            if (ctx->transform_instruction)
+               ctx->transform_instruction(ctx, fullinst);
+            else
+               ctx->emit_instruction(ctx, fullinst);
+         }
+         break;
+
+      case TGSI_TOKEN_TYPE_DECLARATION:
+         {
+            struct tgsi_full_declaration *fulldecl
+               = &parse.FullToken.FullDeclaration;
+
+            if (ctx->transform_declaration)
+               ctx->transform_declaration(ctx, fulldecl);
+            else
+               ctx->emit_declaration(ctx, fulldecl);
+         }
+         break;
+
+      case TGSI_TOKEN_TYPE_IMMEDIATE:
+         {
+            struct tgsi_full_immediate *fullimm
+               = &parse.FullToken.FullImmediate;
+
+            if (ctx->transform_immediate)
+               ctx->transform_immediate(ctx, fullimm);
+            else
+               ctx->emit_immediate(ctx, fullimm);
+         }
+         break;
+
+      default:
+         assert( 0 );
+      }
+   }
+
+   if (ctx->epilog) {
+      ctx->epilog(ctx);
+   }
+
+   tgsi_parse_free (&parse);
+
+   return ctx->ti;
+}
diff --git a/src/gallium/auxiliary/tgsi/tgsi_transform.h b/src/gallium/auxiliary/tgsi/tgsi_transform.h
new file mode 100644
index 00000000000..a121adbaef4
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_transform.h
@@ -0,0 +1,92 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef TGSI_TRANSFORM_H
+#define TGSI_TRANSFORM_H
+
+
+#include "pipe/p_shader_tokens.h"
+#include "tgsi/tgsi_parse.h"
+#include "tgsi/tgsi_build.h"
+
+
+
+/**
+ * Subclass this to add caller-specific data
+ */
+struct tgsi_transform_context
+{
+/**** PUBLIC ***/
+
+   /**
+    * User-defined callbacks invoked per instruction.
+    */
+   void (*transform_instruction)(struct tgsi_transform_context *ctx,
+                                 struct tgsi_full_instruction *inst);
+
+   void (*transform_declaration)(struct tgsi_transform_context *ctx,
+                                 struct tgsi_full_declaration *decl);
+
+   void (*transform_immediate)(struct tgsi_transform_context *ctx,
+                               struct tgsi_full_immediate *imm);
+
+   /**
+    * Called at end of input program to allow caller to append extra
+    * instructions.  Return number of tokens emitted.
+    */
+   void (*epilog)(struct tgsi_transform_context *ctx);
+
+
+/*** PRIVATE ***/
+
+   /**
+    * These are setup by tgsi_transform_shader() and cannot be overridden.
+    * Meant to be called from in the above user callback functions.
+    */
+   void (*emit_instruction)(struct tgsi_transform_context *ctx,
+                            const struct tgsi_full_instruction *inst);
+   void (*emit_declaration)(struct tgsi_transform_context *ctx,
+                            const struct tgsi_full_declaration *decl);
+   void (*emit_immediate)(struct tgsi_transform_context *ctx,
+                          const struct tgsi_full_immediate *imm);
+
+   struct tgsi_header *header;
+   uint max_tokens_out;
+   struct tgsi_token *tokens_out;
+   uint ti;
+};
+
+
+
+extern int
+tgsi_transform_shader(const struct tgsi_token *tokens_in,
+                      struct tgsi_token *tokens_out,
+                      uint max_tokens_out,
+                      struct tgsi_transform_context *ctx);
+
+
+#endif /* TGSI_TRANSFORM_H */
diff --git a/src/gallium/auxiliary/tgsi/tgsi_util.c b/src/gallium/auxiliary/tgsi/tgsi_util.c
new file mode 100644
index 00000000000..50101a9bb0c
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_util.c
@@ -0,0 +1,299 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "pipe/p_debug.h"
+#include "pipe/p_shader_tokens.h"
+#include "tgsi_parse.h"
+#include "tgsi_build.h"
+#include "tgsi_util.h"
+
+union pointer_hack
+{
+   void *pointer;
+   uint64_t uint64;
+};
+
+void *
+tgsi_align_128bit(
+   void *unaligned )
+{
+   union pointer_hack ph;
+
+   ph.uint64 = 0;
+   ph.pointer = unaligned;
+   ph.uint64 = (ph.uint64 + 15) & ~15;
+   return ph.pointer;
+}
+
+unsigned
+tgsi_util_get_src_register_swizzle(
+   const struct tgsi_src_register *reg,
+   unsigned component )
+{
+   switch( component ) {
+   case 0:
+      return reg->SwizzleX;
+   case 1:
+      return reg->SwizzleY;
+   case 2:
+      return reg->SwizzleZ;
+   case 3:
+      return reg->SwizzleW;
+   default:
+      assert( 0 );
+   }
+   return 0;
+}
+
+unsigned
+tgsi_util_get_src_register_extswizzle(
+   const struct tgsi_src_register_ext_swz *reg,
+   unsigned component )
+{
+   switch( component ) {
+   case 0:
+      return reg->ExtSwizzleX;
+   case 1:
+      return reg->ExtSwizzleY;
+   case 2:
+      return reg->ExtSwizzleZ;
+   case 3:
+      return reg->ExtSwizzleW;
+   default:
+      assert( 0 );
+   }
+   return 0;
+}
+
+unsigned
+tgsi_util_get_full_src_register_extswizzle(
+   const struct tgsi_full_src_register  *reg,
+   unsigned component )
+{
+   unsigned swizzle;
+
+   /*
+    * First, calculate  the   extended swizzle for a given channel. This will give
+    * us either a channel index into the simple swizzle or  a constant 1 or   0.
+    */
+   swizzle = tgsi_util_get_src_register_extswizzle(
+      &reg->SrcRegisterExtSwz,
+      component );
+
+   assert (TGSI_SWIZZLE_X == TGSI_EXTSWIZZLE_X);
+   assert (TGSI_SWIZZLE_Y == TGSI_EXTSWIZZLE_Y);
+   assert (TGSI_SWIZZLE_Z == TGSI_EXTSWIZZLE_Z);
+   assert (TGSI_SWIZZLE_W == TGSI_EXTSWIZZLE_W);
+   assert (TGSI_EXTSWIZZLE_ZERO > TGSI_SWIZZLE_W);
+   assert (TGSI_EXTSWIZZLE_ONE > TGSI_SWIZZLE_W);
+
+   /*
+    * Second, calculate the simple  swizzle  for   the   unswizzled channel index.
+    * Leave the constants intact, they are   not   affected by the   simple swizzle.
+    */
+   if( swizzle <= TGSI_SWIZZLE_W ) {
+      swizzle = tgsi_util_get_src_register_swizzle(
+         &reg->SrcRegister,
+         swizzle );
+   }
+
+   return swizzle;
+}
+
+void
+tgsi_util_set_src_register_swizzle(
+   struct tgsi_src_register *reg,
+   unsigned swizzle,
+   unsigned component )
+{
+   switch( component ) {
+   case 0:
+      reg->SwizzleX = swizzle;
+      break;
+   case 1:
+      reg->SwizzleY = swizzle;
+      break;
+   case 2:
+      reg->SwizzleZ = swizzle;
+      break;
+   case 3:
+      reg->SwizzleW = swizzle;
+      break;
+   default:
+      assert( 0 );
+   }
+}
+
+void
+tgsi_util_set_src_register_extswizzle(
+   struct tgsi_src_register_ext_swz *reg,
+   unsigned swizzle,
+   unsigned component )
+{
+   switch( component ) {
+   case 0:
+      reg->ExtSwizzleX = swizzle;
+      break;
+   case 1:
+      reg->ExtSwizzleY = swizzle;
+      break;
+   case 2:
+      reg->ExtSwizzleZ = swizzle;
+      break;
+   case 3:
+      reg->ExtSwizzleW = swizzle;
+      break;
+   default:
+      assert( 0 );
+   }
+}
+
+unsigned
+tgsi_util_get_src_register_extnegate(
+   const  struct tgsi_src_register_ext_swz *reg,
+   unsigned component )
+{
+   switch( component ) {
+   case 0:
+      return reg->NegateX;
+   case 1:
+      return reg->NegateY;
+   case 2:
+      return reg->NegateZ;
+   case 3:
+      return reg->NegateW;
+   default:
+      assert( 0 );
+   }
+   return 0;
+}
+
+void
+tgsi_util_set_src_register_extnegate(
+   struct tgsi_src_register_ext_swz *reg,
+   unsigned negate,
+   unsigned component )
+{
+   switch( component ) {
+   case 0:
+      reg->NegateX = negate;
+      break;
+   case 1:
+      reg->NegateY = negate;
+      break;
+   case 2:
+      reg->NegateZ = negate;
+      break;
+   case 3:
+      reg->NegateW = negate;
+      break;
+   default:
+      assert( 0 );
+   }
+}
+
+unsigned
+tgsi_util_get_full_src_register_sign_mode(
+   const struct  tgsi_full_src_register *reg,
+   unsigned component )
+{
+   unsigned sign_mode;
+
+   if( reg->SrcRegisterExtMod.Absolute ) {
+      /* Consider only the post-abs negation. */
+
+      if( reg->SrcRegisterExtMod.Negate ) {
+         sign_mode = TGSI_UTIL_SIGN_SET;
+      }
+      else {
+         sign_mode = TGSI_UTIL_SIGN_CLEAR;
+      }
+   }
+   else {
+      /* Accumulate the three negations. */
+
+      unsigned negate;
+
+      negate = reg->SrcRegister.Negate;
+      if( tgsi_util_get_src_register_extnegate( &reg->SrcRegisterExtSwz, component ) ) {
+         negate = !negate;
+      }
+      if( reg->SrcRegisterExtMod.Negate ) {
+         negate = !negate;
+      }
+
+      if( negate ) {
+         sign_mode = TGSI_UTIL_SIGN_TOGGLE;
+      }
+      else {
+         sign_mode = TGSI_UTIL_SIGN_KEEP;
+      }
+   }
+
+   return sign_mode;
+}
+
+void
+tgsi_util_set_full_src_register_sign_mode(
+   struct tgsi_full_src_register *reg,
+   unsigned sign_mode )
+{
+   reg->SrcRegisterExtSwz.NegateX = 0;
+   reg->SrcRegisterExtSwz.NegateY = 0;
+   reg->SrcRegisterExtSwz.NegateZ = 0;
+   reg->SrcRegisterExtSwz.NegateW = 0;
+
+   switch (sign_mode)
+   {
+   case TGSI_UTIL_SIGN_CLEAR:
+      reg->SrcRegister.Negate = 0;
+      reg->SrcRegisterExtMod.Absolute = 1;
+      reg->SrcRegisterExtMod.Negate = 0;
+      break;
+
+   case TGSI_UTIL_SIGN_SET:
+      reg->SrcRegister.Negate = 0;
+      reg->SrcRegisterExtMod.Absolute = 1;
+      reg->SrcRegisterExtMod.Negate = 1;
+      break;
+
+   case TGSI_UTIL_SIGN_TOGGLE:
+      reg->SrcRegister.Negate = 1;
+      reg->SrcRegisterExtMod.Absolute = 0;
+      reg->SrcRegisterExtMod.Negate = 0;
+      break;
+
+   case TGSI_UTIL_SIGN_KEEP:
+      reg->SrcRegister.Negate = 0;
+      reg->SrcRegisterExtMod.Absolute = 0;
+      reg->SrcRegisterExtMod.Negate = 0;
+      break;
+
+   default:
+      assert( 0 );
+   }
+}
diff --git a/src/gallium/auxiliary/tgsi/tgsi_util.h b/src/gallium/auxiliary/tgsi/tgsi_util.h
new file mode 100644
index 00000000000..7877f345587
--- /dev/null
+++ b/src/gallium/auxiliary/tgsi/tgsi_util.h
@@ -0,0 +1,96 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef TGSI_UTIL_H
+#define TGSI_UTIL_H
+
+#if defined __cplusplus
+extern "C" {
+#endif
+
+void *
+tgsi_align_128bit(
+   void *unaligned );
+
+unsigned
+tgsi_util_get_src_register_swizzle(
+   const struct tgsi_src_register *reg,
+   unsigned component );
+
+unsigned
+tgsi_util_get_src_register_extswizzle(
+   const struct tgsi_src_register_ext_swz *reg,
+   unsigned component);
+
+unsigned
+tgsi_util_get_full_src_register_extswizzle(
+   const struct tgsi_full_src_register *reg,
+   unsigned component );
+
+void
+tgsi_util_set_src_register_swizzle(
+   struct tgsi_src_register *reg,
+   unsigned swizzle,
+   unsigned component );
+
+void
+tgsi_util_set_src_register_extswizzle(
+   struct tgsi_src_register_ext_swz *reg,
+   unsigned swizzle,
+   unsigned component );
+
+unsigned
+tgsi_util_get_src_register_extnegate(
+   const struct tgsi_src_register_ext_swz *reg,
+   unsigned component );
+
+void
+tgsi_util_set_src_register_extnegate(
+   struct tgsi_src_register_ext_swz *reg,
+   unsigned negate,
+   unsigned component );
+
+#define TGSI_UTIL_SIGN_CLEAR    0   /* Force positive */
+#define TGSI_UTIL_SIGN_SET      1   /* Force negative */
+#define TGSI_UTIL_SIGN_TOGGLE   2   /* Negate */
+#define TGSI_UTIL_SIGN_KEEP     3   /* No change */
+
+unsigned
+tgsi_util_get_full_src_register_sign_mode(
+   const struct tgsi_full_src_register *reg,
+   unsigned component );
+
+void
+tgsi_util_set_full_src_register_sign_mode(
+   struct tgsi_full_src_register *reg,
+   unsigned sign_mode );
+
+#if defined __cplusplus
+}
+#endif
+
+#endif /* TGSI_UTIL_H */
diff --git a/src/gallium/auxiliary/translate/Makefile b/src/gallium/auxiliary/translate/Makefile
new file mode 100644
index 00000000000..ad2a5b705e4
--- /dev/null
+++ b/src/gallium/auxiliary/translate/Makefile
@@ -0,0 +1,15 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = translate
+
+C_SOURCES = \
+	translate_generic.c \
+	translate_sse.c \
+	translate.c \
+        translate_cache.c
+
+include ../../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/auxiliary/translate/SConscript b/src/gallium/auxiliary/translate/SConscript
new file mode 100644
index 00000000000..9553a675372
--- /dev/null
+++ b/src/gallium/auxiliary/translate/SConscript
@@ -0,0 +1,12 @@
+Import('*')
+
+translate = env.ConvenienceLibrary(
+	target = 'translate',
+	source = [
+		'translate_generic.c',
+		'translate_sse.c',
+		'translate.c',
+		'translate_cache.c',
+	])
+
+auxiliaries.insert(0, translate)
diff --git a/src/gallium/auxiliary/translate/translate.c b/src/gallium/auxiliary/translate/translate.c
new file mode 100644
index 00000000000..7678903f75c
--- /dev/null
+++ b/src/gallium/auxiliary/translate/translate.c
@@ -0,0 +1,48 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  */
+
+#include "pipe/p_config.h"
+#include "pipe/p_state.h"
+#include "translate.h"
+
+struct translate *translate_create( const struct translate_key *key )
+{
+   struct translate *translate = NULL;
+
+#if defined(PIPE_ARCH_X86)
+   translate = translate_sse2_create( key );
+   if (translate)
+      return translate;
+#endif
+
+   return translate_generic_create( key );
+}
diff --git a/src/gallium/auxiliary/translate/translate.h b/src/gallium/auxiliary/translate/translate.h
new file mode 100644
index 00000000000..34526eb0617
--- /dev/null
+++ b/src/gallium/auxiliary/translate/translate.h
@@ -0,0 +1,127 @@
+/*
+ * Copyright 2008 Tungsten Graphics, inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+/**
+ * Vertex fetch/store/convert code.  This functionality is used in two places:
+ * 1. Vertex fetch/convert - to grab vertex data from incoming vertex
+ *    arrays and convert to format needed by vertex shaders.
+ * 2. Vertex store/emit - to convert simple float[][4] vertex attributes
+ *    (which is the organization used throughout the draw/prim pipeline) to
+ *    hardware-specific formats and emit into hardware vertex buffers.
+ *
+ *
+ * Authors:
+ *    Keith Whitwell <[email protected]>
+ */
+
+#ifndef _TRANSLATE_H
+#define _TRANSLATE_H
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_format.h"
+#include "pipe/p_state.h"
+
+struct translate_element 
+{
+   enum pipe_format input_format;
+   enum pipe_format output_format;
+   unsigned input_buffer:8;
+   unsigned input_offset:24;
+   unsigned output_offset;
+};
+
+
+struct translate_key {
+   unsigned output_stride;
+   unsigned nr_elements;
+   struct translate_element element[PIPE_MAX_ATTRIBS + 1];
+};
+
+
+struct translate {
+   struct translate_key key;
+
+   void (*release)( struct translate * );
+
+   void (*set_buffer)( struct translate *,
+		       unsigned i,
+		       const void *ptr,
+		       unsigned stride );
+
+   void (PIPE_CDECL *run_elts)( struct translate *,
+                                const unsigned *elts,
+                                unsigned count,
+                                void *output_buffer);
+
+   void (PIPE_CDECL *run)( struct translate *,
+                           unsigned start,
+                           unsigned count,
+                           void *output_buffer);
+};
+
+
+
+#if 0
+struct translate_context *translate_context_create( void );
+void translate_context_destroy( struct translate_context * );
+
+struct translate *translate_lookup_or_create( struct translate_context *tctx,
+					      const struct translate_key *key );
+#endif
+
+
+struct translate *translate_create( const struct translate_key *key );
+
+static INLINE int translate_keysize( const struct translate_key *key )
+{
+   return 2 * sizeof(int) + key->nr_elements * sizeof(struct translate_element);
+}
+
+static INLINE int translate_key_compare( const struct translate_key *a,
+                                         const struct translate_key *b )
+{
+   int keysize = translate_keysize(a);
+   return memcmp(a, b, keysize);
+}
+
+
+static INLINE void translate_key_sanitize( struct translate_key *a )
+{
+   int keysize = translate_keysize(a);
+   char *ptr = (char *)a;
+   memset(ptr + keysize, 0, sizeof(*a) - keysize);
+}
+
+
+/*******************************************************************************
+ *  Private:
+ */
+struct translate *translate_sse2_create( const struct translate_key *key );
+
+struct translate *translate_generic_create( const struct translate_key *key );
+
+
+#endif
diff --git a/src/gallium/auxiliary/translate/translate_cache.c b/src/gallium/auxiliary/translate/translate_cache.c
new file mode 100644
index 00000000000..d8069a149cf
--- /dev/null
+++ b/src/gallium/auxiliary/translate/translate_cache.c
@@ -0,0 +1,102 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include "util/u_memory.h"
+#include "pipe/p_state.h"
+#include "translate.h"
+#include "translate_cache.h"
+
+#include "cso_cache/cso_cache.h"
+#include "cso_cache/cso_hash.h"
+
+struct translate_cache {
+   struct cso_hash *hash;
+};
+
+struct translate_cache * translate_cache_create( void )
+{
+   struct translate_cache *cache = MALLOC_STRUCT(translate_cache);
+   cache->hash = cso_hash_create();
+   return cache;
+}
+
+
+static INLINE void delete_translates(struct translate_cache *cache)
+{
+   struct cso_hash *hash = cache->hash;
+   struct cso_hash_iter iter = cso_hash_first_node(hash);
+   while (!cso_hash_iter_is_null(iter)) {
+      struct translate *state = (struct translate*)cso_hash_iter_data(iter);
+      iter = cso_hash_iter_next(iter);
+      if (state) {
+         state->release(state);
+      }
+   }
+}
+
+void translate_cache_destroy(struct translate_cache *cache)
+{
+   delete_translates(cache);
+   cso_hash_delete(cache->hash);
+   FREE(cache);
+}
+
+
+static INLINE unsigned translate_hash_key_size(struct translate_key *key)
+{
+   unsigned size = sizeof(struct translate_key) -
+                   sizeof(struct translate_element) * (PIPE_MAX_ATTRIBS - key->nr_elements);
+   return size;
+}
+
+static INLINE unsigned create_key(struct translate_key *key)
+{
+   unsigned hash_key;
+   unsigned size = translate_hash_key_size(key);
+   /*debug_printf("key size = %d, (els = %d)\n",
+     size, key->nr_elements);*/
+   hash_key = cso_construct_key(key, size);
+   return hash_key;
+}
+
+struct translate * translate_cache_find(struct translate_cache *cache,
+                                        struct translate_key *key)
+{
+   unsigned hash_key = create_key(key);
+   struct translate *translate = (struct translate*)
+      cso_hash_find_data_from_template(cache->hash,
+                                       hash_key,
+                                       key, sizeof(*key));
+
+   if (!translate) {
+      /* create/insert */
+      translate = translate_create(key);
+      cso_hash_insert(cache->hash, hash_key, translate);
+   }
+
+   return translate;
+}
diff --git a/src/gallium/auxiliary/translate/translate_cache.h b/src/gallium/auxiliary/translate/translate_cache.h
new file mode 100644
index 00000000000..7dba871e579
--- /dev/null
+++ b/src/gallium/auxiliary/translate/translate_cache.h
@@ -0,0 +1,54 @@
+/*
+ * Copyright 2008 Tungsten Graphics, inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+#ifndef _TRANSLATE_CACHE_H
+#define _TRANSLATE_CACHE_H
+
+
+/*******************************************************************************
+ * Translate cache.
+ * Simply used to cache created translates. Avoids unecessary creation of
+ * translate's if one suitable for a given translate_key has already been
+ * created.
+ *
+ * Note: this functionality depends and requires the CSO module.
+ */
+struct translate_cache;
+
+struct translate_key;
+struct translate;
+
+struct translate_cache *translate_cache_create( void );
+void translate_cache_destroy(struct translate_cache *cache);
+
+/**
+ * Will try to find a translate structure matched by the given key.
+ * If such a structure doesn't exist in the cache the function
+ * will automatically create it, insert it in the cache and
+ * return the created version.
+ *
+ */
+struct translate *translate_cache_find(struct translate_cache *cache,
+                                       struct translate_key *key);
+
+#endif
diff --git a/src/gallium/auxiliary/translate/translate_generic.c b/src/gallium/auxiliary/translate/translate_generic.c
new file mode 100644
index 00000000000..8d39b64c6c1
--- /dev/null
+++ b/src/gallium/auxiliary/translate/translate_generic.c
@@ -0,0 +1,700 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <[email protected]>
+  */
+
+#include "util/u_memory.h"
+#include "pipe/p_state.h"
+#include "translate.h"
+
+
+#define DRAW_DBG 0
+
+typedef void (*fetch_func)(const void *ptr, float *attrib);
+typedef void (*emit_func)(const float *attrib, void *ptr);
+
+
+
+struct translate_generic {
+   struct translate translate;
+
+   struct {
+      fetch_func fetch;
+      unsigned buffer;
+      unsigned input_offset;
+
+      emit_func emit;
+      unsigned output_offset;
+      
+      char *input_ptr;
+      unsigned input_stride;
+
+   } attrib[PIPE_MAX_ATTRIBS];
+
+   unsigned nr_attrib;
+};
+
+
+static struct translate_generic *translate_generic( struct translate *translate )
+{
+   return (struct translate_generic *)translate;
+}
+
+/**
+ * Fetch a float[4] vertex attribute from memory, doing format/type
+ * conversion as needed.
+ *
+ * This is probably needed/dupliocated elsewhere, eg format
+ * conversion, texture sampling etc.
+ */
+#define ATTRIB( NAME, SZ, TYPE, FROM, TO )		\
+static void						\
+fetch_##NAME(const void *ptr, float *attrib)		\
+{							\
+   const float defaults[4] = { 0.0f,0.0f,0.0f,1.0f };	\
+   unsigned i;						\
+							\
+   for (i = 0; i < SZ; i++) {				\
+      attrib[i] = FROM(i);				\
+   }							\
+							\
+   for (; i < 4; i++) {					\
+      attrib[i] = defaults[i];				\
+   }							\
+}							\
+							\
+static void						\
+emit_##NAME(const float *attrib, void *ptr)		\
+{  \
+   unsigned i;						\
+   TYPE *out = (TYPE *)ptr;				\
+							\
+   for (i = 0; i < SZ; i++) {				\
+      out[i] = TO(attrib[i]);				\
+   }							\
+}
+
+
+#define FROM_64_FLOAT(i)   ((float) ((double *) ptr)[i])
+#define FROM_32_FLOAT(i)   (((float *) ptr)[i])
+
+#define FROM_8_USCALED(i)  ((float) ((unsigned char *) ptr)[i])
+#define FROM_16_USCALED(i) ((float) ((unsigned short *) ptr)[i])
+#define FROM_32_USCALED(i) ((float) ((unsigned int *) ptr)[i])
+
+#define FROM_8_SSCALED(i)  ((float) ((char *) ptr)[i])
+#define FROM_16_SSCALED(i) ((float) ((short *) ptr)[i])
+#define FROM_32_SSCALED(i) ((float) ((int *) ptr)[i])
+
+#define FROM_8_UNORM(i)    ((float) ((unsigned char *) ptr)[i] / 255.0f)
+#define FROM_16_UNORM(i)   ((float) ((unsigned short *) ptr)[i] / 65535.0f)
+#define FROM_32_UNORM(i)   ((float) ((unsigned int *) ptr)[i] / 4294967295.0f)
+
+#define FROM_8_SNORM(i)    ((float) ((char *) ptr)[i] / 127.0f)
+#define FROM_16_SNORM(i)   ((float) ((short *) ptr)[i] / 32767.0f)
+#define FROM_32_SNORM(i)   ((float) ((int *) ptr)[i] / 2147483647.0f)
+
+#define FROM_32_FIXED(i)   (((int *) ptr)[i] / 65536.0f)
+
+#define TO_64_FLOAT(x)   ((double) x)
+#define TO_32_FLOAT(x)   (x)
+
+#define TO_8_USCALED(x)  ((unsigned char) x)
+#define TO_16_USCALED(x) ((unsigned short) x)
+#define TO_32_USCALED(x) ((unsigned int) x)
+
+#define TO_8_SSCALED(x)  ((char) x)
+#define TO_16_SSCALED(x) ((short) x)
+#define TO_32_SSCALED(x) ((int) x)
+
+#define TO_8_UNORM(x)    ((unsigned char) (x * 255.0f))
+#define TO_16_UNORM(x)   ((unsigned short) (x * 65535.0f))
+#define TO_32_UNORM(x)   ((unsigned int) (x * 4294967295.0f))
+
+#define TO_8_SNORM(x)    ((char) (x * 127.0f))
+#define TO_16_SNORM(x)   ((short) (x * 32767.0f))
+#define TO_32_SNORM(x)   ((int) (x * 2147483647.0f))
+
+#define TO_32_FIXED(x)   ((int) (x * 65536.0f))
+
+
+
+ATTRIB( R64G64B64A64_FLOAT,   4, double, FROM_64_FLOAT, TO_64_FLOAT )
+ATTRIB( R64G64B64_FLOAT,      3, double, FROM_64_FLOAT, TO_64_FLOAT )
+ATTRIB( R64G64_FLOAT,         2, double, FROM_64_FLOAT, TO_64_FLOAT )
+ATTRIB( R64_FLOAT,            1, double, FROM_64_FLOAT, TO_64_FLOAT )
+
+ATTRIB( R32G32B32A32_FLOAT,   4, float, FROM_32_FLOAT, TO_32_FLOAT )
+ATTRIB( R32G32B32_FLOAT,      3, float, FROM_32_FLOAT, TO_32_FLOAT )
+ATTRIB( R32G32_FLOAT,         2, float, FROM_32_FLOAT, TO_32_FLOAT )
+ATTRIB( R32_FLOAT,            1, float, FROM_32_FLOAT, TO_32_FLOAT )
+
+ATTRIB( R32G32B32A32_USCALED, 4, unsigned, FROM_32_USCALED, TO_32_USCALED )
+ATTRIB( R32G32B32_USCALED,    3, unsigned, FROM_32_USCALED, TO_32_USCALED )
+ATTRIB( R32G32_USCALED,       2, unsigned, FROM_32_USCALED, TO_32_USCALED )
+ATTRIB( R32_USCALED,          1, unsigned, FROM_32_USCALED, TO_32_USCALED )
+
+ATTRIB( R32G32B32A32_SSCALED, 4, int, FROM_32_SSCALED, TO_32_SSCALED )
+ATTRIB( R32G32B32_SSCALED,    3, int, FROM_32_SSCALED, TO_32_SSCALED )
+ATTRIB( R32G32_SSCALED,       2, int, FROM_32_SSCALED, TO_32_SSCALED )
+ATTRIB( R32_SSCALED,          1, int, FROM_32_SSCALED, TO_32_SSCALED )
+
+ATTRIB( R32G32B32A32_UNORM, 4, unsigned, FROM_32_UNORM, TO_32_UNORM )
+ATTRIB( R32G32B32_UNORM,    3, unsigned, FROM_32_UNORM, TO_32_UNORM )
+ATTRIB( R32G32_UNORM,       2, unsigned, FROM_32_UNORM, TO_32_UNORM )
+ATTRIB( R32_UNORM,          1, unsigned, FROM_32_UNORM, TO_32_UNORM )
+
+ATTRIB( R32G32B32A32_SNORM, 4, int, FROM_32_SNORM, TO_32_SNORM )
+ATTRIB( R32G32B32_SNORM,    3, int, FROM_32_SNORM, TO_32_SNORM )
+ATTRIB( R32G32_SNORM,       2, int, FROM_32_SNORM, TO_32_SNORM )
+ATTRIB( R32_SNORM,          1, int, FROM_32_SNORM, TO_32_SNORM )
+
+ATTRIB( R16G16B16A16_USCALED, 4, ushort, FROM_16_USCALED, TO_16_USCALED )
+ATTRIB( R16G16B16_USCALED,    3, ushort, FROM_16_USCALED, TO_16_USCALED )
+ATTRIB( R16G16_USCALED,       2, ushort, FROM_16_USCALED, TO_16_USCALED )
+ATTRIB( R16_USCALED,          1, ushort, FROM_16_USCALED, TO_16_USCALED )
+
+ATTRIB( R16G16B16A16_SSCALED, 4, short, FROM_16_SSCALED, TO_16_SSCALED )
+ATTRIB( R16G16B16_SSCALED,    3, short, FROM_16_SSCALED, TO_16_SSCALED )
+ATTRIB( R16G16_SSCALED,       2, short, FROM_16_SSCALED, TO_16_SSCALED )
+ATTRIB( R16_SSCALED,          1, short, FROM_16_SSCALED, TO_16_SSCALED )
+
+ATTRIB( R16G16B16A16_UNORM, 4, ushort, FROM_16_UNORM, TO_16_UNORM )
+ATTRIB( R16G16B16_UNORM,    3, ushort, FROM_16_UNORM, TO_16_UNORM )
+ATTRIB( R16G16_UNORM,       2, ushort, FROM_16_UNORM, TO_16_UNORM )
+ATTRIB( R16_UNORM,          1, ushort, FROM_16_UNORM, TO_16_UNORM )
+
+ATTRIB( R16G16B16A16_SNORM, 4, short, FROM_16_SNORM, TO_16_SNORM )
+ATTRIB( R16G16B16_SNORM,    3, short, FROM_16_SNORM, TO_16_SNORM )
+ATTRIB( R16G16_SNORM,       2, short, FROM_16_SNORM, TO_16_SNORM )
+ATTRIB( R16_SNORM,          1, short, FROM_16_SNORM, TO_16_SNORM )
+
+ATTRIB( R8G8B8A8_USCALED,   4, ubyte, FROM_8_USCALED, TO_8_USCALED )
+ATTRIB( R8G8B8_USCALED,     3, ubyte, FROM_8_USCALED, TO_8_USCALED )
+ATTRIB( R8G8_USCALED,       2, ubyte, FROM_8_USCALED, TO_8_USCALED )
+ATTRIB( R8_USCALED,         1, ubyte, FROM_8_USCALED, TO_8_USCALED )
+
+ATTRIB( R8G8B8A8_SSCALED,  4, char, FROM_8_SSCALED, TO_8_SSCALED )
+ATTRIB( R8G8B8_SSCALED,    3, char, FROM_8_SSCALED, TO_8_SSCALED )
+ATTRIB( R8G8_SSCALED,      2, char, FROM_8_SSCALED, TO_8_SSCALED )
+ATTRIB( R8_SSCALED,        1, char, FROM_8_SSCALED, TO_8_SSCALED )
+
+ATTRIB( R8G8B8A8_UNORM,  4, ubyte, FROM_8_UNORM, TO_8_UNORM )
+ATTRIB( R8G8B8_UNORM,    3, ubyte, FROM_8_UNORM, TO_8_UNORM )
+ATTRIB( R8G8_UNORM,      2, ubyte, FROM_8_UNORM, TO_8_UNORM )
+ATTRIB( R8_UNORM,        1, ubyte, FROM_8_UNORM, TO_8_UNORM )
+
+ATTRIB( R8G8B8A8_SNORM,  4, char, FROM_8_SNORM, TO_8_SNORM )
+ATTRIB( R8G8B8_SNORM,    3, char, FROM_8_SNORM, TO_8_SNORM )
+ATTRIB( R8G8_SNORM,      2, char, FROM_8_SNORM, TO_8_SNORM )
+ATTRIB( R8_SNORM,        1, char, FROM_8_SNORM, TO_8_SNORM )
+
+ATTRIB( A8R8G8B8_UNORM,       4, ubyte, FROM_8_UNORM, TO_8_UNORM )
+//ATTRIB( R8G8B8A8_UNORM,       4, ubyte, FROM_8_UNORM, TO_8_UNORM )
+
+ATTRIB( R32G32B32A32_FIXED,   4, int, FROM_32_FIXED, TO_32_FIXED )
+ATTRIB( R32G32B32_FIXED,      3, int, FROM_32_FIXED, TO_32_FIXED )
+ATTRIB( R32G32_FIXED,         2, int, FROM_32_FIXED, TO_32_FIXED )
+ATTRIB( R32_FIXED,            1, int, FROM_32_FIXED, TO_32_FIXED )
+
+
+
+static void
+fetch_B8G8R8A8_UNORM(const void *ptr, float *attrib)
+{
+   attrib[2] = FROM_8_UNORM(0);
+   attrib[1] = FROM_8_UNORM(1);
+   attrib[0] = FROM_8_UNORM(2);
+   attrib[3] = FROM_8_UNORM(3);
+}
+
+static void
+emit_B8G8R8A8_UNORM( const float *attrib, void *ptr)
+{
+   ubyte *out = (ubyte *)ptr;
+   out[2] = TO_8_UNORM(attrib[0]);
+   out[1] = TO_8_UNORM(attrib[1]);
+   out[0] = TO_8_UNORM(attrib[2]);
+   out[3] = TO_8_UNORM(attrib[3]);
+}
+
+static void 
+fetch_NULL( const void *ptr, float *attrib )
+{
+   attrib[0] = 0;
+   attrib[1] = 0;
+   attrib[2] = 0;
+   attrib[3] = 1;
+}
+
+static void 
+emit_NULL( const float *attrib, void *ptr )
+{
+   /* do nothing is the only sensible option */
+}
+
+static fetch_func get_fetch_func( enum pipe_format format )
+{
+   switch (format) {
+   case PIPE_FORMAT_R64_FLOAT:
+      return &fetch_R64_FLOAT;
+   case PIPE_FORMAT_R64G64_FLOAT:
+      return &fetch_R64G64_FLOAT;
+   case PIPE_FORMAT_R64G64B64_FLOAT:
+      return &fetch_R64G64B64_FLOAT;
+   case PIPE_FORMAT_R64G64B64A64_FLOAT:
+      return &fetch_R64G64B64A64_FLOAT;
+
+   case PIPE_FORMAT_R32_FLOAT:
+      return &fetch_R32_FLOAT;
+   case PIPE_FORMAT_R32G32_FLOAT:
+      return &fetch_R32G32_FLOAT;
+   case PIPE_FORMAT_R32G32B32_FLOAT:
+      return &fetch_R32G32B32_FLOAT;
+   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+      return &fetch_R32G32B32A32_FLOAT;
+
+   case PIPE_FORMAT_R32_UNORM:
+      return &fetch_R32_UNORM;
+   case PIPE_FORMAT_R32G32_UNORM:
+      return &fetch_R32G32_UNORM;
+   case PIPE_FORMAT_R32G32B32_UNORM:
+      return &fetch_R32G32B32_UNORM;
+   case PIPE_FORMAT_R32G32B32A32_UNORM:
+      return &fetch_R32G32B32A32_UNORM;
+
+   case PIPE_FORMAT_R32_USCALED:
+      return &fetch_R32_USCALED;
+   case PIPE_FORMAT_R32G32_USCALED:
+      return &fetch_R32G32_USCALED;
+   case PIPE_FORMAT_R32G32B32_USCALED:
+      return &fetch_R32G32B32_USCALED;
+   case PIPE_FORMAT_R32G32B32A32_USCALED:
+      return &fetch_R32G32B32A32_USCALED;
+
+   case PIPE_FORMAT_R32_SNORM:
+      return &fetch_R32_SNORM;
+   case PIPE_FORMAT_R32G32_SNORM:
+      return &fetch_R32G32_SNORM;
+   case PIPE_FORMAT_R32G32B32_SNORM:
+      return &fetch_R32G32B32_SNORM;
+   case PIPE_FORMAT_R32G32B32A32_SNORM:
+      return &fetch_R32G32B32A32_SNORM;
+
+   case PIPE_FORMAT_R32_SSCALED:
+      return &fetch_R32_SSCALED;
+   case PIPE_FORMAT_R32G32_SSCALED:
+      return &fetch_R32G32_SSCALED;
+   case PIPE_FORMAT_R32G32B32_SSCALED:
+      return &fetch_R32G32B32_SSCALED;
+   case PIPE_FORMAT_R32G32B32A32_SSCALED:
+      return &fetch_R32G32B32A32_SSCALED;
+
+   case PIPE_FORMAT_R16_UNORM:
+      return &fetch_R16_UNORM;
+   case PIPE_FORMAT_R16G16_UNORM:
+      return &fetch_R16G16_UNORM;
+   case PIPE_FORMAT_R16G16B16_UNORM:
+      return &fetch_R16G16B16_UNORM;
+   case PIPE_FORMAT_R16G16B16A16_UNORM:
+      return &fetch_R16G16B16A16_UNORM;
+
+   case PIPE_FORMAT_R16_USCALED:
+      return &fetch_R16_USCALED;
+   case PIPE_FORMAT_R16G16_USCALED:
+      return &fetch_R16G16_USCALED;
+   case PIPE_FORMAT_R16G16B16_USCALED:
+      return &fetch_R16G16B16_USCALED;
+   case PIPE_FORMAT_R16G16B16A16_USCALED:
+      return &fetch_R16G16B16A16_USCALED;
+
+   case PIPE_FORMAT_R16_SNORM:
+      return &fetch_R16_SNORM;
+   case PIPE_FORMAT_R16G16_SNORM:
+      return &fetch_R16G16_SNORM;
+   case PIPE_FORMAT_R16G16B16_SNORM:
+      return &fetch_R16G16B16_SNORM;
+   case PIPE_FORMAT_R16G16B16A16_SNORM:
+      return &fetch_R16G16B16A16_SNORM;
+
+   case PIPE_FORMAT_R16_SSCALED:
+      return &fetch_R16_SSCALED;
+   case PIPE_FORMAT_R16G16_SSCALED:
+      return &fetch_R16G16_SSCALED;
+   case PIPE_FORMAT_R16G16B16_SSCALED:
+      return &fetch_R16G16B16_SSCALED;
+   case PIPE_FORMAT_R16G16B16A16_SSCALED:
+      return &fetch_R16G16B16A16_SSCALED;
+
+   case PIPE_FORMAT_R8_UNORM:
+      return &fetch_R8_UNORM;
+   case PIPE_FORMAT_R8G8_UNORM:
+      return &fetch_R8G8_UNORM;
+   case PIPE_FORMAT_R8G8B8_UNORM:
+      return &fetch_R8G8B8_UNORM;
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      return &fetch_R8G8B8A8_UNORM;
+
+   case PIPE_FORMAT_R8_USCALED:
+      return &fetch_R8_USCALED;
+   case PIPE_FORMAT_R8G8_USCALED:
+      return &fetch_R8G8_USCALED;
+   case PIPE_FORMAT_R8G8B8_USCALED:
+      return &fetch_R8G8B8_USCALED;
+   case PIPE_FORMAT_R8G8B8A8_USCALED:
+      return &fetch_R8G8B8A8_USCALED;
+
+   case PIPE_FORMAT_R8_SNORM:
+      return &fetch_R8_SNORM;
+   case PIPE_FORMAT_R8G8_SNORM:
+      return &fetch_R8G8_SNORM;
+   case PIPE_FORMAT_R8G8B8_SNORM:
+      return &fetch_R8G8B8_SNORM;
+   case PIPE_FORMAT_R8G8B8A8_SNORM:
+      return &fetch_R8G8B8A8_SNORM;
+
+   case PIPE_FORMAT_R8_SSCALED:
+      return &fetch_R8_SSCALED;
+   case PIPE_FORMAT_R8G8_SSCALED:
+      return &fetch_R8G8_SSCALED;
+   case PIPE_FORMAT_R8G8B8_SSCALED:
+      return &fetch_R8G8B8_SSCALED;
+   case PIPE_FORMAT_R8G8B8A8_SSCALED:
+      return &fetch_R8G8B8A8_SSCALED;
+
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+      return &fetch_A8R8G8B8_UNORM;
+
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      return &fetch_B8G8R8A8_UNORM;
+
+   case PIPE_FORMAT_R32_FIXED:
+      return &fetch_R32_FIXED;
+   case PIPE_FORMAT_R32G32_FIXED:
+      return &fetch_R32G32_FIXED;
+   case PIPE_FORMAT_R32G32B32_FIXED:
+      return &fetch_R32G32B32_FIXED;
+   case PIPE_FORMAT_R32G32B32A32_FIXED:
+      return &fetch_R32G32B32A32_FIXED;
+
+   default:
+      assert(0); 
+      return &fetch_NULL;
+   }
+}
+
+
+
+
+static emit_func get_emit_func( enum pipe_format format )
+{
+   /* silence warnings */
+   (void) emit_R32G32B32A32_FIXED;
+   (void) emit_R32G32B32_FIXED;
+   (void) emit_R32G32_FIXED;
+   (void) emit_R32_FIXED;
+
+   switch (format) {
+   case PIPE_FORMAT_R64_FLOAT:
+      return &emit_R64_FLOAT;
+   case PIPE_FORMAT_R64G64_FLOAT:
+      return &emit_R64G64_FLOAT;
+   case PIPE_FORMAT_R64G64B64_FLOAT:
+      return &emit_R64G64B64_FLOAT;
+   case PIPE_FORMAT_R64G64B64A64_FLOAT:
+      return &emit_R64G64B64A64_FLOAT;
+
+   case PIPE_FORMAT_R32_FLOAT:
+      return &emit_R32_FLOAT;
+   case PIPE_FORMAT_R32G32_FLOAT:
+      return &emit_R32G32_FLOAT;
+   case PIPE_FORMAT_R32G32B32_FLOAT:
+      return &emit_R32G32B32_FLOAT;
+   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+      return &emit_R32G32B32A32_FLOAT;
+
+   case PIPE_FORMAT_R32_UNORM:
+      return &emit_R32_UNORM;
+   case PIPE_FORMAT_R32G32_UNORM:
+      return &emit_R32G32_UNORM;
+   case PIPE_FORMAT_R32G32B32_UNORM:
+      return &emit_R32G32B32_UNORM;
+   case PIPE_FORMAT_R32G32B32A32_UNORM:
+      return &emit_R32G32B32A32_UNORM;
+
+   case PIPE_FORMAT_R32_USCALED:
+      return &emit_R32_USCALED;
+   case PIPE_FORMAT_R32G32_USCALED:
+      return &emit_R32G32_USCALED;
+   case PIPE_FORMAT_R32G32B32_USCALED:
+      return &emit_R32G32B32_USCALED;
+   case PIPE_FORMAT_R32G32B32A32_USCALED:
+      return &emit_R32G32B32A32_USCALED;
+
+   case PIPE_FORMAT_R32_SNORM:
+      return &emit_R32_SNORM;
+   case PIPE_FORMAT_R32G32_SNORM:
+      return &emit_R32G32_SNORM;
+   case PIPE_FORMAT_R32G32B32_SNORM:
+      return &emit_R32G32B32_SNORM;
+   case PIPE_FORMAT_R32G32B32A32_SNORM:
+      return &emit_R32G32B32A32_SNORM;
+
+   case PIPE_FORMAT_R32_SSCALED:
+      return &emit_R32_SSCALED;
+   case PIPE_FORMAT_R32G32_SSCALED:
+      return &emit_R32G32_SSCALED;
+   case PIPE_FORMAT_R32G32B32_SSCALED:
+      return &emit_R32G32B32_SSCALED;
+   case PIPE_FORMAT_R32G32B32A32_SSCALED:
+      return &emit_R32G32B32A32_SSCALED;
+
+   case PIPE_FORMAT_R16_UNORM:
+      return &emit_R16_UNORM;
+   case PIPE_FORMAT_R16G16_UNORM:
+      return &emit_R16G16_UNORM;
+   case PIPE_FORMAT_R16G16B16_UNORM:
+      return &emit_R16G16B16_UNORM;
+   case PIPE_FORMAT_R16G16B16A16_UNORM:
+      return &emit_R16G16B16A16_UNORM;
+
+   case PIPE_FORMAT_R16_USCALED:
+      return &emit_R16_USCALED;
+   case PIPE_FORMAT_R16G16_USCALED:
+      return &emit_R16G16_USCALED;
+   case PIPE_FORMAT_R16G16B16_USCALED:
+      return &emit_R16G16B16_USCALED;
+   case PIPE_FORMAT_R16G16B16A16_USCALED:
+      return &emit_R16G16B16A16_USCALED;
+
+   case PIPE_FORMAT_R16_SNORM:
+      return &emit_R16_SNORM;
+   case PIPE_FORMAT_R16G16_SNORM:
+      return &emit_R16G16_SNORM;
+   case PIPE_FORMAT_R16G16B16_SNORM:
+      return &emit_R16G16B16_SNORM;
+   case PIPE_FORMAT_R16G16B16A16_SNORM:
+      return &emit_R16G16B16A16_SNORM;
+
+   case PIPE_FORMAT_R16_SSCALED:
+      return &emit_R16_SSCALED;
+   case PIPE_FORMAT_R16G16_SSCALED:
+      return &emit_R16G16_SSCALED;
+   case PIPE_FORMAT_R16G16B16_SSCALED:
+      return &emit_R16G16B16_SSCALED;
+   case PIPE_FORMAT_R16G16B16A16_SSCALED:
+      return &emit_R16G16B16A16_SSCALED;
+
+   case PIPE_FORMAT_R8_UNORM:
+      return &emit_R8_UNORM;
+   case PIPE_FORMAT_R8G8_UNORM:
+      return &emit_R8G8_UNORM;
+   case PIPE_FORMAT_R8G8B8_UNORM:
+      return &emit_R8G8B8_UNORM;
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      return &emit_R8G8B8A8_UNORM;
+
+   case PIPE_FORMAT_R8_USCALED:
+      return &emit_R8_USCALED;
+   case PIPE_FORMAT_R8G8_USCALED:
+      return &emit_R8G8_USCALED;
+   case PIPE_FORMAT_R8G8B8_USCALED:
+      return &emit_R8G8B8_USCALED;
+   case PIPE_FORMAT_R8G8B8A8_USCALED:
+      return &emit_R8G8B8A8_USCALED;
+
+   case PIPE_FORMAT_R8_SNORM:
+      return &emit_R8_SNORM;
+   case PIPE_FORMAT_R8G8_SNORM:
+      return &emit_R8G8_SNORM;
+   case PIPE_FORMAT_R8G8B8_SNORM:
+      return &emit_R8G8B8_SNORM;
+   case PIPE_FORMAT_R8G8B8A8_SNORM:
+      return &emit_R8G8B8A8_SNORM;
+
+   case PIPE_FORMAT_R8_SSCALED:
+      return &emit_R8_SSCALED;
+   case PIPE_FORMAT_R8G8_SSCALED:
+      return &emit_R8G8_SSCALED;
+   case PIPE_FORMAT_R8G8B8_SSCALED:
+      return &emit_R8G8B8_SSCALED;
+   case PIPE_FORMAT_R8G8B8A8_SSCALED:
+      return &emit_R8G8B8A8_SSCALED;
+
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+      return &emit_A8R8G8B8_UNORM;
+
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      return &emit_B8G8R8A8_UNORM;
+
+   default:
+      assert(0); 
+      return &emit_NULL;
+   }
+}
+
+
+
+/**
+ * Fetch vertex attributes for 'count' vertices.
+ */
+static void PIPE_CDECL generic_run_elts( struct translate *translate,
+                                         const unsigned *elts,
+                                         unsigned count,
+                                         void *output_buffer )
+{
+   struct translate_generic *tg = translate_generic(translate);
+   char *vert = output_buffer;
+   unsigned nr_attrs = tg->nr_attrib;
+   unsigned attr;
+   unsigned i;
+
+   /* loop over vertex attributes (vertex shader inputs)
+    */
+   for (i = 0; i < count; i++) {
+      unsigned elt = *elts++;
+
+      for (attr = 0; attr < nr_attrs; attr++) {
+	 float data[4];
+
+	 const char *src = (tg->attrib[attr].input_ptr + 
+			    tg->attrib[attr].input_stride * elt);
+
+	 char *dst = (vert + 
+		      tg->attrib[attr].output_offset);
+
+	 tg->attrib[attr].fetch( src, data );
+
+         if (0) debug_printf("vert %d/%d attr %d: %f %f %f %f\n",
+                             i, elt, attr, data[0], data[1], data[2], data[3]);
+
+	 tg->attrib[attr].emit( data, dst );
+      }
+      
+      vert += tg->translate.key.output_stride;
+   }
+}
+
+
+
+static void PIPE_CDECL generic_run( struct translate *translate,
+                                    unsigned start,
+                                    unsigned count,
+                                    void *output_buffer )
+{
+   struct translate_generic *tg = translate_generic(translate);
+   char *vert = output_buffer;
+   unsigned nr_attrs = tg->nr_attrib;
+   unsigned attr;
+   unsigned i;
+
+   /* loop over vertex attributes (vertex shader inputs)
+    */
+   for (i = 0; i < count; i++) {
+      unsigned elt = start + i;
+
+      for (attr = 0; attr < nr_attrs; attr++) {
+	 float data[4];
+
+	 const char *src = (tg->attrib[attr].input_ptr + 
+			    tg->attrib[attr].input_stride * elt);
+
+	 char *dst = (vert + 
+		      tg->attrib[attr].output_offset);
+
+	 tg->attrib[attr].fetch( src, data );
+
+         if (0) debug_printf("vert %d attr %d: %f %f %f %f\n",
+                             i, attr, data[0], data[1], data[2], data[3]);
+
+	 tg->attrib[attr].emit( data, dst );
+      }
+      
+      vert += tg->translate.key.output_stride;
+   }
+}
+
+
+			       
+static void generic_set_buffer( struct translate *translate,
+				unsigned buf,
+				const void *ptr,
+				unsigned stride )
+{
+   struct translate_generic *tg = translate_generic(translate);
+   unsigned i;
+
+   for (i = 0; i < tg->nr_attrib; i++) {
+      if (tg->attrib[i].buffer == buf) {
+	 tg->attrib[i].input_ptr = ((char *)ptr +
+				    tg->attrib[i].input_offset);
+	 tg->attrib[i].input_stride = stride;
+      }
+   }
+}
+
+
+static void generic_release( struct translate *translate )
+{
+   /* Refcount?
+    */
+   FREE(translate);
+}
+
+struct translate *translate_generic_create( const struct translate_key *key )
+{
+   struct translate_generic *tg = CALLOC_STRUCT(translate_generic);
+   unsigned i;
+
+   if (tg == NULL)
+      return NULL;
+
+   tg->translate.key = *key;
+   tg->translate.release = generic_release;
+   tg->translate.set_buffer = generic_set_buffer;
+   tg->translate.run_elts = generic_run_elts;
+   tg->translate.run = generic_run;
+
+   for (i = 0; i < key->nr_elements; i++) {
+
+      tg->attrib[i].fetch = get_fetch_func(key->element[i].input_format);
+      tg->attrib[i].buffer = key->element[i].input_buffer;
+      tg->attrib[i].input_offset = key->element[i].input_offset;
+
+      tg->attrib[i].emit = get_emit_func(key->element[i].output_format);
+      tg->attrib[i].output_offset = key->element[i].output_offset;
+
+   }
+
+   tg->nr_attrib = key->nr_elements;
+
+
+   return &tg->translate;
+}
diff --git a/src/gallium/auxiliary/translate/translate_sse.c b/src/gallium/auxiliary/translate/translate_sse.c
new file mode 100644
index 00000000000..b62db8d8f33
--- /dev/null
+++ b/src/gallium/auxiliary/translate/translate_sse.c
@@ -0,0 +1,706 @@
+/*
+ * Copyright 2003 Tungsten Graphics, inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <[email protected]>
+ */
+
+
+#include "pipe/p_config.h"
+#include "pipe/p_compiler.h"
+#include "util/u_memory.h"
+#include "util/u_math.h"
+
+#include "translate.h"
+
+
+#if defined(PIPE_ARCH_X86)
+
+#include "rtasm/rtasm_cpu.h"
+#include "rtasm/rtasm_x86sse.h"
+
+
+#define X    0
+#define Y    1
+#define Z    2
+#define W    3
+
+
+typedef void (PIPE_CDECL *run_func)( struct translate *translate,
+                                     unsigned start,
+                                     unsigned count,
+                                     void *output_buffer );
+
+typedef void (PIPE_CDECL *run_elts_func)( struct translate *translate,
+                                          const unsigned *elts,
+                                          unsigned count,
+                                          void *output_buffer );
+
+struct translate_buffer {
+   const void *base_ptr;
+   unsigned stride;
+   void *ptr;                   /* updated per vertex */
+};
+
+
+struct translate_sse {
+   struct translate translate;
+
+   struct x86_function linear_func;
+   struct x86_function elt_func;
+   struct x86_function *func;
+
+   boolean loaded_identity;
+   boolean loaded_255;
+   boolean loaded_inv_255;
+
+   float identity[4];
+   float float_255[4];
+   float inv_255[4];
+
+   struct translate_buffer buffer[PIPE_MAX_ATTRIBS];
+   unsigned nr_buffers;
+
+   run_func      gen_run;
+   run_elts_func gen_run_elts;
+
+   /* these are actually known values, but putting them in a struct
+    * like this is helpful to keep them in sync across the file.
+    */
+   struct x86_reg tmp_EAX;
+   struct x86_reg idx_EBX;     /* either start+i or &elt[i] */
+   struct x86_reg outbuf_ECX;
+   struct x86_reg machine_EDX;
+   struct x86_reg count_ESI;    /* decrements to zero */
+};
+
+static int get_offset( const void *a, const void *b )
+{
+   return (const char *)b - (const char *)a;
+}
+
+
+
+static struct x86_reg get_identity( struct translate_sse *p )
+{
+   struct x86_reg reg = x86_make_reg(file_XMM, 6);
+
+   if (!p->loaded_identity) {
+      p->loaded_identity = TRUE;
+      p->identity[0] = 0;
+      p->identity[1] = 0;
+      p->identity[2] = 0;
+      p->identity[3] = 1;
+
+      sse_movups(p->func, reg, 
+		 x86_make_disp(p->machine_EDX, 
+			       get_offset(p, &p->identity[0])));
+   }
+
+   return reg;
+}
+
+static struct x86_reg get_255( struct translate_sse *p )
+{
+   struct x86_reg reg = x86_make_reg(file_XMM, 7);
+
+   if (!p->loaded_255) {
+      p->loaded_255 = TRUE;
+      p->float_255[0] =
+	 p->float_255[1] =
+	 p->float_255[2] =
+	 p->float_255[3] = 255.0f;
+
+      sse_movups(p->func, reg, 
+		 x86_make_disp(p->machine_EDX, 
+			       get_offset(p, &p->float_255[0])));
+   }
+
+   return reg;
+}
+
+static struct x86_reg get_inv_255( struct translate_sse *p )
+{
+   struct x86_reg reg = x86_make_reg(file_XMM, 5);
+
+   if (!p->loaded_inv_255) {
+      p->loaded_inv_255 = TRUE;
+      p->inv_255[0] =
+	 p->inv_255[1] =
+	 p->inv_255[2] =
+	 p->inv_255[3] = 1.0f / 255.0f;
+
+      sse_movups(p->func, reg, 
+		 x86_make_disp(p->machine_EDX, 
+			       get_offset(p, &p->inv_255[0])));
+   }
+
+   return reg;
+}
+
+
+static void emit_load_R32G32B32A32( struct translate_sse *p, 			   
+				    struct x86_reg data,
+				    struct x86_reg arg0 )
+{
+   sse_movups(p->func, data, arg0);
+}
+
+static void emit_load_R32G32B32( struct translate_sse *p, 			   
+				 struct x86_reg data,
+				 struct x86_reg arg0 )
+{
+   /* Have to jump through some hoops:
+    *
+    * c 0 0 0
+    * c 0 0 1
+    * 0 0 c 1
+    * a b c 1
+    */
+   sse_movss(p->func, data, x86_make_disp(arg0, 8));
+   sse_shufps(p->func, data, get_identity(p), SHUF(X,Y,Z,W) );
+   sse_shufps(p->func, data, data, SHUF(Y,Z,X,W) );
+   sse_movlps(p->func, data, arg0);
+}
+
+static void emit_load_R32G32( struct translate_sse *p, 
+			   struct x86_reg data,
+			   struct x86_reg arg0 )
+{
+   /* 0 0 0 1
+    * a b 0 1
+    */
+   sse_movups(p->func, data, get_identity(p) );
+   sse_movlps(p->func, data, arg0);
+}
+
+
+static void emit_load_R32( struct translate_sse *p, 
+			   struct x86_reg data,
+			   struct x86_reg arg0 )
+{
+   /* a 0 0 0
+    * a 0 0 1
+    */
+   sse_movss(p->func, data, arg0);
+   sse_orps(p->func, data, get_identity(p) );
+}
+
+
+static void emit_load_R8G8B8A8_UNORM( struct translate_sse *p,
+				       struct x86_reg data,
+				       struct x86_reg src )
+{
+
+   /* Load and unpack twice:
+    */
+   sse_movss(p->func, data, src);
+   sse2_punpcklbw(p->func, data, get_identity(p));
+   sse2_punpcklbw(p->func, data, get_identity(p));
+
+   /* Convert to float:
+    */
+   sse2_cvtdq2ps(p->func, data, data);
+
+
+   /* Scale by 1/255.0
+    */
+   sse_mulps(p->func, data, get_inv_255(p));
+}
+
+
+
+
+static void emit_store_R32G32B32A32( struct translate_sse *p, 			   
+				     struct x86_reg dest,
+				     struct x86_reg dataXMM )
+{
+   sse_movups(p->func, dest, dataXMM);
+}
+
+static void emit_store_R32G32B32( struct translate_sse *p, 
+				  struct x86_reg dest,
+				  struct x86_reg dataXMM )
+{
+   /* Emit two, shuffle, emit one.
+    */
+   sse_movlps(p->func, dest, dataXMM);
+   sse_shufps(p->func, dataXMM, dataXMM, SHUF(Z,Z,Z,Z) ); /* NOTE! destructive */
+   sse_movss(p->func, x86_make_disp(dest,8), dataXMM);
+}
+
+static void emit_store_R32G32( struct translate_sse *p, 
+			       struct x86_reg dest,
+			       struct x86_reg dataXMM )
+{
+   sse_movlps(p->func, dest, dataXMM);
+}
+
+static void emit_store_R32( struct translate_sse *p, 
+			    struct x86_reg dest,
+			    struct x86_reg dataXMM )
+{
+   sse_movss(p->func, dest, dataXMM);
+}
+
+
+
+static void emit_store_R8G8B8A8_UNORM( struct translate_sse *p,
+				       struct x86_reg dest,
+				       struct x86_reg dataXMM )
+{
+   /* Scale by 255.0
+    */
+   sse_mulps(p->func, dataXMM, get_255(p));
+
+   /* Pack and emit:
+    */
+   sse2_cvtps2dq(p->func, dataXMM, dataXMM);
+   sse2_packssdw(p->func, dataXMM, dataXMM);
+   sse2_packuswb(p->func, dataXMM, dataXMM);
+   sse_movss(p->func, dest, dataXMM);
+}
+
+
+
+
+
+/* Extended swizzles?  Maybe later.
+ */  
+static void emit_swizzle( struct translate_sse *p,
+			  struct x86_reg dest,
+			  struct x86_reg src,
+			  unsigned char shuffle )
+{
+   sse_shufps(p->func, dest, src, shuffle);
+}
+
+
+static boolean translate_attr( struct translate_sse *p,
+			       const struct translate_element *a,
+			       struct x86_reg srcECX,
+			       struct x86_reg dstEAX)
+{
+   struct x86_reg dataXMM = x86_make_reg(file_XMM, 0);
+
+   switch (a->input_format) {
+   case PIPE_FORMAT_R32_FLOAT:
+      emit_load_R32(p, dataXMM, srcECX);
+      break;
+   case PIPE_FORMAT_R32G32_FLOAT:
+      emit_load_R32G32(p, dataXMM, srcECX);
+      break;
+   case PIPE_FORMAT_R32G32B32_FLOAT:
+      emit_load_R32G32B32(p, dataXMM, srcECX);
+      break;
+   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+      emit_load_R32G32B32A32(p, dataXMM, srcECX);
+      break;
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      emit_load_R8G8B8A8_UNORM(p, dataXMM, srcECX);
+      emit_swizzle(p, dataXMM, dataXMM, SHUF(Z,Y,X,W));
+      break;
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      emit_load_R8G8B8A8_UNORM(p, dataXMM, srcECX);
+      break;
+   default:
+      return FALSE;
+   }
+
+   switch (a->output_format) {
+   case PIPE_FORMAT_R32_FLOAT:
+      emit_store_R32(p, dstEAX, dataXMM);
+      break;
+   case PIPE_FORMAT_R32G32_FLOAT:
+      emit_store_R32G32(p, dstEAX, dataXMM);
+      break;
+   case PIPE_FORMAT_R32G32B32_FLOAT:
+      emit_store_R32G32B32(p, dstEAX, dataXMM);
+      break;
+   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+      emit_store_R32G32B32A32(p, dstEAX, dataXMM);
+      break;
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      emit_swizzle(p, dataXMM, dataXMM, SHUF(Z,Y,X,W));
+      emit_store_R8G8B8A8_UNORM(p, dstEAX, dataXMM);
+      break;
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      emit_store_R8G8B8A8_UNORM(p, dstEAX, dataXMM);
+      break;
+   default:
+      return FALSE;
+   }
+
+   return TRUE;
+}
+
+
+static boolean init_inputs( struct translate_sse *p,
+                            boolean linear )
+{
+   unsigned i;
+   if (linear) {
+      for (i = 0; i < p->nr_buffers; i++) {
+         struct x86_reg buf_stride   = x86_make_disp(p->machine_EDX,
+                                                     get_offset(p, &p->buffer[i].stride));
+         struct x86_reg buf_ptr      = x86_make_disp(p->machine_EDX,
+                                                     get_offset(p, &p->buffer[i].ptr));
+         struct x86_reg buf_base_ptr = x86_make_disp(p->machine_EDX,
+                                                     get_offset(p, &p->buffer[i].base_ptr));
+         struct x86_reg elt = p->idx_EBX;
+         struct x86_reg tmp = p->tmp_EAX;
+
+
+         /* Calculate pointer to first attrib:
+          */
+         x86_mov(p->func, tmp, buf_stride);
+         x86_imul(p->func, tmp, elt);
+         x86_add(p->func, tmp, buf_base_ptr);
+
+
+         /* In the linear case, keep the buffer pointer instead of the
+          * index number.
+          */
+         if (p->nr_buffers == 1) 
+            x86_mov( p->func, elt, tmp );
+         else
+            x86_mov( p->func, buf_ptr, tmp );
+      }
+   }
+
+   return TRUE;
+}
+
+
+static struct x86_reg get_buffer_ptr( struct translate_sse *p,
+                                      boolean linear,
+                                      unsigned buf_idx,
+                                      struct x86_reg elt )
+{
+   if (linear && p->nr_buffers == 1) {
+      return p->idx_EBX;
+   }
+   else if (linear) {
+      struct x86_reg ptr = p->tmp_EAX;
+      struct x86_reg buf_ptr = 
+         x86_make_disp(p->machine_EDX, 
+                       get_offset(p, &p->buffer[buf_idx].ptr));
+      
+      x86_mov(p->func, ptr, buf_ptr);
+      return ptr;
+   }
+   else {
+      struct x86_reg ptr = p->tmp_EAX;
+
+      struct x86_reg buf_stride = 
+         x86_make_disp(p->machine_EDX, 
+                       get_offset(p, &p->buffer[buf_idx].stride));
+
+      struct x86_reg buf_base_ptr = 
+         x86_make_disp(p->machine_EDX, 
+                       get_offset(p, &p->buffer[buf_idx].base_ptr));
+
+
+
+      /* Calculate pointer to current attrib:
+       */
+      x86_mov(p->func, ptr, buf_stride);
+      x86_imul(p->func, ptr, elt);
+      x86_add(p->func, ptr, buf_base_ptr);
+      return ptr;
+   }
+}
+
+
+
+static boolean incr_inputs( struct translate_sse *p, 
+                            boolean linear )
+{
+   if (linear && p->nr_buffers == 1) {
+      struct x86_reg stride = x86_make_disp(p->machine_EDX,
+                                            get_offset(p, &p->buffer[0].stride));
+
+      x86_add(p->func, p->idx_EBX, stride);
+      sse_prefetchnta(p->func, x86_make_disp(p->idx_EBX, 192));
+   }
+   else if (linear) {
+      unsigned i;
+
+      /* Is this worthwhile??
+       */
+      for (i = 0; i < p->nr_buffers; i++) {
+         struct x86_reg buf_ptr = x86_make_disp(p->machine_EDX,
+                                                get_offset(p, &p->buffer[i].ptr));
+         struct x86_reg buf_stride = x86_make_disp(p->machine_EDX,
+                                                   get_offset(p, &p->buffer[i].stride));
+
+         x86_mov(p->func, p->tmp_EAX, buf_ptr);
+         x86_add(p->func, p->tmp_EAX, buf_stride);
+         if (i == 0) sse_prefetchnta(p->func, x86_make_disp(p->tmp_EAX, 192));
+         x86_mov(p->func, buf_ptr, p->tmp_EAX);
+      }
+   } 
+   else {
+      x86_lea(p->func, p->idx_EBX, x86_make_disp(p->idx_EBX, 4));
+   }
+   
+   return TRUE;
+}
+
+
+/* Build run( struct translate *machine,
+ *            unsigned start,
+ *            unsigned count,
+ *            void *output_buffer )
+ * or
+ *  run_elts( struct translate *machine,
+ *            unsigned *elts,
+ *            unsigned count,
+ *            void *output_buffer )
+ *
+ *  Lots of hardcoding
+ *
+ * EAX -- pointer to current output vertex
+ * ECX -- pointer to current attribute 
+ * 
+ */
+static boolean build_vertex_emit( struct translate_sse *p,
+				  struct x86_function *func,
+				  boolean linear )
+{
+   int fixup, label;
+   unsigned j;
+
+   p->tmp_EAX       = x86_make_reg(file_REG32, reg_AX);
+   p->idx_EBX       = x86_make_reg(file_REG32, reg_BX);
+   p->outbuf_ECX    = x86_make_reg(file_REG32, reg_CX);
+   p->machine_EDX   = x86_make_reg(file_REG32, reg_DX);
+   p->count_ESI     = x86_make_reg(file_REG32, reg_SI);
+
+   p->func = func;
+   p->loaded_inv_255 = FALSE;
+   p->loaded_255 = FALSE;
+   p->loaded_identity = FALSE;
+
+   x86_init_func(p->func);
+
+   /* Push a few regs?
+    */
+   x86_push(p->func, p->idx_EBX);
+   x86_push(p->func, p->count_ESI);
+
+   /* Load arguments into regs:
+    */
+   x86_mov(p->func, p->machine_EDX, x86_fn_arg(p->func, 1));
+   x86_mov(p->func, p->idx_EBX, x86_fn_arg(p->func, 2));
+   x86_mov(p->func, p->count_ESI, x86_fn_arg(p->func, 3));
+   x86_mov(p->func, p->outbuf_ECX, x86_fn_arg(p->func, 4));
+
+   /* Get vertex count, compare to zero
+    */
+   x86_xor(p->func, p->tmp_EAX, p->tmp_EAX);
+   x86_cmp(p->func, p->count_ESI, p->tmp_EAX);
+   fixup = x86_jcc_forward(p->func, cc_E);
+
+   /* always load, needed or not:
+    */
+   init_inputs(p, linear);
+
+   /* Note address for loop jump
+    */
+   label = x86_get_label(p->func);
+   {
+      struct x86_reg elt = linear ? p->idx_EBX : x86_deref(p->idx_EBX);
+      int last_vb = -1;
+      struct x86_reg vb;
+
+      for (j = 0; j < p->translate.key.nr_elements; j++) {
+         const struct translate_element *a = &p->translate.key.element[j];
+
+         /* Figure out source pointer address:
+          */
+         if (a->input_buffer != last_vb) {
+            last_vb = a->input_buffer;
+            vb = get_buffer_ptr(p, linear, a->input_buffer, elt);
+         }
+         
+         if (!translate_attr( p, a, 
+                              x86_make_disp(vb, a->input_offset), 
+                              x86_make_disp(p->outbuf_ECX, a->output_offset)))
+            return FALSE;
+      }
+
+      /* Next output vertex:
+       */
+      x86_lea(p->func, 
+              p->outbuf_ECX, 
+              x86_make_disp(p->outbuf_ECX, 
+                            p->translate.key.output_stride));
+
+      /* Incr index
+       */ 
+      incr_inputs( p, linear );
+   }
+
+   /* decr count, loop if not zero
+    */
+   x86_dec(p->func, p->count_ESI);
+   x86_jcc(p->func, cc_NZ, label);
+
+   /* Exit mmx state?
+    */
+   if (p->func->need_emms)
+      mmx_emms(p->func);
+
+   /* Land forward jump here:
+    */
+   x86_fixup_fwd_jump(p->func, fixup);
+
+   /* Pop regs and return
+    */
+   
+   x86_pop(p->func, p->count_ESI);
+   x86_pop(p->func, p->idx_EBX);
+   x86_ret(p->func);
+
+   return TRUE;
+}
+
+
+
+
+
+
+			       
+static void translate_sse_set_buffer( struct translate *translate,
+				unsigned buf,
+				const void *ptr,
+				unsigned stride )
+{
+   struct translate_sse *p = (struct translate_sse *)translate;
+
+   if (buf < p->nr_buffers) {
+      p->buffer[buf].base_ptr = (char *)ptr;
+      p->buffer[buf].stride = stride;
+   }
+
+   if (0) debug_printf("%s %d/%d: %p %d\n", 
+                       __FUNCTION__, buf, 
+                       p->nr_buffers, 
+                       ptr, stride);
+}
+
+
+static void translate_sse_release( struct translate *translate )
+{
+   struct translate_sse *p = (struct translate_sse *)translate;
+
+   x86_release_func( &p->linear_func );
+   x86_release_func( &p->elt_func );
+
+   FREE(p);
+}
+
+static void PIPE_CDECL translate_sse_run_elts( struct translate *translate,
+			      const unsigned *elts,
+			      unsigned count,
+			      void *output_buffer )
+{
+   struct translate_sse *p = (struct translate_sse *)translate;
+
+   p->gen_run_elts( translate,
+		    elts,
+		    count,
+		    output_buffer );
+}
+
+static void PIPE_CDECL translate_sse_run( struct translate *translate,
+			 unsigned start,
+			 unsigned count,
+			 void *output_buffer )
+{
+   struct translate_sse *p = (struct translate_sse *)translate;
+
+   p->gen_run( translate,
+	       start,
+	       count,
+	       output_buffer );
+}
+
+
+struct translate *translate_sse2_create( const struct translate_key *key )
+{
+   struct translate_sse *p = NULL;
+   unsigned i;
+
+   if (!rtasm_cpu_has_sse() || !rtasm_cpu_has_sse2())
+      goto fail;
+
+   p = CALLOC_STRUCT( translate_sse );
+   if (p == NULL) 
+      goto fail;
+
+   p->translate.key = *key;
+   p->translate.release = translate_sse_release;
+   p->translate.set_buffer = translate_sse_set_buffer;
+   p->translate.run_elts = translate_sse_run_elts;
+   p->translate.run = translate_sse_run;
+
+   for (i = 0; i < key->nr_elements; i++) 
+      p->nr_buffers = MAX2( p->nr_buffers, key->element[i].input_buffer + 1 );
+
+   if (0) debug_printf("nr_buffers: %d\n", p->nr_buffers);
+
+   if (!build_vertex_emit(p, &p->linear_func, TRUE))
+      goto fail;
+
+   if (!build_vertex_emit(p, &p->elt_func, FALSE))
+      goto fail;
+
+   p->gen_run = (run_func)x86_get_func(&p->linear_func);
+   if (p->gen_run == NULL)
+      goto fail;
+
+   p->gen_run_elts = (run_elts_func)x86_get_func(&p->elt_func);
+   if (p->gen_run_elts == NULL)
+      goto fail;
+
+   return &p->translate;
+
+ fail:
+   if (p)
+      translate_sse_release( &p->translate );
+
+   return NULL;
+}
+
+
+
+#else
+
+struct translate *translate_sse2_create( const struct translate_key *key )
+{
+   return NULL;
+}
+
+#endif
diff --git a/src/gallium/auxiliary/util/Makefile b/src/gallium/auxiliary/util/Makefile
new file mode 100644
index 00000000000..44c23777218
--- /dev/null
+++ b/src/gallium/auxiliary/util/Makefile
@@ -0,0 +1,32 @@
+TOP = ../../../..
+include $(TOP)/configs/current
+
+LIBNAME = util
+
+C_SOURCES = \
+	p_debug.c \
+	u_blit.c \
+	u_cache.c \
+	u_draw_quad.c \
+	u_gen_mipmap.c \
+	u_handle_table.c \
+	u_hash_table.c \
+	u_hash.c \
+	u_keymap.c \
+	u_linear.c \
+	u_math.c \
+	u_mm.c \
+	u_rect.c \
+	u_simple_shaders.c \
+	u_snprintf.c \
+	u_stream_stdc.c \
+	u_stream_wd.c \
+	u_tile.c \
+	u_time.c \
+	u_timed_winsys.c \
+	u_simple_screen.c
+
+include ../../Makefile.template
+
+symlinks:
+
diff --git a/src/gallium/auxiliary/util/SConscript b/src/gallium/auxiliary/util/SConscript
new file mode 100644
index 00000000000..35f683fb8e6
--- /dev/null
+++ b/src/gallium/auxiliary/util/SConscript
@@ -0,0 +1,30 @@
+Import('*')
+
+util = env.ConvenienceLibrary(
+	target = 'util',
+	source = [
+		'p_debug.c',
+		'p_debug_mem.c',
+		'p_debug_prof.c',
+		'u_blit.c',
+		'u_cache.c',
+		'u_draw_quad.c',
+		'u_gen_mipmap.c',
+		'u_handle_table.c',
+		'u_hash.c',
+		'u_hash_table.c',
+		'u_keymap.c',
+		'u_math.c',
+		'u_mm.c',
+		'u_rect.c',
+		'u_simple_shaders.c',
+		'u_snprintf.c',
+		'u_stream_stdc.c',
+		'u_stream_wd.c',
+		'u_tile.c',
+		'u_time.c',
+		'u_timed_winsys.c',
+		'u_simple_screen.c',
+	])
+
+auxiliaries.insert(0, util)
diff --git a/src/gallium/auxiliary/util/p_debug.c b/src/gallium/auxiliary/util/p_debug.c
new file mode 100644
index 00000000000..f373f941dd8
--- /dev/null
+++ b/src/gallium/auxiliary/util/p_debug.c
@@ -0,0 +1,782 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright (c) 2008 VMware, Inc.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "pipe/p_config.h" 
+
+#include <stdarg.h>
+
+
+#ifdef PIPE_SUBSYSTEM_WINDOWS_DISPLAY
+
+#include <windows.h>
+#include <winddi.h>
+
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_CE)
+
+#include <stdio.h> 
+#include <stdlib.h> 
+#include <windows.h> 
+#include <types.h> 
+
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_USER)
+
+#ifndef WIN32_LEAN_AND_MEAN
+#define WIN32_LEAN_AND_MEAN      // Exclude rarely-used stuff from Windows headers
+#endif
+#include <windows.h>
+
+#else
+
+#include <stdio.h>
+#include <stdlib.h>
+
+#endif
+
+#include "pipe/p_compiler.h" 
+#include "pipe/p_debug.h" 
+#include "pipe/p_format.h" 
+#include "pipe/p_state.h" 
+#include "pipe/p_inlines.h" 
+#include "util/u_memory.h" 
+#include "util/u_string.h" 
+#include "util/u_stream.h" 
+#include "util/u_math.h" 
+#include "util/u_tile.h" 
+
+
+#ifdef PIPE_SUBSYSTEM_WINDOWS_DISPLAY
+static INLINE void 
+_EngDebugPrint(const char *format, ...)
+{
+   va_list ap;
+   va_start(ap, format);
+   EngDebugPrint("", (PCHAR)format, ap);
+   va_end(ap);
+}
+#endif
+
+
+void _debug_vprintf(const char *format, va_list ap)
+{
+#if defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY)
+   /* EngDebugPrint does not handle float point arguments, so we need to use
+    * our own vsnprintf implementation. It is also very slow, so buffer until
+    * we find a newline. */
+   static char buf[512] = {'\0'};
+   size_t len = strlen(buf);
+   int ret = util_vsnprintf(buf + len, sizeof(buf) - len, format, ap);
+   if(ret > (int)(sizeof(buf) - len - 1) || util_strchr(buf + len, '\n')) {
+      _EngDebugPrint("%s", buf);
+      buf[0] = '\0';
+   }
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_USER)
+   /* EngDebugPrint does not handle float point arguments, so we need to use
+    * our own vsnprintf implementation. It is also very slow, so buffer until
+    * we find a newline. */
+   static char buf[512 + 1] = {'\0'};
+   size_t len = strlen(buf);
+   int ret = util_vsnprintf(buf + len, sizeof(buf) - len, format, ap);
+   if(ret > (int)(sizeof(buf) - len - 1) || util_strchr(buf + len, '\n')) {
+      OutputDebugStringA(buf);
+      buf[0] = '\0';
+   }
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_CE)
+   wchar_t *wide_format;
+   long wide_str_len;   
+   char buf[512];   
+   int ret;   
+#if (_WIN32_WCE < 600)
+   ret = vsprintf(buf, format, ap);   
+   if(ret < 0){   
+       sprintf(buf, "Cant handle debug print!");   
+       ret = 25;
+   }
+#else
+   ret = vsprintf_s(buf, 512, format, ap);   
+   if(ret < 0){   
+       sprintf_s(buf, 512, "Cant handle debug print!");   
+       ret = 25;
+   }
+#endif
+   buf[ret] = '\0';   
+   /* Format is ascii - needs to be converted to wchar_t for printing */   
+   wide_str_len = MultiByteToWideChar(CP_ACP, 0, (const char *) buf, -1, NULL, 0);   
+   wide_format = (wchar_t *) malloc((wide_str_len+1) * sizeof(wchar_t));   
+   if (wide_format) {   
+      MultiByteToWideChar(CP_ACP, 0, (const char *) buf, -1,   
+            wide_format, wide_str_len);   
+      NKDbgPrintfW(wide_format, wide_format);   
+      free(wide_format);   
+   } 
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT)
+   /* TODO */
+#else /* !PIPE_SUBSYSTEM_WINDOWS */
+#ifdef DEBUG
+   vfprintf(stderr, format, ap);
+#endif
+#endif
+}
+
+
+#ifdef DEBUG
+void debug_print_blob( const char *name,
+                       const void *blob,
+                       unsigned size )
+{
+   const unsigned *ublob = (const unsigned *)blob;
+   unsigned i;
+
+   debug_printf("%s (%d dwords%s)\n", name, size/4,
+                size%4 ? "... plus a few bytes" : "");
+
+   for (i = 0; i < size/4; i++) {
+      debug_printf("%d:\t%08x\n", i, ublob[i]);
+   }
+}
+#endif
+
+
+void _debug_break(void) 
+{
+#if defined(PIPE_ARCH_X86) && defined(PIPE_CC_GCC)
+   __asm("int3");
+#elif defined(PIPE_ARCH_X86) && defined(PIPE_CC_MSVC)
+   _asm {int 3};
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY)
+   EngDebugBreak();
+#else
+   abort();
+#endif
+}
+
+
+#ifdef PIPE_SUBSYSTEM_WINDOWS_DISPLAY
+static const char *
+find(const char *start, const char *end, char c) 
+{
+   const char *p;
+   for(p = start; !end || p != end; ++p) {
+      if(*p == c)
+	 return p;
+      if(*p < 32)
+	 break;
+   }
+   return NULL;
+}
+
+static int 
+compare(const char *start, const char *end, const char *s)
+{
+   const char *p, *q;
+   for(p = start, q = s; p != end && *q != '\0'; ++p, ++q) {
+      if(*p != *q)
+	 return 0;
+   }
+   return p == end && *q == '\0';
+}
+
+static void 
+copy(char *dst, const char *start, const char *end, size_t n) 
+{
+   const char *p;
+   char *q;
+   for(p = start, q = dst, n = n - 1; p != end && n; ++p, ++q, --n)
+      *q = *p;
+   *q = '\0';
+}
+#endif
+
+
+static INLINE const char *
+_debug_get_option(const char *name)
+{
+#if defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY)
+   /* EngMapFile creates the file if it does not exists, so it must either be
+    * disabled on release versions (or put in a less conspicuous place). */
+#ifdef DEBUG
+   const char *result = NULL;
+   ULONG_PTR iFile = 0;
+   const void *pMap = NULL;
+   const char *sol, *eol, *sep;
+   static char output[1024];
+   
+   pMap = EngMapFile(L"\\??\\c:\\gallium.cfg", 0, &iFile);
+   if(pMap) {
+      sol = (const char *)pMap;
+      while(1) {
+	 /* TODO: handle LF line endings */
+	 eol = find(sol, NULL, '\r');
+	 if(!eol || eol == sol)
+	    break;
+	 sep = find(sol, eol, '=');
+	 if(!sep)
+	    break;
+	 if(compare(sol, sep, name)) {
+	    copy(output, sep + 1, eol, sizeof(output));
+	    result = output;
+	    break;
+	 }
+	 sol = eol + 2;
+      }
+      EngUnmapFile(iFile);
+   }
+   return result;
+#else
+   return NULL;
+#endif
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_CE) || defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT) 
+   /* TODO: implement */
+   return NULL;
+#else
+   return getenv(name);
+#endif
+}
+
+const char *
+debug_get_option(const char *name, const char *dfault)
+{
+   const char *result;
+
+   result = _debug_get_option(name);
+   if(!result)
+      result = dfault;
+      
+   debug_printf("%s: %s = %s\n", __FUNCTION__, name, result ? result : "(null)");
+   
+   return result;
+}
+
+boolean
+debug_get_bool_option(const char *name, boolean dfault)
+{
+   const char *str = _debug_get_option(name);
+   boolean result;
+   
+   if(str == NULL)
+      result = dfault;
+   else if(!util_strcmp(str, "n"))
+      result = FALSE;
+   else if(!util_strcmp(str, "no"))
+      result = FALSE;
+   else if(!util_strcmp(str, "0"))
+      result = FALSE;
+   else if(!util_strcmp(str, "f"))
+      result = FALSE;
+   else if(!util_strcmp(str, "false"))
+      result = FALSE;
+   else
+      result = TRUE;
+
+   debug_printf("%s: %s = %s\n", __FUNCTION__, name, result ? "TRUE" : "FALSE");
+   
+   return result;
+}
+
+
+long
+debug_get_num_option(const char *name, long dfault)
+{
+   long result;
+   const char *str;
+   
+   str = _debug_get_option(name);
+   if(!str)
+      result = dfault;
+   else {
+      long sign;
+      char c;
+      c = *str++;
+      if(c == '-') {
+	 sign = -1;
+	 c = *str++;
+      } 
+      else {
+	 sign = 1;
+      }
+      result = 0;
+      while('0' <= c && c <= '9') {
+	 result = result*10 + (c - '0');
+	 c = *str++;
+      }
+      result *= sign;
+   }
+   
+   debug_printf("%s: %s = %li\n", __FUNCTION__, name, result);
+
+   return result;
+}
+
+
+unsigned long
+debug_get_flags_option(const char *name, 
+                       const struct debug_named_value *flags,
+                       unsigned long dfault)
+{
+   unsigned long result;
+   const char *str;
+   
+   str = _debug_get_option(name);
+   if(!str)
+      result = dfault;
+   else if (!util_strcmp(str, "help")) {
+      result = dfault;
+      while (flags->name) {
+         debug_printf("%s: help for %s: %s [0x%lx]\n", __FUNCTION__, name, flags->name, flags->value);
+         flags++;
+      }
+   }
+   else {
+      result = 0;
+      while( flags->name ) {
+	 if (!util_strcmp(str, "all") || util_strstr(str, flags->name ))
+	    result |= flags->value;
+	 ++flags;
+      }
+   }
+
+   if (str) {
+      debug_printf("%s: %s = 0x%lx (%s)\n", __FUNCTION__, name, result, str);
+   }
+   else {
+      debug_printf("%s: %s = 0x%lx\n", __FUNCTION__, name, result);
+   }
+
+   return result;
+}
+
+
+void _debug_assert_fail(const char *expr, 
+                        const char *file, 
+                        unsigned line, 
+                        const char *function) 
+{
+   _debug_printf("%s:%u:%s: Assertion `%s' failed.\n", file, line, function, expr);
+#if defined(PIPE_OS_WINDOWS) && !defined(PIPE_SUBSYSTEM_WINDOWS_USER)
+   if (debug_get_bool_option("GALLIUM_ABORT_ON_ASSERT", FALSE))
+#else
+   if (debug_get_bool_option("GALLIUM_ABORT_ON_ASSERT", TRUE))
+#endif
+      debug_break();
+   else
+      _debug_printf("continuing...\n");
+}
+
+
+const char *
+debug_dump_enum(const struct debug_named_value *names, 
+                unsigned long value)
+{
+   static char rest[64];
+   
+   while(names->name) {
+      if(names->value == value)
+	 return names->name;
+      ++names;
+   }
+
+   util_snprintf(rest, sizeof(rest), "0x%08lx", value);
+   return rest;
+}
+
+
+const char *
+debug_dump_enum_noprefix(const struct debug_named_value *names, 
+                         const char *prefix,
+                         unsigned long value)
+{
+   static char rest[64];
+   
+   while(names->name) {
+      if(names->value == value) {
+         const char *name = names->name;
+         while (*name == *prefix) {
+            name++;
+            prefix++;
+         }
+         return name;
+      }
+      ++names;
+   }
+
+   
+
+   util_snprintf(rest, sizeof(rest), "0x%08lx", value);
+   return rest;
+}
+
+
+const char *
+debug_dump_flags(const struct debug_named_value *names, 
+                 unsigned long value)
+{
+   static char output[4096];
+   static char rest[256];
+   int first = 1;
+
+   output[0] = '\0';
+
+   while(names->name) {
+      if((names->value & value) == names->value) {
+	 if (!first)
+	    util_strncat(output, "|", sizeof(output));
+	 else
+	    first = 0;
+	 util_strncat(output, names->name, sizeof(output));
+	 value &= ~names->value;
+      }
+      ++names;
+   }
+   
+   if (value) {
+      if (!first)
+	 util_strncat(output, "|", sizeof(output));
+      else
+	 first = 0;
+      
+      util_snprintf(rest, sizeof(rest), "0x%08lx", value);
+      util_strncat(output, rest, sizeof(output));
+   }
+   
+   if(first)
+      return "0";
+   
+   return output;
+}
+
+
+static const struct debug_named_value pipe_format_names[] = {
+#ifdef DEBUG
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_NONE),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_A8R8G8B8_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_X8R8G8B8_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_B8G8R8A8_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_B8G8R8X8_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_A1R5G5B5_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_A4R4G4B4_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R5G6B5_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_A2B10G10R10_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_L8_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_A8_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_I8_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_A8L8_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_L16_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_YCBCR),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_YCBCR_REV),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_Z16_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_Z32_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_Z32_FLOAT),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_S8Z24_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_Z24S8_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_X8Z24_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_Z24X8_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_S8_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R64_FLOAT),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R64G64_FLOAT),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R64G64B64_FLOAT),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R64G64B64A64_FLOAT),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32_FLOAT),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32_FLOAT),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32_FLOAT),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32A32_FLOAT),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32A32_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32_USCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32_USCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32_USCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32A32_USCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32_SNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32_SNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32_SNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32A32_SNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32_SSCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32_SSCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32_SSCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R32G32B32A32_SSCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R16_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16B16_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16B16A16_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R16_USCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16_USCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16B16_USCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16B16A16_USCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R16_SNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16_SNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16B16_SNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16B16A16_SNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R16_SSCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16_SSCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16B16_SSCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R16G16B16A16_SSCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8A8_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8X8_UNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8_USCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8_USCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8_USCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8A8_USCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8X8_USCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8_SNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8_SNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8_SNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8A8_SNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8X8_SNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_B6G5R5_SNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_A8B8G8R8_SNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_X8B8G8R8_SNORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8_SSCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8_SSCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8_SSCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8A8_SSCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8X8_SSCALED),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_L8_SRGB),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_A8L8_SRGB),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8_SRGB),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8A8_SRGB),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_R8G8B8X8_SRGB),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_A8R8G8B8_SRGB),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_X8R8G8B8_SRGB),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_B8G8R8A8_SRGB),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_B8G8R8X8_SRGB),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_X8UB8UG8SR8S_NORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_B6UG5SR5S_NORM),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_DXT1_RGB),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_DXT1_RGBA),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_DXT3_RGBA),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_DXT5_RGBA),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_DXT1_SRGB),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_DXT1_SRGBA),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_DXT3_SRGBA),
+   DEBUG_NAMED_VALUE(PIPE_FORMAT_DXT5_SRGBA),
+#endif
+   DEBUG_NAMED_VALUE_END
+};
+
+#ifdef DEBUG
+void debug_print_format(const char *msg, unsigned fmt )
+{
+   debug_printf("%s: %s\n", msg, debug_dump_enum(pipe_format_names, fmt)); 
+}
+#endif
+
+const char *pf_name( enum pipe_format format )
+{
+   return debug_dump_enum(pipe_format_names, format);
+}
+
+
+#ifdef DEBUG
+void debug_dump_image(const char *prefix,
+                      unsigned format, unsigned cpp,
+                      unsigned width, unsigned height,
+                      unsigned stride,
+                      const void *data)     
+{
+#ifdef PIPE_SUBSYSTEM_WINDOWS_DISPLAY
+   static unsigned no = 0; 
+   char filename[256];
+   WCHAR wfilename[sizeof(filename)];
+   ULONG_PTR iFile = 0;
+   struct {
+      unsigned format;
+      unsigned cpp;
+      unsigned width;
+      unsigned height;
+   } header;
+   unsigned char *pMap = NULL;
+   unsigned i;
+
+   util_snprintf(filename, sizeof(filename), "\\??\\c:\\%03u%s.raw", ++no, prefix);
+   for(i = 0; i < sizeof(filename); ++i)
+      wfilename[i] = (WCHAR)filename[i];
+   
+   pMap = (unsigned char *)EngMapFile(wfilename, sizeof(header) + height*width*cpp, &iFile);
+   if(!pMap)
+      return;
+   
+   header.format = format;
+   header.cpp = cpp;
+   header.width = width;
+   header.height = height;
+   memcpy(pMap, &header, sizeof(header));
+   pMap += sizeof(header);
+   
+   for(i = 0; i < height; ++i) {
+      memcpy(pMap, (unsigned char *)data + stride*i, cpp*width);
+      pMap += cpp*width;
+   }
+      
+   EngUnmapFile(iFile);
+#endif
+}
+
+void debug_dump_surface(const char *prefix,
+                        struct pipe_surface *surface)     
+{
+   unsigned surface_usage;
+   void *data;
+
+   if (!surface)
+      goto error1;
+
+   /* XXX: force mappable surface */
+   surface_usage = surface->usage;
+   surface->usage |= PIPE_BUFFER_USAGE_CPU_READ;
+   
+   data = pipe_surface_map(surface, 
+                           PIPE_BUFFER_USAGE_CPU_READ);
+   if(!data)
+      goto error2;
+   
+   debug_dump_image(prefix, 
+                    surface->format,
+                    surface->block.size, 
+                    surface->nblocksx,
+                    surface->nblocksy,
+                    surface->stride,
+                    data);
+   
+   pipe_surface_unmap(surface);
+error2:
+   surface->usage = surface_usage;
+error1:
+   ;
+}
+
+
+#pragma pack(push,2)
+struct bmp_file_header {
+   uint16_t bfType;
+   uint32_t bfSize;
+   uint16_t bfReserved1;
+   uint16_t bfReserved2;
+   uint32_t bfOffBits;
+};
+#pragma pack(pop)
+
+struct bmp_info_header {
+   uint32_t biSize;
+   int32_t biWidth;
+   int32_t biHeight;
+   uint16_t biPlanes;
+   uint16_t biBitCount;
+   uint32_t biCompression;
+   uint32_t biSizeImage;
+   int32_t biXPelsPerMeter;
+   int32_t biYPelsPerMeter;
+   uint32_t biClrUsed;
+   uint32_t biClrImportant;
+};
+
+struct bmp_rgb_quad {
+   uint8_t rgbBlue;
+   uint8_t rgbGreen;
+   uint8_t rgbRed;
+   uint8_t rgbAlpha;
+};
+
+void 
+debug_dump_surface_bmp(const char *filename,
+                       struct pipe_surface *surface)
+{
+#ifndef PIPE_SUBSYSTEM_WINDOWS_MINIPORT
+   struct util_stream *stream;
+   unsigned surface_usage;
+   struct bmp_file_header bmfh;
+   struct bmp_info_header bmih;
+   float *rgba;
+   unsigned x, y;
+
+   if (!surface)
+      goto error1;
+
+   rgba = MALLOC(surface->width*4*sizeof(float));
+   if(!rgba)
+      goto error1;
+   
+   bmfh.bfType = 0x4d42;
+   bmfh.bfSize = 14 + 40 + surface->height*surface->width*4;
+   bmfh.bfReserved1 = 0;
+   bmfh.bfReserved2 = 0;
+   bmfh.bfOffBits = 14 + 40;
+   
+   bmih.biSize = 40;
+   bmih.biWidth = surface->width;
+   bmih.biHeight = surface->height;
+   bmih.biPlanes = 1;
+   bmih.biBitCount = 32;
+   bmih.biCompression = 0;
+   bmih.biSizeImage = surface->height*surface->width*4;
+   bmih.biXPelsPerMeter = 0;
+   bmih.biYPelsPerMeter = 0;
+   bmih.biClrUsed = 0;
+   bmih.biClrImportant = 0;
+   
+   stream = util_stream_create(filename, bmfh.bfSize);
+   if(!stream)
+      goto error2;
+   
+   util_stream_write(stream, &bmfh, 14);
+   util_stream_write(stream, &bmih, 40);
+   
+   /* XXX: force mappable surface */
+   surface_usage = surface->usage;
+   surface->usage |= PIPE_BUFFER_USAGE_CPU_READ;
+
+   y = surface->height;
+   while(y--) {
+      pipe_get_tile_rgba(surface,
+                         0, y, surface->width, 1,
+                         rgba);
+      for(x = 0; x < surface->width; ++x)
+      {
+         struct bmp_rgb_quad pixel;
+         pixel.rgbRed   = float_to_ubyte(rgba[x*4 + 0]);
+         pixel.rgbGreen = float_to_ubyte(rgba[x*4 + 1]);
+         pixel.rgbBlue  = float_to_ubyte(rgba[x*4 + 2]);
+         pixel.rgbAlpha = float_to_ubyte(rgba[x*4 + 3]);
+         util_stream_write(stream, &pixel, 4);
+      }  
+   }
+   
+   surface->usage = surface_usage;
+
+   util_stream_close(stream);
+error2:
+   FREE(rgba);
+error1:
+   ;
+#endif
+}
+
+#endif
diff --git a/src/gallium/auxiliary/util/p_debug_mem.c b/src/gallium/auxiliary/util/p_debug_mem.c
new file mode 100644
index 00000000000..250fd60f634
--- /dev/null
+++ b/src/gallium/auxiliary/util/p_debug_mem.c
@@ -0,0 +1,311 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * @file
+ * Memory debugging.
+ * 
+ * @author José Fonseca <[email protected]>
+ */
+
+#include "pipe/p_config.h" 
+
+#if defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY)
+#include <windows.h>
+#include <winddi.h>
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT)
+#include <wdm.h>
+#else
+#include <stdio.h>
+#include <stdlib.h>
+#endif
+
+#include "pipe/p_debug.h" 
+#include "util/u_double_list.h" 
+
+
+#define DEBUG_MEMORY_MAGIC 0x6e34090aU 
+
+
+#if defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) && !defined(WINCE)
+#define real_malloc(_size) EngAllocMem(0, _size, 'D3AG')
+#define real_free(_ptr) EngFreeMem(_ptr)
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT)
+#define real_malloc(_size) ExAllocatePool(0, _size)
+#define real_free(_ptr) ExFreePool(_ptr)
+#else
+#define real_malloc(_size) malloc(_size)
+#define real_free(_ptr) free(_ptr)
+#endif
+
+
+struct debug_memory_header 
+{
+   struct list_head head;
+   
+   unsigned long no;
+   const char *file;
+   unsigned line;
+   const char *function;
+   size_t size;
+   unsigned magic;
+};
+
+struct debug_memory_footer
+{
+   unsigned magic;
+};
+
+
+static struct list_head list = { &list, &list };
+
+static unsigned long last_no = 0;
+
+
+static INLINE struct debug_memory_header *
+header_from_data(void *data)
+{
+   if(data)
+      return (struct debug_memory_header *)((char *)data - sizeof(struct debug_memory_header));
+   else
+      return NULL;
+}
+
+static INLINE void *
+data_from_header(struct debug_memory_header *hdr)
+{
+   if(hdr)
+      return (void *)((char *)hdr + sizeof(struct debug_memory_header));
+   else
+      return NULL;
+}
+
+static INLINE struct debug_memory_footer *
+footer_from_header(struct debug_memory_header *hdr)
+{
+   if(hdr)
+      return (struct debug_memory_footer *)((char *)hdr + sizeof(struct debug_memory_header) + hdr->size);
+   else
+      return NULL;
+}
+
+
+void *
+debug_malloc(const char *file, unsigned line, const char *function,
+             size_t size) 
+{
+   struct debug_memory_header *hdr;
+   struct debug_memory_footer *ftr;
+   
+   hdr = real_malloc(sizeof(*hdr) + size + sizeof(*ftr));
+   if(!hdr) {
+      debug_printf("%s:%u:%s: out of memory when trying to allocate %lu bytes\n",
+                   file, line, function,
+                   (long unsigned)size);
+      return NULL;
+   }
+ 
+   hdr->no = last_no++;
+   hdr->file = file;
+   hdr->line = line;
+   hdr->function = function;
+   hdr->size = size;
+   hdr->magic = DEBUG_MEMORY_MAGIC;
+
+   ftr = footer_from_header(hdr);
+   ftr->magic = DEBUG_MEMORY_MAGIC;
+   
+   LIST_ADDTAIL(&hdr->head, &list);
+   
+   return data_from_header(hdr);
+}
+
+void
+debug_free(const char *file, unsigned line, const char *function,
+           void *ptr) 
+{
+   struct debug_memory_header *hdr;
+   struct debug_memory_footer *ftr;
+   
+   if(!ptr)
+      return;
+   
+   hdr = header_from_data(ptr);
+   if(hdr->magic != DEBUG_MEMORY_MAGIC) {
+      debug_printf("%s:%u:%s: freeing bad or corrupted memory %p\n",
+                   file, line, function,
+                   ptr);
+      debug_assert(0);
+      return;
+   }
+
+   ftr = footer_from_header(hdr);
+   if(ftr->magic != DEBUG_MEMORY_MAGIC) {
+      debug_printf("%s:%u:%s: buffer overflow %p\n",
+                   hdr->file, hdr->line, hdr->function,
+                   ptr);
+      debug_assert(0);
+   }
+
+   LIST_DEL(&hdr->head);
+   hdr->magic = 0;
+   ftr->magic = 0;
+   
+   real_free(hdr);
+}
+
+void *
+debug_calloc(const char *file, unsigned line, const char *function,
+             size_t count, size_t size )
+{
+   void *ptr = debug_malloc( file, line, function, count * size );
+   if( ptr )
+      memset( ptr, 0, count * size );
+   return ptr;
+}
+
+void *
+debug_realloc(const char *file, unsigned line, const char *function,
+              void *old_ptr, size_t old_size, size_t new_size )
+{
+   struct debug_memory_header *old_hdr, *new_hdr;
+   struct debug_memory_footer *old_ftr, *new_ftr;
+   void *new_ptr;
+   
+   if(!old_ptr)
+      return debug_malloc( file, line, function, new_size );
+   
+   if(!new_size) {
+      debug_free( file, line, function, old_ptr );
+      return NULL;
+   }
+   
+   old_hdr = header_from_data(old_ptr);
+   if(old_hdr->magic != DEBUG_MEMORY_MAGIC) {
+      debug_printf("%s:%u:%s: reallocating bad or corrupted memory %p\n",
+                   file, line, function,
+                   old_ptr);
+      debug_assert(0);
+      return NULL;
+   }
+   
+   old_ftr = footer_from_header(old_hdr);
+   if(old_ftr->magic != DEBUG_MEMORY_MAGIC) {
+      debug_printf("%s:%u:%s: buffer overflow %p\n",
+                   old_hdr->file, old_hdr->line, old_hdr->function,
+                   old_ptr);
+      debug_assert(0);
+   }
+
+   /* alloc new */
+   new_hdr = real_malloc(sizeof(*new_hdr) + new_size + sizeof(*new_ftr));
+   if(!new_hdr) {
+      debug_printf("%s:%u:%s: out of memory when trying to allocate %lu bytes\n",
+                   file, line, function,
+                   (long unsigned)new_size);
+      return NULL;
+   }
+   new_hdr->no = old_hdr->no;
+   new_hdr->file = old_hdr->file;
+   new_hdr->line = old_hdr->line;
+   new_hdr->function = old_hdr->function;
+   new_hdr->size = new_size;
+   new_hdr->magic = DEBUG_MEMORY_MAGIC;
+   
+   new_ftr = footer_from_header(new_hdr);
+   new_ftr->magic = DEBUG_MEMORY_MAGIC;
+   
+   LIST_REPLACE(&old_hdr->head, &new_hdr->head);
+
+   /* copy data */
+   new_ptr = data_from_header(new_hdr);
+   memcpy( new_ptr, old_ptr, old_size < new_size ? old_size : new_size );
+
+   /* free old */
+   old_hdr->magic = 0;
+   old_ftr->magic = 0;
+   real_free(old_hdr);
+
+   return new_ptr;
+}
+
+unsigned long
+debug_memory_begin(void)
+{
+   return last_no;
+}
+
+void 
+debug_memory_end(unsigned long start_no)
+{
+   size_t total_size = 0;
+   struct list_head *entry;
+
+   if(start_no == last_no)
+      return;
+
+   entry = list.prev;
+   for (; entry != &list; entry = entry->prev) {
+      struct debug_memory_header *hdr;
+      void *ptr;
+      struct debug_memory_footer *ftr;
+
+      hdr = LIST_ENTRY(struct debug_memory_header, entry, head);
+      ptr = data_from_header(hdr);
+      ftr = footer_from_header(hdr);
+
+      if(hdr->magic != DEBUG_MEMORY_MAGIC) {
+         debug_printf("%s:%u:%s: bad or corrupted memory %p\n",
+                      hdr->file, hdr->line, hdr->function,
+                      ptr);
+         debug_assert(0);
+      }
+
+      if((start_no <= hdr->no && hdr->no < last_no) ||
+	 (last_no < start_no && (hdr->no < last_no || start_no <= hdr->no))) {
+	 debug_printf("%s:%u:%s: %u bytes at %p not freed\n",
+		      hdr->file, hdr->line, hdr->function,
+		      hdr->size, ptr);
+	 total_size += hdr->size;
+      }
+
+      if(ftr->magic != DEBUG_MEMORY_MAGIC) {
+         debug_printf("%s:%u:%s: buffer overflow %p\n",
+                      hdr->file, hdr->line, hdr->function,
+                      ptr);
+         debug_assert(0);
+      }
+   }
+
+   if(total_size) {
+      debug_printf("Total of %u KB of system memory apparently leaked\n",
+		   (total_size + 1023)/1024);
+   }
+   else {
+      debug_printf("No memory leaks detected.\n");
+   }
+}
diff --git a/src/gallium/auxiliary/util/p_debug_prof.c b/src/gallium/auxiliary/util/p_debug_prof.c
new file mode 100644
index 00000000000..5f9772ef917
--- /dev/null
+++ b/src/gallium/auxiliary/util/p_debug_prof.c
@@ -0,0 +1,320 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * @file
+ * Poor-man profiling.
+ * 
+ * @author José Fonseca <[email protected]>
+ * 
+ * @sa http://blogs.msdn.com/joshpoley/archive/2008/03/12/poor-man-s-profiler.aspx
+ * @sa http://www.johnpanzer.com/aci_cuj/index.html
+ */
+
+#include "pipe/p_config.h" 
+
+#if defined(PROFILE) && defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY)
+
+#include <windows.h>
+#include <winddi.h>
+
+#include "pipe/p_debug.h" 
+#include "util/u_string.h" 
+
+
+#define PROFILE_TABLE_SIZE (1024*1024)
+#define FILE_NAME_SIZE 256
+
+struct debug_profile_entry
+{
+   uintptr_t caller;
+   uintptr_t callee;
+   uint64_t samples;
+};
+
+static unsigned long enabled = 0;
+
+static WCHAR wFileName[FILE_NAME_SIZE] = L"\\??\\c:\\00000000.prof";
+static ULONG_PTR iFile = 0;
+
+static struct debug_profile_entry *table = NULL;
+static unsigned long free_table_entries = 0;
+static unsigned long max_table_entries = 0;
+
+uint64_t start_stamp = 0;
+uint64_t end_stamp = 0;
+
+
+static void
+debug_profile_entry(uintptr_t caller, uintptr_t callee, uint64_t samples)
+{
+   unsigned hash = ( caller + callee ) & PROFILE_TABLE_SIZE - 1;
+   
+   while(1) {
+      if(table[hash].caller == 0 && table[hash].callee == 0) {
+         table[hash].caller = caller;
+         table[hash].callee = callee;
+         table[hash].samples = samples;
+         --free_table_entries;
+         break;
+      }
+      else if(table[hash].caller == caller && table[hash].callee == callee) {
+         table[hash].samples += samples;
+         break;
+      }
+      else {
+         ++hash;
+      }
+   }
+}
+
+
+static uintptr_t caller_stack[1024];
+static unsigned last_caller = 0;
+
+
+static int64_t delta(void) {
+   int64_t result = end_stamp - start_stamp;
+   if(result > UINT64_C(0xffffffff))
+      result = 0;
+   return result;
+}
+
+
+static void __cdecl 
+debug_profile_enter(uintptr_t callee)
+{
+   uintptr_t caller = last_caller ? caller_stack[last_caller - 1] : 0;
+                
+   if (caller)
+      debug_profile_entry(caller, 0, delta());
+   debug_profile_entry(caller, callee, 1);
+   caller_stack[last_caller++] = callee;
+}
+
+
+static void __cdecl
+debug_profile_exit(uintptr_t callee)
+{
+   debug_profile_entry(callee, 0, delta());
+   if(last_caller)
+      --last_caller;
+}
+   
+   
+/**
+ * Called at the start of every method or function.
+ * 
+ * @sa http://msdn.microsoft.com/en-us/library/c63a9b7h.aspx
+ */
+void __declspec(naked) __cdecl 
+_penter(void) {
+   _asm {
+      push eax
+      mov eax, [enabled]
+      test eax, eax
+      jz skip
+
+      push edx
+      
+      rdtsc
+      mov dword ptr [end_stamp], eax
+      mov dword ptr [end_stamp+4], edx
+
+      xor eax, eax
+      mov [enabled], eax
+
+      mov eax, [esp+8]
+
+      push ebx
+      push ecx
+      push ebp
+      push edi
+      push esi
+
+      push eax
+      call debug_profile_enter
+      add esp, 4
+
+      pop esi
+      pop edi
+      pop ebp
+      pop ecx
+      pop ebx
+
+      mov eax, 1
+      mov [enabled], eax 
+
+      rdtsc
+      mov dword ptr [start_stamp], eax
+      mov dword ptr [start_stamp+4], edx
+      
+      pop edx
+skip:
+      pop eax
+      ret
+   }
+}
+
+
+/**
+ * Called at the end of Calls the end of every method or function.
+ * 
+ * @sa http://msdn.microsoft.com/en-us/library/xc11y76y.aspx
+ */
+void __declspec(naked) __cdecl 
+_pexit(void) {
+   _asm {
+      push eax
+      mov eax, [enabled]
+      test eax, eax
+      jz skip
+
+      push edx
+      
+      rdtsc
+      mov dword ptr [end_stamp], eax
+      mov dword ptr [end_stamp+4], edx
+
+      xor eax, eax
+      mov [enabled], eax
+
+      mov eax, [esp+8]
+
+      push ebx
+      push ecx
+      push ebp
+      push edi
+      push esi
+
+      push eax
+      call debug_profile_exit
+      add esp, 4
+
+      pop esi
+      pop edi
+      pop ebp
+      pop ecx
+      pop ebx
+
+      mov eax, 1
+      mov [enabled], eax 
+
+      rdtsc
+      mov dword ptr [start_stamp], eax
+      mov dword ptr [start_stamp+4], edx
+      
+      pop edx
+skip:
+      pop eax
+      ret
+   }
+}
+
+
+/**
+ * Reference function for calibration. 
+ */
+void __declspec(naked) 
+__debug_profile_reference(void) {
+   _asm {
+      call _penter
+      call _pexit
+      ret
+   }
+}
+
+
+void
+debug_profile_start(void)
+{
+   WCHAR *p;
+
+   // increment starting from the less significant digit
+   p = &wFileName[14];
+   while(1) {
+      if(*p == '9') {
+         *p-- = '0';
+      }
+      else {
+         *p += 1;
+         break;
+      }
+   }
+
+   table = EngMapFile(wFileName, 
+                      PROFILE_TABLE_SIZE*sizeof(struct debug_profile_entry), 
+                      &iFile);
+   if(table) {
+      unsigned i;
+      
+      free_table_entries = max_table_entries = PROFILE_TABLE_SIZE;
+      memset(table, 0, PROFILE_TABLE_SIZE*sizeof(struct debug_profile_entry));
+      
+      table[0].caller = (uintptr_t)&__debug_profile_reference;
+      table[0].callee = 0;
+      table[0].samples = 0;
+      --free_table_entries;
+
+      _asm {
+         push edx
+         push eax
+      
+         rdtsc
+         mov dword ptr [start_stamp], eax
+         mov dword ptr [start_stamp+4], edx
+         
+         pop edx
+         pop eax
+      }
+
+      last_caller = 0;
+      
+      enabled = 1;
+
+      for(i = 0; i < 8; ++i) {
+         _asm {
+            call __debug_profile_reference
+         }
+      }
+   }
+}
+
+
+void 
+debug_profile_stop(void)
+{
+   enabled = 0;
+
+   if(iFile)
+      EngUnmapFile(iFile);
+   iFile = 0;
+   table = NULL;
+   free_table_entries = max_table_entries = 0;
+}
+
+#endif /* PROFILE */
diff --git a/src/gallium/auxiliary/util/u_blit.c b/src/gallium/auxiliary/util/u_blit.c
new file mode 100644
index 00000000000..841e9c01e7e
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_blit.c
@@ -0,0 +1,555 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Copy/blit pixel rect between surfaces
+ *  
+ * @author Brian Paul
+ */
+
+
+#include "pipe/p_context.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "util/u_blit.h"
+#include "util/u_draw_quad.h"
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "util/u_simple_shaders.h"
+
+#include "cso_cache/cso_context.h"
+
+
+struct blit_state
+{
+   struct pipe_context *pipe;
+   struct cso_context *cso;
+
+   struct pipe_blend_state blend;
+   struct pipe_depth_stencil_alpha_state depthstencil;
+   struct pipe_rasterizer_state rasterizer;
+   struct pipe_sampler_state sampler;
+   struct pipe_viewport_state viewport;
+
+   struct pipe_shader_state vert_shader;
+   struct pipe_shader_state frag_shader;
+   void *vs;
+   void *fs;
+
+   struct pipe_buffer *vbuf;  /**< quad vertices */
+   unsigned vbuf_slot;
+
+   float vertices[4][2][4];   /**< vertex/texcoords for quad */
+};
+
+
+/**
+ * Create state object for blit.
+ * Intended to be created once and re-used for many blit() calls.
+ */
+struct blit_state *
+util_create_blit(struct pipe_context *pipe, struct cso_context *cso)
+{
+   struct blit_state *ctx;
+   uint i;
+
+   ctx = CALLOC_STRUCT(blit_state);
+   if (!ctx)
+      return NULL;
+
+   ctx->pipe = pipe;
+   ctx->cso = cso;
+
+   /* disabled blending/masking */
+   memset(&ctx->blend, 0, sizeof(ctx->blend));
+   ctx->blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+   ctx->blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+   ctx->blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
+   ctx->blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+   ctx->blend.colormask = PIPE_MASK_RGBA;
+
+   /* no-op depth/stencil/alpha */
+   memset(&ctx->depthstencil, 0, sizeof(ctx->depthstencil));
+
+   /* rasterizer */
+   memset(&ctx->rasterizer, 0, sizeof(ctx->rasterizer));
+   ctx->rasterizer.front_winding = PIPE_WINDING_CW;
+   ctx->rasterizer.cull_mode = PIPE_WINDING_NONE;
+   ctx->rasterizer.bypass_clipping = 1;
+   /*ctx->rasterizer.bypass_vs = 1;*/
+   ctx->rasterizer.gl_rasterization_rules = 1;
+
+   /* samplers */
+   memset(&ctx->sampler, 0, sizeof(ctx->sampler));
+   ctx->sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   ctx->sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   ctx->sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   ctx->sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NONE;
+   ctx->sampler.min_img_filter = 0; /* set later */
+   ctx->sampler.mag_img_filter = 0; /* set later */
+   ctx->sampler.normalized_coords = 1;
+
+   /* viewport (identity, we setup vertices in wincoords) */
+   ctx->viewport.scale[0] = 1.0;
+   ctx->viewport.scale[1] = 1.0;
+   ctx->viewport.scale[2] = 1.0;
+   ctx->viewport.scale[3] = 1.0;
+   ctx->viewport.translate[0] = 0.0;
+   ctx->viewport.translate[1] = 0.0;
+   ctx->viewport.translate[2] = 0.0;
+   ctx->viewport.translate[3] = 0.0;
+
+   /* vertex shader */
+   {
+      const uint semantic_names[] = { TGSI_SEMANTIC_POSITION,
+                                      TGSI_SEMANTIC_GENERIC };
+      const uint semantic_indexes[] = { 0, 0 };
+      ctx->vs = util_make_vertex_passthrough_shader(pipe, 2, semantic_names,
+                                                    semantic_indexes,
+                                                    &ctx->vert_shader);
+   }
+
+   /* fragment shader */
+   ctx->fs = util_make_fragment_tex_shader(pipe, &ctx->frag_shader);
+   ctx->vbuf = NULL;
+
+   /* init vertex data that doesn't change */
+   for (i = 0; i < 4; i++) {
+      ctx->vertices[i][0][3] = 1.0f; /* w */
+      ctx->vertices[i][1][2] = 0.0f; /* r */
+      ctx->vertices[i][1][3] = 1.0f; /* q */
+   }
+
+   return ctx;
+}
+
+
+/**
+ * Destroy a blit context
+ */
+void
+util_destroy_blit(struct blit_state *ctx)
+{
+   struct pipe_context *pipe = ctx->pipe;
+
+   pipe->delete_vs_state(pipe, ctx->vs);
+   pipe->delete_fs_state(pipe, ctx->fs);
+
+   FREE((void*) ctx->vert_shader.tokens);
+   FREE((void*) ctx->frag_shader.tokens);
+
+   pipe_buffer_reference(pipe->screen, &ctx->vbuf, NULL);
+
+   FREE(ctx);
+}
+
+
+static unsigned get_next_slot( struct blit_state *ctx )
+{
+   const unsigned max_slots = 4096 / sizeof ctx->vertices;
+
+   if (ctx->vbuf_slot >= max_slots) 
+      util_blit_flush( ctx );
+
+   if (!ctx->vbuf) {
+      ctx->vbuf = pipe_buffer_create(ctx->pipe->screen,
+                                     32,
+                                     PIPE_BUFFER_USAGE_VERTEX,
+                                     max_slots * sizeof ctx->vertices);
+   }
+   
+   return ctx->vbuf_slot++ * sizeof ctx->vertices;
+}
+                               
+
+
+/**
+ * Setup vertex data for the textured quad we'll draw.
+ * Note: y=0=top
+ */
+static unsigned
+setup_vertex_data(struct blit_state *ctx,
+                  float x0, float y0, float x1, float y1, float z)
+{
+   void *buf;
+   unsigned offset;
+
+   ctx->vertices[0][0][0] = x0;
+   ctx->vertices[0][0][1] = y0;
+   ctx->vertices[0][0][2] = z;
+   ctx->vertices[0][1][0] = 0.0f; /*s*/
+   ctx->vertices[0][1][1] = 0.0f; /*t*/
+
+   ctx->vertices[1][0][0] = x1;
+   ctx->vertices[1][0][1] = y0;
+   ctx->vertices[1][0][2] = z;
+   ctx->vertices[1][1][0] = 1.0f; /*s*/
+   ctx->vertices[1][1][1] = 0.0f; /*t*/
+
+   ctx->vertices[2][0][0] = x1;
+   ctx->vertices[2][0][1] = y1;
+   ctx->vertices[2][0][2] = z;
+   ctx->vertices[2][1][0] = 1.0f;
+   ctx->vertices[2][1][1] = 1.0f;
+
+   ctx->vertices[3][0][0] = x0;
+   ctx->vertices[3][0][1] = y1;
+   ctx->vertices[3][0][2] = z;
+   ctx->vertices[3][1][0] = 0.0f;
+   ctx->vertices[3][1][1] = 1.0f;
+
+   offset = get_next_slot( ctx );
+
+   buf = pipe_buffer_map(ctx->pipe->screen, ctx->vbuf,
+                         PIPE_BUFFER_USAGE_CPU_WRITE);
+
+   memcpy((char *)buf + offset, ctx->vertices, sizeof(ctx->vertices));
+
+   pipe_buffer_unmap(ctx->pipe->screen, ctx->vbuf);
+
+   return offset;
+}
+
+
+/**
+ * Setup vertex data for the textured quad we'll draw.
+ * Note: y=0=top
+ */
+static unsigned
+setup_vertex_data_tex(struct blit_state *ctx,
+                      float x0, float y0, float x1, float y1,
+                      float s0, float t0, float s1, float t1,
+                      float z)
+{
+   void *buf;
+   unsigned offset;
+
+   ctx->vertices[0][0][0] = x0;
+   ctx->vertices[0][0][1] = y0;
+   ctx->vertices[0][0][2] = z;
+   ctx->vertices[0][1][0] = s0; /*s*/
+   ctx->vertices[0][1][1] = t0; /*t*/
+
+   ctx->vertices[1][0][0] = x1;
+   ctx->vertices[1][0][1] = y0;
+   ctx->vertices[1][0][2] = z;
+   ctx->vertices[1][1][0] = s1; /*s*/
+   ctx->vertices[1][1][1] = t0; /*t*/
+
+   ctx->vertices[2][0][0] = x1;
+   ctx->vertices[2][0][1] = y1;
+   ctx->vertices[2][0][2] = z;
+   ctx->vertices[2][1][0] = s1;
+   ctx->vertices[2][1][1] = t1;
+
+   ctx->vertices[3][0][0] = x0;
+   ctx->vertices[3][0][1] = y1;
+   ctx->vertices[3][0][2] = z;
+   ctx->vertices[3][1][0] = s0;
+   ctx->vertices[3][1][1] = t1;
+
+   offset = get_next_slot( ctx );
+
+   buf = pipe_buffer_map(ctx->pipe->screen, ctx->vbuf,
+                         PIPE_BUFFER_USAGE_CPU_WRITE);
+
+   memcpy((char *)buf + offset, ctx->vertices, sizeof(ctx->vertices));
+
+   pipe_buffer_unmap(ctx->pipe->screen, ctx->vbuf);
+
+   return offset;
+}
+/**
+ * Copy pixel block from src surface to dst surface.
+ * Overlapping regions are acceptable.
+ * XXX need some control over blitting Z and/or stencil.
+ */
+void
+util_blit_pixels(struct blit_state *ctx,
+                 struct pipe_surface *src,
+                 int srcX0, int srcY0,
+                 int srcX1, int srcY1,
+                 struct pipe_surface *dst,
+                 int dstX0, int dstY0,
+                 int dstX1, int dstY1,
+                 float z, uint filter)
+{
+   struct pipe_context *pipe = ctx->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   struct pipe_texture texTemp, *tex;
+   struct pipe_surface *texSurf;
+   struct pipe_framebuffer_state fb;
+   const int srcW = abs(srcX1 - srcX0);
+   const int srcH = abs(srcY1 - srcY0);
+   const int srcLeft = MIN2(srcX0, srcX1);
+   const int srcTop = MIN2(srcY0, srcY1);
+   unsigned offset;
+
+   assert(filter == PIPE_TEX_MIPFILTER_NEAREST ||
+          filter == PIPE_TEX_MIPFILTER_LINEAR);
+
+   if (srcLeft != srcX0) {
+      /* left-right flip */
+      int tmp = dstX0;
+      dstX0 = dstX1;
+      dstX1 = tmp;
+   }
+
+   if (srcTop != srcY0) {
+      /* up-down flip */
+      int tmp = dstY0;
+      dstY0 = dstY1;
+      dstY1 = tmp;
+   }
+
+   assert(screen->is_format_supported(screen, src->format, PIPE_TEXTURE_2D,
+                                      PIPE_TEXTURE_USAGE_SAMPLER, 0));
+   assert(screen->is_format_supported(screen, dst->format, PIPE_TEXTURE_2D,
+                                      PIPE_TEXTURE_USAGE_SAMPLER, 0));
+
+   if(dst->format == src->format && (dstX1 - dstX0) == srcW && (dstY1 - dstY0) == srcH) {
+      /* FIXME: this will most surely fail for overlapping rectangles */
+      pipe->surface_copy(pipe, FALSE,
+			 dst, dstX0, dstY0,   /* dest */
+			 src, srcX0, srcY0, /* src */
+			 srcW, srcH);     /* size */
+      return;
+   }
+   
+   assert(screen->is_format_supported(screen, dst->format, PIPE_TEXTURE_2D,
+                                      PIPE_TEXTURE_USAGE_RENDER_TARGET, 0));
+
+   /*
+    * XXX for now we're always creating a temporary texture.
+    * Strictly speaking that's not always needed.
+    */
+
+   /* create temp texture */
+   memset(&texTemp, 0, sizeof(texTemp));
+   texTemp.target = PIPE_TEXTURE_2D;
+   texTemp.format = src->format;
+   texTemp.last_level = 0;
+   texTemp.width[0] = srcW;
+   texTemp.height[0] = srcH;
+   texTemp.depth[0] = 1;
+   texTemp.compressed = 0;
+   pf_get_block(src->format, &texTemp.block);
+
+   tex = screen->texture_create(screen, &texTemp);
+   if (!tex)
+      return;
+
+   texSurf = screen->get_tex_surface(screen, tex, 0, 0, 0, 
+                                     PIPE_BUFFER_USAGE_GPU_WRITE);
+
+   /* load temp texture */
+   pipe->surface_copy(pipe, FALSE,
+                      texSurf, 0, 0,   /* dest */
+                      src, srcLeft, srcTop, /* src */
+                      srcW, srcH);     /* size */
+
+   /* free the surface, update the texture if necessary.
+    */
+   screen->tex_surface_release(screen, &texSurf);
+
+   /* save state (restored below) */
+   cso_save_blend(ctx->cso);
+   cso_save_depth_stencil_alpha(ctx->cso);
+   cso_save_rasterizer(ctx->cso);
+   cso_save_samplers(ctx->cso);
+   cso_save_sampler_textures(ctx->cso);
+   cso_save_framebuffer(ctx->cso);
+   cso_save_fragment_shader(ctx->cso);
+   cso_save_vertex_shader(ctx->cso);
+   cso_save_viewport(ctx->cso);
+
+   /* set misc state we care about */
+   cso_set_blend(ctx->cso, &ctx->blend);
+   cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil);
+   cso_set_rasterizer(ctx->cso, &ctx->rasterizer);
+   cso_set_viewport(ctx->cso, &ctx->viewport);
+
+   /* sampler */
+   ctx->sampler.min_img_filter = filter;
+   ctx->sampler.mag_img_filter = filter;
+   cso_single_sampler(ctx->cso, 0, &ctx->sampler);
+   cso_single_sampler_done(ctx->cso);
+
+   /* texture */
+   cso_set_sampler_textures(ctx->cso, 1, &tex);
+
+   /* shaders */
+   cso_set_fragment_shader_handle(ctx->cso, ctx->fs);
+   cso_set_vertex_shader_handle(ctx->cso, ctx->vs);
+
+   /* drawing dest */
+   memset(&fb, 0, sizeof(fb));
+   fb.width = dst->width;
+   fb.height = dst->height;
+   fb.nr_cbufs = 1;
+   fb.cbufs[0] = dst;
+   cso_set_framebuffer(ctx->cso, &fb);
+
+   /* draw quad */
+   offset = setup_vertex_data(ctx,
+                              (float) dstX0, (float) dstY0, 
+                              (float) dstX1, (float) dstY1, z);
+
+   util_draw_vertex_buffer(ctx->pipe, ctx->vbuf, offset,
+                           PIPE_PRIM_TRIANGLE_FAN,
+                           4,  /* verts */
+                           2); /* attribs/vert */
+
+   /* restore state we changed */
+   cso_restore_blend(ctx->cso);
+   cso_restore_depth_stencil_alpha(ctx->cso);
+   cso_restore_rasterizer(ctx->cso);
+   cso_restore_samplers(ctx->cso);
+   cso_restore_sampler_textures(ctx->cso);
+   cso_restore_framebuffer(ctx->cso);
+   cso_restore_fragment_shader(ctx->cso);
+   cso_restore_vertex_shader(ctx->cso);
+   cso_restore_viewport(ctx->cso);
+
+   screen->texture_release(screen, &tex);
+}
+
+
+/* Release vertex buffer at end of frame to avoid synchronous
+ * rendering.
+ */
+void util_blit_flush( struct blit_state *ctx )
+{
+   pipe_buffer_reference(ctx->pipe->screen, &ctx->vbuf, NULL);
+   ctx->vbuf_slot = 0;
+} 
+
+
+
+/**
+ * Copy pixel block from src texture to dst surface.
+ * Overlapping regions are acceptable.
+ *
+ * XXX Should support selection of level.
+ * XXX need some control over blitting Z and/or stencil.
+ */
+void
+util_blit_pixels_tex(struct blit_state *ctx,
+                 struct pipe_texture *tex,
+                 int srcX0, int srcY0,
+                 int srcX1, int srcY1,
+                 struct pipe_surface *dst,
+                 int dstX0, int dstY0,
+                 int dstX1, int dstY1,
+                 float z, uint filter)
+{
+   struct pipe_context *pipe = ctx->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   struct pipe_framebuffer_state fb;
+   float s0, t0, s1, t1;
+   unsigned offset;
+
+   assert(filter == PIPE_TEX_MIPFILTER_NEAREST ||
+          filter == PIPE_TEX_MIPFILTER_LINEAR);
+
+   assert(tex->width[0] != 0);
+   assert(tex->height[0] != 0);
+
+   s0 = srcX0 / (float)tex->width[0];
+   s1 = srcX1 / (float)tex->width[0];
+   t0 = srcY0 / (float)tex->height[0];
+   t1 = srcY1 / (float)tex->height[0];
+
+   assert(screen->is_format_supported(screen, dst->format, PIPE_TEXTURE_2D,
+                                      PIPE_TEXTURE_USAGE_RENDER_TARGET, 0));
+
+   /* save state (restored below) */
+   cso_save_blend(ctx->cso);
+   cso_save_depth_stencil_alpha(ctx->cso);
+   cso_save_rasterizer(ctx->cso);
+   cso_save_samplers(ctx->cso);
+   cso_save_sampler_textures(ctx->cso);
+   cso_save_framebuffer(ctx->cso);
+   cso_save_fragment_shader(ctx->cso);
+   cso_save_vertex_shader(ctx->cso);
+   cso_save_viewport(ctx->cso);
+
+   /* set misc state we care about */
+   cso_set_blend(ctx->cso, &ctx->blend);
+   cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil);
+   cso_set_rasterizer(ctx->cso, &ctx->rasterizer);
+   cso_set_viewport(ctx->cso, &ctx->viewport);
+
+   /* sampler */
+   ctx->sampler.min_img_filter = filter;
+   ctx->sampler.mag_img_filter = filter;
+   cso_single_sampler(ctx->cso, 0, &ctx->sampler);
+   cso_single_sampler_done(ctx->cso);
+
+   /* texture */
+   cso_set_sampler_textures(ctx->cso, 1, &tex);
+
+   /* shaders */
+   cso_set_fragment_shader_handle(ctx->cso, ctx->fs);
+   cso_set_vertex_shader_handle(ctx->cso, ctx->vs);
+
+   /* drawing dest */
+   memset(&fb, 0, sizeof(fb));
+   fb.width = dst->width;
+   fb.height = dst->height;
+   fb.nr_cbufs = 1;
+   fb.cbufs[0] = dst;
+   cso_set_framebuffer(ctx->cso, &fb);
+
+   /* draw quad */
+   offset = setup_vertex_data_tex(ctx,
+                                  (float) dstX0, (float) dstY0,
+                                  (float) dstX1, (float) dstY1,
+                                  s0, t0, s1, t1,
+                                  z);
+
+   util_draw_vertex_buffer(ctx->pipe, 
+                           ctx->vbuf, offset,
+                           PIPE_PRIM_TRIANGLE_FAN,
+                           4,  /* verts */
+                           2); /* attribs/vert */
+
+   /* restore state we changed */
+   cso_restore_blend(ctx->cso);
+   cso_restore_depth_stencil_alpha(ctx->cso);
+   cso_restore_rasterizer(ctx->cso);
+   cso_restore_samplers(ctx->cso);
+   cso_restore_sampler_textures(ctx->cso);
+   cso_restore_framebuffer(ctx->cso);
+   cso_restore_fragment_shader(ctx->cso);
+   cso_restore_vertex_shader(ctx->cso);
+   cso_restore_viewport(ctx->cso);
+}
diff --git a/src/gallium/auxiliary/util/u_blit.h b/src/gallium/auxiliary/util/u_blit.h
new file mode 100644
index 00000000000..c35beceda8d
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_blit.h
@@ -0,0 +1,82 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+
+#ifndef U_BLIT_H
+#define U_BLIT_H
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+   
+struct pipe_context;
+struct pipe_surface;
+struct pipe_texture;
+struct cso_context;
+
+
+struct blit_state;
+
+
+extern struct blit_state *
+util_create_blit(struct pipe_context *pipe, struct cso_context *cso);
+
+extern void
+util_destroy_blit(struct blit_state *ctx);
+
+extern void
+util_blit_pixels(struct blit_state *ctx,
+                 struct pipe_surface *src,
+                 int srcX0, int srcY0,
+                 int srcX1, int srcY1,
+                 struct pipe_surface *dst,
+                 int dstX0, int dstY0,
+                 int dstX1, int dstY1,
+                 float z, uint filter);
+
+extern void
+util_blit_pixels_tex(struct blit_state *ctx,
+                     struct pipe_texture *tex,
+                     int srcX0, int srcY0,
+                     int srcX1, int srcY1,
+                     struct pipe_surface *dst,
+                     int dstX0, int dstY0,
+                     int dstX1, int dstY1,
+                     float z, uint filter);
+
+/* Call at end of frame to avoid synchronous rendering.
+ */
+extern void
+util_blit_flush( struct blit_state *ctx );
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_cache.c b/src/gallium/auxiliary/util/u_cache.c
new file mode 100644
index 00000000000..0a1a64259f1
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_cache.c
@@ -0,0 +1,209 @@
+/**************************************************************************
+ *
+ * Copyright 2008 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Simple cache implementation.
+ *
+ * We simply have fixed size array, and destroy previous values on collision. 
+ * 
+ * @author Jose Fonseca <[email protected]>
+ */
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "util/u_cache.h"
+
+
+struct util_cache_entry
+{
+   void *key;
+   void *value;
+   
+#ifdef DEBUG
+   unsigned count;
+#endif
+};
+
+
+struct util_cache
+{
+   /** Hash function */
+   uint32_t (*hash)(const void *key);
+   
+   /** Compare two keys */
+   int (*compare)(const void *key1, const void *key2);
+
+   /** Destroy a (key, value) pair */
+   void (*destroy)(void *key, void *value);
+
+   uint32_t size;
+   
+   struct util_cache_entry *entries;
+   
+#ifdef DEBUG
+   unsigned count;
+#endif
+};
+
+
+struct util_cache *
+util_cache_create(uint32_t (*hash)(const void *key),
+                  int (*compare)(const void *key1, const void *key2),
+                  void (*destroy)(void *key, void *value),
+                  uint32_t size)
+{
+   struct util_cache *cache;
+   
+   cache = CALLOC_STRUCT(util_cache);
+   if(!cache)
+      return NULL;
+   
+   cache->hash = hash;
+   cache->compare = compare;
+   cache->destroy = destroy;
+   cache->size = size;
+   
+   cache->entries = CALLOC(size, sizeof(struct util_cache_entry));
+   if(!cache->entries) {
+      FREE(cache);
+      return NULL;
+   }
+   
+   return cache;
+}
+
+
+static INLINE struct util_cache_entry *
+util_cache_entry_get(struct util_cache *cache,
+                     const void *key)
+{
+   uint32_t hash;
+   
+   hash = cache->hash(key);
+
+   return &cache->entries[hash % cache->size];
+}
+
+static INLINE void
+util_cache_entry_destroy(struct util_cache *cache,
+                         struct util_cache_entry *entry)
+{
+   void *key = entry->key;
+   void *value = entry->value;
+   
+   entry->key = NULL;
+   entry->value = NULL;
+   
+   if(key || value)
+      if(cache->destroy)
+         cache->destroy(key, value);
+}
+
+
+void
+util_cache_set(struct util_cache *cache,
+               void *key,
+               void *value)
+{
+   struct util_cache_entry *entry;
+
+   assert(cache);
+
+   entry = util_cache_entry_get(cache, key);
+   util_cache_entry_destroy(cache, entry);
+   
+#ifdef DEBUG
+   ++entry->count;
+   ++cache->count;
+#endif
+   
+   entry->key = key;
+   entry->value = value;
+}
+
+
+void *
+util_cache_get(struct util_cache *cache, 
+               const void *key)
+{
+   struct util_cache_entry *entry;
+
+   assert(cache);
+
+   entry = util_cache_entry_get(cache, key);
+   if(!entry->key && !entry->value)
+      return NULL;
+   
+   if(cache->compare(key, entry->key) != 0)
+      return NULL;
+   
+   return entry->value;
+}
+
+
+void 
+util_cache_clear(struct util_cache *cache)
+{
+   uint32_t i;
+
+   assert(cache);
+   
+   for(i = 0; i < cache->size; ++i)
+      util_cache_entry_destroy(cache, &cache->entries[i]);
+}
+
+
+void
+util_cache_destroy(struct util_cache *cache)
+{
+   assert(cache);
+
+#ifdef DEBUG
+   if(cache->count >= 20*cache->size) {
+      /* Normal approximation of the Poisson distribution */
+      double mean = (double)cache->count/(double)cache->size;
+      double stddev = sqrt(mean);
+      unsigned i;
+      for(i = 0; i < cache->size; ++i) {
+         double z = fabs(cache->count - mean)/stddev;
+         /* This assert should not fail 99.9999998027% of the times, unless 
+          * the hash function is a poor one */
+         assert(z <= 6.0);
+      }
+   }
+#endif
+      
+   util_cache_clear(cache);
+   
+   FREE(cache->entries);
+   FREE(cache);
+}
diff --git a/src/gallium/auxiliary/util/u_cache.h b/src/gallium/auxiliary/util/u_cache.h
new file mode 100644
index 00000000000..8a612c6585f
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_cache.h
@@ -0,0 +1,87 @@
+/**************************************************************************
+ *
+ * Copyright 2008 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Simple cache.
+ *
+ * @author Jose Fonseca <[email protected]>
+ */
+
+#ifndef U_CACHE_H_
+#define U_CACHE_H_
+
+
+#include "pipe/p_compiler.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+   
+/**
+ * Least Recently Used (LRU) cache.
+ */
+struct util_cache;
+
+
+/**
+ * Create a cache.
+ * 
+ * @param hash hash function
+ * @param compare should return 0 for two equal keys
+ * @param destroy destruction callback (optional)
+ * @param size maximum number of entries
+ */
+struct util_cache *
+util_cache_create(uint32_t (*hash)(const void *key),
+                  int (*compare)(const void *key1, const void *key2),
+                  void (*destroy)(void *key, void *value),
+                  uint32_t size);
+
+void
+util_cache_set(struct util_cache *cache,
+               void *key,
+               void *value);
+
+void *
+util_cache_get(struct util_cache *cache, 
+               const void *key);
+
+void
+util_cache_clear(struct util_cache *cache);
+
+void
+util_cache_destroy(struct util_cache *cache);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* U_CACHE_H_ */
diff --git a/src/gallium/auxiliary/util/u_cpu_detect.c b/src/gallium/auxiliary/util/u_cpu_detect.c
new file mode 100644
index 00000000000..d9f2f8fc288
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_cpu_detect.c
@@ -0,0 +1,506 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Dennis Smit
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * AUTHORS, COPYRIGHT HOLDERS, AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/*
+ * Based on the work of Eric Anholt <[email protected]>
+ */
+
+/* FIXME: clean this entire file up */
+
+#include "u_cpu_detect.h"
+
+#ifdef __linux__
+#define OS_LINUX
+#endif
+#ifdef WIN32
+#define OS_WIN32
+#endif
+
+#if defined(ARCH_POWERPC)
+#if defined(OS_DARWIN)
+#include <sys/sysctl.h>
+#else
+#include <signal.h>
+#include <setjmp.h>
+#endif
+#endif
+
+#if defined(OS_NETBSD) || defined(OS_OPENBSD)
+#include <sys/param.h>
+#include <sys/sysctl.h>
+#include <machine/cpu.h>
+#endif
+
+#if defined(OS_FREEBSD)
+#include <sys/types.h>
+#include <sys/sysctl.h>
+#endif
+
+#if defined(OS_LINUX)
+#include <signal.h>
+#endif
+
+#if defined(OS_WIN32)
+#include <windows.h>
+#endif
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+#include <string.h>
+
+
+static struct cpu_detect_caps __cpu_detect_caps;
+static int __cpu_detect_initialized = 0;
+
+static int has_cpuid(void);
+static int cpuid(unsigned int ax, unsigned int *p);
+
+/* The sigill handlers */
+#if defined(ARCH_X86) /* x86 (linux katmai handler check thing) */
+#if defined(OS_LINUX) && defined(_POSIX_SOURCE) && defined(X86_FXSR_MAGIC)
+static void sigill_handler_sse(int signal, struct sigcontext sc)
+{
+	/* Both the "xorps %%xmm0,%%xmm0" and "divps %xmm0,%%xmm1"
+	 * instructions are 3 bytes long.  We must increment the instruction
+	 * pointer manually to avoid repeated execution of the offending
+	 * instruction.
+	 *
+	 * If the SIGILL is caused by a divide-by-zero when unmasked
+	 * exceptions aren't supported, the SIMD FPU status and control
+	 * word will be restored at the end of the test, so we don't need
+	 * to worry about doing it here.  Besides, we may not be able to...
+	 */
+	sc.eip += 3;
+
+	__cpu_detect_caps.hasSSE=0;
+}
+
+static void sigfpe_handler_sse(int signal, struct sigcontext sc)
+{
+	if (sc.fpstate->magic != 0xffff) {
+		/* Our signal context has the extended FPU state, so reset the
+		 * divide-by-zero exception mask and clear the divide-by-zero
+		 * exception bit.
+		 */
+		sc.fpstate->mxcsr |= 0x00000200;
+		sc.fpstate->mxcsr &= 0xfffffffb;
+	} else {
+		/* If we ever get here, we're completely hosed.
+		*/
+	}
+}
+#endif
+#endif /* OS_LINUX && _POSIX_SOURCE && X86_FXSR_MAGIC */
+
+#if defined(OS_WIN32)
+LONG CALLBACK win32_sig_handler_sse(EXCEPTION_POINTERS* ep)
+{
+	if(ep->ExceptionRecord->ExceptionCode==EXCEPTION_ILLEGAL_INSTRUCTION){
+		ep->ContextRecord->Eip +=3;
+		__cpu_detect_caps.hasSSE=0;
+		return EXCEPTION_CONTINUE_EXECUTION;
+	}
+	return EXCEPTION_CONTINUE_SEARCH;
+}
+#endif /* OS_WIN32 */
+
+
+#if defined(ARCH_POWERPC) && !defined(OS_DARWIN)
+static sigjmp_buf __lv_powerpc_jmpbuf;
+static volatile sig_atomic_t __lv_powerpc_canjump = 0;
+
+static void sigill_handler (int sig);
+
+static void sigill_handler (int sig)
+{
+	if (!__lv_powerpc_canjump) {
+		signal (sig, SIG_DFL);
+		raise (sig);
+	}
+
+	__lv_powerpc_canjump = 0;
+	siglongjmp(__lv_powerpc_jmpbuf, 1);
+}
+
+static void check_os_altivec_support(void)
+{
+#if defined(OS_DARWIN)
+	int sels[2] = {CTL_HW, HW_VECTORUNIT};
+	int has_vu = 0;
+	int len = sizeof (has_vu);
+	int err;
+
+	err = sysctl(sels, 2, &has_vu, &len, NULL, 0);
+
+	if (err == 0) {
+		if (has_vu != 0) {
+			__cpu_detect_caps.hasAltiVec = 1;
+		}
+	}
+#else /* !OS_DARWIN */
+	/* no Darwin, do it the brute-force way */
+	/* this is borrowed from the libmpeg2 library */
+	signal(SIGILL, sigill_handler);
+	if (sigsetjmp(__lv_powerpc_jmpbuf, 1)) {
+		signal(SIGILL, SIG_DFL);
+	} else {
+		__lv_powerpc_canjump = 1;
+
+		__asm __volatile
+			("mtspr 256, %0\n\t"
+			 "vand %%v0, %%v0, %%v0"
+			 :
+			 : "r" (-1));
+
+		signal(SIGILL, SIG_DFL);
+		__cpu_detect_caps.hasAltiVec = 1;
+	}
+#endif
+}
+#endif
+
+/* If we're running on a processor that can do SSE, let's see if we
+ * are allowed to or not.  This will catch 2.4.0 or later kernels that
+ * haven't been configured for a Pentium III but are running on one,
+ * and RedHat patched 2.2 kernels that have broken exception handling
+ * support for user space apps that do SSE.
+ */
+static void check_os_katmai_support(void)
+{
+#if defined(ARCH_X86)
+#if defined(OS_FREEBSD)
+	int has_sse=0, ret;
+	int len = sizeof (has_sse);
+
+	ret = sysctlbyname("hw.instruction_sse", &has_sse, &len, NULL, 0);
+	if (ret || !has_sse)
+		__cpu_detect_caps.hasSSE=0;
+
+#elif defined(OS_NETBSD) || defined(OS_OPENBSD)
+	int has_sse, has_sse2, ret, mib[2];
+	int varlen;
+
+	mib[0] = CTL_MACHDEP;
+	mib[1] = CPU_SSE;
+	varlen = sizeof (has_sse);
+
+	ret = sysctl(mib, 2, &has_sse, &varlen, NULL, 0);
+	if (ret < 0 || !has_sse) {
+		__cpu_detect_caps.hasSSE = 0;
+	} else {
+		__cpu_detect_caps.hasSSE = 1;
+	}
+
+	mib[1] = CPU_SSE2;
+	varlen = sizeof (has_sse2);
+	ret = sysctl(mib, 2, &has_sse2, &varlen, NULL, 0);
+	if (ret < 0 || !has_sse2) {
+		__cpu_detect_caps.hasSSE2 = 0;
+	} else {
+		__cpu_detect_caps.hasSSE2 = 1;
+	}
+	__cpu_detect_caps.hasSSE = 0; /* FIXME ?!?!? */
+
+#elif defined(OS_WIN32)
+	LPTOP_LEVEL_EXCEPTION_FILTER exc_fil;
+	if (__cpu_detect_caps.hasSSE) {
+		exc_fil = SetUnhandledExceptionFilter(win32_sig_handler_sse);
+		__asm __volatile ("xorps %xmm0, %xmm0");
+		SetUnhandledExceptionFilter(exc_fil);
+	}
+#elif defined(OS_LINUX)
+	struct sigaction saved_sigill;
+	struct sigaction saved_sigfpe;
+
+	/* Save the original signal handlers.
+	*/
+	sigaction(SIGILL, NULL, &saved_sigill);
+	sigaction(SIGFPE, NULL, &saved_sigfpe);
+
+	signal(SIGILL, (void (*)(int))sigill_handler_sse);
+	signal(SIGFPE, (void (*)(int))sigfpe_handler_sse);
+
+	/* Emulate test for OSFXSR in CR4.  The OS will set this bit if it
+	 * supports the extended FPU save and restore required for SSE.  If
+	 * we execute an SSE instruction on a PIII and get a SIGILL, the OS
+	 * doesn't support Streaming SIMD Exceptions, even if the processor
+	 * does.
+	 */
+	if (__cpu_detect_caps.hasSSE) {
+		__asm __volatile ("xorps %xmm1, %xmm0");
+	}
+
+	/* Emulate test for OSXMMEXCPT in CR4.  The OS will set this bit if
+	 * it supports unmasked SIMD FPU exceptions.  If we unmask the
+	 * exceptions, do a SIMD divide-by-zero and get a SIGILL, the OS
+	 * doesn't support unmasked SIMD FPU exceptions.  If we get a SIGFPE
+	 * as expected, we're okay but we need to clean up after it.
+	 *
+	 * Are we being too stringent in our requirement that the OS support
+	 * unmasked exceptions?  Certain RedHat 2.2 kernels enable SSE by
+	 * setting CR4.OSFXSR but don't support unmasked exceptions.  Win98
+	 * doesn't even support them.  We at least know the user-space SSE
+	 * support is good in kernels that do support unmasked exceptions,
+	 * and therefore to be safe I'm going to leave this test in here.
+	 */
+	if (__cpu_detect_caps.hasSSE) {
+		//      test_os_katmai_exception_support();
+	}
+
+	/* Restore the original signal handlers.
+	*/
+	sigaction(SIGILL, &saved_sigill, NULL);
+	sigaction(SIGFPE, &saved_sigfpe, NULL);
+
+#else
+	/* We can't use POSIX signal handling to test the availability of
+	 * SSE, so we disable it by default.
+	 */
+	__cpu_detect_caps.hasSSE = 0;
+#endif /* __linux__ */
+#endif
+}
+
+
+static int has_cpuid(void)
+{
+#if defined(ARCH_X86)
+	int a, c;
+
+	__asm __volatile
+		("pushf\n"
+		 "popl %0\n"
+		 "movl %0, %1\n"
+		 "xorl $0x200000, %0\n"
+		 "push %0\n"
+		 "popf\n"
+		 "pushf\n"
+		 "popl %0\n"
+		 : "=a" (a), "=c" (c)
+		 :
+		 : "cc");
+
+	return a != c;
+#else
+	return 0;
+#endif
+}
+
+static int cpuid(unsigned int ax, unsigned int *p)
+{
+#if defined(ARCH_X86)
+	unsigned int flags;
+
+	__asm __volatile
+		("movl %%ebx, %%esi\n\t"
+		 "cpuid\n\t"
+		 "xchgl %%ebx, %%esi"
+		 : "=a" (p[0]), "=S" (p[1]),
+		 "=c" (p[2]), "=d" (p[3])
+		 : "0" (ax));
+
+	return 0;
+#else
+	return -1;
+#endif
+}
+
+void cpu_detect_initialize()
+{
+	unsigned int regs[4];
+	unsigned int regs2[4];
+
+	int mib[2], ncpu;
+	int len;
+
+	memset(&__cpu_detect_caps, 0, sizeof (struct cpu_detect_caps));
+
+	/* Check for arch type */
+#if defined(ARCH_MIPS)
+	__cpu_detect_caps.type = CPU_DETECT_TYPE_MIPS;
+#elif defined(ARCH_ALPHA)
+	__cpu_detect_caps.type = CPU_DETECT_TYPE_ALPHA;
+#elif defined(ARCH_SPARC)
+	__cpu_detect_caps.type = CPU_DETECT_TYPE_SPARC;
+#elif defined(ARCH_X86)
+	__cpu_detect_caps.type = CPU_DETECT_TYPE_X86;
+#elif defined(ARCH_POWERPC)
+	__cpu_detect_caps.type = CPU_DETECT_TYPE_POWERPC;
+#else
+	__cpu_detect_caps.type = CPU_DETECT_TYPE_OTHER;
+#endif
+
+	/* Count the number of CPUs in system */
+#if !defined(OS_WIN32) && !defined(OS_UNKNOWN) && defined(_SC_NPROCESSORS_ONLN)
+	__cpu_detect_caps.nrcpu = sysconf(_SC_NPROCESSORS_ONLN);
+	if (__cpu_detect_caps.nrcpu == -1)
+		__cpu_detect_caps.nrcpu = 1;
+
+#elif defined(OS_NETBSD) || defined(OS_FREEBSD) || defined(OS_OPENBSD)
+
+	mib[0] = CTL_HW;
+	mib[1] = HW_NCPU;
+
+	len = sizeof (ncpu);
+	sysctl(mib, 2, &ncpu, &len, NULL, 0);
+	__cpu_detect_caps.nrcpu = ncpu;
+
+#else
+	__cpu_detect_caps.nrcpu = 1;
+#endif
+
+#if defined(ARCH_X86)
+	/* No cpuid, old 486 or lower */
+	if (has_cpuid() == 0)
+		return;
+
+	__cpu_detect_caps.cacheline = 32;
+
+	/* Get max cpuid level */
+	cpuid(0x00000000, regs);
+
+	if (regs[0] >= 0x00000001) {
+		unsigned int cacheline;
+
+		cpuid (0x00000001, regs2);
+
+		__cpu_detect_caps.x86cpuType = (regs2[0] >> 8) & 0xf;
+		if (__cpu_detect_caps.x86cpuType == 0xf)
+		    __cpu_detect_caps.x86cpuType = 8 + ((regs2[0] >> 20) & 255); /* use extended family (P4, IA64) */
+
+		/* general feature flags */
+		__cpu_detect_caps.hasTSC  = (regs2[3] & (1 << 8  )) >>  8; /* 0x0000010 */
+		__cpu_detect_caps.hasMMX  = (regs2[3] & (1 << 23 )) >> 23; /* 0x0800000 */
+		__cpu_detect_caps.hasSSE  = (regs2[3] & (1 << 25 )) >> 25; /* 0x2000000 */
+		__cpu_detect_caps.hasSSE2 = (regs2[3] & (1 << 26 )) >> 26; /* 0x4000000 */
+		__cpu_detect_caps.hasSSE3 = (regs2[2] & (1));	       /* 0x0000001 */
+		__cpu_detect_caps.hasSSSE3 = (regs2[2] & (1 << 9 )) >> 9;   /* 0x0000020 */
+		__cpu_detect_caps.hasMMX2 = __cpu_detect_caps.hasSSE; /* SSE cpus supports mmxext too */
+
+		cacheline = ((regs2[1] >> 8) & 0xFF) * 8;
+		if (cacheline > 0)
+			__cpu_detect_caps.cacheline = cacheline;
+	}
+
+	cpuid(0x80000000, regs);
+
+	if (regs[0] >= 0x80000001) {
+
+		cpuid(0x80000001, regs2);
+
+		__cpu_detect_caps.hasMMX  |= (regs2[3] & (1 << 23 )) >> 23; /* 0x0800000 */
+		__cpu_detect_caps.hasMMX2 |= (regs2[3] & (1 << 22 )) >> 22; /* 0x400000 */
+		__cpu_detect_caps.has3DNow    = (regs2[3] & (1 << 31 )) >> 31; /* 0x80000000 */
+		__cpu_detect_caps.has3DNowExt = (regs2[3] & (1 << 30 )) >> 30;
+	}
+
+	if (regs[0] >= 0x80000006) {
+		cpuid(0x80000006, regs2);
+		__cpu_detect_caps.cacheline = regs2[2] & 0xFF;
+	}
+
+
+#if defined(OS_LINUX) || defined(OS_FREEBSD) || defined(OS_NETBSD) || defined(OS_CYGWIN) || defined(OS_OPENBSD)
+	if (__cpu_detect_caps.hasSSE)
+		check_os_katmai_support();
+
+	if (!__cpu_detect_caps.hasSSE) {
+		__cpu_detect_caps.hasSSE2 = 0;
+		__cpu_detect_caps.hasSSE3 = 0;
+		__cpu_detect_caps.hasSSSE3 = 0;
+	}
+#else
+	__cpu_detect_caps.hasSSE = 0;
+	__cpu_detect_caps.hasSSE2 = 0;
+	__cpu_detect_caps.hasSSE3 = 0;
+	__cpu_detect_caps.hasSSSE3 = 0;
+#endif
+#endif /* ARCH_X86 */
+
+#if defined(ARCH_POWERPC)
+	check_os_altivec_support();
+#endif /* ARCH_POWERPC */
+
+	__cpu_detect_initialized = 1;
+}
+
+struct cpu_detect_caps *cpu_detect_get_caps()
+{
+	return &__cpu_detect_caps;
+}
+
+/* The getters and setters for feature flags */
+int cpu_detect_get_tsc()
+{
+	return __cpu_detect_caps.hasTSC;
+}
+
+int cpu_detect_get_mmx()
+{
+	return __cpu_detect_caps.hasMMX;
+}
+
+int cpu_detect_get_mmx2()
+{
+	return __cpu_detect_caps.hasMMX2;
+}
+
+int cpu_detect_get_sse()
+{
+	return __cpu_detect_caps.hasSSE;
+}
+
+int cpu_detect_get_sse2()
+{
+	return __cpu_detect_caps.hasSSE2;
+}
+
+int cpu_detect_get_sse3()
+{
+	return __cpu_detect_caps.hasSSE3;
+}
+
+int cpu_detect_get_ssse3()
+{
+	return __cpu_detect_caps.hasSSSE3;
+}
+
+int cpu_detect_get_3dnow()
+{
+	return __cpu_detect_caps.has3DNow;
+}
+
+int cpu_detect_get_3dnow2()
+{
+	return __cpu_detect_caps.has3DNowExt;
+}
+
+int cpu_detect_get_altivec()
+{
+	return __cpu_detect_caps.hasAltiVec;
+}
+
diff --git a/src/gallium/auxiliary/util/u_cpu_detect.h b/src/gallium/auxiliary/util/u_cpu_detect.h
new file mode 100644
index 00000000000..1612d49286a
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_cpu_detect.h
@@ -0,0 +1,78 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Dennis Smit
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * AUTHORS, COPYRIGHT HOLDERS, AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ ***************************************************************************/
+
+/*
+ * Based on the work of Eric Anholt <[email protected]>
+ */
+
+#ifndef _CPU_DETECT_H
+#define _CPU_DETECT_H
+
+typedef enum {
+	CPU_DETECT_TYPE_MIPS,
+	CPU_DETECT_TYPE_ALPHA,
+	CPU_DETECT_TYPE_SPARC,
+	CPU_DETECT_TYPE_X86,
+	CPU_DETECT_TYPE_POWERPC,
+	CPU_DETECT_TYPE_OTHER
+} cpu_detect_type;
+
+struct cpu_detect_caps {
+	cpu_detect_type	type;
+	int		nrcpu;
+
+	/* Feature flags */
+	int		x86cpuType;
+	int		cacheline;
+
+	int		hasTSC;
+	int		hasMMX;
+	int		hasMMX2;
+	int		hasSSE;
+	int		hasSSE2;
+	int		hasSSE3;
+	int		hasSSSE3;
+	int		has3DNow;
+	int		has3DNowExt;
+	int		hasAltiVec;
+};
+
+/* prototypes */
+void cpu_detect_initialize(void);
+struct cpu_detect_caps *cpu_detect_get_caps(void);
+
+int cpu_detect_get_tsc(void);
+int cpu_detect_get_mmx(void);
+int cpu_detect_get_mmx2(void);
+int cpu_detect_get_sse(void);
+int cpu_detect_get_sse2(void);
+int cpu_detect_get_sse3(void);
+int cpu_detect_get_ssse3(void);
+int cpu_detect_get_3dnow(void);
+int cpu_detect_get_3dnow2(void);
+int cpu_detect_get_altivec(void);
+
+#endif /* _CPU_DETECT_H */
diff --git a/src/gallium/auxiliary/util/u_double_list.h b/src/gallium/auxiliary/util/u_double_list.h
new file mode 100644
index 00000000000..d108d92e52b
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_double_list.h
@@ -0,0 +1,99 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND. USA.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, 
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ **************************************************************************/
+
+/**
+ * \file
+ * List macros heavily inspired by the Linux kernel
+ * list handling. No list looping yet.
+ * 
+ * Is not threadsafe, so common operations need to
+ * be protected using an external mutex.
+ */
+
+#ifndef _U_DOUBLE_LIST_H_
+#define _U_DOUBLE_LIST_H_
+
+
+#include <stddef.h>
+
+
+struct list_head
+{
+    struct list_head *prev;
+    struct list_head *next;
+};
+
+
+#define LIST_INITHEAD(__item)			\
+  do {						\
+    (__item)->prev = (__item);			\
+    (__item)->next = (__item);			\
+  } while (0)
+
+#define LIST_ADD(__item, __list)		\
+  do {						\
+    (__item)->prev = (__list);			\
+    (__item)->next = (__list)->next;		\
+    (__list)->next->prev = (__item);		\
+    (__list)->next = (__item);			\
+  } while (0)
+
+#define LIST_ADDTAIL(__item, __list)		\
+  do {						\
+    (__item)->next = (__list);			\
+    (__item)->prev = (__list)->prev;		\
+    (__list)->prev->next = (__item);		\
+    (__list)->prev = (__item);			\
+  } while(0)
+
+#define LIST_REPLACE(__from, __to)		\
+  do {						\
+    (__to)->prev = (__from)->prev;		\
+    (__to)->next = (__from)->next;		\
+    (__from)->next->prev = (__to);		\
+    (__from)->prev->next = (__to);		\
+  } while (0)
+
+#define LIST_DEL(__item)			\
+  do {						\
+    (__item)->prev->next = (__item)->next;	\
+    (__item)->next->prev = (__item)->prev;	\
+  } while(0)
+
+#define LIST_DELINIT(__item)			\
+  do {						\
+    (__item)->prev->next = (__item)->next;	\
+    (__item)->next->prev = (__item)->prev;	\
+    (__item)->next = (__item);			\
+    (__item)->prev = (__item);			\
+  } while(0)
+
+#define LIST_ENTRY(__type, __item, __field)   \
+    ((__type *)(((char *)(__item)) - offsetof(__type, __field)))
+
+
+#endif /*_U_DOUBLE_LIST_H_*/
diff --git a/src/gallium/auxiliary/util/u_draw_quad.c b/src/gallium/auxiliary/util/u_draw_quad.c
new file mode 100644
index 00000000000..f282f3d2891
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_draw_quad.c
@@ -0,0 +1,132 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+
+#include "pipe/p_context.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_inlines.h"
+#include "util/u_draw_quad.h"
+
+
+/**
+ * Draw a simple vertex buffer / primitive.
+ * Limited to float[4] vertex attribs, tightly packed.
+ */
+void 
+util_draw_vertex_buffer(struct pipe_context *pipe,
+                        struct pipe_buffer *vbuf,
+                        uint offset,
+                        uint prim_type,
+                        uint num_verts,
+                        uint num_attribs)
+{
+   struct pipe_vertex_buffer vbuffer;
+   struct pipe_vertex_element velements[PIPE_MAX_ATTRIBS];
+   uint i;
+
+   assert(num_attribs <= PIPE_MAX_ATTRIBS);
+
+   /* tell pipe about the vertex buffer */
+   vbuffer.buffer = vbuf;
+   vbuffer.stride = num_attribs * 4 * sizeof(float);  /* vertex size */
+   vbuffer.buffer_offset = offset;
+   pipe->set_vertex_buffers(pipe, 1, &vbuffer);
+
+   /* tell pipe about the vertex attributes */
+   for (i = 0; i < num_attribs; i++) {
+      velements[i].src_offset = i * 4 * sizeof(float);
+      velements[i].vertex_buffer_index = 0;
+      velements[i].src_format = PIPE_FORMAT_R32G32B32A32_FLOAT;
+      velements[i].nr_components = 4;
+   }
+   pipe->set_vertex_elements(pipe, num_attribs, velements);
+
+   /* draw */
+   pipe->draw_arrays(pipe, prim_type, 0, num_verts);
+}
+
+
+
+/**
+ * Draw screen-aligned textured quad.
+ * Note: this function allocs/destroys a vertex buffer and isn't especially
+ * efficient.
+ */
+void 
+util_draw_texquad(struct pipe_context *pipe,
+                  float x0, float y0, float x1, float y1, float z)
+{
+   struct pipe_buffer *vbuf;
+   uint numAttribs = 2, vertexBytes, i, j;
+
+   vertexBytes = 4 * (4 * numAttribs * sizeof(float));
+
+   /* XXX create one-time */
+   vbuf = pipe_buffer_create(pipe->screen, 32,
+                             PIPE_BUFFER_USAGE_VERTEX, vertexBytes);
+   if (vbuf) {
+      float *v = (float *) pipe_buffer_map(pipe->screen, vbuf,
+                                           PIPE_BUFFER_USAGE_CPU_WRITE);
+      if (v) {
+         /*
+          * Load vertex buffer
+          */
+         for (i = j = 0; i < 4; i++) {
+            v[j + 2] = z;   /* z */
+            v[j + 3] = 1.0; /* w */
+            v[j + 6] = 0.0; /* r */
+            v[j + 7] = 1.0; /* q */
+            j += 8;
+         }
+
+         v[0] = x0;
+         v[1] = y0;
+         v[4] = 0.0; /*s*/
+         v[5] = 0.0; /*t*/
+
+         v[8] = x1;
+         v[9] = y0;
+         v[12] = 1.0;
+         v[13] = 0.0;
+
+         v[16] = x1;
+         v[17] = y1;
+         v[20] = 1.0;
+         v[21] = 1.0;
+
+         v[24] = x0;
+         v[25] = y1;
+         v[28] = 0.0;
+         v[29] = 1.0;
+
+         pipe_buffer_unmap(pipe->screen, vbuf);
+         util_draw_vertex_buffer(pipe, vbuf, 0, PIPE_PRIM_TRIANGLE_FAN, 4, 2);
+      }
+
+      pipe_buffer_reference(pipe->screen, &vbuf, NULL);
+   }
+}
diff --git a/src/gallium/auxiliary/util/u_draw_quad.h b/src/gallium/auxiliary/util/u_draw_quad.h
new file mode 100644
index 00000000000..00d3f5b7158
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_draw_quad.h
@@ -0,0 +1,54 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#ifndef U_DRAWQUAD_H
+#define U_DRAWQUAD_H
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+struct pipe_buffer;
+
+extern void 
+util_draw_vertex_buffer(struct pipe_context *pipe,
+                        struct pipe_buffer *vbuf, uint offset,
+                        uint num_attribs, uint num_verts, uint prim_type);
+
+
+extern void 
+util_draw_texquad(struct pipe_context *pipe,
+                  float x0, float y0, float x1, float y1, float z);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_gen_mipmap.c b/src/gallium/auxiliary/util/u_gen_mipmap.c
new file mode 100644
index 00000000000..2b4cdab6cf3
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_gen_mipmap.c
@@ -0,0 +1,1559 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * Copyright 2008  VMware, Inc.  All rights reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Mipmap generation utility
+ *  
+ * @author Brian Paul
+ */
+
+
+#include "pipe/p_context.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "util/u_memory.h"
+#include "util/u_draw_quad.h"
+#include "util/u_gen_mipmap.h"
+#include "util/u_simple_shaders.h"
+
+#include "tgsi/tgsi_build.h"
+#include "tgsi/tgsi_dump.h"
+#include "tgsi/tgsi_parse.h"
+
+#include "cso_cache/cso_context.h"
+
+
+struct gen_mipmap_state
+{
+   struct pipe_context *pipe;
+   struct cso_context *cso;
+
+   struct pipe_blend_state blend;
+   struct pipe_depth_stencil_alpha_state depthstencil;
+   struct pipe_rasterizer_state rasterizer;
+   struct pipe_sampler_state sampler;
+   struct pipe_viewport_state viewport;
+
+   struct pipe_shader_state vert_shader;
+   struct pipe_shader_state frag_shader;
+   void *vs;
+   void *fs;
+
+   struct pipe_buffer *vbuf;  /**< quad vertices */
+   unsigned vbuf_slot;
+
+   float vertices[4][2][4];   /**< vertex/texcoords for quad */
+};
+
+
+
+enum dtype
+{
+   UBYTE,
+   UBYTE_3_3_2,
+   USHORT,
+   USHORT_4_4_4_4,
+   USHORT_5_6_5,
+   USHORT_1_5_5_5_REV,
+   UINT,
+   FLOAT,
+   HALF_FLOAT
+};
+
+
+typedef ushort half_float;
+
+
+static half_float
+float_to_half(float f)
+{
+   /* XXX fix this */
+   return 0;
+}
+
+static float
+half_to_float(half_float h)
+{
+   /* XXX fix this */
+   return 0.0f;
+}
+
+
+
+
+/**
+ * \name Support macros for do_row and do_row_3d
+ *
+ * The macro madness is here for two reasons.  First, it compacts the code
+ * slightly.  Second, it makes it much easier to adjust the specifics of the
+ * filter to tune the rounding characteristics.
+ */
+/*@{*/
+#define DECLARE_ROW_POINTERS(t, e) \
+      const t(*rowA)[e] = (const t(*)[e]) srcRowA; \
+      const t(*rowB)[e] = (const t(*)[e]) srcRowB; \
+      const t(*rowC)[e] = (const t(*)[e]) srcRowC; \
+      const t(*rowD)[e] = (const t(*)[e]) srcRowD; \
+      t(*dst)[e] = (t(*)[e]) dstRow
+
+#define DECLARE_ROW_POINTERS0(t) \
+      const t *rowA = (const t *) srcRowA; \
+      const t *rowB = (const t *) srcRowB; \
+      const t *rowC = (const t *) srcRowC; \
+      const t *rowD = (const t *) srcRowD; \
+      t *dst = (t *) dstRow
+
+#define FILTER_SUM_3D(Aj, Ak, Bj, Bk, Cj, Ck, Dj, Dk) \
+   ((unsigned) Aj + (unsigned) Ak \
+    + (unsigned) Bj + (unsigned) Bk \
+    + (unsigned) Cj + (unsigned) Ck \
+    + (unsigned) Dj + (unsigned) Dk \
+    + 4) >> 3
+
+#define FILTER_3D(e) \
+   do { \
+      dst[i][e] = FILTER_SUM_3D(rowA[j][e], rowA[k][e], \
+                                rowB[j][e], rowB[k][e], \
+                                rowC[j][e], rowC[k][e], \
+                                rowD[j][e], rowD[k][e]); \
+   } while(0)
+   
+#define FILTER_F_3D(e) \
+   do { \
+      dst[i][e] = (rowA[j][e] + rowA[k][e] \
+                   + rowB[j][e] + rowB[k][e] \
+                   + rowC[j][e] + rowC[k][e] \
+                   + rowD[j][e] + rowD[k][e]) * 0.125F; \
+   } while(0)
+
+#define FILTER_HF_3D(e) \
+   do { \
+      const float aj = half_to_float(rowA[j][e]); \
+      const float ak = half_to_float(rowA[k][e]); \
+      const float bj = half_to_float(rowB[j][e]); \
+      const float bk = half_to_float(rowB[k][e]); \
+      const float cj = half_to_float(rowC[j][e]); \
+      const float ck = half_to_float(rowC[k][e]); \
+      const float dj = half_to_float(rowD[j][e]); \
+      const float dk = half_to_float(rowD[k][e]); \
+      dst[i][e] = float_to_half((aj + ak + bj + bk + cj + ck + dj + dk) \
+                                      * 0.125F); \
+   } while(0)
+/*@}*/
+
+
+/**
+ * Average together two rows of a source image to produce a single new
+ * row in the dest image.  It's legal for the two source rows to point
+ * to the same data.  The source width must be equal to either the
+ * dest width or two times the dest width.
+ * \param datatype  GL_UNSIGNED_BYTE, GL_UNSIGNED_SHORT, GL_FLOAT, etc.
+ * \param comps  number of components per pixel (1..4)
+ */
+static void
+do_row(enum dtype datatype, uint comps, int srcWidth,
+       const void *srcRowA, const void *srcRowB,
+       int dstWidth, void *dstRow)
+{
+   const uint k0 = (srcWidth == dstWidth) ? 0 : 1;
+   const uint colStride = (srcWidth == dstWidth) ? 1 : 2;
+
+   assert(comps >= 1);
+   assert(comps <= 4);
+
+   /* This assertion is no longer valid with non-power-of-2 textures
+   assert(srcWidth == dstWidth || srcWidth == 2 * dstWidth);
+   */
+
+   if (datatype == UBYTE && comps == 4) {
+      uint i, j, k;
+      const ubyte(*rowA)[4] = (const ubyte(*)[4]) srcRowA;
+      const ubyte(*rowB)[4] = (const ubyte(*)[4]) srcRowB;
+      ubyte(*dst)[4] = (ubyte(*)[4]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
+         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
+         dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
+      }
+   }
+   else if (datatype == UBYTE && comps == 3) {
+      uint i, j, k;
+      const ubyte(*rowA)[3] = (const ubyte(*)[3]) srcRowA;
+      const ubyte(*rowB)[3] = (const ubyte(*)[3]) srcRowB;
+      ubyte(*dst)[3] = (ubyte(*)[3]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
+         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
+      }
+   }
+   else if (datatype == UBYTE && comps == 2) {
+      uint i, j, k;
+      const ubyte(*rowA)[2] = (const ubyte(*)[2]) srcRowA;
+      const ubyte(*rowB)[2] = (const ubyte(*)[2]) srcRowB;
+      ubyte(*dst)[2] = (ubyte(*)[2]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) >> 2;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) >> 2;
+      }
+   }
+   else if (datatype == UBYTE && comps == 1) {
+      uint i, j, k;
+      const ubyte *rowA = (const ubyte *) srcRowA;
+      const ubyte *rowB = (const ubyte *) srcRowB;
+      ubyte *dst = (ubyte *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) >> 2;
+      }
+   }
+
+   else if (datatype == USHORT && comps == 4) {
+      uint i, j, k;
+      const ushort(*rowA)[4] = (const ushort(*)[4]) srcRowA;
+      const ushort(*rowB)[4] = (const ushort(*)[4]) srcRowB;
+      ushort(*dst)[4] = (ushort(*)[4]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
+         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
+         dst[i][3] = (rowA[j][3] + rowA[k][3] + rowB[j][3] + rowB[k][3]) / 4;
+      }
+   }
+   else if (datatype == USHORT && comps == 3) {
+      uint i, j, k;
+      const ushort(*rowA)[3] = (const ushort(*)[3]) srcRowA;
+      const ushort(*rowB)[3] = (const ushort(*)[3]) srcRowB;
+      ushort(*dst)[3] = (ushort(*)[3]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
+         dst[i][2] = (rowA[j][2] + rowA[k][2] + rowB[j][2] + rowB[k][2]) / 4;
+      }
+   }
+   else if (datatype == USHORT && comps == 2) {
+      uint i, j, k;
+      const ushort(*rowA)[2] = (const ushort(*)[2]) srcRowA;
+      const ushort(*rowB)[2] = (const ushort(*)[2]) srcRowB;
+      ushort(*dst)[2] = (ushort(*)[2]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] + rowB[j][0] + rowB[k][0]) / 4;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] + rowB[j][1] + rowB[k][1]) / 4;
+      }
+   }
+   else if (datatype == USHORT && comps == 1) {
+      uint i, j, k;
+      const ushort *rowA = (const ushort *) srcRowA;
+      const ushort *rowB = (const ushort *) srcRowB;
+      ushort *dst = (ushort *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) / 4;
+      }
+   }
+
+   else if (datatype == FLOAT && comps == 4) {
+      uint i, j, k;
+      const float(*rowA)[4] = (const float(*)[4]) srcRowA;
+      const float(*rowB)[4] = (const float(*)[4]) srcRowB;
+      float(*dst)[4] = (float(*)[4]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] +
+                      rowB[j][0] + rowB[k][0]) * 0.25F;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] +
+                      rowB[j][1] + rowB[k][1]) * 0.25F;
+         dst[i][2] = (rowA[j][2] + rowA[k][2] +
+                      rowB[j][2] + rowB[k][2]) * 0.25F;
+         dst[i][3] = (rowA[j][3] + rowA[k][3] +
+                      rowB[j][3] + rowB[k][3]) * 0.25F;
+      }
+   }
+   else if (datatype == FLOAT && comps == 3) {
+      uint i, j, k;
+      const float(*rowA)[3] = (const float(*)[3]) srcRowA;
+      const float(*rowB)[3] = (const float(*)[3]) srcRowB;
+      float(*dst)[3] = (float(*)[3]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] +
+                      rowB[j][0] + rowB[k][0]) * 0.25F;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] +
+                      rowB[j][1] + rowB[k][1]) * 0.25F;
+         dst[i][2] = (rowA[j][2] + rowA[k][2] +
+                      rowB[j][2] + rowB[k][2]) * 0.25F;
+      }
+   }
+   else if (datatype == FLOAT && comps == 2) {
+      uint i, j, k;
+      const float(*rowA)[2] = (const float(*)[2]) srcRowA;
+      const float(*rowB)[2] = (const float(*)[2]) srcRowB;
+      float(*dst)[2] = (float(*)[2]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i][0] = (rowA[j][0] + rowA[k][0] +
+                      rowB[j][0] + rowB[k][0]) * 0.25F;
+         dst[i][1] = (rowA[j][1] + rowA[k][1] +
+                      rowB[j][1] + rowB[k][1]) * 0.25F;
+      }
+   }
+   else if (datatype == FLOAT && comps == 1) {
+      uint i, j, k;
+      const float *rowA = (const float *) srcRowA;
+      const float *rowB = (const float *) srcRowB;
+      float *dst = (float *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i] = (rowA[j] + rowA[k] + rowB[j] + rowB[k]) * 0.25F;
+      }
+   }
+
+#if 0
+   else if (datatype == HALF_FLOAT && comps == 4) {
+      uint i, j, k, comp;
+      const half_float(*rowA)[4] = (const half_float(*)[4]) srcRowA;
+      const half_float(*rowB)[4] = (const half_float(*)[4]) srcRowB;
+      half_float(*dst)[4] = (half_float(*)[4]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         for (comp = 0; comp < 4; comp++) {
+            float aj, ak, bj, bk;
+            aj = half_to_float(rowA[j][comp]);
+            ak = half_to_float(rowA[k][comp]);
+            bj = half_to_float(rowB[j][comp]);
+            bk = half_to_float(rowB[k][comp]);
+            dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
+         }
+      }
+   }
+   else if (datatype == HALF_FLOAT && comps == 3) {
+      uint i, j, k, comp;
+      const half_float(*rowA)[3] = (const half_float(*)[3]) srcRowA;
+      const half_float(*rowB)[3] = (const half_float(*)[3]) srcRowB;
+      half_float(*dst)[3] = (half_float(*)[3]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         for (comp = 0; comp < 3; comp++) {
+            float aj, ak, bj, bk;
+            aj = half_to_float(rowA[j][comp]);
+            ak = half_to_float(rowA[k][comp]);
+            bj = half_to_float(rowB[j][comp]);
+            bk = half_to_float(rowB[k][comp]);
+            dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
+         }
+      }
+   }
+   else if (datatype == HALF_FLOAT && comps == 2) {
+      uint i, j, k, comp;
+      const half_float(*rowA)[2] = (const half_float(*)[2]) srcRowA;
+      const half_float(*rowB)[2] = (const half_float(*)[2]) srcRowB;
+      half_float(*dst)[2] = (half_float(*)[2]) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         for (comp = 0; comp < 2; comp++) {
+            float aj, ak, bj, bk;
+            aj = half_to_float(rowA[j][comp]);
+            ak = half_to_float(rowA[k][comp]);
+            bj = half_to_float(rowB[j][comp]);
+            bk = half_to_float(rowB[k][comp]);
+            dst[i][comp] = float_to_half((aj + ak + bj + bk) * 0.25F);
+         }
+      }
+   }
+   else if (datatype == HALF_FLOAT && comps == 1) {
+      uint i, j, k;
+      const half_float *rowA = (const half_float *) srcRowA;
+      const half_float *rowB = (const half_float *) srcRowB;
+      half_float *dst = (half_float *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         float aj, ak, bj, bk;
+         aj = half_to_float(rowA[j]);
+         ak = half_to_float(rowA[k]);
+         bj = half_to_float(rowB[j]);
+         bk = half_to_float(rowB[k]);
+         dst[i] = float_to_half((aj + ak + bj + bk) * 0.25F);
+      }
+   }
+#endif
+
+   else if (datatype == UINT && comps == 1) {
+      uint i, j, k;
+      const uint *rowA = (const uint *) srcRowA;
+      const uint *rowB = (const uint *) srcRowB;
+      uint *dst = (uint *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         dst[i] = rowA[j] / 4 + rowA[k] / 4 + rowB[j] / 4 + rowB[k] / 4;
+      }
+   }
+
+   else if (datatype == USHORT_5_6_5 && comps == 3) {
+      uint i, j, k;
+      const ushort *rowA = (const ushort *) srcRowA;
+      const ushort *rowB = (const ushort *) srcRowB;
+      ushort *dst = (ushort *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         const int rowAr0 = rowA[j] & 0x1f;
+         const int rowAr1 = rowA[k] & 0x1f;
+         const int rowBr0 = rowB[j] & 0x1f;
+         const int rowBr1 = rowB[k] & 0x1f;
+         const int rowAg0 = (rowA[j] >> 5) & 0x3f;
+         const int rowAg1 = (rowA[k] >> 5) & 0x3f;
+         const int rowBg0 = (rowB[j] >> 5) & 0x3f;
+         const int rowBg1 = (rowB[k] >> 5) & 0x3f;
+         const int rowAb0 = (rowA[j] >> 11) & 0x1f;
+         const int rowAb1 = (rowA[k] >> 11) & 0x1f;
+         const int rowBb0 = (rowB[j] >> 11) & 0x1f;
+         const int rowBb1 = (rowB[k] >> 11) & 0x1f;
+         const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
+         const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
+         const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
+         dst[i] = (blue << 11) | (green << 5) | red;
+      }
+   }
+   else if (datatype == USHORT_4_4_4_4 && comps == 4) {
+      uint i, j, k;
+      const ushort *rowA = (const ushort *) srcRowA;
+      const ushort *rowB = (const ushort *) srcRowB;
+      ushort *dst = (ushort *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         const int rowAr0 = rowA[j] & 0xf;
+         const int rowAr1 = rowA[k] & 0xf;
+         const int rowBr0 = rowB[j] & 0xf;
+         const int rowBr1 = rowB[k] & 0xf;
+         const int rowAg0 = (rowA[j] >> 4) & 0xf;
+         const int rowAg1 = (rowA[k] >> 4) & 0xf;
+         const int rowBg0 = (rowB[j] >> 4) & 0xf;
+         const int rowBg1 = (rowB[k] >> 4) & 0xf;
+         const int rowAb0 = (rowA[j] >> 8) & 0xf;
+         const int rowAb1 = (rowA[k] >> 8) & 0xf;
+         const int rowBb0 = (rowB[j] >> 8) & 0xf;
+         const int rowBb1 = (rowB[k] >> 8) & 0xf;
+         const int rowAa0 = (rowA[j] >> 12) & 0xf;
+         const int rowAa1 = (rowA[k] >> 12) & 0xf;
+         const int rowBa0 = (rowB[j] >> 12) & 0xf;
+         const int rowBa1 = (rowB[k] >> 12) & 0xf;
+         const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
+         const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
+         const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
+         const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
+         dst[i] = (alpha << 12) | (blue << 8) | (green << 4) | red;
+      }
+   }
+   else if (datatype == USHORT_1_5_5_5_REV && comps == 4) {
+      uint i, j, k;
+      const ushort *rowA = (const ushort *) srcRowA;
+      const ushort *rowB = (const ushort *) srcRowB;
+      ushort *dst = (ushort *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         const int rowAr0 = rowA[j] & 0x1f;
+         const int rowAr1 = rowA[k] & 0x1f;
+         const int rowBr0 = rowB[j] & 0x1f;
+         const int rowBr1 = rowB[k] & 0x1f;
+         const int rowAg0 = (rowA[j] >> 5) & 0x1f;
+         const int rowAg1 = (rowA[k] >> 5) & 0x1f;
+         const int rowBg0 = (rowB[j] >> 5) & 0x1f;
+         const int rowBg1 = (rowB[k] >> 5) & 0x1f;
+         const int rowAb0 = (rowA[j] >> 10) & 0x1f;
+         const int rowAb1 = (rowA[k] >> 10) & 0x1f;
+         const int rowBb0 = (rowB[j] >> 10) & 0x1f;
+         const int rowBb1 = (rowB[k] >> 10) & 0x1f;
+         const int rowAa0 = (rowA[j] >> 15) & 0x1;
+         const int rowAa1 = (rowA[k] >> 15) & 0x1;
+         const int rowBa0 = (rowB[j] >> 15) & 0x1;
+         const int rowBa1 = (rowB[k] >> 15) & 0x1;
+         const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
+         const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
+         const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
+         const int alpha = (rowAa0 + rowAa1 + rowBa0 + rowBa1) >> 2;
+         dst[i] = (alpha << 15) | (blue << 10) | (green << 5) | red;
+      }
+   }
+   else if (datatype == UBYTE_3_3_2 && comps == 3) {
+      uint i, j, k;
+      const ubyte *rowA = (const ubyte *) srcRowA;
+      const ubyte *rowB = (const ubyte *) srcRowB;
+      ubyte *dst = (ubyte *) dstRow;
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         const int rowAr0 = rowA[j] & 0x3;
+         const int rowAr1 = rowA[k] & 0x3;
+         const int rowBr0 = rowB[j] & 0x3;
+         const int rowBr1 = rowB[k] & 0x3;
+         const int rowAg0 = (rowA[j] >> 2) & 0x7;
+         const int rowAg1 = (rowA[k] >> 2) & 0x7;
+         const int rowBg0 = (rowB[j] >> 2) & 0x7;
+         const int rowBg1 = (rowB[k] >> 2) & 0x7;
+         const int rowAb0 = (rowA[j] >> 5) & 0x7;
+         const int rowAb1 = (rowA[k] >> 5) & 0x7;
+         const int rowBb0 = (rowB[j] >> 5) & 0x7;
+         const int rowBb1 = (rowB[k] >> 5) & 0x7;
+         const int red = (rowAr0 + rowAr1 + rowBr0 + rowBr1) >> 2;
+         const int green = (rowAg0 + rowAg1 + rowBg0 + rowBg1) >> 2;
+         const int blue = (rowAb0 + rowAb1 + rowBb0 + rowBb1) >> 2;
+         dst[i] = (blue << 5) | (green << 2) | red;
+      }
+   }
+   else {
+      debug_printf("bad format in do_row()");
+   }
+}
+
+
+/**
+ * Average together four rows of a source image to produce a single new
+ * row in the dest image.  It's legal for the two source rows to point
+ * to the same data.  The source width must be equal to either the
+ * dest width or two times the dest width.
+ *
+ * \param datatype  GL pixel type \c GL_UNSIGNED_BYTE, \c GL_UNSIGNED_SHORT,
+ *                  \c GL_FLOAT, etc.
+ * \param comps     number of components per pixel (1..4)
+ * \param srcWidth  Width of a row in the source data
+ * \param srcRowA   Pointer to one of the rows of source data
+ * \param srcRowB   Pointer to one of the rows of source data
+ * \param srcRowC   Pointer to one of the rows of source data
+ * \param srcRowD   Pointer to one of the rows of source data
+ * \param dstWidth  Width of a row in the destination data
+ * \param srcRowA   Pointer to the row of destination data
+ */
+static void
+do_row_3D(enum dtype datatype, uint comps, int srcWidth,
+          const void *srcRowA, const void *srcRowB,
+          const void *srcRowC, const void *srcRowD,
+          int dstWidth, void *dstRow)
+{
+   const uint k0 = (srcWidth == dstWidth) ? 0 : 1;
+   const uint colStride = (srcWidth == dstWidth) ? 1 : 2;
+   uint i, j, k;
+
+   assert(comps >= 1);
+   assert(comps <= 4);
+
+   if ((datatype == UBYTE) && (comps == 4)) {
+      DECLARE_ROW_POINTERS(ubyte, 4);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         FILTER_3D(0);
+         FILTER_3D(1);
+         FILTER_3D(2);
+         FILTER_3D(3);
+      }
+   }
+   else if ((datatype == UBYTE) && (comps == 3)) {
+      DECLARE_ROW_POINTERS(ubyte, 3);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         FILTER_3D(0);
+         FILTER_3D(1);
+         FILTER_3D(2);
+      }
+   }
+   else if ((datatype == UBYTE) && (comps == 2)) {
+      DECLARE_ROW_POINTERS(ubyte, 2);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         FILTER_3D(0);
+         FILTER_3D(1);
+      }
+   }
+   else if ((datatype == UBYTE) && (comps == 1)) {
+      DECLARE_ROW_POINTERS(ubyte, 1);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         FILTER_3D(0);
+      }
+   }
+   else if ((datatype == USHORT) && (comps == 4)) {
+      DECLARE_ROW_POINTERS(ushort, 4);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         FILTER_3D(0);
+         FILTER_3D(1);
+         FILTER_3D(2);
+         FILTER_3D(3);
+      }
+   }
+   else if ((datatype == USHORT) && (comps == 3)) {
+      DECLARE_ROW_POINTERS(ushort, 3);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         FILTER_3D(0);
+         FILTER_3D(1);
+         FILTER_3D(2);
+      }
+   }
+   else if ((datatype == USHORT) && (comps == 2)) {
+      DECLARE_ROW_POINTERS(ushort, 2);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         FILTER_3D(0);
+         FILTER_3D(1);
+      }
+   }
+   else if ((datatype == USHORT) && (comps == 1)) {
+      DECLARE_ROW_POINTERS(ushort, 1);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         FILTER_3D(0);
+      }
+   }
+   else if ((datatype == FLOAT) && (comps == 4)) {
+      DECLARE_ROW_POINTERS(float, 4);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         FILTER_F_3D(0);
+         FILTER_F_3D(1);
+         FILTER_F_3D(2);
+         FILTER_F_3D(3);
+      }
+   }
+   else if ((datatype == FLOAT) && (comps == 3)) {
+      DECLARE_ROW_POINTERS(float, 3);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         FILTER_F_3D(0);
+         FILTER_F_3D(1);
+         FILTER_F_3D(2);
+      }
+   }
+   else if ((datatype == FLOAT) && (comps == 2)) {
+      DECLARE_ROW_POINTERS(float, 2);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         FILTER_F_3D(0);
+         FILTER_F_3D(1);
+      }
+   }
+   else if ((datatype == FLOAT) && (comps == 1)) {
+      DECLARE_ROW_POINTERS(float, 1);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         FILTER_F_3D(0);
+      }
+   }
+   else if ((datatype == HALF_FLOAT) && (comps == 4)) {
+      DECLARE_ROW_POINTERS(half_float, 4);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         FILTER_HF_3D(0);
+         FILTER_HF_3D(1);
+         FILTER_HF_3D(2);
+         FILTER_HF_3D(3);
+      }
+   }
+   else if ((datatype == HALF_FLOAT) && (comps == 3)) {
+      DECLARE_ROW_POINTERS(half_float, 4);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         FILTER_HF_3D(0);
+         FILTER_HF_3D(1);
+         FILTER_HF_3D(2);
+      }
+   }
+   else if ((datatype == HALF_FLOAT) && (comps == 2)) {
+      DECLARE_ROW_POINTERS(half_float, 4);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         FILTER_HF_3D(0);
+         FILTER_HF_3D(1);
+      }
+   }
+   else if ((datatype == HALF_FLOAT) && (comps == 1)) {
+      DECLARE_ROW_POINTERS(half_float, 4);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         FILTER_HF_3D(0);
+      }
+   }
+   else if ((datatype == UINT) && (comps == 1)) {
+      const uint *rowA = (const uint *) srcRowA;
+      const uint *rowB = (const uint *) srcRowB;
+      const uint *rowC = (const uint *) srcRowC;
+      const uint *rowD = (const uint *) srcRowD;
+      float *dst = (float *) dstRow;
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         const uint64_t tmp = (((uint64_t) rowA[j] + (uint64_t) rowA[k])
+                               + ((uint64_t) rowB[j] + (uint64_t) rowB[k])
+                               + ((uint64_t) rowC[j] + (uint64_t) rowC[k])
+                               + ((uint64_t) rowD[j] + (uint64_t) rowD[k]));
+         dst[i] = (float)((double) tmp * 0.125);
+      }
+   }
+   else if ((datatype == USHORT_5_6_5) && (comps == 3)) {
+      DECLARE_ROW_POINTERS0(ushort);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         const int rowAr0 = rowA[j] & 0x1f;
+         const int rowAr1 = rowA[k] & 0x1f;
+         const int rowBr0 = rowB[j] & 0x1f;
+         const int rowBr1 = rowB[k] & 0x1f;
+         const int rowCr0 = rowC[j] & 0x1f;
+         const int rowCr1 = rowC[k] & 0x1f;
+         const int rowDr0 = rowD[j] & 0x1f;
+         const int rowDr1 = rowD[k] & 0x1f;
+         const int rowAg0 = (rowA[j] >> 5) & 0x3f;
+         const int rowAg1 = (rowA[k] >> 5) & 0x3f;
+         const int rowBg0 = (rowB[j] >> 5) & 0x3f;
+         const int rowBg1 = (rowB[k] >> 5) & 0x3f;
+         const int rowCg0 = (rowC[j] >> 5) & 0x3f;
+         const int rowCg1 = (rowC[k] >> 5) & 0x3f;
+         const int rowDg0 = (rowD[j] >> 5) & 0x3f;
+         const int rowDg1 = (rowD[k] >> 5) & 0x3f;
+         const int rowAb0 = (rowA[j] >> 11) & 0x1f;
+         const int rowAb1 = (rowA[k] >> 11) & 0x1f;
+         const int rowBb0 = (rowB[j] >> 11) & 0x1f;
+         const int rowBb1 = (rowB[k] >> 11) & 0x1f;
+         const int rowCb0 = (rowC[j] >> 11) & 0x1f;
+         const int rowCb1 = (rowC[k] >> 11) & 0x1f;
+         const int rowDb0 = (rowD[j] >> 11) & 0x1f;
+         const int rowDb1 = (rowD[k] >> 11) & 0x1f;
+         const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
+                                       rowCr0, rowCr1, rowDr0, rowDr1);
+         const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
+                                       rowCg0, rowCg1, rowDg0, rowDg1);
+         const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
+                                       rowCb0, rowCb1, rowDb0, rowDb1);
+         dst[i] = (b << 11) | (g << 5) | r;
+      }
+   }
+   else if ((datatype == USHORT_4_4_4_4) && (comps == 4)) {
+      DECLARE_ROW_POINTERS0(ushort);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         const int rowAr0 = rowA[j] & 0xf;
+         const int rowAr1 = rowA[k] & 0xf;
+         const int rowBr0 = rowB[j] & 0xf;
+         const int rowBr1 = rowB[k] & 0xf;
+         const int rowCr0 = rowC[j] & 0xf;
+         const int rowCr1 = rowC[k] & 0xf;
+         const int rowDr0 = rowD[j] & 0xf;
+         const int rowDr1 = rowD[k] & 0xf;
+         const int rowAg0 = (rowA[j] >> 4) & 0xf;
+         const int rowAg1 = (rowA[k] >> 4) & 0xf;
+         const int rowBg0 = (rowB[j] >> 4) & 0xf;
+         const int rowBg1 = (rowB[k] >> 4) & 0xf;
+         const int rowCg0 = (rowC[j] >> 4) & 0xf;
+         const int rowCg1 = (rowC[k] >> 4) & 0xf;
+         const int rowDg0 = (rowD[j] >> 4) & 0xf;
+         const int rowDg1 = (rowD[k] >> 4) & 0xf;
+         const int rowAb0 = (rowA[j] >> 8) & 0xf;
+         const int rowAb1 = (rowA[k] >> 8) & 0xf;
+         const int rowBb0 = (rowB[j] >> 8) & 0xf;
+         const int rowBb1 = (rowB[k] >> 8) & 0xf;
+         const int rowCb0 = (rowC[j] >> 8) & 0xf;
+         const int rowCb1 = (rowC[k] >> 8) & 0xf;
+         const int rowDb0 = (rowD[j] >> 8) & 0xf;
+         const int rowDb1 = (rowD[k] >> 8) & 0xf;
+         const int rowAa0 = (rowA[j] >> 12) & 0xf;
+         const int rowAa1 = (rowA[k] >> 12) & 0xf;
+         const int rowBa0 = (rowB[j] >> 12) & 0xf;
+         const int rowBa1 = (rowB[k] >> 12) & 0xf;
+         const int rowCa0 = (rowC[j] >> 12) & 0xf;
+         const int rowCa1 = (rowC[k] >> 12) & 0xf;
+         const int rowDa0 = (rowD[j] >> 12) & 0xf;
+         const int rowDa1 = (rowD[k] >> 12) & 0xf;
+         const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
+                                       rowCr0, rowCr1, rowDr0, rowDr1);
+         const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
+                                       rowCg0, rowCg1, rowDg0, rowDg1);
+         const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
+                                       rowCb0, rowCb1, rowDb0, rowDb1);
+         const int a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
+                                       rowCa0, rowCa1, rowDa0, rowDa1);
+
+         dst[i] = (a << 12) | (b << 8) | (g << 4) | r;
+      }
+   }
+   else if ((datatype == USHORT_1_5_5_5_REV) && (comps == 4)) {
+      DECLARE_ROW_POINTERS0(ushort);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         const int rowAr0 = rowA[j] & 0x1f;
+         const int rowAr1 = rowA[k] & 0x1f;
+         const int rowBr0 = rowB[j] & 0x1f;
+         const int rowBr1 = rowB[k] & 0x1f;
+         const int rowCr0 = rowC[j] & 0x1f;
+         const int rowCr1 = rowC[k] & 0x1f;
+         const int rowDr0 = rowD[j] & 0x1f;
+         const int rowDr1 = rowD[k] & 0x1f;
+         const int rowAg0 = (rowA[j] >> 5) & 0x1f;
+         const int rowAg1 = (rowA[k] >> 5) & 0x1f;
+         const int rowBg0 = (rowB[j] >> 5) & 0x1f;
+         const int rowBg1 = (rowB[k] >> 5) & 0x1f;
+         const int rowCg0 = (rowC[j] >> 5) & 0x1f;
+         const int rowCg1 = (rowC[k] >> 5) & 0x1f;
+         const int rowDg0 = (rowD[j] >> 5) & 0x1f;
+         const int rowDg1 = (rowD[k] >> 5) & 0x1f;
+         const int rowAb0 = (rowA[j] >> 10) & 0x1f;
+         const int rowAb1 = (rowA[k] >> 10) & 0x1f;
+         const int rowBb0 = (rowB[j] >> 10) & 0x1f;
+         const int rowBb1 = (rowB[k] >> 10) & 0x1f;
+         const int rowCb0 = (rowC[j] >> 10) & 0x1f;
+         const int rowCb1 = (rowC[k] >> 10) & 0x1f;
+         const int rowDb0 = (rowD[j] >> 10) & 0x1f;
+         const int rowDb1 = (rowD[k] >> 10) & 0x1f;
+         const int rowAa0 = (rowA[j] >> 15) & 0x1;
+         const int rowAa1 = (rowA[k] >> 15) & 0x1;
+         const int rowBa0 = (rowB[j] >> 15) & 0x1;
+         const int rowBa1 = (rowB[k] >> 15) & 0x1;
+         const int rowCa0 = (rowC[j] >> 15) & 0x1;
+         const int rowCa1 = (rowC[k] >> 15) & 0x1;
+         const int rowDa0 = (rowD[j] >> 15) & 0x1;
+         const int rowDa1 = (rowD[k] >> 15) & 0x1;
+         const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
+                                       rowCr0, rowCr1, rowDr0, rowDr1);
+         const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
+                                       rowCg0, rowCg1, rowDg0, rowDg1);
+         const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
+                                       rowCb0, rowCb1, rowDb0, rowDb1);
+         const int a = FILTER_SUM_3D(rowAa0, rowAa1, rowBa0, rowBa1,
+                                       rowCa0, rowCa1, rowDa0, rowDa1);
+
+         dst[i] = (a << 15) | (b << 10) | (g << 5) | r;
+      }
+   }
+   else if ((datatype == UBYTE_3_3_2) && (comps == 3)) {
+      DECLARE_ROW_POINTERS0(ushort);
+
+      for (i = j = 0, k = k0; i < (uint) dstWidth;
+           i++, j += colStride, k += colStride) {
+         const int rowAr0 = rowA[j] & 0x3;
+         const int rowAr1 = rowA[k] & 0x3;
+         const int rowBr0 = rowB[j] & 0x3;
+         const int rowBr1 = rowB[k] & 0x3;
+         const int rowCr0 = rowC[j] & 0x3;
+         const int rowCr1 = rowC[k] & 0x3;
+         const int rowDr0 = rowD[j] & 0x3;
+         const int rowDr1 = rowD[k] & 0x3;
+         const int rowAg0 = (rowA[j] >> 2) & 0x7;
+         const int rowAg1 = (rowA[k] >> 2) & 0x7;
+         const int rowBg0 = (rowB[j] >> 2) & 0x7;
+         const int rowBg1 = (rowB[k] >> 2) & 0x7;
+         const int rowCg0 = (rowC[j] >> 2) & 0x7;
+         const int rowCg1 = (rowC[k] >> 2) & 0x7;
+         const int rowDg0 = (rowD[j] >> 2) & 0x7;
+         const int rowDg1 = (rowD[k] >> 2) & 0x7;
+         const int rowAb0 = (rowA[j] >> 5) & 0x7;
+         const int rowAb1 = (rowA[k] >> 5) & 0x7;
+         const int rowBb0 = (rowB[j] >> 5) & 0x7;
+         const int rowBb1 = (rowB[k] >> 5) & 0x7;
+         const int rowCb0 = (rowC[j] >> 5) & 0x7;
+         const int rowCb1 = (rowC[k] >> 5) & 0x7;
+         const int rowDb0 = (rowD[j] >> 5) & 0x7;
+         const int rowDb1 = (rowD[k] >> 5) & 0x7;
+         const int r = FILTER_SUM_3D(rowAr0, rowAr1, rowBr0, rowBr1,
+                                       rowCr0, rowCr1, rowDr0, rowDr1);
+         const int g = FILTER_SUM_3D(rowAg0, rowAg1, rowBg0, rowBg1,
+                                       rowCg0, rowCg1, rowDg0, rowDg1);
+         const int b = FILTER_SUM_3D(rowAb0, rowAb1, rowBb0, rowBb1,
+                                       rowCb0, rowCb1, rowDb0, rowDb1);
+         dst[i] = (b << 5) | (g << 2) | r;
+      }
+   }
+   else {
+      debug_printf("bad format in do_row_3D()");
+   }
+}
+
+
+
+static void
+format_to_type_comps(enum pipe_format pformat,
+                     enum dtype *datatype, uint *comps)
+{
+   switch (pformat) {
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+   case PIPE_FORMAT_X8R8G8B8_UNORM:
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+   case PIPE_FORMAT_B8G8R8X8_UNORM:
+      *datatype = UBYTE;
+      *comps = 4;
+      return;
+   case PIPE_FORMAT_A1R5G5B5_UNORM:
+      *datatype = USHORT_1_5_5_5_REV;
+      *comps = 4;
+      return;
+   case PIPE_FORMAT_A4R4G4B4_UNORM:
+      *datatype = USHORT_4_4_4_4;
+      *comps = 4;
+      return;
+   case PIPE_FORMAT_R5G6B5_UNORM:
+      *datatype = USHORT_5_6_5;
+      *comps = 3;
+      return;
+   case PIPE_FORMAT_L8_UNORM:
+   case PIPE_FORMAT_A8_UNORM:
+   case PIPE_FORMAT_I8_UNORM:
+      *datatype = UBYTE;
+      *comps = 1;
+      return;
+   case PIPE_FORMAT_A8L8_UNORM:
+      *datatype = UBYTE;
+      *comps = 2;
+      return;
+   default:
+      assert(0);
+      *datatype = UBYTE;
+      *comps = 0;
+      break;
+   }
+}
+
+
+static void
+reduce_1d(enum pipe_format pformat,
+          int srcWidth, const ubyte *srcPtr,
+          int dstWidth, ubyte *dstPtr)
+{
+   enum dtype datatype;
+   uint comps;
+
+   format_to_type_comps(pformat, &datatype, &comps);
+
+   /* we just duplicate the input row, kind of hack, saves code */
+   do_row(datatype, comps,
+          srcWidth, srcPtr, srcPtr,
+          dstWidth, dstPtr);
+}
+
+
+/**
+ * Strides are in bytes.  If zero, it'll be computed as width * bpp.
+ */
+static void
+reduce_2d(enum pipe_format pformat,
+          int srcWidth, int srcHeight,
+          int srcRowStride, const ubyte *srcPtr,
+          int dstWidth, int dstHeight,
+          int dstRowStride, ubyte *dstPtr)
+{
+   enum dtype datatype;
+   uint comps;
+   const int bpt = pf_get_size(pformat);
+   const ubyte *srcA, *srcB;
+   ubyte *dst;
+   int row;
+
+   format_to_type_comps(pformat, &datatype, &comps);
+
+   if (!srcRowStride)
+      srcRowStride = bpt * srcWidth;
+
+   if (!dstRowStride)
+      dstRowStride = bpt * dstWidth;
+
+   /* Compute src and dst pointers */
+   srcA = srcPtr;
+   if (srcHeight > 1) 
+      srcB = srcA + srcRowStride;
+   else
+      srcB = srcA;
+   dst = dstPtr;
+
+   for (row = 0; row < dstHeight; row++) {
+      do_row(datatype, comps,
+             srcWidth, srcA, srcB,
+             dstWidth, dst);
+      srcA += 2 * srcRowStride;
+      srcB += 2 * srcRowStride;
+      dst += dstRowStride;
+   }
+}
+
+
+static void
+reduce_3d(enum pipe_format pformat,
+          int srcWidth, int srcHeight, int srcDepth,
+          int srcRowStride, const ubyte *srcPtr,
+          int dstWidth, int dstHeight, int dstDepth,
+          int dstRowStride, ubyte *dstPtr)
+{
+   const int bpt = pf_get_size(pformat);
+   const int border = 0;
+   int img, row;
+   int bytesPerSrcImage, bytesPerDstImage;
+   int bytesPerSrcRow, bytesPerDstRow;
+   int srcImageOffset, srcRowOffset;
+   enum dtype datatype;
+   uint comps;
+
+   format_to_type_comps(pformat, &datatype, &comps);
+
+   bytesPerSrcImage = srcWidth * srcHeight * bpt;
+   bytesPerDstImage = dstWidth * dstHeight * bpt;
+
+   bytesPerSrcRow = srcWidth * bpt;
+   bytesPerDstRow = dstWidth * bpt;
+
+   /* Offset between adjacent src images to be averaged together */
+   srcImageOffset = (srcDepth == dstDepth) ? 0 : bytesPerSrcImage;
+
+   /* Offset between adjacent src rows to be averaged together */
+   srcRowOffset = (srcHeight == dstHeight) ? 0 : srcWidth * bpt;
+
+   /*
+    * Need to average together up to 8 src pixels for each dest pixel.
+    * Break that down into 3 operations:
+    *   1. take two rows from source image and average them together.
+    *   2. take two rows from next source image and average them together.
+    *   3. take the two averaged rows and average them for the final dst row.
+    */
+
+   /*
+   _mesa_printf("mip3d %d x %d x %d  ->  %d x %d x %d\n",
+          srcWidth, srcHeight, srcDepth, dstWidth, dstHeight, dstDepth);
+   */
+
+   for (img = 0; img < dstDepth; img++) {
+      /* first source image pointer, skipping border */
+      const ubyte *imgSrcA = srcPtr
+         + (bytesPerSrcImage + bytesPerSrcRow + border) * bpt * border
+         + img * (bytesPerSrcImage + srcImageOffset);
+      /* second source image pointer, skipping border */
+      const ubyte *imgSrcB = imgSrcA + srcImageOffset;
+      /* address of the dest image, skipping border */
+      ubyte *imgDst = dstPtr
+         + (bytesPerDstImage + bytesPerDstRow + border) * bpt * border
+         + img * bytesPerDstImage;
+
+      /* setup the four source row pointers and the dest row pointer */
+      const ubyte *srcImgARowA = imgSrcA;
+      const ubyte *srcImgARowB = imgSrcA + srcRowOffset;
+      const ubyte *srcImgBRowA = imgSrcB;
+      const ubyte *srcImgBRowB = imgSrcB + srcRowOffset;
+      ubyte *dstImgRow = imgDst;
+
+      for (row = 0; row < dstHeight; row++) {
+         do_row_3D(datatype, comps, srcWidth, 
+                   srcImgARowA, srcImgARowB,
+                   srcImgBRowA, srcImgBRowB,
+                   dstWidth, dstImgRow);
+
+         /* advance to next rows */
+         srcImgARowA += bytesPerSrcRow + srcRowOffset;
+         srcImgARowB += bytesPerSrcRow + srcRowOffset;
+         srcImgBRowA += bytesPerSrcRow + srcRowOffset;
+         srcImgBRowB += bytesPerSrcRow + srcRowOffset;
+         dstImgRow += bytesPerDstRow;
+      }
+   }
+}
+
+
+
+
+static void
+make_1d_mipmap(struct gen_mipmap_state *ctx,
+               struct pipe_texture *pt,
+               uint face, uint baseLevel, uint lastLevel)
+{
+   struct pipe_context *pipe = ctx->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   const uint zslice = 0;
+   uint dstLevel;
+
+   for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
+      const uint srcLevel = dstLevel - 1;
+      struct pipe_surface *srcSurf, *dstSurf;
+      void *srcMap, *dstMap;
+      
+      srcSurf = screen->get_tex_surface(screen, pt, face, srcLevel, zslice,
+                                        PIPE_BUFFER_USAGE_CPU_READ);
+
+      dstSurf = screen->get_tex_surface(screen, pt, face, dstLevel, zslice,
+                                        PIPE_BUFFER_USAGE_CPU_WRITE);
+
+      srcMap = ((ubyte *) pipe_surface_map(srcSurf,
+                                           PIPE_BUFFER_USAGE_CPU_READ)
+                + srcSurf->offset);
+      dstMap = ((ubyte *) pipe_surface_map(dstSurf,
+                                           PIPE_BUFFER_USAGE_CPU_WRITE)
+                + dstSurf->offset);
+
+      reduce_1d(pt->format,
+                srcSurf->width, srcMap,
+                dstSurf->width, dstMap);
+
+      pipe_surface_unmap(srcSurf);
+      pipe_surface_unmap(dstSurf);
+
+      pipe_surface_reference(&srcSurf, NULL);
+      pipe_surface_reference(&dstSurf, NULL);
+   }
+}
+
+
+static void
+make_2d_mipmap(struct gen_mipmap_state *ctx,
+               struct pipe_texture *pt,
+               uint face, uint baseLevel, uint lastLevel)
+{
+   struct pipe_context *pipe = ctx->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   const uint zslice = 0;
+   uint dstLevel;
+   
+   assert(pt->block.width == 1);
+   assert(pt->block.height == 1);
+
+   for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
+      const uint srcLevel = dstLevel - 1;
+      struct pipe_surface *srcSurf, *dstSurf;
+      ubyte *srcMap, *dstMap;
+      
+      srcSurf = screen->get_tex_surface(screen, pt, face, srcLevel, zslice,
+                                        PIPE_BUFFER_USAGE_CPU_READ);
+      dstSurf = screen->get_tex_surface(screen, pt, face, dstLevel, zslice,
+                                        PIPE_BUFFER_USAGE_CPU_WRITE);
+
+      srcMap = ((ubyte *) pipe_surface_map(srcSurf,
+                                           PIPE_BUFFER_USAGE_CPU_READ)
+                + srcSurf->offset);
+      dstMap = ((ubyte *) pipe_surface_map(dstSurf,
+                                           PIPE_BUFFER_USAGE_CPU_WRITE)
+                + dstSurf->offset);
+
+      reduce_2d(pt->format,
+                srcSurf->width, srcSurf->height,
+                srcSurf->stride, srcMap,
+                dstSurf->width, dstSurf->height,
+                dstSurf->stride, dstMap);
+
+      pipe_surface_unmap(srcSurf);
+      pipe_surface_unmap(dstSurf);
+
+      pipe_surface_reference(&srcSurf, NULL);
+      pipe_surface_reference(&dstSurf, NULL);
+   }
+}
+
+
+static void
+make_3d_mipmap(struct gen_mipmap_state *ctx,
+               struct pipe_texture *pt,
+               uint face, uint baseLevel, uint lastLevel)
+{
+   struct pipe_context *pipe = ctx->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   uint dstLevel, zslice = 0;
+
+   assert(pt->block.width == 1);
+   assert(pt->block.height == 1);
+
+   for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
+      const uint srcLevel = dstLevel - 1;
+      struct pipe_surface *srcSurf, *dstSurf;
+      ubyte *srcMap, *dstMap;
+      
+      srcSurf = screen->get_tex_surface(screen, pt, face, srcLevel, zslice,
+                                        PIPE_BUFFER_USAGE_CPU_READ);
+      dstSurf = screen->get_tex_surface(screen, pt, face, dstLevel, zslice,
+                                        PIPE_BUFFER_USAGE_CPU_WRITE);
+
+      srcMap = ((ubyte *) pipe_surface_map(srcSurf,
+                                           PIPE_BUFFER_USAGE_CPU_READ)
+                + srcSurf->offset);
+      dstMap = ((ubyte *) pipe_surface_map(dstSurf,
+                                           PIPE_BUFFER_USAGE_CPU_WRITE)
+                + dstSurf->offset);
+
+#if 0
+      reduce_3d(pt->format,
+                srcSurf->width, srcSurf->height,
+                srcSurf->stride, srcMap,
+                dstSurf->width, dstSurf->height,
+                dstSurf->stride, dstMap);
+#else
+      (void) reduce_3d;
+#endif
+
+      pipe_surface_unmap(srcSurf);
+      pipe_surface_unmap(dstSurf);
+
+      pipe_surface_reference(&srcSurf, NULL);
+      pipe_surface_reference(&dstSurf, NULL);
+   }
+}
+
+
+static void
+fallback_gen_mipmap(struct gen_mipmap_state *ctx,
+                    struct pipe_texture *pt,
+                    uint face, uint baseLevel, uint lastLevel)
+{
+   switch (pt->target) {
+   case PIPE_TEXTURE_1D:
+      make_1d_mipmap(ctx, pt, face, baseLevel, lastLevel);
+      break;
+   case PIPE_TEXTURE_2D:
+   case PIPE_TEXTURE_CUBE:
+      make_2d_mipmap(ctx, pt, face, baseLevel, lastLevel);
+      break;
+   case PIPE_TEXTURE_3D:
+      make_3d_mipmap(ctx, pt, face, baseLevel, lastLevel);
+      break;
+   default:
+      assert(0);
+   }
+}
+
+
+/**
+ * Create a mipmap generation context.
+ * The idea is to create one of these and re-use it each time we need to
+ * generate a mipmap.
+ */
+struct gen_mipmap_state *
+util_create_gen_mipmap(struct pipe_context *pipe,
+                       struct cso_context *cso)
+{
+   struct gen_mipmap_state *ctx;
+   uint i;
+
+   ctx = CALLOC_STRUCT(gen_mipmap_state);
+   if (!ctx)
+      return NULL;
+
+   ctx->pipe = pipe;
+   ctx->cso = cso;
+
+   /* disabled blending/masking */
+   memset(&ctx->blend, 0, sizeof(ctx->blend));
+   ctx->blend.rgb_src_factor = PIPE_BLENDFACTOR_ONE;
+   ctx->blend.alpha_src_factor = PIPE_BLENDFACTOR_ONE;
+   ctx->blend.rgb_dst_factor = PIPE_BLENDFACTOR_ZERO;
+   ctx->blend.alpha_dst_factor = PIPE_BLENDFACTOR_ZERO;
+   ctx->blend.colormask = PIPE_MASK_RGBA;
+
+   /* no-op depth/stencil/alpha */
+   memset(&ctx->depthstencil, 0, sizeof(ctx->depthstencil));
+
+   /* rasterizer */
+   memset(&ctx->rasterizer, 0, sizeof(ctx->rasterizer));
+   ctx->rasterizer.front_winding = PIPE_WINDING_CW;
+   ctx->rasterizer.cull_mode = PIPE_WINDING_NONE;
+   ctx->rasterizer.bypass_clipping = 1;
+   /*ctx->rasterizer.bypass_vs = 1;*/
+   ctx->rasterizer.gl_rasterization_rules = 1;
+
+   /* sampler state */
+   memset(&ctx->sampler, 0, sizeof(ctx->sampler));
+   ctx->sampler.wrap_s = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   ctx->sampler.wrap_t = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   ctx->sampler.wrap_r = PIPE_TEX_WRAP_CLAMP_TO_EDGE;
+   ctx->sampler.min_mip_filter = PIPE_TEX_MIPFILTER_NEAREST;
+   ctx->sampler.normalized_coords = 1;
+
+   /* viewport state (identity, verts are in wincoords) */
+   ctx->viewport.scale[0] = 1.0;
+   ctx->viewport.scale[1] = 1.0;
+   ctx->viewport.scale[2] = 1.0;
+   ctx->viewport.scale[3] = 1.0;
+   ctx->viewport.translate[0] = 0.0;
+   ctx->viewport.translate[1] = 0.0;
+   ctx->viewport.translate[2] = 0.0;
+   ctx->viewport.translate[3] = 0.0;
+
+   /* vertex shader */
+   {
+      const uint semantic_names[] = { TGSI_SEMANTIC_POSITION,
+                                      TGSI_SEMANTIC_GENERIC };
+      const uint semantic_indexes[] = { 0, 0 };
+      ctx->vs = util_make_vertex_passthrough_shader(pipe, 2, semantic_names,
+                                                    semantic_indexes,
+                                                    &ctx->vert_shader);
+   }
+
+   /* fragment shader */
+   ctx->fs = util_make_fragment_tex_shader(pipe, &ctx->frag_shader);
+
+   /* vertex data that doesn't change */
+   for (i = 0; i < 4; i++) {
+      ctx->vertices[i][0][2] = 0.0f; /* z */
+      ctx->vertices[i][0][3] = 1.0f; /* w */
+      ctx->vertices[i][1][2] = 0.0f; /* r */
+      ctx->vertices[i][1][3] = 1.0f; /* q */
+   }
+
+   /* Note: the actual vertex buffer is allocated as needed below */
+
+   return ctx;
+}
+
+
+/**
+ * Get next "slot" of vertex space in the vertex buffer.
+ * We're allocating one large vertex buffer and using it piece by piece.
+ */
+static unsigned
+get_next_slot(struct gen_mipmap_state *ctx)
+{
+   const unsigned max_slots = 4096 / sizeof ctx->vertices;
+
+   if (ctx->vbuf_slot >= max_slots) 
+      util_gen_mipmap_flush( ctx );
+
+   if (!ctx->vbuf) {
+      ctx->vbuf = pipe_buffer_create(ctx->pipe->screen,
+                                     32,
+                                     PIPE_BUFFER_USAGE_VERTEX,
+                                     max_slots * sizeof ctx->vertices);
+   }
+   
+   return ctx->vbuf_slot++ * sizeof ctx->vertices;
+}
+
+
+static unsigned
+set_vertex_data(struct gen_mipmap_state *ctx, float width, float height)
+{
+   void *buf;
+   unsigned offset;
+
+   ctx->vertices[0][0][0] = 0.0f; /*x*/
+   ctx->vertices[0][0][1] = 0.0f; /*y*/
+   ctx->vertices[0][1][0] = 0.0f; /*s*/
+   ctx->vertices[0][1][1] = 0.0f; /*t*/
+
+   ctx->vertices[1][0][0] = width;
+   ctx->vertices[1][0][1] = 0.0f;
+   ctx->vertices[1][1][0] = 1.0f;
+   ctx->vertices[1][1][1] = 0.0f;
+
+   ctx->vertices[2][0][0] = width;
+   ctx->vertices[2][0][1] = height;
+   ctx->vertices[2][1][0] = 1.0f;
+   ctx->vertices[2][1][1] = 1.0f;
+
+   ctx->vertices[3][0][0] = 0.0f;
+   ctx->vertices[3][0][1] = height;
+   ctx->vertices[3][1][0] = 0.0f;
+   ctx->vertices[3][1][1] = 1.0f;
+
+   offset = get_next_slot( ctx );
+
+   buf = pipe_buffer_map(ctx->pipe->screen, ctx->vbuf,
+                         PIPE_BUFFER_USAGE_CPU_WRITE);
+
+   memcpy((char *)buf + offset, ctx->vertices, sizeof(ctx->vertices));
+
+   pipe_buffer_unmap(ctx->pipe->screen, ctx->vbuf);
+
+   return offset;
+}
+
+
+
+/**
+ * Destroy a mipmap generation context
+ */
+void
+util_destroy_gen_mipmap(struct gen_mipmap_state *ctx)
+{
+   struct pipe_context *pipe = ctx->pipe;
+
+   pipe->delete_vs_state(pipe, ctx->vs);
+   pipe->delete_fs_state(pipe, ctx->fs);
+
+   FREE((void*) ctx->vert_shader.tokens);
+   FREE((void*) ctx->frag_shader.tokens);
+
+   pipe_buffer_reference(pipe->screen, &ctx->vbuf, NULL);
+
+   FREE(ctx);
+}
+
+
+
+/* Release vertex buffer at end of frame to avoid synchronous
+ * rendering.
+ */
+void util_gen_mipmap_flush( struct gen_mipmap_state *ctx )
+{
+   pipe_buffer_reference(ctx->pipe->screen, &ctx->vbuf, NULL);
+   ctx->vbuf_slot = 0;
+} 
+
+
+/**
+ * Generate mipmap images.  It's assumed all needed texture memory is
+ * already allocated.
+ *
+ * \param pt  the texture to generate mipmap levels for
+ * \param face  which cube face to generate mipmaps for (0 for non-cube maps)
+ * \param baseLevel  the first mipmap level to use as a src
+ * \param lastLevel  the last mipmap level to generate
+ * \param filter  the minification filter used to generate mipmap levels with
+ * \param filter  one of PIPE_TEX_FILTER_LINEAR, PIPE_TEX_FILTER_NEAREST
+ */
+void
+util_gen_mipmap(struct gen_mipmap_state *ctx,
+                struct pipe_texture *pt,
+                uint face, uint baseLevel, uint lastLevel, uint filter)
+{
+   struct pipe_context *pipe = ctx->pipe;
+   struct pipe_screen *screen = pipe->screen;
+   struct pipe_framebuffer_state fb;
+   uint dstLevel;
+   uint zslice = 0;
+   uint offset;
+
+   /* check if we can render in the texture's format */
+   if (!screen->is_format_supported(screen, pt->format, PIPE_TEXTURE_2D,
+                                    PIPE_TEXTURE_USAGE_RENDER_TARGET, 0)) {
+      fallback_gen_mipmap(ctx, pt, face, baseLevel, lastLevel);
+      return;
+   }
+
+   /* save state (restored below) */
+   cso_save_blend(ctx->cso);
+   cso_save_depth_stencil_alpha(ctx->cso);
+   cso_save_rasterizer(ctx->cso);
+   cso_save_samplers(ctx->cso);
+   cso_save_sampler_textures(ctx->cso);
+   cso_save_framebuffer(ctx->cso);
+   cso_save_fragment_shader(ctx->cso);
+   cso_save_vertex_shader(ctx->cso);
+   cso_save_viewport(ctx->cso);
+
+   /* bind our state */
+   cso_set_blend(ctx->cso, &ctx->blend);
+   cso_set_depth_stencil_alpha(ctx->cso, &ctx->depthstencil);
+   cso_set_rasterizer(ctx->cso, &ctx->rasterizer);
+   cso_set_viewport(ctx->cso, &ctx->viewport);
+
+   cso_set_fragment_shader_handle(ctx->cso, ctx->fs);
+   cso_set_vertex_shader_handle(ctx->cso, ctx->vs);
+
+   /* init framebuffer state */
+   memset(&fb, 0, sizeof(fb));
+   fb.nr_cbufs = 1;
+
+   /* set min/mag to same filter for faster sw speed */
+   ctx->sampler.mag_img_filter = filter;
+   ctx->sampler.min_img_filter = filter;
+
+   /*
+    * XXX for small mipmap levels, it may be faster to use the software
+    * fallback path...
+    */
+   for (dstLevel = baseLevel + 1; dstLevel <= lastLevel; dstLevel++) {
+      const uint srcLevel = dstLevel - 1;
+
+      struct pipe_surface *surf = 
+         screen->get_tex_surface(screen, pt, face, dstLevel, zslice,
+                                 PIPE_BUFFER_USAGE_GPU_WRITE);
+
+      /*
+       * Setup framebuffer / dest surface
+       */
+      fb.cbufs[0] = surf;
+      fb.width = pt->width[dstLevel];
+      fb.height = pt->height[dstLevel];
+      cso_set_framebuffer(ctx->cso, &fb);
+
+      /*
+       * Setup sampler state
+       * Note: we should only have to set the min/max LOD clamps to ensure
+       * we grab texels from the right mipmap level.  But some hardware
+       * has trouble with min clamping so we also set the lod_bias to
+       * try to work around that.
+       */
+      ctx->sampler.min_lod = ctx->sampler.max_lod = (float) srcLevel;
+      ctx->sampler.lod_bias = (float) srcLevel;
+      cso_single_sampler(ctx->cso, 0, &ctx->sampler);
+      cso_single_sampler_done(ctx->cso);
+
+      cso_set_sampler_textures(ctx->cso, 1, &pt);
+
+      /* quad coords in window coords (bypassing clipping, viewport mapping) */
+      offset = set_vertex_data(ctx,
+                               (float) pt->width[dstLevel],
+                               (float) pt->height[dstLevel]);
+
+      util_draw_vertex_buffer(ctx->pipe, 
+                              ctx->vbuf,
+                              offset,
+                              PIPE_PRIM_TRIANGLE_FAN,
+                              4,  /* verts */
+                              2); /* attribs/vert */
+
+      pipe->flush(pipe, PIPE_FLUSH_RENDER_CACHE, NULL);
+
+      /* need to signal that the texture has changed _after_ rendering to it */
+      pipe_surface_reference( &surf, NULL );
+   }
+
+   /* restore state we changed */
+   cso_restore_blend(ctx->cso);
+   cso_restore_depth_stencil_alpha(ctx->cso);
+   cso_restore_rasterizer(ctx->cso);
+   cso_restore_samplers(ctx->cso);
+   cso_restore_sampler_textures(ctx->cso);
+   cso_restore_framebuffer(ctx->cso);
+   cso_restore_fragment_shader(ctx->cso);
+   cso_restore_vertex_shader(ctx->cso);
+   cso_restore_viewport(ctx->cso);
+}
diff --git a/src/gallium/auxiliary/util/u_gen_mipmap.h b/src/gallium/auxiliary/util/u_gen_mipmap.h
new file mode 100644
index 00000000000..54608f9466d
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_gen_mipmap.h
@@ -0,0 +1,70 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#ifndef U_GENMIPMAP_H
+#define U_GENMIPMAP_H
+
+#include "pipe/p_state.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+   
+struct pipe_context;
+struct pipe_texture;
+struct cso_context;
+
+struct gen_mipmap_state;
+
+
+extern struct gen_mipmap_state *
+util_create_gen_mipmap(struct pipe_context *pipe, struct cso_context *cso);
+
+
+extern void
+util_destroy_gen_mipmap(struct gen_mipmap_state *ctx);
+
+/* Release vertex buffer at end of frame to avoid synchronous
+ * rendering.
+ */
+extern void 
+util_gen_mipmap_flush( struct gen_mipmap_state *ctx );
+
+
+extern void
+util_gen_mipmap(struct gen_mipmap_state *ctx,
+                struct pipe_texture *pt,
+                uint face, uint baseLevel, uint lastLevel, uint filter);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_handle_table.c b/src/gallium/auxiliary/util/u_handle_table.c
new file mode 100644
index 00000000000..2d15932ce3b
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_handle_table.c
@@ -0,0 +1,293 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Generic handle table implementation.
+ *  
+ * @author José Fonseca <[email protected]>
+ */
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+
+#include "util/u_memory.h"
+#include "util/u_handle_table.h"
+
+
+#define HANDLE_TABLE_INITIAL_SIZE 16  
+
+
+struct handle_table
+{
+   /** Object array. Empty handles have a null object */
+   void **objects;
+   
+   /** Number of objects the handle can currently hold */
+   unsigned size;
+   /** Number of consecutive objects allocated at the start of the table */
+   unsigned filled;
+   
+   /** Optional object destructor */
+   void (*destroy)(void *object);
+};
+
+
+struct handle_table *
+handle_table_create(void)
+{
+   struct handle_table *ht;
+   
+   ht = MALLOC_STRUCT(handle_table);
+   if(!ht)
+      return NULL;
+   
+   ht->objects = (void **)CALLOC(HANDLE_TABLE_INITIAL_SIZE, sizeof(void *));
+   if(!ht->objects) {
+      FREE(ht);
+      return NULL;
+   }
+   
+   ht->size = HANDLE_TABLE_INITIAL_SIZE;
+   ht->filled = 0;
+   
+   ht->destroy = NULL;
+   
+   return ht;
+}
+
+
+void
+handle_table_set_destroy(struct handle_table *ht,
+                         void (*destroy)(void *object))
+{
+   assert(ht);
+   ht->destroy = destroy;
+}
+
+
+/**
+ * Resize the table if necessary 
+ */
+static INLINE int
+handle_table_resize(struct handle_table *ht,
+                    unsigned minimum_size)
+{
+   unsigned new_size;
+   void **new_objects;
+
+   if(ht->size > minimum_size)
+      return ht->size;
+
+   new_size = ht->size;
+   while(!(new_size > minimum_size))
+      new_size *= 2;
+   assert(new_size);
+   
+   new_objects = (void **)REALLOC((void *)ht->objects,
+				  ht->size*sizeof(void *),
+				  new_size*sizeof(void *));
+   if(!new_objects)
+      return 0;
+   
+   memset(new_objects + ht->size, 0, (new_size - ht->size)*sizeof(void *));
+   
+   ht->size = new_size;
+   ht->objects = new_objects;
+   
+   return ht->size;
+}
+
+
+static INLINE void
+handle_table_clear(struct handle_table *ht, 
+                   unsigned index)
+{
+   void *object;
+   
+   /* The order here is important so that the object being destroyed is not
+    * present in the table when seen by the destroy callback, because the 
+    * destroy callback may directly or indirectly call the other functions in 
+    * this module.
+    */
+
+   object = ht->objects[index];
+   if(object) {
+      ht->objects[index] = NULL;
+      
+      if(ht->destroy)
+         ht->destroy(object);
+   }   
+}
+
+
+unsigned
+handle_table_add(struct handle_table *ht, 
+                 void *object)
+{
+   unsigned index;
+   unsigned handle;
+   
+   assert(ht);
+   assert(object);
+   if(!object)
+      return 0;
+
+   /* linear search for an empty handle */
+   while(ht->filled < ht->size) {
+      if(!ht->objects[ht->filled])
+	 break;
+      ++ht->filled;
+   }
+  
+   index = ht->filled;
+   handle = index + 1;
+   
+   /* check integer overflow */
+   if(!handle)
+      return 0;
+   
+   /* grow the table if necessary */
+   if(!handle_table_resize(ht, index))
+      return 0;
+
+   assert(!ht->objects[index]);
+   ht->objects[index] = object;
+   ++ht->filled;
+   
+   return handle;
+}
+
+
+unsigned
+handle_table_set(struct handle_table *ht, 
+                 unsigned handle,
+                 void *object)
+{
+   unsigned index;
+   
+   assert(ht);
+   assert(handle);
+   if(!handle)
+      return 0;
+
+   assert(object);
+   if(!object)
+      return 0;
+   
+   index = handle - 1;
+
+   /* grow the table if necessary */
+   if(!handle_table_resize(ht, index))
+      return 0;
+
+   handle_table_clear(ht, index);
+
+   ht->objects[index] = object;
+   
+   return handle;
+}
+
+
+void *
+handle_table_get(struct handle_table *ht, 
+                 unsigned handle)
+{
+   void *object;
+   
+   assert(ht);
+   assert(handle);
+   if(!handle || handle > ht->size)
+      return NULL;
+
+   object = ht->objects[handle - 1];
+   
+   return object;
+}
+
+
+void
+handle_table_remove(struct handle_table *ht, 
+                    unsigned handle)
+{
+   void *object;
+   unsigned index;
+   
+   assert(ht);
+   assert(handle);
+   if(!handle || handle > ht->size)
+      return;
+
+   index = handle - 1;
+   object = ht->objects[index];
+   if(!object)
+      return;
+   
+   handle_table_clear(ht, index);
+
+   if(index < ht->filled)
+      ht->filled = index;
+}
+
+
+unsigned
+handle_table_get_next_handle(struct handle_table *ht, 
+                             unsigned handle)
+{
+   unsigned index;
+   
+   for(index = handle; index < ht->size; ++index) {
+      if(ht->objects[index])
+	 return index + 1;
+   }
+
+   return 0;
+}
+
+
+unsigned
+handle_table_get_first_handle(struct handle_table *ht)
+{
+   return handle_table_get_next_handle(ht, 0);
+}
+
+
+void
+handle_table_destroy(struct handle_table *ht)
+{
+   unsigned index;
+   assert(ht);
+
+   if(ht->destroy)
+      for(index = 0; index < ht->size; ++index)
+         handle_table_clear(ht, index);
+   
+   FREE(ht->objects);
+   FREE(ht);
+}
+
diff --git a/src/gallium/auxiliary/util/u_handle_table.h b/src/gallium/auxiliary/util/u_handle_table.h
new file mode 100644
index 00000000000..d080135c9f4
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_handle_table.h
@@ -0,0 +1,116 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Generic handle table.
+ *  
+ * @author José Fonseca <[email protected]>
+ */
+
+#ifndef U_HANDLE_TABLE_H_
+#define U_HANDLE_TABLE_H_
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+   
+/**
+ * Abstract data type to map integer handles to objects.
+ * 
+ * Also referred as "pointer array".
+ */
+struct handle_table;
+
+
+struct handle_table *
+handle_table_create(void);
+
+
+/**
+ * Set an optional destructor callback.
+ * 
+ * If set, it will be called during handle_table_remove and 
+ * handle_table_destroy calls.
+ */
+void
+handle_table_set_destroy(struct handle_table *ht,
+                         void (*destroy)(void *object));
+
+
+/**
+ * Add a new object.
+ * 
+ * Returns a zero handle on failure (out of memory).
+ */
+unsigned
+handle_table_add(struct handle_table *ht, 
+                 void *object);
+
+/**
+ * Returns zero on failure (out of memory).
+ */
+unsigned
+handle_table_set(struct handle_table *ht, 
+                 unsigned handle,
+                 void *object);
+
+/**
+ * Fetch an existing object.
+ * 
+ * Returns NULL for an invalid handle.
+ */
+void *
+handle_table_get(struct handle_table *ht, 
+                 unsigned handle);
+
+
+void
+handle_table_remove(struct handle_table *ht, 
+                    unsigned handle);
+
+
+void
+handle_table_destroy(struct handle_table *ht);
+
+
+unsigned
+handle_table_get_first_handle(struct handle_table *ht);
+
+
+unsigned
+handle_table_get_next_handle(struct handle_table *ht,
+                             unsigned handle);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* U_HANDLE_TABLE_H_ */
diff --git a/src/gallium/auxiliary/util/u_hash.c b/src/gallium/auxiliary/util/u_hash.c
new file mode 100644
index 00000000000..b67653ec708
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_hash.c
@@ -0,0 +1,121 @@
+/**************************************************************************
+ *
+ * Copyright 2008 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Hash functions implementation.
+ * 
+ * @author Jose Fonseca
+ */
+
+
+#include "u_hash.h"
+
+
+static const uint32_t 
+util_crc32_table[256] = {
+   0x00000000, 0x77073096, 0xee0e612c, 0x990951ba, 
+   0x076dc419, 0x706af48f, 0xe963a535, 0x9e6495a3, 
+   0x0edb8832, 0x79dcb8a4, 0xe0d5e91e, 0x97d2d988, 
+   0x09b64c2b, 0x7eb17cbd, 0xe7b82d07, 0x90bf1d91, 
+   0x1db71064, 0x6ab020f2, 0xf3b97148, 0x84be41de, 
+   0x1adad47d, 0x6ddde4eb, 0xf4d4b551, 0x83d385c7, 
+   0x136c9856, 0x646ba8c0, 0xfd62f97a, 0x8a65c9ec, 
+   0x14015c4f, 0x63066cd9, 0xfa0f3d63, 0x8d080df5, 
+   0x3b6e20c8, 0x4c69105e, 0xd56041e4, 0xa2677172, 
+   0x3c03e4d1, 0x4b04d447, 0xd20d85fd, 0xa50ab56b, 
+   0x35b5a8fa, 0x42b2986c, 0xdbbbc9d6, 0xacbcf940, 
+   0x32d86ce3, 0x45df5c75, 0xdcd60dcf, 0xabd13d59, 
+   0x26d930ac, 0x51de003a, 0xc8d75180, 0xbfd06116, 
+   0x21b4f4b5, 0x56b3c423, 0xcfba9599, 0xb8bda50f, 
+   0x2802b89e, 0x5f058808, 0xc60cd9b2, 0xb10be924, 
+   0x2f6f7c87, 0x58684c11, 0xc1611dab, 0xb6662d3d, 
+   0x76dc4190, 0x01db7106, 0x98d220bc, 0xefd5102a, 
+   0x71b18589, 0x06b6b51f, 0x9fbfe4a5, 0xe8b8d433, 
+   0x7807c9a2, 0x0f00f934, 0x9609a88e, 0xe10e9818, 
+   0x7f6a0dbb, 0x086d3d2d, 0x91646c97, 0xe6635c01, 
+   0x6b6b51f4, 0x1c6c6162, 0x856530d8, 0xf262004e, 
+   0x6c0695ed, 0x1b01a57b, 0x8208f4c1, 0xf50fc457, 
+   0x65b0d9c6, 0x12b7e950, 0x8bbeb8ea, 0xfcb9887c, 
+   0x62dd1ddf, 0x15da2d49, 0x8cd37cf3, 0xfbd44c65, 
+   0x4db26158, 0x3ab551ce, 0xa3bc0074, 0xd4bb30e2, 
+   0x4adfa541, 0x3dd895d7, 0xa4d1c46d, 0xd3d6f4fb, 
+   0x4369e96a, 0x346ed9fc, 0xad678846, 0xda60b8d0, 
+   0x44042d73, 0x33031de5, 0xaa0a4c5f, 0xdd0d7cc9, 
+   0x5005713c, 0x270241aa, 0xbe0b1010, 0xc90c2086, 
+   0x5768b525, 0x206f85b3, 0xb966d409, 0xce61e49f, 
+   0x5edef90e, 0x29d9c998, 0xb0d09822, 0xc7d7a8b4, 
+   0x59b33d17, 0x2eb40d81, 0xb7bd5c3b, 0xc0ba6cad, 
+   0xedb88320, 0x9abfb3b6, 0x03b6e20c, 0x74b1d29a, 
+   0xead54739, 0x9dd277af, 0x04db2615, 0x73dc1683, 
+   0xe3630b12, 0x94643b84, 0x0d6d6a3e, 0x7a6a5aa8, 
+   0xe40ecf0b, 0x9309ff9d, 0x0a00ae27, 0x7d079eb1, 
+   0xf00f9344, 0x8708a3d2, 0x1e01f268, 0x6906c2fe, 
+   0xf762575d, 0x806567cb, 0x196c3671, 0x6e6b06e7, 
+   0xfed41b76, 0x89d32be0, 0x10da7a5a, 0x67dd4acc, 
+   0xf9b9df6f, 0x8ebeeff9, 0x17b7be43, 0x60b08ed5, 
+   0xd6d6a3e8, 0xa1d1937e, 0x38d8c2c4, 0x4fdff252, 
+   0xd1bb67f1, 0xa6bc5767, 0x3fb506dd, 0x48b2364b, 
+   0xd80d2bda, 0xaf0a1b4c, 0x36034af6, 0x41047a60, 
+   0xdf60efc3, 0xa867df55, 0x316e8eef, 0x4669be79, 
+   0xcb61b38c, 0xbc66831a, 0x256fd2a0, 0x5268e236, 
+   0xcc0c7795, 0xbb0b4703, 0x220216b9, 0x5505262f, 
+   0xc5ba3bbe, 0xb2bd0b28, 0x2bb45a92, 0x5cb36a04, 
+   0xc2d7ffa7, 0xb5d0cf31, 0x2cd99e8b, 0x5bdeae1d, 
+   0x9b64c2b0, 0xec63f226, 0x756aa39c, 0x026d930a, 
+   0x9c0906a9, 0xeb0e363f, 0x72076785, 0x05005713, 
+   0x95bf4a82, 0xe2b87a14, 0x7bb12bae, 0x0cb61b38, 
+   0x92d28e9b, 0xe5d5be0d, 0x7cdcefb7, 0x0bdbdf21, 
+   0x86d3d2d4, 0xf1d4e242, 0x68ddb3f8, 0x1fda836e, 
+   0x81be16cd, 0xf6b9265b, 0x6fb077e1, 0x18b74777, 
+   0x88085ae6, 0xff0f6a70, 0x66063bca, 0x11010b5c, 
+   0x8f659eff, 0xf862ae69, 0x616bffd3, 0x166ccf45, 
+   0xa00ae278, 0xd70dd2ee, 0x4e048354, 0x3903b3c2, 
+   0xa7672661, 0xd06016f7, 0x4969474d, 0x3e6e77db, 
+   0xaed16a4a, 0xd9d65adc, 0x40df0b66, 0x37d83bf0, 
+   0xa9bcae53, 0xdebb9ec5, 0x47b2cf7f, 0x30b5ffe9, 
+   0xbdbdf21c, 0xcabac28a, 0x53b39330, 0x24b4a3a6, 
+   0xbad03605, 0xcdd70693, 0x54de5729, 0x23d967bf, 
+   0xb3667a2e, 0xc4614ab8, 0x5d681b02, 0x2a6f2b94, 
+   0xb40bbe37, 0xc30c8ea1, 0x5a05df1b, 0x2d02ef8d
+};
+
+
+/**
+ * @sa http://www.w3.org/TR/PNG/#D-CRCAppendix
+ */
+uint32_t
+util_hash_crc32(const void *data, size_t size)
+{
+   uint8_t *p = (uint8_t *)data;
+   uint32_t crc = 0xffffffff;
+ 
+   while (size--)
+      crc = util_crc32_table[(crc ^ *p++) & 0xff] ^ (crc >> 8);
+   
+   return crc;
+}
diff --git a/src/gallium/auxiliary/util/u_hash.h b/src/gallium/auxiliary/util/u_hash.h
new file mode 100644
index 00000000000..8d92b07c85e
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_hash.h
@@ -0,0 +1,55 @@
+/**************************************************************************
+ *
+ * Copyright 2008 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL VMWARE AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Hash functions.
+ * 
+ * @author Jose Fonseca <[email protected]>
+ */
+
+#ifndef U_HASH_H_
+#define U_HASH_H_
+
+
+#include "pipe/p_compiler.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+   
+uint32_t
+util_hash_crc32(const void *data, size_t size);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* U_HASH_H_ */
diff --git a/src/gallium/auxiliary/util/u_hash_table.c b/src/gallium/auxiliary/util/u_hash_table.c
new file mode 100644
index 00000000000..0bc8de9632c
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_hash_table.c
@@ -0,0 +1,280 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * General purpose hash table implementation.
+ * 
+ * Just uses the cso_hash for now, but it might be better switch to a linear 
+ * probing hash table implementation at some point -- as it is said they have 
+ * better lookup and cache performance and it appears to be possible to write 
+ * a lock-free implementation of such hash tables . 
+ * 
+ * @author José Fonseca <[email protected]>
+ */
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+
+#include "cso_cache/cso_hash.h"
+
+#include "util/u_memory.h"
+#include "util/u_hash_table.h"
+
+
+struct hash_table
+{
+   struct cso_hash *cso;   
+   
+   /** Hash function */
+   unsigned (*hash)(void *key);
+   
+   /** Compare two keys */
+   int (*compare)(void *key1, void *key2);
+   
+   /* TODO: key, value destructors? */
+};
+
+
+struct hash_table_item
+{
+   void *key;
+   void *value;
+};
+
+
+static INLINE struct hash_table_item *
+hash_table_item(struct cso_hash_iter iter)
+{
+   return (struct hash_table_item *)cso_hash_iter_data(iter);
+}
+
+
+struct hash_table *
+hash_table_create(unsigned (*hash)(void *key),
+                  int (*compare)(void *key1, void *key2))
+{
+   struct hash_table *ht;
+   
+   ht = MALLOC_STRUCT(hash_table);
+   if(!ht)
+      return NULL;
+   
+   ht->cso = cso_hash_create();
+   if(!ht->cso) {
+      FREE(ht);
+      return NULL;
+   }
+   
+   ht->hash = hash;
+   ht->compare = compare;
+   
+   return ht;
+}
+
+
+static INLINE struct cso_hash_iter
+hash_table_find_iter(struct hash_table *ht,
+                     void *key, 
+                     unsigned key_hash)
+{
+   struct cso_hash_iter iter;
+   struct hash_table_item *item;
+   
+   iter = cso_hash_find(ht->cso, key_hash);
+   while (!cso_hash_iter_is_null(iter)) {
+      item = (struct hash_table_item *)cso_hash_iter_data(iter);
+      if (!ht->compare(item->key, key))
+         break;
+      iter = cso_hash_iter_next(iter);
+   }
+   
+   return iter;
+}
+
+
+static INLINE struct hash_table_item *
+hash_table_find_item(struct hash_table *ht,
+                     void *key, 
+                     unsigned key_hash)
+{
+   struct cso_hash_iter iter;
+   struct hash_table_item *item;
+   
+   iter = cso_hash_find(ht->cso, key_hash);
+   while (!cso_hash_iter_is_null(iter)) {
+      item = (struct hash_table_item *)cso_hash_iter_data(iter);
+      if (!ht->compare(item->key, key))
+         return item;
+      iter = cso_hash_iter_next(iter);
+   }
+   
+   return NULL;
+}
+
+
+enum pipe_error
+hash_table_set(struct hash_table *ht,
+               void *key,
+               void *value)
+{
+   unsigned key_hash;
+   struct hash_table_item *item;
+   struct cso_hash_iter iter;
+
+   assert(ht);
+
+   key_hash = ht->hash(key);
+
+   item = hash_table_find_item(ht, key, key_hash);
+   if(item) {
+      /* TODO: key/value destruction? */
+      item->value = value;
+      return PIPE_OK;
+   }
+   
+   item = MALLOC_STRUCT(hash_table_item);
+   if(!item)
+      return PIPE_ERROR_OUT_OF_MEMORY;
+   
+   item->key = key;
+   item->value = value;
+   
+   iter = cso_hash_insert(ht->cso, key_hash, item);
+   if(cso_hash_iter_is_null(iter)) {
+      FREE(item);
+      return PIPE_ERROR_OUT_OF_MEMORY;
+   }
+
+   return PIPE_OK;
+}
+
+
+void *
+hash_table_get(struct hash_table *ht, 
+               void *key)
+{
+   unsigned key_hash;
+   struct hash_table_item *item;
+
+   assert(ht);
+
+   key_hash = ht->hash(key);
+
+   item = hash_table_find_item(ht, key, key_hash);
+   if(!item)
+      return NULL;
+   
+   return item->value;
+}
+
+
+void
+hash_table_remove(struct hash_table *ht, 
+                  void *key)
+{
+   unsigned key_hash;
+   struct cso_hash_iter iter;
+   struct hash_table_item *item;
+
+   assert(ht);
+
+   key_hash = ht->hash(key);
+
+   iter = hash_table_find_iter(ht, key, key_hash);
+   if(cso_hash_iter_is_null(iter))
+      return;
+   
+   item = hash_table_item(iter);
+   assert(item);
+   FREE(item);
+   
+   cso_hash_erase(ht->cso, iter);
+}
+
+
+void 
+hash_table_clear(struct hash_table *ht)
+{
+   struct cso_hash_iter iter;
+   struct hash_table_item *item;
+
+   assert(ht);
+   
+   iter = cso_hash_first_node(ht->cso);
+   while (!cso_hash_iter_is_null(iter)) {
+      item = (struct hash_table_item *)cso_hash_take(ht->cso, cso_hash_iter_key(iter));
+      FREE(item);
+      iter = cso_hash_first_node(ht->cso);
+   }
+}
+
+
+enum pipe_error
+hash_table_foreach(struct hash_table *ht,
+                   enum pipe_error (*callback)(void *key, void *value, void *data),
+                   void *data)
+{
+   struct cso_hash_iter iter;
+   struct hash_table_item *item;
+   enum pipe_error result;
+   
+   assert(ht);
+   
+   iter = cso_hash_first_node(ht->cso);
+   while (!cso_hash_iter_is_null(iter)) {
+      item = (struct hash_table_item *)cso_hash_iter_data(iter);
+      result = callback(item->key, item->value, data);
+      if(result != PIPE_OK)
+	 return result;
+      iter = cso_hash_iter_next(iter);
+   }
+
+   return PIPE_OK;
+}
+
+
+void
+hash_table_destroy(struct hash_table *ht)
+{
+   struct cso_hash_iter iter;
+   struct hash_table_item *item;
+   
+   assert(ht);
+   
+   iter = cso_hash_first_node(ht->cso);
+   while (!cso_hash_iter_is_null(iter)) {
+      item = (struct hash_table_item *)cso_hash_iter_data(iter);
+      FREE(item);
+      iter = cso_hash_iter_next(iter);
+   }
+
+   cso_hash_delete(ht->cso);
+   
+   FREE(ht);
+}
diff --git a/src/gallium/auxiliary/util/u_hash_table.h b/src/gallium/auxiliary/util/u_hash_table.h
new file mode 100644
index 00000000000..feee881582e
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_hash_table.h
@@ -0,0 +1,95 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * General purpose hash table.
+ *  
+ * @author José Fonseca <[email protected]>
+ */
+
+#ifndef U_HASH_TABLE_H_
+#define U_HASH_TABLE_H_
+
+
+#include "pipe/p_error.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+   
+/**
+ * Generic purpose hash table.
+ */
+struct hash_table;
+
+
+/**
+ * Create an hash table.
+ * 
+ * @param hash hash function
+ * @param compare should return 0 for two equal keys.
+ */
+struct hash_table *
+hash_table_create(unsigned (*hash)(void *key),
+                  int (*compare)(void *key1, void *key2));
+
+
+enum pipe_error
+hash_table_set(struct hash_table *ht,
+               void *key,
+               void *value);
+
+void *
+hash_table_get(struct hash_table *ht, 
+               void *key);
+
+
+void
+hash_table_remove(struct hash_table *ht, 
+                  void *key);
+
+
+void
+hash_table_clear(struct hash_table *ht);
+
+
+enum pipe_error
+hash_table_foreach(struct hash_table *ht,
+                   enum pipe_error (*callback)(void *key, void *value, void *data),
+                   void *data);
+
+void
+hash_table_destroy(struct hash_table *ht);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* U_HASH_TABLE_H_ */
diff --git a/src/gallium/auxiliary/util/u_keymap.c b/src/gallium/auxiliary/util/u_keymap.c
new file mode 100644
index 00000000000..01b17ddb1b3
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_keymap.c
@@ -0,0 +1,309 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * Key lookup/associative container.
+ *
+ * Like Jose's u_hash_table, based on CSO cache code for now.
+ *
+ * Author: Brian Paul
+ */
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_error.h"
+
+#include "cso_cache/cso_hash.h"
+
+#include "util/u_memory.h"
+#include "util/u_keymap.h"
+
+
+struct keymap
+{
+   struct cso_hash *cso;   
+   unsigned key_size;
+   unsigned max_entries; /* XXX not obeyed net */
+   unsigned num_entries;
+   keymap_delete_func delete_func;
+};
+
+
+struct keymap_item
+{
+   void *key, *value;
+};
+
+
+/**
+ * This the default key-delete function used when the client doesn't
+ * provide one.
+ */
+static void
+default_delete_func(const struct keymap *map,
+                    const void *key, void *data, void *user)
+{
+   FREE((void*) data);
+}
+
+
+static INLINE struct keymap_item *
+hash_table_item(struct cso_hash_iter iter)
+{
+   return (struct keymap_item *) cso_hash_iter_data(iter);
+}
+
+
+/**
+ * Return 4-byte hash key for a block of bytes.
+ */
+static unsigned
+hash(const void *key, unsigned keySize)
+{
+   unsigned i, hash;
+
+   keySize /= 4; /* convert from bytes to uints */
+
+   hash = 0;
+   for (i = 0; i < keySize; i++) {
+      hash ^= (i + 1) * ((const unsigned *) key)[i];
+   }
+
+   /*hash = hash ^ (hash >> 11) ^ (hash >> 22);*/
+
+   return hash;
+}
+
+
+/**
+ * Create a new map.
+ * \param keySize  size of the keys in bytes
+ * \param maxEntries  max number of entries to allow (~0 = infinity)
+ * \param deleteFunc  optional callback to call when entries
+ *                    are deleted/replaced
+ */
+struct keymap *
+util_new_keymap(unsigned keySize, unsigned maxEntries,
+                 keymap_delete_func deleteFunc)
+{
+   struct keymap *map = MALLOC_STRUCT(keymap);
+   if (!map)
+      return NULL;
+   
+   map->cso = cso_hash_create();
+   if (!map->cso) {
+      FREE(map);
+      return NULL;
+   }
+   
+   map->max_entries = maxEntries;
+   map->num_entries = 0;
+   map->key_size = keySize;
+   map->delete_func = deleteFunc ? deleteFunc : default_delete_func;
+
+   return map;
+}
+
+
+/**
+ * Delete/free a keymap and all entries.  The deleteFunc that was given at
+ * create time will be called for each entry.
+ * \param user  user-provided pointer passed through to the delete callback
+ */
+void
+util_delete_keymap(struct keymap *map, void *user)
+{
+   util_keymap_remove_all(map, user);
+   cso_hash_delete(map->cso);
+   FREE(map);
+}
+
+
+static INLINE struct cso_hash_iter
+hash_table_find_iter(const struct keymap *map, const void *key,
+                     unsigned key_hash)
+{
+   struct cso_hash_iter iter;
+   struct keymap_item *item;
+   
+   iter = cso_hash_find(map->cso, key_hash);
+   while (!cso_hash_iter_is_null(iter)) {
+      item = (struct keymap_item *) cso_hash_iter_data(iter);
+      if (!memcmp(item->key, key, map->key_size))
+         break;
+      iter = cso_hash_iter_next(iter);
+   }
+   
+   return iter;
+}
+
+
+static INLINE struct keymap_item *
+hash_table_find_item(const struct keymap *map, const void *key,
+                     unsigned key_hash)
+{
+   struct cso_hash_iter iter = hash_table_find_iter(map, key, key_hash);
+   if (cso_hash_iter_is_null(iter)) {
+      return NULL;
+   }
+   else {
+      return hash_table_item(iter);
+   }
+}
+
+
+/**
+ * Insert a new key + data pointer into the table.
+ * Note: we create a copy of the key, but not the data!
+ * If the key is already present in the table, replace the existing
+ * entry (calling the delete callback on the previous entry).
+ * If the maximum capacity of the map is reached an old entry
+ * will be deleted (the delete callback will be called).
+ */
+boolean
+util_keymap_insert(struct keymap *map, const void *key,
+                   const void *data, void *user)
+{
+   unsigned key_hash;
+   struct keymap_item *item;
+   struct cso_hash_iter iter;
+
+   assert(map);
+
+   key_hash = hash(key, map->key_size);
+
+   item = hash_table_find_item(map, key, key_hash);
+   if (item) {
+      /* call delete callback for old entry/item */
+      map->delete_func(map, item->key, item->value, user);
+      item->value = (void *) data;
+      return TRUE;
+   }
+   
+   item = MALLOC_STRUCT(keymap_item);
+   if (!item)
+      return FALSE;
+
+   item->key = mem_dup(key, map->key_size);
+   item->value = (void *) data;
+   
+   iter = cso_hash_insert(map->cso, key_hash, item);
+   if (cso_hash_iter_is_null(iter)) {
+      FREE(item);
+      return FALSE;
+   }
+
+   map->num_entries++;
+
+   return TRUE;
+}
+
+
+/**
+ * Look up a key in the map and return the associated data pointer.
+ */
+const void *
+util_keymap_lookup(const struct keymap *map, const void *key)
+{
+   unsigned key_hash;
+   struct keymap_item *item;
+
+   assert(map);
+
+   key_hash = hash(key, map->key_size);
+
+   item = hash_table_find_item(map, key, key_hash);
+   if (!item)
+      return NULL;
+   
+   return item->value;
+}
+
+
+/**
+ * Remove an entry from the map.
+ * The delete callback will be called if the given key/entry is found.
+ * \param user  passed to the delete callback as the last param.
+ */
+void
+util_keymap_remove(struct keymap *map, const void *key, void *user)
+{
+   unsigned key_hash;
+   struct cso_hash_iter iter;
+   struct keymap_item *item;
+
+   assert(map);
+
+   key_hash = hash(key, map->key_size);
+
+   iter = hash_table_find_iter(map, key, key_hash);
+   if (cso_hash_iter_is_null(iter))
+      return;
+   
+   item = hash_table_item(iter);
+   assert(item);
+   map->delete_func(map, item->key, item->value, user);
+   FREE(item->key);
+   FREE(item);
+   
+   map->num_entries--;
+
+   cso_hash_erase(map->cso, iter);
+}
+
+
+/**
+ * Remove all entries from the map, calling the delete callback for each.
+ * \param user  passed to the delete callback as the last param.
+ */
+void
+util_keymap_remove_all(struct keymap *map, void *user)
+{
+   struct cso_hash_iter iter;
+   struct keymap_item *item;
+
+   assert(map);
+   
+   iter = cso_hash_first_node(map->cso);
+   while (!cso_hash_iter_is_null(iter)) {
+      item = (struct keymap_item *)
+         cso_hash_take(map->cso, cso_hash_iter_key(iter));
+      map->delete_func(map, item->key, item->value, user);
+      FREE(item->key);
+      FREE(item);
+      iter = cso_hash_first_node(map->cso);
+   }
+}
+
+
+extern void
+util_keymap_info(const struct keymap *map)
+{
+   debug_printf("Keymap %p: %u of max %u entries\n",
+                (void *) map, map->num_entries, map->max_entries);
+}
diff --git a/src/gallium/auxiliary/util/u_keymap.h b/src/gallium/auxiliary/util/u_keymap.h
new file mode 100644
index 00000000000..8d60a76fc3c
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_keymap.h
@@ -0,0 +1,68 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef U_KEYMAP_H
+#define U_KEYMAP_H
+
+#include "pipe/p_compiler.h"
+
+
+/** opaque keymap type */
+struct keymap;
+
+
+/** Delete/callback function type */
+typedef void (*keymap_delete_func)(const struct keymap *map,
+                                   const void *key, void *data,
+                                   void *user);
+
+
+extern struct keymap *
+util_new_keymap(unsigned keySize, unsigned maxEntries,
+                keymap_delete_func deleteFunc);
+
+extern void
+util_delete_keymap(struct keymap *map, void *user);
+
+extern boolean
+util_keymap_insert(struct keymap *map, const void *key,
+                   const void *data, void *user);
+
+extern const void *
+util_keymap_lookup(const struct keymap *map, const void *key);
+
+extern void
+util_keymap_remove(struct keymap *map, const void *key, void *user);
+
+extern void
+util_keymap_remove_all(struct keymap *map, void *user);
+
+extern void
+util_keymap_info(const struct keymap *map);
+
+
+#endif /* U_KEYMAP_H */
diff --git a/src/gallium/auxiliary/util/u_linear.c b/src/gallium/auxiliary/util/u_linear.c
new file mode 100644
index 00000000000..e999cefe748
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_linear.c
@@ -0,0 +1,70 @@
+
+#include "pipe/p_debug.h"
+#include "u_linear.h"
+
+void
+pipe_linear_to_tile(size_t src_stride, const void *src_ptr,
+		    struct pipe_tile_info *t, void *dst_ptr)
+{
+   int x, y, z;
+   char *ptr;
+   size_t bytes = t->cols * t->block.size;
+   char *dst_ptr2 = (char *) dst_ptr;
+
+   assert(pipe_linear_check_tile(t));
+
+   /* lets write lineary to the tiled buffer */
+   for (y = 0; y < t->tiles_y; y++) {
+      for (x = 0; x < t->tiles_x; x++) {
+	 /* this inner loop could be replace with SSE magic */
+	 ptr = (char*)src_ptr + src_stride * t->rows * y + bytes * x;
+	 for (z = 0; z < t->rows; z++) {
+	    memcpy(dst_ptr2, ptr, bytes);
+	    dst_ptr2 += bytes;
+	    ptr += src_stride;
+	 }
+      }
+   }
+}
+
+void pipe_linear_from_tile(struct pipe_tile_info *t, const void *src_ptr,
+			   size_t dst_stride, void *dst_ptr)
+{
+   int x, y, z;
+   char *ptr;
+   size_t bytes = t->cols * t->block.size;
+   const char *src_ptr2 = (const char *) src_ptr;
+
+   /* lets read lineary from the tiled buffer */
+   for (y = 0; y < t->tiles_y; y++) {
+      for (x = 0; x < t->tiles_x; x++) {
+	 /* this inner loop could be replace with SSE magic */
+	 ptr = (char*)dst_ptr + dst_stride * t->rows * y + bytes * x;
+	 for (z = 0; z < t->rows; z++) {
+	    memcpy(ptr, src_ptr2, bytes);
+	    src_ptr2 += bytes;
+	    ptr += dst_stride;
+	 }
+      }
+   }
+}
+
+void
+pipe_linear_fill_info(struct pipe_tile_info *t,
+		      const struct pipe_format_block *block,
+		      unsigned tile_width, unsigned tile_height,
+		      unsigned tiles_x, unsigned tiles_y)
+{
+   t->block = *block;
+
+   t->tile.width = tile_width;
+   t->tile.height = tile_height;
+   t->cols = t->tile.width / t->block.width;
+   t->rows = t->tile.height / t->block.height;
+   t->tile.size = t->cols * t->rows * t->block.size;
+
+   t->tiles_x = tiles_x;
+   t->tiles_y = tiles_y;
+   t->stride = t->cols * t->tiles_x * t->block.size;
+   t->size = t->tiles_x * t->tiles_y * t->tile.size;
+}
diff --git a/src/gallium/auxiliary/util/u_linear.h b/src/gallium/auxiliary/util/u_linear.h
new file mode 100644
index 00000000000..1589f029bc4
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_linear.h
@@ -0,0 +1,61 @@
+
+#ifndef U_LINEAR_H
+#define U_LINEAR_H
+
+#include "pipe/p_format.h"
+
+struct pipe_tile_info
+{
+   unsigned size;
+   unsigned stride;
+
+   /* The number of tiles */
+   unsigned tiles_x;
+   unsigned tiles_y;
+
+   /* size of each tile expressed in blocks */
+   unsigned cols;
+   unsigned rows;
+
+   /* Describe the tile in pixels */
+   struct pipe_format_block tile;
+
+   /* Describe each block within the tile */
+   struct pipe_format_block block;
+};
+
+void pipe_linear_to_tile(size_t src_stride, const void *src_ptr,
+			 struct pipe_tile_info *t, void  *dst_ptr);
+
+void pipe_linear_from_tile(struct pipe_tile_info *t, const void *src_ptr,
+			   size_t dst_stride, void *dst_ptr);
+
+/**
+ * Convenience function to fillout a pipe_tile_info struct.
+ * @t info to fill out.
+ * @block block info about pixel layout
+ * @tile_width the width of the tile in pixels
+ * @tile_height the height of the tile in pixels
+ * @tiles_x number of tiles in x axis
+ * @tiles_y number of tiles in y axis
+ */
+void pipe_linear_fill_info(struct pipe_tile_info *t,
+			   const struct pipe_format_block *block,
+			   unsigned tile_width, unsigned tile_height,
+			   unsigned tiles_x, unsigned tiles_y);
+
+static INLINE boolean pipe_linear_check_tile(const struct pipe_tile_info *t)
+{
+   if (t->tile.size != t->block.size * t->cols * t->rows)
+      return FALSE;
+
+   if (t->stride != t->block.size * t->cols * t->tiles_x)
+      return FALSE;
+
+   if (t->size < t->stride * t->rows * t->tiles_y)
+      return FALSE;
+
+   return TRUE;
+}
+
+#endif /* U_LINEAR_H */
diff --git a/src/gallium/auxiliary/util/u_math.c b/src/gallium/auxiliary/util/u_math.c
new file mode 100644
index 00000000000..2811475fa00
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_math.c
@@ -0,0 +1,73 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+
+#include "util/u_math.h"
+
+
+/** 2^x, for x in [-1.0, 1.0) */
+float pow2_table[POW2_TABLE_SIZE];
+
+
+static void
+init_pow2_table(void)
+{
+   int i;
+   for (i = 0; i < POW2_TABLE_SIZE; i++)
+      pow2_table[i] = (float) pow(2.0, (i - POW2_TABLE_OFFSET) / POW2_TABLE_SCALE);
+}
+
+
+/** log2(x), for x in [1.0, 2.0) */
+float log2_table[LOG2_TABLE_SIZE];
+
+
+static void 
+init_log2_table(void)
+{
+   unsigned i;
+   for (i = 0; i < LOG2_TABLE_SIZE; i++)
+      log2_table[i] = (float) log2(1.0 + i * (1.0 / LOG2_TABLE_SCALE));
+}
+
+
+/**
+ * One time init for math utilities.
+ */
+void
+util_init_math(void)
+{
+   static boolean initialized = FALSE;
+   if (!initialized) {
+      init_pow2_table();
+      init_log2_table();
+      initialized = TRUE;
+   }
+}
+
+
diff --git a/src/gallium/auxiliary/util/u_math.h b/src/gallium/auxiliary/util/u_math.h
new file mode 100644
index 00000000000..ab6f39ac31f
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_math.h
@@ -0,0 +1,437 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+/**
+ * Math utilities and approximations for common math functions.
+ * Reduced precision is usually acceptable in shaders...
+ *
+ * "fast" is used in the names of functions which are low-precision,
+ * or at least lower-precision than the normal C lib functions.
+ */
+
+
+#ifndef U_MATH_H
+#define U_MATH_H
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+#if defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT)
+__inline double ceil(double val)
+{
+   double ceil_val;
+
+   if((val - (long) val) == 0) {
+      ceil_val = val;
+   }
+   else {
+      if(val > 0) {
+         ceil_val = (long) val + 1;
+      }
+      else {
+         ceil_val = (long) val;
+      }
+   }
+
+   return ceil_val;
+}
+
+#ifndef PIPE_SUBSYSTEM_WINDOWS_CE_OGL
+__inline double floor(double val)
+{
+   double floor_val;
+
+   if((val - (long) val) == 0) {
+      floor_val = val;
+   }
+   else {
+      if(val > 0) {
+         floor_val = (long) val;
+      }
+      else {
+         floor_val = (long) val - 1;
+      }
+   }
+
+   return floor_val;
+}
+#endif
+
+#pragma function(pow)
+__inline double __cdecl pow(double val, double exponent)
+{
+   /* XXX */
+   assert(0);
+   return 0;
+}
+
+#pragma function(log)
+__inline double __cdecl log(double val)
+{
+   /* XXX */
+   assert(0);
+   return 0;
+}
+
+#pragma function(atan2)
+__inline double __cdecl atan2(double val)
+{
+   /* XXX */
+   assert(0);
+   return 0;
+}
+#else
+#include <math.h>
+#include <stdarg.h>
+#endif
+
+
+#if defined(_MSC_VER) 
+
+#if _MSC_VER < 1400 && !defined(__cplusplus) || defined(PIPE_SUBSYSTEM_WINDOWS_CE)
+ 
+static INLINE float cosf( float f ) 
+{
+   return (float) cos( (double) f );
+}
+
+static INLINE float sinf( float f ) 
+{
+   return (float) sin( (double) f );
+}
+
+static INLINE float ceilf( float f ) 
+{
+   return (float) ceil( (double) f );
+}
+
+static INLINE float floorf( float f ) 
+{
+   return (float) floor( (double) f );
+}
+
+static INLINE float powf( float f, float g ) 
+{
+   return (float) pow( (double) f, (double) g );
+}
+
+static INLINE float sqrtf( float f ) 
+{
+   return (float) sqrt( (double) f );
+}
+
+static INLINE float fabsf( float f ) 
+{
+   return (float) fabs( (double) f );
+}
+
+static INLINE float logf( float f ) 
+{
+   return (float) log( (double) f );
+}
+
+#else
+/* Work-around an extra semi-colon in VS 2005 logf definition */
+#ifdef logf
+#undef logf
+#define logf(x) ((float)log((double)(x)))
+#endif /* logf */
+#endif
+
+static INLINE double log2( double x )
+{
+   const double invln2 = 1.442695041;
+   return log( x ) * invln2;
+}
+
+#endif /* _MSC_VER */
+
+
+
+
+
+#define POW2_TABLE_SIZE_LOG2 9
+#define POW2_TABLE_SIZE (1 << POW2_TABLE_SIZE_LOG2)
+#define POW2_TABLE_OFFSET (POW2_TABLE_SIZE/2)
+#define POW2_TABLE_SCALE ((float)(POW2_TABLE_SIZE/2))
+extern float pow2_table[POW2_TABLE_SIZE];
+
+
+
+extern void
+util_init_math(void);
+
+
+union fi {
+   float f;
+   int32_t i;
+   uint32_t ui;
+};
+
+
+/**
+ * Fast version of 2^x
+ * Identity: exp2(a + b) = exp2(a) * exp2(b)
+ * Let ipart = int(x)
+ * Let fpart = x - ipart;
+ * So, exp2(x) = exp2(ipart) * exp2(fpart)
+ * Compute exp2(ipart) with i << ipart
+ * Compute exp2(fpart) with lookup table.
+ */
+static INLINE float
+util_fast_exp2(float x)
+{
+   int32_t ipart;
+   float fpart, mpart;
+   union fi epart;
+   
+   if(x > 129.00000f)
+      return 3.402823466e+38f;
+   
+   if(x < -126.99999f)
+      return 0.0f;
+
+   ipart = (int32_t) x;
+   fpart = x - (float) ipart;
+   
+   /* same as
+    *   epart.f = (float) (1 << ipart)
+    * but faster and without integer overflow for ipart > 31 */
+   epart.i = (ipart + 127 ) << 23;
+   
+   mpart = pow2_table[POW2_TABLE_OFFSET + (int)(fpart * POW2_TABLE_SCALE)];
+   
+   return epart.f * mpart;
+}
+
+
+/**
+ * Fast approximation to exp(x).
+ */
+static INLINE float
+util_fast_exp(float x)
+{
+   const float k = 1.44269f; /* = log2(e) */
+   return util_fast_exp2(k * x);
+}
+
+
+#define LOG2_TABLE_SIZE_LOG2 16
+#define LOG2_TABLE_SCALE (1 << LOG2_TABLE_SIZE_LOG2)
+#define LOG2_TABLE_SIZE (LOG2_TABLE_SCALE + 1)
+extern float log2_table[LOG2_TABLE_SIZE];
+
+
+static INLINE float
+util_fast_log2(float x)
+{
+   union fi num;
+   float epart, mpart;
+   num.f = x;
+   epart = (float)(((num.i & 0x7f800000) >> 23) - 127);
+   /* mpart = log2_table[mantissa*LOG2_TABLE_SCALE + 0.5] */
+   mpart = log2_table[((num.i & 0x007fffff) + (1 << (22 - LOG2_TABLE_SIZE_LOG2))) >> (23 - LOG2_TABLE_SIZE_LOG2)];
+   return epart + mpart;
+}
+
+
+static INLINE float
+util_fast_pow(float x, float y)
+{
+   return util_fast_exp2(util_fast_log2(x) * y);
+}
+
+
+
+/**
+ * Floor(x), returned as int.
+ */
+static INLINE int
+util_ifloor(float f)
+{
+   int ai, bi;
+   double af, bf;
+   union fi u;
+   af = (3 << 22) + 0.5 + (double)f;
+   bf = (3 << 22) + 0.5 - (double)f;
+   u.f = (float) af;  ai = u.i;
+   u.f = (float) bf;  bi = u.i;
+   return (ai - bi) >> 1;
+}
+
+
+/**
+ * Round float to nearest int.
+ */
+static INLINE int
+util_iround(float f)
+{
+#if defined(PIPE_CC_GCC) && defined(PIPE_ARCH_X86) 
+   int r;
+   __asm__ ("fistpl %0" : "=m" (r) : "t" (f) : "st");
+   return r;
+#elif defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86)
+   int r;
+   _asm {
+	 fld f
+	 fistp r
+	}
+   return r;
+#else
+   if (f >= 0.0f)
+      return (int) (f + 0.5f);
+   else
+      return (int) (f - 0.5f);
+#endif
+}
+
+
+
+#if defined(PIPE_CC_MSVC) && defined(PIPE_ARCH_X86)
+/**
+ * Find first bit set in word.  Least significant bit is 1.
+ * Return 0 if no bits set.
+ */
+static INLINE
+unsigned ffs( unsigned u )
+{
+   unsigned i;
+
+   if( u == 0 ) {
+      return 0;
+   }
+
+   __asm bsf eax, [u]
+   __asm inc eax
+   __asm mov [i], eax
+
+   return i;
+}
+#endif
+
+#ifdef __MINGW32__
+#define ffs __builtin_ffs
+#endif
+
+
+/**
+ * Return float bits.
+ */
+static INLINE unsigned
+fui( float f )
+{
+   union fi fi;
+   fi.f = f;
+   return fi.ui;
+}
+
+
+
+static INLINE float
+ubyte_to_float(ubyte ub)
+{
+   return (float) ub * (1.0f / 255.0f);
+}
+
+
+/**
+ * Convert float in [0,1] to ubyte in [0,255] with clamping.
+ */
+static INLINE ubyte
+float_to_ubyte(float f)
+{
+   const int ieee_0996 = 0x3f7f0000;   /* 0.996 or so */
+   union fi tmp;
+
+   tmp.f = f;
+   if (tmp.i < 0) {
+      return (ubyte) 0;
+   }
+   else if (tmp.i >= ieee_0996) {
+      return (ubyte) 255;
+   }
+   else {
+      tmp.f = tmp.f * (255.0f/256.0f) + 32768.0f;
+      return (ubyte) tmp.i;
+   }
+}
+
+
+
+#define CLAMP( X, MIN, MAX )  ( (X)<(MIN) ? (MIN) : ((X)>(MAX) ? (MAX) : (X)) )
+
+#define MIN2( A, B )   ( (A)<(B) ? (A) : (B) )
+#define MAX2( A, B )   ( (A)>(B) ? (A) : (B) )
+
+
+static INLINE int
+align(int value, int alignment)
+{
+   return (value + alignment - 1) & ~(alignment - 1);
+}
+
+
+#ifndef COPY_4V
+#define COPY_4V( DST, SRC )         \
+do {                                \
+   (DST)[0] = (SRC)[0];             \
+   (DST)[1] = (SRC)[1];             \
+   (DST)[2] = (SRC)[2];             \
+   (DST)[3] = (SRC)[3];             \
+} while (0)
+#endif
+
+
+#ifndef COPY_4FV
+#define COPY_4FV( DST, SRC )  COPY_4V(DST, SRC)
+#endif
+
+
+#ifndef ASSIGN_4V
+#define ASSIGN_4V( DST, V0, V1, V2, V3 ) \
+do {                                     \
+   (DST)[0] = (V0);                      \
+   (DST)[1] = (V1);                      \
+   (DST)[2] = (V2);                      \
+   (DST)[3] = (V3);                      \
+} while (0)
+#endif
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* U_MATH_H */
diff --git a/src/gallium/auxiliary/util/u_memory.h b/src/gallium/auxiliary/util/u_memory.h
new file mode 100644
index 00000000000..1a6b596421f
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_memory.h
@@ -0,0 +1,234 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+/**
+ * Memory functions
+ */
+
+
+#ifndef U_MEMORY_H
+#define U_MEMORY_H
+
+
+#include "util/u_pointer.h"
+#include "pipe/p_debug.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+/* Define ENOMEM for WINCE */ 
+#if (_WIN32_WCE < 600)
+#ifndef ENOMEM
+#define ENOMEM 12
+#endif
+#endif
+
+
+#if defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) && defined(DEBUG) 
+
+/* memory debugging */
+
+#include "pipe/p_debug.h"
+
+#define MALLOC( _size ) \
+   debug_malloc( __FILE__, __LINE__, __FUNCTION__, _size )
+#define CALLOC( _count, _size ) \
+   debug_calloc(__FILE__, __LINE__, __FUNCTION__, _count, _size )
+#define FREE( _ptr ) \
+   debug_free( __FILE__, __LINE__, __FUNCTION__,  _ptr )
+#define REALLOC( _ptr, _old_size, _size ) \
+   debug_realloc( __FILE__, __LINE__, __FUNCTION__,  _ptr, _old_size, _size )
+
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY)
+
+void * __stdcall
+EngAllocMem(
+    unsigned long Flags,
+    unsigned long MemSize,
+    unsigned long Tag );
+
+void __stdcall
+EngFreeMem(
+    void *Mem );
+
+#define MALLOC( _size ) EngAllocMem( 0, _size, 'D3AG' )
+#define _FREE( _ptr ) EngFreeMem( _ptr )
+
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT)
+
+void *
+ExAllocatePool(
+    unsigned long PoolType, 
+    size_t NumberOfBytes);
+
+void 
+ExFreePool(void *P);
+
+#define MALLOC(_size) ExAllocatePool(0, _size)
+#define _FREE(_ptr) ExFreePool(_ptr)
+
+#else
+
+#define MALLOC( SIZE )  malloc( SIZE )
+#define CALLOC( COUNT, SIZE )   calloc( COUNT, SIZE )
+#define FREE( PTR )  free( PTR )
+#define REALLOC( OLDPTR, OLDSIZE, NEWSIZE )  realloc( OLDPTR, NEWSIZE )
+
+#endif
+
+
+#ifndef CALLOC
+static INLINE void *
+CALLOC( unsigned count, unsigned size )
+{
+   void *ptr = MALLOC( count * size );
+   if( ptr ) {
+      memset( ptr, 0, count * size );
+   }
+   return ptr;
+}
+#endif /* !CALLOC */
+
+#ifndef FREE
+static INLINE void
+FREE( void *ptr )
+{
+   if( ptr ) {
+      _FREE( ptr );
+   }
+}
+#endif /* !FREE */
+
+#ifndef REALLOC
+static INLINE void *
+REALLOC( void *old_ptr, unsigned old_size, unsigned new_size )
+{
+   void *new_ptr = NULL;
+
+   if (new_size != 0) {
+      unsigned copy_size = old_size < new_size ? old_size : new_size;
+      new_ptr = MALLOC( new_size );
+      if (new_ptr && old_ptr && copy_size) {
+         memcpy( new_ptr, old_ptr, copy_size );
+      }
+   }
+
+   FREE( old_ptr );
+   return new_ptr;
+}
+#endif /* !REALLOC */
+
+
+#define MALLOC_STRUCT(T)   (struct T *) MALLOC(sizeof(struct T))
+
+#define CALLOC_STRUCT(T)   (struct T *) CALLOC(1, sizeof(struct T))
+
+#define CALLOC_VARIANT_LENGTH_STRUCT(T,more_size)   ((struct T *) CALLOC(1, sizeof(struct T) + more_size))
+
+
+/**
+ * Return memory on given byte alignment
+ */
+static INLINE void *
+align_malloc(size_t bytes, uint alignment)
+{
+#if defined(HAVE_POSIX_MEMALIGN)
+   void *mem;
+   alignment = (alignment + (uint)sizeof(void*) - 1) & ~((uint)sizeof(void*) - 1);
+   if(posix_memalign(& mem, alignment, bytes) != 0)
+      return NULL;
+   return mem;
+#else
+   char *ptr, *buf;
+
+   assert( alignment > 0 );
+
+   ptr = (char *) MALLOC(bytes + alignment + sizeof(void *));
+   if (!ptr)
+      return NULL;
+
+   buf = (char *) align_pointer( ptr + sizeof(void *), alignment );
+   *(char **)(buf - sizeof(void *)) = ptr;
+
+   return buf;
+#endif /* defined(HAVE_POSIX_MEMALIGN) */
+}
+
+/**
+ * Free memory returned by align_malloc().
+ */
+static INLINE void
+align_free(void *ptr)
+{
+#if defined(HAVE_POSIX_MEMALIGN)
+   FREE(ptr);
+#else
+   void **cubbyHole = (void **) ((char *) ptr - sizeof(void *));
+   void *realAddr = *cubbyHole;
+   FREE(realAddr);
+#endif /* defined(HAVE_POSIX_MEMALIGN) */
+}
+
+
+/**
+ * Duplicate a block of memory.
+ */
+static INLINE void *
+mem_dup(const void *src, uint size)
+{
+   void *dup = MALLOC(size);
+   if (dup)
+      memcpy(dup, src, size);
+   return dup;
+}
+
+
+/**
+ * Number of elements in an array.
+ */
+#ifndef Elements
+#define Elements(x) (sizeof(x)/sizeof((x)[0]))
+#endif
+
+
+/**
+ * Offset of a field in a struct, in bytes.
+ */
+#define Offset(TYPE, MEMBER) ((unsigned)&(((TYPE *)NULL)->MEMBER))
+
+
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif /* U_MEMORY_H */
diff --git a/src/gallium/auxiliary/util/u_mm.c b/src/gallium/auxiliary/util/u_mm.c
new file mode 100644
index 00000000000..45ce257b5e5
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_mm.c
@@ -0,0 +1,287 @@
+/**************************************************************************
+ *
+ * Copyright (C) 1999 Wittawat Yamwong
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * WITTAWAT YAMWONG, OR ANY OTHER CONTRIBUTORS BE LIABLE FOR ANY CLAIM, 
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE 
+ * OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_debug.h"
+
+#include "util/u_memory.h"
+#include "util/u_mm.h"
+
+
+void
+u_mmDumpMemInfo(const struct mem_block *heap)
+{
+   debug_printf("Memory heap %p:\n", (void *)heap);
+   if (heap == 0) {
+      debug_printf("  heap == 0\n");
+   } else {
+      const struct mem_block *p;
+
+      for(p = heap->next; p != heap; p = p->next) {
+	 debug_printf("  Offset:%08x, Size:%08x, %c%c\n",p->ofs,p->size,
+		 p->free ? 'F':'.',
+		 p->reserved ? 'R':'.');
+      }
+
+      debug_printf("\nFree list:\n");
+
+      for(p = heap->next_free; p != heap; p = p->next_free) {
+	 debug_printf(" FREE Offset:%08x, Size:%08x, %c%c\n",p->ofs,p->size,
+		 p->free ? 'F':'.',
+		 p->reserved ? 'R':'.');
+      }
+
+   }
+   debug_printf("End of memory blocks\n");
+}
+
+struct mem_block *
+u_mmInit(int ofs, int size)
+{
+   struct mem_block *heap, *block;
+  
+   if (size <= 0) 
+      return NULL;
+
+   heap = CALLOC_STRUCT(mem_block);
+   if (!heap) 
+      return NULL;
+   
+   block = CALLOC_STRUCT(mem_block);
+   if (!block) {
+      FREE(heap);
+      return NULL;
+   }
+
+   heap->next = block;
+   heap->prev = block;
+   heap->next_free = block;
+   heap->prev_free = block;
+
+   block->heap = heap;
+   block->next = heap;
+   block->prev = heap;
+   block->next_free = heap;
+   block->prev_free = heap;
+
+   block->ofs = ofs;
+   block->size = size;
+   block->free = 1;
+
+   return heap;
+}
+
+
+static struct mem_block *
+SliceBlock(struct mem_block *p, 
+           int startofs, int size, 
+           int reserved, int alignment)
+{
+   struct mem_block *newblock;
+
+   /* break left  [p, newblock, p->next], then p = newblock */
+   if (startofs > p->ofs) {
+      newblock = CALLOC_STRUCT(mem_block);
+      if (!newblock)
+	 return NULL;
+      newblock->ofs = startofs;
+      newblock->size = p->size - (startofs - p->ofs);
+      newblock->free = 1;
+      newblock->heap = p->heap;
+
+      newblock->next = p->next;
+      newblock->prev = p;
+      p->next->prev = newblock;
+      p->next = newblock;
+
+      newblock->next_free = p->next_free;
+      newblock->prev_free = p;
+      p->next_free->prev_free = newblock;
+      p->next_free = newblock;
+
+      p->size -= newblock->size;
+      p = newblock;
+   }
+
+   /* break right, also [p, newblock, p->next] */
+   if (size < p->size) {
+      newblock = CALLOC_STRUCT(mem_block);
+      if (!newblock)
+	 return NULL;
+      newblock->ofs = startofs + size;
+      newblock->size = p->size - size;
+      newblock->free = 1;
+      newblock->heap = p->heap;
+
+      newblock->next = p->next;
+      newblock->prev = p;
+      p->next->prev = newblock;
+      p->next = newblock;
+
+      newblock->next_free = p->next_free;
+      newblock->prev_free = p;
+      p->next_free->prev_free = newblock;
+      p->next_free = newblock;
+	 
+      p->size = size;
+   }
+
+   /* p = middle block */
+   p->free = 0;
+
+   /* Remove p from the free list: 
+    */
+   p->next_free->prev_free = p->prev_free;
+   p->prev_free->next_free = p->next_free;
+
+   p->next_free = 0;
+   p->prev_free = 0;
+
+   p->reserved = reserved;
+   return p;
+}
+
+
+struct mem_block *
+u_mmAllocMem(struct mem_block *heap, int size, int align2, int startSearch)
+{
+   struct mem_block *p;
+   const int mask = (1 << align2)-1;
+   int startofs = 0;
+   int endofs;
+
+   assert(size >= 0);
+   assert(align2 >= 0);
+   assert(align2 <= 12); /* sanity check, 2^12 (4KB) enough? */
+
+   if (!heap || align2 < 0 || size <= 0)
+      return NULL;
+
+   for (p = heap->next_free; p != heap; p = p->next_free) {
+      assert(p->free);
+
+      startofs = (p->ofs + mask) & ~mask;
+      if ( startofs < startSearch ) {
+	 startofs = startSearch;
+      }
+      endofs = startofs+size;
+      if (endofs <= (p->ofs+p->size))
+	 break;
+   }
+
+   if (p == heap) 
+      return NULL;
+
+   assert(p->free);
+   p = SliceBlock(p,startofs,size,0,mask+1);
+
+   return p;
+}
+
+
+struct mem_block *
+u_mmFindBlock(struct mem_block *heap, int start)
+{
+   struct mem_block *p;
+
+   for (p = heap->next; p != heap; p = p->next) {
+      if (p->ofs == start) 
+	 return p;
+   }
+
+   return NULL;
+}
+
+
+static INLINE int
+Join2Blocks(struct mem_block *p)
+{
+   /* XXX there should be some assertions here */
+
+   /* NOTE: heap->free == 0 */
+
+   if (p->free && p->next->free) {
+      struct mem_block *q = p->next;
+
+      assert(p->ofs + p->size == q->ofs);
+      p->size += q->size;
+
+      p->next = q->next;
+      q->next->prev = p;
+
+      q->next_free->prev_free = q->prev_free; 
+      q->prev_free->next_free = q->next_free;
+     
+      FREE(q);
+      return 1;
+   }
+   return 0;
+}
+
+int
+u_mmFreeMem(struct mem_block *b)
+{
+   if (!b)
+      return 0;
+
+   if (b->free) {
+      debug_printf("block already free\n");
+      return -1;
+   }
+   if (b->reserved) {
+      debug_printf("block is reserved\n");
+      return -1;
+   }
+
+   b->free = 1;
+   b->next_free = b->heap->next_free;
+   b->prev_free = b->heap;
+   b->next_free->prev_free = b;
+   b->prev_free->next_free = b;
+
+   Join2Blocks(b);
+   if (b->prev != b->heap)
+      Join2Blocks(b->prev);
+
+   return 0;
+}
+
+
+void
+u_mmDestroy(struct mem_block *heap)
+{
+   struct mem_block *p;
+
+   if (!heap)
+      return;
+
+   for (p = heap->next; p != heap; ) {
+      struct mem_block *next = p->next;
+      FREE(p);
+      p = next;
+   }
+
+   FREE(heap);
+}
diff --git a/src/gallium/auxiliary/util/u_mm.h b/src/gallium/auxiliary/util/u_mm.h
new file mode 100644
index 00000000000..ce20e487635
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_mm.h
@@ -0,0 +1,91 @@
+/**************************************************************************
+ *
+ * Copyright (C) 1999 Wittawat Yamwong
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * KEITH WHITWELL, OR ANY OTHER CONTRIBUTORS BE LIABLE FOR ANY CLAIM, 
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE 
+ * OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+
+/**
+ * @file
+ * Memory manager code.  Primarily used by device drivers to manage texture
+ * heaps, etc.
+ */
+
+
+#ifndef _U_MM_H_
+#define _U_MM_H_
+
+
+struct mem_block {
+   struct mem_block *next, *prev;
+   struct mem_block *next_free, *prev_free;
+   struct mem_block *heap;
+   int ofs,size;
+   unsigned int free:1;
+   unsigned int reserved:1;
+};
+
+
+
+/** 
+ * input: total size in bytes
+ * return: a heap pointer if OK, NULL if error
+ */
+extern struct mem_block *u_mmInit(int ofs, int size);
+
+/**
+ * Allocate 'size' bytes with 2^align2 bytes alignment,
+ * restrict the search to free memory after 'startSearch'
+ * depth and back buffers should be in different 4mb banks
+ * to get better page hits if possible
+ * input:	size = size of block
+ *       	align2 = 2^align2 bytes alignment
+ *		startSearch = linear offset from start of heap to begin search
+ * return: pointer to the allocated block, 0 if error
+ */
+extern struct mem_block *u_mmAllocMem(struct mem_block *heap, int size, int align2, 
+                            int startSearch);
+
+/**
+ * Free block starts at offset
+ * input: pointer to a block
+ * return: 0 if OK, -1 if error
+ */
+extern int u_mmFreeMem(struct mem_block *b);
+
+/**
+ * Free block starts at offset
+ * input: pointer to a heap, start offset
+ * return: pointer to a block
+ */
+extern struct mem_block *u_mmFindBlock(struct mem_block *heap, int start);
+
+/**
+ * destroy MM
+ */
+extern void u_mmDestroy(struct mem_block *mmInit);
+
+/**
+ * For debuging purpose.
+ */
+extern void u_mmDumpMemInfo(const struct mem_block *mmInit);
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_pack_color.h b/src/gallium/auxiliary/util/u_pack_color.h
new file mode 100644
index 00000000000..e0e8aa8e9fe
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_pack_color.h
@@ -0,0 +1,472 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Functions to produce packed colors/Z from floats.
+ */
+
+
+#ifndef U_PACK_COLOR_H
+#define U_PACK_COLOR_H
+
+
+#include "pipe/p_compiler.h"
+#include "pipe/p_format.h"
+#include "util/u_math.h"
+
+
+/**
+ * Pack ubyte R,G,B,A into dest pixel.
+ */
+static INLINE void
+util_pack_color_ub(ubyte r, ubyte g, ubyte b, ubyte a,
+                   enum pipe_format format, void *dest)
+{
+   switch (format) {
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      {
+         uint *d = (uint *) dest;
+         *d = (r << 24) | (g << 16) | (b << 8) | a;
+      }
+      return;
+   case PIPE_FORMAT_R8G8B8X8_UNORM:
+      {
+         uint *d = (uint *) dest;
+         *d = (r << 24) | (g << 16) | (b << 8) | 0xff;
+      }
+      return;
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+      {
+         uint *d = (uint *) dest;
+         *d = (a << 24) | (r << 16) | (g << 8) | b;
+      }
+      return;
+   case PIPE_FORMAT_X8R8G8B8_UNORM:
+      {
+         uint *d = (uint *) dest;
+         *d = (0xff << 24) | (r << 16) | (g << 8) | b;
+      }
+      return;
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      {
+         uint *d = (uint *) dest;
+         *d = (b << 24) | (g << 16) | (r << 8) | a;
+      }
+      return;
+   case PIPE_FORMAT_B8G8R8X8_UNORM:
+      {
+         uint *d = (uint *) dest;
+         *d = (b << 24) | (g << 16) | (r << 8) | 0xff;
+      }
+      return;
+   case PIPE_FORMAT_R5G6B5_UNORM:
+      {
+         ushort *d = (ushort *) dest;
+         *d = ((r & 0xf8) << 8) | ((g & 0xfc) << 3) | (b >> 3);
+      }
+      return;
+   case PIPE_FORMAT_A1R5G5B5_UNORM:
+      {
+         ushort *d = (ushort *) dest;
+         *d = ((a & 0x80) << 8) | ((r & 0xf8) << 7) | ((g & 0xf8) << 2) | (b >> 3);
+      }
+      return;
+   case PIPE_FORMAT_A4R4G4B4_UNORM:
+      {
+         ushort *d = (ushort *) dest;
+         *d = ((a & 0xf0) << 8) | ((r & 0xf0) << 4) | ((g & 0xf0) << 0) | (b >> 4);
+      }
+      return;
+   case PIPE_FORMAT_A8_UNORM:
+      {
+         ubyte *d = (ubyte *) dest;
+         *d = a;
+      }
+      return;
+   case PIPE_FORMAT_L8_UNORM:
+   case PIPE_FORMAT_I8_UNORM:
+      {
+         ubyte *d = (ubyte *) dest;
+         *d = r;
+      }
+      return;
+   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+      {
+         float *d = (float *) dest;
+         d[0] = (float)r / 255.0f;
+         d[1] = (float)g / 255.0f;
+         d[2] = (float)b / 255.0f;
+         d[3] = (float)a / 255.0f;
+      }
+      return;
+   case PIPE_FORMAT_R32G32B32_FLOAT:
+      {
+         float *d = (float *) dest;
+         d[0] = (float)r / 255.0f;
+         d[1] = (float)g / 255.0f;
+         d[2] = (float)b / 255.0f;
+      }
+      return;
+
+   /* XXX lots more cases to add */
+   default:
+      debug_print_format("gallium: unhandled format in util_pack_color_ub()", format);
+      assert(0);
+   }
+}
+ 
+
+/**
+ * Unpack RGBA from a packed pixel, returning values as ubytes in [0,255].
+ */
+static INLINE void
+util_unpack_color_ub(enum pipe_format format, const void *src,
+                     ubyte *r, ubyte *g, ubyte *b, ubyte *a)
+{
+   switch (format) {
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      {
+         uint p = ((const uint *) src)[0];
+         *r = (ubyte) ((p >> 24) & 0xff);
+         *g = (ubyte) ((p >> 16) & 0xff);
+         *b = (ubyte) ((p >>  8) & 0xff);
+         *a = (ubyte) ((p >>  0) & 0xff);
+      }
+      return;
+   case PIPE_FORMAT_R8G8B8X8_UNORM:
+      {
+         uint p = ((const uint *) src)[0];
+         *r = (ubyte) ((p >> 24) & 0xff);
+         *g = (ubyte) ((p >> 16) & 0xff);
+         *b = (ubyte) ((p >>  8) & 0xff);
+         *a = (ubyte) 0xff;
+      }
+      return;
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+      {
+         uint p = ((const uint *) src)[0];
+         *r = (ubyte) ((p >> 16) & 0xff);
+         *g = (ubyte) ((p >>  8) & 0xff);
+         *b = (ubyte) ((p >>  0) & 0xff);
+         *a = (ubyte) ((p >> 24) & 0xff);
+      }
+      return;
+   case PIPE_FORMAT_X8R8G8B8_UNORM:
+      {
+         uint p = ((const uint *) src)[0];
+         *r = (ubyte) ((p >> 16) & 0xff);
+         *g = (ubyte) ((p >>  8) & 0xff);
+         *b = (ubyte) ((p >>  0) & 0xff);
+         *a = (ubyte) 0xff;
+      }
+      return;
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      {
+         uint p = ((const uint *) src)[0];
+         *r = (ubyte) ((p >>  8) & 0xff);
+         *g = (ubyte) ((p >> 16) & 0xff);
+         *b = (ubyte) ((p >> 24) & 0xff);
+         *a = (ubyte) ((p >>  0) & 0xff);
+      }
+      return;
+   case PIPE_FORMAT_B8G8R8X8_UNORM:
+      {
+         uint p = ((const uint *) src)[0];
+         *r = (ubyte) ((p >>  8) & 0xff);
+         *g = (ubyte) ((p >> 16) & 0xff);
+         *b = (ubyte) ((p >> 24) & 0xff);
+         *a = (ubyte) 0xff;
+      }
+      return;
+   case PIPE_FORMAT_R5G6B5_UNORM:
+      {
+         ushort p = ((const ushort *) src)[0];
+         *r = (ubyte) (((p >> 8) & 0xf8) | ((p >> 13) & 0x7));
+         *g = (ubyte) (((p >> 3) & 0xfc) | ((p >>  9) & 0x3));
+         *b = (ubyte) (((p << 3) & 0xf8) | ((p >>  2) & 0x7));
+         *a = (ubyte) 0xff;
+      }
+      return;
+   case PIPE_FORMAT_A1R5G5B5_UNORM:
+      {
+         ushort p = ((const ushort *) src)[0];
+         *r = (ubyte) (((p >>  7) & 0xf8) | ((p >> 12) & 0x7));
+         *g = (ubyte) (((p >>  2) & 0xf8) | ((p >>  7) & 0x7));
+         *b = (ubyte) (((p <<  3) & 0xf8) | ((p >>  2) & 0x7));
+         *a = (ubyte) (0xff * (p >> 15));
+      }
+      return;
+   case PIPE_FORMAT_A4R4G4B4_UNORM:
+      {
+         ushort p = ((const ushort *) src)[0];
+         *r = (ubyte) (((p >> 4) & 0xf0) | ((p >>  8) & 0xf));
+         *g = (ubyte) (((p >> 0) & 0xf0) | ((p >>  4) & 0xf));
+         *b = (ubyte) (((p << 4) & 0xf0) | ((p >>  0) & 0xf));
+         *a = (ubyte) (((p >> 8) & 0xf0) | ((p >> 12) & 0xf));
+      }
+      return;
+   case PIPE_FORMAT_A8_UNORM:
+      {
+         ubyte p = ((const ubyte *) src)[0];
+         *r = *g = *b = (ubyte) 0xff;
+         *a = p;
+      }
+      return;
+   case PIPE_FORMAT_L8_UNORM:
+      {
+         ubyte p = ((const ubyte *) src)[0];
+         *r = *g = *b = p;
+         *a = (ubyte) 0xff;
+      }
+      return;
+   case PIPE_FORMAT_I8_UNORM:
+      {
+         ubyte p = ((const ubyte *) src)[0];
+         *r = *g = *b = *a = p;
+      }
+      return;
+   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+      {
+         const float *p = (const float *) src;
+         *r = float_to_ubyte(p[0]);
+         *g = float_to_ubyte(p[1]);
+         *b = float_to_ubyte(p[2]);
+         *a = float_to_ubyte(p[3]);
+      }
+      return;
+   case PIPE_FORMAT_R32G32B32_FLOAT:
+      {
+         const float *p = (const float *) src;
+         *r = float_to_ubyte(p[0]);
+         *g = float_to_ubyte(p[1]);
+         *b = float_to_ubyte(p[2]);
+         *a = (ubyte) 0xff;
+      }
+      return;
+
+   case PIPE_FORMAT_R32G32_FLOAT:
+      {
+         const float *p = (const float *) src;
+         *r = float_to_ubyte(p[0]);
+         *g = float_to_ubyte(p[1]);
+         *b = *a = (ubyte) 0xff;
+      }
+      return;
+
+   case PIPE_FORMAT_R32_FLOAT:
+      {
+         const float *p = (const float *) src;
+         *r = float_to_ubyte(p[0]);
+         *g = *b = *a = (ubyte) 0xff;
+      }
+      return;
+
+   /* XXX lots more cases to add */
+   default:
+      debug_print_format("gallium: unhandled format in util_unpack_color_ub()",
+                         format);
+      assert(0);
+   }
+}
+ 
+
+
+/**
+ * Note rgba outside [0,1] will be clamped for int pixel formats.
+ */
+static INLINE void
+util_pack_color(const float rgba[4], enum pipe_format format, void *dest)
+{
+   ubyte r, g, b, a;
+
+   if (pf_size_x(format) <= 8) {
+      /* format uses 8-bit components or less */
+      r = float_to_ubyte(rgba[0]);
+      g = float_to_ubyte(rgba[1]);
+      b = float_to_ubyte(rgba[2]);
+      a = float_to_ubyte(rgba[3]);
+   }
+
+   switch (format) {
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      {
+         uint *d = (uint *) dest;
+         *d = (r << 24) | (g << 16) | (b << 8) | a;
+      }
+      return;
+   case PIPE_FORMAT_R8G8B8X8_UNORM:
+      {
+         uint *d = (uint *) dest;
+         *d = (r << 24) | (g << 16) | (b << 8) | 0xff;
+      }
+      return;
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+      {
+         uint *d = (uint *) dest;
+         *d = (a << 24) | (r << 16) | (g << 8) | b;
+      }
+      return;
+   case PIPE_FORMAT_X8R8G8B8_UNORM:
+      {
+         uint *d = (uint *) dest;
+         *d = (0xff << 24) | (r << 16) | (g << 8) | b;
+      }
+      return;
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      {
+         uint *d = (uint *) dest;
+         *d = (b << 24) | (g << 16) | (r << 8) | a;
+      }
+      return;
+   case PIPE_FORMAT_B8G8R8X8_UNORM:
+      {
+         uint *d = (uint *) dest;
+         *d = (b << 24) | (g << 16) | (r << 8) | 0xff;
+      }
+      return;
+   case PIPE_FORMAT_R5G6B5_UNORM:
+      {
+         ushort *d = (ushort *) dest;
+         *d = ((r & 0xf8) << 8) | ((g & 0xfc) << 3) | (b >> 3);
+      }
+      return;
+   case PIPE_FORMAT_A1R5G5B5_UNORM:
+      {
+         ushort *d = (ushort *) dest;
+         *d = ((a & 0x80) << 8) | ((r & 0xf8) << 7) | ((g & 0xf8) << 2) | (b >> 3);
+      }
+      return;
+   case PIPE_FORMAT_A4R4G4B4_UNORM:
+      {
+         ushort *d = (ushort *) dest;
+         *d = ((a & 0xf0) << 8) | ((r & 0xf0) << 4) | ((g & 0xf0) << 0) | (b >> 4);
+      }
+      return;
+   case PIPE_FORMAT_A8_UNORM:
+      {
+         ubyte *d = (ubyte *) dest;
+         *d = a;
+      }
+      return;
+   case PIPE_FORMAT_L8_UNORM:
+   case PIPE_FORMAT_I8_UNORM:
+      {
+         ubyte *d = (ubyte *) dest;
+         *d = r;
+      }
+      return;
+   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+      {
+         float *d = (float *) dest;
+         d[0] = rgba[0];
+         d[1] = rgba[1];
+         d[2] = rgba[2];
+         d[3] = rgba[3];
+      }
+      return;
+   case PIPE_FORMAT_R32G32B32_FLOAT:
+      {
+         float *d = (float *) dest;
+         d[0] = rgba[0];
+         d[1] = rgba[1];
+         d[2] = rgba[2];
+      }
+      return;
+   /* XXX lots more cases to add */
+   default:
+      debug_print_format("gallium: unhandled format in util_pack_color()", format);
+      assert(0);
+   }
+}
+ 
+
+/**
+ * Note: it's assumed that z is in [0,1]
+ */
+static INLINE uint
+util_pack_z(enum pipe_format format, double z)
+{
+   if (z == 0.0)
+      return 0;
+
+   switch (format) {
+   case PIPE_FORMAT_Z16_UNORM:
+      if (z == 1.0)
+         return 0xffff;
+      return (uint) (z * 0xffff);
+   case PIPE_FORMAT_Z32_UNORM:
+      /* special-case to avoid overflow */
+      if (z == 1.0)
+         return 0xffffffff;
+      return (uint) (z * 0xffffffff);
+   case PIPE_FORMAT_S8Z24_UNORM:
+   case PIPE_FORMAT_X8Z24_UNORM:
+      if (z == 1.0)
+         return 0xffffff;
+      return (uint) (z * 0xffffff);
+   case PIPE_FORMAT_Z24S8_UNORM:
+   case PIPE_FORMAT_Z24X8_UNORM:
+      if (z == 1.0)
+         return 0xffffff00;
+      return ((uint) (z * 0xffffff)) << 8;
+   default:
+      debug_print_format("gallium: unhandled format in util_pack_z()", format);
+      assert(0);
+      return 0;
+   }
+}
+
+
+/**
+ * Pack 4 ubytes into a 4-byte word
+ */
+static INLINE unsigned
+pack_ub4(ubyte b0, ubyte b1, ubyte b2, ubyte b3)
+{
+   return ((((unsigned int)b0) << 0) |
+	   (((unsigned int)b1) << 8) |
+	   (((unsigned int)b2) << 16) |
+	   (((unsigned int)b3) << 24));
+}
+
+
+/**
+ * Pack/convert 4 floats into one 4-byte word.
+ */
+static INLINE unsigned
+pack_ui32_float4(float a, float b, float c, float d)
+{
+   return pack_ub4( float_to_ubyte(a),
+		    float_to_ubyte(b),
+		    float_to_ubyte(c),
+		    float_to_ubyte(d) );
+}
+
+
+
+#endif /* U_PACK_COLOR_H */
diff --git a/src/gallium/auxiliary/util/u_pointer.h b/src/gallium/auxiliary/util/u_pointer.h
new file mode 100644
index 00000000000..e1af9f11cb9
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_pointer.h
@@ -0,0 +1,107 @@
+/**************************************************************************
+ * 
+ * Copyright 2007-2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef U_POINTER_H
+#define U_POINTER_H
+
+#include "pipe/p_compiler.h"
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+static INLINE intptr_t
+pointer_to_intptr( const void *p )
+{
+   union {
+      const void *p;
+      intptr_t i;
+   } pi;
+   pi.p = p;
+   return pi.i;
+}
+
+static INLINE void *
+intptr_to_pointer( intptr_t i )
+{
+   union {
+      void *p;
+      intptr_t i;
+   } pi;
+   pi.i = i;
+   return pi.p;
+}
+
+static INLINE uintptr_t
+pointer_to_uintptr( const void *ptr )
+{
+   union {
+      const void *p;
+      uintptr_t u;
+   } pu;
+   pu.p = ptr;
+   return pu.u;
+}
+
+static INLINE void *
+uintptr_to_pointer( uintptr_t u )
+{
+   union {
+      void *p;
+      uintptr_t u;
+   } pu;
+   pu.u = u;
+   return pu.p;
+}
+
+/**
+ * Return a pointer aligned to next multiple of N bytes.
+ */
+static INLINE void *
+align_pointer( const void *unaligned, uintptr_t alignment )
+{
+   uintptr_t aligned = (pointer_to_uintptr( unaligned ) + alignment - 1) & ~(alignment - 1);
+   return uintptr_to_pointer( aligned );
+}
+
+
+/**
+ * Return a pointer aligned to next multiple of 16 bytes.
+ */
+static INLINE void *
+align16( void *unaligned )
+{
+   return align_pointer( unaligned, 16 );
+}
+
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* U_POINTER_H */
diff --git a/src/gallium/auxiliary/util/u_prim.h b/src/gallium/auxiliary/util/u_prim.h
new file mode 100644
index 00000000000..d7c3995dbf0
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_prim.h
@@ -0,0 +1,138 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+
+#ifndef U_BLIT_H
+#define U_BLIT_H
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+#include "pipe/p_defines.h"
+
+static INLINE boolean u_validate_pipe_prim( unsigned pipe_prim, unsigned nr )
+{
+   boolean ok = TRUE;
+
+   switch (pipe_prim) {
+   case PIPE_PRIM_POINTS:
+      ok = (nr >= 1);
+      break;
+   case PIPE_PRIM_LINES:
+      ok = (nr >= 2);
+      break;
+   case PIPE_PRIM_LINE_STRIP:
+   case PIPE_PRIM_LINE_LOOP:
+      ok = (nr >= 2);
+      break;
+   case PIPE_PRIM_TRIANGLES:
+      ok = (nr >= 3);
+      break;
+   case PIPE_PRIM_TRIANGLE_STRIP:
+   case PIPE_PRIM_TRIANGLE_FAN:
+   case PIPE_PRIM_POLYGON:
+      ok = (nr >= 3);
+      break;
+   case PIPE_PRIM_QUADS:
+      ok = (nr >= 4);
+      break;
+   case PIPE_PRIM_QUAD_STRIP:
+      ok = (nr >= 4);
+      break;
+   default:
+      ok = 0;
+      break;
+   }
+
+   return ok;
+}
+
+
+static INLINE boolean u_trim_pipe_prim( unsigned pipe_prim, unsigned *nr )
+{
+   boolean ok = TRUE;
+
+   switch (pipe_prim) {
+   case PIPE_PRIM_POINTS:
+      ok = (*nr >= 1);
+      break;
+   case PIPE_PRIM_LINES:
+      ok = (*nr >= 2);
+      *nr -= (*nr % 2);
+      break;
+   case PIPE_PRIM_LINE_STRIP:
+   case PIPE_PRIM_LINE_LOOP:
+      ok = (*nr >= 2);
+      break;
+   case PIPE_PRIM_TRIANGLES:
+      ok = (*nr >= 3);
+      *nr -= (*nr % 3);
+      break;
+   case PIPE_PRIM_TRIANGLE_STRIP:
+   case PIPE_PRIM_TRIANGLE_FAN:
+   case PIPE_PRIM_POLYGON:
+      ok = (*nr >= 3);
+      break;
+   case PIPE_PRIM_QUADS:
+      ok = (*nr >= 4);
+      *nr -= (*nr % 4);
+      break;
+   case PIPE_PRIM_QUAD_STRIP:
+      ok = (*nr >= 4);
+      *nr -= (*nr % 2);
+      break;
+   default:
+      ok = 0;
+      break;
+   }
+
+   if (!ok)
+      *nr = 0;
+
+   return ok;
+}
+
+
+static INLINE boolean u_reduced_prim( unsigned pipe_prim )
+{
+   switch (pipe_prim) {
+   case PIPE_PRIM_POINTS:
+      return PIPE_PRIM_POINTS;
+
+   case PIPE_PRIM_LINES:
+   case PIPE_PRIM_LINE_STRIP:
+   case PIPE_PRIM_LINE_LOOP:
+      return PIPE_PRIM_LINES;
+
+   default:
+      return PIPE_PRIM_TRIANGLES;
+   }
+}
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_rect.c b/src/gallium/auxiliary/util/u_rect.c
new file mode 100644
index 00000000000..fe81a685be1
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_rect.c
@@ -0,0 +1,329 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * Rectangle-related helper functions.
+ */
+
+
+#include "pipe/p_defines.h"
+#include "pipe/p_format.h"
+#include "pipe/p_context.h"
+#include "pipe/p_screen.h"
+#include "util/u_rect.h"
+
+
+/**
+ * Copy 2D rect from one place to another.
+ * Position and sizes are in pixels.
+ * src_pitch may be negative to do vertical flip of pixels from source.
+ */
+void
+pipe_copy_rect(ubyte * dst,
+               const struct pipe_format_block *block,
+               unsigned dst_stride,
+               unsigned dst_x,
+               unsigned dst_y,
+               unsigned width,
+               unsigned height,
+               const ubyte * src,
+               int src_stride,
+               unsigned src_x, 
+               int src_y)
+{
+   unsigned i;
+   int src_stride_pos = src_stride < 0 ? -src_stride : src_stride;
+
+   assert(block->size > 0);
+   assert(block->width > 0);
+   assert(block->height > 0);
+   assert(src_x >= 0);
+   assert(src_y >= 0);
+   assert(dst_x >= 0);
+   assert(dst_y >= 0);
+
+   dst_x /= block->width;
+   dst_y /= block->height;
+   width = (width + block->width - 1)/block->width;
+   height = (height + block->height - 1)/block->height;
+   src_x /= block->width;
+   src_y /= block->height;
+   
+   dst += dst_x * block->size;
+   src += src_x * block->size;
+   dst += dst_y * dst_stride;
+   src += src_y * src_stride_pos;
+   width *= block->size;
+
+   if (width == dst_stride && width == src_stride)
+      memcpy(dst, src, height * width);
+   else {
+      for (i = 0; i < height; i++) {
+         memcpy(dst, src, width);
+         dst += dst_stride;
+         src += src_stride;
+      }
+   }
+}
+
+void
+pipe_fill_rect(ubyte * dst,
+               const struct pipe_format_block *block,
+               unsigned dst_stride,
+               unsigned dst_x,
+               unsigned dst_y,
+               unsigned width,
+               unsigned height,
+               uint32_t value)
+{
+   unsigned i, j;
+   unsigned width_size;
+
+   assert(block->size > 0);
+   assert(block->width > 0);
+   assert(block->height > 0);
+   assert(dst_x >= 0);
+   assert(dst_y >= 0);
+
+   dst_x /= block->width;
+   dst_y /= block->height;
+   width = (width + block->width - 1)/block->width;
+   height = (height + block->height - 1)/block->height;
+   
+   dst += dst_x * block->size;
+   dst += dst_y * dst_stride;
+   width_size = width * block->size;
+   
+   switch (block->size) {
+   case 1:
+      if(dst_stride == width_size)
+	 memset(dst, (ubyte) value, height * width_size);
+      else {
+	 for (i = 0; i < height; i++) {
+	    memset(dst, (ubyte) value, width_size);
+	    dst += dst_stride;
+	 }
+      }
+      break;
+   case 2:
+      for (i = 0; i < height; i++) {
+	 uint16_t *row = (uint16_t *)dst;
+	 for (j = 0; j < width; j++)
+	    *row++ = (uint16_t) value;
+	 dst += dst_stride;
+      }
+      break;
+   case 4:
+      for (i = 0; i < height; i++) {
+	 uint32_t *row = (uint32_t *)dst;
+	 for (j = 0; j < width; j++)
+	    *row++ = value;
+	 dst += dst_stride;
+      }
+      break;
+   default:
+	 assert(0);
+	 break;
+   }
+}
+
+
+
+/**
+ * Fallback function for pipe->surface_copy().
+ * Note: (X,Y)=(0,0) is always the upper-left corner.
+ * if do_flip, flip the image vertically on its way from src rect to dst rect.
+ * XXX should probably put this in new u_surface.c file...
+ */
+void
+util_surface_copy(struct pipe_context *pipe,
+                  boolean do_flip,
+                  struct pipe_surface *dst,
+                  unsigned dst_x, unsigned dst_y,
+                  struct pipe_surface *src,
+                  unsigned src_x, unsigned src_y, 
+                  unsigned w, unsigned h)
+{
+   struct pipe_screen *screen = pipe->screen;
+   struct pipe_surface *new_src = NULL, *new_dst = NULL;
+   void *dst_map;
+   const void *src_map;
+
+   assert(dst->block.size == src->block.size);
+   assert(dst->block.width == src->block.width);
+   assert(dst->block.height == src->block.height);
+
+   if ((src->usage & PIPE_BUFFER_USAGE_CPU_READ) == 0) {
+      /* Need to create new src surface which is CPU readable */
+      assert(src->texture);
+      if (!src->texture)
+         return;
+      new_src = screen->get_tex_surface(screen,
+                                        src->texture,
+                                        src->face,
+                                        src->level,
+                                        src->zslice,
+                                        PIPE_BUFFER_USAGE_CPU_READ);
+      src = new_src;
+   }
+
+   if ((dst->usage & PIPE_BUFFER_USAGE_CPU_WRITE) == 0) {
+      /* Need to create new dst surface which is CPU writable */
+      assert(dst->texture);
+      if (!dst->texture)
+         return;
+      new_dst = screen->get_tex_surface(screen,
+                                        dst->texture,
+                                        dst->face,
+                                        dst->level,
+                                        dst->zslice,
+                                        PIPE_BUFFER_USAGE_CPU_WRITE);
+      dst = new_dst;
+   }
+
+   src_map = pipe->screen->surface_map(screen,
+                                       src, PIPE_BUFFER_USAGE_CPU_READ);
+   dst_map = pipe->screen->surface_map(screen,
+                                       dst, PIPE_BUFFER_USAGE_CPU_WRITE);
+
+   assert(src_map);
+   assert(dst_map);
+
+   if (src_map && dst_map) {
+      /* If do_flip, invert src_y position and pass negative src stride */
+      pipe_copy_rect(dst_map,
+                     &dst->block,
+                     dst->stride,
+                     dst_x, dst_y,
+                     w, h,
+                     src_map,
+                     do_flip ? -(int) src->stride : src->stride,
+                     src_x,
+                     do_flip ? src_y + h - 1 : src_y);
+   }
+
+   pipe->screen->surface_unmap(pipe->screen, src);
+   pipe->screen->surface_unmap(pipe->screen, dst);
+
+   if (new_src)
+      screen->tex_surface_release(screen, &new_src);
+   if (new_dst)
+      screen->tex_surface_release(screen, &new_dst);
+}
+
+
+
+static void *
+get_pointer(struct pipe_surface *dst, void *dst_map, unsigned x, unsigned y)
+{
+   return (char *)dst_map
+      + y / dst->block.height * dst->stride
+      + x / dst->block.width * dst->block.size;
+}
+
+
+#define UBYTE_TO_USHORT(B) ((B) | ((B) << 8))
+
+
+/**
+ * Fallback for pipe->surface_fill() function.
+ * XXX should probably put this in new u_surface.c file...
+ */
+void
+util_surface_fill(struct pipe_context *pipe,
+                  struct pipe_surface *dst,
+                  unsigned dstx, unsigned dsty,
+                  unsigned width, unsigned height, unsigned value)
+{
+   struct pipe_screen *screen = pipe->screen;
+   struct pipe_surface *new_dst = NULL;
+   void *dst_map;
+
+   if ((dst->usage & PIPE_BUFFER_USAGE_CPU_WRITE) == 0) {
+      /* Need to create new dst surface which is CPU writable */
+      assert(dst->texture);
+      if (!dst->texture)
+         return;
+      new_dst = screen->get_tex_surface(screen,
+                                        dst->texture,
+                                        dst->face,
+                                        dst->level,
+                                        dst->zslice,
+                                        PIPE_BUFFER_USAGE_CPU_WRITE);
+      dst = new_dst;
+   }
+
+   dst_map = pipe->screen->surface_map(screen,
+                                       dst, PIPE_BUFFER_USAGE_CPU_WRITE);
+
+   assert(dst_map);
+
+   if (dst_map) {
+      assert(dst->stride > 0);
+
+      switch (dst->block.size) {
+      case 1:
+      case 2:
+      case 4:
+         pipe_fill_rect(dst_map, &dst->block, dst->stride,
+                        dstx, dsty, width, height, value);
+         break;
+      case 8:
+         {
+            /* expand the 4-byte clear value to an 8-byte value */
+            ushort *row = (ushort *) get_pointer(dst, dst_map, dstx, dsty);
+            ushort val0 = UBYTE_TO_USHORT((value >>  0) & 0xff);
+            ushort val1 = UBYTE_TO_USHORT((value >>  8) & 0xff);
+            ushort val2 = UBYTE_TO_USHORT((value >> 16) & 0xff);
+            ushort val3 = UBYTE_TO_USHORT((value >> 24) & 0xff);
+            unsigned i, j;
+            val0 = (val0 << 8) | val0;
+            val1 = (val1 << 8) | val1;
+            val2 = (val2 << 8) | val2;
+            val3 = (val3 << 8) | val3;
+            for (i = 0; i < height; i++) {
+               for (j = 0; j < width; j++) {
+                  row[j*4+0] = val0;
+                  row[j*4+1] = val1;
+                  row[j*4+2] = val2;
+                  row[j*4+3] = val3;
+               }
+               row += dst->stride/2;
+            }
+         }
+         break;
+      default:
+         assert(0);
+         break;
+      }
+   }
+
+   pipe->screen->surface_unmap(pipe->screen, dst);
+
+   if (new_dst)
+      screen->tex_surface_release(screen, &new_dst);
+}
diff --git a/src/gallium/auxiliary/util/u_rect.h b/src/gallium/auxiliary/util/u_rect.h
new file mode 100644
index 00000000000..59e842e16d1
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_rect.h
@@ -0,0 +1,72 @@
+/**************************************************************************
+ * 
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+/**
+ * Pipe copy/fill rect helpers.
+ */
+
+
+#ifndef U_RECT_H
+#define U_RECT_H
+
+
+#include "pipe/p_format.h"
+
+struct pipe_context;
+struct pipe_surface;
+
+
+extern void
+pipe_copy_rect(ubyte * dst, const struct pipe_format_block *block,
+               unsigned dst_stride, unsigned dst_x, unsigned dst_y,
+               unsigned width, unsigned height, const ubyte * src,
+               int src_stride, unsigned src_x, int src_y);
+
+extern void
+pipe_fill_rect(ubyte * dst, const struct pipe_format_block *block,
+               unsigned dst_stride, unsigned dst_x, unsigned dst_y,
+               unsigned width, unsigned height, uint32_t value);
+
+
+extern void
+util_surface_copy(struct pipe_context *pipe,
+                  boolean do_flip,
+                  struct pipe_surface *dst,
+                  unsigned dst_x, unsigned dst_y,
+                  struct pipe_surface *src,
+                  unsigned src_x, unsigned src_y, 
+                  unsigned w, unsigned h);
+
+extern void
+util_surface_fill(struct pipe_context *pipe,
+                  struct pipe_surface *dst,
+                  unsigned dstx, unsigned dsty,
+                  unsigned width, unsigned height, unsigned value);
+
+
+#endif /* U_RECT_H */
diff --git a/src/gallium/auxiliary/util/u_simple_list.h b/src/gallium/auxiliary/util/u_simple_list.h
new file mode 100644
index 00000000000..f5f43b0faa2
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_simple_list.h
@@ -0,0 +1,197 @@
+/**
+ * \file simple_list.h
+ * Simple macros for type-safe, intrusive lists.
+ *
+ *  Intended to work with a list sentinal which is created as an empty
+ *  list.  Insert & delete are O(1).
+ *  
+ * \author
+ *  (C) 1997, Keith Whitwell
+ */
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999-2001  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+
+#ifndef _U_SIMPLE_LIST_H_
+#define _U_SIMPLE_LIST_H_
+
+/**
+ * Remove an element from list.
+ *
+ * \param elem element to remove.
+ */
+#define remove_from_list(elem)			\
+do {						\
+   (elem)->next->prev = (elem)->prev;		\
+   (elem)->prev->next = (elem)->next;		\
+} while (0)
+
+/**
+ * Insert an element to the list head.
+ *
+ * \param list list.
+ * \param elem element to insert.
+ */
+#define insert_at_head(list, elem)		\
+do {						\
+   (elem)->prev = list;				\
+   (elem)->next = (list)->next;			\
+   (list)->next->prev = elem;			\
+   (list)->next = elem;				\
+} while(0)
+
+/**
+ * Insert an element to the list tail.
+ *
+ * \param list list.
+ * \param elem element to insert.
+ */
+#define insert_at_tail(list, elem)		\
+do {						\
+   (elem)->next = list;				\
+   (elem)->prev = (list)->prev;			\
+   (list)->prev->next = elem;			\
+   (list)->prev = elem;				\
+} while(0)
+
+/**
+ * Move an element to the list head.
+ *
+ * \param list list.
+ * \param elem element to move.
+ */
+#define move_to_head(list, elem)		\
+do {						\
+   remove_from_list(elem);			\
+   insert_at_head(list, elem);			\
+} while (0)
+
+/**
+ * Move an element to the list tail.
+ *
+ * \param list list.
+ * \param elem element to move.
+ */
+#define move_to_tail(list, elem)		\
+do {						\
+   remove_from_list(elem);			\
+   insert_at_tail(list, elem);			\
+} while (0)
+
+/**
+ * Make a empty list empty.
+ *
+ * \param sentinal list (sentinal element).
+ */
+#define make_empty_list(sentinal)		\
+do {						\
+   (sentinal)->next = sentinal;			\
+   (sentinal)->prev = sentinal;			\
+} while (0)
+
+/**
+ * Get list first element.
+ *
+ * \param list list.
+ *
+ * \return pointer to first element.
+ */
+#define first_elem(list)       ((list)->next)
+
+/**
+ * Get list last element.
+ *
+ * \param list list.
+ *
+ * \return pointer to last element.
+ */
+#define last_elem(list)        ((list)->prev)
+
+/**
+ * Get next element.
+ *
+ * \param elem element.
+ *
+ * \return pointer to next element.
+ */
+#define next_elem(elem)        ((elem)->next)
+
+/**
+ * Get previous element.
+ *
+ * \param elem element.
+ *
+ * \return pointer to previous element.
+ */
+#define prev_elem(elem)        ((elem)->prev)
+
+/**
+ * Test whether element is at end of the list.
+ * 
+ * \param list list.
+ * \param elem element.
+ * 
+ * \return non-zero if element is at end of list, or zero otherwise.
+ */
+#define at_end(list, elem)     ((elem) == (list))
+
+/**
+ * Test if a list is empty.
+ * 
+ * \param list list.
+ * 
+ * \return non-zero if list empty, or zero otherwise.
+ */
+#define is_empty_list(list)    ((list)->next == (list))
+
+/**
+ * Walk through the elements of a list.
+ *
+ * \param ptr pointer to the current element.
+ * \param list list.
+ *
+ * \note It should be followed by a { } block or a single statement, as in a \c
+ * for loop.
+ */
+#define foreach(ptr, list)     \
+        for( ptr=(list)->next ;  ptr!=list ;  ptr=(ptr)->next )
+
+/**
+ * Walk through the elements of a list.
+ *
+ * Same as #foreach but lets you unlink the current value during a list
+ * traversal.  Useful for freeing a list, element by element.
+ * 
+ * \param ptr pointer to the current element.
+ * \param t temporary pointer.
+ * \param list list.
+ *
+ * \note It should be followed by a { } block or a single statement, as in a \c
+ * for loop.
+ */
+#define foreach_s(ptr, t, list)   \
+        for(ptr=(list)->next,t=(ptr)->next; list != ptr; ptr=t, t=(t)->next)
+
+#endif /* _U_SIMPLE_LIST_H_ */
diff --git a/src/gallium/auxiliary/util/u_simple_screen.c b/src/gallium/auxiliary/util/u_simple_screen.c
new file mode 100644
index 00000000000..089bbbc48a8
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_simple_screen.c
@@ -0,0 +1,143 @@
+/**************************************************************************
+ *
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#include "u_simple_screen.h"
+
+#include "pipe/p_screen.h"
+#include "pipe/internal/p_winsys_screen.h"
+
+
+static struct pipe_buffer *
+pass_buffer_create(struct pipe_screen *screen,
+                   unsigned alignment,
+                   unsigned usage,
+                   unsigned size)
+{
+   return screen->winsys->buffer_create(screen->winsys,
+                                        alignment, usage, size);
+}
+
+static struct pipe_buffer *
+pass_user_buffer_create(struct pipe_screen *screen,
+                        void *ptr,
+                        unsigned bytes)
+{
+   return screen->winsys->user_buffer_create(screen->winsys,
+                                             ptr, bytes);
+}
+
+static struct pipe_buffer *
+pass_surface_buffer_create(struct pipe_screen *screen,
+                           unsigned width, unsigned height,
+                           enum pipe_format format,
+                           unsigned usage,
+                           unsigned *stride)
+{
+   return screen->winsys->surface_buffer_create(screen->winsys,
+                                                width, height,
+                                                format, usage, stride);
+}
+
+static void *
+pass_buffer_map(struct pipe_screen *screen,
+                struct pipe_buffer *buf,
+                unsigned usage)
+{
+   return screen->winsys->buffer_map(screen->winsys,
+                                     buf, usage);
+}
+
+static void
+pass_buffer_unmap(struct pipe_screen *screen,
+                  struct pipe_buffer *buf)
+{
+   screen->winsys->buffer_unmap(screen->winsys, buf);
+}
+
+static void
+pass_buffer_destroy(struct pipe_screen *screen,
+                    struct pipe_buffer *buf)
+{
+   screen->winsys->buffer_destroy(screen->winsys, buf);
+}
+
+
+static void
+pass_flush_frontbuffer(struct pipe_screen *screen,
+                       struct pipe_surface *surf,
+                       void *context_private)
+{
+   screen->winsys->flush_frontbuffer(screen->winsys,
+                                     surf, context_private);
+}
+
+static void
+pass_fence_reference(struct pipe_screen *screen,
+                     struct pipe_fence_handle **ptr,
+                     struct pipe_fence_handle *fence)
+{
+   screen->winsys->fence_reference(screen->winsys,
+                                   ptr, fence);
+}
+
+static int
+pass_fence_signalled(struct pipe_screen *screen,
+                     struct pipe_fence_handle *fence,
+                     unsigned flag)
+{
+   return screen->winsys->fence_signalled(screen->winsys,
+                                          fence, flag);
+}
+
+static int
+pass_fence_finish(struct pipe_screen *screen,
+                  struct pipe_fence_handle *fence,
+                  unsigned flag)
+{
+   return screen->winsys->fence_finish(screen->winsys,
+                                       fence, flag);
+}
+
+void u_simple_screen_init(struct pipe_screen *screen)
+{
+   screen->buffer_create = pass_buffer_create;
+   screen->user_buffer_create = pass_user_buffer_create;
+   screen->surface_buffer_create = pass_surface_buffer_create;
+
+   screen->buffer_map = pass_buffer_map;
+   screen->buffer_unmap = pass_buffer_unmap;
+   screen->buffer_destroy = pass_buffer_destroy;
+   screen->flush_frontbuffer = pass_flush_frontbuffer;
+   screen->fence_reference = pass_fence_reference;
+   screen->fence_signalled = pass_fence_signalled;
+   screen->fence_finish = pass_fence_finish;
+}
+
+const char* u_simple_screen_winsys_name(struct pipe_screen *screen)
+{
+   return screen->winsys->get_name(screen->winsys);
+}
diff --git a/src/gallium/auxiliary/util/u_simple_screen.h b/src/gallium/auxiliary/util/u_simple_screen.h
new file mode 100644
index 00000000000..6612a8a7c09
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_simple_screen.h
@@ -0,0 +1,47 @@
+/**************************************************************************
+ *
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+#ifndef U_SIMPLE_SCREEN_H
+#define U_SIMPLE_SCREEN_H
+
+struct pipe_screen;
+struct pipe_winsys;
+
+/**
+ * The following function initializes a simple passthrough screen.
+ *
+ * All the relevant screen function pointers will forwarded to the
+ * winsys.
+ */
+void u_simple_screen_init(struct pipe_screen *screen);
+
+/**
+ * Returns the name of the winsys associated with this screen.
+ */
+const char* u_simple_screen_winsys_name(struct pipe_screen *screen);
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_simple_shaders.c b/src/gallium/auxiliary/util/u_simple_shaders.c
new file mode 100644
index 00000000000..706155e99a7
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_simple_shaders.c
@@ -0,0 +1,361 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Simple vertex/fragment shader generators.
+ *  
+ * @author Brian Paul
+ */
+
+
+#include "pipe/p_context.h"
+#include "pipe/p_debug.h"
+#include "pipe/p_defines.h"
+#include "pipe/p_inlines.h"
+#include "pipe/p_screen.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "util/u_memory.h"
+#include "util/u_simple_shaders.h"
+
+#include "tgsi/tgsi_build.h"
+#include "tgsi/tgsi_dump.h"
+#include "tgsi/tgsi_parse.h"
+
+
+
+/**
+ * Make simple vertex pass-through shader.
+ */
+void *
+util_make_vertex_passthrough_shader(struct pipe_context *pipe,
+                                    uint num_attribs,
+                                    const uint *semantic_names,
+                                    const uint *semantic_indexes,
+                                    struct pipe_shader_state *shader)
+                                    
+{
+   uint maxTokens = 100;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+   struct tgsi_processor *processor;
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+   const uint procType = TGSI_PROCESSOR_VERTEX;
+   uint ti, i;
+
+   tokens = (struct tgsi_token *) MALLOC(maxTokens * sizeof(tokens[0]));
+
+   /* shader header
+    */
+   *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
+
+   header = (struct tgsi_header *) &tokens[1];
+   *header = tgsi_build_header();
+
+   processor = (struct tgsi_processor *) &tokens[2];
+   *processor = tgsi_build_processor( procType, header );
+
+   ti = 3;
+
+   /* declare inputs */
+   for (i = 0; i < num_attribs; i++) {
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_INPUT;
+
+      decl.Declaration.Semantic = 1;
+      decl.Semantic.SemanticName = semantic_names[i];
+      decl.Semantic.SemanticIndex = semantic_indexes[i];
+
+      decl.DeclarationRange.First = 
+      decl.DeclarationRange.Last = i;
+      ti += tgsi_build_full_declaration(&decl,
+                                        &tokens[ti],
+                                        header,
+                                        maxTokens - ti);
+   }
+
+   /* declare outputs */
+   for (i = 0; i < num_attribs; i++) {
+      decl = tgsi_default_full_declaration();
+      decl.Declaration.File = TGSI_FILE_OUTPUT;
+      decl.Declaration.Semantic = 1;
+      decl.Semantic.SemanticName = semantic_names[i];
+      decl.Semantic.SemanticIndex = semantic_indexes[i];
+      decl.DeclarationRange.First = 
+         decl.DeclarationRange.Last = i;
+      ti += tgsi_build_full_declaration(&decl,
+                                        &tokens[ti],
+                                        header,
+                                        maxTokens - ti);
+   }
+
+   /* emit MOV instructions */
+   for (i = 0; i < num_attribs; i++) {
+      /* MOVE out[i], in[i]; */
+      inst = tgsi_default_full_instruction();
+      inst.Instruction.Opcode = TGSI_OPCODE_MOV;
+      inst.Instruction.NumDstRegs = 1;
+      inst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_OUTPUT;
+      inst.FullDstRegisters[0].DstRegister.Index = i;
+      inst.Instruction.NumSrcRegs = 1;
+      inst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT;
+      inst.FullSrcRegisters[0].SrcRegister.Index = i;
+      ti += tgsi_build_full_instruction(&inst,
+                                        &tokens[ti],
+                                        header,
+                                        maxTokens - ti );
+   }
+
+   /* END instruction */
+   inst = tgsi_default_full_instruction();
+   inst.Instruction.Opcode = TGSI_OPCODE_END;
+   inst.Instruction.NumDstRegs = 0;
+   inst.Instruction.NumSrcRegs = 0;
+   ti += tgsi_build_full_instruction(&inst,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti );
+
+#if 0 /*debug*/
+   tgsi_dump(tokens, 0);
+#endif
+
+   shader->tokens = tokens;
+   /*shader->num_tokens = ti;*/
+
+   return pipe->create_vs_state(pipe, shader);
+}
+
+
+
+
+/**
+ * Make simple fragment texture shader:
+ *  TEX OUT[0], IN[0], SAMP[0], 2D;
+ *  END;
+ */
+void *
+util_make_fragment_tex_shader(struct pipe_context *pipe,
+                              struct pipe_shader_state *shader)
+{
+   uint maxTokens = 100;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+   struct tgsi_processor *processor;
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+   const uint procType = TGSI_PROCESSOR_FRAGMENT;
+   uint ti;
+
+   tokens = (struct tgsi_token *) MALLOC(maxTokens * sizeof(tokens[0]));
+
+   /* shader header
+    */
+   *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
+
+   header = (struct tgsi_header *) &tokens[1];
+   *header = tgsi_build_header();
+
+   processor = (struct tgsi_processor *) &tokens[2];
+   *processor = tgsi_build_processor( procType, header );
+
+   ti = 3;
+
+   /* declare TEX[0] input */
+   decl = tgsi_default_full_declaration();
+   decl.Declaration.File = TGSI_FILE_INPUT;
+   /* XXX this could be linear... */
+   decl.Declaration.Interpolate = TGSI_INTERPOLATE_PERSPECTIVE;
+   decl.Declaration.Semantic = 1;
+   decl.Semantic.SemanticName = TGSI_SEMANTIC_GENERIC;
+   decl.Semantic.SemanticIndex = 0;
+   decl.DeclarationRange.First = 
+   decl.DeclarationRange.Last = 0;
+   ti += tgsi_build_full_declaration(&decl,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti);
+
+   /* declare color[0] output */
+   decl = tgsi_default_full_declaration();
+   decl.Declaration.File = TGSI_FILE_OUTPUT;
+   decl.Declaration.Semantic = 1;
+   decl.Semantic.SemanticName = TGSI_SEMANTIC_COLOR;
+   decl.Semantic.SemanticIndex = 0;
+   decl.DeclarationRange.First = 
+   decl.DeclarationRange.Last = 0;
+   ti += tgsi_build_full_declaration(&decl,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti);
+
+   /* declare sampler */
+   decl = tgsi_default_full_declaration();
+   decl.Declaration.File = TGSI_FILE_SAMPLER;
+   decl.DeclarationRange.First = 
+   decl.DeclarationRange.Last = 0;
+   ti += tgsi_build_full_declaration(&decl,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti);
+
+   /* TEX instruction */
+   inst = tgsi_default_full_instruction();
+   inst.Instruction.Opcode = TGSI_OPCODE_TEX;
+   inst.Instruction.NumDstRegs = 1;
+   inst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_OUTPUT;
+   inst.FullDstRegisters[0].DstRegister.Index = 0;
+   inst.Instruction.NumSrcRegs = 2;
+   inst.InstructionExtTexture.Texture = TGSI_TEXTURE_2D;
+   inst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT;
+   inst.FullSrcRegisters[0].SrcRegister.Index = 0;
+   inst.FullSrcRegisters[1].SrcRegister.File = TGSI_FILE_SAMPLER;
+   inst.FullSrcRegisters[1].SrcRegister.Index = 0;
+   ti += tgsi_build_full_instruction(&inst,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti );
+
+   /* END instruction */
+   inst = tgsi_default_full_instruction();
+   inst.Instruction.Opcode = TGSI_OPCODE_END;
+   inst.Instruction.NumDstRegs = 0;
+   inst.Instruction.NumSrcRegs = 0;
+   ti += tgsi_build_full_instruction(&inst,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti );
+
+#if 0 /*debug*/
+   tgsi_dump(tokens, 0);
+#endif
+
+   shader->tokens = tokens;
+   /*shader->num_tokens = ti;*/
+
+   return pipe->create_fs_state(pipe, shader);
+}
+
+
+
+
+
+/**
+ * Make simple fragment color pass-through shader.
+ */
+void *
+util_make_fragment_passthrough_shader(struct pipe_context *pipe,
+                                      struct pipe_shader_state *shader)
+{
+   uint maxTokens = 40;
+   struct tgsi_token *tokens;
+   struct tgsi_header *header;
+   struct tgsi_processor *processor;
+   struct tgsi_full_declaration decl;
+   struct tgsi_full_instruction inst;
+   const uint procType = TGSI_PROCESSOR_FRAGMENT;
+   uint ti;
+
+   tokens = (struct tgsi_token *) MALLOC(maxTokens * sizeof(tokens[0]));
+
+   /* shader header
+    */
+   *(struct tgsi_version *) &tokens[0] = tgsi_build_version();
+
+   header = (struct tgsi_header *) &tokens[1];
+   *header = tgsi_build_header();
+
+   processor = (struct tgsi_processor *) &tokens[2];
+   *processor = tgsi_build_processor( procType, header );
+
+   ti = 3;
+
+   /* declare input */
+   decl = tgsi_default_full_declaration();
+   decl.Declaration.File = TGSI_FILE_INPUT;
+   decl.Declaration.Semantic = 1;
+   decl.Semantic.SemanticName = TGSI_SEMANTIC_COLOR;
+   decl.Semantic.SemanticIndex = 0;
+   decl.DeclarationRange.First = 
+      decl.DeclarationRange.Last = 0;
+   ti += tgsi_build_full_declaration(&decl,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti);
+
+   /* declare output */
+   decl = tgsi_default_full_declaration();
+   decl.Declaration.File = TGSI_FILE_OUTPUT;
+   decl.Declaration.Semantic = 1;
+   decl.Semantic.SemanticName = TGSI_SEMANTIC_COLOR;
+   decl.Semantic.SemanticIndex = 0;
+   decl.DeclarationRange.First = 
+      decl.DeclarationRange.Last = 0;
+   ti += tgsi_build_full_declaration(&decl,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti);
+
+
+   /* MOVE out[0], in[0]; */
+   inst = tgsi_default_full_instruction();
+   inst.Instruction.Opcode = TGSI_OPCODE_MOV;
+   inst.Instruction.NumDstRegs = 1;
+   inst.FullDstRegisters[0].DstRegister.File = TGSI_FILE_OUTPUT;
+   inst.FullDstRegisters[0].DstRegister.Index = 0;
+   inst.Instruction.NumSrcRegs = 1;
+   inst.FullSrcRegisters[0].SrcRegister.File = TGSI_FILE_INPUT;
+   inst.FullSrcRegisters[0].SrcRegister.Index = 0;
+   ti += tgsi_build_full_instruction(&inst,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti );
+
+   /* END instruction */
+   inst = tgsi_default_full_instruction();
+   inst.Instruction.Opcode = TGSI_OPCODE_END;
+   inst.Instruction.NumDstRegs = 0;
+   inst.Instruction.NumSrcRegs = 0;
+   ti += tgsi_build_full_instruction(&inst,
+                                     &tokens[ti],
+                                     header,
+                                     maxTokens - ti );
+
+   assert(ti < maxTokens);
+
+#if 0 /*debug*/
+   tgsi_dump(tokens, 0);
+#endif
+
+   shader->tokens = tokens;
+   /*shader->num_tokens = ti;*/
+
+   return pipe->create_fs_state(pipe, shader);
+}
+
diff --git a/src/gallium/auxiliary/util/u_simple_shaders.h b/src/gallium/auxiliary/util/u_simple_shaders.h
new file mode 100644
index 00000000000..8ca4977d715
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_simple_shaders.h
@@ -0,0 +1,68 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+
+#ifndef U_SIMPLE_SHADERS_H
+#define U_SIMPLE_SHADERS_H
+
+
+#include "pipe/p_compiler.h"
+
+
+struct pipe_context;
+struct pipe_shader_state;
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+extern void *
+util_make_vertex_passthrough_shader(struct pipe_context *pipe,
+                                    uint num_attribs,
+                                    const uint *semantic_names,
+                                    const uint *semantic_indexes,
+                                    struct pipe_shader_state *shader);
+
+
+extern void *
+util_make_fragment_tex_shader(struct pipe_context *pipe,
+                              struct pipe_shader_state *shader);
+
+
+extern void *
+util_make_fragment_passthrough_shader(struct pipe_context *pipe,
+                                      struct pipe_shader_state *shader);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_snprintf.c b/src/gallium/auxiliary/util/u_snprintf.c
new file mode 100644
index 00000000000..0d54299b28a
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_snprintf.c
@@ -0,0 +1,1497 @@
+/*
+ * Copyright (c) 1995 Patrick Powell.
+ *
+ * This code is based on code written by Patrick Powell <[email protected]>.
+ * It may be used for any purpose as long as this notice remains intact on all
+ * source code distributions.
+ */
+
+/*
+ * Copyright (c) 2008 Holger Weiss.
+ *
+ * This version of the code is maintained by Holger Weiss <[email protected]>.
+ * My changes to the code may freely be used, modified and/or redistributed for
+ * any purpose.  It would be nice if additions and fixes to this file (including
+ * trivial code cleanups) would be sent back in order to let me include them in
+ * the version available at <http://www.jhweiss.de/software/snprintf.html>.
+ * However, this is not a requirement for using or redistributing (possibly
+ * modified) versions of this file, nor is leaving this notice intact mandatory.
+ */
+
+/*
+ * History
+ *
+ * 2008-01-20 Holger Weiss <[email protected]> for C99-snprintf 1.1:
+ *
+ * 	Fixed the detection of infinite floating point values on IRIX (and
+ * 	possibly other systems) and applied another few minor cleanups.
+ *
+ * 2008-01-06 Holger Weiss <[email protected]> for C99-snprintf 1.0:
+ *
+ * 	Added a lot of new features, fixed many bugs, and incorporated various
+ * 	improvements done by Andrew Tridgell <[email protected]>, Russ Allbery
+ * 	<[email protected]>, Hrvoje Niksic <[email protected]>, Damien Miller
+ * 	<[email protected]>, and others for the Samba, INN, Wget, and OpenSSH
+ * 	projects.  The additions include: support the "e", "E", "g", "G", and
+ * 	"F" conversion specifiers (and use conversion style "f" or "F" for the
+ * 	still unsupported "a" and "A" specifiers); support the "hh", "ll", "j",
+ * 	"t", and "z" length modifiers; support the "#" flag and the (non-C99)
+ * 	"'" flag; use localeconv(3) (if available) to get both the current
+ * 	locale's decimal point character and the separator between groups of
+ * 	digits; fix the handling of various corner cases of field width and
+ * 	precision specifications; fix various floating point conversion bugs;
+ * 	handle infinite and NaN floating point values; don't attempt to write to
+ * 	the output buffer (which may be NULL) if a size of zero was specified;
+ * 	check for integer overflow of the field width, precision, and return
+ * 	values and during the floating point conversion; use the OUTCHAR() macro
+ * 	instead of a function for better performance; provide asprintf(3) and
+ * 	vasprintf(3) functions; add new test cases.  The replacement functions
+ * 	have been renamed to use an "rpl_" prefix, the function calls in the
+ * 	main project (and in this file) must be redefined accordingly for each
+ * 	replacement function which is needed (by using Autoconf or other means).
+ * 	Various other minor improvements have been applied and the coding style
+ * 	was cleaned up for consistency.
+ *
+ * 2007-07-23 Holger Weiss <[email protected]> for Mutt 1.5.13:
+ *
+ * 	C99 compliant snprintf(3) and vsnprintf(3) functions return the number
+ * 	of characters that would have been written to a sufficiently sized
+ * 	buffer (excluding the '\0').  The original code simply returned the
+ * 	length of the resulting output string, so that's been fixed.
+ *
+ * 1998-03-05 Michael Elkins <[email protected]> for Mutt 0.90.8:
+ *
+ * 	The original code assumed that both snprintf(3) and vsnprintf(3) were
+ * 	missing.  Some systems only have snprintf(3) but not vsnprintf(3), so
+ * 	the code is now broken down under HAVE_SNPRINTF and HAVE_VSNPRINTF.
+ *
+ * 1998-01-27 Thomas Roessler <[email protected]> for Mutt 0.89i:
+ *
+ * 	The PGP code was using unsigned hexadecimal formats.  Unfortunately,
+ * 	unsigned formats simply didn't work.
+ *
+ * 1997-10-22 Brandon Long <[email protected]> for Mutt 0.87.1:
+ *
+ * 	Ok, added some minimal floating point support, which means this probably
+ * 	requires libm on most operating systems.  Don't yet support the exponent
+ * 	(e,E) and sigfig (g,G).  Also, fmtint() was pretty badly broken, it just
+ * 	wasn't being exercised in ways which showed it, so that's been fixed.
+ * 	Also, formatted the code to Mutt conventions, and removed dead code left
+ * 	over from the original.  Also, there is now a builtin-test, run with:
+ * 	gcc -DTEST_SNPRINTF -o snprintf snprintf.c -lm && ./snprintf
+ *
+ * 2996-09-15 Brandon Long <[email protected]> for Mutt 0.43:
+ *
+ * 	This was ugly.  It is still ugly.  I opted out of floating point
+ * 	numbers, but the formatter understands just about everything from the
+ * 	normal C string format, at least as far as I can tell from the Solaris
+ * 	2.5 printf(3S) man page.
+ */
+
+/*
+ * ToDo
+ *
+ * - Add wide character support.
+ * - Add support for "%a" and "%A" conversions.
+ * - Create test routines which predefine the expected results.  Our test cases
+ *   usually expose bugs in system implementations rather than in ours :-)
+ */
+
+/*
+ * Usage
+ *
+ * 1) The following preprocessor macros should be defined to 1 if the feature or
+ *    file in question is available on the target system (by using Autoconf or
+ *    other means), though basic functionality should be available as long as
+ *    HAVE_STDARG_H and HAVE_STDLIB_H are defined correctly:
+ *
+ *    	HAVE_VSNPRINTF
+ *    	HAVE_SNPRINTF
+ *    	HAVE_VASPRINTF
+ *    	HAVE_ASPRINTF
+ *    	HAVE_STDARG_H
+ *    	HAVE_STDDEF_H
+ *    	HAVE_STDINT_H
+ *    	HAVE_STDLIB_H
+ *    	HAVE_INTTYPES_H
+ *    	HAVE_LOCALE_H
+ *    	HAVE_LOCALECONV
+ *    	HAVE_LCONV_DECIMAL_POINT
+ *    	HAVE_LCONV_THOUSANDS_SEP
+ *    	HAVE_LONG_DOUBLE
+ *    	HAVE_LONG_LONG_INT
+ *    	HAVE_UNSIGNED_LONG_LONG_INT
+ *    	HAVE_INTMAX_T
+ *    	HAVE_UINTMAX_T
+ *    	HAVE_UINTPTR_T
+ *    	HAVE_PTRDIFF_T
+ *    	HAVE_VA_COPY
+ *    	HAVE___VA_COPY
+ *
+ * 2) The calls to the functions which should be replaced must be redefined
+ *    throughout the project files (by using Autoconf or other means):
+ *
+ *    	#define vsnprintf rpl_vsnprintf
+ *    	#define snprintf rpl_snprintf
+ *    	#define vasprintf rpl_vasprintf
+ *    	#define asprintf rpl_asprintf
+ *
+ * 3) The required replacement functions should be declared in some header file
+ *    included throughout the project files:
+ *
+ *    	#if HAVE_CONFIG_H
+ *    	#include <config.h>
+ *    	#endif
+ *    	#if HAVE_STDARG_H
+ *    	#include <stdarg.h>
+ *    	#if !HAVE_VSNPRINTF
+ *    	int rpl_vsnprintf(char *, size_t, const char *, va_list);
+ *    	#endif
+ *    	#if !HAVE_SNPRINTF
+ *    	int rpl_snprintf(char *, size_t, const char *, ...);
+ *    	#endif
+ *    	#if !HAVE_VASPRINTF
+ *    	int rpl_vasprintf(char **, const char *, va_list);
+ *    	#endif
+ *    	#if !HAVE_ASPRINTF
+ *    	int rpl_asprintf(char **, const char *, ...);
+ *    	#endif
+ *    	#endif
+ *
+ * Autoconf macros for handling step 1 and step 2 are available at
+ * <http://www.jhweiss.de/software/snprintf.html>.
+ */
+
+#include "pipe/p_config.h"
+
+#if HAVE_CONFIG_H
+#include <config.h>
+#else
+#ifdef WIN32
+#define vsnprintf util_vsnprintf
+#define snprintf util_snprintf
+#define HAVE_VSNPRINTF 0
+#define HAVE_SNPRINTF 0
+#define HAVE_VASPRINTF 1 /* not needed */
+#define HAVE_ASPRINTF 1 /* not needed */
+#define HAVE_STDARG_H 1
+#define HAVE_STDDEF_H 1
+#define HAVE_STDINT_H 0
+#define HAVE_STDLIB_H 1
+#define HAVE_INTTYPES_H 0
+#define HAVE_LOCALE_H 0
+#define HAVE_LOCALECONV 0
+#define HAVE_LCONV_DECIMAL_POINT 0
+#define HAVE_LCONV_THOUSANDS_SEP 0
+#define HAVE_LONG_DOUBLE 0
+#define HAVE_LONG_LONG_INT 1
+#define HAVE_UNSIGNED_LONG_LONG_INT 1
+#define HAVE_INTMAX_T 0
+#define HAVE_UINTMAX_T 0
+#define HAVE_UINTPTR_T 1
+#define HAVE_PTRDIFF_T 1
+#define HAVE_VA_COPY 0
+#define HAVE___VA_COPY 0
+#else
+#define HAVE_VSNPRINTF 1
+#define HAVE_SNPRINTF 1
+#define HAVE_VASPRINTF 1
+#define HAVE_ASPRINTF 1
+#endif
+#endif	/* HAVE_CONFIG_H */
+
+#if !HAVE_SNPRINTF || !HAVE_VSNPRINTF || !HAVE_ASPRINTF || !HAVE_VASPRINTF
+#include <stdio.h>	/* For NULL, size_t, vsnprintf(3), and vasprintf(3). */
+#ifdef VA_START
+#undef VA_START
+#endif	/* defined(VA_START) */
+#ifdef VA_SHIFT
+#undef VA_SHIFT
+#endif	/* defined(VA_SHIFT) */
+#if HAVE_STDARG_H
+#include <stdarg.h>
+#define VA_START(ap, last) va_start(ap, last)
+#define VA_SHIFT(ap, value, type) /* No-op for ANSI C. */
+#else	/* Assume <varargs.h> is available. */
+#include <varargs.h>
+#define VA_START(ap, last) va_start(ap)	/* "last" is ignored. */
+#define VA_SHIFT(ap, value, type) value = va_arg(ap, type)
+#endif	/* HAVE_STDARG_H */
+
+#if !HAVE_VASPRINTF
+#if HAVE_STDLIB_H
+#include <stdlib.h>	/* For malloc(3). */
+#endif	/* HAVE_STDLIB_H */
+#ifdef VA_COPY
+#undef VA_COPY
+#endif	/* defined(VA_COPY) */
+#ifdef VA_END_COPY
+#undef VA_END_COPY
+#endif	/* defined(VA_END_COPY) */
+#if HAVE_VA_COPY
+#define VA_COPY(dest, src) va_copy(dest, src)
+#define VA_END_COPY(ap) va_end(ap)
+#elif HAVE___VA_COPY
+#define VA_COPY(dest, src) __va_copy(dest, src)
+#define VA_END_COPY(ap) va_end(ap)
+#else
+#define VA_COPY(dest, src) (void)mymemcpy(&dest, &src, sizeof(va_list))
+#define VA_END_COPY(ap) /* No-op. */
+#define NEED_MYMEMCPY 1
+static void *mymemcpy(void *, void *, size_t);
+#endif	/* HAVE_VA_COPY */
+#endif	/* !HAVE_VASPRINTF */
+
+#if !HAVE_VSNPRINTF
+#include <limits.h>	/* For *_MAX. */
+#if HAVE_INTTYPES_H
+#include <inttypes.h>	/* For intmax_t (if not defined in <stdint.h>). */
+#endif	/* HAVE_INTTYPES_H */
+#if HAVE_LOCALE_H
+#include <locale.h>	/* For localeconv(3). */
+#endif	/* HAVE_LOCALE_H */
+#if HAVE_STDDEF_H
+#include <stddef.h>	/* For ptrdiff_t. */
+#endif	/* HAVE_STDDEF_H */
+#if HAVE_STDINT_H
+#include <stdint.h>	/* For intmax_t. */
+#endif	/* HAVE_STDINT_H */
+
+/* Support for unsigned long long int.  We may also need ULLONG_MAX. */
+#ifndef ULONG_MAX	/* We may need ULONG_MAX as a fallback. */
+#ifdef UINT_MAX
+#define ULONG_MAX UINT_MAX
+#else
+#define ULONG_MAX INT_MAX
+#endif	/* defined(UINT_MAX) */
+#endif	/* !defined(ULONG_MAX) */
+#ifdef ULLONG
+#undef ULLONG
+#endif	/* defined(ULLONG) */
+#if HAVE_UNSIGNED_LONG_LONG_INT
+#define ULLONG unsigned long long int
+#ifndef ULLONG_MAX
+#define ULLONG_MAX ULONG_MAX
+#endif	/* !defined(ULLONG_MAX) */
+#else
+#define ULLONG unsigned long int
+#ifdef ULLONG_MAX
+#undef ULLONG_MAX
+#endif	/* defined(ULLONG_MAX) */
+#define ULLONG_MAX ULONG_MAX
+#endif	/* HAVE_LONG_LONG_INT */
+
+/* Support for uintmax_t.  We also need UINTMAX_MAX. */
+#ifdef UINTMAX_T
+#undef UINTMAX_T
+#endif	/* defined(UINTMAX_T) */
+#if HAVE_UINTMAX_T || defined(uintmax_t)
+#define UINTMAX_T uintmax_t
+#ifndef UINTMAX_MAX
+#define UINTMAX_MAX ULLONG_MAX
+#endif	/* !defined(UINTMAX_MAX) */
+#else
+#define UINTMAX_T ULLONG
+#ifdef UINTMAX_MAX
+#undef UINTMAX_MAX
+#endif	/* defined(UINTMAX_MAX) */
+#define UINTMAX_MAX ULLONG_MAX
+#endif	/* HAVE_UINTMAX_T || defined(uintmax_t) */
+
+/* Support for long double. */
+#ifndef LDOUBLE
+#if HAVE_LONG_DOUBLE
+#define LDOUBLE long double
+#else
+#define LDOUBLE double
+#endif	/* HAVE_LONG_DOUBLE */
+#endif	/* !defined(LDOUBLE) */
+
+/* Support for long long int. */
+#ifndef LLONG
+#if HAVE_LONG_LONG_INT
+#define LLONG long long int
+#else
+#define LLONG long int
+#endif	/* HAVE_LONG_LONG_INT */
+#endif	/* !defined(LLONG) */
+
+/* Support for intmax_t. */
+#ifndef INTMAX_T
+#if HAVE_INTMAX_T || defined(intmax_t)
+#define INTMAX_T intmax_t
+#else
+#define INTMAX_T LLONG
+#endif	/* HAVE_INTMAX_T || defined(intmax_t) */
+#endif	/* !defined(INTMAX_T) */
+
+/* Support for uintptr_t. */
+#ifndef UINTPTR_T
+#if HAVE_UINTPTR_T || defined(uintptr_t)
+#define UINTPTR_T uintptr_t
+#else
+#define UINTPTR_T unsigned long int
+#endif	/* HAVE_UINTPTR_T || defined(uintptr_t) */
+#endif	/* !defined(UINTPTR_T) */
+
+/* WinCE5.0 does not have uintptr_t defined */ 
+#if (_WIN32_WCE < 600) 
+#ifdef UINTPTR_T 
+#undef UINTPTR_T 
+#endif 
+#define UINTPTR_T unsigned long int 
+#endif 
+
+
+/* Support for ptrdiff_t. */
+#ifndef PTRDIFF_T
+#if HAVE_PTRDIFF_T || defined(ptrdiff_t)
+#define PTRDIFF_T ptrdiff_t
+#else
+#define PTRDIFF_T long int
+#endif	/* HAVE_PTRDIFF_T || defined(ptrdiff_t) */
+#endif	/* !defined(PTRDIFF_T) */
+
+/*
+ * We need an unsigned integer type corresponding to ptrdiff_t (cf. C99:
+ * 7.19.6.1, 7).  However, we'll simply use PTRDIFF_T and convert it to an
+ * unsigned type if necessary.  This should work just fine in practice.
+ */
+#ifndef UPTRDIFF_T
+#define UPTRDIFF_T PTRDIFF_T
+#endif	/* !defined(UPTRDIFF_T) */
+
+/*
+ * We need a signed integer type corresponding to size_t (cf. C99: 7.19.6.1, 7).
+ * However, we'll simply use size_t and convert it to a signed type if
+ * necessary.  This should work just fine in practice.
+ */
+#ifndef SSIZE_T
+#define SSIZE_T size_t
+#endif	/* !defined(SSIZE_T) */
+
+/* Either ERANGE or E2BIG should be available everywhere. */
+#ifndef ERANGE
+#define ERANGE E2BIG
+#endif	/* !defined(ERANGE) */
+#ifndef EOVERFLOW
+#define EOVERFLOW ERANGE
+#endif	/* !defined(EOVERFLOW) */
+
+/*
+ * Buffer size to hold the octal string representation of UINT128_MAX without
+ * nul-termination ("3777777777777777777777777777777777777777777").
+ */
+#ifdef MAX_CONVERT_LENGTH
+#undef MAX_CONVERT_LENGTH
+#endif	/* defined(MAX_CONVERT_LENGTH) */
+#define MAX_CONVERT_LENGTH      43
+
+/* Format read states. */
+#define PRINT_S_DEFAULT         0
+#define PRINT_S_FLAGS           1
+#define PRINT_S_WIDTH           2
+#define PRINT_S_DOT             3
+#define PRINT_S_PRECISION       4
+#define PRINT_S_MOD             5
+#define PRINT_S_CONV            6
+
+/* Format flags. */
+#define PRINT_F_MINUS           (1 << 0)
+#define PRINT_F_PLUS            (1 << 1)
+#define PRINT_F_SPACE           (1 << 2)
+#define PRINT_F_NUM             (1 << 3)
+#define PRINT_F_ZERO            (1 << 4)
+#define PRINT_F_QUOTE           (1 << 5)
+#define PRINT_F_UP              (1 << 6)
+#define PRINT_F_UNSIGNED        (1 << 7)
+#define PRINT_F_TYPE_G          (1 << 8)
+#define PRINT_F_TYPE_E          (1 << 9)
+
+/* Conversion flags. */
+#define PRINT_C_CHAR            1
+#define PRINT_C_SHORT           2
+#define PRINT_C_LONG            3
+#define PRINT_C_LLONG           4
+#define PRINT_C_LDOUBLE         5
+#define PRINT_C_SIZE            6
+#define PRINT_C_PTRDIFF         7
+#define PRINT_C_INTMAX          8
+
+#ifndef MAX
+#define MAX(x, y) ((x >= y) ? x : y)
+#endif	/* !defined(MAX) */
+#ifndef CHARTOINT
+#define CHARTOINT(ch) (ch - '0')
+#endif	/* !defined(CHARTOINT) */
+#ifndef ISDIGIT
+#define ISDIGIT(ch) ('0' <= (unsigned char)ch && (unsigned char)ch <= '9')
+#endif	/* !defined(ISDIGIT) */
+#ifndef ISNAN
+#define ISNAN(x) (x != x)
+#endif	/* !defined(ISNAN) */
+#ifndef ISINF
+#define ISINF(x) (x != 0.0 && x + x == x)
+#endif	/* !defined(ISINF) */
+
+#ifdef OUTCHAR
+#undef OUTCHAR
+#endif	/* defined(OUTCHAR) */
+#define OUTCHAR(str, len, size, ch)                                          \
+do {                                                                         \
+	if (len + 1 < size)                                                  \
+		str[len] = ch;                                               \
+	(len)++;                                                             \
+} while (/* CONSTCOND */ 0)
+
+static void fmtstr(char *, size_t *, size_t, const char *, int, int, int);
+static void fmtint(char *, size_t *, size_t, INTMAX_T, int, int, int, int);
+static void fmtflt(char *, size_t *, size_t, LDOUBLE, int, int, int, int *);
+static void printsep(char *, size_t *, size_t);
+static int getnumsep(int);
+static int getexponent(LDOUBLE);
+static int convert(UINTMAX_T, char *, size_t, int, int);
+static UINTMAX_T cast(LDOUBLE);
+static UINTMAX_T myround(LDOUBLE);
+static LDOUBLE mypow10(int);
+
+int
+util_vsnprintf(char *str, size_t size, const char *format, va_list args)
+{
+	LDOUBLE fvalue;
+	INTMAX_T value;
+	unsigned char cvalue;
+	const char *strvalue;
+	INTMAX_T *intmaxptr;
+	PTRDIFF_T *ptrdiffptr;
+	SSIZE_T *sizeptr;
+	LLONG *llongptr;
+	long int *longptr;
+	int *intptr;
+	short int *shortptr;
+	signed char *charptr;
+	size_t len = 0;
+	int overflow = 0;
+	int base = 0;
+	int cflags = 0;
+	int flags = 0;
+	int width = 0;
+	int precision = -1;
+	int state = PRINT_S_DEFAULT;
+	char ch = *format++;
+
+	/*
+	 * C99 says: "If `n' is zero, nothing is written, and `s' may be a null
+	 * pointer." (7.19.6.5, 2)  We're forgiving and allow a NULL pointer
+	 * even if a size larger than zero was specified.  At least NetBSD's
+	 * snprintf(3) does the same, as well as other versions of this file.
+	 * (Though some of these versions will write to a non-NULL buffer even
+	 * if a size of zero was specified, which violates the standard.)
+	 */
+	if (str == NULL && size != 0)
+		size = 0;
+
+	while (ch != '\0')
+		switch (state) {
+		case PRINT_S_DEFAULT:
+			if (ch == '%')
+				state = PRINT_S_FLAGS;
+			else
+				OUTCHAR(str, len, size, ch);
+			ch = *format++;
+			break;
+		case PRINT_S_FLAGS:
+			switch (ch) {
+			case '-':
+				flags |= PRINT_F_MINUS;
+				ch = *format++;
+				break;
+			case '+':
+				flags |= PRINT_F_PLUS;
+				ch = *format++;
+				break;
+			case ' ':
+				flags |= PRINT_F_SPACE;
+				ch = *format++;
+				break;
+			case '#':
+				flags |= PRINT_F_NUM;
+				ch = *format++;
+				break;
+			case '0':
+				flags |= PRINT_F_ZERO;
+				ch = *format++;
+				break;
+			case '\'':	/* SUSv2 flag (not in C99). */
+				flags |= PRINT_F_QUOTE;
+				ch = *format++;
+				break;
+			default:
+				state = PRINT_S_WIDTH;
+				break;
+			}
+			break;
+		case PRINT_S_WIDTH:
+			if (ISDIGIT(ch)) {
+				ch = CHARTOINT(ch);
+				if (width > (INT_MAX - ch) / 10) {
+					overflow = 1;
+					goto out;
+				}
+				width = 10 * width + ch;
+				ch = *format++;
+			} else if (ch == '*') {
+				/*
+				 * C99 says: "A negative field width argument is
+				 * taken as a `-' flag followed by a positive
+				 * field width." (7.19.6.1, 5)
+				 */
+				if ((width = va_arg(args, int)) < 0) {
+					flags |= PRINT_F_MINUS;
+					width = -width;
+				}
+				ch = *format++;
+				state = PRINT_S_DOT;
+			} else
+				state = PRINT_S_DOT;
+			break;
+		case PRINT_S_DOT:
+			if (ch == '.') {
+				state = PRINT_S_PRECISION;
+				ch = *format++;
+			} else
+				state = PRINT_S_MOD;
+			break;
+		case PRINT_S_PRECISION:
+			if (precision == -1)
+				precision = 0;
+			if (ISDIGIT(ch)) {
+				ch = CHARTOINT(ch);
+				if (precision > (INT_MAX - ch) / 10) {
+					overflow = 1;
+					goto out;
+				}
+				precision = 10 * precision + ch;
+				ch = *format++;
+			} else if (ch == '*') {
+				/*
+				 * C99 says: "A negative precision argument is
+				 * taken as if the precision were omitted."
+				 * (7.19.6.1, 5)
+				 */
+				if ((precision = va_arg(args, int)) < 0)
+					precision = -1;
+				ch = *format++;
+				state = PRINT_S_MOD;
+			} else
+				state = PRINT_S_MOD;
+			break;
+		case PRINT_S_MOD:
+			switch (ch) {
+			case 'h':
+				ch = *format++;
+				if (ch == 'h') {	/* It's a char. */
+					ch = *format++;
+					cflags = PRINT_C_CHAR;
+				} else
+					cflags = PRINT_C_SHORT;
+				break;
+			case 'l':
+				ch = *format++;
+				if (ch == 'l') {	/* It's a long long. */
+					ch = *format++;
+					cflags = PRINT_C_LLONG;
+				} else
+					cflags = PRINT_C_LONG;
+				break;
+			case 'L':
+				cflags = PRINT_C_LDOUBLE;
+				ch = *format++;
+				break;
+			case 'j':
+				cflags = PRINT_C_INTMAX;
+				ch = *format++;
+				break;
+			case 't':
+				cflags = PRINT_C_PTRDIFF;
+				ch = *format++;
+				break;
+			case 'z':
+				cflags = PRINT_C_SIZE;
+				ch = *format++;
+				break;
+			}
+			state = PRINT_S_CONV;
+			break;
+		case PRINT_S_CONV:
+			switch (ch) {
+			case 'd':
+				/* FALLTHROUGH */
+			case 'i':
+				switch (cflags) {
+				case PRINT_C_CHAR:
+					value = (signed char)va_arg(args, int);
+					break;
+				case PRINT_C_SHORT:
+					value = (short int)va_arg(args, int);
+					break;
+				case PRINT_C_LONG:
+					value = va_arg(args, long int);
+					break;
+				case PRINT_C_LLONG:
+					value = va_arg(args, LLONG);
+					break;
+				case PRINT_C_SIZE:
+					value = va_arg(args, SSIZE_T);
+					break;
+				case PRINT_C_INTMAX:
+					value = va_arg(args, INTMAX_T);
+					break;
+				case PRINT_C_PTRDIFF:
+					value = va_arg(args, PTRDIFF_T);
+					break;
+				default:
+					value = va_arg(args, int);
+					break;
+				}
+				fmtint(str, &len, size, value, 10, width,
+				    precision, flags);
+				break;
+			case 'X':
+				flags |= PRINT_F_UP;
+				/* FALLTHROUGH */
+			case 'x':
+				base = 16;
+				/* FALLTHROUGH */
+			case 'o':
+				if (base == 0)
+					base = 8;
+				/* FALLTHROUGH */
+			case 'u':
+				if (base == 0)
+					base = 10;
+				flags |= PRINT_F_UNSIGNED;
+				switch (cflags) {
+				case PRINT_C_CHAR:
+					value = (unsigned char)va_arg(args,
+					    unsigned int);
+					break;
+				case PRINT_C_SHORT:
+					value = (unsigned short int)va_arg(args,
+					    unsigned int);
+					break;
+				case PRINT_C_LONG:
+					value = va_arg(args, unsigned long int);
+					break;
+				case PRINT_C_LLONG:
+					value = va_arg(args, ULLONG);
+					break;
+				case PRINT_C_SIZE:
+					value = va_arg(args, size_t);
+					break;
+				case PRINT_C_INTMAX:
+					value = va_arg(args, UINTMAX_T);
+					break;
+				case PRINT_C_PTRDIFF:
+					value = va_arg(args, UPTRDIFF_T);
+					break;
+				default:
+					value = va_arg(args, unsigned int);
+					break;
+				}
+				fmtint(str, &len, size, value, base, width,
+				    precision, flags);
+				break;
+			case 'A':
+				/* Not yet supported, we'll use "%F". */
+				/* FALLTHROUGH */
+			case 'F':
+				flags |= PRINT_F_UP;
+			case 'a':
+				/* Not yet supported, we'll use "%f". */
+				/* FALLTHROUGH */
+			case 'f':
+				if (cflags == PRINT_C_LDOUBLE)
+					fvalue = va_arg(args, LDOUBLE);
+				else
+					fvalue = va_arg(args, double);
+				fmtflt(str, &len, size, fvalue, width,
+				    precision, flags, &overflow);
+				if (overflow)
+					goto out;
+				break;
+			case 'E':
+				flags |= PRINT_F_UP;
+				/* FALLTHROUGH */
+			case 'e':
+				flags |= PRINT_F_TYPE_E;
+				if (cflags == PRINT_C_LDOUBLE)
+					fvalue = va_arg(args, LDOUBLE);
+				else
+					fvalue = va_arg(args, double);
+				fmtflt(str, &len, size, fvalue, width,
+				    precision, flags, &overflow);
+				if (overflow)
+					goto out;
+				break;
+			case 'G':
+				flags |= PRINT_F_UP;
+				/* FALLTHROUGH */
+			case 'g':
+				flags |= PRINT_F_TYPE_G;
+				if (cflags == PRINT_C_LDOUBLE)
+					fvalue = va_arg(args, LDOUBLE);
+				else
+					fvalue = va_arg(args, double);
+				/*
+				 * If the precision is zero, it is treated as
+				 * one (cf. C99: 7.19.6.1, 8).
+				 */
+				if (precision == 0)
+					precision = 1;
+				fmtflt(str, &len, size, fvalue, width,
+				    precision, flags, &overflow);
+				if (overflow)
+					goto out;
+				break;
+			case 'c':
+				cvalue = (unsigned char)va_arg(args, int);
+				OUTCHAR(str, len, size, cvalue);
+				break;
+			case 's':
+				strvalue = va_arg(args, char *);
+				fmtstr(str, &len, size, strvalue, width,
+				    precision, flags);
+				break;
+			case 'p':
+				/*
+				 * C99 says: "The value of the pointer is
+				 * converted to a sequence of printing
+				 * characters, in an implementation-defined
+				 * manner." (C99: 7.19.6.1, 8)
+				 */
+				if ((strvalue = va_arg(args, void *)) == NULL)
+					/*
+					 * We use the glibc format.  BSD prints
+					 * "0x0", SysV "0".
+					 */
+					fmtstr(str, &len, size, "(nil)", width,
+					    -1, flags);
+				else {
+					/*
+					 * We use the BSD/glibc format.  SysV
+					 * omits the "0x" prefix (which we emit
+					 * using the PRINT_F_NUM flag).
+					 */
+					flags |= PRINT_F_NUM;
+					flags |= PRINT_F_UNSIGNED;
+					fmtint(str, &len, size,
+					    (UINTPTR_T)strvalue, 16, width,
+					    precision, flags);
+				}
+				break;
+			case 'n':
+				switch (cflags) {
+				case PRINT_C_CHAR:
+					charptr = va_arg(args, signed char *);
+					*charptr = (signed char)len;
+					break;
+				case PRINT_C_SHORT:
+					shortptr = va_arg(args, short int *);
+					*shortptr = (short int)len;
+					break;
+				case PRINT_C_LONG:
+					longptr = va_arg(args, long int *);
+					*longptr = (long int)len;
+					break;
+				case PRINT_C_LLONG:
+					llongptr = va_arg(args, LLONG *);
+					*llongptr = (LLONG)len;
+					break;
+				case PRINT_C_SIZE:
+					/*
+					 * C99 says that with the "z" length
+					 * modifier, "a following `n' conversion
+					 * specifier applies to a pointer to a
+					 * signed integer type corresponding to
+					 * size_t argument." (7.19.6.1, 7)
+					 */
+					sizeptr = va_arg(args, SSIZE_T *);
+					*sizeptr = len;
+					break;
+				case PRINT_C_INTMAX:
+					intmaxptr = va_arg(args, INTMAX_T *);
+					*intmaxptr = len;
+					break;
+				case PRINT_C_PTRDIFF:
+					ptrdiffptr = va_arg(args, PTRDIFF_T *);
+					*ptrdiffptr = len;
+					break;
+				default:
+					intptr = va_arg(args, int *);
+					*intptr = len;
+					break;
+				}
+				break;
+			case '%':	/* Print a "%" character verbatim. */
+				OUTCHAR(str, len, size, ch);
+				break;
+			default:	/* Skip other characters. */
+				break;
+			}
+			ch = *format++;
+			state = PRINT_S_DEFAULT;
+			base = cflags = flags = width = 0;
+			precision = -1;
+			break;
+		}
+out:
+	if (len < size)
+		str[len] = '\0';
+	else if (size > 0)
+		str[size - 1] = '\0';
+
+	if (overflow || len >= INT_MAX) {
+		return -1;
+	}
+	return (int)len;
+}
+
+static void
+fmtstr(char *str, size_t *len, size_t size, const char *value, int width,
+       int precision, int flags)
+{
+	int padlen, strln;	/* Amount to pad. */
+	int noprecision = (precision == -1);
+
+	if (value == NULL)	/* We're forgiving. */
+		value = "(null)";
+
+	/* If a precision was specified, don't read the string past it. */
+	for (strln = 0; value[strln] != '\0' &&
+	    (noprecision || strln < precision); strln++)
+		continue;
+
+	if ((padlen = width - strln) < 0)
+		padlen = 0;
+	if (flags & PRINT_F_MINUS)	/* Left justify. */
+		padlen = -padlen;
+
+	while (padlen > 0) {	/* Leading spaces. */
+		OUTCHAR(str, *len, size, ' ');
+		padlen--;
+	}
+	while (*value != '\0' && (noprecision || precision-- > 0)) {
+		OUTCHAR(str, *len, size, *value);
+		value++;
+	}
+	while (padlen < 0) {	/* Trailing spaces. */
+		OUTCHAR(str, *len, size, ' ');
+		padlen++;
+	}
+}
+
+static void
+fmtint(char *str, size_t *len, size_t size, INTMAX_T value, int base, int width,
+       int precision, int flags)
+{
+	UINTMAX_T uvalue;
+	char iconvert[MAX_CONVERT_LENGTH];
+	char sign = 0;
+	char hexprefix = 0;
+	int spadlen = 0;	/* Amount to space pad. */
+	int zpadlen = 0;	/* Amount to zero pad. */
+	int pos;
+	int separators = (flags & PRINT_F_QUOTE);
+	int noprecision = (precision == -1);
+
+	if (flags & PRINT_F_UNSIGNED)
+		uvalue = value;
+	else {
+		uvalue = (value >= 0) ? value : -value;
+		if (value < 0)
+			sign = '-';
+		else if (flags & PRINT_F_PLUS)	/* Do a sign. */
+			sign = '+';
+		else if (flags & PRINT_F_SPACE)
+			sign = ' ';
+	}
+
+	pos = convert(uvalue, iconvert, sizeof(iconvert), base,
+	    flags & PRINT_F_UP);
+
+	if (flags & PRINT_F_NUM && uvalue != 0) {
+		/*
+		 * C99 says: "The result is converted to an `alternative form'.
+		 * For `o' conversion, it increases the precision, if and only
+		 * if necessary, to force the first digit of the result to be a
+		 * zero (if the value and precision are both 0, a single 0 is
+		 * printed).  For `x' (or `X') conversion, a nonzero result has
+		 * `0x' (or `0X') prefixed to it." (7.19.6.1, 6)
+		 */
+		switch (base) {
+		case 8:
+			if (precision <= pos)
+				precision = pos + 1;
+			break;
+		case 16:
+			hexprefix = (flags & PRINT_F_UP) ? 'X' : 'x';
+			break;
+		}
+	}
+
+	if (separators)	/* Get the number of group separators we'll print. */
+		separators = getnumsep(pos);
+
+	zpadlen = precision - pos - separators;
+	spadlen = width                         /* Minimum field width. */
+	    - separators                        /* Number of separators. */
+	    - MAX(precision, pos)               /* Number of integer digits. */
+	    - ((sign != 0) ? 1 : 0)             /* Will we print a sign? */
+	    - ((hexprefix != 0) ? 2 : 0);       /* Will we print a prefix? */
+
+	if (zpadlen < 0)
+		zpadlen = 0;
+	if (spadlen < 0)
+		spadlen = 0;
+
+	/*
+	 * C99 says: "If the `0' and `-' flags both appear, the `0' flag is
+	 * ignored.  For `d', `i', `o', `u', `x', and `X' conversions, if a
+	 * precision is specified, the `0' flag is ignored." (7.19.6.1, 6)
+	 */
+	if (flags & PRINT_F_MINUS)	/* Left justify. */
+		spadlen = -spadlen;
+	else if (flags & PRINT_F_ZERO && noprecision) {
+		zpadlen += spadlen;
+		spadlen = 0;
+	}
+	while (spadlen > 0) {	/* Leading spaces. */
+		OUTCHAR(str, *len, size, ' ');
+		spadlen--;
+	}
+	if (sign != 0)	/* Sign. */
+		OUTCHAR(str, *len, size, sign);
+	if (hexprefix != 0) {	/* A "0x" or "0X" prefix. */
+		OUTCHAR(str, *len, size, '0');
+		OUTCHAR(str, *len, size, hexprefix);
+	}
+	while (zpadlen > 0) {	/* Leading zeros. */
+		OUTCHAR(str, *len, size, '0');
+		zpadlen--;
+	}
+	while (pos > 0) {	/* The actual digits. */
+		pos--;
+		OUTCHAR(str, *len, size, iconvert[pos]);
+		if (separators > 0 && pos > 0 && pos % 3 == 0)
+			printsep(str, len, size);
+	}
+	while (spadlen < 0) {	/* Trailing spaces. */
+		OUTCHAR(str, *len, size, ' ');
+		spadlen++;
+	}
+}
+
+static void
+fmtflt(char *str, size_t *len, size_t size, LDOUBLE fvalue, int width,
+       int precision, int flags, int *overflow)
+{
+	LDOUBLE ufvalue;
+	UINTMAX_T intpart;
+	UINTMAX_T fracpart;
+	UINTMAX_T mask;
+	const char *infnan = NULL;
+	char iconvert[MAX_CONVERT_LENGTH];
+	char fconvert[MAX_CONVERT_LENGTH];
+	char econvert[4];	/* "e-12" (without nul-termination). */
+	char esign = 0;
+	char sign = 0;
+	int leadfraczeros = 0;
+	int exponent = 0;
+	int emitpoint = 0;
+	int omitzeros = 0;
+	int omitcount = 0;
+	int padlen = 0;
+	int epos = 0;
+	int fpos = 0;
+	int ipos = 0;
+	int separators = (flags & PRINT_F_QUOTE);
+	int estyle = (flags & PRINT_F_TYPE_E);
+#if HAVE_LOCALECONV && HAVE_LCONV_DECIMAL_POINT
+	struct lconv *lc = localeconv();
+#endif	/* HAVE_LOCALECONV && HAVE_LCONV_DECIMAL_POINT */
+
+	/*
+	 * AIX' man page says the default is 0, but C99 and at least Solaris'
+	 * and NetBSD's man pages say the default is 6, and sprintf(3) on AIX
+	 * defaults to 6.
+	 */
+	if (precision == -1)
+		precision = 6;
+
+	if (fvalue < 0.0)
+		sign = '-';
+	else if (flags & PRINT_F_PLUS)	/* Do a sign. */
+		sign = '+';
+	else if (flags & PRINT_F_SPACE)
+		sign = ' ';
+
+	if (ISNAN(fvalue))
+		infnan = (flags & PRINT_F_UP) ? "NAN" : "nan";
+	else if (ISINF(fvalue))
+		infnan = (flags & PRINT_F_UP) ? "INF" : "inf";
+
+	if (infnan != NULL) {
+		if (sign != 0)
+			iconvert[ipos++] = sign;
+		while (*infnan != '\0')
+			iconvert[ipos++] = *infnan++;
+		fmtstr(str, len, size, iconvert, width, ipos, flags);
+		return;
+	}
+
+	/* "%e" (or "%E") or "%g" (or "%G") conversion. */
+	if (flags & PRINT_F_TYPE_E || flags & PRINT_F_TYPE_G) {
+		if (flags & PRINT_F_TYPE_G) {
+			/*
+			 * For "%g" (and "%G") conversions, the precision
+			 * specifies the number of significant digits, which
+			 * includes the digits in the integer part.  The
+			 * conversion will or will not be using "e-style" (like
+			 * "%e" or "%E" conversions) depending on the precision
+			 * and on the exponent.  However, the exponent can be
+			 * affected by rounding the converted value, so we'll
+			 * leave this decision for later.  Until then, we'll
+			 * assume that we're going to do an "e-style" conversion
+			 * (in order to get the exponent calculated).  For
+			 * "e-style", the precision must be decremented by one.
+			 */
+			precision--;
+			/*
+			 * For "%g" (and "%G") conversions, trailing zeros are
+			 * removed from the fractional portion of the result
+			 * unless the "#" flag was specified.
+			 */
+			if (!(flags & PRINT_F_NUM))
+				omitzeros = 1;
+		}
+		exponent = getexponent(fvalue);
+		estyle = 1;
+	}
+
+again:
+	/*
+	 * Sorry, we only support 9, 19, or 38 digits (that is, the number of
+	 * digits of the 32-bit, the 64-bit, or the 128-bit UINTMAX_MAX value
+	 * minus one) past the decimal point due to our conversion method.
+	 */
+	switch (sizeof(UINTMAX_T)) {
+	case 16:
+		if (precision > 38)
+			precision = 38;
+		break;
+	case 8:
+		if (precision > 19)
+			precision = 19;
+		break;
+	default:
+		if (precision > 9)
+			precision = 9;
+		break;
+	}
+
+	ufvalue = (fvalue >= 0.0) ? fvalue : -fvalue;
+	if (estyle)	/* We want exactly one integer digit. */
+		ufvalue /= mypow10(exponent);
+
+	if ((intpart = cast(ufvalue)) == UINTMAX_MAX) {
+		*overflow = 1;
+		return;
+	}
+
+	/*
+	 * Factor of ten with the number of digits needed for the fractional
+	 * part.  For example, if the precision is 3, the mask will be 1000.
+	 */
+#if defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT)
+	mask = (unsigned long)mypow10(precision);
+#else
+	mask = (UINTMAX_T)mypow10(precision);
+#endif
+	/*
+	 * We "cheat" by converting the fractional part to integer by
+	 * multiplying by a factor of ten.
+	 */
+	if ((fracpart = myround(mask * (ufvalue - intpart))) >= mask) {
+		/*
+		 * For example, ufvalue = 2.99962, intpart = 2, and mask = 1000
+		 * (because precision = 3).  Now, myround(1000 * 0.99962) will
+		 * return 1000.  So, the integer part must be incremented by one
+		 * and the fractional part must be set to zero.
+		 */
+		intpart++;
+		fracpart = 0;
+		if (estyle && intpart == 10) {
+			/*
+			 * The value was rounded up to ten, but we only want one
+			 * integer digit if using "e-style".  So, the integer
+			 * part must be set to one and the exponent must be
+			 * incremented by one.
+			 */
+			intpart = 1;
+			exponent++;
+		}
+	}
+
+	/*
+	 * Now that we know the real exponent, we can check whether or not to
+	 * use "e-style" for "%g" (and "%G") conversions.  If we don't need
+	 * "e-style", the precision must be adjusted and the integer and
+	 * fractional parts must be recalculated from the original value.
+	 *
+	 * C99 says: "Let P equal the precision if nonzero, 6 if the precision
+	 * is omitted, or 1 if the precision is zero.  Then, if a conversion
+	 * with style `E' would have an exponent of X:
+	 *
+	 * - if P > X >= -4, the conversion is with style `f' (or `F') and
+	 *   precision P - (X + 1).
+	 *
+	 * - otherwise, the conversion is with style `e' (or `E') and precision
+	 *   P - 1." (7.19.6.1, 8)
+	 *
+	 * Note that we had decremented the precision by one.
+	 */
+	if (flags & PRINT_F_TYPE_G && estyle &&
+	    precision + 1 > exponent && exponent >= -4) {
+		precision -= exponent;
+		estyle = 0;
+		goto again;
+	}
+
+	if (estyle) {
+		if (exponent < 0) {
+			exponent = -exponent;
+			esign = '-';
+		} else
+			esign = '+';
+
+		/*
+		 * Convert the exponent.  The sizeof(econvert) is 4.  So, the
+		 * econvert buffer can hold e.g. "e+99" and "e-99".  We don't
+		 * support an exponent which contains more than two digits.
+		 * Therefore, the following stores are safe.
+		 */
+		epos = convert(exponent, econvert, 2, 10, 0);
+		/*
+		 * C99 says: "The exponent always contains at least two digits,
+		 * and only as many more digits as necessary to represent the
+		 * exponent." (7.19.6.1, 8)
+		 */
+		if (epos == 1)
+			econvert[epos++] = '0';
+		econvert[epos++] = esign;
+		econvert[epos++] = (flags & PRINT_F_UP) ? 'E' : 'e';
+	}
+
+	/* Convert the integer part and the fractional part. */
+	ipos = convert(intpart, iconvert, sizeof(iconvert), 10, 0);
+	if (fracpart != 0)	/* convert() would return 1 if fracpart == 0. */
+		fpos = convert(fracpart, fconvert, sizeof(fconvert), 10, 0);
+
+	leadfraczeros = precision - fpos;
+
+	if (omitzeros) {
+		if (fpos > 0)	/* Omit trailing fractional part zeros. */
+			while (omitcount < fpos && fconvert[omitcount] == '0')
+				omitcount++;
+		else {	/* The fractional part is zero, omit it completely. */
+			omitcount = precision;
+			leadfraczeros = 0;
+		}
+		precision -= omitcount;
+	}
+
+	/*
+	 * Print a decimal point if either the fractional part is non-zero
+	 * and/or the "#" flag was specified.
+	 */
+	if (precision > 0 || flags & PRINT_F_NUM)
+		emitpoint = 1;
+	if (separators)	/* Get the number of group separators we'll print. */
+		separators = getnumsep(ipos);
+
+	padlen = width                  /* Minimum field width. */
+	    - ipos                      /* Number of integer digits. */
+	    - epos                      /* Number of exponent characters. */
+	    - precision                 /* Number of fractional digits. */
+	    - separators                /* Number of group separators. */
+	    - (emitpoint ? 1 : 0)       /* Will we print a decimal point? */
+	    - ((sign != 0) ? 1 : 0);    /* Will we print a sign character? */
+
+	if (padlen < 0)
+		padlen = 0;
+
+	/*
+	 * C99 says: "If the `0' and `-' flags both appear, the `0' flag is
+	 * ignored." (7.19.6.1, 6)
+	 */
+	if (flags & PRINT_F_MINUS)	/* Left justifty. */
+		padlen = -padlen;
+	else if (flags & PRINT_F_ZERO && padlen > 0) {
+		if (sign != 0) {	/* Sign. */
+			OUTCHAR(str, *len, size, sign);
+			sign = 0;
+		}
+		while (padlen > 0) {	/* Leading zeros. */
+			OUTCHAR(str, *len, size, '0');
+			padlen--;
+		}
+	}
+	while (padlen > 0) {	/* Leading spaces. */
+		OUTCHAR(str, *len, size, ' ');
+		padlen--;
+	}
+	if (sign != 0)	/* Sign. */
+		OUTCHAR(str, *len, size, sign);
+	while (ipos > 0) {	/* Integer part. */
+		ipos--;
+		OUTCHAR(str, *len, size, iconvert[ipos]);
+		if (separators > 0 && ipos > 0 && ipos % 3 == 0)
+			printsep(str, len, size);
+	}
+	if (emitpoint) {	/* Decimal point. */
+#if HAVE_LOCALECONV && HAVE_LCONV_DECIMAL_POINT
+		if (lc->decimal_point != NULL && *lc->decimal_point != '\0')
+			OUTCHAR(str, *len, size, *lc->decimal_point);
+		else	/* We'll always print some decimal point character. */
+#endif	/* HAVE_LOCALECONV && HAVE_LCONV_DECIMAL_POINT */
+			OUTCHAR(str, *len, size, '.');
+	}
+	while (leadfraczeros > 0) {	/* Leading fractional part zeros. */
+		OUTCHAR(str, *len, size, '0');
+		leadfraczeros--;
+	}
+	while (fpos > omitcount) {	/* The remaining fractional part. */
+		fpos--;
+		OUTCHAR(str, *len, size, fconvert[fpos]);
+	}
+	while (epos > 0) {	/* Exponent. */
+		epos--;
+		OUTCHAR(str, *len, size, econvert[epos]);
+	}
+	while (padlen < 0) {	/* Trailing spaces. */
+		OUTCHAR(str, *len, size, ' ');
+		padlen++;
+	}
+}
+
+static void
+printsep(char *str, size_t *len, size_t size)
+{
+#if HAVE_LOCALECONV && HAVE_LCONV_THOUSANDS_SEP
+	struct lconv *lc = localeconv();
+	int i;
+
+	if (lc->thousands_sep != NULL)
+		for (i = 0; lc->thousands_sep[i] != '\0'; i++)
+			OUTCHAR(str, *len, size, lc->thousands_sep[i]);
+	else
+#endif	/* HAVE_LOCALECONV && HAVE_LCONV_THOUSANDS_SEP */
+		OUTCHAR(str, *len, size, ',');
+}
+
+static int
+getnumsep(int digits)
+{
+	int separators = (digits - ((digits % 3 == 0) ? 1 : 0)) / 3;
+#if HAVE_LOCALECONV && HAVE_LCONV_THOUSANDS_SEP
+	int strln;
+	struct lconv *lc = localeconv();
+
+	/* We support an arbitrary separator length (including zero). */
+	if (lc->thousands_sep != NULL) {
+		for (strln = 0; lc->thousands_sep[strln] != '\0'; strln++)
+			continue;
+		separators *= strln;
+	}
+#endif	/* HAVE_LOCALECONV && HAVE_LCONV_THOUSANDS_SEP */
+	return separators;
+}
+
+static int
+getexponent(LDOUBLE value)
+{
+	LDOUBLE tmp = (value >= 0.0) ? value : -value;
+	int exponent = 0;
+
+	/*
+	 * We check for 99 > exponent > -99 in order to work around possible
+	 * endless loops which could happen (at least) in the second loop (at
+	 * least) if we're called with an infinite value.  However, we checked
+	 * for infinity before calling this function using our ISINF() macro, so
+	 * this might be somewhat paranoid.
+	 */
+	while (tmp < 1.0 && tmp > 0.0 && --exponent > -99)
+		tmp *= 10;
+	while (tmp >= 10.0 && ++exponent < 99)
+		tmp /= 10;
+
+	return exponent;
+}
+
+static int
+convert(UINTMAX_T value, char *buf, size_t size, int base, int caps)
+{
+	const char *digits = caps ? "0123456789ABCDEF" : "0123456789abcdef";
+	size_t pos = 0;
+
+	/* We return an unterminated buffer with the digits in reverse order. */
+	do {
+		buf[pos++] = digits[value % base];
+		value /= base;
+	} while (value != 0 && pos < size);
+
+	return (int)pos;
+}
+
+static UINTMAX_T
+cast(LDOUBLE value)
+{
+	UINTMAX_T result;
+
+	/*
+	 * We check for ">=" and not for ">" because if UINTMAX_MAX cannot be
+	 * represented exactly as an LDOUBLE value (but is less than LDBL_MAX),
+	 * it may be increased to the nearest higher representable value for the
+	 * comparison (cf. C99: 6.3.1.4, 2).  It might then equal the LDOUBLE
+	 * value although converting the latter to UINTMAX_T would overflow.
+	 */
+	if (value >= UINTMAX_MAX)
+		return UINTMAX_MAX;
+
+#if defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT)
+	result = (unsigned long)value;
+#else
+	result = (UINTMAX_T)value;
+#endif
+	/*
+	 * At least on NetBSD/sparc64 3.0.2 and 4.99.30, casting long double to
+	 * an integer type converts e.g. 1.9 to 2 instead of 1 (which violates
+	 * the standard).  Sigh.
+	 */
+	return (result <= value) ? result : result - 1;
+}
+
+static UINTMAX_T
+myround(LDOUBLE value)
+{
+	UINTMAX_T intpart = cast(value);
+
+	return ((value -= intpart) < 0.5) ? intpart : intpart + 1;
+}
+
+static LDOUBLE
+mypow10(int exponent)
+{
+	LDOUBLE result = 1;
+
+	while (exponent > 0) {
+		result *= 10;
+		exponent--;
+	}
+	while (exponent < 0) {
+		result /= 10;
+		exponent++;
+	}
+	return result;
+}
+#endif	/* !HAVE_VSNPRINTF */
+
+#if !HAVE_VASPRINTF
+#if NEED_MYMEMCPY
+void *
+mymemcpy(void *dst, void *src, size_t len)
+{
+	const char *from = src;
+	char *to = dst;
+
+	/* No need for optimization, we use this only to replace va_copy(3). */
+	while (len-- > 0)
+		*to++ = *from++;
+	return dst;
+}
+#endif	/* NEED_MYMEMCPY */
+
+int
+util_vasprintf(char **ret, const char *format, va_list ap)
+{
+	size_t size;
+	int len;
+	va_list aq;
+
+	VA_COPY(aq, ap);
+	len = vsnprintf(NULL, 0, format, aq);
+	VA_END_COPY(aq);
+	if (len < 0 || (*ret = malloc(size = len + 1)) == NULL)
+		return -1;
+	return vsnprintf(*ret, size, format, ap);
+}
+#endif	/* !HAVE_VASPRINTF */
+
+#if !HAVE_SNPRINTF
+#if HAVE_STDARG_H
+int
+util_snprintf(char *str, size_t size, const char *format, ...)
+#else
+int
+util_snprintf(va_alist) va_dcl
+#endif	/* HAVE_STDARG_H */
+{
+#if !HAVE_STDARG_H
+	char *str;
+	size_t size;
+	char *format;
+#endif	/* HAVE_STDARG_H */
+	va_list ap;
+	int len;
+
+	VA_START(ap, format);
+	VA_SHIFT(ap, str, char *);
+	VA_SHIFT(ap, size, size_t);
+	VA_SHIFT(ap, format, const char *);
+	len = vsnprintf(str, size, format, ap);
+	va_end(ap);
+	return len;
+}
+#endif	/* !HAVE_SNPRINTF */
+
+#if !HAVE_ASPRINTF
+#if HAVE_STDARG_H
+int
+util_asprintf(char **ret, const char *format, ...)
+#else
+int
+util_asprintf(va_alist) va_dcl
+#endif	/* HAVE_STDARG_H */
+{
+#if !HAVE_STDARG_H
+	char **ret;
+	char *format;
+#endif	/* HAVE_STDARG_H */
+	va_list ap;
+	int len;
+
+	VA_START(ap, format);
+	VA_SHIFT(ap, ret, char **);
+	VA_SHIFT(ap, format, const char *);
+	len = vasprintf(ret, format, ap);
+	va_end(ap);
+	return len;
+}
+#endif	/* !HAVE_ASPRINTF */
+#else	/* Dummy declaration to avoid empty translation unit warnings. */
+int main(void);
+#endif	/* !HAVE_SNPRINTF || !HAVE_VSNPRINTF || !HAVE_ASPRINTF || [...] */
+
+
+/* vim: set joinspaces textwidth=80: */
diff --git a/src/gallium/auxiliary/util/u_sse.h b/src/gallium/auxiliary/util/u_sse.h
new file mode 100644
index 00000000000..e2a8491e62c
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_sse.h
@@ -0,0 +1,77 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * SSE intrinsics portability header.
+ * 
+ * Although the SSE intrinsics are support by all modern x86 and x86-64 
+ * compilers, there are some intrisincs missing in some implementations 
+ * (especially older MSVC versions). This header abstracts that away.
+ */
+
+#ifndef U_SSE_H_
+#define U_SSE_H_
+
+#include "pipe/p_config.h"
+
+#if defined(PIPE_ARCH_SSE)
+
+#include <xmmintrin.h>
+#include <emmintrin.h>
+
+
+/* MSVC before VC8 does not support the _mm_castxxx_yyy */
+#if defined(_MSC_VER) && _MSC_VER < 1500
+
+union __declspec(align(16)) m128_types {
+   __m128 m128;
+   __m128i m128i;
+   __m128d m128d;
+};
+
+static __inline __m128
+_mm_castsi128_ps(__m128i a)
+{
+   union m128_types u;
+   u.m128i = a;
+   return u.m128;
+}
+
+static __inline __m128i
+_mm_castps_si128(__m128 a)
+{
+   union m128_types u;
+   u.m128 = a;
+   return u.m128i;
+}
+
+#endif /* defined(_MSC_VER) && _MSC_VER < 1500 */
+
+#endif /* PIPE_ARCH_X86 || PIPE_ARCH_X86_64 */
+
+#endif /* U_SSE_H_ */
diff --git a/src/gallium/auxiliary/util/u_stream.h b/src/gallium/auxiliary/util/u_stream.h
new file mode 100644
index 00000000000..a9d0f0121a6
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_stream.h
@@ -0,0 +1,61 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Cross-platform sequential access stream abstraction.
+ */
+
+#ifndef U_STREAM_H
+#define U_STREAM_H
+
+
+#include "pipe/p_compiler.h"
+
+
+struct util_stream;
+
+
+/**
+ * Create a stream
+ * @param filename relative or absolute path (necessary for windows)  
+ * @param optional maximum file size (0 for a growable size).
+ */
+struct util_stream *
+util_stream_create(const char *filename, size_t max_size);
+
+boolean
+util_stream_write(struct util_stream *stream, const void *data, size_t size);
+
+void
+util_stream_flush(struct util_stream *stream);
+
+void
+util_stream_close(struct util_stream *stream);
+
+
+#endif /* U_STREAM_H */
diff --git a/src/gallium/auxiliary/util/u_stream_stdc.c b/src/gallium/auxiliary/util/u_stream_stdc.c
new file mode 100644
index 00000000000..ca80bef0f3d
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_stream_stdc.c
@@ -0,0 +1,106 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Stream implementation based on the Standard C Library.
+ */
+
+#include "pipe/p_config.h"
+
+#if defined(PIPE_OS_LINUX) || defined(PIPE_SUBSYSTEM_WINDOWS_USER)
+
+#include <stdio.h>
+
+#include "util/u_memory.h"
+
+#include "u_stream.h"
+
+
+struct util_stream 
+{
+   FILE *file;
+};
+
+
+struct util_stream *
+util_stream_create(const char *filename, size_t max_size)
+{
+   struct util_stream *stream;
+   
+   (void)max_size;
+   
+   stream = CALLOC_STRUCT(util_stream);
+   if(!stream)
+      goto error1;
+   
+   stream->file = fopen(filename, "w");
+   if(!stream->file)
+      goto error2;
+   
+   return stream;
+   
+error2:
+   FREE(stream);
+error1:
+   return NULL;
+}
+
+
+boolean
+util_stream_write(struct util_stream *stream, const void *data, size_t size)
+{
+   if(!stream)
+      return FALSE;
+   
+   return fwrite(data, size, 1, stream->file) == size ? TRUE : FALSE;
+}
+
+
+void
+util_stream_flush(struct util_stream *stream) 
+{
+   if(!stream)
+      return;
+   
+   fflush(stream->file);
+}
+
+
+void
+util_stream_close(struct util_stream *stream) 
+{
+   if(!stream)
+      return;
+   
+   fclose(stream->file);
+
+   FREE(stream);
+}
+
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_stream_wd.c b/src/gallium/auxiliary/util/u_stream_wd.c
new file mode 100644
index 00000000000..864489e7755
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_stream_wd.c
@@ -0,0 +1,224 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Stream implementation for the Windows Display driver.
+ */
+
+#include "pipe/p_config.h"
+
+#if defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY)
+
+#include <windows.h>
+#include <winddi.h>
+
+#include "util/u_memory.h"
+#include "util/u_string.h"
+
+#include "u_stream.h"
+
+
+#define MAP_FILE_SIZE (4*1024*1024)
+
+
+struct util_stream 
+{
+   char filename[MAX_PATH + 1];
+   WCHAR wFileName[MAX_PATH + 1];
+   boolean growable;
+   size_t map_size;
+   ULONG_PTR iFile;
+   char *pMap;
+   size_t written;
+   unsigned suffix;
+};
+
+
+static INLINE boolean
+util_stream_map(struct util_stream *stream)
+{
+   ULONG BytesInUnicodeString;
+   static char filename[MAX_PATH + 1];
+   unsigned filename_len;
+
+   if(stream->growable)
+      filename_len = util_snprintf(filename,
+                                   sizeof(filename),
+                                   "%s.%04x",
+                                   stream->filename,
+                                   stream->suffix++);
+   else
+      filename_len = util_snprintf(filename,
+                                   sizeof(filename),
+                                   "%s",
+                                   stream->filename);
+
+   EngMultiByteToUnicodeN(
+         stream->wFileName,
+         sizeof(stream->wFileName),
+         &BytesInUnicodeString,
+         filename,
+         filename_len);
+   
+   stream->pMap = EngMapFile(stream->wFileName, stream->map_size, &stream->iFile);
+   if(!stream->pMap)
+      return FALSE;
+   
+   memset(stream->pMap, 0, stream->map_size);
+   stream->written = 0;
+   
+   return TRUE;
+}
+
+
+static INLINE void
+util_stream_unmap(struct util_stream *stream)
+{
+   EngUnmapFile(stream->iFile);
+   if(stream->written < stream->map_size) {
+      /* Truncate file size */
+      stream->pMap = EngMapFile(stream->wFileName, stream->written, &stream->iFile);
+      if(stream->pMap)
+         EngUnmapFile(stream->iFile);
+   }
+   
+   stream->pMap = NULL;
+}
+
+
+static INLINE void
+util_stream_full_qualified_filename(char *dst, size_t size, const char *src)
+{
+   boolean need_drive, need_root;
+   
+   if((('A' <= src[0] && src[0] <= 'Z') || ('a' <= src[0] && src[0] <= 'z')) && src[1] == ':') {
+      need_drive = FALSE;
+      need_root = src[2] == '\\' ? FALSE : TRUE;
+   }
+   else {
+      need_drive = TRUE;
+      need_root = src[0] == '\\' ? FALSE : TRUE;
+   }
+   
+   util_snprintf(dst, size, 
+                 "\\??\\%s%s%s",
+                 need_drive ? "C:" : "",
+                 need_root ? "\\" : "",
+                 src);
+}
+
+
+struct util_stream *
+util_stream_create(const char *filename, size_t max_size)
+{
+   struct util_stream *stream;
+   
+   stream = CALLOC_STRUCT(util_stream);
+   if(!stream)
+      goto error1;
+   
+   util_stream_full_qualified_filename(stream->filename,
+                                       sizeof(stream->filename),
+                                       filename);
+   
+   if(max_size) {
+      stream->growable = FALSE;
+      stream->map_size = max_size;
+   }
+   else {
+      stream->growable = TRUE;
+      stream->map_size = MAP_FILE_SIZE;
+   }
+   
+   if(!util_stream_map(stream))
+      goto error2;
+   
+   return stream;
+   
+error2:
+   FREE(stream);
+error1:
+   return NULL;
+}
+
+
+static INLINE void
+util_stream_copy(struct util_stream *stream, const char *data, size_t size)
+{
+   assert(stream->written + size <= stream->map_size);
+   memcpy(stream->pMap + stream->written, data, size);
+   stream->written += size;
+}
+
+
+boolean
+util_stream_write(struct util_stream *stream, const void *data, size_t size)
+{
+   if(!stream)
+      return FALSE;
+   
+   if(!stream->pMap)
+      return FALSE;
+   
+   while(stream->written + size > stream->map_size) {
+      size_t step = stream->map_size - stream->written;
+      util_stream_copy(stream, data, step);
+      data = (const char *)data + step;
+      size -= step;
+      
+      util_stream_unmap(stream);
+      if(!stream->growable || !util_stream_map(stream))
+         return FALSE;
+   }
+
+   util_stream_copy(stream, data, size);
+   
+   return TRUE;
+}
+
+
+void
+util_stream_flush(struct util_stream *stream) 
+{
+   (void)stream;
+}
+
+
+void
+util_stream_close(struct util_stream *stream) 
+{
+   if(!stream)
+      return;
+   
+   util_stream_unmap(stream);
+
+   FREE(stream);
+}
+
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_string.h b/src/gallium/auxiliary/util/u_string.h
new file mode 100644
index 00000000000..08c89bbf770
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_string.h
@@ -0,0 +1,220 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * Platform independent functions for string manipulation.
+ *
+ * @author Jose Fonseca <[email protected]>
+ */
+
+#ifndef U_STRING_H_
+#define U_STRING_H_
+
+#if !defined(WIN32) && !defined(XF86_LIBC_H)
+#include <stdio.h>
+#endif
+#include <stddef.h>
+#include <stdarg.h>
+
+#include "pipe/p_compiler.h"
+
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+
+#ifdef WIN32
+
+int util_vsnprintf(char *, size_t, const char *, va_list);
+int util_snprintf(char *str, size_t size, const char *format, ...);
+
+static INLINE void
+util_vsprintf(char *str, const char *format, va_list ap)
+{
+   util_vsnprintf(str, (size_t)-1, format, ap);
+}
+
+static INLINE void
+util_sprintf(char *str, const char *format, ...)
+{
+   va_list ap;
+   va_start(ap, format);
+   util_vsnprintf(str, (size_t)-1, format, ap);
+   va_end(ap);
+}
+
+static INLINE char *
+util_strchr(const char *s, char c)
+{
+   while(*s) {
+      if(*s == c)
+	 return (char *)s;
+      ++s;
+   }
+   return NULL;
+}
+
+static INLINE char*
+util_strncat(char *dst, const char *src, size_t n)
+{
+   char *p = dst + strlen(dst);
+   const char *q = src;
+   size_t i;
+
+   for (i = 0; i < n && *q != '\0'; ++i)
+       *p++ = *q++;
+   *p = '\0';
+
+   return dst;
+}
+
+static INLINE int
+util_strcmp(const char *s1, const char *s2)
+{
+   unsigned char u1, u2;
+
+   while (1) {
+      u1 = (unsigned char) *s1++;
+      u2 = (unsigned char) *s2++;
+      if (u1 != u2)
+	 return u1 - u2;
+      if (u1 == '\0')
+	 return 0;
+   }
+   return 0;
+}
+
+static INLINE int
+util_strncmp(const char *s1, const char *s2, size_t n)
+{
+   unsigned char u1, u2;
+
+   while (n-- > 0) {
+      u1 = (unsigned char) *s1++;
+      u2 = (unsigned char) *s2++;
+      if (u1 != u2)
+	 return u1 - u2;
+      if (u1 == '\0')
+	 return 0;
+   }
+   return 0;
+}
+
+static INLINE char *
+util_strstr(const char *haystack, const char *needle)
+{
+   const char *p = haystack;
+   int len = strlen(needle);
+
+   for (; (p = util_strchr(p, *needle)) != 0; p++) {
+      if (util_strncmp(p, needle, len) == 0) {
+	 return (char *)p;
+      }
+   }
+   return NULL;
+}
+
+static INLINE void *
+util_memmove(void *dest, const void *src, size_t n)
+{
+   char *p = (char *)dest;
+   const char *q = (const char *)src;
+   if (dest < src) {
+      while (n--)
+	 *p++ = *q++;
+   }
+   else
+   {
+      p += n;
+      q += n;
+      while (n--)
+	 *--p = *--q;
+   }
+   return dest;
+}
+
+
+#else
+
+#define util_vsnprintf vsnprintf
+#define util_snprintf snprintf
+#define util_vsprintf vsprintf
+#define util_sprintf sprintf
+#define util_strchr strchr
+#define util_strcmp strcmp
+#define util_strncmp strncmp
+#define util_strncat strncat
+#define util_strstr strstr
+#define util_memmove memmove
+
+#endif
+
+
+/**
+ * Printable string buffer
+ */
+struct util_strbuf
+{
+   char *str;
+   char *ptr;
+   size_t left;
+};
+
+
+static INLINE void
+util_strbuf_init(struct util_strbuf *sbuf, char *str, size_t size) 
+{
+   sbuf->str = str;
+   sbuf->str[0] = 0;
+   sbuf->ptr = sbuf->str;
+   sbuf->left = size;
+}
+
+
+static INLINE void
+util_strbuf_printf(struct util_strbuf *sbuf, const char *format, ...)
+{
+   if(sbuf->left > 1) {
+      size_t written;
+      va_list ap;
+      va_start(ap, format);
+      written = util_vsnprintf(sbuf->ptr, sbuf->left, format, ap);
+      va_end(ap);
+      sbuf->ptr += written;
+      sbuf->left -= written;
+   }
+}
+
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* U_STRING_H_ */
diff --git a/src/gallium/auxiliary/util/u_tile.c b/src/gallium/auxiliary/util/u_tile.c
new file mode 100644
index 00000000000..32f6b072a00
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_tile.c
@@ -0,0 +1,1198 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/**
+ * RGBA/float tile get/put functions.
+ * Usable both by drivers and state trackers.
+ * Surfaces should already be in a mapped state.
+ */
+
+
+#include "pipe/p_defines.h"
+#include "pipe/p_inlines.h"
+
+#include "util/u_math.h"
+#include "util/u_memory.h"
+#include "util/u_rect.h"
+#include "util/u_tile.h"
+
+
+/**
+ * Move raw block of pixels from surface to user memory.
+ * This should be usable by any hw driver that has mappable surfaces.
+ */
+void
+pipe_get_tile_raw(struct pipe_surface *ps,
+                  uint x, uint y, uint w, uint h,
+                  void *dst, int dst_stride)
+{
+   const void *src;
+
+   if (dst_stride == 0)
+      dst_stride = pf_get_nblocksx(&ps->block, w) * ps->block.size;
+
+   if (pipe_clip_tile(x, y, &w, &h, ps))
+      return;
+
+   src = pipe_surface_map(ps, PIPE_BUFFER_USAGE_CPU_READ);
+   assert(src);
+   if(!src)
+      return;
+
+   pipe_copy_rect(dst, &ps->block, dst_stride, 0, 0, w, h, src, ps->stride, x, y);
+
+   pipe_surface_unmap(ps);
+}
+
+
+/**
+ * Move raw block of pixels from user memory to surface.
+ * This should be usable by any hw driver that has mappable surfaces.
+ */
+void
+pipe_put_tile_raw(struct pipe_surface *ps,
+                  uint x, uint y, uint w, uint h,
+                  const void *src, int src_stride)
+{
+   void *dst;
+
+   if (src_stride == 0)
+      src_stride = pf_get_nblocksx(&ps->block, w) * ps->block.size;
+
+   if (pipe_clip_tile(x, y, &w, &h, ps))
+      return;
+
+   dst = pipe_surface_map(ps, PIPE_BUFFER_USAGE_CPU_WRITE);
+   assert(dst);
+   if(!dst)
+      return;
+
+   pipe_copy_rect(dst, &ps->block, ps->stride, x, y, w, h, src, src_stride, 0, 0);
+
+   pipe_surface_unmap(ps);
+}
+
+
+
+
+/** Convert short in [-32768,32767] to GLfloat in [-1.0,1.0] */
+#define SHORT_TO_FLOAT(S)   ((2.0F * (S) + 1.0F) * (1.0F/65535.0F))
+
+#define UNCLAMPED_FLOAT_TO_SHORT(us, f)  \
+   us = ( (short) ( CLAMP((f), -1.0, 1.0) * 32767.0F) )
+
+
+
+/*** PIPE_FORMAT_A8R8G8B8_UNORM ***/
+
+static void
+a8r8g8b8_get_tile_rgba(const unsigned *src,
+                       unsigned w, unsigned h,
+                       float *p,
+                       unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         const unsigned pixel = *src++;
+         pRow[0] = ubyte_to_float((pixel >> 16) & 0xff);
+         pRow[1] = ubyte_to_float((pixel >>  8) & 0xff);
+         pRow[2] = ubyte_to_float((pixel >>  0) & 0xff);
+         pRow[3] = ubyte_to_float((pixel >> 24) & 0xff);
+      }
+      p += dst_stride;
+   }
+}
+
+
+static void
+a8r8g8b8_put_tile_rgba(unsigned *dst,
+                       unsigned w, unsigned h,
+                       const float *p,
+                       unsigned src_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      const float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         unsigned r, g, b, a;
+         r = float_to_ubyte(pRow[0]);
+         g = float_to_ubyte(pRow[1]);
+         b = float_to_ubyte(pRow[2]);
+         a = float_to_ubyte(pRow[3]);
+         *dst++ = (a << 24) | (r << 16) | (g << 8) | b;
+      }
+      p += src_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_A8R8G8B8_UNORM ***/
+
+static void
+x8r8g8b8_get_tile_rgba(const unsigned *src,
+                       unsigned w, unsigned h,
+                       float *p,
+                       unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         const unsigned pixel = *src++;
+         pRow[0] = ubyte_to_float((pixel >> 16) & 0xff);
+         pRow[1] = ubyte_to_float((pixel >>  8) & 0xff);
+         pRow[2] = ubyte_to_float((pixel >>  0) & 0xff);
+         pRow[3] = ubyte_to_float(0xff);
+      }
+      p += dst_stride;
+   }
+}
+
+
+static void
+x8r8g8b8_put_tile_rgba(unsigned *dst,
+                       unsigned w, unsigned h,
+                       const float *p,
+                       unsigned src_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      const float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         unsigned r, g, b;
+         r = float_to_ubyte(pRow[0]);
+         g = float_to_ubyte(pRow[1]);
+         b = float_to_ubyte(pRow[2]);
+         *dst++ = (0xff << 24) | (r << 16) | (g << 8) | b;
+      }
+      p += src_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_B8G8R8A8_UNORM ***/
+
+static void
+b8g8r8a8_get_tile_rgba(const unsigned *src,
+                       unsigned w, unsigned h,
+                       float *p,
+                       unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         const unsigned pixel = *src++;
+         pRow[0] = ubyte_to_float((pixel >>  8) & 0xff);
+         pRow[1] = ubyte_to_float((pixel >> 16) & 0xff);
+         pRow[2] = ubyte_to_float((pixel >> 24) & 0xff);
+         pRow[3] = ubyte_to_float((pixel >>  0) & 0xff);
+      }
+      p += dst_stride;
+   }
+}
+
+
+static void
+b8g8r8a8_put_tile_rgba(unsigned *dst,
+                       unsigned w, unsigned h,
+                       const float *p,
+                       unsigned src_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      const float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         unsigned r, g, b, a;
+         r = float_to_ubyte(pRow[0]);
+         g = float_to_ubyte(pRow[1]);
+         b = float_to_ubyte(pRow[2]);
+         a = float_to_ubyte(pRow[3]);
+         *dst++ = (b << 24) | (g << 16) | (r << 8) | a;
+      }
+      p += src_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_A1R5G5B5_UNORM ***/
+
+static void
+a1r5g5b5_get_tile_rgba(const ushort *src,
+                       unsigned w, unsigned h,
+                       float *p,
+                       unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         const ushort pixel = *src++;
+         pRow[0] = ((pixel >> 10) & 0x1f) * (1.0f / 31.0f);
+         pRow[1] = ((pixel >>  5) & 0x1f) * (1.0f / 31.0f);
+         pRow[2] = ((pixel      ) & 0x1f) * (1.0f / 31.0f);
+         pRow[3] = ((pixel >> 15)       ) * 1.0f;
+      }
+      p += dst_stride;
+   }
+}
+
+
+static void
+a1r5g5b5_put_tile_rgba(ushort *dst,
+                       unsigned w, unsigned h,
+                       const float *p,
+                       unsigned src_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      const float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         unsigned r, g, b, a;
+         r = float_to_ubyte(pRow[0]);
+         g = float_to_ubyte(pRow[1]);
+         b = float_to_ubyte(pRow[2]);
+         a = float_to_ubyte(pRow[3]);
+         r = r >> 3;  /* 5 bits */
+         g = g >> 3;  /* 5 bits */
+         b = b >> 3;  /* 5 bits */
+         a = a >> 7;  /* 1 bit */
+         *dst++ = (a << 15) | (r << 10) | (g << 5) | b;
+      }
+      p += src_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_A4R4G4B4_UNORM ***/
+
+static void
+a4r4g4b4_get_tile_rgba(const ushort *src,
+                       unsigned w, unsigned h,
+                       float *p,
+                       unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         const ushort pixel = *src++;
+         pRow[0] = ((pixel >>  8) & 0xf) * (1.0f / 15.0f);
+         pRow[1] = ((pixel >>  4) & 0xf) * (1.0f / 15.0f);
+         pRow[2] = ((pixel      ) & 0xf) * (1.0f / 15.0f);
+         pRow[3] = ((pixel >> 12)      ) * (1.0f / 15.0f);
+      }
+      p += dst_stride;
+   }
+}
+
+
+static void
+a4r4g4b4_put_tile_rgba(ushort *dst,
+                       unsigned w, unsigned h,
+                       const float *p,
+                       unsigned src_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      const float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         unsigned r, g, b, a;
+         r = float_to_ubyte(pRow[0]);
+         g = float_to_ubyte(pRow[1]);
+         b = float_to_ubyte(pRow[2]);
+         a = float_to_ubyte(pRow[3]);
+         r >>= 4;
+         g >>= 4;
+         b >>= 4;
+         a >>= 4;
+         *dst++ = (a << 12) | (r << 16) | (g << 4) | b;
+      }
+      p += src_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_R5G6B5_UNORM ***/
+
+static void
+r5g6b5_get_tile_rgba(const ushort *src,
+                     unsigned w, unsigned h,
+                     float *p,
+                     unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         const ushort pixel = *src++;
+         pRow[0] = ((pixel >> 11) & 0x1f) * (1.0f / 31.0f);
+         pRow[1] = ((pixel >>  5) & 0x3f) * (1.0f / 63.0f);
+         pRow[2] = ((pixel      ) & 0x1f) * (1.0f / 31.0f);
+         pRow[3] = 1.0f;
+      }
+      p += dst_stride;
+   }
+}
+
+
+static void
+r5g6b5_put_tile_rgba(ushort *dst,
+                     unsigned w, unsigned h,
+                     const float *p,
+                     unsigned src_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      const float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         uint r = (uint) (CLAMP(pRow[0], 0.0, 1.0) * 31.0);
+         uint g = (uint) (CLAMP(pRow[1], 0.0, 1.0) * 63.0);
+         uint b = (uint) (CLAMP(pRow[2], 0.0, 1.0) * 31.0);
+         *dst++ = (r << 11) | (g << 5) | (b);
+      }
+      p += src_stride;
+   }
+}
+
+
+
+/*** PIPE_FORMAT_Z16_UNORM ***/
+
+/**
+ * Return each Z value as four floats in [0,1].
+ */
+static void
+z16_get_tile_rgba(const ushort *src,
+                  unsigned w, unsigned h,
+                  float *p,
+                  unsigned dst_stride)
+{
+   const float scale = 1.0f / 65535.0f;
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] =
+         pRow[3] = *src++ * scale;
+      }
+      p += dst_stride;
+   }
+}
+
+
+
+
+/*** PIPE_FORMAT_L8_UNORM ***/
+
+static void
+l8_get_tile_rgba(const ubyte *src,
+                 unsigned w, unsigned h,
+                 float *p,
+                 unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, src++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] = ubyte_to_float(*src);
+         pRow[3] = 1.0;
+      }
+      p += dst_stride;
+   }
+}
+
+
+static void
+l8_put_tile_rgba(ubyte *dst,
+                 unsigned w, unsigned h,
+                 const float *p,
+                 unsigned src_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      const float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         unsigned r;
+         r = float_to_ubyte(pRow[0]);
+         *dst++ = (ubyte) r;
+      }
+      p += src_stride;
+   }
+}
+
+
+
+/*** PIPE_FORMAT_A8_UNORM ***/
+
+static void
+a8_get_tile_rgba(const ubyte *src,
+                 unsigned w, unsigned h,
+                 float *p,
+                 unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, src++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] = 0.0;
+         pRow[3] = ubyte_to_float(*src);
+      }
+      p += dst_stride;
+   }
+}
+
+
+static void
+a8_put_tile_rgba(ubyte *dst,
+                 unsigned w, unsigned h,
+                 const float *p,
+                 unsigned src_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      const float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         unsigned a;
+         a = float_to_ubyte(pRow[3]);
+         *dst++ = (ubyte) a;
+      }
+      p += src_stride;
+   }
+}
+
+
+
+/*** PIPE_FORMAT_R16_SNORM ***/
+
+static void
+r16_get_tile_rgba(const short *src,
+                  unsigned w, unsigned h,
+                  float *p,
+                  unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, src++, pRow += 4) {
+         pRow[0] = SHORT_TO_FLOAT(src[0]);
+         pRow[1] =
+         pRow[2] = 0.0;
+         pRow[3] = 1.0;
+      }
+      p += dst_stride;
+   }
+}
+
+
+static void
+r16_put_tile_rgba(short *dst,
+                  unsigned w, unsigned h,
+                  const float *p,
+                  unsigned src_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      const float *pRow = p;
+      for (j = 0; j < w; j++, dst++, pRow += 4) {
+         UNCLAMPED_FLOAT_TO_SHORT(dst[0], pRow[0]);
+      }
+      p += src_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_R16G16B16A16_SNORM ***/
+
+static void
+r16g16b16a16_get_tile_rgba(const short *src,
+                           unsigned w, unsigned h,
+                           float *p,
+                           unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, src += 4, pRow += 4) {
+         pRow[0] = SHORT_TO_FLOAT(src[0]);
+         pRow[1] = SHORT_TO_FLOAT(src[1]);
+         pRow[2] = SHORT_TO_FLOAT(src[2]);
+         pRow[3] = SHORT_TO_FLOAT(src[3]);
+      }
+      p += dst_stride;
+   }
+}
+
+
+static void
+r16g16b16a16_put_tile_rgba(short *dst,
+                           unsigned w, unsigned h,
+                           const float *p,
+                           unsigned src_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      const float *pRow = p;
+      for (j = 0; j < w; j++, dst += 4, pRow += 4) {
+         UNCLAMPED_FLOAT_TO_SHORT(dst[0], pRow[0]);
+         UNCLAMPED_FLOAT_TO_SHORT(dst[1], pRow[1]);
+         UNCLAMPED_FLOAT_TO_SHORT(dst[2], pRow[2]);
+         UNCLAMPED_FLOAT_TO_SHORT(dst[3], pRow[3]);
+      }
+      p += src_stride;
+   }
+}
+
+
+
+/*** PIPE_FORMAT_I8_UNORM ***/
+
+static void
+i8_get_tile_rgba(const ubyte *src,
+                 unsigned w, unsigned h,
+                 float *p,
+                 unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, src++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] =
+         pRow[3] = ubyte_to_float(*src);
+      }
+      p += dst_stride;
+   }
+}
+
+
+static void
+i8_put_tile_rgba(ubyte *dst,
+                 unsigned w, unsigned h,
+                 const float *p,
+                 unsigned src_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      const float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         unsigned r;
+         r = float_to_ubyte(pRow[0]);
+         *dst++ = (ubyte) r;
+      }
+      p += src_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_A8L8_UNORM ***/
+
+static void
+a8l8_get_tile_rgba(const ushort *src,
+                   unsigned w, unsigned h,
+                   float *p,
+                   unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         ushort p = *src++;
+         pRow[0] =
+         pRow[1] =
+         pRow[2] = ubyte_to_float(p & 0xff);
+         pRow[3] = ubyte_to_float(p >> 8);
+      }
+      p += dst_stride;
+   }
+}
+
+
+static void
+a8l8_put_tile_rgba(ushort *dst,
+                   unsigned w, unsigned h,
+                   const float *p,
+                   unsigned src_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      const float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         unsigned r, a;
+         r = float_to_ubyte(pRow[0]);
+         a = float_to_ubyte(pRow[3]);
+         *dst++ = (a << 8) | r;
+      }
+      p += src_stride;
+   }
+}
+
+
+
+
+/*** PIPE_FORMAT_Z32_UNORM ***/
+
+/**
+ * Return each Z value as four floats in [0,1].
+ */
+static void
+z32_get_tile_rgba(const unsigned *src,
+                  unsigned w, unsigned h,
+                  float *p,
+                  unsigned dst_stride)
+{
+   const double scale = 1.0 / (double) 0xffffffff;
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] =
+         pRow[3] = (float) (*src++ * scale);
+      }
+      p += dst_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_S8Z24_UNORM ***/
+
+/**
+ * Return Z component as four float in [0,1].  Stencil part ignored.
+ */
+static void
+s8z24_get_tile_rgba(const unsigned *src,
+                    unsigned w, unsigned h,
+                    float *p,
+                    unsigned dst_stride)
+{
+   const double scale = 1.0 / ((1 << 24) - 1);
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] =
+         pRow[3] = (float) (scale * (*src++ & 0xffffff));
+      }
+      p += dst_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_Z24S8_UNORM ***/
+
+/**
+ * Return Z component as four float in [0,1].  Stencil part ignored.
+ */
+static void
+z24s8_get_tile_rgba(const unsigned *src,
+                    unsigned w, unsigned h,
+                    float *p,
+                    unsigned dst_stride)
+{
+   const double scale = 1.0 / ((1 << 24) - 1);
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] =
+         pRow[3] = (float) (scale * (*src++ >> 8));
+      }
+      p += dst_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_Z32_FLOAT ***/
+
+/**
+ * Return each Z value as four floats in [0,1].
+ */
+static void
+z32f_get_tile_rgba(const float *src,
+                   unsigned w, unsigned h,
+                   float *p,
+                   unsigned dst_stride)
+{
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      for (j = 0; j < w; j++, pRow += 4) {
+         pRow[0] =
+         pRow[1] =
+         pRow[2] =
+         pRow[3] = *src++;
+      }
+      p += dst_stride;
+   }
+}
+
+
+/*** PIPE_FORMAT_YCBCR / PIPE_FORMAT_YCBCR_REV ***/
+
+/**
+ * Convert YCbCr (or YCrCb) to RGBA.
+ */
+static void
+ycbcr_get_tile_rgba(const ushort *src,
+                    unsigned w, unsigned h,
+                    float *p,
+                    unsigned dst_stride,
+                    boolean rev)
+{
+   const float scale = 1.0f / 255.0f;
+   unsigned i, j;
+
+   for (i = 0; i < h; i++) {
+      float *pRow = p;
+      /* do two texels at a time */
+      for (j = 0; j < (w & ~1); j += 2, src += 2) {
+         const ushort t0 = src[0];
+         const ushort t1 = src[1];
+         const ubyte y0 = (t0 >> 8) & 0xff;  /* luminance */
+         const ubyte y1 = (t1 >> 8) & 0xff;  /* luminance */
+         ubyte cb, cr;
+         float r, g, b;
+
+         if (rev) {
+            cb = t1 & 0xff;         /* chroma U */
+            cr = t0 & 0xff;         /* chroma V */
+         }
+         else {
+            cb = t0 & 0xff;         /* chroma U */
+            cr = t1 & 0xff;         /* chroma V */
+         }
+
+         /* even pixel: y0,cr,cb */
+         r = 1.164f * (y0-16) + 1.596f * (cr-128);
+         g = 1.164f * (y0-16) - 0.813f * (cr-128) - 0.391f * (cb-128);
+         b = 1.164f * (y0-16) + 2.018f * (cb-128);
+         pRow[0] = r * scale;
+         pRow[1] = g * scale;
+         pRow[2] = b * scale;
+         pRow[3] = 1.0f;
+         pRow += 4;
+
+         /* odd pixel: use y1,cr,cb */
+         r = 1.164f * (y1-16) + 1.596f * (cr-128);
+         g = 1.164f * (y1-16) - 0.813f * (cr-128) - 0.391f * (cb-128);
+         b = 1.164f * (y1-16) + 2.018f * (cb-128);
+         pRow[0] = r * scale;
+         pRow[1] = g * scale;
+         pRow[2] = b * scale;
+         pRow[3] = 1.0f;
+         pRow += 4;
+
+      }
+      /* do the last texel */
+      if (w & 1) {
+         const ushort t0 = src[0];
+         const ushort t1 = src[1];
+         const ubyte y0 = (t0 >> 8) & 0xff;  /* luminance */
+         ubyte cb, cr;
+         float r, g, b;
+
+         if (rev) {
+            cb = t1 & 0xff;         /* chroma U */
+            cr = t0 & 0xff;         /* chroma V */
+         }
+         else {
+            cb = t0 & 0xff;         /* chroma U */
+            cr = t1 & 0xff;         /* chroma V */
+         }
+
+         /* even pixel: y0,cr,cb */
+         r = 1.164f * (y0-16) + 1.596f * (cr-128);
+         g = 1.164f * (y0-16) - 0.813f * (cr-128) - 0.391f * (cb-128);
+         b = 1.164f * (y0-16) + 2.018f * (cb-128);
+         pRow[0] = r * scale;
+         pRow[1] = g * scale;
+         pRow[2] = b * scale;
+         pRow[3] = 1.0f;
+         pRow += 4;
+      }
+      p += dst_stride;
+   }
+}
+
+
+void
+pipe_tile_raw_to_rgba(enum pipe_format format,
+                      void *src,
+                      uint w, uint h,
+                      float *dst, unsigned dst_stride)
+{
+   switch (format) {
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+      a8r8g8b8_get_tile_rgba((unsigned *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_X8R8G8B8_UNORM:
+      x8r8g8b8_get_tile_rgba((unsigned *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      b8g8r8a8_get_tile_rgba((unsigned *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_A1R5G5B5_UNORM:
+      a1r5g5b5_get_tile_rgba((ushort *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_A4R4G4B4_UNORM:
+      a4r4g4b4_get_tile_rgba((ushort *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_R5G6B5_UNORM:
+      r5g6b5_get_tile_rgba((ushort *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_L8_UNORM:
+      l8_get_tile_rgba((ubyte *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_A8_UNORM:
+      a8_get_tile_rgba((ubyte *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_I8_UNORM:
+      i8_get_tile_rgba((ubyte *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_A8L8_UNORM:
+      a8l8_get_tile_rgba((ushort *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_R16_SNORM:
+      r16_get_tile_rgba((short *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_R16G16B16A16_SNORM:
+      r16g16b16a16_get_tile_rgba((short *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_Z16_UNORM:
+      z16_get_tile_rgba((ushort *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_Z32_UNORM:
+      z32_get_tile_rgba((unsigned *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_S8Z24_UNORM:
+   case PIPE_FORMAT_X8Z24_UNORM:
+      s8z24_get_tile_rgba((unsigned *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_Z24S8_UNORM:
+      z24s8_get_tile_rgba((unsigned *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_Z32_FLOAT:
+      z32f_get_tile_rgba((float *) src, w, h, dst, dst_stride);
+      break;
+   case PIPE_FORMAT_YCBCR:
+      ycbcr_get_tile_rgba((ushort *) src, w, h, dst, dst_stride, FALSE);
+      break;
+   case PIPE_FORMAT_YCBCR_REV:
+      ycbcr_get_tile_rgba((ushort *) src, w, h, dst, dst_stride, TRUE);
+      break;
+   default:
+      assert(0);
+   }
+}
+
+
+void
+pipe_get_tile_rgba(struct pipe_surface *ps,
+                   uint x, uint y, uint w, uint h,
+                   float *p)
+{
+   unsigned dst_stride = w * 4;
+   void *packed;
+
+   if (pipe_clip_tile(x, y, &w, &h, ps))
+      return;
+
+   packed = MALLOC(pf_get_nblocks(&ps->block, w, h) * ps->block.size);
+
+   if (!packed)
+      return;
+
+   if(ps->format == PIPE_FORMAT_YCBCR || ps->format == PIPE_FORMAT_YCBCR_REV)
+      assert((x & 1) == 0);
+
+   pipe_get_tile_raw(ps, x, y, w, h, packed, 0);
+
+   pipe_tile_raw_to_rgba(ps->format, packed, w, h, p, dst_stride);
+
+   FREE(packed);
+}
+
+
+void
+pipe_put_tile_rgba(struct pipe_surface *ps,
+                   uint x, uint y, uint w, uint h,
+                   const float *p)
+{
+   unsigned src_stride = w * 4;
+   void *packed;
+
+   if (pipe_clip_tile(x, y, &w, &h, ps))
+      return;
+
+   packed = MALLOC(pf_get_nblocks(&ps->block, w, h) * ps->block.size);
+
+   if (!packed)
+      return;
+
+   switch (ps->format) {
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
+      a8r8g8b8_put_tile_rgba((unsigned *) packed, w, h, p, src_stride);
+      break;
+   case PIPE_FORMAT_X8R8G8B8_UNORM:
+      x8r8g8b8_put_tile_rgba((unsigned *) packed, w, h, p, src_stride);
+      break;
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+      b8g8r8a8_put_tile_rgba((unsigned *) packed, w, h, p, src_stride);
+      break;
+   case PIPE_FORMAT_A1R5G5B5_UNORM:
+      a1r5g5b5_put_tile_rgba((ushort *) packed, w, h, p, src_stride);
+      break;
+   case PIPE_FORMAT_R5G6B5_UNORM:
+      r5g6b5_put_tile_rgba((ushort *) packed, w, h, p, src_stride);
+      break;
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      assert(0);
+      break;
+   case PIPE_FORMAT_A4R4G4B4_UNORM:
+      a4r4g4b4_put_tile_rgba((ushort *) packed, w, h, p, src_stride);
+      break;
+   case PIPE_FORMAT_L8_UNORM:
+      l8_put_tile_rgba((ubyte *) packed, w, h, p, src_stride);
+      break;
+   case PIPE_FORMAT_A8_UNORM:
+      a8_put_tile_rgba((ubyte *) packed, w, h, p, src_stride);
+      break;
+   case PIPE_FORMAT_I8_UNORM:
+      i8_put_tile_rgba((ubyte *) packed, w, h, p, src_stride);
+      break;
+   case PIPE_FORMAT_A8L8_UNORM:
+      a8l8_put_tile_rgba((ushort *) packed, w, h, p, src_stride);
+      break;
+   case PIPE_FORMAT_R16_SNORM:
+      r16_put_tile_rgba((short *) packed, w, h, p, src_stride);
+      break;
+   case PIPE_FORMAT_R16G16B16A16_SNORM:
+      r16g16b16a16_put_tile_rgba((short *) packed, w, h, p, src_stride);
+      break;
+   case PIPE_FORMAT_Z16_UNORM:
+      /*z16_put_tile_rgba((ushort *) packed, w, h, p, src_stride);*/
+      break;
+   case PIPE_FORMAT_Z32_UNORM:
+      /*z32_put_tile_rgba((unsigned *) packed, w, h, p, src_stride);*/
+      break;
+   case PIPE_FORMAT_S8Z24_UNORM:
+   case PIPE_FORMAT_X8Z24_UNORM:
+      /*s8z24_put_tile_rgba((unsigned *) packed, w, h, p, src_stride);*/
+      break;
+   case PIPE_FORMAT_Z24S8_UNORM:
+      /*z24s8_put_tile_rgba((unsigned *) packed, w, h, p, src_stride);*/
+      break;
+   default:
+      assert(0);
+   }
+
+   pipe_put_tile_raw(ps, x, y, w, h, packed, 0);
+
+   FREE(packed);
+}
+
+
+/**
+ * Get a block of Z values, converted to 32-bit range.
+ */
+void
+pipe_get_tile_z(struct pipe_surface *ps,
+                uint x, uint y, uint w, uint h,
+                uint *z)
+{
+   const uint dstStride = w;
+   ubyte *map;
+   uint *pDest = z;
+   uint i, j;
+
+   if (pipe_clip_tile(x, y, &w, &h, ps))
+      return;
+
+   map = (ubyte *)pipe_surface_map(ps, PIPE_BUFFER_USAGE_CPU_READ);
+   if (!map) {
+      assert(0);
+      return;
+   }
+
+   switch (ps->format) {
+   case PIPE_FORMAT_Z32_UNORM:
+      {
+         const uint *pSrc
+            = (const uint *)(map  + y * ps->stride + x*4);
+         for (i = 0; i < h; i++) {
+            memcpy(pDest, pSrc, 4 * w);
+            pDest += dstStride;
+            pSrc += ps->stride/4;
+         }
+      }
+      break;
+   case PIPE_FORMAT_S8Z24_UNORM:
+   case PIPE_FORMAT_X8Z24_UNORM:
+      {
+         const uint *pSrc
+            = (const uint *)(map + y * ps->stride + x*4);
+         for (i = 0; i < h; i++) {
+            for (j = 0; j < w; j++) {
+               /* convert 24-bit Z to 32-bit Z */
+               pDest[j] = (pSrc[j] << 8) | (pSrc[j] & 0xff);
+            }
+            pDest += dstStride;
+            pSrc += ps->stride/4;
+         }
+      }
+      break;
+   case PIPE_FORMAT_Z16_UNORM:
+      {
+         const ushort *pSrc
+            = (const ushort *)(map + y * ps->stride + x*2);
+         for (i = 0; i < h; i++) {
+            for (j = 0; j < w; j++) {
+               /* convert 16-bit Z to 32-bit Z */
+               pDest[j] = (pSrc[j] << 16) | pSrc[j];
+            }
+            pDest += dstStride;
+            pSrc += ps->stride/2;
+         }
+      }
+      break;
+   default:
+      assert(0);
+   }
+
+   pipe_surface_unmap(ps);
+}
+
+
+void
+pipe_put_tile_z(struct pipe_surface *ps,
+                uint x, uint y, uint w, uint h,
+                const uint *zSrc)
+{
+   const uint srcStride = w;
+   const uint *pSrc = zSrc;
+   ubyte *map;
+   uint i, j;
+
+   if (pipe_clip_tile(x, y, &w, &h, ps))
+      return;
+
+   map = (ubyte *)pipe_surface_map(ps, PIPE_BUFFER_USAGE_CPU_WRITE);
+   if (!map) {
+      assert(0);
+      return;
+   }
+
+   switch (ps->format) {
+   case PIPE_FORMAT_Z32_UNORM:
+      {
+         uint *pDest = (uint *) (map + y * ps->stride + x*4);
+         for (i = 0; i < h; i++) {
+            memcpy(pDest, pSrc, 4 * w);
+            pDest += ps->stride/4;
+            pSrc += srcStride;
+         }
+      }
+      break;
+   case PIPE_FORMAT_S8Z24_UNORM:
+   case PIPE_FORMAT_X8Z24_UNORM:
+      {
+         uint *pDest = (uint *) (map + y * ps->stride + x*4);
+         for (i = 0; i < h; i++) {
+            for (j = 0; j < w; j++) {
+               /* convert 32-bit Z to 24-bit Z (0 stencil) */
+               pDest[j] = pSrc[j] >> 8;
+            }
+            pDest += ps->stride/4;
+            pSrc += srcStride;
+         }
+      }
+      break;
+   case PIPE_FORMAT_Z16_UNORM:
+      {
+         ushort *pDest = (ushort *) (map + y * ps->stride + x*2);
+         for (i = 0; i < h; i++) {
+            for (j = 0; j < w; j++) {
+               /* convert 32-bit Z to 16-bit Z */
+               pDest[j] = pSrc[j] >> 16;
+            }
+            pDest += ps->stride/2;
+            pSrc += srcStride;
+         }
+      }
+      break;
+   default:
+      assert(0);
+   }
+
+   pipe_surface_unmap(ps);
+}
+
+
diff --git a/src/gallium/auxiliary/util/u_tile.h b/src/gallium/auxiliary/util/u_tile.h
new file mode 100644
index 00000000000..a8ac8053083
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_tile.h
@@ -0,0 +1,101 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#ifndef P_TILE_H
+#define P_TILE_H
+
+#include "pipe/p_compiler.h"
+
+struct pipe_surface;
+
+
+/**
+ * Clip tile against surface dims.
+ * \return TRUE if tile is totally clipped, FALSE otherwise
+ */
+static INLINE boolean
+pipe_clip_tile(uint x, uint y, uint *w, uint *h, const struct pipe_surface *ps)
+{
+   if (x >= ps->width)
+      return TRUE;
+   if (y >= ps->height)
+      return TRUE;
+   if (x + *w > ps->width)
+      *w = ps->width - x;
+   if (y + *h > ps->height)
+      *h = ps->height - y;
+   return FALSE;
+}
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+void
+pipe_get_tile_raw(struct pipe_surface *ps,
+                  uint x, uint y, uint w, uint h,
+                  void *p, int dst_stride);
+
+void
+pipe_put_tile_raw(struct pipe_surface *ps,
+                  uint x, uint y, uint w, uint h,
+                  const void *p, int src_stride);
+
+
+void
+pipe_get_tile_rgba(struct pipe_surface *ps,
+                   uint x, uint y, uint w, uint h,
+                   float *p);
+
+void
+pipe_put_tile_rgba(struct pipe_surface *ps,
+                   uint x, uint y, uint w, uint h,
+                   const float *p);
+
+
+void
+pipe_get_tile_z(struct pipe_surface *ps,
+                uint x, uint y, uint w, uint h,
+                uint *z);
+
+void
+pipe_put_tile_z(struct pipe_surface *ps,
+                uint x, uint y, uint w, uint h,
+                const uint *z);
+
+void
+pipe_tile_raw_to_rgba(enum pipe_format format,
+                      void *src,
+                      uint w, uint h,
+                      float *dst, unsigned dst_stride);
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
diff --git a/src/gallium/auxiliary/util/u_time.c b/src/gallium/auxiliary/util/u_time.c
new file mode 100644
index 00000000000..dde2c74fa83
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_time.c
@@ -0,0 +1,220 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * OS independent time-manipulation functions.
+ * 
+ * @author Jose Fonseca <[email protected]>
+ */
+
+
+#include "pipe/p_config.h"
+
+#if defined(PIPE_OS_LINUX)
+#include <sys/time.h>
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY)
+#include <windows.h>
+#include <winddi.h>
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT)
+#include <windows.h>
+extern VOID KeQuerySystemTime(PLARGE_INTEGER);
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_USER) || defined(PIPE_SUBSYSTEM_WINDOWS_CE)
+#include <windows.h>
+#else
+#error Unsupported OS
+#endif
+
+#include "util/u_time.h"
+
+
+#if defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) || defined(PIPE_SUBSYSTEM_WINDOWS_USER) || defined(PIPE_SUBSYSTEM_WINDOWS_CE)
+
+static int64_t frequency = 0;
+
+static INLINE void 
+util_time_get_frequency(void)
+{
+   if(!frequency) {
+#if defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY)
+      LONGLONG temp;
+      EngQueryPerformanceFrequency(&temp);
+      frequency = temp;
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_USER) || defined(PIPE_SUBSYSTEM_WINDOWS_CE)
+      LARGE_INTEGER temp;
+      QueryPerformanceFrequency(&temp);
+      frequency = temp.QuadPart;
+#endif
+   }
+}
+#endif
+
+
+void 
+util_time_get(struct util_time *t)
+{
+#if defined(PIPE_OS_LINUX)
+   gettimeofday(&t->tv, NULL);
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY)
+   LONGLONG temp;
+   EngQueryPerformanceCounter(&temp);
+   t->counter = temp;
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT)
+   /* Updated every 10 miliseconds, measured in units of 100 nanoseconds.
+    * http://msdn.microsoft.com/en-us/library/ms801642.aspx */
+   LARGE_INTEGER temp;
+   KeQuerySystemTime(&temp);
+   t->counter = temp.QuadPart;
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_USER) || defined(PIPE_SUBSYSTEM_WINDOWS_CE)
+   LARGE_INTEGER temp;
+   QueryPerformanceCounter(&temp);
+   t->counter = temp.QuadPart;
+#endif
+}
+
+
+void 
+util_time_add(const struct util_time *t1,
+              int64_t usecs,
+              struct util_time *t2)
+{
+#if defined(PIPE_OS_LINUX)
+   t2->tv.tv_sec = t1->tv.tv_sec + usecs / 1000000;
+   t2->tv.tv_usec = t1->tv.tv_usec + usecs % 1000000;
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) || defined(PIPE_SUBSYSTEM_WINDOWS_USER) || defined(PIPE_SUBSYSTEM_WINDOWS_CE)
+   util_time_get_frequency();
+   t2->counter = t1->counter + (usecs * frequency + INT64_C(999999))/INT64_C(1000000);
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT)
+   /* 1 tick = 100 nano seconds. */
+   t2->counter = t1->counter + usecs * 10;
+#else
+   LARGE_INTEGER temp;
+   LONGLONG freq;
+   freq = temp.QuadPart;
+   t2->counter = t1->counter + (usecs * freq)/1000000L;
+#endif
+}
+
+
+int64_t
+util_time_diff(const struct util_time *t1, 
+               const struct util_time *t2)
+{
+#if defined(PIPE_OS_LINUX)
+   return (t2->tv.tv_usec - t1->tv.tv_usec) + 
+          (t2->tv.tv_sec - t1->tv.tv_sec)*1000000;
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) || defined(PIPE_SUBSYSTEM_WINDOWS_USER) || defined(PIPE_SUBSYSTEM_WINDOWS_CE)
+   util_time_get_frequency();
+   return (t2->counter - t1->counter)*INT64_C(1000000)/frequency;
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT)
+   return (t2->counter - t1->counter)/10;
+#endif
+}
+
+
+
+uint64_t
+util_time_micros( void )
+{
+   struct util_time t1;
+   
+   util_time_get(&t1);
+   
+#if defined(PIPE_OS_LINUX)
+   return t1.tv.tv_usec + t1.tv.tv_sec*1000000LL;
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY) || defined(PIPE_SUBSYSTEM_WINDOWS_USER) || defined(PIPE_SUBSYSTEM_WINDOWS_CE)
+   util_time_get_frequency();
+   return t1.counter*INT64_C(1000000)/frequency;
+#elif defined(PIPE_SUBSYSTEM_WINDOWS_MINIPORT)
+   return t1.counter/10;
+#endif
+}
+
+
+
+/**
+ * Compare two time values.
+ * 
+ * Not publicly available because it does not take in account wrap-arounds. 
+ * Use util_time_timeout instead.
+ */
+static INLINE int
+util_time_compare(const struct util_time *t1, 
+                  const struct util_time *t2)
+{
+#if defined(PIPE_OS_LINUX)
+   if (t1->tv.tv_sec < t2->tv.tv_sec)
+      return -1;
+   else if(t1->tv.tv_sec > t2->tv.tv_sec)
+      return 1;
+   else if (t1->tv.tv_usec < t2->tv.tv_usec)
+      return -1;
+   else if(t1->tv.tv_usec > t2->tv.tv_usec)
+      return 1;
+   else 
+      return 0;
+#elif defined(PIPE_OS_WINDOWS)
+   if (t1->counter < t2->counter)
+      return -1;
+   else if(t1->counter > t2->counter)
+      return 1;
+   else 
+      return 0;
+#endif
+}
+
+
+boolean 
+util_time_timeout(const struct util_time *start, 
+                  const struct util_time *end,
+                  const struct util_time *curr) 
+{
+   if(util_time_compare(start, end) <= 0)
+      return !(util_time_compare(start, curr) <= 0 && util_time_compare(curr, end) < 0);
+   else
+      return !(util_time_compare(start, curr) <= 0 || util_time_compare(curr, end) < 0);
+}
+
+
+#if defined(PIPE_SUBSYSTEM_WINDOWS_DISPLAY)
+void util_time_sleep(unsigned usecs)
+{
+   LONGLONG start, curr, end;
+   
+   EngQueryPerformanceCounter(&start);
+   
+   if(!frequency)
+      EngQueryPerformanceFrequency(&frequency);
+   
+   end = start + (usecs * frequency + 999999LL)/1000000LL;
+   
+   do {
+      EngQueryPerformanceCounter(&curr);
+   } while(start <= curr && curr < end || 
+	   end < start && (curr < end || start <= curr));
+}
+#endif
diff --git a/src/gallium/auxiliary/util/u_time.h b/src/gallium/auxiliary/util/u_time.h
new file mode 100644
index 00000000000..35d97d16c73
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_time.h
@@ -0,0 +1,104 @@
+/**************************************************************************
+ *
+ * Copyright 2008 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ **************************************************************************/
+
+/**
+ * @file
+ * OS independent time-manipulation functions.
+ * 
+ * @author Jose Fonseca <[email protected]>
+ */
+
+#ifndef U_TIME_H_
+#define U_TIME_H_
+
+
+#include "pipe/p_config.h"
+
+#if defined(PIPE_OS_LINUX)
+#include <time.h> /* timeval */
+#include <unistd.h> /* usleep */
+#endif
+
+#include "pipe/p_compiler.h"
+
+
+#ifdef	__cplusplus
+extern "C" {
+#endif
+
+
+/**
+ * Time abstraction.
+ * 
+ * Do not access this structure directly. Use the provided function instead.
+ */
+struct util_time 
+{
+#if defined(PIPE_OS_LINUX)
+   struct timeval tv;
+#else
+   int64_t counter;
+#endif
+};
+   
+
+void 
+util_time_get(struct util_time *t);
+
+void 
+util_time_add(const struct util_time *t1,
+              int64_t usecs,
+              struct util_time *t2);
+
+uint64_t
+util_time_micros( void );
+
+int64_t
+util_time_diff(const struct util_time *t1, 
+               const struct util_time *t2);
+
+/**
+ * Returns non-zero when the timeout expires.
+ */
+boolean 
+util_time_timeout(const struct util_time *start, 
+                  const struct util_time *end,
+                  const struct util_time *curr);
+
+#if defined(PIPE_OS_LINUX)
+#define util_time_sleep usleep
+#else
+void
+util_time_sleep(unsigned usecs);
+#endif
+
+
+#ifdef	__cplusplus
+}
+#endif
+
+#endif /* U_TIME_H_ */
diff --git a/src/gallium/auxiliary/util/u_timed_winsys.c b/src/gallium/auxiliary/util/u_timed_winsys.c
new file mode 100644
index 00000000000..f237e12d735
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_timed_winsys.c
@@ -0,0 +1,316 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+/*
+ * Authors: Keith Whitwell <keithw-at-tungstengraphics-dot-com>
+ */
+
+#include "pipe/internal/p_winsys_screen.h"
+#include "u_timed_winsys.h"
+#include "util/u_memory.h"
+#include "util/u_time.h"
+
+
+struct timed_winsys {
+   struct pipe_winsys base;
+   struct pipe_winsys *backend;
+   uint64_t last_dump;
+   struct {
+      const char *name_key;
+      double total;
+      unsigned calls;
+   } funcs[13];
+};
+
+
+static struct timed_winsys *timed_winsys( struct pipe_winsys *winsys )
+{
+   return (struct timed_winsys *)winsys;
+}
+
+
+static uint64_t time_start( void )
+{
+   return util_time_micros();
+}
+
+
+static void time_display( struct pipe_winsys *winsys )
+{
+   struct timed_winsys *tws = timed_winsys(winsys);
+   unsigned i;
+   double overall = 0;
+
+   for (i = 0; i < Elements(tws->funcs); i++) {
+      if (tws->funcs[i].name_key) {
+         debug_printf("*** %-25s %5.3fms (%d calls, avg %.3fms)\n", 
+                      tws->funcs[i].name_key,
+                      tws->funcs[i].total,
+                      tws->funcs[i].calls,
+                      tws->funcs[i].total / tws->funcs[i].calls);
+         overall += tws->funcs[i].total;
+         tws->funcs[i].calls = 0;
+         tws->funcs[i].total = 0;
+      }
+   }
+
+   debug_printf("*** %-25s %5.3fms\n", 
+                "OVERALL WINSYS",
+                overall);
+}
+
+static void time_finish( struct pipe_winsys *winsys,
+                         long long startval, 
+                         unsigned idx,
+                         const char *name ) 
+{
+   struct timed_winsys *tws = timed_winsys(winsys);
+   uint64_t endval = util_time_micros();
+   double elapsed = (endval - startval)/1000.0;
+
+   if (endval - startval > 1000LL) 
+      debug_printf("*** %s %.3f\n", name, elapsed );
+
+   assert( tws->funcs[idx].name_key == name ||
+           tws->funcs[idx].name_key == NULL);
+
+   tws->funcs[idx].name_key = name;
+   tws->funcs[idx].total += elapsed;
+   tws->funcs[idx].calls++;
+
+   if (endval - tws->last_dump > 10LL * 1000LL * 1000LL) {
+      time_display( winsys );
+      tws->last_dump = endval;
+   }
+}
+
+
+/* Pipe has no concept of pools, but the psb driver passes a flag that
+ * can be mapped onto pools in the backend.
+ */
+static struct pipe_buffer *
+timed_buffer_create(struct pipe_winsys *winsys, 
+                    unsigned alignment, 
+                    unsigned usage, 
+                    unsigned size )
+{
+   struct pipe_winsys *backend = timed_winsys(winsys)->backend;
+   uint64_t start = time_start();
+
+   struct pipe_buffer *buf =
+      backend->buffer_create( backend, alignment, usage, size );
+
+   time_finish(winsys, start, 0, __FUNCTION__);
+   
+   return buf;
+}
+
+
+
+
+static struct pipe_buffer *
+timed_user_buffer_create(struct pipe_winsys *winsys,
+                             void *data, 
+                             unsigned bytes) 
+{
+   struct pipe_winsys *backend = timed_winsys(winsys)->backend;
+   uint64_t start = time_start();
+
+   struct pipe_buffer *buf = backend->user_buffer_create( backend, data, bytes );
+
+   time_finish(winsys, start, 1, __FUNCTION__);
+   
+   return buf;
+}
+
+
+static void *
+timed_buffer_map(struct pipe_winsys *winsys,
+                     struct pipe_buffer *buf,
+                     unsigned flags)
+{
+   struct pipe_winsys *backend = timed_winsys(winsys)->backend;
+   uint64_t start = time_start();
+
+   void *map = backend->buffer_map( backend, buf, flags );
+
+   time_finish(winsys, start, 2, __FUNCTION__);
+   
+   return map;
+}
+
+
+static void
+timed_buffer_unmap(struct pipe_winsys *winsys,
+                       struct pipe_buffer *buf)
+{
+   struct pipe_winsys *backend = timed_winsys(winsys)->backend;
+   uint64_t start = time_start();
+
+   backend->buffer_unmap( backend, buf );
+
+   time_finish(winsys, start, 3, __FUNCTION__);
+}
+
+
+static void
+timed_buffer_destroy(struct pipe_winsys *winsys,
+                         struct pipe_buffer *buf)
+{
+   struct pipe_winsys *backend = timed_winsys(winsys)->backend;
+   uint64_t start = time_start();
+
+   backend->buffer_destroy( backend, buf );
+
+   time_finish(winsys, start, 4, __FUNCTION__);
+}
+
+
+static void
+timed_flush_frontbuffer( struct pipe_winsys *winsys,
+                         struct pipe_surface *surf,
+                         void *context_private)
+{
+   struct pipe_winsys *backend = timed_winsys(winsys)->backend;
+   uint64_t start = time_start();
+
+   backend->flush_frontbuffer( backend, surf, context_private );
+
+   time_finish(winsys, start, 5, __FUNCTION__);
+}
+
+
+
+
+static struct pipe_buffer *
+timed_surface_buffer_create(struct pipe_winsys *winsys,
+                              unsigned width, unsigned height,
+                              enum pipe_format format, 
+                              unsigned usage,
+                              unsigned *stride)
+{
+   struct pipe_winsys *backend = timed_winsys(winsys)->backend;
+   uint64_t start = time_start();
+
+   struct pipe_buffer *ret = backend->surface_buffer_create( backend, width, height, 
+                                                             format, usage, stride );
+
+   time_finish(winsys, start, 7, __FUNCTION__);
+   
+   return ret;
+}
+
+
+static const char *
+timed_get_name( struct pipe_winsys *winsys )
+{
+   struct pipe_winsys *backend = timed_winsys(winsys)->backend;
+   uint64_t start = time_start();
+
+   const char *ret = backend->get_name( backend );
+
+   time_finish(winsys, start, 9, __FUNCTION__);
+   
+   return ret;
+}
+
+static void
+timed_fence_reference(struct pipe_winsys *winsys,
+                    struct pipe_fence_handle **ptr,
+                    struct pipe_fence_handle *fence)
+{
+   struct pipe_winsys *backend = timed_winsys(winsys)->backend;
+   uint64_t start = time_start();
+
+   backend->fence_reference( backend, ptr, fence );
+
+   time_finish(winsys, start, 10, __FUNCTION__);
+}
+
+
+static int
+timed_fence_signalled( struct pipe_winsys *winsys,
+                       struct pipe_fence_handle *fence,
+                       unsigned flag )
+{
+   struct pipe_winsys *backend = timed_winsys(winsys)->backend;
+   uint64_t start = time_start();
+
+   int ret = backend->fence_signalled( backend, fence, flag );
+
+   time_finish(winsys, start, 11, __FUNCTION__);
+   
+   return ret;
+}
+
+static int
+timed_fence_finish( struct pipe_winsys *winsys,
+                     struct pipe_fence_handle *fence,
+                     unsigned flag )
+{
+   struct pipe_winsys *backend = timed_winsys(winsys)->backend;
+   uint64_t start = time_start();
+
+   int ret = backend->fence_finish( backend, fence, flag );
+
+   time_finish(winsys, start, 12, __FUNCTION__);
+   
+   return ret;
+}
+
+static void
+timed_winsys_destroy( struct pipe_winsys *winsys )
+{
+   struct pipe_winsys *backend = timed_winsys(winsys)->backend;
+   backend->destroy( backend );
+   FREE(winsys);
+}
+
+
+
+struct pipe_winsys *u_timed_winsys_create( struct pipe_winsys *backend )
+{
+   struct timed_winsys *ws = CALLOC_STRUCT(timed_winsys);
+   
+   ws->base.user_buffer_create = timed_user_buffer_create;
+   ws->base.buffer_map = timed_buffer_map;
+   ws->base.buffer_unmap = timed_buffer_unmap;
+   ws->base.buffer_destroy = timed_buffer_destroy;
+   ws->base.buffer_create = timed_buffer_create;
+   ws->base.surface_buffer_create = timed_surface_buffer_create;
+   ws->base.flush_frontbuffer = timed_flush_frontbuffer;
+   ws->base.get_name = timed_get_name;
+   ws->base.fence_reference = timed_fence_reference;
+   ws->base.fence_signalled = timed_fence_signalled;
+   ws->base.fence_finish = timed_fence_finish;
+   ws->base.destroy = timed_winsys_destroy;
+   
+   ws->backend = backend;
+
+   return &ws->base;
+}
+
diff --git a/src/gallium/auxiliary/util/u_timed_winsys.h b/src/gallium/auxiliary/util/u_timed_winsys.h
new file mode 100644
index 00000000000..542365112d7
--- /dev/null
+++ b/src/gallium/auxiliary/util/u_timed_winsys.h
@@ -0,0 +1,41 @@
+/**************************************************************************
+ * 
+ * Copyright 2006 Tungsten Graphics, Inc., Bismarck, ND., USA
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+ * THE COPYRIGHT HOLDERS, AUTHORS AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR 
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE 
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * 
+ **************************************************************************/
+/*
+ * Authors: Keith Whitwell <keithw-at-tungstengraphics-dot-com>
+ */
+
+
+#ifndef U_TIMED_WINSYS_H
+#define U_TIMED_WINSYS_H
+
+
+struct pipe_winsys;
+struct pipe_winsys *u_timed_winsys_create( struct pipe_winsys *backend );
+
+
+#endif