Merge branch 'master' into gallium-sampler-view

Conflicts:
	src/gallium/auxiliary/Makefile
	src/gallium/auxiliary/SConscript
	src/gallium/auxiliary/tgsi/tgsi_exec.c
	src/gallium/auxiliary/util/u_blitter.c
	src/gallium/drivers/i915/i915_context.h
	src/gallium/drivers/i965/brw_context.h
	src/gallium/drivers/llvmpipe/lp_context.h
	src/gallium/drivers/nv50/nv50_context.h
	src/gallium/drivers/nv50/nv50_state_validate.c
	src/gallium/drivers/nv50/nv50_tex.c
	src/gallium/drivers/r300/r300_blit.c
	src/gallium/drivers/r300/r300_context.h
	src/gallium/drivers/r300/r300_emit.c
	src/gallium/drivers/r300/r300_state.c
	src/gallium/drivers/softpipe/sp_context.h
	src/gallium/drivers/svga/svga_context.h
	src/gallium/drivers/svga/svga_pipe_sampler.c

200 files changed, 6888 insertions, 6051 deletions

diff --git a/src/gallium/drivers/cell/ppu/Makefile b/src/gallium/drivers/cell/ppu/Makefile
index c92f8e5cba2..8769b826b5f 100644
--- a/src/gallium/drivers/cell/ppu/Makefile
+++ b/src/gallium/drivers/cell/ppu/Makefile
@@ -21,6 +21,7 @@ SPU_CODE_MODULE = ../spu/g3d_spu.a
 
 SOURCES = \
 	cell_batch.c \
+	cell_buffer.c \
 	cell_clear.c \
 	cell_context.c \
 	cell_draw_arrays.c \
diff --git a/src/gallium/drivers/cell/ppu/cell_buffer.c b/src/gallium/drivers/cell/ppu/cell_buffer.c
new file mode 100644
index 00000000000..f56a28d62e2
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_buffer.c
@@ -0,0 +1,118 @@
+/**************************************************************************
+ *
+ * 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 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.
+ *
+ **************************************************************************/
+
+
+#include "util/u_inlines.h"
+#include "util/u_memory.h"
+#include "util/u_math.h"
+
+#include "cell_screen.h"
+#include "cell_buffer.h"
+
+
+static void *
+cell_buffer_map(struct pipe_screen *screen,
+                    struct pipe_buffer *buf,
+                    unsigned flags)
+{
+   struct cell_buffer *cell_buf = cell_buffer(buf);
+   return cell_buf->data;
+}
+
+
+static void
+cell_buffer_unmap(struct pipe_screen *screen,
+                      struct pipe_buffer *buf)
+{
+}
+
+
+static void
+cell_buffer_destroy(struct pipe_buffer *buf)
+{
+   struct cell_buffer *sbuf = cell_buffer(buf);
+
+   if (!sbuf->userBuffer)
+      align_free(sbuf->data);
+      
+   FREE(sbuf);
+}
+
+
+static struct pipe_buffer *
+cell_buffer_create(struct pipe_screen *screen,
+                       unsigned alignment,
+                       unsigned usage,
+                       unsigned size)
+{
+   struct cell_buffer *buffer = CALLOC_STRUCT(cell_buffer);
+
+   pipe_reference_init(&buffer->base.reference, 1);
+   buffer->base.screen = screen;
+   buffer->base.alignment = MAX2(alignment, 16);
+   buffer->base.usage = usage;
+   buffer->base.size = size;
+
+   buffer->data = align_malloc(size, alignment);
+
+   return &buffer->base;
+}
+
+
+/**
+ * Create buffer which wraps user-space data.
+ */
+static struct pipe_buffer *
+cell_user_buffer_create(struct pipe_screen *screen,
+                            void *ptr,
+                            unsigned bytes)
+{
+   struct cell_buffer *buffer;
+
+   buffer = CALLOC_STRUCT(cell_buffer);
+   if(!buffer)
+      return NULL;
+
+   pipe_reference_init(&buffer->base.reference, 1);
+   buffer->base.screen = screen;
+   buffer->base.size = bytes;
+   buffer->userBuffer = TRUE;
+   buffer->data = ptr;
+
+   return &buffer->base;
+}
+
+
+void
+cell_init_screen_buffer_funcs(struct pipe_screen *screen)
+{
+   screen->buffer_create = cell_buffer_create;
+   screen->user_buffer_create = cell_user_buffer_create;
+   screen->buffer_map = cell_buffer_map;
+   screen->buffer_unmap = cell_buffer_unmap;
+   screen->buffer_destroy = cell_buffer_destroy;
+}
diff --git a/src/gallium/drivers/cell/ppu/cell_winsys.h b/src/gallium/drivers/cell/ppu/cell_buffer.h
index e227e065ff3..ef0a8a70e59 100644
--- a/src/gallium/drivers/cell/ppu/cell_winsys.h
+++ b/src/gallium/drivers/cell/ppu/cell_buffer.h
@@ -1,8 +1,8 @@
 /**************************************************************************
- * 
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ *
+ * 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
@@ -10,38 +10,46 @@
  * 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
+ * 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.
- * 
+ *
  **************************************************************************/
 
-
-#ifndef CELL_WINSYS_H
-#define CELL_WINSYS_H
+#ifndef SP_BUFFER_H
+#define SP_BUFFER_H
 
 #include "pipe/p_compiler.h"
+#include "pipe/p_state.h"
 
 
-/**
- * Very simple winsys at this time.
- * Will probably eventually add SPU control info.
- */
-struct cell_winsys
+struct cell_buffer
 {
-   uint dummy;
+   struct pipe_buffer base;
+   boolean userBuffer;  /** Is this a user-space buffer? */
+   void *data;
 };
 
 
+/** Cast wrapper */
+static INLINE struct cell_buffer *
+cell_buffer( struct pipe_buffer *buf )
+{
+   return (struct cell_buffer *)buf;
+}
+
+
+void
+cell_init_screen_buffer_funcs(struct pipe_screen *screen);
 
 
-#endif
+#endif /* SP_BUFFER_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_context.c b/src/gallium/drivers/cell/ppu/cell_context.c
index 5bff9869fd0..afcea616d59 100644
--- a/src/gallium/drivers/cell/ppu/cell_context.c
+++ b/src/gallium/drivers/cell/ppu/cell_context.c
@@ -36,7 +36,6 @@
 #include "pipe/p_defines.h"
 #include "pipe/p_format.h"
 #include "util/u_memory.h"
-#include "util/u_simple_screen.h"
 #include "pipe/p_screen.h"
 
 #include "draw/draw_context.h"
@@ -137,7 +136,7 @@ cell_create_context(struct pipe_screen *screen,
    memset(cell, 0, sizeof(*cell));
 
    cell->winsys = NULL;		/* XXX: fixme - get this from screen? */
-   cell->pipe.winsys = screen->winsys;
+   cell->pipe.winsys = NULL;
    cell->pipe.screen = screen;
    cell->pipe.priv = priv;
    cell->pipe.destroy = cell_destroy_context;
diff --git a/src/gallium/drivers/cell/ppu/cell_context.h b/src/gallium/drivers/cell/ppu/cell_context.h
index 0724da9a7aa..4d87f9a038a 100644
--- a/src/gallium/drivers/cell/ppu/cell_context.h
+++ b/src/gallium/drivers/cell/ppu/cell_context.h
@@ -34,7 +34,7 @@
 #include "pipe/p_defines.h"
 #include "draw/draw_vertex.h"
 #include "draw/draw_vbuf.h"
-#include "cell_winsys.h"
+/*#include "cell_winsys.h"*/
 #include "cell/common.h"
 #include "rtasm/rtasm_ppc_spe.h"
 #include "tgsi/tgsi_scan.h"
@@ -93,6 +93,11 @@ struct cell_buffer_list
    struct cell_buffer_node *head;
 };
 
+struct cell_velems_state
+{
+   unsigned count;
+   struct pipe_vertex_element velem[PIPE_MAX_ATTRIBS];
+}
 
 /**
  * Per-context state, subclass of pipe_context.
@@ -110,6 +115,7 @@ struct cell_context
    const struct pipe_rasterizer_state *rasterizer;
    const struct cell_vertex_shader_state *vs;
    const struct cell_fragment_shader_state *fs;
+   const struct cell_velems_state *velems;
 
    struct spe_function logic_op;
 
@@ -126,8 +132,6 @@ struct cell_context
    struct pipe_viewport_state viewport;
    struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
    uint num_vertex_buffers;
-   struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS];
-   uint num_vertex_elements;
 
    ubyte *cbuf_map[PIPE_MAX_COLOR_BUFS];
    ubyte *zsbuf_map;
diff --git a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
index bffd0fac6fe..15d4e8338f4 100644
--- a/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
+++ b/src/gallium/drivers/cell/ppu/cell_draw_arrays.c
@@ -33,48 +33,18 @@
 
 #include "pipe/p_defines.h"
 #include "pipe/p_context.h"
-#include "util/u_simple_screen.h"
 #include "util/u_inlines.h"
 
 #include "cell_context.h"
 #include "cell_draw_arrays.h"
 #include "cell_state.h"
 #include "cell_flush.h"
+#include "cell_buffer.h"
 
 #include "draw/draw_context.h"
 
 
 
-static void
-cell_map_constant_buffers(struct cell_context *sp)
-{
-   struct pipe_winsys *ws = sp->pipe.winsys;
-   uint i;
-   for (i = 0; i < 2; i++) {
-      if (sp->constants[i] && sp->constants[i]->size) {
-         sp->mapped_constants[i] = ws->buffer_map(ws, sp->constants[i],
-                                                   PIPE_BUFFER_USAGE_CPU_READ);
-         cell_flush_buffer_range(sp, sp->mapped_constants[i], 
-                                 sp->constants[i]->size);
-      }
-   }
-
-   draw_set_mapped_constant_buffer(sp->draw, PIPE_SHADER_VERTEX, 0,
-                                   sp->mapped_constants[PIPE_SHADER_VERTEX],
-                                   sp->constants[PIPE_SHADER_VERTEX]->size);
-}
-
-static void
-cell_unmap_constant_buffers(struct cell_context *sp)
-{
-   struct pipe_winsys *ws = sp->pipe.winsys;
-   uint i;
-   for (i = 0; i < 2; i++) {
-      if (sp->constants[i] && sp->constants[i]->size)
-         ws->buffer_unmap(ws, sp->constants[i]);
-      sp->mapped_constants[i] = NULL;
-   }
-}
 
 
 
@@ -93,33 +63,27 @@ cell_draw_range_elements(struct pipe_context *pipe,
                          unsigned max_index,
                          unsigned mode, unsigned start, unsigned count)
 {
-   struct cell_context *sp = cell_context(pipe);
-   struct draw_context *draw = sp->draw;
+   struct cell_context *cell = cell_context(pipe);
+   struct draw_context *draw = cell->draw;
    unsigned i;
 
-   if (sp->dirty)
-      cell_update_derived( sp );
+   if (cell->dirty)
+      cell_update_derived( cell );
 
 #if 0
-   cell_map_surfaces(sp);
+   cell_map_surfaces(cell);
 #endif
-   cell_map_constant_buffers(sp);
 
    /*
     * Map vertex buffers
     */
-   for (i = 0; i < sp->num_vertex_buffers; i++) {
-      void *buf = pipe_buffer_map(pipe->screen,
-                                           sp->vertex_buffer[i].buffer,
-                                           PIPE_BUFFER_USAGE_CPU_READ);
-      cell_flush_buffer_range(sp, buf, sp->vertex_buffer[i].buffer->size);
+   for (i = 0; i < cell->num_vertex_buffers; i++) {
+      void *buf = cell_buffer(cell->vertex_buffer[i].buffer)->data;
       draw_set_mapped_vertex_buffer(draw, i, buf);
    }
    /* Map index buffer, if present */
    if (indexBuffer) {
-      void *mapped_indexes = pipe_buffer_map(pipe->screen,
-                                                      indexBuffer,
-                                                      PIPE_BUFFER_USAGE_CPU_READ);
+      void *mapped_indexes = cell_buffer(indexBuffer)->data;
       draw_set_mapped_element_buffer(draw, indexSize, mapped_indexes);
    }
    else {
@@ -134,17 +98,19 @@ cell_draw_range_elements(struct pipe_context *pipe,
    /*
     * unmap vertex/index buffers - will cause draw module to flush
     */
-   for (i = 0; i < sp->num_vertex_buffers; i++) {
+   for (i = 0; i < cell->num_vertex_buffers; i++) {
       draw_set_mapped_vertex_buffer(draw, i, NULL);
-      pipe_buffer_unmap(pipe->screen, sp->vertex_buffer[i].buffer);
    }
    if (indexBuffer) {
       draw_set_mapped_element_buffer(draw, 0, NULL);
-      pipe_buffer_unmap(pipe->screen, indexBuffer);
    }
 
-   /* Note: leave drawing surfaces mapped */
-   cell_unmap_constant_buffers(sp);
+   /*
+    * TODO: Flush only when a user vertex/index buffer is present
+    * (or even better, modify draw module to do this
+    * internally when this condition is seen?)
+    */
+   draw_flush(draw);
 }
 
 
diff --git a/src/gallium/drivers/cell/ppu/cell_fence.c b/src/gallium/drivers/cell/ppu/cell_fence.c
index e10071529a8..035ef41b898 100644
--- a/src/gallium/drivers/cell/ppu/cell_fence.c
+++ b/src/gallium/drivers/cell/ppu/cell_fence.c
@@ -92,7 +92,6 @@ cell_add_buffer_to_list(struct cell_context *cell,
                         struct cell_buffer_list *list,
                         struct pipe_buffer *buffer)
 {
-   struct pipe_screen *ps = cell->pipe.screen;
    struct cell_buffer_node *node = CALLOC_STRUCT(cell_buffer_node);
    /* create new list node which references the buffer, insert at head */
    if (node) {
@@ -157,8 +156,13 @@ cell_add_fenced_textures(struct cell_context *cell)
          printf("Adding texture %p buffer %p to list\n",
                 ct, ct->tiled_buffer[level]);
 #endif
-         if (ct->buffer)
+#if 00
+         /* XXX this needs to be fixed/restored!
+          * Maybe keep pointers to textures, not buffers.
+          */
+         if (ct->base.buffer)
             cell_add_buffer_to_list(cell, list, ct->buffer);
+#endif
       }
    }
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_gen_fragment.c b/src/gallium/drivers/cell/ppu/cell_gen_fragment.c
index 70683bb3676..576d514741d 100644
--- a/src/gallium/drivers/cell/ppu/cell_gen_fragment.c
+++ b/src/gallium/drivers/cell/ppu/cell_gen_fragment.c
@@ -304,7 +304,7 @@ unpack_colors(struct spe_function *f,
    spe_comment(f, 0, "Unpack framebuffer colors, convert to floats");
 
    switch (color_format) {
-   case PIPE_FORMAT_A8R8G8B8_UNORM:
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
       /* fbB = fbRGBA & mask */
       spe_and(f, fbB_reg, fbRGBA_reg, mask0_reg);
 
@@ -327,7 +327,7 @@ unpack_colors(struct spe_function *f,
       spe_roti(f, fbA_reg, fbA_reg, -24);
       break;
 
-   case PIPE_FORMAT_B8G8R8A8_UNORM:
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
       /* fbA = fbRGBA & mask */
       spe_and(f, fbA_reg, fbRGBA_reg, mask0_reg);
 
@@ -1043,12 +1043,12 @@ gen_pack_colors(struct spe_function *f,
    spe_rotmi(f, a_reg, a_reg, -24);
 
    /* Shift the color bytes according to the surface format */
-   if (color_format == PIPE_FORMAT_A8R8G8B8_UNORM) {
+   if (color_format == PIPE_FORMAT_B8G8R8A8_UNORM) {
       spe_roti(f, g_reg, g_reg, 8);   /* green <<= 8 */
       spe_roti(f, r_reg, r_reg, 16);  /* red <<= 16 */
       spe_roti(f, a_reg, a_reg, 24);  /* alpha <<= 24 */
    }
-   else if (color_format == PIPE_FORMAT_B8G8R8A8_UNORM) {
+   else if (color_format == PIPE_FORMAT_A8R8G8B8_UNORM) {
       spe_roti(f, r_reg, r_reg, 8);   /* red <<= 8 */
       spe_roti(f, g_reg, g_reg, 16);  /* green <<= 16 */
       spe_roti(f, b_reg, b_reg, 24);  /* blue <<= 24 */
@@ -1096,14 +1096,14 @@ gen_colormask(struct spe_function *f,
     * end up, so we can mask them correctly.
     */
    switch(color_format) {
-      case PIPE_FORMAT_A8R8G8B8_UNORM:
+      case PIPE_FORMAT_B8G8R8A8_UNORM:
          /* ARGB */
          a_mask = 0xff000000;
          r_mask = 0x00ff0000;
          g_mask = 0x0000ff00;
          b_mask = 0x000000ff;
          break;
-      case PIPE_FORMAT_B8G8R8A8_UNORM:
+      case PIPE_FORMAT_A8R8G8B8_UNORM:
          /* BGRA */
          b_mask = 0xff000000;
          g_mask = 0x00ff0000;
@@ -1859,8 +1859,8 @@ gen_depth_stencil(struct cell_context *cell,
    spe_comment(f, 0, "Fetch Z/stencil quad from tile");
 
    switch(zs_format) {
-   case PIPE_FORMAT_S8Z24_UNORM: /* fall through */
-   case PIPE_FORMAT_X8Z24_UNORM:
+   case PIPE_FORMAT_Z24S8_UNORM: /* fall through */
+   case PIPE_FORMAT_Z24X8_UNORM:
       /* prepare mask to extract Z vals from ZS vals */
       spe_load_uint(f, zmask_reg, 0x00ffffff);
 
@@ -1880,8 +1880,8 @@ gen_depth_stencil(struct cell_context *cell,
       spe_rotmi(f, fbS_reg, fbZS_reg, -24);
       break;
 
-   case PIPE_FORMAT_Z24S8_UNORM: /* fall through */
-   case PIPE_FORMAT_Z24X8_UNORM:
+   case PIPE_FORMAT_S8Z24_UNORM: /* fall through */
+   case PIPE_FORMAT_X8Z24_UNORM:
       /* convert fragment Z from [0,1] to 32-bit ints */
       spe_cfltu(f, fragZ_reg, fragZ_reg, 32);
 
@@ -1969,13 +1969,13 @@ gen_depth_stencil(struct cell_context *cell,
        * fbS_reg has four 8-bit Z values in bits [7..0].
        */
       spe_comment(f, 0, "Store quad's depth/stencil values in tile");
-      if (zs_format == PIPE_FORMAT_S8Z24_UNORM ||
-          zs_format == PIPE_FORMAT_X8Z24_UNORM) {
+      if (zs_format == PIPE_FORMAT_Z24S8_UNORM ||
+          zs_format == PIPE_FORMAT_Z24X8_UNORM) {
          spe_shli(f, fbS_reg, fbS_reg, 24); /* fbS = fbS << 24 */
          spe_or(f, fbZS_reg, fbS_reg, fbZ_reg); /* fbZS = fbS | fbZ */
       }
-      else if (zs_format == PIPE_FORMAT_Z24S8_UNORM ||
-               zs_format == PIPE_FORMAT_Z24X8_UNORM) {
+      else if (zs_format == PIPE_FORMAT_S8Z24_UNORM ||
+               zs_format == PIPE_FORMAT_X8Z24_UNORM) {
          spe_shli(f, fbZ_reg, fbZ_reg, 8); /* fbZ = fbZ << 8 */
          spe_or(f, fbZS_reg, fbS_reg, fbZ_reg); /* fbZS = fbS | fbZ */
       }
@@ -2015,7 +2015,7 @@ gen_depth_stencil(struct cell_context *cell,
  * code before the fragment shader to cull fragments/quads that are
  * totally occluded/discarded.
  *
- * XXX we only support PIPE_FORMAT_Z24S8_UNORM z/stencil buffer right now.
+ * XXX we only support PIPE_FORMAT_S8Z24_UNORM z/stencil buffer right now.
  *
  * See the spu_default_fragment_ops() function to see how the per-fragment
  * operations would be done with ordinary C code.
diff --git a/src/gallium/drivers/cell/ppu/cell_pipe_state.c b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
index e1cf9dbbee8..2fc82933e4d 100644
--- a/src/gallium/drivers/cell/ppu/cell_pipe_state.c
+++ b/src/gallium/drivers/cell/ppu/cell_pipe_state.c
@@ -323,17 +323,23 @@ cell_sampler_view_destroy(struct pipe_context *pipe,
 static void
 cell_map_surfaces(struct cell_context *cell)
 {
+#if 0
    struct pipe_screen *screen = cell->pipe.screen;
+#endif
    uint i;
 
    for (i = 0; i < 1; i++) {
       struct pipe_surface *ps = cell->framebuffer.cbufs[i];
       if (ps) {
          struct cell_texture *ct = cell_texture(ps->texture);
+#if 0
          cell->cbuf_map[i] = screen->buffer_map(screen,
                                                 ct->buffer,
                                                 (PIPE_BUFFER_USAGE_GPU_READ |
                                                  PIPE_BUFFER_USAGE_GPU_WRITE));
+#else
+         cell->cbuf_map[i] = ct->data;
+#endif
       }
    }
 
@@ -341,10 +347,14 @@ cell_map_surfaces(struct cell_context *cell)
       struct pipe_surface *ps = cell->framebuffer.zsbuf;
       if (ps) {
          struct cell_texture *ct = cell_texture(ps->texture);
+#if 0
          cell->zsbuf_map = screen->buffer_map(screen,
                                               ct->buffer,
                                               (PIPE_BUFFER_USAGE_GPU_READ |
                                                PIPE_BUFFER_USAGE_GPU_WRITE));
+#else
+         cell->zsbuf_map = ct->data;
+#endif
       }
    }
 }
@@ -356,17 +366,17 @@ cell_map_surfaces(struct cell_context *cell)
 static void
 cell_unmap_surfaces(struct cell_context *cell)
 {
-   struct pipe_screen *screen = cell->pipe.screen;
+   /*struct pipe_screen *screen = cell->pipe.screen;*/
    uint i;
 
    for (i = 0; i < PIPE_MAX_COLOR_BUFS; i++) {
       struct pipe_surface *ps = cell->framebuffer.cbufs[i];
       if (ps && cell->cbuf_map[i]) {
-         struct cell_texture *ct = cell_texture(ps->texture);
+         /*struct cell_texture *ct = cell_texture(ps->texture);*/
          assert(ps->texture);
-         assert(ct->buffer);
+         /*assert(ct->buffer);*/
 
-         screen->buffer_unmap(screen, ct->buffer);
+         /*screen->buffer_unmap(screen, ct->buffer);*/
          cell->cbuf_map[i] = NULL;
       }
    }
@@ -374,8 +384,8 @@ cell_unmap_surfaces(struct cell_context *cell)
    {
       struct pipe_surface *ps = cell->framebuffer.zsbuf;
       if (ps && cell->zsbuf_map) {
-         struct cell_texture *ct = cell_texture(ps->texture);
-         screen->buffer_unmap(screen, ct->buffer);
+         /*struct cell_texture *ct = cell_texture(ps->texture);*/
+         /*screen->buffer_unmap(screen, ct->buffer);*/
          cell->zsbuf_map = NULL;
       }
    }
diff --git a/src/gallium/drivers/cell/ppu/cell_public.h b/src/gallium/drivers/cell/ppu/cell_public.h
new file mode 100644
index 00000000000..7e2e093565d
--- /dev/null
+++ b/src/gallium/drivers/cell/ppu/cell_public.h
@@ -0,0 +1,10 @@
+#ifndef CELL_PUBLIC_H
+#define CELL_PUBLIC_H
+
+struct pipe_screen;
+struct sw_winsys;
+
+struct pipe_screen *
+cell_create_screen(struct sw_winsys *winsys);
+
+#endif
diff --git a/src/gallium/drivers/cell/ppu/cell_screen.c b/src/gallium/drivers/cell/ppu/cell_screen.c
index 449855f539d..31fd963d19a 100644
--- a/src/gallium/drivers/cell/ppu/cell_screen.c
+++ b/src/gallium/drivers/cell/ppu/cell_screen.c
@@ -28,7 +28,6 @@
 
 #include "util/u_memory.h"
 #include "util/u_simple_screen.h"
-#include "util/u_simple_screen.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_screen.h"
 
@@ -36,7 +35,10 @@
 #include "cell_context.h"
 #include "cell_screen.h"
 #include "cell_texture.h"
-#include "cell_winsys.h"
+#include "cell_buffer.h"
+#include "cell_public.h"
+
+#include "state_tracker/sw_winsys.h"
 
 
 static const char *
@@ -134,19 +136,28 @@ cell_is_format_supported( struct pipe_screen *screen,
                           unsigned tex_usage, 
                           unsigned geom_flags )
 {
-   /* cell supports most formats, XXX for now anyway */
+   struct sw_winsys *winsys = cell_screen(screen)->winsys;
+
    if (format == PIPE_FORMAT_DXT5_RGBA ||
-       format == PIPE_FORMAT_R8G8B8A8_SRGB)
+       format == PIPE_FORMAT_A8B8G8R8_SRGB)
       return FALSE;
-   else
-      return TRUE;
+
+   if (tex_usage & PIPE_TEXTURE_USAGE_DISPLAY_TARGET) {
+      if (!winsys->is_displaytarget_format_supported(winsys, format))
+         return FALSE;
+   }
+
+   /* This is often a lie.  Pull in logic from llvmpipe to fix.
+    */
+   return TRUE;
 }
 
 
 static void
 cell_destroy_screen( struct pipe_screen *screen )
 {
-   struct pipe_winsys *winsys = screen->winsys;
+   struct cell_screen *sp_screen = cell_screen(screen);
+   struct sw_winsys *winsys = sp_screen->winsys;
 
    if(winsys->destroy)
       winsys->destroy(winsys);
@@ -155,32 +166,34 @@ cell_destroy_screen( struct pipe_screen *screen )
 }
 
 
+
 /**
  * Create a new pipe_screen object
  * Note: we're not presently subclassing pipe_screen (no cell_screen) but
  * that would be the place to put SPU thread/context info...
  */
 struct pipe_screen *
-cell_create_screen(struct pipe_winsys *winsys)
+cell_create_screen(struct sw_winsys *winsys)
 {
-   struct pipe_screen *screen = CALLOC_STRUCT(pipe_screen);
+   struct cell_screen *screen = CALLOC_STRUCT(cell_screen);
 
    if (!screen)
       return NULL;
 
    screen->winsys = winsys;
+   screen->base.winsys = NULL;
 
-   screen->destroy = cell_destroy_screen;
+   screen->base.destroy = cell_destroy_screen;
 
-   screen->get_name = cell_get_name;
-   screen->get_vendor = cell_get_vendor;
-   screen->get_param = cell_get_param;
-   screen->get_paramf = cell_get_paramf;
-   screen->is_format_supported = cell_is_format_supported;
-   screen->context_create = cell_create_context;
+   screen->base.get_name = cell_get_name;
+   screen->base.get_vendor = cell_get_vendor;
+   screen->base.get_param = cell_get_param;
+   screen->base.get_paramf = cell_get_paramf;
+   screen->base.is_format_supported = cell_is_format_supported;
+   screen->base.context_create = cell_create_context;
 
-   cell_init_screen_texture_funcs(screen);
-   u_simple_screen_init(screen);
+   cell_init_screen_texture_funcs(&screen->base);
+   cell_init_screen_buffer_funcs(&screen->base);
 
-   return screen;
+   return &screen->base;
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_screen.h b/src/gallium/drivers/cell/ppu/cell_screen.h
index c7e15889d66..baff9d3b7d4 100644
--- a/src/gallium/drivers/cell/ppu/cell_screen.h
+++ b/src/gallium/drivers/cell/ppu/cell_screen.h
@@ -30,12 +30,26 @@
 #define CELL_SCREEN_H
 
 
-struct pipe_screen;
-struct pipe_winsys;
+#include "pipe/p_screen.h"
 
+struct sw_winsys;
 
-extern struct pipe_screen *
-cell_create_screen(struct pipe_winsys *winsys);
+struct cell_screen {
+   struct pipe_screen base;
+
+   struct sw_winsys *winsys;
+
+   /* Increments whenever textures are modified.  Contexts can track
+    * this.
+    */
+   unsigned timestamp;          
+};
+
+static INLINE struct cell_screen *
+cell_screen( struct pipe_screen *pipe )
+{
+   return (struct cell_screen *)pipe;
+}
 
 
 #endif /* CELL_SCREEN_H */
diff --git a/src/gallium/drivers/cell/ppu/cell_state_emit.c b/src/gallium/drivers/cell/ppu/cell_state_emit.c
index a59c7828ac3..424e2628a95 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_emit.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_emit.c
@@ -245,16 +245,13 @@ cell_emit_state(struct cell_context *cell)
       uint i, j;
       float *buf = cell_batch_alloc16(cell, ROUNDUP16(32 + num_const * sizeof(float)));
       uint32_t *ibuf = (uint32_t *) buf;
-      const float *constants = pipe_buffer_map(cell->pipe.screen,
-                                               cell->constants[shader],
-                                               PIPE_BUFFER_USAGE_CPU_READ);
+      const float *constants = cell->mapped_constants[shader];
       ibuf[0] = CELL_CMD_STATE_FS_CONSTANTS;
       ibuf[4] = num_const;
       j = 8;
       for (i = 0; i < num_const; i++) {
          buf[j++] = constants[i];
       }
-      pipe_buffer_unmap(cell->pipe.screen, cell->constants[shader]);
    }
 
    if (cell->dirty & (CELL_NEW_FRAMEBUFFER |
diff --git a/src/gallium/drivers/cell/ppu/cell_state_per_fragment.c b/src/gallium/drivers/cell/ppu/cell_state_per_fragment.c
index 07be5e92ea7..dc33e7ccc2c 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_per_fragment.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_per_fragment.c
@@ -1251,7 +1251,7 @@ cell_generate_logic_op(struct spe_function *f,
    /* Convert fragment colors to framebuffer format in AoS layout.
     */
    switch (surf->format) {
-   case PIPE_FORMAT_A8R8G8B8_UNORM:
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
       data[0] = 0x00010203;
       data[1] = 0x10111213;
       data[2] = 0x04050607;
@@ -1261,7 +1261,7 @@ cell_generate_logic_op(struct spe_function *f,
       data[6] = 0x80808080;
       data[7] = 0x80808080;
       break;
-   case PIPE_FORMAT_B8G8R8A8_UNORM:
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
       data[0] = 0x03020100;
       data[1] = 0x13121110;
       data[2] = 0x07060504;
diff --git a/src/gallium/drivers/cell/ppu/cell_state_shader.c b/src/gallium/drivers/cell/ppu/cell_state_shader.c
index 9b2f86fdfba..9e29ddc2d45 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_shader.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_shader.c
@@ -28,13 +28,13 @@
 #include "pipe/p_defines.h"
 #include "util/u_memory.h"
 #include "util/u_inlines.h"
-#include "util/u_simple_screen.h"
 #include "draw/draw_context.h"
 #include "tgsi/tgsi_parse.h"
 
 #include "cell_context.h"
 #include "cell_state.h"
 #include "cell_gen_fp.h"
+#include "cell_buffer.h"
 
 
 /** cast wrapper */
@@ -183,17 +183,29 @@ cell_delete_vs_state(struct pipe_context *pipe, void *vs)
 static void
 cell_set_constant_buffer(struct pipe_context *pipe,
                          uint shader, uint index,
-                         struct pipe_buffer *buf)
+                         struct pipe_buffer *constants)
 {
    struct cell_context *cell = cell_context(pipe);
+   unsigned size = constants ? constants->size : 0;
+   const void *data = constants ? cell_buffer(constants)->data : NULL;
 
    assert(shader < PIPE_SHADER_TYPES);
    assert(index == 0);
 
+   if (cell->constants[shader] == constants)
+      return;
+
    draw_flush(cell->draw);
 
    /* note: reference counting */
-   pipe_buffer_reference(&cell->constants[shader], buf);
+   pipe_buffer_reference(&cell->constants[shader], constants);
+
+   if(shader == PIPE_SHADER_VERTEX) {
+      draw_set_mapped_constant_buffer(cell->draw, PIPE_SHADER_VERTEX, 0,
+                                      data, size);
+   }
+
+   cell->mapped_constants[shader] = data;
 
    if (shader == PIPE_SHADER_VERTEX)
       cell->dirty |= CELL_NEW_VS_CONSTANTS;
diff --git a/src/gallium/drivers/cell/ppu/cell_state_vertex.c b/src/gallium/drivers/cell/ppu/cell_state_vertex.c
index fbe55c84721..d3efb8ecea2 100644
--- a/src/gallium/drivers/cell/ppu/cell_state_vertex.c
+++ b/src/gallium/drivers/cell/ppu/cell_state_vertex.c
@@ -32,24 +32,43 @@
 #include "cell_context.h"
 #include "cell_state.h"
 
+#include "util/u_memory.h"
 #include "draw/draw_context.h"
 
 
-static void
-cell_set_vertex_elements(struct pipe_context *pipe,
-                         unsigned count,
-                         const struct pipe_vertex_element *elements)
+void *
+cell_create_vertex_elements_state(struct pipe_context *pipe,
+                                  unsigned count,
+                                  const struct pipe_vertex_element *attribs)
 {
-   struct cell_context *cell = cell_context(pipe);
-
+   struct cell_velems_state *velems;
    assert(count <= PIPE_MAX_ATTRIBS);
+   velems = (struct cell_velems_state *) MALLOC(sizeof(struct cell_velems_state));
+   if (velems) {
+      velems->count = count;
+      memcpy(velems->velem, attribs, sizeof(*attribs) * count);
+   }
+   return velems;
+}
+
+void
+cell_bind_vertex_elements_state(struct pipe_context *pipe,
+                                void *velems)
+{
+   struct cell_context *cell = cell_context(pipe);
+   struct cell_velems_state *cell_velems = (struct cell_velems_state *) velems;
 
-   memcpy(cell->vertex_element, elements, count * sizeof(elements[0]));
-   cell->num_vertex_elements = count;
+   cell->velems = cell_velems;
 
    cell->dirty |= CELL_NEW_VERTEX;
 
-   draw_set_vertex_elements(cell->draw, count, elements);
+   draw_set_vertex_elements(cell->draw, cell_velems->count, cell_velems->velem);
+}
+
+void
+cell_delete_vertex_elements_state(struct pipe_context *pipe, void *velems)
+{
+   FREE( velems );
 }
 
 
@@ -75,5 +94,7 @@ void
 cell_init_vertex_functions(struct cell_context *cell)
 {
    cell->pipe.set_vertex_buffers = cell_set_vertex_buffers;
-   cell->pipe.set_vertex_elements = cell_set_vertex_elements;
+   cell->pipe.create_vertex_elements_state = cell_create_vertex_elements_state;
+   cell->pipe.bind_vertex_elements_state = cell_bind_vertex_elements_state;
+   cell->pipe.delete_vertex_elements_state = cell_delete_vertex_elements_state;
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.c b/src/gallium/drivers/cell/ppu/cell_texture.c
index fad290dfa0e..c8a1acd86a9 100644
--- a/src/gallium/drivers/cell/ppu/cell_texture.c
+++ b/src/gallium/drivers/cell/ppu/cell_texture.c
@@ -34,20 +34,23 @@
 #include "pipe/p_context.h"
 #include "pipe/p_defines.h"
 #include "util/u_inlines.h"
-#include "util/u_simple_screen.h"
 
 #include "util/u_format.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
 
 #include "cell_context.h"
+#include "cell_screen.h"
 #include "cell_state.h"
 #include "cell_texture.h"
 
+#include "state_tracker/sw_winsys.h"
 
 
-static void
-cell_texture_layout(struct cell_texture *ct)
+
+static boolean
+cell_texture_layout(struct pipe_screen *screen, 
+                    struct cell_texture *ct)
 {
    struct pipe_texture *pt = &ct->base;
    unsigned level;
@@ -83,9 +86,34 @@ cell_texture_layout(struct cell_texture *ct)
       height = u_minify(height, 1);
       depth = u_minify(depth, 1);
    }
+
+   ct->data = align_malloc(ct->buffer_size, 16);
+ 
+   return ct->data != NULL;
 }
 
 
+/**
+ * Texture layout for simple color buffers.
+ */
+static boolean
+cell_displaytarget_layout(struct pipe_screen *screen,
+                          struct cell_texture * ct)
+{
+   struct sw_winsys *winsys = cell_screen(screen)->winsys;
+
+   /* Round up the surface size to a multiple of the tile size?
+    */
+   ct->dt = winsys->displaytarget_create(winsys,
+                                          ct->base.format,
+                                          ct->base.width0, 
+                                          ct->base.height0,
+                                          16,
+                                          &ct->dt_stride );
+
+   return ct->dt != NULL;
+}
+
 static struct pipe_texture *
 cell_texture_create(struct pipe_screen *screen,
                     const struct pipe_texture *templat)
@@ -98,31 +126,46 @@ cell_texture_create(struct pipe_screen *screen,
    pipe_reference_init(&ct->base.reference, 1);
    ct->base.screen = screen;
 
-   cell_texture_layout(ct);
+   /* Create both a displaytarget (linear) and regular texture
+    * (twiddled).  Convert twiddled->linear at flush_frontbuffer time.
+    */
+   if (ct->base.tex_usage & (PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
+                             PIPE_TEXTURE_USAGE_SCANOUT |
+                             PIPE_TEXTURE_USAGE_SHARED)) {
+      if (!cell_displaytarget_layout(screen, ct))
+         goto fail;
+   }
 
-   ct->buffer = screen->buffer_create(screen, 32, PIPE_BUFFER_USAGE_PIXEL,
-                                   ct->buffer_size);
+   if (!cell_texture_layout(screen, ct))
+      goto fail;
 
-   if (!ct->buffer) {
-      FREE(ct);
-      return NULL;
+   return &ct->base;
+
+fail:
+   if (ct->dt) {
+      struct sw_winsys *winsys = cell_screen(screen)->winsys;
+      winsys->displaytarget_destroy(winsys, ct->dt);
    }
 
-   return &ct->base;
+   FREE(ct);
+
+   return NULL;
 }
 
 
 static void
 cell_texture_destroy(struct pipe_texture *pt)
 {
+   struct cell_screen *screen = cell_screen(pt->screen);
+   struct sw_winsys *winsys = screen->winsys;
    struct cell_texture *ct = cell_texture(pt);
 
-   if (ct->mapped) {
-      pipe_buffer_unmap(ct->buffer->screen, ct->buffer);
-      ct->mapped = NULL;
+   if (ct->dt) {
+      /* display target */
+      winsys->displaytarget_destroy(winsys, ct->dt);
    }
 
-   pipe_buffer_reference(&ct->buffer, NULL);
+   align_free(ct->data);
 
    FREE(ct);
 }
@@ -389,10 +432,8 @@ cell_transfer_map(struct pipe_screen *screen, struct pipe_transfer *transfer)
 
    assert(transfer->texture);
 
-   if (!ct->mapped) {
-      /* map now */
-      ct->mapped = pipe_buffer_map(screen, ct->buffer,
-                                   pipe_transfer_buffer_flags(transfer));
+   if (ct->mapped == NULL) {
+      ct->mapped = ct->data;
    }
 
    /*
@@ -442,9 +483,8 @@ cell_transfer_unmap(struct pipe_screen *screen,
    const uint stride = ct->stride[level];
 
    if (!ct->mapped) {
-      /* map now */
-      ct->mapped = pipe_buffer_map(screen, ct->buffer,
-                                   PIPE_BUFFER_USAGE_CPU_READ);
+      assert(0);
+      return;
    }
 
    if (transfer->usage & PIPE_TRANSFER_WRITE) {
@@ -467,6 +507,50 @@ cell_transfer_unmap(struct pipe_screen *screen,
 }
 
 
+
+/* This used to be overriden by the co-state tracker, but really needs
+ * to be active with sw_winsys.
+ *
+ * Contrasting with llvmpipe and softpipe, this is the only place
+ * where we use the ct->dt display target in any real sense.
+ *
+ * Basically just untwiddle our local data into the linear
+ * displaytarget.
+ */
+static void
+cell_flush_frontbuffer(struct pipe_screen *_screen,
+                       struct pipe_surface *surface,
+                       void *context_private)
+{
+   struct cell_screen *screen = cell_screen(_screen);
+   struct sw_winsys *winsys = screen->winsys;
+   struct cell_texture *ct = cell_texture(surface->texture);
+
+   if (!ct->dt)
+      return;
+
+   /* Need to untwiddle from our internal representation here:
+    */
+   {
+      unsigned *map = winsys->displaytarget_map(winsys, ct->dt,
+                                                (PIPE_BUFFER_USAGE_CPU_READ |
+                                                 PIPE_BUFFER_USAGE_CPU_WRITE));
+      unsigned *src = (unsigned *)(ct->data + ct->level_offset[surface->level]);
+
+      untwiddle_image_uint(surface->width,
+                           surface->height,
+                           TILE_SIZE,
+                           map,
+                           ct->dt_stride,
+                           src);
+
+      winsys->displaytarget_unmap(winsys, ct->dt);
+   }
+
+   winsys->displaytarget_display(winsys, ct->dt, context_private);
+}
+
+
 void
 cell_init_screen_texture_funcs(struct pipe_screen *screen)
 {
@@ -481,4 +565,6 @@ cell_init_screen_texture_funcs(struct pipe_screen *screen)
 
    screen->transfer_map = cell_transfer_map;
    screen->transfer_unmap = cell_transfer_unmap;
+
+   screen->flush_frontbuffer = cell_flush_frontbuffer;
 }
diff --git a/src/gallium/drivers/cell/ppu/cell_texture.h b/src/gallium/drivers/cell/ppu/cell_texture.h
index 3ffc0bfdb51..2be05793122 100644
--- a/src/gallium/drivers/cell/ppu/cell_texture.h
+++ b/src/gallium/drivers/cell/ppu/cell_texture.h
@@ -28,6 +28,7 @@
 #ifndef CELL_TEXTURE_H
 #define CELL_TEXTURE_H
 
+#include "cell/common.h"
 
 struct cell_context;
 struct pipe_texture;
@@ -43,8 +44,20 @@ struct cell_texture
    unsigned long level_offset[CELL_MAX_TEXTURE_LEVELS];
    unsigned long stride[CELL_MAX_TEXTURE_LEVELS];
 
-   /** The tiled texture data is held in this buffer */
-   struct pipe_buffer *buffer;
+   /**
+    * Display target, for textures with the PIPE_TEXTURE_USAGE_DISPLAY_TARGET
+    * usage.
+    */
+   struct sw_displaytarget *dt;
+   unsigned dt_stride;
+
+   /**
+    * Malloc'ed data for regular textures, or a mapping to dt above.
+    */
+   void *data;
+
+   /* Size of the linear buffer??
+    */
    unsigned long buffer_size;
 
    /** The buffer above, mapped.  This is the memory from which the
diff --git a/src/gallium/drivers/cell/ppu/cell_vertex_shader.c b/src/gallium/drivers/cell/ppu/cell_vertex_shader.c
index cf8cd411598..3d389d6ea36 100644
--- a/src/gallium/drivers/cell/ppu/cell_vertex_shader.c
+++ b/src/gallium/drivers/cell/ppu/cell_vertex_shader.c
@@ -31,7 +31,6 @@
 
 #include "pipe/p_defines.h"
 #include "pipe/p_context.h"
-#include "util/u_simple_screen.h"
 #include "util/u_math.h"
 
 #include "cell_context.h"
diff --git a/src/gallium/drivers/cell/spu/spu_command.c b/src/gallium/drivers/cell/spu/spu_command.c
index 55bd85bde2b..79f1fb7fb28 100644
--- a/src/gallium/drivers/cell/spu/spu_command.c
+++ b/src/gallium/drivers/cell/spu/spu_command.c
@@ -337,10 +337,10 @@ cmd_state_framebuffer(const struct cell_command_framebuffer *cmd)
       spu.fb.zsize = 4;
       spu.fb.zscale = (float) 0xffffffffu;
       break;
-   case PIPE_FORMAT_Z24S8_UNORM:
    case PIPE_FORMAT_S8Z24_UNORM:
-   case PIPE_FORMAT_Z24X8_UNORM:
+   case PIPE_FORMAT_Z24S8_UNORM:
    case PIPE_FORMAT_X8Z24_UNORM:
+   case PIPE_FORMAT_Z24X8_UNORM:
       spu.fb.zsize = 4;
       spu.fb.zscale = (float) 0x00ffffffu;
       break;
diff --git a/src/gallium/drivers/cell/spu/spu_per_fragment_op.c b/src/gallium/drivers/cell/spu/spu_per_fragment_op.c
index 53283740805..2c9e7458afe 100644
--- a/src/gallium/drivers/cell/spu/spu_per_fragment_op.c
+++ b/src/gallium/drivers/cell/spu/spu_per_fragment_op.c
@@ -138,14 +138,14 @@ spu_fallback_fragment_ops(uint x, uint y,
 
       if (spu.depth_stencil_alpha.stencil[0].enabled) {
          /* do stencil test */
-         ASSERT(spu.fb.depth_format == PIPE_FORMAT_S8Z24_UNORM);
+         ASSERT(spu.fb.depth_format == PIPE_FORMAT_Z24S8_UNORM);
 
       }
       else if (spu.depth_stencil_alpha.depth.enabled) {
          /* do depth test */
 
-         ASSERT(spu.fb.depth_format == PIPE_FORMAT_S8Z24_UNORM ||
-                spu.fb.depth_format == PIPE_FORMAT_X8Z24_UNORM);
+         ASSERT(spu.fb.depth_format == PIPE_FORMAT_Z24S8_UNORM ||
+                spu.fb.depth_format == PIPE_FORMAT_Z24X8_UNORM);
 
          vector unsigned int ifragZ;
          vector unsigned int zmask;
@@ -240,13 +240,13 @@ spu_fallback_fragment_ops(uint x, uint y,
       {
          vector float temp[4]; /* float colors in AOS form */
          switch (spu.fb.color_format) {
-         case PIPE_FORMAT_B8G8R8A8_UNORM:
+         case PIPE_FORMAT_A8R8G8B8_UNORM:
             temp[0] = spu_unpack_B8G8R8A8(fbc0);
             temp[1] = spu_unpack_B8G8R8A8(fbc1);
             temp[2] = spu_unpack_B8G8R8A8(fbc2);
             temp[3] = spu_unpack_B8G8R8A8(fbc3);
             break;
-         case PIPE_FORMAT_A8R8G8B8_UNORM:
+         case PIPE_FORMAT_B8G8R8A8_UNORM:
             temp[0] = spu_unpack_A8R8G8B8(fbc0);
             temp[1] = spu_unpack_A8R8G8B8(fbc1);
             temp[2] = spu_unpack_A8R8G8B8(fbc2);
@@ -506,13 +506,13 @@ spu_fallback_fragment_ops(uint x, uint y,
     * Pack fragment float colors into 32-bit RGBA words.
     */
    switch (spu.fb.color_format) {
-   case PIPE_FORMAT_A8R8G8B8_UNORM:
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
       fragc0 = spu_pack_A8R8G8B8(frag_aos[0]);
       fragc1 = spu_pack_A8R8G8B8(frag_aos[1]);
       fragc2 = spu_pack_A8R8G8B8(frag_aos[2]);
       fragc3 = spu_pack_A8R8G8B8(frag_aos[3]);
       break;
-   case PIPE_FORMAT_B8G8R8A8_UNORM:
+   case PIPE_FORMAT_A8R8G8B8_UNORM:
       fragc0 = spu_pack_B8G8R8A8(frag_aos[0]);
       fragc1 = spu_pack_B8G8R8A8(frag_aos[1]);
       fragc2 = spu_pack_B8G8R8A8(frag_aos[2]);
@@ -532,7 +532,7 @@ spu_fallback_fragment_ops(uint x, uint y,
 
       /* Form bitmask depending on color buffer format and colormask bits */
       switch (spu.fb.color_format) {
-      case PIPE_FORMAT_A8R8G8B8_UNORM:
+      case PIPE_FORMAT_B8G8R8A8_UNORM:
          if (spu.blend.rt[0].colormask & PIPE_MASK_R)
             cmask |= 0x00ff0000; /* red */
          if (spu.blend.rt[0].colormask & PIPE_MASK_G)
@@ -542,7 +542,7 @@ spu_fallback_fragment_ops(uint x, uint y,
          if (spu.blend.rt[0].colormask & PIPE_MASK_A)
             cmask |= 0xff000000; /* alpha */
          break;
-      case PIPE_FORMAT_B8G8R8A8_UNORM:
+      case PIPE_FORMAT_A8R8G8B8_UNORM:
          if (spu.blend.rt[0].colormask & PIPE_MASK_R)
             cmask |= 0x0000ff00; /* red */
          if (spu.blend.rt[0].colormask & PIPE_MASK_G)
diff --git a/src/gallium/drivers/cell/spu/spu_tri.c b/src/gallium/drivers/cell/spu/spu_tri.c
index 58be001be4c..f619380d807 100644
--- a/src/gallium/drivers/cell/spu/spu_tri.c
+++ b/src/gallium/drivers/cell/spu/spu_tri.c
@@ -752,7 +752,7 @@ subtriangle(struct edge *eleft, struct edge *eright, unsigned lines)
    finish_y -= sy;
 
    /*
-   _mesa_printf("%s %d %d\n", __FUNCTION__, start_y, finish_y);  
+   printf("%s %d %d\n", __FUNCTION__, start_y, finish_y);  
    */
 
    for (y = start_y; y < finish_y; y++) {
diff --git a/src/gallium/drivers/failover/fo_context.c b/src/gallium/drivers/failover/fo_context.c
index 2ccc5d3e605..659e40cbf03 100644
--- a/src/gallium/drivers/failover/fo_context.c
+++ b/src/gallium/drivers/failover/fo_context.c
@@ -27,7 +27,6 @@
 
 
 #include "pipe/p_defines.h"
-#include "util/u_simple_screen.h"
 #include "util/u_memory.h"
 #include "pipe/p_context.h"
 
diff --git a/src/gallium/drivers/failover/fo_context.h b/src/gallium/drivers/failover/fo_context.h
index 53e1a02bcba..73031321d98 100644
--- a/src/gallium/drivers/failover/fo_context.h
+++ b/src/gallium/drivers/failover/fo_context.h
@@ -85,6 +85,7 @@ struct failover_context {
    const struct fo_state     *rasterizer;
    const struct fo_state     *fragment_shader;
    const struct fo_state     *vertex_shader;
+   const struct fo_state     *vertex_elements;
 
    struct pipe_blend_color blend_color;
    struct pipe_stencil_ref stencil_ref;
@@ -94,10 +95,8 @@ struct failover_context {
    struct pipe_scissor_state scissor;
    struct pipe_viewport_state viewport;
    struct pipe_vertex_buffer vertex_buffers[PIPE_MAX_ATTRIBS];
-   struct pipe_vertex_element vertex_elements[PIPE_MAX_ATTRIBS];
 
    uint num_vertex_buffers;
-   uint num_vertex_elements;
 
    void *sw_sampler_state[PIPE_MAX_SAMPLERS];
    void *hw_sampler_state[PIPE_MAX_SAMPLERS];
diff --git a/src/gallium/drivers/failover/fo_state.c b/src/gallium/drivers/failover/fo_state.c
index d0c2f1474ae..25c62735705 100644
--- a/src/gallium/drivers/failover/fo_state.c
+++ b/src/gallium/drivers/failover/fo_state.c
@@ -255,9 +255,52 @@ failover_delete_vs_state(struct pipe_context *pipe,
    free(state);
 }
 
+
+
+static void *
+failover_create_vertex_elements_state( struct pipe_context *pipe,
+                                       unsigned count,
+                                       const struct pipe_vertex_element *velems )
+{
+   struct fo_state *state = malloc(sizeof(struct fo_state));
+   struct failover_context *failover = failover_context(pipe);
+
+   state->sw_state = failover->sw->create_vertex_elements_state(failover->sw, count, velems);
+   state->hw_state = failover->hw->create_vertex_elements_state(failover->hw, count, velems);
+
+   return state;
+}
+
+static void
+failover_bind_vertex_elements_state(struct pipe_context *pipe,
+                                    void *velems )
+{
+   struct failover_context *failover = failover_context(pipe);
+   struct fo_state *state = (struct fo_state*)velems;
+
+   failover->vertex_elements = state;
+   failover->dirty |= FO_NEW_VERTEX_ELEMENT;
+   failover->sw->bind_vertex_elements_state( failover->sw, velems );
+   failover->hw->bind_vertex_elements_state( failover->hw, velems );
+}
+
+static void
+failover_delete_vertex_elements_state( struct pipe_context *pipe,
+                                       void *velems )
+{
+   struct fo_state *state = (struct fo_state*)velems;
+   struct failover_context *failover = failover_context(pipe);
+
+   failover->sw->delete_vertex_elements_state(failover->sw, state->sw_state);
+   failover->hw->delete_vertex_elements_state(failover->hw, state->hw_state);
+   state->sw_state = 0;
+   state->hw_state = 0;
+   free(state);
+}
+
 static void 
 failover_set_polygon_stipple( struct pipe_context *pipe,
-			      const struct pipe_poly_stipple *stipple )
+                              const struct pipe_poly_stipple *stipple )
 {
    struct failover_context *failover = failover_context(pipe);
 
@@ -526,22 +569,6 @@ failover_set_vertex_buffers(struct pipe_context *pipe,
 }
 
 
-static void
-failover_set_vertex_elements(struct pipe_context *pipe,
-                             unsigned count,
-                             const struct pipe_vertex_element *vertex_elements)
-{
-   struct failover_context *failover = failover_context(pipe);
-
-   memcpy(failover->vertex_elements, vertex_elements,
-          count * sizeof(vertex_elements[0]));
-
-   failover->dirty |= FO_NEW_VERTEX_ELEMENT;
-   failover->num_vertex_elements = count;
-   failover->sw->set_vertex_elements( failover->sw, count, vertex_elements );
-   failover->hw->set_vertex_elements( failover->hw, count, vertex_elements );
-}
-
 void
 failover_set_constant_buffer(struct pipe_context *pipe,
                              uint shader, uint index,
@@ -579,6 +606,9 @@ failover_init_state_functions( struct failover_context *failover )
    failover->pipe.create_vs_state = failover_create_vs_state;
    failover->pipe.bind_vs_state   = failover_bind_vs_state;
    failover->pipe.delete_vs_state = failover_delete_vs_state;
+   failover->pipe.create_vertex_elements_state = failover_create_vertex_elements_state;
+   failover->pipe.bind_vertex_elements_state = failover_bind_vertex_elements_state;
+   failover->pipe.delete_vertex_elements_state = failover_delete_vertex_elements_state;
 
    failover->pipe.set_blend_color = failover_set_blend_color;
    failover->pipe.set_stencil_ref = failover_set_stencil_ref;
@@ -590,7 +620,6 @@ failover_init_state_functions( struct failover_context *failover )
    failover->pipe.set_vertex_sampler_views = failover_set_vertex_sampler_views;
    failover->pipe.set_viewport_state = failover_set_viewport_state;
    failover->pipe.set_vertex_buffers = failover_set_vertex_buffers;
-   failover->pipe.set_vertex_elements = failover_set_vertex_elements;
    failover->pipe.set_constant_buffer = failover_set_constant_buffer;
    failover->pipe.create_sampler_view = failover_create_sampler_view;
    failover->pipe.sampler_view_destroy = failover_sampler_view_destroy;
diff --git a/src/gallium/drivers/failover/fo_state_emit.c b/src/gallium/drivers/failover/fo_state_emit.c
index 171151e1ee2..42bd6929a7f 100644
--- a/src/gallium/drivers/failover/fo_state_emit.c
+++ b/src/gallium/drivers/failover/fo_state_emit.c
@@ -81,6 +81,10 @@ failover_state_emit( struct failover_context *failover )
       failover->sw->bind_vs_state( failover->sw,
                                    failover->vertex_shader->sw_state );
 
+   if (failover->dirty & FO_NEW_VERTEX_ELEMENT)
+      failover->sw->bind_vertex_elements_state( failover->sw,
+                                                failover->vertex_elements->sw_state );
+
    if (failover->dirty & FO_NEW_STIPPLE)
       failover->sw->set_polygon_stipple( failover->sw, &failover->poly_stipple );
 
@@ -128,11 +132,5 @@ failover_state_emit( struct failover_context *failover )
                                         failover->vertex_buffers );
    }
 
-   if (failover->dirty & FO_NEW_VERTEX_ELEMENT) {
-      failover->sw->set_vertex_elements( failover->sw,
-                                         failover->num_vertex_elements,
-                                         failover->vertex_elements );
-   }
-
    failover->dirty = 0;
 }
diff --git a/src/gallium/drivers/i915/i915_context.h b/src/gallium/drivers/i915/i915_context.h
index cd6dcd59a9f..9d03ca37fda 100644
--- a/src/gallium/drivers/i915/i915_context.h
+++ b/src/gallium/drivers/i915/i915_context.h
@@ -148,7 +148,7 @@ struct i915_state
 
    /** Describes the current hardware vertex layout */
    struct vertex_info vertex_info;
-   
+
    unsigned id;			/* track lost context events */
 };
 
@@ -187,6 +187,14 @@ struct i915_sampler_state {
    unsigned maxlod;
 };
 
+struct i915_velems_state {
+   unsigned count;
+   struct pipe_vertex_element velem[PIPE_MAX_ATTRIBS];
+};
+
+#define I915_MAX_TEXTURE_2D_LEVELS 11  /* max 1024x1024 */
+#define I915_MAX_TEXTURE_3D_LEVELS  8  /* max 128x128x128 */
+
 struct i915_texture {
    struct pipe_texture base;
 
@@ -199,7 +207,7 @@ struct i915_texture {
    unsigned sw_tiled; /**< tiled with software flags */
    unsigned hw_tiled; /**< tiled with hardware fences */
 
-   unsigned nr_images[PIPE_MAX_TEXTURE_LEVELS];
+   unsigned nr_images[I915_MAX_TEXTURE_2D_LEVELS];
 
    /* Explicitly store the offset of each image for each cube face or
     * depth value.  Pretty much have to accept that hardware formats
@@ -207,7 +215,7 @@ struct i915_texture {
     * compute the offsets of depth/cube images within a mipmap level,
     * so have to store them as a lookup table:
     */
-   unsigned *image_offset[PIPE_MAX_TEXTURE_LEVELS];   /**< array [depth] of offsets */
+   unsigned *image_offset[I915_MAX_TEXTURE_2D_LEVELS];   /**< array [depth] of offsets */
 
    /* The data is held here:
     */
@@ -247,7 +255,6 @@ struct i915_context
 
    unsigned num_samplers;
    unsigned num_fragment_sampler_views;
-   unsigned num_vertex_elements;
    unsigned num_vertex_buffers;
 
    struct intel_batchbuffer *batch;
diff --git a/src/gallium/drivers/i915/i915_debug_fp.c b/src/gallium/drivers/i915/i915_debug_fp.c
index 066e7392d18..f41c51f2991 100644
--- a/src/gallium/drivers/i915/i915_debug_fp.c
+++ b/src/gallium/drivers/i915/i915_debug_fp.c
@@ -28,7 +28,6 @@
 
 #include "i915_reg.h"
 #include "i915_debug.h"
-#include "util/u_simple_screen.h"
 #include "util/u_debug.h"
 
 
diff --git a/src/gallium/drivers/i915/i915_screen.c b/src/gallium/drivers/i915/i915_screen.c
index c450854c982..e5bf4a20bd0 100644
--- a/src/gallium/drivers/i915/i915_screen.c
+++ b/src/gallium/drivers/i915/i915_screen.c
@@ -95,6 +95,10 @@ i915_get_param(struct pipe_screen *screen, int param)
    switch (param) {
    case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS:
       return 8;
+   case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS:
+      return 0;
+   case PIPE_CAP_MAX_COMBINED_SAMPLERS:
+      return 8;
    case PIPE_CAP_NPOT_TEXTURES:
       return 1;
    case PIPE_CAP_TWO_SIDED_STENCIL:
@@ -112,11 +116,11 @@ i915_get_param(struct pipe_screen *screen, int param)
    case PIPE_CAP_TEXTURE_SHADOW_MAP:
       return 1;
    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
-      return 11; /* max 1024x1024 */
+      return I915_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
-      return 8;  /* max 128x128x128 */
+      return I915_MAX_TEXTURE_3D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
-      return 11; /* max 1024x1024 */
+      return I915_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
       return 1;
@@ -161,22 +165,22 @@ i915_is_format_supported(struct pipe_screen *screen,
                          unsigned geom_flags)
 {
    static const enum pipe_format tex_supported[] = {
-      PIPE_FORMAT_R8G8B8A8_UNORM,
-      PIPE_FORMAT_A8R8G8B8_UNORM,
-      PIPE_FORMAT_R5G6B5_UNORM,
+      PIPE_FORMAT_A8B8G8R8_UNORM,
+      PIPE_FORMAT_B8G8R8A8_UNORM,
+      PIPE_FORMAT_B5G6R5_UNORM,
       PIPE_FORMAT_L8_UNORM,
       PIPE_FORMAT_A8_UNORM,
       PIPE_FORMAT_I8_UNORM,
-      PIPE_FORMAT_A8L8_UNORM,
-      PIPE_FORMAT_YCBCR,
-      PIPE_FORMAT_YCBCR_REV,
-      PIPE_FORMAT_S8Z24_UNORM,
+      PIPE_FORMAT_L8A8_UNORM,
+      PIPE_FORMAT_UYVY,
+      PIPE_FORMAT_YUYV,
+      PIPE_FORMAT_Z24S8_UNORM,
       PIPE_FORMAT_NONE  /* list terminator */
    };
    static const enum pipe_format surface_supported[] = {
-      PIPE_FORMAT_A8R8G8B8_UNORM,
-      PIPE_FORMAT_R5G6B5_UNORM,
-      PIPE_FORMAT_S8Z24_UNORM,
+      PIPE_FORMAT_B8G8R8A8_UNORM,
+      PIPE_FORMAT_B5G6R5_UNORM,
+      PIPE_FORMAT_Z24S8_UNORM,
       PIPE_FORMAT_NONE  /* list terminator */
    };
    const enum pipe_format *list;
diff --git a/src/gallium/drivers/i915/i915_state.c b/src/gallium/drivers/i915/i915_state.c
index 46703dce350..884abe622b3 100644
--- a/src/gallium/drivers/i915/i915_state.c
+++ b/src/gallium/drivers/i915/i915_state.c
@@ -767,21 +767,42 @@ static void i915_set_vertex_buffers(struct pipe_context *pipe,
    draw_set_vertex_buffers(i915->draw, count, buffers);
 }
 
-static void i915_set_vertex_elements(struct pipe_context *pipe,
-                                     unsigned count,
-                                     const struct pipe_vertex_element *elements)
+static void *
+i915_create_vertex_elements_state(struct pipe_context *pipe,
+                                  unsigned count,
+                                  const struct pipe_vertex_element *attribs)
+{
+   struct i915_velems_state *velems;
+   assert(count <= PIPE_MAX_ATTRIBS);
+   velems = (struct i915_velems_state *) MALLOC(sizeof(struct i915_velems_state));
+   if (velems) {
+      velems->count = count;
+      memcpy(velems->velem, attribs, sizeof(*attribs) * count);
+   }
+   return velems;
+}
+
+static void
+i915_bind_vertex_elements_state(struct pipe_context *pipe,
+                                void *velems)
 {
    struct i915_context *i915 = i915_context(pipe);
+   struct i915_velems_state *i915_velems = (struct i915_velems_state *) velems;
+
    /* Because we change state before the draw_set_vertex_buffers call
     * we need a flush here, just to be sure.
     */
    draw_flush(i915->draw);
 
-   i915->num_vertex_elements = count;
    /* pass-through to draw module */
-   draw_set_vertex_elements(i915->draw, count, elements);
+   draw_set_vertex_elements(i915->draw, i915_velems->count, i915_velems->velem);
 }
 
+static void
+i915_delete_vertex_elements_state(struct pipe_context *pipe, void *velems)
+{
+   FREE( velems );
+}
 
 void
 i915_init_state_functions( struct i915_context *i915 )
@@ -807,6 +828,9 @@ i915_init_state_functions( struct i915_context *i915 )
    i915->base.create_vs_state = i915_create_vs_state;
    i915->base.bind_vs_state = i915_bind_vs_state;
    i915->base.delete_vs_state = i915_delete_vs_state;
+   i915->base.create_vertex_elements_state = i915_create_vertex_elements_state;
+   i915->base.bind_vertex_elements_state = i915_bind_vertex_elements_state;
+   i915->base.delete_vertex_elements_state = i915_delete_vertex_elements_state;
 
    i915->base.set_blend_color = i915_set_blend_color;
    i915->base.set_stencil_ref = i915_set_stencil_ref;
@@ -821,5 +845,4 @@ i915_init_state_functions( struct i915_context *i915 )
    i915->base.sampler_view_destroy = i915_sampler_view_destroy;
    i915->base.set_viewport_state = i915_set_viewport_state;
    i915->base.set_vertex_buffers = i915_set_vertex_buffers;
-   i915->base.set_vertex_elements = i915_set_vertex_elements;
 }
diff --git a/src/gallium/drivers/i915/i915_state_emit.c b/src/gallium/drivers/i915/i915_state_emit.c
index 202607a6563..d79c1ca0b2c 100644
--- a/src/gallium/drivers/i915/i915_state_emit.c
+++ b/src/gallium/drivers/i915/i915_state_emit.c
@@ -37,9 +37,9 @@
 static unsigned translate_format( enum pipe_format format )
 {
    switch (format) {
-   case PIPE_FORMAT_A8R8G8B8_UNORM:
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
       return COLOR_BUF_ARGB8888;
-   case PIPE_FORMAT_R5G6B5_UNORM:
+   case PIPE_FORMAT_B5G6R5_UNORM:
       return COLOR_BUF_RGB565;
    default:
       assert(0);
@@ -50,7 +50,7 @@ static unsigned translate_format( enum pipe_format format )
 static unsigned translate_depth_format( enum pipe_format zformat )
 {
    switch (zformat) {
-   case PIPE_FORMAT_S8Z24_UNORM:
+   case PIPE_FORMAT_Z24S8_UNORM:
       return DEPTH_FRMT_24_FIXED_8_OTHER;
    case PIPE_FORMAT_Z16_UNORM:
       return DEPTH_FRMT_16_FIXED;
@@ -244,6 +244,7 @@ i915_emit_hardware_state(struct i915_context *i915 )
 
          OUT_BATCH(_3DSTATE_BUF_INFO_CMD);
 
+         assert(tex);
          OUT_BATCH(BUF_3D_ID_DEPTH |
                    BUF_3D_PITCH(tex->stride) |  /* pitch in bytes */
                    ztile);
@@ -259,7 +260,7 @@ i915_emit_hardware_state(struct i915_context *i915 )
          if (cbuf_surface)
             cformat = cbuf_surface->format;
          else
-            cformat = PIPE_FORMAT_A8R8G8B8_UNORM; /* arbitrary */
+            cformat = PIPE_FORMAT_B8G8R8A8_UNORM; /* arbitrary */
          cformat = translate_format(cformat);
 
          if (depth_surface) 
diff --git a/src/gallium/drivers/i915/i915_state_sampler.c b/src/gallium/drivers/i915/i915_state_sampler.c
index 483947ece7f..d6da8225490 100644
--- a/src/gallium/drivers/i915/i915_state_sampler.c
+++ b/src/gallium/drivers/i915/i915_state_sampler.c
@@ -87,8 +87,8 @@ static void update_sampler(struct i915_context *i915,
    state[1] = sampler->state[1];
    state[2] = sampler->state[2];
 
-   if (pt->format == PIPE_FORMAT_YCBCR ||
-       pt->format == PIPE_FORMAT_YCBCR_REV)
+   if (pt->format == PIPE_FORMAT_UYVY ||
+       pt->format == PIPE_FORMAT_YUYV)
       state[0] |= SS2_COLORSPACE_CONVERSION;
 
    /* 3D textures don't seem to respect the border color.
@@ -182,19 +182,19 @@ translate_texture_format(enum pipe_format pipeFormat)
       return MAPSURF_8BIT | MT_8BIT_I8;
    case PIPE_FORMAT_A8_UNORM:
       return MAPSURF_8BIT | MT_8BIT_A8;
-   case PIPE_FORMAT_A8L8_UNORM:
+   case PIPE_FORMAT_L8A8_UNORM:
       return MAPSURF_16BIT | MT_16BIT_AY88;
-   case PIPE_FORMAT_R5G6B5_UNORM:
+   case PIPE_FORMAT_B5G6R5_UNORM:
       return MAPSURF_16BIT | MT_16BIT_RGB565;
-   case PIPE_FORMAT_A1R5G5B5_UNORM:
+   case PIPE_FORMAT_B5G5R5A1_UNORM:
       return MAPSURF_16BIT | MT_16BIT_ARGB1555;
-   case PIPE_FORMAT_A4R4G4B4_UNORM:
+   case PIPE_FORMAT_B4G4R4A4_UNORM:
       return MAPSURF_16BIT | MT_16BIT_ARGB4444;
-   case PIPE_FORMAT_A8R8G8B8_UNORM:
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
       return MAPSURF_32BIT | MT_32BIT_ARGB8888;
-   case PIPE_FORMAT_YCBCR_REV:
+   case PIPE_FORMAT_YUYV:
       return (MAPSURF_422 | MT_422_YCRCB_NORMAL);
-   case PIPE_FORMAT_YCBCR:
+   case PIPE_FORMAT_UYVY:
       return (MAPSURF_422 | MT_422_YCRCB_SWAPY);
 #if 0
    case PIPE_FORMAT_RGB_FXT1:
@@ -212,7 +212,7 @@ translate_texture_format(enum pipe_format pipeFormat)
    case PIPE_FORMAT_RGBA_DXT5:
       return (MAPSURF_COMPRESSED | MT_COMPRESS_DXT4_5);
 #endif
-   case PIPE_FORMAT_S8Z24_UNORM:
+   case PIPE_FORMAT_Z24S8_UNORM:
       return (MAPSURF_32BIT | MT_32BIT_xI824);
    default:
       debug_printf("i915: translate_texture_format() bad image format %x\n",
diff --git a/src/gallium/drivers/i915/i915_texture.c b/src/gallium/drivers/i915/i915_texture.c
index e101c8683ec..3ce52cdcdbd 100644
--- a/src/gallium/drivers/i915/i915_texture.c
+++ b/src/gallium/drivers/i915/i915_texture.c
@@ -96,7 +96,7 @@ i915_miptree_set_level_info(struct i915_texture *tex,
                              unsigned nr_images,
                              unsigned w, unsigned h, unsigned d)
 {
-   assert(level < PIPE_MAX_TEXTURE_LEVELS);
+   assert(level < Elements(tex->nr_images));
 
    tex->nr_images[level] = nr_images;
 
@@ -219,12 +219,12 @@ i915_miptree_layout_2d(struct i915_texture *tex)
    unsigned nblocksy = util_format_get_nblocksy(pt->format, pt->width0);
 
    /* used for scanouts that need special layouts */
-   if (pt->tex_usage & PIPE_TEXTURE_USAGE_PRIMARY)
+   if (pt->tex_usage & PIPE_TEXTURE_USAGE_SCANOUT)
       if (i915_scanout_layout(tex))
          return;
 
-   /* for shared buffers we use some very like scanout */
-   if (pt->tex_usage & PIPE_TEXTURE_USAGE_DISPLAY_TARGET)
+   /* shared buffers needs to be compatible with X servers */
+   if (pt->tex_usage & PIPE_TEXTURE_USAGE_SHARED)
       if (i915_display_target_layout(tex))
          return;
 
@@ -369,12 +369,12 @@ i945_miptree_layout_2d(struct i915_texture *tex)
    unsigned nblocksy = util_format_get_nblocksy(pt->format, pt->height0);
 
    /* used for scanouts that need special layouts */
-   if (tex->base.tex_usage & PIPE_TEXTURE_USAGE_PRIMARY)
+   if (tex->base.tex_usage & PIPE_TEXTURE_USAGE_SCANOUT)
       if (i915_scanout_layout(tex))
          return;
 
-   /* for shared buffers we use some very like scanout */
-   if (tex->base.tex_usage & PIPE_TEXTURE_USAGE_DISPLAY_TARGET)
+   /* shared buffers needs to be compatible with X servers */
+   if (tex->base.tex_usage & PIPE_TEXTURE_USAGE_SHARED)
       if (i915_display_target_layout(tex))
          return;
 
@@ -642,7 +642,7 @@ i915_texture_create(struct pipe_screen *screen,
 
 
    /* for scanouts and cursors, cursors arn't scanouts */
-   if (templat->tex_usage & PIPE_TEXTURE_USAGE_PRIMARY && templat->width0 != 64)
+   if (templat->tex_usage & PIPE_TEXTURE_USAGE_SCANOUT && templat->width0 != 64)
       buf_usage = INTEL_NEW_SCANOUT;
    else
       buf_usage = INTEL_NEW_TEXTURE;
@@ -673,19 +673,24 @@ fail:
 }
 
 static struct pipe_texture *
-i915_texture_blanket(struct pipe_screen * screen,
-                     const struct pipe_texture *base,
-                     const unsigned *stride,
-                     struct pipe_buffer *buffer)
+i915_texture_from_handle(struct pipe_screen * screen,
+                         const struct pipe_texture *templat,
+                         struct winsys_handle *whandle)
 {
-#if 0
+   struct i915_screen *is = i915_screen(screen);
    struct i915_texture *tex;
+   struct intel_winsys *iws = is->iws;
+   struct intel_buffer *buffer;
+   unsigned stride;
+
    assert(screen);
 
+   buffer = iws->buffer_from_handle(iws, whandle, &stride);
+
    /* Only supports one type */
-   if (base->target != PIPE_TEXTURE_2D ||
-       base->last_level != 0 ||
-       base->depth0 != 1) {
+   if (templat->target != PIPE_TEXTURE_2D ||
+       templat->last_level != 0 ||
+       templat->depth0 != 1) {
       return NULL;
    }
 
@@ -693,23 +698,33 @@ i915_texture_blanket(struct pipe_screen * screen,
    if (!tex)
       return NULL;
 
-   tex->base = *base;
+   tex->base = *templat;
    pipe_reference_init(&tex->base.reference, 1);
    tex->base.screen = screen;
 
-   tex->stride = stride[0];
+   tex->stride = stride;
 
-   i915_miptree_set_level_info(tex, 0, 1, base->width0, base->height0, 1);
+   i915_miptree_set_level_info(tex, 0, 1, templat->width0, templat->height0, 1);
    i915_miptree_set_image_offset(tex, 0, 0, 0, 0);
 
-   pipe_buffer_reference(&tex->buffer, buffer);
+   tex->buffer = buffer;
 
    return &tex->base;
-#else
-   return NULL;
-#endif
 }
 
+static boolean
+i915_texture_get_handle(struct pipe_screen * screen,
+                        struct pipe_texture *texture,
+                        struct winsys_handle *whandle)
+{
+   struct i915_screen *is = i915_screen(screen);
+   struct i915_texture *tex = (struct i915_texture *)texture;
+   struct intel_winsys *iws = is->iws;
+
+   return iws->buffer_get_handle(iws, tex->buffer, whandle, tex->stride);
+}
+
+
 static void
 i915_texture_destroy(struct pipe_texture *pt)
 {
@@ -723,7 +738,7 @@ i915_texture_destroy(struct pipe_texture *pt)
 
    iws->buffer_destroy(iws, tex->buffer);
 
-   for (i = 0; i < PIPE_MAX_TEXTURE_LEVELS; i++)
+   for (i = 0; i < Elements(tex->image_offset); i++)
       if (tex->image_offset[i])
          FREE(tex->image_offset[i]);
 
@@ -869,7 +884,8 @@ void
 i915_init_screen_texture_functions(struct i915_screen *is)
 {
    is->base.texture_create = i915_texture_create;
-   is->base.texture_blanket = i915_texture_blanket;
+   is->base.texture_from_handle = i915_texture_from_handle;
+   is->base.texture_get_handle = i915_texture_get_handle;
    is->base.texture_destroy = i915_texture_destroy;
    is->base.get_tex_surface = i915_get_tex_surface;
    is->base.tex_surface_destroy = i915_tex_surface_destroy;
@@ -878,53 +894,3 @@ i915_init_screen_texture_functions(struct i915_screen *is)
    is->base.transfer_unmap = i915_transfer_unmap;
    is->base.tex_transfer_destroy = i915_tex_transfer_destroy;
 }
-
-struct pipe_texture *
-i915_texture_blanket_intel(struct pipe_screen *screen,
-                           struct pipe_texture *base,
-                           unsigned stride,
-                           struct intel_buffer *buffer)
-{
-   struct i915_texture *tex;
-   assert(screen);
-
-   /* Only supports one type */
-   if (base->target != PIPE_TEXTURE_2D ||
-       base->last_level != 0 ||
-       base->depth0 != 1) {
-      return NULL;
-   }
-
-   tex = CALLOC_STRUCT(i915_texture);
-   if (!tex)
-      return NULL;
-
-   tex->base = *base;
-   pipe_reference_init(&tex->base.reference, 1);
-   tex->base.screen = screen;
-
-   tex->stride = stride;
-
-   i915_miptree_set_level_info(tex, 0, 1, base->width0, base->height0, 1);
-   i915_miptree_set_image_offset(tex, 0, 0, 0, 0);
-
-   tex->buffer = buffer;
-
-   return &tex->base;
-}
-
-boolean
-i915_get_texture_buffer_intel(struct pipe_texture *texture,
-                              struct intel_buffer **buffer,
-                              unsigned *stride)
-{
-   struct i915_texture *tex = (struct i915_texture *)texture;
-
-   if (!texture)
-      return FALSE;
-
-   *stride = tex->stride;
-   *buffer = tex->buffer;
-
-   return TRUE;
-}
diff --git a/src/gallium/drivers/i915/intel_winsys.h b/src/gallium/drivers/i915/intel_winsys.h
index b3a802b0e29..00fd0c1efea 100644
--- a/src/gallium/drivers/i915/intel_winsys.h
+++ b/src/gallium/drivers/i915/intel_winsys.h
@@ -33,6 +33,7 @@ struct intel_buffer;
 struct intel_batchbuffer;
 struct pipe_texture;
 struct pipe_fence_handle;
+struct winsys_handle;
 
 enum intel_buffer_usage
 {
@@ -129,6 +130,25 @@ struct intel_winsys {
                                          enum intel_buffer_type type);
 
    /**
+    * Creates a buffer from a handle.
+    * Used to implement pipe_screen::texture_from_handle.
+    * Also provides the stride information needed for the
+    * texture via the stride argument.
+    */
+   struct intel_buffer *(*buffer_from_handle)(struct intel_winsys *iws,
+                                              struct winsys_handle *whandle,
+                                              unsigned *stride);
+
+   /**
+    * Used to implement pipe_screen::texture_get_handle.
+    * The winsys might need the stride information.
+    */
+   boolean (*buffer_get_handle)(struct intel_winsys *iws,
+                                struct intel_buffer *buffer,
+                                struct winsys_handle *whandle,
+                                unsigned stride);
+
+   /**
     * Fence a buffer with a fence reg.
     * Not to be confused with pipe_fence_handle.
     */
@@ -204,23 +224,4 @@ struct intel_winsys {
 struct pipe_screen *i915_create_screen(struct intel_winsys *iws, unsigned pci_id);
 
 
-/**
- * Get the intel_winsys buffer backing the texture.
- *
- * TODO UGLY
- */
-boolean i915_get_texture_buffer_intel(struct pipe_texture *texture,
-                                      struct intel_buffer **buffer,
-                                      unsigned *stride);
-
-/**
- * Wrap a intel_winsys buffer with a texture blanket.
- *
- * TODO UGLY
- */
-struct pipe_texture * i915_texture_blanket_intel(struct pipe_screen *screen,
-                                                 struct pipe_texture *tmplt,
-                                                 unsigned pitch,
-                                                 struct intel_buffer *buffer);
-
 #endif
diff --git a/src/gallium/drivers/i965/brw_context.h b/src/gallium/drivers/i965/brw_context.h
index 8816714a7aa..dab881fea24 100644
--- a/src/gallium/drivers/i965/brw_context.h
+++ b/src/gallium/drivers/i965/brw_context.h
@@ -351,7 +351,7 @@ struct brw_vs_prog_data {
 
 /* Size == 0 if output either not written, or always [0,0,0,1]
  */
-struct brw_vs_ouput_sizes {
+struct brw_vs_output_sizes {
    GLubyte output_size[PIPE_MAX_SHADER_OUTPUTS];
 };
 
@@ -546,14 +546,13 @@ struct brw_context
       const struct brw_blend_state *blend;
       const struct brw_rasterizer_state *rast;
       const struct brw_depth_stencil_state *zstencil;
+      const struct brw_vertex_element_packet *velems;
 
       const struct brw_sampler *sampler[PIPE_MAX_SAMPLERS];
       unsigned num_samplers;
 
       struct pipe_sampler_view *fragment_sampler_views[PIPE_MAX_SAMPLERS];
       struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
-      struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS];
-      unsigned num_vertex_elements;
       unsigned num_fragment_sampler_views;
       unsigned num_vertex_buffers;
 
diff --git a/src/gallium/drivers/i965/brw_draw_upload.c b/src/gallium/drivers/i965/brw_draw_upload.c
index d59261557b5..0820ba20a08 100644
--- a/src/gallium/drivers/i965/brw_draw_upload.c
+++ b/src/gallium/drivers/i965/brw_draw_upload.c
@@ -42,141 +42,6 @@
 
 
 
-static unsigned brw_translate_surface_format( unsigned id )
-{
-   switch (id) {
-   case PIPE_FORMAT_R64_FLOAT:
-      return BRW_SURFACEFORMAT_R64_FLOAT;
-   case PIPE_FORMAT_R64G64_FLOAT:
-      return BRW_SURFACEFORMAT_R64G64_FLOAT;
-   case PIPE_FORMAT_R64G64B64_FLOAT:
-      return BRW_SURFACEFORMAT_R64G64B64_FLOAT;
-   case PIPE_FORMAT_R64G64B64A64_FLOAT:
-      return BRW_SURFACEFORMAT_R64G64B64A64_FLOAT;
-
-   case PIPE_FORMAT_R32_FLOAT:
-      return BRW_SURFACEFORMAT_R32_FLOAT;
-   case PIPE_FORMAT_R32G32_FLOAT:
-      return BRW_SURFACEFORMAT_R32G32_FLOAT;
-   case PIPE_FORMAT_R32G32B32_FLOAT:
-      return BRW_SURFACEFORMAT_R32G32B32_FLOAT;
-   case PIPE_FORMAT_R32G32B32A32_FLOAT:
-      return BRW_SURFACEFORMAT_R32G32B32A32_FLOAT;
-
-   case PIPE_FORMAT_R32_UNORM:
-      return BRW_SURFACEFORMAT_R32_UNORM;
-   case PIPE_FORMAT_R32G32_UNORM:
-      return BRW_SURFACEFORMAT_R32G32_UNORM;
-   case PIPE_FORMAT_R32G32B32_UNORM:
-      return BRW_SURFACEFORMAT_R32G32B32_UNORM;
-   case PIPE_FORMAT_R32G32B32A32_UNORM:
-      return BRW_SURFACEFORMAT_R32G32B32A32_UNORM;
-
-   case PIPE_FORMAT_R32_USCALED:
-      return BRW_SURFACEFORMAT_R32_USCALED;
-   case PIPE_FORMAT_R32G32_USCALED:
-      return BRW_SURFACEFORMAT_R32G32_USCALED;
-   case PIPE_FORMAT_R32G32B32_USCALED:
-      return BRW_SURFACEFORMAT_R32G32B32_USCALED;
-   case PIPE_FORMAT_R32G32B32A32_USCALED:
-      return BRW_SURFACEFORMAT_R32G32B32A32_USCALED;
-
-   case PIPE_FORMAT_R32_SNORM:
-      return BRW_SURFACEFORMAT_R32_SNORM;
-   case PIPE_FORMAT_R32G32_SNORM:
-      return BRW_SURFACEFORMAT_R32G32_SNORM;
-   case PIPE_FORMAT_R32G32B32_SNORM:
-      return BRW_SURFACEFORMAT_R32G32B32_SNORM;
-   case PIPE_FORMAT_R32G32B32A32_SNORM:
-      return BRW_SURFACEFORMAT_R32G32B32A32_SNORM;
-
-   case PIPE_FORMAT_R32_SSCALED:
-      return BRW_SURFACEFORMAT_R32_SSCALED;
-   case PIPE_FORMAT_R32G32_SSCALED:
-      return BRW_SURFACEFORMAT_R32G32_SSCALED;
-   case PIPE_FORMAT_R32G32B32_SSCALED:
-      return BRW_SURFACEFORMAT_R32G32B32_SSCALED;
-   case PIPE_FORMAT_R32G32B32A32_SSCALED:
-      return BRW_SURFACEFORMAT_R32G32B32A32_SSCALED;
-
-   case PIPE_FORMAT_R16_UNORM:
-      return BRW_SURFACEFORMAT_R16_UNORM;
-   case PIPE_FORMAT_R16G16_UNORM:
-      return BRW_SURFACEFORMAT_R16G16_UNORM;
-   case PIPE_FORMAT_R16G16B16_UNORM:
-      return BRW_SURFACEFORMAT_R16G16B16_UNORM;
-   case PIPE_FORMAT_R16G16B16A16_UNORM:
-      return BRW_SURFACEFORMAT_R16G16B16A16_UNORM;
-
-   case PIPE_FORMAT_R16_USCALED:
-      return BRW_SURFACEFORMAT_R16_USCALED;
-   case PIPE_FORMAT_R16G16_USCALED:
-      return BRW_SURFACEFORMAT_R16G16_USCALED;
-   case PIPE_FORMAT_R16G16B16_USCALED:
-      return BRW_SURFACEFORMAT_R16G16B16_USCALED;
-   case PIPE_FORMAT_R16G16B16A16_USCALED:
-      return BRW_SURFACEFORMAT_R16G16B16A16_USCALED;
-
-   case PIPE_FORMAT_R16_SNORM:
-      return BRW_SURFACEFORMAT_R16_SNORM;
-   case PIPE_FORMAT_R16G16_SNORM:
-      return BRW_SURFACEFORMAT_R16G16_SNORM;
-   case PIPE_FORMAT_R16G16B16_SNORM:
-      return BRW_SURFACEFORMAT_R16G16B16_SNORM;
-   case PIPE_FORMAT_R16G16B16A16_SNORM:
-      return BRW_SURFACEFORMAT_R16G16B16A16_SNORM;
-
-   case PIPE_FORMAT_R16_SSCALED:
-      return BRW_SURFACEFORMAT_R16_SSCALED;
-   case PIPE_FORMAT_R16G16_SSCALED:
-      return BRW_SURFACEFORMAT_R16G16_SSCALED;
-   case PIPE_FORMAT_R16G16B16_SSCALED:
-      return BRW_SURFACEFORMAT_R16G16B16_SSCALED;
-   case PIPE_FORMAT_R16G16B16A16_SSCALED:
-      return BRW_SURFACEFORMAT_R16G16B16A16_SSCALED;
-
-   case PIPE_FORMAT_R8_UNORM:
-      return BRW_SURFACEFORMAT_R8_UNORM;
-   case PIPE_FORMAT_R8G8_UNORM:
-      return BRW_SURFACEFORMAT_R8G8_UNORM;
-   case PIPE_FORMAT_R8G8B8_UNORM:
-      return BRW_SURFACEFORMAT_R8G8B8_UNORM;
-   case PIPE_FORMAT_R8G8B8A8_UNORM:
-      return BRW_SURFACEFORMAT_R8G8B8A8_UNORM;
-
-   case PIPE_FORMAT_R8_USCALED:
-      return BRW_SURFACEFORMAT_R8_USCALED;
-   case PIPE_FORMAT_R8G8_USCALED:
-      return BRW_SURFACEFORMAT_R8G8_USCALED;
-   case PIPE_FORMAT_R8G8B8_USCALED:
-      return BRW_SURFACEFORMAT_R8G8B8_USCALED;
-   case PIPE_FORMAT_R8G8B8A8_USCALED:
-      return BRW_SURFACEFORMAT_R8G8B8A8_USCALED;
-
-   case PIPE_FORMAT_R8_SNORM:
-      return BRW_SURFACEFORMAT_R8_SNORM;
-   case PIPE_FORMAT_R8G8_SNORM:
-      return BRW_SURFACEFORMAT_R8G8_SNORM;
-   case PIPE_FORMAT_R8G8B8_SNORM:
-      return BRW_SURFACEFORMAT_R8G8B8_SNORM;
-   case PIPE_FORMAT_R8G8B8A8_SNORM:
-      return BRW_SURFACEFORMAT_R8G8B8A8_SNORM;
-
-   case PIPE_FORMAT_R8_SSCALED:
-      return BRW_SURFACEFORMAT_R8_SSCALED;
-   case PIPE_FORMAT_R8G8_SSCALED:
-      return BRW_SURFACEFORMAT_R8G8_SSCALED;
-   case PIPE_FORMAT_R8G8B8_SSCALED:
-      return BRW_SURFACEFORMAT_R8G8B8_SSCALED;
-   case PIPE_FORMAT_R8G8B8A8_SSCALED:
-      return BRW_SURFACEFORMAT_R8G8B8A8_SSCALED;
-
-   default:
-      assert(0);
-      return 0;
-   }
-}
-
 static unsigned get_index_type(int type)
 {
    switch (type) {
@@ -315,75 +180,16 @@ static int brw_emit_vertex_buffers( struct brw_context *brw )
 
 
 
-
 static int brw_emit_vertex_elements(struct brw_context *brw)
 {
-   GLuint nr = brw->curr.num_vertex_elements;
-   GLuint i;
+   const struct brw_vertex_element_packet *brw_velems = brw->curr.velems;
+   unsigned size = brw_velems->header.length + 2;
 
+   /* why is this here */
    brw_emit_query_begin(brw);
 
-   /* If the VS doesn't read any inputs (calculating vertex position from
-    * a state variable for some reason, for example), emit a single pad
-    * VERTEX_ELEMENT struct and bail.
-    *
-    * The stale VB state stays in place, but they don't do anything unless
-    * a VE loads from them.
-    */
-   if (nr == 0) {
-      BEGIN_BATCH(3, IGNORE_CLIPRECTS);
-      OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | 1);
-      OUT_BATCH((0 << BRW_VE0_INDEX_SHIFT) |
-		BRW_VE0_VALID |
-		(BRW_SURFACEFORMAT_R32G32B32A32_FLOAT << BRW_VE0_FORMAT_SHIFT) |
-		(0 << BRW_VE0_SRC_OFFSET_SHIFT));
-      OUT_BATCH((BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_0_SHIFT) |
-		(BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_1_SHIFT) |
-		(BRW_VE1_COMPONENT_STORE_0 << BRW_VE1_COMPONENT_2_SHIFT) |
-		(BRW_VE1_COMPONENT_STORE_1_FLT << BRW_VE1_COMPONENT_3_SHIFT));
-      ADVANCE_BATCH();
-      return 0;
-   }
-
-   /* Now emit vertex element (VEP) state packets.
-    *
-    */
-   BEGIN_BATCH(1 + nr * 2, IGNORE_CLIPRECTS);
-   OUT_BATCH((CMD_VERTEX_ELEMENT << 16) | ((1 + nr * 2) - 2));
-   for (i = 0; i < nr; i++) {
-      const struct pipe_vertex_element *input = &brw->curr.vertex_element[i];
-      uint32_t format = brw_translate_surface_format( input->src_format );
-      uint32_t comp0 = BRW_VE1_COMPONENT_STORE_SRC;
-      uint32_t comp1 = BRW_VE1_COMPONENT_STORE_SRC;
-      uint32_t comp2 = BRW_VE1_COMPONENT_STORE_SRC;
-      uint32_t comp3 = BRW_VE1_COMPONENT_STORE_SRC;
-
-      switch (input->nr_components) {
-      case 0: comp0 = BRW_VE1_COMPONENT_STORE_0;
-      case 1: comp1 = BRW_VE1_COMPONENT_STORE_0;
-      case 2: comp2 = BRW_VE1_COMPONENT_STORE_0;
-      case 3: comp3 = BRW_VE1_COMPONENT_STORE_1_FLT;
-	 break;
-      }
-
-      OUT_BATCH((input->vertex_buffer_index << BRW_VE0_INDEX_SHIFT) |
-		BRW_VE0_VALID |
-		(format << BRW_VE0_FORMAT_SHIFT) |
-		(input->src_offset << BRW_VE0_SRC_OFFSET_SHIFT));
+   brw_batchbuffer_data(brw->batch, brw_velems, size * 4, IGNORE_CLIPRECTS);
 
-      if (BRW_IS_IGDNG(brw))
-          OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
-                    (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
-                    (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
-                    (comp3 << BRW_VE1_COMPONENT_3_SHIFT));
-      else
-          OUT_BATCH((comp0 << BRW_VE1_COMPONENT_0_SHIFT) |
-                    (comp1 << BRW_VE1_COMPONENT_1_SHIFT) |
-                    (comp2 << BRW_VE1_COMPONENT_2_SHIFT) |
-                    (comp3 << BRW_VE1_COMPONENT_3_SHIFT) |
-                    ((i * 4) << BRW_VE1_DST_OFFSET_SHIFT));
-   }
-   ADVANCE_BATCH();
    return 0;
 }
 
@@ -396,10 +202,11 @@ static int brw_emit_vertices( struct brw_context *brw )
    if (ret)
       return ret;
 
+   /* XXX should separate this? */
    ret = brw_emit_vertex_elements( brw );
    if (ret)
       return ret;
-   
+
    return 0;
 }
 
@@ -407,7 +214,8 @@ static int brw_emit_vertices( struct brw_context *brw )
 const struct brw_tracked_state brw_vertices = {
    .dirty = {
       .mesa = (PIPE_NEW_INDEX_RANGE |
-               PIPE_NEW_VERTEX_BUFFER),
+               PIPE_NEW_VERTEX_BUFFER |
+               PIPE_NEW_VERTEX_ELEMENT),
       .brw = BRW_NEW_BATCH,
       .cache = 0,
    },
diff --git a/src/gallium/drivers/i965/brw_misc_state.c b/src/gallium/drivers/i965/brw_misc_state.c
index e4b24229db3..e3f25bdf622 100644
--- a/src/gallium/drivers/i965/brw_misc_state.c
+++ b/src/gallium/drivers/i965/brw_misc_state.c
@@ -265,8 +265,8 @@ static int emit_depthbuffer(struct brw_context *brw)
 	 format = BRW_DEPTHFORMAT_D16_UNORM;
 	 cpp = 2;
 	 break;
-      case PIPE_FORMAT_X8Z24_UNORM:
-      case PIPE_FORMAT_S8Z24_UNORM:
+      case PIPE_FORMAT_Z24X8_UNORM:
+      case PIPE_FORMAT_Z24S8_UNORM:
 	 format = BRW_DEPTHFORMAT_D24_UNORM_S8_UINT;
 	 cpp = 4;
 	 break;
diff --git a/src/gallium/drivers/i965/brw_pipe_clear.c b/src/gallium/drivers/i965/brw_pipe_clear.c
index 452e1e89f93..d7048f769b2 100644
--- a/src/gallium/drivers/i965/brw_pipe_clear.c
+++ b/src/gallium/drivers/i965/brw_pipe_clear.c
@@ -139,8 +139,8 @@ static void zstencil_clear(struct brw_context *brw,
    unsigned value;
 
    switch (bsurface->base.format) {
-   case PIPE_FORMAT_X8Z24_UNORM:
-   case PIPE_FORMAT_S8Z24_UNORM:
+   case PIPE_FORMAT_Z24X8_UNORM:
+   case PIPE_FORMAT_Z24S8_UNORM:
       value = ((unsigned)(depth * MASK24) & MASK24);
       break;
    case PIPE_FORMAT_Z16_UNORM:
@@ -152,8 +152,8 @@ static void zstencil_clear(struct brw_context *brw,
    }
 
    switch (bsurface->base.format) {
-   case PIPE_FORMAT_X8Z24_UNORM:
-   case PIPE_FORMAT_S8Z24_UNORM:
+   case PIPE_FORMAT_Z24X8_UNORM:
+   case PIPE_FORMAT_Z24S8_UNORM:
       value = value | (stencil << 24);
       break;
 
diff --git a/src/gallium/drivers/i965/brw_pipe_vertex.c b/src/gallium/drivers/i965/brw_pipe_vertex.c
index e3c48e31493..d6a840857ec 100644
--- a/src/gallium/drivers/i965/brw_pipe_vertex.c
+++ b/src/gallium/drivers/i965/brw_pipe_vertex.c
@@ -1,22 +1,251 @@
 #include "brw_context.h"
+#include "brw_defines.h"
+#include "brw_structs.h"
 
+#include "util/u_memory.h"
+#include "util/u_format.h"
 
-static void brw_set_vertex_elements( struct pipe_context *pipe,
-				     unsigned count,
-				     const struct pipe_vertex_element *elements )
+
+static unsigned brw_translate_surface_format( unsigned id )
+{
+   switch (id) {
+   case PIPE_FORMAT_R64_FLOAT:
+      return BRW_SURFACEFORMAT_R64_FLOAT;
+   case PIPE_FORMAT_R64G64_FLOAT:
+      return BRW_SURFACEFORMAT_R64G64_FLOAT;
+   case PIPE_FORMAT_R64G64B64_FLOAT:
+      return BRW_SURFACEFORMAT_R64G64B64_FLOAT;
+   case PIPE_FORMAT_R64G64B64A64_FLOAT:
+      return BRW_SURFACEFORMAT_R64G64B64A64_FLOAT;
+
+   case PIPE_FORMAT_R32_FLOAT:
+      return BRW_SURFACEFORMAT_R32_FLOAT;
+   case PIPE_FORMAT_R32G32_FLOAT:
+      return BRW_SURFACEFORMAT_R32G32_FLOAT;
+   case PIPE_FORMAT_R32G32B32_FLOAT:
+      return BRW_SURFACEFORMAT_R32G32B32_FLOAT;
+   case PIPE_FORMAT_R32G32B32A32_FLOAT:
+      return BRW_SURFACEFORMAT_R32G32B32A32_FLOAT;
+
+   case PIPE_FORMAT_R32_UNORM:
+      return BRW_SURFACEFORMAT_R32_UNORM;
+   case PIPE_FORMAT_R32G32_UNORM:
+      return BRW_SURFACEFORMAT_R32G32_UNORM;
+   case PIPE_FORMAT_R32G32B32_UNORM:
+      return BRW_SURFACEFORMAT_R32G32B32_UNORM;
+   case PIPE_FORMAT_R32G32B32A32_UNORM:
+      return BRW_SURFACEFORMAT_R32G32B32A32_UNORM;
+
+   case PIPE_FORMAT_R32_USCALED:
+      return BRW_SURFACEFORMAT_R32_USCALED;
+   case PIPE_FORMAT_R32G32_USCALED:
+      return BRW_SURFACEFORMAT_R32G32_USCALED;
+   case PIPE_FORMAT_R32G32B32_USCALED:
+      return BRW_SURFACEFORMAT_R32G32B32_USCALED;
+   case PIPE_FORMAT_R32G32B32A32_USCALED:
+      return BRW_SURFACEFORMAT_R32G32B32A32_USCALED;
+
+   case PIPE_FORMAT_R32_SNORM:
+      return BRW_SURFACEFORMAT_R32_SNORM;
+   case PIPE_FORMAT_R32G32_SNORM:
+      return BRW_SURFACEFORMAT_R32G32_SNORM;
+   case PIPE_FORMAT_R32G32B32_SNORM:
+      return BRW_SURFACEFORMAT_R32G32B32_SNORM;
+   case PIPE_FORMAT_R32G32B32A32_SNORM:
+      return BRW_SURFACEFORMAT_R32G32B32A32_SNORM;
+
+   case PIPE_FORMAT_R32_SSCALED:
+      return BRW_SURFACEFORMAT_R32_SSCALED;
+   case PIPE_FORMAT_R32G32_SSCALED:
+      return BRW_SURFACEFORMAT_R32G32_SSCALED;
+   case PIPE_FORMAT_R32G32B32_SSCALED:
+      return BRW_SURFACEFORMAT_R32G32B32_SSCALED;
+   case PIPE_FORMAT_R32G32B32A32_SSCALED:
+      return BRW_SURFACEFORMAT_R32G32B32A32_SSCALED;
+
+   case PIPE_FORMAT_R16_UNORM:
+      return BRW_SURFACEFORMAT_R16_UNORM;
+   case PIPE_FORMAT_R16G16_UNORM:
+      return BRW_SURFACEFORMAT_R16G16_UNORM;
+   case PIPE_FORMAT_R16G16B16_UNORM:
+      return BRW_SURFACEFORMAT_R16G16B16_UNORM;
+   case PIPE_FORMAT_R16G16B16A16_UNORM:
+      return BRW_SURFACEFORMAT_R16G16B16A16_UNORM;
+
+   case PIPE_FORMAT_R16_USCALED:
+      return BRW_SURFACEFORMAT_R16_USCALED;
+   case PIPE_FORMAT_R16G16_USCALED:
+      return BRW_SURFACEFORMAT_R16G16_USCALED;
+   case PIPE_FORMAT_R16G16B16_USCALED:
+      return BRW_SURFACEFORMAT_R16G16B16_USCALED;
+   case PIPE_FORMAT_R16G16B16A16_USCALED:
+      return BRW_SURFACEFORMAT_R16G16B16A16_USCALED;
+
+   case PIPE_FORMAT_R16_SNORM:
+      return BRW_SURFACEFORMAT_R16_SNORM;
+   case PIPE_FORMAT_R16G16_SNORM:
+      return BRW_SURFACEFORMAT_R16G16_SNORM;
+   case PIPE_FORMAT_R16G16B16_SNORM:
+      return BRW_SURFACEFORMAT_R16G16B16_SNORM;
+   case PIPE_FORMAT_R16G16B16A16_SNORM:
+      return BRW_SURFACEFORMAT_R16G16B16A16_SNORM;
+
+   case PIPE_FORMAT_R16_SSCALED:
+      return BRW_SURFACEFORMAT_R16_SSCALED;
+   case PIPE_FORMAT_R16G16_SSCALED:
+      return BRW_SURFACEFORMAT_R16G16_SSCALED;
+   case PIPE_FORMAT_R16G16B16_SSCALED:
+      return BRW_SURFACEFORMAT_R16G16B16_SSCALED;
+   case PIPE_FORMAT_R16G16B16A16_SSCALED:
+      return BRW_SURFACEFORMAT_R16G16B16A16_SSCALED;
+
+   case PIPE_FORMAT_R8_UNORM:
+      return BRW_SURFACEFORMAT_R8_UNORM;
+   case PIPE_FORMAT_R8G8_UNORM:
+      return BRW_SURFACEFORMAT_R8G8_UNORM;
+   case PIPE_FORMAT_R8G8B8_UNORM:
+      return BRW_SURFACEFORMAT_R8G8B8_UNORM;
+   case PIPE_FORMAT_R8G8B8A8_UNORM:
+      return BRW_SURFACEFORMAT_R8G8B8A8_UNORM;
+
+   case PIPE_FORMAT_R8_USCALED:
+      return BRW_SURFACEFORMAT_R8_USCALED;
+   case PIPE_FORMAT_R8G8_USCALED:
+      return BRW_SURFACEFORMAT_R8G8_USCALED;
+   case PIPE_FORMAT_R8G8B8_USCALED:
+      return BRW_SURFACEFORMAT_R8G8B8_USCALED;
+   case PIPE_FORMAT_R8G8B8A8_USCALED:
+      return BRW_SURFACEFORMAT_R8G8B8A8_USCALED;
+
+   case PIPE_FORMAT_R8_SNORM:
+      return BRW_SURFACEFORMAT_R8_SNORM;
+   case PIPE_FORMAT_R8G8_SNORM:
+      return BRW_SURFACEFORMAT_R8G8_SNORM;
+   case PIPE_FORMAT_R8G8B8_SNORM:
+      return BRW_SURFACEFORMAT_R8G8B8_SNORM;
+   case PIPE_FORMAT_R8G8B8A8_SNORM:
+      return BRW_SURFACEFORMAT_R8G8B8A8_SNORM;
+
+   case PIPE_FORMAT_R8_SSCALED:
+      return BRW_SURFACEFORMAT_R8_SSCALED;
+   case PIPE_FORMAT_R8G8_SSCALED:
+      return BRW_SURFACEFORMAT_R8G8_SSCALED;
+   case PIPE_FORMAT_R8G8B8_SSCALED:
+      return BRW_SURFACEFORMAT_R8G8B8_SSCALED;
+   case PIPE_FORMAT_R8G8B8A8_SSCALED:
+      return BRW_SURFACEFORMAT_R8G8B8A8_SSCALED;
+
+   default:
+      assert(0);
+      return 0;
+   }
+}
+
+static void brw_translate_vertex_elements(struct brw_context *brw,
+                                          struct brw_vertex_element_packet *brw_velems,
+                                          const struct pipe_vertex_element *attribs,
+                                          unsigned count)
+{
+   unsigned i;
+
+   /* If the VS doesn't read any inputs (calculating vertex position from
+    * a state variable for some reason, for example), emit a single pad
+    * VERTEX_ELEMENT struct and bail.
+    *
+    * The stale VB state stays in place, but they don't do anything unless
+    * a VE loads from them.
+    */
+   brw_velems->header.opcode = CMD_VERTEX_ELEMENT;
+
+   if (count == 0) {
+      brw_velems->header.length = 1;
+      brw_velems->ve[0].ve0.src_offset = 0;
+      brw_velems->ve[0].ve0.src_format = BRW_SURFACEFORMAT_R32G32B32A32_FLOAT;
+      brw_velems->ve[0].ve0.valid = 1;
+      brw_velems->ve[0].ve0.vertex_buffer_index = 0;
+      brw_velems->ve[0].ve1.dst_offset = 0;
+      brw_velems->ve[0].ve1.vfcomponent0 = BRW_VE1_COMPONENT_STORE_0;
+      brw_velems->ve[0].ve1.vfcomponent1 = BRW_VE1_COMPONENT_STORE_0;
+      brw_velems->ve[0].ve1.vfcomponent2 = BRW_VE1_COMPONENT_STORE_0;
+      brw_velems->ve[0].ve1.vfcomponent3 = BRW_VE1_COMPONENT_STORE_1_FLT;
+      return;
+   }
+
+
+   /* Now emit vertex element (VEP) state packets.
+    *
+    */
+   brw_velems->header.length = (1 + count * 2) - 2;
+   for (i = 0; i < count; i++) {
+      const struct pipe_vertex_element *input = &attribs[i];
+      unsigned nr_components = util_format_get_nr_components(input->src_format);
+
+      uint32_t format = brw_translate_surface_format( input->src_format );
+      uint32_t comp0 = BRW_VE1_COMPONENT_STORE_SRC;
+      uint32_t comp1 = BRW_VE1_COMPONENT_STORE_SRC;
+      uint32_t comp2 = BRW_VE1_COMPONENT_STORE_SRC;
+      uint32_t comp3 = BRW_VE1_COMPONENT_STORE_SRC;
+
+      switch (nr_components) {
+      case 0: comp0 = BRW_VE1_COMPONENT_STORE_0; /* fallthrough */
+      case 1: comp1 = BRW_VE1_COMPONENT_STORE_0; /* fallthrough */
+      case 2: comp2 = BRW_VE1_COMPONENT_STORE_0; /* fallthrough */
+      case 3: comp3 = BRW_VE1_COMPONENT_STORE_1_FLT;
+         break;
+      }
+
+      brw_velems->ve[i].ve0.src_offset = input->src_offset;
+      brw_velems->ve[i].ve0.src_format = format;
+      brw_velems->ve[i].ve0.valid = 1;
+      brw_velems->ve[i].ve0.vertex_buffer_index = input->vertex_buffer_index;
+      brw_velems->ve[i].ve1.vfcomponent0 = comp0;
+      brw_velems->ve[i].ve1.vfcomponent1 = comp1;
+      brw_velems->ve[i].ve1.vfcomponent2 = comp2;
+      brw_velems->ve[i].ve1.vfcomponent3 = comp3;
+
+      if (BRW_IS_IGDNG(brw))
+         brw_velems->ve[i].ve1.dst_offset = 0;
+      else
+         brw_velems->ve[i].ve1.dst_offset = i * 4;
+   }
+}
+
+static void* brw_create_vertex_elements_state( struct pipe_context *pipe,
+                                               unsigned count,
+                                               const struct pipe_vertex_element *attribs )
 {
+   /* note: for the brw_swtnl.c code (if ever we need draw fallback) we'd also need
+      to store the original data */
    struct brw_context *brw = brw_context(pipe);
+   struct brw_vertex_element_packet *velems;
+   assert(count <= BRW_VEP_MAX);
+   velems = (struct brw_vertex_element_packet *) MALLOC(sizeof(struct brw_vertex_element_packet));
+   if (velems) {
+      brw_translate_vertex_elements(brw, velems, attribs, count);
+   }
+   return velems;
+}
 
-   memcpy(brw->curr.vertex_element, elements, count * sizeof(elements[0]));
-   brw->curr.num_vertex_elements = count;
+static void brw_bind_vertex_elements_state(struct pipe_context *pipe,
+                                           void *velems)
+{
+   struct brw_context *brw = brw_context(pipe);
+   struct brw_vertex_element_packet *brw_velems = (struct brw_vertex_element_packet *) velems;
+
+   brw->curr.velems = brw_velems;
 
    brw->state.dirty.mesa |= PIPE_NEW_VERTEX_ELEMENT;
 }
 
+static void brw_delete_vertex_elements_state(struct pipe_context *pipe, void *velems)
+{
+   FREE( velems );
+}
+
 
 static void brw_set_vertex_buffers(struct pipe_context *pipe,
-				   unsigned count,
-				   const struct pipe_vertex_buffer *buffers)
+                                   unsigned count,
+                                   const struct pipe_vertex_buffer *buffers)
 {
    struct brw_context *brw = brw_context(pipe);
    unsigned i;
@@ -49,7 +278,9 @@ void
 brw_pipe_vertex_init( struct brw_context *brw )
 {
    brw->base.set_vertex_buffers = brw_set_vertex_buffers;
-   brw->base.set_vertex_elements = brw_set_vertex_elements;
+   brw->base.create_vertex_elements_state = brw_create_vertex_elements_state;
+   brw->base.bind_vertex_elements_state = brw_bind_vertex_elements_state;
+   brw->base.delete_vertex_elements_state = brw_delete_vertex_elements_state;
 }
 
 
diff --git a/src/gallium/drivers/i965/brw_screen.c b/src/gallium/drivers/i965/brw_screen.c
index 184cd490e55..cef83ffea80 100644
--- a/src/gallium/drivers/i965/brw_screen.c
+++ b/src/gallium/drivers/i965/brw_screen.c
@@ -153,6 +153,10 @@ brw_get_param(struct pipe_screen *screen, int param)
    switch (param) {
    case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS:
       return 8;
+   case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS:
+      return 8;
+   case PIPE_CAP_MAX_COMBINED_SAMPLERS:
+      return 16; /* XXX correct? */
    case PIPE_CAP_NPOT_TEXTURES:
       return 1;
    case PIPE_CAP_TWO_SIDED_STENCIL:
@@ -170,11 +174,11 @@ brw_get_param(struct pipe_screen *screen, int param)
    case PIPE_CAP_TEXTURE_SHADOW_MAP:
       return 1;
    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
-      return 11; /* max 1024x1024 */
+      return BRW_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
-      return 8;  /* max 128x128x128 */
+      return BRW_MAX_TEXTURE_3D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
-      return 11; /* max 1024x1024 */
+      return BRW_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_TGSI_FS_COORD_ORIGIN_UPPER_LEFT:
    case PIPE_CAP_TGSI_FS_COORD_PIXEL_CENTER_HALF_INTEGER:
       return 1;
@@ -225,15 +229,15 @@ brw_is_format_supported(struct pipe_screen *screen,
       PIPE_FORMAT_L16_UNORM,
       /*PIPE_FORMAT_I16_UNORM,*/
       /*PIPE_FORMAT_A16_UNORM,*/
-      PIPE_FORMAT_A8L8_UNORM,
-      PIPE_FORMAT_R5G6B5_UNORM,
-      PIPE_FORMAT_A1R5G5B5_UNORM,
-      PIPE_FORMAT_A4R4G4B4_UNORM,
-      PIPE_FORMAT_X8R8G8B8_UNORM,
-      PIPE_FORMAT_A8R8G8B8_UNORM,
+      PIPE_FORMAT_L8A8_UNORM,
+      PIPE_FORMAT_B5G6R5_UNORM,
+      PIPE_FORMAT_B5G5R5A1_UNORM,
+      PIPE_FORMAT_B4G4R4A4_UNORM,
+      PIPE_FORMAT_B8G8R8X8_UNORM,
+      PIPE_FORMAT_B8G8R8A8_UNORM,
       /* video */
-      PIPE_FORMAT_YCBCR,
-      PIPE_FORMAT_YCBCR_REV,
+      PIPE_FORMAT_UYVY,
+      PIPE_FORMAT_YUYV,
       /* compressed */
       /*PIPE_FORMAT_FXT1_RGBA,*/
       PIPE_FORMAT_DXT1_RGB,
@@ -241,14 +245,14 @@ brw_is_format_supported(struct pipe_screen *screen,
       PIPE_FORMAT_DXT3_RGBA,
       PIPE_FORMAT_DXT5_RGBA,
       /* sRGB */
-      PIPE_FORMAT_R8G8B8A8_SRGB,
-      PIPE_FORMAT_A8L8_SRGB,
+      PIPE_FORMAT_A8B8G8R8_SRGB,
+      PIPE_FORMAT_L8A8_SRGB,
       PIPE_FORMAT_L8_SRGB,
       PIPE_FORMAT_DXT1_SRGB,
       /* depth */
       PIPE_FORMAT_Z32_FLOAT,
-      PIPE_FORMAT_X8Z24_UNORM,
-      PIPE_FORMAT_S8Z24_UNORM,
+      PIPE_FORMAT_Z24X8_UNORM,
+      PIPE_FORMAT_Z24S8_UNORM,
       PIPE_FORMAT_Z16_UNORM,
       /* signed */
       PIPE_FORMAT_R8G8_SNORM,
@@ -256,15 +260,15 @@ brw_is_format_supported(struct pipe_screen *screen,
       PIPE_FORMAT_NONE  /* list terminator */
    };
    static const enum pipe_format render_supported[] = {
-      PIPE_FORMAT_X8R8G8B8_UNORM,
-      PIPE_FORMAT_A8R8G8B8_UNORM,
-      PIPE_FORMAT_R5G6B5_UNORM,
+      PIPE_FORMAT_B8G8R8X8_UNORM,
+      PIPE_FORMAT_B8G8R8A8_UNORM,
+      PIPE_FORMAT_B5G6R5_UNORM,
       PIPE_FORMAT_NONE  /* list terminator */
    };
    static const enum pipe_format depth_supported[] = {
       PIPE_FORMAT_Z32_FLOAT,
-      PIPE_FORMAT_X8Z24_UNORM,
-      PIPE_FORMAT_S8Z24_UNORM,
+      PIPE_FORMAT_Z24X8_UNORM,
+      PIPE_FORMAT_Z24S8_UNORM,
       PIPE_FORMAT_Z16_UNORM,
       PIPE_FORMAT_NONE  /* list terminator */
    };
diff --git a/src/gallium/drivers/i965/brw_screen.h b/src/gallium/drivers/i965/brw_screen.h
index 7226d9228b7..e0f3cd2a9fe 100644
--- a/src/gallium/drivers/i965/brw_screen.h
+++ b/src/gallium/drivers/i965/brw_screen.h
@@ -100,6 +100,9 @@ struct brw_surface
 };
 
 
+#define BRW_MAX_TEXTURE_2D_LEVELS 11  /* max 1024x1024 */
+#define BRW_MAX_TEXTURE_3D_LEVELS  8  /* max 128x128x128 */
+
 
 struct brw_texture
 {
@@ -107,9 +110,9 @@ struct brw_texture
    struct brw_winsys_buffer *bo;
    struct brw_surface_state ss;
 
-   unsigned *image_offset[PIPE_MAX_TEXTURE_LEVELS];
-   unsigned nr_images[PIPE_MAX_TEXTURE_LEVELS];
-   unsigned level_offset[PIPE_MAX_TEXTURE_LEVELS];
+   unsigned *image_offset[BRW_MAX_TEXTURE_2D_LEVELS];
+   unsigned nr_images[BRW_MAX_TEXTURE_2D_LEVELS];
+   unsigned level_offset[BRW_MAX_TEXTURE_2D_LEVELS];
 
    boolean compressed;
    unsigned brw_target;
diff --git a/src/gallium/drivers/i965/brw_screen_surface.c b/src/gallium/drivers/i965/brw_screen_surface.c
index e2b9954e596..904df813dda 100644
--- a/src/gallium/drivers/i965/brw_screen_surface.c
+++ b/src/gallium/drivers/i965/brw_screen_surface.c
@@ -231,8 +231,6 @@ static struct pipe_surface *brw_get_tex_surface(struct pipe_screen *screen,
    case BRW_VIEW_IN_PLACE:
       surface = create_in_place_view( bscreen, tex, id, usage );
       break;
-   default:
-      return NULL;
    }
 
    insert_at_head( &tex->views[type], surface );
diff --git a/src/gallium/drivers/i965/brw_screen_texture.c b/src/gallium/drivers/i965/brw_screen_texture.c
index feb9d5f7657..cc79bfc7715 100644
--- a/src/gallium/drivers/i965/brw_screen_texture.c
+++ b/src/gallium/drivers/i965/brw_screen_texture.c
@@ -85,32 +85,32 @@ static GLuint translate_tex_format( enum pipe_format pf )
       return BRW_SURFACEFORMAT_A16_UNORM; 
       */
 
-   case PIPE_FORMAT_A8L8_UNORM:
+   case PIPE_FORMAT_L8A8_UNORM:
       return BRW_SURFACEFORMAT_L8A8_UNORM;
 
-   case PIPE_FORMAT_R5G6B5_UNORM:
+   case PIPE_FORMAT_B5G6R5_UNORM:
       return BRW_SURFACEFORMAT_B5G6R5_UNORM;
 
-   case PIPE_FORMAT_A1R5G5B5_UNORM:
+   case PIPE_FORMAT_B5G5R5A1_UNORM:
       return BRW_SURFACEFORMAT_B5G5R5A1_UNORM;
 
-   case PIPE_FORMAT_A4R4G4B4_UNORM:
+   case PIPE_FORMAT_B4G4R4A4_UNORM:
       return BRW_SURFACEFORMAT_B4G4R4A4_UNORM;
 
-   case PIPE_FORMAT_X8R8G8B8_UNORM:
+   case PIPE_FORMAT_B8G8R8X8_UNORM:
       return BRW_SURFACEFORMAT_R8G8B8X8_UNORM;
 
-   case PIPE_FORMAT_A8R8G8B8_UNORM:
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
       return BRW_SURFACEFORMAT_B8G8R8A8_UNORM;
 
    /*
     * Video formats
     */
 
-   case PIPE_FORMAT_YCBCR_REV:
+   case PIPE_FORMAT_YUYV:
       return BRW_SURFACEFORMAT_YCRCB_NORMAL;
 
-   case PIPE_FORMAT_YCBCR:
+   case PIPE_FORMAT_UYVY:
       return BRW_SURFACEFORMAT_YCRCB_SWAPUVY;
 
    /*
@@ -137,10 +137,10 @@ static GLuint translate_tex_format( enum pipe_format pf )
     * sRGB formats
     */
 
-   case PIPE_FORMAT_R8G8B8A8_SRGB:
+   case PIPE_FORMAT_A8B8G8R8_SRGB:
       return BRW_SURFACEFORMAT_B8G8R8A8_UNORM_SRGB;
 
-   case PIPE_FORMAT_A8L8_SRGB:
+   case PIPE_FORMAT_L8A8_SRGB:
       return BRW_SURFACEFORMAT_L8A8_UNORM_SRGB;
 
    case PIPE_FORMAT_L8_SRGB:
@@ -156,8 +156,8 @@ static GLuint translate_tex_format( enum pipe_format pf )
    case PIPE_FORMAT_Z16_UNORM:
          return BRW_SURFACEFORMAT_I16_UNORM;
 
-   case PIPE_FORMAT_S8Z24_UNORM:
-   case PIPE_FORMAT_X8Z24_UNORM:
+   case PIPE_FORMAT_Z24S8_UNORM:
+   case PIPE_FORMAT_Z24X8_UNORM:
          return BRW_SURFACEFORMAT_I24X8_UNORM;
 
    case PIPE_FORMAT_Z32_FLOAT:
@@ -191,6 +191,7 @@ static struct pipe_texture *brw_texture_create( struct pipe_screen *screen,
    struct brw_texture *tex;
    enum brw_buffer_type buffer_type;
    enum pipe_error ret;
+   GLuint format;
    
    tex = CALLOC_STRUCT(brw_texture);
    if (tex == NULL)
@@ -230,8 +231,8 @@ static struct pipe_texture *brw_texture_create( struct pipe_screen *screen,
       goto fail;
 
    
-   if (templ->tex_usage & (PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
-                           PIPE_TEXTURE_USAGE_PRIMARY)) {
+   if (templ->tex_usage & (PIPE_TEXTURE_USAGE_SCANOUT |
+                           PIPE_TEXTURE_USAGE_SHARED)) {
       buffer_type = BRW_BUFFER_TYPE_SCANOUT;
    }
    else {
@@ -248,8 +249,10 @@ static struct pipe_texture *brw_texture_create( struct pipe_screen *screen,
 
    tex->ss.ss0.mipmap_layout_mode = BRW_SURFACE_MIPMAPLAYOUT_BELOW;
    tex->ss.ss0.surface_type = translate_tex_target(tex->base.target);
-   tex->ss.ss0.surface_format = translate_tex_format(tex->base.format);
-   assert(tex->ss.ss0.surface_format != BRW_SURFACEFORMAT_INVALID);
+
+   format = translate_tex_format(tex->base.format);
+   assert(format != BRW_SURFACEFORMAT_INVALID);
+   tex->ss.ss0.surface_format = format;
 
    /* This is ok for all textures with channel width 8bit or less:
     */
@@ -300,14 +303,121 @@ fail:
    return NULL;
 }
 
-static struct pipe_texture *brw_texture_blanket(struct pipe_screen *screen,
-						const struct pipe_texture *templ,
-						const unsigned *stride,
-						struct pipe_buffer *buffer)
+static struct pipe_texture * 
+brw_texture_from_handle(struct pipe_screen *screen,
+                        const struct pipe_texture *templ,
+                        struct winsys_handle *whandle)
 {
+   struct brw_screen *bscreen = brw_screen(screen);
+   struct brw_texture *tex;
+   struct brw_winsys_buffer *buffer;
+   unsigned tiling;
+   unsigned pitch;
+
+   if (templ->target != PIPE_TEXTURE_2D ||
+       templ->last_level != 0 ||
+       templ->depth0 != 1)
+      return NULL;
+
+   if (util_format_is_compressed(templ->format))
+      return NULL;
+
+   tex = CALLOC_STRUCT(brw_texture);
+   if (!tex)
+      return NULL;
+
+   if (bscreen->sws->bo_from_handle(bscreen->sws, whandle, &pitch, &tiling, &buffer) != PIPE_OK)
+      goto fail;
+
+   memcpy(&tex->base, templ, sizeof *templ);
+   pipe_reference_init(&tex->base.reference, 1);
+   tex->base.screen = screen;
+
+   /* XXX: cpp vs. blocksize
+    */
+   tex->cpp = util_format_get_blocksize(tex->base.format);
+   tex->tiling = tiling;
+
+   make_empty_list(&tex->views[0]);
+   make_empty_list(&tex->views[1]);
+
+   if (!brw_texture_layout(bscreen, tex))
+      goto fail;
+
+   /* XXX Maybe some more checks? */
+   if ((pitch / tex->cpp) < tex->pitch)
+      goto fail;
+
+   tex->pitch = pitch / tex->cpp;
+
+   tex->bo = buffer;
+
+   /* fix this warning */
+#if 0
+   if (tex->size > buffer->size)
+      goto fail;
+#endif
+
+   tex->ss.ss0.mipmap_layout_mode = BRW_SURFACE_MIPMAPLAYOUT_BELOW;
+   tex->ss.ss0.surface_type = translate_tex_target(tex->base.target);
+   tex->ss.ss0.surface_format = translate_tex_format(tex->base.format);
+   assert(tex->ss.ss0.surface_format != BRW_SURFACEFORMAT_INVALID);
+
+   /* This is ok for all textures with channel width 8bit or less:
+    */
+/*    tex->ss.ss0.data_return_format = BRW_SURFACERETURNFORMAT_S1; */
+
+
+   /* XXX: what happens when tex->bo->offset changes???
+    */
+   tex->ss.ss1.base_addr = 0; /* reloc */
+   tex->ss.ss2.mip_count = tex->base.last_level;
+   tex->ss.ss2.width = tex->base.width0 - 1;
+   tex->ss.ss2.height = tex->base.height0 - 1;
+
+   switch (tex->tiling) {
+   case BRW_TILING_NONE:
+      tex->ss.ss3.tiled_surface = 0;
+      tex->ss.ss3.tile_walk = 0;
+      break;
+   case BRW_TILING_X:
+      tex->ss.ss3.tiled_surface = 1;
+      tex->ss.ss3.tile_walk = BRW_TILEWALK_XMAJOR;
+      break;
+   case BRW_TILING_Y:
+      tex->ss.ss3.tiled_surface = 1;
+      tex->ss.ss3.tile_walk = BRW_TILEWALK_YMAJOR;
+      break;
+   }
+
+   tex->ss.ss3.pitch = (tex->pitch * tex->cpp) - 1;
+   tex->ss.ss3.depth = tex->base.depth0 - 1;
+
+   tex->ss.ss4.min_lod = 0;
+
+   return &tex->base;
+
+fail:
+   FREE(tex);
    return NULL;
 }
 
+static boolean
+brw_texture_get_handle(struct pipe_screen *screen,
+                       struct pipe_texture *texture,
+                       struct winsys_handle *whandle)
+{
+   struct brw_screen *bscreen = brw_screen(screen);
+   struct brw_texture *tex = brw_texture(texture);
+   unsigned stride;
+
+   stride = tex->pitch * tex->cpp;
+
+   return bscreen->sws->bo_get_handle(tex->bo, whandle, stride);
+}
+
+
+
 static void brw_texture_destroy(struct pipe_texture *pt)
 {
    struct brw_texture *tex = brw_texture(pt);
@@ -448,124 +558,13 @@ brw_tex_transfer_destroy(struct pipe_transfer *trans)
 }
 
 
-/*
- * Functions exported to the winsys
- */
-
-boolean brw_texture_get_winsys_buffer(struct pipe_texture *texture,
-                                      struct brw_winsys_buffer **buffer,
-                                      unsigned *stride)
-{
-   struct brw_texture *tex = brw_texture(texture);
-
-   *buffer = tex->bo;
-   if (stride)
-      *stride = tex->pitch * tex->cpp;
-
-   return TRUE;
-}
-
-struct pipe_texture * 
-brw_texture_blanket_winsys_buffer(struct pipe_screen *screen,
-                                  const struct pipe_texture *templ,
-                                  unsigned pitch,
-				  unsigned tiling,
-                                  struct brw_winsys_buffer *buffer)
-{
-   struct brw_screen *bscreen = brw_screen(screen);
-   struct brw_texture *tex;
-
-   if (templ->target != PIPE_TEXTURE_2D ||
-       templ->last_level != 0 ||
-       templ->depth0 != 1)
-      return NULL;
-
-   if (util_format_is_compressed(templ->format))
-      return NULL;
-
-   tex = CALLOC_STRUCT(brw_texture);
-   if (!tex)
-      return NULL;
-
-   memcpy(&tex->base, templ, sizeof *templ);
-   pipe_reference_init(&tex->base.reference, 1);
-   tex->base.screen = screen;
-
-   /* XXX: cpp vs. blocksize
-    */
-   tex->cpp = util_format_get_blocksize(tex->base.format);
-   tex->tiling = tiling;
-
-   make_empty_list(&tex->views[0]);
-   make_empty_list(&tex->views[1]);
-
-   if (!brw_texture_layout(bscreen, tex))
-      goto fail;
-
-   /* XXX Maybe some more checks? */
-   if ((pitch / tex->cpp) < tex->pitch)
-      goto fail;
-
-   tex->pitch = pitch / tex->cpp;
-
-   tex->bo = buffer;
-
-   /* fix this warning */
-#if 0
-   if (tex->size > buffer->size)
-      goto fail;
-#endif
-
-   tex->ss.ss0.mipmap_layout_mode = BRW_SURFACE_MIPMAPLAYOUT_BELOW;
-   tex->ss.ss0.surface_type = translate_tex_target(tex->base.target);
-   tex->ss.ss0.surface_format = translate_tex_format(tex->base.format);
-   assert(tex->ss.ss0.surface_format != BRW_SURFACEFORMAT_INVALID);
-
-   /* This is ok for all textures with channel width 8bit or less:
-    */
-/*    tex->ss.ss0.data_return_format = BRW_SURFACERETURNFORMAT_S1; */
-
-
-   /* XXX: what happens when tex->bo->offset changes???
-    */
-   tex->ss.ss1.base_addr = 0; /* reloc */
-   tex->ss.ss2.mip_count = tex->base.last_level;
-   tex->ss.ss2.width = tex->base.width0 - 1;
-   tex->ss.ss2.height = tex->base.height0 - 1;
-
-   switch (tex->tiling) {
-   case BRW_TILING_NONE:
-      tex->ss.ss3.tiled_surface = 0;
-      tex->ss.ss3.tile_walk = 0;
-      break;
-   case BRW_TILING_X:
-      tex->ss.ss3.tiled_surface = 1;
-      tex->ss.ss3.tile_walk = BRW_TILEWALK_XMAJOR;
-      break;
-   case BRW_TILING_Y:
-      tex->ss.ss3.tiled_surface = 1;
-      tex->ss.ss3.tile_walk = BRW_TILEWALK_YMAJOR;
-      break;
-   }
-
-   tex->ss.ss3.pitch = (tex->pitch * tex->cpp) - 1;
-   tex->ss.ss3.depth = tex->base.depth0 - 1;
-
-   tex->ss.ss4.min_lod = 0;
-
-   return &tex->base;
-
-fail:
-   FREE(tex);
-   return NULL;
-}
-
 void brw_screen_tex_init( struct brw_screen *brw_screen )
 {
    brw_screen->base.is_format_supported = brw_is_format_supported;
    brw_screen->base.texture_create = brw_texture_create;
+   brw_screen->base.texture_from_handle = brw_texture_from_handle;
+   brw_screen->base.texture_get_handle = brw_texture_get_handle;
    brw_screen->base.texture_destroy = brw_texture_destroy;
-   brw_screen->base.texture_blanket = brw_texture_blanket;
    brw_screen->base.get_tex_transfer = brw_get_tex_transfer;
    brw_screen->base.transfer_map = brw_transfer_map;
    brw_screen->base.transfer_unmap = brw_transfer_unmap;
diff --git a/src/gallium/drivers/i965/brw_state_cache.c b/src/gallium/drivers/i965/brw_state_cache.c
index 85c20076fb8..c911f3997d1 100644
--- a/src/gallium/drivers/i965/brw_state_cache.c
+++ b/src/gallium/drivers/i965/brw_state_cache.c
@@ -213,7 +213,7 @@ brw_upload_cache( struct brw_cache *cache,
 		  void *aux_return,
                   struct brw_winsys_buffer **bo_out)
 {
-   struct brw_cache_item *item = CALLOC_STRUCT(brw_cache_item);
+   struct brw_cache_item *item;
    GLuint hash = hash_key(key, key_size, relocs, nr_relocs);
    GLuint relocs_size = nr_relocs * sizeof relocs[0];
    GLuint aux_size = cache->aux_size[cache_id];
@@ -233,6 +233,7 @@ brw_upload_cache( struct brw_cache *cache,
    if (ret)
       return ret;
 
+   item = CALLOC_STRUCT(brw_cache_item);
 
    /* Set up the memory containing the key, aux_data, and relocs */
    tmp = MALLOC(key_size + aux_size + relocs_size);
diff --git a/src/gallium/drivers/i965/brw_structs.h b/src/gallium/drivers/i965/brw_structs.h
index bf10bc04de7..e97ddeb5e1c 100644
--- a/src/gallium/drivers/i965/brw_structs.h
+++ b/src/gallium/drivers/i965/brw_structs.h
@@ -28,7 +28,7 @@
   * Authors:
   *   Keith Whitwell <[email protected]>
   */
-        
+
 
 #ifndef BRW_STRUCTS_H
 #define BRW_STRUCTS_H
@@ -1149,7 +1149,7 @@ struct brw_vertex_element_state
       GLuint valid:1; 
       GLuint vertex_buffer_index:5; 
    } ve0;
-   
+
    struct
    {
       GLuint dst_offset:8; 
diff --git a/src/gallium/drivers/i965/brw_winsys.h b/src/gallium/drivers/i965/brw_winsys.h
index c82d00f4a47..f30c7f18132 100644
--- a/src/gallium/drivers/i965/brw_winsys.h
+++ b/src/gallium/drivers/i965/brw_winsys.h
@@ -162,6 +162,16 @@ struct brw_winsys_screen {
                                unsigned alignment,
                                struct brw_winsys_buffer **bo_out);
 
+   enum pipe_error (*bo_from_handle)(struct brw_winsys_screen *sws,
+                                     struct winsys_handle *whandle,
+                                     unsigned *stride,
+                                     unsigned *tiling,
+                                     struct brw_winsys_buffer **bo_out);
+
+   enum pipe_error (*bo_get_handle)(struct brw_winsys_buffer *buffer,
+                                    struct winsys_handle *whandle,
+                                    unsigned stride);
+
    /* Destroy a buffer when our refcount goes to zero:
     */
    void (*bo_destroy)(struct brw_winsys_buffer *buffer);
@@ -257,28 +267,6 @@ bo_reference(struct brw_winsys_buffer **ptr, struct brw_winsys_buffer *buf)
 struct pipe_screen *brw_create_screen(struct brw_winsys_screen *iws, unsigned pci_id);
 
 
-/**
- * Get the brw_winsys buffer backing the texture.
- *
- * TODO UGLY
- */
-struct pipe_texture;
-boolean brw_texture_get_winsys_buffer(struct pipe_texture *texture,
-                                      struct brw_winsys_buffer **buffer,
-                                      unsigned *stride);
-
-/**
- * Wrap a brw_winsys buffer with a texture blanket.
- *
- * TODO UGLY
- */
-struct pipe_texture * 
-brw_texture_blanket_winsys_buffer(struct pipe_screen *screen,
-                                  const struct pipe_texture *template,
-                                  unsigned pitch,
-				  unsigned tiling,
-                                  struct brw_winsys_buffer *buffer);
-
 
 /*************************************************************************
  * Cooperative dumping between winsys and driver.  TODO: make this
diff --git a/src/gallium/drivers/i965/brw_wm.c b/src/gallium/drivers/i965/brw_wm.c
index 3724adc040e..7ed2378ec04 100644
--- a/src/gallium/drivers/i965/brw_wm.c
+++ b/src/gallium/drivers/i965/brw_wm.c
@@ -254,10 +254,10 @@ static void brw_wm_populate_key( struct brw_context *brw,
    for (i = 0; i < brw->curr.num_fragment_sampler_views; i++) {
       const struct brw_texture *tex = brw_texture(brw->curr.fragment_sampler_views[i]->texture);
 	 
-      if (tex->base.format == PIPE_FORMAT_YCBCR)
+      if (tex->base.format == PIPE_FORMAT_UYVY)
 	 key->yuvtex_mask |= 1 << i;
 
-      if (tex->base.format == PIPE_FORMAT_YCBCR_REV)
+      if (tex->base.format == PIPE_FORMAT_YUYV)
 	 key->yuvtex_swap_mask |= 1 << i;
 
       /* XXX: shadow texture
diff --git a/src/gallium/drivers/i965/brw_wm_sampler_state.c b/src/gallium/drivers/i965/brw_wm_sampler_state.c
index a4bfa61ab30..3f18062c584 100644
--- a/src/gallium/drivers/i965/brw_wm_sampler_state.c
+++ b/src/gallium/drivers/i965/brw_wm_sampler_state.c
@@ -130,15 +130,14 @@ brw_wm_sampler_update_default_colors(struct brw_context *brw)
       const struct brw_texture *tex = brw_texture(brw->curr.fragment_sampler_views[i]->texture);
       const struct brw_sampler *sampler = brw->curr.sampler[i];
       const float *bc;
-
+      float bordercolor[4] = {
+         sampler->border_color[0],
+         sampler->border_color[0],
+         sampler->border_color[0],
+         sampler->border_color[0]
+      };
+      
       if (util_format_is_depth_or_stencil(tex->base.format)) {
-	 float bordercolor[4] = {
-	    sampler->border_color[0],
-	    sampler->border_color[0],
-	    sampler->border_color[0],
-	    sampler->border_color[0]
-	 };
-         
          bc = bordercolor;
       }
       else {
diff --git a/src/gallium/drivers/identity/id_context.c b/src/gallium/drivers/identity/id_context.c
index 442d851c135..ef5b428161c 100644
--- a/src/gallium/drivers/identity/id_context.c
+++ b/src/gallium/drivers/identity/id_context.c
@@ -377,6 +377,42 @@ identity_delete_vs_state(struct pipe_context *_pipe,
                          vs);
 }
 
+
+static void *
+identity_create_vertex_elements_state(struct pipe_context *_pipe,
+                                      unsigned num_elements,
+                                      const struct pipe_vertex_element *vertex_elements)
+{
+   struct identity_context *id_pipe = identity_context(_pipe);
+   struct pipe_context *pipe = id_pipe->pipe;
+
+   return pipe->create_vertex_elements_state(pipe,
+                                             num_elements,
+                                             vertex_elements);
+}
+
+static void
+identity_bind_vertex_elements_state(struct pipe_context *_pipe,
+                                    void *velems)
+{
+   struct identity_context *id_pipe = identity_context(_pipe);
+   struct pipe_context *pipe = id_pipe->pipe;
+
+   pipe->bind_vertex_elements_state(pipe,
+                                    velems);
+}
+
+static void
+identity_delete_vertex_elements_state(struct pipe_context *_pipe,
+                                      void *velems)
+{
+   struct identity_context *id_pipe = identity_context(_pipe);
+   struct pipe_context *pipe = id_pipe->pipe;
+
+   pipe->delete_vertex_elements_state(pipe,
+                                      velems);
+}
+
 static void
 identity_set_blend_color(struct pipe_context *_pipe,
                          const struct pipe_blend_color *blend_color)
@@ -559,20 +595,6 @@ identity_set_vertex_buffers(struct pipe_context *_pipe,
                             num_buffers,
                             buffers);
 }
-
-static void
-identity_set_vertex_elements(struct pipe_context *_pipe,
-                             unsigned num_elements,
-                             const struct pipe_vertex_element *vertex_elements)
-{
-   struct identity_context *id_pipe = identity_context(_pipe);
-   struct pipe_context *pipe = id_pipe->pipe;
-
-   pipe->set_vertex_elements(pipe,
-                             num_elements,
-                             vertex_elements);
-}
-
 static void
 identity_surface_copy(struct pipe_context *_pipe,
                       struct pipe_surface *_dst,
@@ -769,6 +791,9 @@ identity_context_create(struct pipe_screen *_screen, struct pipe_context *pipe)
    id_pipe->base.create_vs_state = identity_create_vs_state;
    id_pipe->base.bind_vs_state = identity_bind_vs_state;
    id_pipe->base.delete_vs_state = identity_delete_vs_state;
+   id_pipe->base.create_vertex_elements_state = identity_create_vertex_elements_state;
+   id_pipe->base.bind_vertex_elements_state = identity_bind_vertex_elements_state;
+   id_pipe->base.delete_vertex_elements_state = identity_delete_vertex_elements_state;
    id_pipe->base.set_blend_color = identity_set_blend_color;
    id_pipe->base.set_stencil_ref = identity_set_stencil_ref;
    id_pipe->base.set_clip_state = identity_set_clip_state;
@@ -780,7 +805,6 @@ identity_context_create(struct pipe_screen *_screen, struct pipe_context *pipe)
    id_pipe->base.set_fragment_sampler_views = identity_set_fragment_sampler_views;
    id_pipe->base.set_vertex_sampler_views = identity_set_vertex_sampler_views;
    id_pipe->base.set_vertex_buffers = identity_set_vertex_buffers;
-   id_pipe->base.set_vertex_elements = identity_set_vertex_elements;
    id_pipe->base.surface_copy = identity_surface_copy;
    id_pipe->base.surface_fill = identity_surface_fill;
    id_pipe->base.clear = identity_clear;
diff --git a/src/gallium/drivers/identity/id_drm.c b/src/gallium/drivers/identity/id_drm.c
index 12b516b445f..936ccc444a8 100644
--- a/src/gallium/drivers/identity/id_drm.c
+++ b/src/gallium/drivers/identity/id_drm.c
@@ -28,11 +28,11 @@
 #include "state_tracker/drm_api.h"
 
 #include "util/u_memory.h"
-#include "identity/id_drm.h"
-#include "identity/id_screen.h"
-#include "identity/id_public.h"
-#include "identity/id_screen.h"
-#include "identity/id_objects.h"
+#include "id_drm.h"
+#include "id_screen.h"
+#include "id_public.h"
+#include "id_screen.h"
+#include "id_objects.h"
 
 struct identity_drm_api
 {
@@ -63,62 +63,6 @@ identity_drm_create_screen(struct drm_api *_api, int fd,
    return identity_screen_create(screen);
 }
 
-
-static struct pipe_texture *
-identity_drm_texture_from_shared_handle(struct drm_api *_api,
-                                        struct pipe_screen *_screen,
-                                        struct pipe_texture *templ,
-                                        const char *name,
-                                        unsigned stride,
-                                        unsigned handle)
-{
-   struct identity_screen *id_screen = identity_screen(_screen);
-   struct identity_drm_api *id_api = identity_drm_api(_api);
-   struct pipe_screen *screen = id_screen->screen;
-   struct drm_api *api = id_api->api;
-   struct pipe_texture *result;
-
-   result = api->texture_from_shared_handle(api, screen, templ, name, stride, handle);
-
-   result = identity_texture_create(identity_screen(_screen), result);
-
-   return result;
-}
-
-static boolean
-identity_drm_shared_handle_from_texture(struct drm_api *_api,
-                                        struct pipe_screen *_screen,
-                                        struct pipe_texture *_texture,
-                                        unsigned *stride,
-                                        unsigned *handle)
-{
-   struct identity_screen *id_screen = identity_screen(_screen);
-   struct identity_texture *id_texture = identity_texture(_texture);
-   struct identity_drm_api *id_api = identity_drm_api(_api);
-   struct pipe_screen *screen = id_screen->screen;
-   struct pipe_texture *texture = id_texture->texture;
-   struct drm_api *api = id_api->api;
-
-   return api->shared_handle_from_texture(api, screen, texture, stride, handle);
-}
-
-static boolean
-identity_drm_local_handle_from_texture(struct drm_api *_api,
-                                       struct pipe_screen *_screen,
-                                       struct pipe_texture *_texture,
-                                       unsigned *stride,
-                                       unsigned *handle)
-{
-   struct identity_screen *id_screen = identity_screen(_screen);
-   struct identity_texture *id_texture = identity_texture(_texture);
-   struct identity_drm_api *id_api = identity_drm_api(_api);
-   struct pipe_screen *screen = id_screen->screen;
-   struct pipe_texture *texture = id_texture->texture;
-   struct drm_api *api = id_api->api;
-
-   return api->local_handle_from_texture(api, screen, texture, stride, handle);
-}
-
 static void
 identity_drm_destroy(struct drm_api *_api)
 {
@@ -142,10 +86,9 @@ identity_drm_create(struct drm_api *api)
    if (!id_api)
       goto error;
 
+   id_api->base.name = api->name;
+   id_api->base.driver_name = api->driver_name;
    id_api->base.create_screen = identity_drm_create_screen;
-   id_api->base.texture_from_shared_handle = identity_drm_texture_from_shared_handle;
-   id_api->base.shared_handle_from_texture = identity_drm_shared_handle_from_texture;
-   id_api->base.local_handle_from_texture = identity_drm_local_handle_from_texture;
    id_api->base.destroy = identity_drm_destroy;
    id_api->api = api;
 
diff --git a/src/gallium/drivers/identity/id_screen.c b/src/gallium/drivers/identity/id_screen.c
index b85492114a3..b9d0f003d74 100644
--- a/src/gallium/drivers/identity/id_screen.c
+++ b/src/gallium/drivers/identity/id_screen.c
@@ -135,27 +135,40 @@ identity_screen_texture_create(struct pipe_screen *_screen,
 }
 
 static struct pipe_texture *
-identity_screen_texture_blanket(struct pipe_screen *_screen,
-                                const struct pipe_texture *templat,
-                                const unsigned *stride,
-                                struct pipe_buffer *_buffer)
+identity_screen_texture_from_handle(struct pipe_screen *_screen,
+                                    const struct pipe_texture *templ,
+                                    struct winsys_handle *handle)
 {
    struct identity_screen *id_screen = identity_screen(_screen);
-   struct identity_buffer *id_buffer = identity_buffer(_buffer);
    struct pipe_screen *screen = id_screen->screen;
-   struct pipe_buffer *buffer = id_buffer->buffer;
    struct pipe_texture *result;
 
-   result = screen->texture_blanket(screen,
-                                    templat,
-                                    stride,
-                                    buffer);
+   /* TODO trace call */
 
-   if (result)
-      return identity_texture_create(id_screen, result);
-   return NULL;
+   result = screen->texture_from_handle(screen, templ, handle);
+
+   result = identity_texture_create(identity_screen(_screen), result);
+
+   return result;
+}
+
+static boolean
+identity_screen_texture_get_handle(struct pipe_screen *_screen,
+                                   struct pipe_texture *_texture,
+                                   struct winsys_handle *handle)
+{
+   struct identity_screen *id_screen = identity_screen(_screen);
+   struct identity_texture *id_texture = identity_texture(_texture);
+   struct pipe_screen *screen = id_screen->screen;
+   struct pipe_texture *texture = id_texture->texture;
+
+   /* TODO trace call */
+
+   return screen->texture_get_handle(screen, texture, handle);
 }
 
+
+
 static void
 identity_screen_texture_destroy(struct pipe_texture *_texture)
 {
@@ -298,31 +311,6 @@ identity_screen_user_buffer_create(struct pipe_screen *_screen,
    return NULL;
 }
 
-static struct pipe_buffer *
-identity_screen_surface_buffer_create(struct pipe_screen *_screen,
-                                      unsigned width,
-                                      unsigned height,
-                                      enum pipe_format format,
-                                      unsigned usage,
-                                      unsigned tex_usage,
-                                      unsigned *stride)
-{
-   struct identity_screen *id_screen = identity_screen(_screen);
-   struct pipe_screen *screen = id_screen->screen;
-   struct pipe_buffer *result;
-
-   result = screen->surface_buffer_create(screen,
-                                          width,
-                                          height,
-                                          format,
-                                          usage,
-                                          tex_usage,
-                                          stride);
-
-   if (result)
-      return identity_buffer_create(id_screen, result);
-   return NULL;
-}
 
 static void *
 identity_screen_buffer_map(struct pipe_screen *_screen,
@@ -495,7 +483,8 @@ identity_screen_create(struct pipe_screen *screen)
    id_screen->base.is_format_supported = identity_screen_is_format_supported;
    id_screen->base.context_create = identity_screen_context_create;
    id_screen->base.texture_create = identity_screen_texture_create;
-   id_screen->base.texture_blanket = identity_screen_texture_blanket;
+   id_screen->base.texture_from_handle = identity_screen_texture_from_handle;
+   id_screen->base.texture_get_handle = identity_screen_texture_get_handle;
    id_screen->base.texture_destroy = identity_screen_texture_destroy;
    id_screen->base.get_tex_surface = identity_screen_get_tex_surface;
    id_screen->base.tex_surface_destroy = identity_screen_tex_surface_destroy;
@@ -505,7 +494,6 @@ identity_screen_create(struct pipe_screen *screen)
    id_screen->base.transfer_unmap = identity_screen_transfer_unmap;
    id_screen->base.buffer_create = identity_screen_buffer_create;
    id_screen->base.user_buffer_create = identity_screen_user_buffer_create;
-   id_screen->base.surface_buffer_create = identity_screen_surface_buffer_create;
    if (screen->buffer_map)
       id_screen->base.buffer_map = identity_screen_buffer_map;
    if (screen->buffer_map_range)
diff --git a/src/gallium/drivers/llvmpipe/Makefile b/src/gallium/drivers/llvmpipe/Makefile
index 31732514379..41ac1cee72d 100644
--- a/src/gallium/drivers/llvmpipe/Makefile
+++ b/src/gallium/drivers/llvmpipe/Makefile
@@ -37,6 +37,7 @@ C_SOURCES = \
 	lp_surface.c \
 	lp_tex_sample_llvm.c \
 	lp_texture.c \
+	lp_tile_image.c \
 	lp_tile_soa.c
 
 CPP_SOURCES = \
diff --git a/src/gallium/drivers/llvmpipe/README b/src/gallium/drivers/llvmpipe/README
index 72d9f39658f..bf4c9a5727c 100644
--- a/src/gallium/drivers/llvmpipe/README
+++ b/src/gallium/drivers/llvmpipe/README
@@ -49,8 +49,6 @@ To do (probably by this order):
 Requirements
 ============
 
- - Linux
- 
  - A x86 or amd64 processor.  64bit mode is preferred.
  
    Support for sse2 is strongly encouraged.  Support for ssse3, and sse4.1 will
@@ -86,7 +84,7 @@ Building
 
 To build everything on Linux invoke scons as:
 
-  scons debug=yes statetrackers=mesa drivers=trace,llvmpipe winsys=xlib dri=false
+  scons debug=yes statetrackers=mesa drivers=llvmpipe winsys=xlib dri=false
 
 Alternatively, you can build it with GNU make, if you prefer, by invoking it as
 
@@ -96,7 +94,7 @@ but the rest of these instructions assume that scons is used.
 
 For windows is everything the except except the winsys:
 
-  scons debug=yes statetrackers=mesa drivers=trace,llvmpipe winsys=gdi dri=false
+  scons debug=yes statetrackers=mesa drivers=llvmpipe winsys=gdi dri=false
 
 Using
 =====
diff --git a/src/gallium/drivers/llvmpipe/SConscript b/src/gallium/drivers/llvmpipe/SConscript
index a39283e5e86..13c1a13e87a 100644
--- a/src/gallium/drivers/llvmpipe/SConscript
+++ b/src/gallium/drivers/llvmpipe/SConscript
@@ -18,6 +18,13 @@ env.CodeGenerate(
 	command = 'python $SCRIPT $SOURCE > $TARGET'
 )
 
+# XXX: Our dependency scanner only finds depended modules in relative dirs.
+env.Depends('lp_tile_soa.c', [
+    '#src/gallium/auxiliary/util/u_format_parse.py', 
+    '#src/gallium/auxiliary/util/u_format_pack.py', 
+    '#src/gallium/auxiliary/util/u_format_access.py',
+])
+
 llvmpipe = env.ConvenienceLibrary(
 	target = 'llvmpipe',
 	source = [
@@ -52,6 +59,7 @@ llvmpipe = env.ConvenienceLibrary(
 		'lp_surface.c',
 		'lp_tex_sample_llvm.c',
 		'lp_texture.c',
+		'lp_tile_image.c',
 		'lp_tile_soa.c',
 	])
 
diff --git a/src/gallium/drivers/llvmpipe/lp_buffer.c b/src/gallium/drivers/llvmpipe/lp_buffer.c
index 9eda9720818..dab20cb6397 100644
--- a/src/gallium/drivers/llvmpipe/lp_buffer.c
+++ b/src/gallium/drivers/llvmpipe/lp_buffer.c
@@ -30,10 +30,10 @@
 #include "util/u_memory.h"
 #include "util/u_math.h"
 
-#include "lp_winsys.h"
 #include "lp_screen.h"
 #include "lp_buffer.h"
 
+#include "state_tracker/sw_winsys.h"
 
 static void *
 llvmpipe_buffer_map(struct pipe_screen *screen,
diff --git a/src/gallium/drivers/llvmpipe/lp_context.c b/src/gallium/drivers/llvmpipe/lp_context.c
index 0e385f843d1..bf7ed118095 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.c
+++ b/src/gallium/drivers/llvmpipe/lp_context.c
@@ -145,6 +145,10 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv )
    llvmpipe->pipe.bind_vs_state   = llvmpipe_bind_vs_state;
    llvmpipe->pipe.delete_vs_state = llvmpipe_delete_vs_state;
 
+   llvmpipe->pipe.create_vertex_elements_state = llvmpipe_create_vertex_elements_state;
+   llvmpipe->pipe.bind_vertex_elements_state = llvmpipe_bind_vertex_elements_state;
+   llvmpipe->pipe.delete_vertex_elements_state = llvmpipe_delete_vertex_elements_state;
+
    llvmpipe->pipe.set_blend_color = llvmpipe_set_blend_color;
    llvmpipe->pipe.set_stencil_ref = llvmpipe_set_stencil_ref;
    llvmpipe->pipe.set_clip_state = llvmpipe_set_clip_state;
@@ -159,7 +163,6 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv )
    llvmpipe->pipe.set_viewport_state = llvmpipe_set_viewport_state;
 
    llvmpipe->pipe.set_vertex_buffers = llvmpipe_set_vertex_buffers;
-   llvmpipe->pipe.set_vertex_elements = llvmpipe_set_vertex_elements;
 
    llvmpipe->pipe.draw_arrays = llvmpipe_draw_arrays;
    llvmpipe->pipe.draw_elements = llvmpipe_draw_elements;
@@ -185,7 +188,7 @@ llvmpipe_create_context( struct pipe_screen *screen, void *priv )
    if (debug_get_bool_option( "LP_NO_RAST", FALSE ))
       llvmpipe->no_rast = TRUE;
 
-   llvmpipe->setup = lp_setup_create( screen,
+   llvmpipe->setup = lp_setup_create( &llvmpipe->pipe,
                                       llvmpipe->draw );
    if (!llvmpipe->setup)
       goto fail;
diff --git a/src/gallium/drivers/llvmpipe/lp_context.h b/src/gallium/drivers/llvmpipe/lp_context.h
index 1b98e3033db..f83e5ffacf8 100644
--- a/src/gallium/drivers/llvmpipe/lp_context.h
+++ b/src/gallium/drivers/llvmpipe/lp_context.h
@@ -46,6 +46,7 @@ struct lp_fragment_shader;
 struct lp_vertex_shader;
 struct lp_blend_state;
 struct setup_context;
+struct lp_velems_state;
 
 struct llvmpipe_context {
    struct pipe_context pipe;  /**< base class */
@@ -58,6 +59,7 @@ struct llvmpipe_context {
    const struct pipe_rasterizer_state *rasterizer;
    struct lp_fragment_shader *fs;
    const struct lp_vertex_shader *vs;
+   const struct lp_velems_state *velems;
 
    /** Other rendering state */
    struct pipe_blend_color blend_color;
@@ -71,13 +73,11 @@ struct llvmpipe_context {
    struct pipe_sampler_view *vertex_sampler_views[PIPE_MAX_VERTEX_SAMPLERS];
    struct pipe_viewport_state viewport;
    struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
-   struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS];
 
    unsigned num_samplers;
    unsigned num_fragment_sampler_views;
    unsigned num_vertex_samplers;
    unsigned num_vertex_sampler_views;
-   unsigned num_vertex_elements;
    unsigned num_vertex_buffers;
 
    unsigned dirty; /**< Mask of LP_NEW_x flags */
diff --git a/src/gallium/drivers/llvmpipe/lp_jit.c b/src/gallium/drivers/llvmpipe/lp_jit.c
index d3d7e26882d..08c8f93794d 100644
--- a/src/gallium/drivers/llvmpipe/lp_jit.c
+++ b/src/gallium/drivers/llvmpipe/lp_jit.c
@@ -37,6 +37,7 @@
 
 #include "util/u_memory.h"
 #include "util/u_cpu_detect.h"
+#include "gallivm/lp_bld_init.h"
 #include "lp_debug.h"
 #include "lp_screen.h"
 #include "gallivm/lp_bld_intr.h"
@@ -50,12 +51,16 @@ lp_jit_init_globals(struct llvmpipe_screen *screen)
 
    /* struct lp_jit_texture */
    {
-      LLVMTypeRef elem_types[4];
+      LLVMTypeRef elem_types[6];
 
       elem_types[LP_JIT_TEXTURE_WIDTH]  = LLVMInt32Type();
       elem_types[LP_JIT_TEXTURE_HEIGHT] = LLVMInt32Type();
+      elem_types[LP_JIT_TEXTURE_DEPTH] = LLVMInt32Type();
+      elem_types[LP_JIT_TEXTURE_LAST_LEVEL] = LLVMInt32Type();
       elem_types[LP_JIT_TEXTURE_STRIDE] = LLVMInt32Type();
-      elem_types[LP_JIT_TEXTURE_DATA]   = LLVMPointerType(LLVMInt8Type(), 0);
+      elem_types[LP_JIT_TEXTURE_DATA] =
+         LLVMArrayType(LLVMPointerType(LLVMInt8Type(), 0),
+                       LP_MAX_TEXTURE_2D_LEVELS);
 
       texture_type = LLVMStructType(elem_types, Elements(elem_types), 0);
 
@@ -65,6 +70,12 @@ lp_jit_init_globals(struct llvmpipe_screen *screen)
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, height,
                              screen->target, texture_type,
                              LP_JIT_TEXTURE_HEIGHT);
+      LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, depth,
+                             screen->target, texture_type,
+                             LP_JIT_TEXTURE_DEPTH);
+      LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, last_level,
+                             screen->target, texture_type,
+                             LP_JIT_TEXTURE_LAST_LEVEL);
       LP_CHECK_MEMBER_OFFSET(struct lp_jit_texture, stride,
                              screen->target, texture_type,
                              LP_JIT_TEXTURE_STRIDE);
@@ -148,8 +159,7 @@ lp_jit_screen_init(struct llvmpipe_screen *screen)
    util_cpu_caps.has_sse4_1 = 0;
 #endif
 
-   LLVMLinkInJIT();
-   LLVMInitializeNativeTarget();
+   lp_build_init();
 
    screen->module = LLVMModuleCreateWithName("llvmpipe");
 
diff --git a/src/gallium/drivers/llvmpipe/lp_jit.h b/src/gallium/drivers/llvmpipe/lp_jit.h
index 8df3015d4b4..5cc7a12c03b 100644
--- a/src/gallium/drivers/llvmpipe/lp_jit.h
+++ b/src/gallium/drivers/llvmpipe/lp_jit.h
@@ -39,6 +39,7 @@
 #include "gallivm/lp_bld_struct.h"
 
 #include "pipe/p_state.h"
+#include "lp_texture.h"
 
 
 struct llvmpipe_screen;
@@ -48,14 +49,18 @@ struct lp_jit_texture
 {
    uint32_t width;
    uint32_t height;
+   uint32_t depth;
+   uint32_t last_level;
    uint32_t stride;
-   const void *data;
+   const void *data[LP_MAX_TEXTURE_2D_LEVELS];
 };
 
 
 enum {
    LP_JIT_TEXTURE_WIDTH = 0,
    LP_JIT_TEXTURE_HEIGHT,
+   LP_JIT_TEXTURE_DEPTH,
+   LP_JIT_TEXTURE_LAST_LEVEL,
    LP_JIT_TEXTURE_STRIDE,
    LP_JIT_TEXTURE_DATA
 };
diff --git a/src/gallium/drivers/llvmpipe/lp_public.h b/src/gallium/drivers/llvmpipe/lp_public.h
new file mode 100644
index 00000000000..ec6b660b48e
--- /dev/null
+++ b/src/gallium/drivers/llvmpipe/lp_public.h
@@ -0,0 +1,10 @@
+#ifndef LP_PUBLIC_H
+#define LP_PUBLIC_H
+
+struct pipe_screen;
+struct sw_winsys;
+
+struct pipe_screen *
+llvmpipe_create_screen(struct sw_winsys *winsys);
+
+#endif
diff --git a/src/gallium/drivers/llvmpipe/lp_rast.c b/src/gallium/drivers/llvmpipe/lp_rast.c
index 6dbcb3c9b31..dd9a8e8856f 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast.c
+++ b/src/gallium/drivers/llvmpipe/lp_rast.c
@@ -42,128 +42,73 @@
 #include "lp_scene.h"
 
 
-/**
- * Begin the rasterization phase.
- * Map the framebuffer surfaces.  Initialize the 'rast' state.
+/* Begin rasterizing a scene:
  */
 static boolean
 lp_rast_begin( struct lp_rasterizer *rast,
-               const struct pipe_framebuffer_state *fb,
-               boolean write_color,
-               boolean write_zstencil )
+               struct lp_scene *scene )
 {
-   struct pipe_screen *screen = rast->screen;
-   struct pipe_surface *cbuf, *zsbuf;
+   const struct pipe_framebuffer_state *fb = &scene->fb;
+   boolean write_color = fb->nr_cbufs != 0;
+   boolean write_zstencil = fb->zsbuf != NULL;
    int i;
 
-   LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
+   rast->curr_scene = scene;
 
-   util_copy_framebuffer_state(&rast->state.fb, fb);
+   LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
 
+   rast->state.nr_cbufs = scene->fb.nr_cbufs;
    rast->state.write_zstencil = write_zstencil;
    rast->state.write_color = write_color;
-
-   rast->check_for_clipped_tiles = (fb->width % TILE_SIZE != 0 ||
-                                    fb->height % TILE_SIZE != 0);
-
    
-   for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
-      cbuf = rast->state.fb.cbufs[i];
-      if (cbuf) {
-	 rast->cbuf_transfer[i] = screen->get_tex_transfer(rast->screen,
-							   cbuf->texture,
-							   cbuf->face,
-							   cbuf->level,
-							   cbuf->zslice,
-							   PIPE_TRANSFER_READ_WRITE,
-							   0, 0,
-							   cbuf->width, 
-							   cbuf->height);
-	 if (!rast->cbuf_transfer[i])
-	    goto fail;
-
-	 rast->cbuf_map[i] = screen->transfer_map(rast->screen, 
-						  rast->cbuf_transfer[i]);
-	 if (!rast->cbuf_map[i])
-	    goto fail;
-      }
+   for (i = 0; i < rast->state.nr_cbufs; i++) {
+      rast->cbuf[i].map = scene->cbuf_map[i];
+      rast->cbuf[i].format = scene->cbuf_transfer[i]->texture->format;
+      rast->cbuf[i].width = scene->cbuf_transfer[i]->width;
+      rast->cbuf[i].height = scene->cbuf_transfer[i]->height;
+      rast->cbuf[i].stride = scene->cbuf_transfer[i]->stride;
    }
 
-   zsbuf = rast->state.fb.zsbuf;
-   if (zsbuf) {
-      rast->zsbuf_transfer = screen->get_tex_transfer(rast->screen,
-                                                      zsbuf->texture,
-                                                      zsbuf->face,
-                                                      zsbuf->level,
-                                                      zsbuf->zslice,
-                                                      PIPE_TRANSFER_READ_WRITE,
-                                                      0, 0,
-                                                      zsbuf->width,
-						      zsbuf->height);
-      if (!rast->zsbuf_transfer)
-         goto fail;
-
-      rast->zsbuf_map = screen->transfer_map(rast->screen, 
-                                            rast->zsbuf_transfer);
-      if (!rast->zsbuf_map)
-	 goto fail;
+   if (write_zstencil) {
+      rast->zsbuf.map = scene->zsbuf_map;
+      rast->zsbuf.stride = scene->zsbuf_transfer->stride;
+      rast->zsbuf.blocksize = 
+         util_format_get_blocksize(scene->zsbuf_transfer->texture->format);
    }
 
+   lp_scene_bin_iter_begin( scene );
+   
    return TRUE;
-
-fail:
-   /* Unmap and release transfers?
-    */
-   return FALSE;
 }
 
 
-/**
- * Finish the rasterization phase.
- * Unmap framebuffer surfaces.
- */
 static void
 lp_rast_end( struct lp_rasterizer *rast )
 {
-   struct pipe_screen *screen = rast->screen;
-   unsigned i;
-
-   for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
-      if (rast->cbuf_map[i]) 
-	 screen->transfer_unmap(screen, rast->cbuf_transfer[i]);
-
-      if (rast->cbuf_transfer[i])
-	 screen->tex_transfer_destroy(rast->cbuf_transfer[i]);
-
-      rast->cbuf_transfer[i] = NULL;
-      rast->cbuf_map[i] = NULL;
-   }
+   int i;
 
-   if (rast->zsbuf_map) 
-      screen->transfer_unmap(screen, rast->zsbuf_transfer);
+   lp_scene_reset( rast->curr_scene );
 
-   if (rast->zsbuf_transfer)
-      screen->tex_transfer_destroy(rast->zsbuf_transfer);
+   for (i = 0; i < rast->state.nr_cbufs; i++)
+      rast->cbuf[i].map = NULL;
 
-   rast->zsbuf_transfer = NULL;
-   rast->zsbuf_map = NULL;
+   rast->zsbuf.map = NULL;
+   rast->curr_scene = NULL;
 }
 
-
 /**
  * Begining rasterization of a tile.
  * \param x  window X position of the tile, in pixels
  * \param y  window Y position of the tile, in pixels
  */
 static void
-lp_rast_start_tile( struct lp_rasterizer *rast,
-                    unsigned thread_index,
-                    unsigned x, unsigned y )
+lp_rast_start_tile(struct lp_rasterizer_task *task,
+                   unsigned x, unsigned y)
 {
    LP_DBG(DEBUG_RAST, "%s %d,%d\n", __FUNCTION__, x, y);
 
-   rast->tasks[thread_index].x = x;
-   rast->tasks[thread_index].y = y;
+   task->x = x;
+   task->y = y;
 }
 
 
@@ -171,12 +116,13 @@ lp_rast_start_tile( struct lp_rasterizer *rast,
  * Clear the rasterizer's current color tile.
  * This is a bin command called during bin processing.
  */
-void lp_rast_clear_color( struct lp_rasterizer *rast,
-                          unsigned thread_index,
-                          const union lp_rast_cmd_arg arg )
+void
+lp_rast_clear_color(struct lp_rasterizer_task *task,
+                    const union lp_rast_cmd_arg arg)
 {
+   struct lp_rasterizer *rast = task->rast;
    const uint8_t *clear_color = arg.clear_color;
-   uint8_t **color_tile = rast->tasks[thread_index].tile.color;
+   uint8_t **color_tile = task->tile.color;
    unsigned i;
 
    LP_DBG(DEBUG_RAST, "%s 0x%x,0x%x,0x%x,0x%x\n", __FUNCTION__, 
@@ -189,7 +135,7 @@ void lp_rast_clear_color( struct lp_rasterizer *rast,
        clear_color[1] == clear_color[2] &&
        clear_color[2] == clear_color[3]) {
       /* clear to grayscale value {x, x, x, x} */
-      for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
+      for (i = 0; i < rast->state.nr_cbufs; i++) {
 	 memset(color_tile[i], clear_color[0], TILE_SIZE * TILE_SIZE * 4);
       }
    }
@@ -200,7 +146,7 @@ void lp_rast_clear_color( struct lp_rasterizer *rast,
        * works.
        */
       const unsigned chunk = TILE_SIZE / 4;
-      for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
+      for (i = 0; i < rast->state.nr_cbufs; i++) {
          uint8_t *c = color_tile[i];
          unsigned j;
          for (j = 0; j < 4 * TILE_SIZE; j++) {
@@ -225,24 +171,24 @@ void lp_rast_clear_color( struct lp_rasterizer *rast,
  * Clear the rasterizer's current z/stencil tile.
  * This is a bin command called during bin processing.
  */
-void lp_rast_clear_zstencil( struct lp_rasterizer *rast,
-                             unsigned thread_index,
-                             const union lp_rast_cmd_arg arg)
+void
+lp_rast_clear_zstencil(struct lp_rasterizer_task *task,
+                       const union lp_rast_cmd_arg arg)
 {
-   struct lp_rasterizer_task *task = &rast->tasks[thread_index];
+   struct lp_rasterizer *rast = task->rast;
    const unsigned tile_x = task->x;
    const unsigned tile_y = task->y;
-   const unsigned height = TILE_SIZE/TILE_VECTOR_HEIGHT;
-   const unsigned width = TILE_SIZE*TILE_VECTOR_HEIGHT;
-   unsigned block_size = util_format_get_blocksize(rast->zsbuf_transfer->texture->format);
+   const unsigned height = TILE_SIZE / TILE_VECTOR_HEIGHT;
+   const unsigned width = TILE_SIZE * TILE_VECTOR_HEIGHT;
+   unsigned block_size = rast->zsbuf.blocksize;
    uint8_t *dst;
-   unsigned dst_stride = rast->zsbuf_transfer->stride*TILE_VECTOR_HEIGHT;
+   unsigned dst_stride = rast->zsbuf.stride * TILE_VECTOR_HEIGHT;
    unsigned i, j;
 
    LP_DBG(DEBUG_RAST, "%s 0x%x\n", __FUNCTION__, arg.clear_zstencil);
 
-   assert(rast->zsbuf_map);
-   if (!rast->zsbuf_map)
+   assert(rast->zsbuf.map);
+   if (!rast->zsbuf.map)
       return;
 
    LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
@@ -278,8 +224,8 @@ void lp_rast_clear_zstencil( struct lp_rasterizer *rast,
       }
       break;
    default:
-         assert(0);
-         break;
+      assert(0);
+      break;
    }
 }
 
@@ -288,55 +234,42 @@ void lp_rast_clear_zstencil( struct lp_rasterizer *rast,
  * Load tile color from the framebuffer surface.
  * This is a bin command called during bin processing.
  */
-void lp_rast_load_color( struct lp_rasterizer *rast,
-                         unsigned thread_index,
-                         const union lp_rast_cmd_arg arg)
+void
+lp_rast_load_color(struct lp_rasterizer_task *task,
+                   const union lp_rast_cmd_arg arg)
 {
-   struct lp_rasterizer_task *task = &rast->tasks[thread_index];
-   const unsigned x = task->x;
-   const unsigned y = task->y;
+   struct lp_rasterizer *rast = task->rast;
+   const unsigned x = task->x, y = task->y;
    unsigned i;
 
    LP_DBG(DEBUG_RAST, "%s at %u, %u\n", __FUNCTION__, x, y);
 
-   for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
-      struct pipe_transfer *transfer = rast->cbuf_transfer[i];
-      int w = TILE_SIZE;
-      int h = TILE_SIZE;
-
-      if (x >= transfer->width)
-	 continue;
-
-      if (y >= transfer->height)
+   for (i = 0; i < rast->state.nr_cbufs; i++) {
+      if (x >= rast->cbuf[i].width || y >= rast->cbuf[i].height)
 	 continue;
 
-      assert(w >= 0);
-      assert(h >= 0);
-      assert(w <= TILE_SIZE);
-      assert(h <= TILE_SIZE);
-
-      lp_tile_read_4ub(transfer->texture->format,
+      lp_tile_read_4ub(rast->cbuf[i].format,
 		       task->tile.color[i],
-		       rast->cbuf_map[i], 
-		       transfer->stride,
+		       rast->cbuf[i].map, 
+		       rast->cbuf[i].stride,
 		       x, y,
-		       w, h);
+		       TILE_SIZE, TILE_SIZE);
 
       LP_COUNT(nr_color_tile_load);
    }
 }
 
 
-void lp_rast_set_state( struct lp_rasterizer *rast,
-                        unsigned thread_index,
-                        const union lp_rast_cmd_arg arg )
+void
+lp_rast_set_state(struct lp_rasterizer_task *task,
+                  const union lp_rast_cmd_arg arg)
 {
    const struct lp_rast_state *state = arg.set_state;
 
    LP_DBG(DEBUG_RAST, "%s %p\n", __FUNCTION__, (void *) state);
 
    /* just set the current state pointer for this rasterizer */
-   rast->tasks[thread_index].current_state = state;
+   task->current_state = state;
 }
 
 
@@ -346,16 +279,15 @@ void lp_rast_set_state( struct lp_rasterizer *rast,
  * completely contained inside a triangle.
  * This is a bin command called during bin processing.
  */
-void lp_rast_shade_tile( struct lp_rasterizer *rast,
-                         unsigned thread_index,
-                         const union lp_rast_cmd_arg arg )
+void
+lp_rast_shade_tile(struct lp_rasterizer_task *task,
+                   const union lp_rast_cmd_arg arg)
 {
-   struct lp_rasterizer_task *task = &rast->tasks[thread_index];
+   struct lp_rasterizer *rast = task->rast;
    const struct lp_rast_state *state = task->current_state;
    struct lp_rast_tile *tile = &task->tile;
    const struct lp_rast_shader_inputs *inputs = arg.shade_tile;
-   const unsigned tile_x = task->x;
-   const unsigned tile_y = task->y;
+   const unsigned tile_x = task->x, tile_y = task->y;
    unsigned x, y;
 
    LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
@@ -371,7 +303,7 @@ void lp_rast_shade_tile( struct lp_rasterizer *rast,
          block_offset = ((y / 4) * (16 * 16) + (x / 4) * 16);
 
          /* color buffer */
-         for (i = 0; i < rast->state.fb.nr_cbufs; i++)
+         for (i = 0; i < rast->state.nr_cbufs; i++)
             color[i] = tile->color[i] + 4 * block_offset;
 
          /* depth buffer */
@@ -396,14 +328,13 @@ void lp_rast_shade_tile( struct lp_rasterizer *rast,
  * Compute shading for a 4x4 block of pixels.
  * This is a bin command called during bin processing.
  */
-void lp_rast_shade_quads( struct lp_rasterizer *rast,
-                          unsigned thread_index,
+void lp_rast_shade_quads( struct lp_rasterizer_task *task,
                           const struct lp_rast_shader_inputs *inputs,
                           unsigned x, unsigned y,
                           int32_t c1, int32_t c2, int32_t c3)
 {
-   struct lp_rasterizer_task *task = &rast->tasks[thread_index];
    const struct lp_rast_state *state = task->current_state;
+   struct lp_rasterizer *rast = task->rast;
    struct lp_rast_tile *tile = &task->tile;
    uint8_t *color[PIPE_MAX_COLOR_BUFS];
    void *depth;
@@ -411,7 +342,6 @@ void lp_rast_shade_quads( struct lp_rasterizer *rast,
    unsigned ix, iy;
    int block_offset;
 
-#ifdef DEBUG
    assert(state);
 
    /* Sanity checks */
@@ -420,7 +350,6 @@ void lp_rast_shade_quads( struct lp_rasterizer *rast,
 
    assert((x % 4) == 0);
    assert((y % 4) == 0);
-#endif
 
    ix = x % TILE_SIZE;
    iy = y % TILE_SIZE;
@@ -429,22 +358,19 @@ void lp_rast_shade_quads( struct lp_rasterizer *rast,
    block_offset = ((iy / 4) * (16 * 16) + (ix / 4) * 16);
 
    /* color buffer */
-   for (i = 0; i < rast->state.fb.nr_cbufs; i++)
+   for (i = 0; i < rast->state.nr_cbufs; i++)
       color[i] = tile->color[i] + 4 * block_offset;
 
    /* depth buffer */
    depth = lp_rast_depth_pointer(rast, x, y);
 
 
-
-#ifdef DEBUG
    assert(lp_check_alignment(tile->color[0], 16));
    assert(lp_check_alignment(state->jit_context.blend_color, 16));
 
    assert(lp_check_alignment(inputs->step[0], 16));
    assert(lp_check_alignment(inputs->step[1], 16));
    assert(lp_check_alignment(inputs->step[2], 16));
-#endif
 
    /* run shader */
    state->jit_function[1]( &state->jit_context,
@@ -515,66 +441,48 @@ outline_subtiles(uint8_t *tile)
 /**
  * Write the rasterizer's color tile to the framebuffer.
  */
-static void lp_rast_store_color( struct lp_rasterizer *rast,
-                                 unsigned thread_index)
+static void
+lp_rast_store_color(struct lp_rasterizer_task *task)
 {
-   struct lp_rasterizer_task *task = &rast->tasks[thread_index];
-   const unsigned x = task->x;
-   const unsigned y = task->y;
+   struct lp_rasterizer *rast = task->rast;
+   const unsigned x = task->x, y = task->y;
    unsigned i;
 
-   for (i = 0; i < rast->state.fb.nr_cbufs; i++) {
-      struct pipe_transfer *transfer = rast->cbuf_transfer[i];
-      int w = TILE_SIZE;
-      int h = TILE_SIZE;
-
-      if (x >= transfer->width)
+   for (i = 0; i < rast->state.nr_cbufs; i++) {
+      if (x >= rast->cbuf[i].width)
 	 continue;
 
-      if (y >= transfer->height)
+      if (y >= rast->cbuf[i].height)
 	 continue;
 
-      LP_DBG(DEBUG_RAST, "%s [%u] %d,%d %dx%d\n", __FUNCTION__,
-	     thread_index, x, y, w, h);
+      LP_DBG(DEBUG_RAST, "%s [%u] %d,%d\n", __FUNCTION__,
+	     task->thread_index, x, y);
 
       if (LP_DEBUG & DEBUG_SHOW_SUBTILES)
          outline_subtiles(task->tile.color[i]);
       else if (LP_DEBUG & DEBUG_SHOW_TILES)
          outline_tile(task->tile.color[i]);
 
-      lp_tile_write_4ub(transfer->texture->format,
+      lp_tile_write_4ub(rast->cbuf[i].format,
 			task->tile.color[i],
-			rast->cbuf_map[i], 
-			transfer->stride,
+			rast->cbuf[i].map, 
+			rast->cbuf[i].stride,
 			x, y,
-			w, h);
+			TILE_SIZE, TILE_SIZE);
 
       LP_COUNT(nr_color_tile_store);
    }
 }
 
 
-/**
- * Write the rasterizer's tiles to the framebuffer.
- */
-static void
-lp_rast_end_tile( struct lp_rasterizer *rast,
-                  unsigned thread_index )
-{
-   LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
-
-   if (rast->state.write_color)
-      lp_rast_store_color(rast, thread_index);
-}
-
 
 /**
  * Signal on a fence.  This is called during bin execution/rasterization.
  * Called per thread.
  */
-void lp_rast_fence( struct lp_rasterizer *rast,
-                    unsigned thread_index,
-                    const union lp_rast_cmd_arg arg )
+void
+lp_rast_fence(struct lp_rasterizer_task *task,
+              const union lp_rast_cmd_arg arg)
 {
    struct lp_fence *fence = arg.fence;
 
@@ -592,20 +500,6 @@ void lp_rast_fence( struct lp_rasterizer *rast,
 }
 
 
-/**
- * When all the threads are done rasterizing a scene, one thread will
- * call this function to reset the scene and put it onto the empty queue.
- */
-static void
-release_scene( struct lp_rasterizer *rast,
-	       struct lp_scene *scene )
-{
-   util_unreference_framebuffer_state( &scene->fb );
-
-   lp_scene_reset( scene );
-   lp_scene_enqueue( rast->empty_scenes, scene );
-   rast->curr_scene = NULL;
-}
 
 
 /**
@@ -615,25 +509,31 @@ release_scene( struct lp_rasterizer *rast,
  * Called per thread.
  */
 static void
-rasterize_bin( struct lp_rasterizer *rast,
-               unsigned thread_index,
-               const struct cmd_bin *bin,
-               int x, int y)
+rasterize_bin(struct lp_rasterizer_task *task,
+              const struct cmd_bin *bin,
+              int x, int y)
 {
    const struct cmd_block_list *commands = &bin->commands;
    struct cmd_block *block;
    unsigned k;
 
-   lp_rast_start_tile( rast, thread_index, x, y );
+   lp_rast_start_tile( task, x * TILE_SIZE, y * TILE_SIZE );
 
    /* simply execute each of the commands in the block list */
    for (block = commands->head; block; block = block->next) {
       for (k = 0; k < block->count; k++) {
-         block->cmd[k]( rast, thread_index, block->arg[k] );
+         block->cmd[k]( task, block->arg[k] );
       }
    }
 
-   lp_rast_end_tile( rast, thread_index );
+   /* Write the rasterizer's tiles to the framebuffer.
+    */
+   if (task->rast->state.write_color)
+      lp_rast_store_color(task);
+
+   /* Free data for this bin.
+    */
+   lp_scene_bin_reset( task->rast->curr_scene, x, y);
 }
 
 
@@ -717,10 +617,8 @@ is_empty_bin( const struct cmd_bin *bin )
  * Called per thread.
  */
 static void
-rasterize_scene( struct lp_rasterizer *rast,
-                unsigned thread_index,
-                struct lp_scene *scene,
-                bool write_depth )
+rasterize_scene(struct lp_rasterizer_task *task,
+                struct lp_scene *scene)
 {
    /* loop over scene bins, rasterize each */
 #if 0
@@ -728,9 +626,8 @@ rasterize_scene( struct lp_rasterizer *rast,
       unsigned i, j;
       for (i = 0; i < scene->tiles_x; i++) {
          for (j = 0; j < scene->tiles_y; j++) {
-            struct cmd_bin *bin = lp_get_bin(scene, i, j);
-            rasterize_bin( rast, thread_index,
-                           bin, i * TILE_SIZE, j * TILE_SIZE );
+            struct cmd_bin *bin = lp_scene_get_bin(scene, i, j);
+            rasterize_bin(task, bin, i, j);
          }
       }
    }
@@ -742,7 +639,7 @@ rasterize_scene( struct lp_rasterizer *rast,
       assert(scene);
       while ((bin = lp_scene_bin_iter_next(scene, &x, &y))) {
          if (!is_empty_bin( bin ))
-            rasterize_bin( rast, thread_index, bin, x * TILE_SIZE, y * TILE_SIZE);
+            rasterize_bin(task, bin, x, y);
       }
    }
 #endif
@@ -753,44 +650,20 @@ rasterize_scene( struct lp_rasterizer *rast,
  * Called by setup module when it has something for us to render.
  */
 void
-lp_rasterize_scene( struct lp_rasterizer *rast,
-                   struct lp_scene *scene,
-                   const struct pipe_framebuffer_state *fb,
-                   bool write_depth )
+lp_rast_queue_scene( struct lp_rasterizer *rast,
+                     struct lp_scene *scene)
 {
-   boolean debug = false;
-
    LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
 
-   if (debug) {
-      unsigned x, y;
-      debug_printf("rasterize scene:\n");
-      debug_printf("  data size: %u\n", lp_scene_data_size(scene));
-      for (y = 0; y < scene->tiles_y; y++) {
-         for (x = 0; x < scene->tiles_x; x++) {
-            debug_printf("  bin %u, %u size: %u\n", x, y,
-                         lp_scene_bin_size(scene, x, y));
-         }
-      }
-   }
-
-   /* save framebuffer state in the bin */
-   util_copy_framebuffer_state(&scene->fb, fb);
-   scene->write_depth = write_depth;
-
    if (rast->num_threads == 0) {
       /* no threading */
 
-      lp_rast_begin( rast, fb,
-                     fb->nr_cbufs != 0, /* always write color if cbufs present */
-                     fb->zsbuf != NULL && write_depth );
+      lp_rast_begin( rast, scene );
 
-      lp_scene_bin_iter_begin( scene );
-      rasterize_scene( rast, 0, scene, write_depth );
+      rasterize_scene( &rast->tasks[0], scene );
 
-      release_scene( rast, scene );
-
-      lp_rast_end( rast );
+      lp_scene_reset( scene );
+      rast->curr_scene = NULL;
    }
    else {
       /* threaded rendering! */
@@ -802,14 +675,26 @@ lp_rasterize_scene( struct lp_rasterizer *rast,
       for (i = 0; i < rast->num_threads; i++) {
          pipe_semaphore_signal(&rast->tasks[i].work_ready);
       }
+   }
+
+   LP_DBG(DEBUG_SETUP, "%s done \n", __FUNCTION__);
+}
+
+
+void
+lp_rast_finish( struct lp_rasterizer *rast )
+{
+   if (rast->num_threads == 0) {
+      /* nothing to do */
+   }
+   else {
+      int i;
 
       /* wait for work to complete */
       for (i = 0; i < rast->num_threads; i++) {
          pipe_semaphore_wait(&rast->tasks[i].work_done);
       }
    }
-
-   LP_DBG(DEBUG_SETUP, "%s done \n", __FUNCTION__);
 }
 
 
@@ -832,24 +717,16 @@ static PIPE_THREAD_ROUTINE( thread_func, init_data )
          debug_printf("thread %d waiting for work\n", task->thread_index);
       pipe_semaphore_wait(&task->work_ready);
 
+      if (rast->exit_flag)
+         break;
+
       if (task->thread_index == 0) {
          /* thread[0]:
           *  - get next scene to rasterize
           *  - map the framebuffer surfaces
           */
-         const struct pipe_framebuffer_state *fb;
-         boolean write_depth;
-
-         rast->curr_scene = lp_scene_dequeue( rast->full_scenes, TRUE );
-
-         lp_scene_bin_iter_begin( rast->curr_scene );
-
-         fb = &rast->curr_scene->fb;
-         write_depth = rast->curr_scene->write_depth;
-
-         lp_rast_begin( rast, fb,
-                        fb->nr_cbufs != 0,
-                        fb->zsbuf != NULL && write_depth );
+         lp_rast_begin( rast, 
+                        lp_scene_dequeue( rast->full_scenes, TRUE ) );
       }
 
       /* Wait for all threads to get here so that threads[1+] don't
@@ -860,26 +737,23 @@ static PIPE_THREAD_ROUTINE( thread_func, init_data )
       /* do work */
       if (debug)
          debug_printf("thread %d doing work\n", task->thread_index);
-      rasterize_scene(rast, 
-		     task->thread_index,
-                     rast->curr_scene, 
-		     rast->curr_scene->write_depth);
+
+      rasterize_scene(task,
+                      rast->curr_scene);
       
       /* wait for all threads to finish with this scene */
       pipe_barrier_wait( &rast->barrier );
 
+      /* XXX: shouldn't be necessary:
+       */
       if (task->thread_index == 0) {
-         /* thread[0]:
-          * - release the scene object
-          * - unmap the framebuffer surfaces
-          */
-         release_scene( rast, rast->curr_scene );
          lp_rast_end( rast );
       }
 
       /* signal done with work */
       if (debug)
          debug_printf("thread %d done working\n", task->thread_index);
+
       pipe_semaphore_signal(&task->work_done);
    }
 
@@ -922,7 +796,7 @@ create_rast_threads(struct lp_rasterizer *rast)
  *               processing them.
  */
 struct lp_rasterizer *
-lp_rast_create( struct pipe_screen *screen, struct lp_scene_queue *empty )
+lp_rast_create( void )
 {
    struct lp_rasterizer *rast;
    unsigned i, cbuf;
@@ -931,9 +805,6 @@ lp_rast_create( struct pipe_screen *screen, struct lp_scene_queue *empty )
    if(!rast)
       return NULL;
 
-   rast->screen = screen;
-
-   rast->empty_scenes = empty;
    rast->full_scenes = lp_scene_queue_create();
 
    for (i = 0; i < Elements(rast->tasks); i++) {
@@ -961,13 +832,31 @@ void lp_rast_destroy( struct lp_rasterizer *rast )
 {
    unsigned i, cbuf;
 
-   util_unreference_framebuffer_state(&rast->state.fb);
-
    for (i = 0; i < Elements(rast->tasks); i++) {
       for (cbuf = 0; cbuf < PIPE_MAX_COLOR_BUFS; cbuf++ )
 	 align_free(rast->tasks[i].tile.color[cbuf]);
    }
 
+   /* Set exit_flag and signal each thread's work_ready semaphore.
+    * Each thread will be woken up, notice that the exit_flag is set and
+    * break out of its main loop.  The thread will then exit.
+    */
+   rast->exit_flag = TRUE;
+   for (i = 0; i < rast->num_threads; i++) {
+      pipe_semaphore_signal(&rast->tasks[i].work_ready);
+   }
+
+   /* Wait for threads to terminate before cleaning up per-thread data */
+   for (i = 0; i < rast->num_threads; i++) {
+      pipe_thread_wait(rast->threads[i]);
+   }
+
+   /* Clean up per-thread data */
+   for (i = 0; i < rast->num_threads; i++) {
+      pipe_semaphore_destroy(&rast->tasks[i].work_ready);
+      pipe_semaphore_destroy(&rast->tasks[i].work_done);
+   }
+
    /* for synchronizing rasterization threads */
    pipe_barrier_destroy( &rast->barrier );
 
diff --git a/src/gallium/drivers/llvmpipe/lp_rast.h b/src/gallium/drivers/llvmpipe/lp_rast.h
index 875f18e0c0c..dc5fc5fc7d6 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast.h
+++ b/src/gallium/drivers/llvmpipe/lp_rast.h
@@ -43,16 +43,17 @@
 
 struct lp_rasterizer;
 struct lp_scene;
-struct lp_scene_queue;
 struct lp_fence;
 struct cmd_bin;
-struct pipe_screen;
 
 /** For sub-pixel positioning */
 #define FIXED_ORDER 4
 #define FIXED_ONE (1<<FIXED_ORDER)
 
 
+struct lp_rasterizer_task;
+
+
 /**
  * Rasterization state.
  * Objects of this type are put into the shared data bin and pointed
@@ -97,6 +98,10 @@ struct lp_rast_shader_inputs {
  * Objects of this type are put into the setup_context::data buffer.
  */
 struct lp_rast_triangle {
+#ifdef DEBUG
+   float v[3][2];
+#endif
+
    /* one-pixel sized trivial accept offsets for each plane */
    int ei1;                   
    int ei2;
@@ -126,18 +131,21 @@ struct lp_rast_triangle {
 
 
 
-struct lp_rasterizer *lp_rast_create( struct pipe_screen *screen,
-                                      struct lp_scene_queue *empty );
+struct lp_rasterizer *
+lp_rast_create( void );
 
-void lp_rast_destroy( struct lp_rasterizer * );
+void
+lp_rast_destroy( struct lp_rasterizer * );
 
-unsigned lp_rast_get_num_threads( struct lp_rasterizer * );
+unsigned
+lp_rast_get_num_threads( struct lp_rasterizer * );
 
-void lp_rasterize_scene( struct lp_rasterizer *rast,
-			 struct lp_scene *scene,
-			 const struct pipe_framebuffer_state *fb,
-			 bool write_depth );
+void 
+lp_rast_queue_scene( struct lp_rasterizer *rast,
+                     struct lp_scene *scene );
 
+void
+lp_rast_finish( struct lp_rasterizer *rast );
 
 
 union lp_rast_cmd_arg {
@@ -201,32 +209,25 @@ lp_rast_arg_null( void )
  * the bins are executed.
  */
 
-void lp_rast_clear_color( struct lp_rasterizer *, 
-                          unsigned thread_index,
+void lp_rast_clear_color( struct lp_rasterizer_task *, 
                           const union lp_rast_cmd_arg );
 
-void lp_rast_clear_zstencil( struct lp_rasterizer *, 
-                             unsigned thread_index,
+void lp_rast_clear_zstencil( struct lp_rasterizer_task *, 
                              const union lp_rast_cmd_arg );
 
-void lp_rast_load_color( struct lp_rasterizer *, 
-                         unsigned thread_index,
+void lp_rast_load_color( struct lp_rasterizer_task *, 
                          const union lp_rast_cmd_arg );
 
-void lp_rast_set_state( struct lp_rasterizer *, 
-                        unsigned thread_index,
+void lp_rast_set_state( struct lp_rasterizer_task *, 
                         const union lp_rast_cmd_arg );
 
-void lp_rast_triangle( struct lp_rasterizer *, 
-                       unsigned thread_index,
+void lp_rast_triangle( struct lp_rasterizer_task *, 
                        const union lp_rast_cmd_arg );
 
-void lp_rast_shade_tile( struct lp_rasterizer *,
-                         unsigned thread_index,
+void lp_rast_shade_tile( struct lp_rasterizer_task *,
                          const union lp_rast_cmd_arg );
 
-void lp_rast_fence( struct lp_rasterizer *,
-                    unsigned thread_index,
+void lp_rast_fence( struct lp_rasterizer_task *,
                     const union lp_rast_cmd_arg );
 
 #endif
diff --git a/src/gallium/drivers/llvmpipe/lp_rast_priv.h b/src/gallium/drivers/llvmpipe/lp_rast_priv.h
index 5c5497e0929..39bf2c25879 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast_priv.h
+++ b/src/gallium/drivers/llvmpipe/lp_rast_priv.h
@@ -38,8 +38,6 @@
 #define MAX_THREADS 8  /* XXX probably temporary here */
 
 
-struct pipe_transfer;
-struct pipe_screen;
 struct lp_rasterizer;
 
 
@@ -82,19 +80,26 @@ struct lp_rasterizer_task
  */
 struct lp_rasterizer
 {
-   boolean clipped_tile;
-   boolean check_for_clipped_tiles;
+   boolean exit_flag;
 
    /* Framebuffer stuff
     */
-   struct pipe_screen *screen;
-   struct pipe_transfer *cbuf_transfer[PIPE_MAX_COLOR_BUFS];
-   struct pipe_transfer *zsbuf_transfer;
-   void *cbuf_map[PIPE_MAX_COLOR_BUFS];
-   uint8_t *zsbuf_map;
+   struct {
+      void *map;
+      unsigned stride;
+      unsigned width;
+      unsigned height;
+      enum pipe_format format;
+   } cbuf[PIPE_MAX_COLOR_BUFS];
+
+   struct {
+      uint8_t *map;
+      unsigned stride;
+      unsigned blocksize;
+   } zsbuf;
 
    struct {
-      struct pipe_framebuffer_state fb;
+      unsigned nr_cbufs;
       boolean write_color;
       boolean write_zstencil;
       unsigned clear_color;
@@ -104,7 +109,14 @@ struct lp_rasterizer
 
    /** The incoming queue of scenes ready to rasterize */
    struct lp_scene_queue *full_scenes;
-   /** The outgoing queue of processed scenes to return to setup modulee */
+
+   /**
+    * The outgoing queue of processed scenes to return to setup module
+    *
+    * XXX: while scenes are per-context but the rasterizer is
+    * (potentially) shared, these empty scenes should be returned to
+    * the context which created them rather than retained here.
+    */
    struct lp_scene_queue *empty_scenes;
 
    /** The scene currently being rasterized by the threads */
@@ -121,8 +133,7 @@ struct lp_rasterizer
 };
 
 
-void lp_rast_shade_quads( struct lp_rasterizer *rast,
-                          unsigned thread_index,
+void lp_rast_shade_quads( struct lp_rasterizer_task *task,
                           const struct lp_rast_shader_inputs *inputs,
                           unsigned x, unsigned y,
                           int32_t c1, int32_t c2, int32_t c3);
@@ -137,17 +148,18 @@ lp_rast_depth_pointer( struct lp_rasterizer *rast,
                        unsigned x, unsigned y )
 {
    void * depth;
+
    assert((x % TILE_VECTOR_WIDTH) == 0);
    assert((y % TILE_VECTOR_HEIGHT) == 0);
-   if(!rast->zsbuf_map)
+
+   if (!rast->zsbuf.map)
       return NULL;
-   assert(rast->zsbuf_transfer);
-   depth = rast->zsbuf_map +
-           y*rast->zsbuf_transfer->stride +
-           TILE_VECTOR_HEIGHT*x*util_format_get_blocksize(rast->zsbuf_transfer->texture->format);
-#ifdef DEBUG
+
+   depth = (rast->zsbuf.map +
+            rast->zsbuf.stride * y +
+            rast->zsbuf.blocksize * x * TILE_VECTOR_HEIGHT);
+
    assert(lp_check_alignment(depth, 16));
-#endif
    return depth;
 }
 
@@ -159,13 +171,13 @@ lp_rast_depth_pointer( struct lp_rasterizer *rast,
  * \param x, y location of 4x4 block in window coords
  */
 static INLINE void
-lp_rast_shade_quads_all( struct lp_rasterizer *rast,
-                         unsigned thread_index,
+lp_rast_shade_quads_all( struct lp_rasterizer_task *task,
                          const struct lp_rast_shader_inputs *inputs,
                          unsigned x, unsigned y )
 {
-   const struct lp_rast_state *state = rast->tasks[thread_index].current_state;
-   struct lp_rast_tile *tile = &rast->tasks[thread_index].tile;
+   struct lp_rasterizer *rast = task->rast;
+   const struct lp_rast_state *state = task->current_state;
+   struct lp_rast_tile *tile = &task->tile;
    const unsigned ix = x % TILE_SIZE, iy = y % TILE_SIZE;
    uint8_t *color[PIPE_MAX_COLOR_BUFS];
    void *depth;
@@ -175,7 +187,7 @@ lp_rast_shade_quads_all( struct lp_rasterizer *rast,
    block_offset = (iy / 4) * (16 * 16) + (ix / 4) * 16;
 
    /* color buffer */
-   for (i = 0; i < rast->state.fb.nr_cbufs; i++)
+   for (i = 0; i < rast->state.nr_cbufs; i++)
       color[i] = tile->color[i] + 4 * block_offset;
 
    depth = lp_rast_depth_pointer(rast, x, y);
diff --git a/src/gallium/drivers/llvmpipe/lp_rast_tri.c b/src/gallium/drivers/llvmpipe/lp_rast_tri.c
index 0334705ef79..a5f0d14c95d 100644
--- a/src/gallium/drivers/llvmpipe/lp_rast_tri.c
+++ b/src/gallium/drivers/llvmpipe/lp_rast_tri.c
@@ -89,14 +89,11 @@ static const int pos_table16[16][2] = {
  * Shade all pixels in a 4x4 block.
  */
 static void
-block_full_4( struct lp_rasterizer_task *rast_task,
-              const struct lp_rast_triangle *tri,
-              int x, int y )
+block_full_4(struct lp_rasterizer_task *task,
+             const struct lp_rast_triangle *tri,
+             int x, int y)
 {
-   lp_rast_shade_quads_all(rast_task->rast,
-                           rast_task->thread_index,
-                           &tri->inputs, 
-                           x, y);
+   lp_rast_shade_quads_all(task, &tri->inputs, x, y);
 }
 
 
@@ -104,16 +101,16 @@ block_full_4( struct lp_rasterizer_task *rast_task,
  * Shade all pixels in a 16x16 block.
  */
 static void
-block_full_16( struct lp_rasterizer_task *rast_task,
-               const struct lp_rast_triangle *tri,
-               int x, int y )
+block_full_16(struct lp_rasterizer_task *task,
+              const struct lp_rast_triangle *tri,
+              int x, int y)
 {
    unsigned ix, iy;
    assert(x % 16 == 0);
    assert(y % 16 == 0);
    for (iy = 0; iy < 16; iy += 4)
       for (ix = 0; ix < 16; ix += 4)
-	 block_full_4(rast_task, tri, x + ix, y + iy);
+	 block_full_4(task, tri, x + ix, y + iy);
 }
 
 
@@ -123,18 +120,15 @@ block_full_16( struct lp_rasterizer_task *rast_task,
  * will be done as part of the fragment shader.
  */
 static void
-do_block_4( struct lp_rasterizer_task *rast_task,
-	    const struct lp_rast_triangle *tri,
-	    int x, int y,
-	    int c1,
-	    int c2,
-	    int c3 )
+do_block_4(struct lp_rasterizer_task *task,
+           const struct lp_rast_triangle *tri,
+           int x, int y,
+           int c1, int c2, int c3)
 {
-   lp_rast_shade_quads(rast_task->rast,
-                       rast_task->thread_index,
-                       &tri->inputs, 
-                       x, y,
-                       -c1, -c2, -c3);
+   assert(x >= 0);
+   assert(y >= 0);
+
+   lp_rast_shade_quads(task, &tri->inputs, x, y, -c1, -c2, -c3);
 }
 
 
@@ -143,18 +137,18 @@ do_block_4( struct lp_rasterizer_task *rast_task,
  * of the triangle's bounds.
  */
 static void
-do_block_16( struct lp_rasterizer_task *rast_task,
-             const struct lp_rast_triangle *tri,
-             int x, int y,
-             int c0,
-             int c1,
-             int c2 )
+do_block_16(struct lp_rasterizer_task *task,
+            const struct lp_rast_triangle *tri,
+            int x, int y,
+            int c0, int c1, int c2)
 {
    unsigned mask = 0;
    int eo[3];
    int c[3];
    int i, j;
 
+   assert(x >= 0);
+   assert(y >= 0);
    assert(x % 16 == 0);
    assert(y % 16 == 0);
 
@@ -193,7 +187,7 @@ do_block_16( struct lp_rasterizer_task *rast_task,
        * the triangle.  It's a little faster to do it in the jit code.
        */
       LP_COUNT(nr_non_empty_4);
-      do_block_4(rast_task, tri, px, py, cx1, cx2, cx3);
+      do_block_4(task, tri, px, py, cx1, cx2, cx3);
    }
 }
 
@@ -203,15 +197,11 @@ do_block_16( struct lp_rasterizer_task *rast_task,
  * for this triangle.
  */
 void
-lp_rast_triangle( struct lp_rasterizer *rast,
-                  unsigned thread_index,
-                  const union lp_rast_cmd_arg arg )
+lp_rast_triangle(struct lp_rasterizer_task *task,
+                 const union lp_rast_cmd_arg arg)
 {
-   struct lp_rasterizer_task *rast_task = &rast->tasks[thread_index];
    const struct lp_rast_triangle *tri = arg.triangle;
-
-   int x = rast_task->x;
-   int y = rast_task->y;
+   const int x = task->x, y = task->y;
    int ei[3], eo[3], c[3];
    unsigned outmask, inmask, partial_mask;
    unsigned i, j;
@@ -272,7 +262,7 @@ lp_rast_triangle( struct lp_rasterizer *rast,
       partial_mask &= ~(1 << i);
 
       LP_COUNT(nr_partially_covered_16);
-      do_block_16(rast_task, tri, px, py, cx1, cx2, cx3);
+      do_block_16(task, tri, px, py, cx1, cx2, cx3);
    }
 
    /* Iterate over fulls: 
@@ -285,6 +275,6 @@ lp_rast_triangle( struct lp_rasterizer *rast,
       inmask &= ~(1 << i);
 
       LP_COUNT(nr_fully_covered_16);
-      block_full_16(rast_task, tri, px, py);
+      block_full_16(task, tri, px, py);
    }
 }
diff --git a/src/gallium/drivers/llvmpipe/lp_scene.c b/src/gallium/drivers/llvmpipe/lp_scene.c
index b7116297ece..72492c0f0ca 100644
--- a/src/gallium/drivers/llvmpipe/lp_scene.c
+++ b/src/gallium/drivers/llvmpipe/lp_scene.c
@@ -29,44 +29,67 @@
 #include "util/u_memory.h"
 #include "util/u_inlines.h"
 #include "util/u_simple_list.h"
+#include "util/u_surface.h"
 #include "lp_scene.h"
+#include "lp_scene_queue.h"
+#include "lp_debug.h"
 
 
 struct lp_scene *
-lp_scene_create(void)
+lp_scene_create( struct pipe_context *pipe,
+                 struct lp_scene_queue *queue )
 {
+   unsigned i, j;
    struct lp_scene *scene = CALLOC_STRUCT(lp_scene);
-   if (scene)
-      lp_scene_init(scene);
+   if (!scene)
+      return NULL;
+
+   scene->pipe = pipe;
+   scene->empty_queue = queue;
+
+   for (i = 0; i < TILES_X; i++) {
+      for (j = 0; j < TILES_Y; j++) {
+         struct cmd_bin *bin = lp_scene_get_bin(scene, i, j);
+         bin->commands.head = bin->commands.tail = CALLOC_STRUCT(cmd_block);
+      }
+   }
+
+   scene->data.head =
+      scene->data.tail = CALLOC_STRUCT(data_block);
+
+   make_empty_list(&scene->textures);
+
+   pipe_mutex_init(scene->mutex);
+
    return scene;
 }
 
 
+/**
+ * Free all data associated with the given scene, and free(scene).
+ */
 void
 lp_scene_destroy(struct lp_scene *scene)
 {
-   lp_scene_reset(scene);
-   lp_scene_free_bin_data(scene);
-   FREE(scene);
-}
+   unsigned i, j;
 
+   lp_scene_reset(scene);
 
-void
-lp_scene_init(struct lp_scene *scene)
-{
-   unsigned i, j;
    for (i = 0; i < TILES_X; i++)
       for (j = 0; j < TILES_Y; j++) {
          struct cmd_bin *bin = lp_scene_get_bin(scene, i, j);
-         bin->commands.head = bin->commands.tail = CALLOC_STRUCT(cmd_block);
+         assert(bin->commands.head == bin->commands.tail);
+         FREE(bin->commands.head);
+         bin->commands.head = NULL;
+         bin->commands.tail = NULL;
       }
 
-   scene->data.head =
-      scene->data.tail = CALLOC_STRUCT(data_block);
+   FREE(scene->data.head);
+   scene->data.head = NULL;
 
-   make_empty_list(&scene->textures);
+   pipe_mutex_destroy(scene->mutex);
 
-   pipe_mutex_init(scene->mutex);
+   FREE(scene);
 }
 
 
@@ -92,6 +115,9 @@ lp_scene_is_empty(struct lp_scene *scene )
 }
 
 
+/* Free data for one particular bin.  May be called from the
+ * rasterizer thread(s).
+ */
 void
 lp_scene_bin_reset(struct lp_scene *scene, unsigned x, unsigned y)
 {
@@ -100,6 +126,9 @@ lp_scene_bin_reset(struct lp_scene *scene, unsigned x, unsigned y)
    struct cmd_block *block;
    struct cmd_block *tmp;
 
+   assert(x < TILES_X);
+   assert(y < TILES_Y);
+
    for (block = list->head; block != list->tail; block = tmp) {
       tmp = block->next;
       FREE(block);
@@ -112,7 +141,8 @@ lp_scene_bin_reset(struct lp_scene *scene, unsigned x, unsigned y)
 
 
 /**
- * Set scene to empty state.
+ * Free all the temporary data in a scene.  May be called from the
+ * rasterizer thread(s).
  */
 void
 lp_scene_reset(struct lp_scene *scene )
@@ -159,40 +189,8 @@ lp_scene_reset(struct lp_scene *scene )
 }
 
 
-/**
- * Free all data associated with the given bin, but don't free(scene).
- */
-void
-lp_scene_free_bin_data(struct lp_scene *scene)
-{
-   unsigned i, j;
 
-   for (i = 0; i < TILES_X; i++)
-      for (j = 0; j < TILES_Y; j++) {
-         struct cmd_bin *bin = lp_scene_get_bin(scene, i, j);
-         /* lp_reset_scene() should have been already called */
-         assert(bin->commands.head == bin->commands.tail);
-         FREE(bin->commands.head);
-         bin->commands.head = NULL;
-         bin->commands.tail = NULL;
-      }
 
-   FREE(scene->data.head);
-   scene->data.head = NULL;
-
-   pipe_mutex_destroy(scene->mutex);
-}
-
-
-void
-lp_scene_set_framebuffer_size( struct lp_scene *scene,
-                               unsigned width, unsigned height )
-{
-   assert(lp_scene_is_empty(scene));
-
-   scene->tiles_x = align(width, TILE_SIZE) / TILE_SIZE;
-   scene->tiles_y = align(height, TILE_SIZE) / TILE_SIZE;
-}
 
 
 void
@@ -390,3 +388,158 @@ end:
    pipe_mutex_unlock(scene->mutex);
    return bin;
 }
+
+
+/**
+ * Prepare this scene for the rasterizer.
+ * Map the framebuffer surfaces.  Initialize the 'rast' state.
+ */
+static boolean
+lp_scene_map_buffers( struct lp_scene *scene )
+{
+   struct pipe_screen *screen = scene->pipe->screen;
+   struct pipe_surface *cbuf, *zsbuf;
+   int i;
+
+   LP_DBG(DEBUG_RAST, "%s\n", __FUNCTION__);
+
+
+   /* Map all color buffers 
+    */
+   for (i = 0; i < scene->fb.nr_cbufs; i++) {
+      cbuf = scene->fb.cbufs[i];
+      if (cbuf) {
+	 scene->cbuf_transfer[i] = screen->get_tex_transfer(screen,
+                                                          cbuf->texture,
+                                                          cbuf->face,
+                                                          cbuf->level,
+                                                          cbuf->zslice,
+                                                          PIPE_TRANSFER_READ_WRITE,
+                                                          0, 0,
+                                                          cbuf->width, 
+                                                          cbuf->height);
+	 if (!scene->cbuf_transfer[i])
+	    goto fail;
+
+	 scene->cbuf_map[i] = screen->transfer_map(screen, 
+                                                 scene->cbuf_transfer[i]);
+	 if (!scene->cbuf_map[i])
+	    goto fail;
+      }
+   }
+
+   /* Map the zsbuffer
+    */
+   zsbuf = scene->fb.zsbuf;
+   if (zsbuf) {
+      scene->zsbuf_transfer = screen->get_tex_transfer(screen,
+                                                       zsbuf->texture,
+                                                       zsbuf->face,
+                                                       zsbuf->level,
+                                                       zsbuf->zslice,
+                                                       PIPE_TRANSFER_READ_WRITE,
+                                                       0, 0,
+                                                       zsbuf->width,
+                                                       zsbuf->height);
+      if (!scene->zsbuf_transfer)
+         goto fail;
+
+      scene->zsbuf_map = screen->transfer_map(screen, 
+                                              scene->zsbuf_transfer);
+      if (!scene->zsbuf_map)
+	 goto fail;
+   }
+
+   return TRUE;
+
+fail:
+   /* Unmap and release transfers?
+    */
+   return FALSE;
+}
+
+
+
+/**
+ * Called after rasterizer as finished rasterizing a scene. 
+ * 
+ * We want to call this from the pipe_context's current thread to
+ * avoid having to have mutexes on the transfer functions.
+ */
+static void
+lp_scene_unmap_buffers( struct lp_scene *scene )
+{
+   struct pipe_screen *screen = scene->pipe->screen;
+   unsigned i;
+
+   for (i = 0; i < scene->fb.nr_cbufs; i++) {
+      if (scene->cbuf_map[i]) 
+	 screen->transfer_unmap(screen, scene->cbuf_transfer[i]);
+
+      if (scene->cbuf_transfer[i])
+	 screen->tex_transfer_destroy(scene->cbuf_transfer[i]);
+
+      scene->cbuf_transfer[i] = NULL;
+      scene->cbuf_map[i] = NULL;
+   }
+
+   if (scene->zsbuf_map) 
+      screen->transfer_unmap(screen, scene->zsbuf_transfer);
+
+   if (scene->zsbuf_transfer)
+      screen->tex_transfer_destroy(scene->zsbuf_transfer);
+
+   scene->zsbuf_transfer = NULL;
+   scene->zsbuf_map = NULL;
+
+   util_unreference_framebuffer_state( &scene->fb );
+}
+
+
+void lp_scene_begin_binning( struct lp_scene *scene,
+                             struct pipe_framebuffer_state *fb )
+{
+   assert(lp_scene_is_empty(scene));
+
+   util_copy_framebuffer_state(&scene->fb, fb);
+
+   scene->tiles_x = align(fb->width, TILE_SIZE) / TILE_SIZE;
+   scene->tiles_y = align(fb->height, TILE_SIZE) / TILE_SIZE;
+}
+
+
+void lp_scene_rasterize( struct lp_scene *scene,
+                         struct lp_rasterizer *rast,
+                         boolean write_depth )
+{
+   if (0) {
+      unsigned x, y;
+      debug_printf("rasterize scene:\n");
+      debug_printf("  data size: %u\n", lp_scene_data_size(scene));
+      for (y = 0; y < scene->tiles_y; y++) {
+         for (x = 0; x < scene->tiles_x; x++) {
+            debug_printf("  bin %u, %u size: %u\n", x, y,
+                         lp_scene_bin_size(scene, x, y));
+         }
+      }
+   }
+
+
+   scene->write_depth = (scene->fb.zsbuf != NULL &&
+                         write_depth);
+
+   lp_scene_map_buffers( scene );
+
+   /* Enqueue the scene for rasterization, then immediately wait for
+    * it to finish.
+    */
+   lp_rast_queue_scene( rast, scene );
+
+   /* Currently just wait for the rasterizer to finish.  Some
+    * threading interactions need to be worked out, particularly once
+    * transfers become per-context:
+    */
+   lp_rast_finish( rast );
+   lp_scene_unmap_buffers( scene );
+   lp_scene_enqueue( scene->empty_queue, scene );
+}
diff --git a/src/gallium/drivers/llvmpipe/lp_scene.h b/src/gallium/drivers/llvmpipe/lp_scene.h
index fb478cc2eb5..739ac229089 100644
--- a/src/gallium/drivers/llvmpipe/lp_scene.h
+++ b/src/gallium/drivers/llvmpipe/lp_scene.h
@@ -39,6 +39,7 @@
 #include "lp_tile_soa.h"
 #include "lp_rast.h"
 
+struct lp_scene_queue;
 
 /* We're limited to 2K by 2K for 32bit fixed point rasterization.
  * Will need a 64-bit version for larger framebuffers.
@@ -56,8 +57,7 @@
 
 /* switch to a non-pointer value for this:
  */
-typedef void (*lp_rast_cmd)( struct lp_rasterizer *,
-                             unsigned thread_index,
+typedef void (*lp_rast_cmd)( struct lp_rasterizer_task *,
                              const union lp_rast_cmd_arg );
 
 struct cmd_block {
@@ -113,8 +113,14 @@ struct texture_ref {
  * scenes:
  */
 struct lp_scene {
-   struct cmd_bin tile[TILES_X][TILES_Y];
-   struct data_block_list data;
+   struct pipe_context *pipe;
+   struct pipe_transfer *cbuf_transfer[PIPE_MAX_COLOR_BUFS];
+   struct pipe_transfer *zsbuf_transfer;
+
+   /* Scene's buffers are mapped at the time the scene is enqueued:
+    */
+   void *cbuf_map[PIPE_MAX_COLOR_BUFS];
+   uint8_t *zsbuf_map;
 
    /** the framebuffer to render the scene into */
    struct pipe_framebuffer_state fb;
@@ -132,25 +138,28 @@ struct lp_scene {
 
    int curr_x, curr_y;  /**< for iterating over bins */
    pipe_mutex mutex;
+
+   /* Where to place this scene once it has been rasterized:
+    */
+   struct lp_scene_queue *empty_queue;
+
+   struct cmd_bin tile[TILES_X][TILES_Y];
+   struct data_block_list data;
 };
 
 
 
-struct lp_scene *lp_scene_create(void);
+struct lp_scene *lp_scene_create(struct pipe_context *pipe,
+                                 struct lp_scene_queue *empty_queue);
 
 void lp_scene_destroy(struct lp_scene *scene);
 
 
-void lp_scene_init(struct lp_scene *scene);
 
 boolean lp_scene_is_empty(struct lp_scene *scene );
 
 void lp_scene_reset(struct lp_scene *scene );
 
-void lp_scene_free_bin_data(struct lp_scene *scene);
-
-void lp_scene_set_framebuffer_size( struct lp_scene *scene,
-                                  unsigned width, unsigned height );
 
 void lp_bin_new_data_block( struct data_block_list *list );
 
@@ -297,5 +306,13 @@ lp_scene_bin_iter_begin( struct lp_scene *scene );
 struct cmd_bin *
 lp_scene_bin_iter_next( struct lp_scene *scene, int *bin_x, int *bin_y );
 
+void
+lp_scene_rasterize( struct lp_scene *scene,
+                    struct lp_rasterizer *rast,
+                    boolean write_depth );
+
+void
+lp_scene_begin_binning( struct lp_scene *scene,
+                        struct pipe_framebuffer_state *fb );
 
 #endif /* LP_BIN_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_screen.c b/src/gallium/drivers/llvmpipe/lp_screen.c
index 1cd3ea9a840..5093f58bb19 100644
--- a/src/gallium/drivers/llvmpipe/lp_screen.c
+++ b/src/gallium/drivers/llvmpipe/lp_screen.c
@@ -34,11 +34,13 @@
 #include "lp_texture.h"
 #include "lp_buffer.h"
 #include "lp_fence.h"
-#include "lp_winsys.h"
 #include "lp_jit.h"
 #include "lp_screen.h"
 #include "lp_context.h"
 #include "lp_debug.h"
+#include "lp_public.h"
+
+#include "state_tracker/sw_winsys.h"
 
 #ifdef DEBUG
 int LP_DEBUG = 0;
@@ -83,7 +85,7 @@ llvmpipe_get_param(struct pipe_screen *screen, int param)
    case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS:
       return PIPE_MAX_SAMPLERS;
    case PIPE_CAP_MAX_VERTEX_TEXTURE_UNITS:
-      return PIPE_MAX_VERTEX_SAMPLERS;
+      return 0;
    case PIPE_CAP_MAX_COMBINED_SAMPLERS:
       return PIPE_MAX_SAMPLERS + PIPE_MAX_VERTEX_SAMPLERS;
    case PIPE_CAP_NPOT_TEXTURES:
@@ -107,11 +109,11 @@ llvmpipe_get_param(struct pipe_screen *screen, int param)
    case PIPE_CAP_TEXTURE_SHADOW_MAP:
       return 1;
    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
-      return 13; /* max 4Kx4K */
+      return LP_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
-      return 9;  /* max 256x256x256 */
+      return LP_MAX_TEXTURE_3D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
-      return 13; /* max 4Kx4K */
+      return LP_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_TGSI_CONT_SUPPORTED:
       return 1;
    case PIPE_CAP_BLEND_EQUATION_SEPARATE:
@@ -167,7 +169,7 @@ llvmpipe_is_format_supported( struct pipe_screen *_screen,
                               unsigned geom_flags )
 {
    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
-   struct llvmpipe_winsys *winsys = screen->winsys;
+   struct sw_winsys *winsys = screen->winsys;
    const struct util_format_description *format_desc;
 
    format_desc = util_format_description(format);
@@ -194,9 +196,7 @@ llvmpipe_is_format_supported( struct pipe_screen *_screen,
          format_desc->block.height != 1)
          return FALSE;
 
-      if(format_desc->layout != UTIL_FORMAT_LAYOUT_SCALAR &&
-         format_desc->layout != UTIL_FORMAT_LAYOUT_ARITH &&
-         format_desc->layout != UTIL_FORMAT_LAYOUT_ARRAY)
+      if(format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN)
          return FALSE;
 
       if(format_desc->colorspace != UTIL_FORMAT_COLORSPACE_RGB &&
@@ -224,33 +224,22 @@ llvmpipe_is_format_supported( struct pipe_screen *_screen,
          format_desc->block.height != 1)
          return FALSE;
 
-      if(format_desc->layout != UTIL_FORMAT_LAYOUT_SCALAR &&
-         format_desc->layout != UTIL_FORMAT_LAYOUT_ARITH &&
-         format_desc->layout != UTIL_FORMAT_LAYOUT_ARRAY)
+      if(format_desc->layout != UTIL_FORMAT_LAYOUT_PLAIN)
          return FALSE;
 
       if(format_desc->colorspace != UTIL_FORMAT_COLORSPACE_RGB &&
-         format_desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB &&
          format_desc->colorspace != UTIL_FORMAT_COLORSPACE_ZS)
          return FALSE;
+
+      /* not supported yet */
+      if (format == PIPE_FORMAT_Z16_UNORM)
+         return FALSE;
    }
 
    return TRUE;
 }
 
 
-static struct pipe_buffer *
-llvmpipe_surface_buffer_create(struct pipe_screen *screen,
-                               unsigned width, unsigned height,
-                               enum pipe_format format,
-                               unsigned tex_usage,
-                               unsigned usage,
-                               unsigned *stride)
-{
-   /* This function should never be used */
-   assert(0);
-   return NULL;
-}
 
 
 static void
@@ -259,7 +248,7 @@ llvmpipe_flush_frontbuffer(struct pipe_screen *_screen,
                            void *context_private)
 {
    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
-   struct llvmpipe_winsys *winsys = screen->winsys;
+   struct sw_winsys *winsys = screen->winsys;
    struct llvmpipe_texture *texture = llvmpipe_texture(surface->texture);
 
    assert(texture->dt);
@@ -272,7 +261,7 @@ static void
 llvmpipe_destroy_screen( struct pipe_screen *_screen )
 {
    struct llvmpipe_screen *screen = llvmpipe_screen(_screen);
-   struct llvmpipe_winsys *winsys = screen->winsys;
+   struct sw_winsys *winsys = screen->winsys;
 
    lp_jit_screen_cleanup(screen);
 
@@ -289,7 +278,7 @@ llvmpipe_destroy_screen( struct pipe_screen *_screen )
  * Note: we're not presently subclassing pipe_screen (no llvmpipe_screen).
  */
 struct pipe_screen *
-llvmpipe_create_screen(struct llvmpipe_winsys *winsys)
+llvmpipe_create_screen(struct sw_winsys *winsys)
 {
    struct llvmpipe_screen *screen = CALLOC_STRUCT(llvmpipe_screen);
 
@@ -310,7 +299,6 @@ llvmpipe_create_screen(struct llvmpipe_winsys *winsys)
    screen->base.get_paramf = llvmpipe_get_paramf;
    screen->base.is_format_supported = llvmpipe_is_format_supported;
 
-   screen->base.surface_buffer_create = llvmpipe_surface_buffer_create;
    screen->base.context_create = llvmpipe_create_context;
    screen->base.flush_frontbuffer = llvmpipe_flush_frontbuffer;
 
diff --git a/src/gallium/drivers/llvmpipe/lp_screen.h b/src/gallium/drivers/llvmpipe/lp_screen.h
index 4a1b4d6f3e2..f4e62cbf08e 100644
--- a/src/gallium/drivers/llvmpipe/lp_screen.h
+++ b/src/gallium/drivers/llvmpipe/lp_screen.h
@@ -43,14 +43,14 @@
 #include "pipe/p_defines.h"
 
 
-struct llvmpipe_winsys;
+struct sw_winsys;
 
 
 struct llvmpipe_screen
 {
    struct pipe_screen base;
 
-   struct llvmpipe_winsys *winsys;
+   struct sw_winsys *winsys;
 
    LLVMModuleRef module;
    LLVMExecutionEngineRef engine;
@@ -76,4 +76,5 @@ llvmpipe_screen( struct pipe_screen *pipe )
 }
 
 
+
 #endif /* LP_SCREEN_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_setup.c b/src/gallium/drivers/llvmpipe/lp_setup.c
index 7d52765970c..059584f2a77 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup.c
@@ -46,7 +46,7 @@
 #include "lp_rast.h"
 #include "lp_setup_context.h"
 #include "lp_screen.h"
-#include "lp_winsys.h"
+#include "state_tracker/sw_winsys.h"
 
 #include "draw/draw_context.h"
 #include "draw/draw_vbuf.h"
@@ -64,11 +64,10 @@ lp_setup_get_current_scene(struct setup_context *setup)
        */
       setup->scene = lp_scene_dequeue(setup->empty_scenes, TRUE);
 
-      if(0)lp_scene_reset( setup->scene ); /* XXX temporary? */
+      assert(lp_scene_is_empty(setup->scene));
 
-      lp_scene_set_framebuffer_size(setup->scene,
-                                    setup->fb.width, 
-                                    setup->fb.height);
+      lp_scene_begin_binning(setup->scene,
+                             &setup->fb );
    }
    return setup->scene;
 }
@@ -133,13 +132,12 @@ static void reset_context( struct setup_context *setup )
 /** Rasterize all scene's bins */
 static void
 lp_setup_rasterize_scene( struct setup_context *setup,
-			 boolean write_depth )
+                          boolean write_depth )
 {
    struct lp_scene *scene = lp_setup_get_current_scene(setup);
 
-   lp_rasterize_scene(setup->rast,
-                      scene,
-                      &setup->fb,
+   lp_scene_rasterize(scene,
+                      setup->rast,
                       write_depth);
 
    reset_context( setup );
@@ -244,19 +242,16 @@ void
 lp_setup_bind_framebuffer( struct setup_context *setup,
                            const struct pipe_framebuffer_state *fb )
 {
-   struct lp_scene *scene = lp_setup_get_current_scene(setup);
-
    LP_DBG(DEBUG_SETUP, "%s\n", __FUNCTION__);
 
+   /* Flush any old scene.
+    */
    set_scene_state( setup, SETUP_FLUSHED );
 
-   /* re-get scene pointer, may have a new scene after flushing */
-   (void) scene;
-   scene = lp_setup_get_current_scene(setup);
-
+   /* Set new state.  This will be picked up later when we next need a
+    * scene.
+    */
    util_copy_framebuffer_state(&setup->fb, fb);
-
-   lp_scene_set_framebuffer_size(scene, setup->fb.width, setup->fb.height);
 }
 
 
@@ -476,20 +471,28 @@ lp_setup_set_fragment_sampler_views(struct setup_context *setup,
          jit_tex = &setup->fs.current.jit_context.textures[i];
          jit_tex->width = tex->width0;
          jit_tex->height = tex->height0;
+         jit_tex->depth = tex->depth0;
+         jit_tex->last_level = tex->last_level;
          jit_tex->stride = lp_tex->stride[0];
-         if(!lp_tex->dt) {
-            jit_tex->data = lp_tex->data;
+         if (!lp_tex->dt) {
+            /* regular texture - setup array of mipmap level pointers */
+            int j;
+            for (j = 0; j < LP_MAX_TEXTURE_2D_LEVELS; j++) {
+               jit_tex->data[j] =
+                  (ubyte *) lp_tex->data + lp_tex->level_offset[j];
+            }
          }
          else {
+            /* display target texture/surface */
             /*
              * XXX: Where should this be unmapped?
              */
 
             struct llvmpipe_screen *screen = llvmpipe_screen(tex->screen);
-            struct llvmpipe_winsys *winsys = screen->winsys;
-            jit_tex->data = winsys->displaytarget_map(winsys, lp_tex->dt,
+            struct sw_winsys *winsys = screen->winsys;
+            jit_tex->data[0] = winsys->displaytarget_map(winsys, lp_tex->dt,
                                                       PIPE_BUFFER_USAGE_CPU_READ);
-            assert(jit_tex->data);
+            assert(jit_tex->data[0]);
          }
 
          /* the scene references this texture */
@@ -683,7 +686,7 @@ lp_setup_destroy( struct setup_context *setup )
  * it.
  */
 struct setup_context *
-lp_setup_create( struct pipe_screen *screen,
+lp_setup_create( struct pipe_context *pipe,
                  struct draw_context *draw )
 {
    unsigned i;
@@ -698,7 +701,9 @@ lp_setup_create( struct pipe_screen *screen,
    if (!setup->empty_scenes)
       goto fail;
 
-   setup->rast = lp_rast_create( screen, setup->empty_scenes );
+   /* XXX: move this to the screen and share between contexts:
+    */
+   setup->rast = lp_rast_create();
    if (!setup->rast) 
       goto fail;
 
@@ -711,7 +716,8 @@ lp_setup_create( struct pipe_screen *screen,
 
    /* create some empty scenes */
    for (i = 0; i < MAX_SCENES; i++) {
-      setup->scenes[i] = lp_scene_create();
+      setup->scenes[i] = lp_scene_create( pipe, setup->empty_scenes );
+
       lp_scene_enqueue(setup->empty_scenes, setup->scenes[i]);
    }
 
diff --git a/src/gallium/drivers/llvmpipe/lp_setup.h b/src/gallium/drivers/llvmpipe/lp_setup.h
index 5a5d3531f95..72116b8c6c4 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup.h
+++ b/src/gallium/drivers/llvmpipe/lp_setup.h
@@ -62,7 +62,7 @@ struct lp_fragment_shader;
 struct lp_jit_context;
 
 struct setup_context *
-lp_setup_create( struct pipe_screen *screen,
+lp_setup_create( struct pipe_context *pipe,
                  struct draw_context *draw );
 
 void
diff --git a/src/gallium/drivers/llvmpipe/lp_setup_tri.c b/src/gallium/drivers/llvmpipe/lp_setup_tri.c
index a8bf540803b..e75412ac9aa 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup_tri.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup_tri.c
@@ -297,6 +297,15 @@ do_triangle_ccw(struct setup_context *setup,
 
    tri = alloc_triangle(scene, setup->fs.nr_inputs, &tri_bytes);
 
+#ifdef DEBUG
+   tri->v[0][0] = v1[0][0];
+   tri->v[1][0] = v2[0][0];
+   tri->v[2][0] = v3[0][0];
+   tri->v[0][1] = v1[0][1];
+   tri->v[1][1] = v2[0][1];
+   tri->v[2][1] = v3[0][1];
+#endif
+
    tri->dx12 = x1 - x2;
    tri->dx23 = x2 - x3;
    tri->dx31 = x3 - x1;
diff --git a/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c b/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
index 24291da91e4..671e74465c0 100644
--- a/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
+++ b/src/gallium/drivers/llvmpipe/lp_setup_vbuf.c
@@ -231,57 +231,29 @@ lp_setup_draw(struct vbuf_render *vbr, const ushort *indices, uint nr)
       break;
 
    case PIPE_PRIM_QUADS:
-      if (setup->flatshade_first) {
-         for (i = 3; i < nr; i += 4) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-2], stride),
-                             get_vert(vertex_buffer, indices[i-1], stride),
-                             get_vert(vertex_buffer, indices[i-3], stride) );
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-1], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride),
-                             get_vert(vertex_buffer, indices[i-3], stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 4) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-3], stride),
-                             get_vert(vertex_buffer, indices[i-2], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride) );
+      for (i = 3; i < nr; i += 4) {
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, indices[i-3], stride),
+                          get_vert(vertex_buffer, indices[i-2], stride),
+                          get_vert(vertex_buffer, indices[i-0], stride) );
 
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-2], stride),
-                             get_vert(vertex_buffer, indices[i-1], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride) );
-         }
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, indices[i-2], stride),
+                          get_vert(vertex_buffer, indices[i-1], stride),
+                          get_vert(vertex_buffer, indices[i-0], stride) );
       }
       break;
 
    case PIPE_PRIM_QUAD_STRIP:
-      if (setup->flatshade_first) {
-         for (i = 3; i < nr; i += 2) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-0], stride),
-                             get_vert(vertex_buffer, indices[i-1], stride),
-                             get_vert(vertex_buffer, indices[i-3], stride));
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-2], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride),
-                             get_vert(vertex_buffer, indices[i-3], stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 2) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-3], stride),
-                             get_vert(vertex_buffer, indices[i-2], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride) );
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, indices[i-1], stride),
-                             get_vert(vertex_buffer, indices[i-3], stride),
-                             get_vert(vertex_buffer, indices[i-0], stride) );
-         }
+      for (i = 3; i < nr; i += 2) {
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, indices[i-3], stride),
+                          get_vert(vertex_buffer, indices[i-2], stride),
+                          get_vert(vertex_buffer, indices[i-0], stride) );
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, indices[i-1], stride),
+                          get_vert(vertex_buffer, indices[i-3], stride),
+                          get_vert(vertex_buffer, indices[i-0], stride) );
       }
       break;
 
@@ -415,57 +387,28 @@ lp_setup_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
       break;
 
    case PIPE_PRIM_QUADS:
-      if (setup->flatshade_first) {
-         for (i = 3; i < nr; i += 4) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-2, stride),
-                             get_vert(vertex_buffer, i-1, stride),
-                             get_vert(vertex_buffer, i-3, stride) );
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-1, stride),
-                             get_vert(vertex_buffer, i-0, stride),
-                             get_vert(vertex_buffer, i-3, stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 4) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-3, stride),
-                             get_vert(vertex_buffer, i-2, stride),
-                             get_vert(vertex_buffer, i-0, stride) );
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-2, stride),
-                             get_vert(vertex_buffer, i-1, stride),
-                             get_vert(vertex_buffer, i-0, stride) );
-         }
+      for (i = 3; i < nr; i += 4) {
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, i-3, stride),
+                          get_vert(vertex_buffer, i-2, stride),
+                          get_vert(vertex_buffer, i-0, stride) );
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, i-2, stride),
+                          get_vert(vertex_buffer, i-1, stride),
+                          get_vert(vertex_buffer, i-0, stride) );
       }
       break;
 
    case PIPE_PRIM_QUAD_STRIP:
-      if (setup->flatshade_first) {
-         for (i = 3; i < nr; i += 2) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-0, stride),
-                             get_vert(vertex_buffer, i-1, stride),
-                             get_vert(vertex_buffer, i-3, stride) );
-            setup->triangle( setup,
-
-                             get_vert(vertex_buffer, i-2, stride),
-                             get_vert(vertex_buffer, i-0, stride),
-                             get_vert(vertex_buffer, i-3, stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 2) {
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-3, stride),
-                             get_vert(vertex_buffer, i-2, stride),
-                             get_vert(vertex_buffer, i-0, stride) );
-            setup->triangle( setup,
-                             get_vert(vertex_buffer, i-1, stride),
-                             get_vert(vertex_buffer, i-3, stride),
-                             get_vert(vertex_buffer, i-0, stride) );
-         }
+      for (i = 3; i < nr; i += 2) {
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, i-3, stride),
+                          get_vert(vertex_buffer, i-2, stride),
+                          get_vert(vertex_buffer, i-0, stride) );
+         setup->triangle( setup,
+                          get_vert(vertex_buffer, i-1, stride),
+                          get_vert(vertex_buffer, i-3, stride),
+                          get_vert(vertex_buffer, i-0, stride) );
       }
       break;
 
diff --git a/src/gallium/drivers/llvmpipe/lp_state.h b/src/gallium/drivers/llvmpipe/lp_state.h
index 8a2a08b0759..9b87cd202e7 100644
--- a/src/gallium/drivers/llvmpipe/lp_state.h
+++ b/src/gallium/drivers/llvmpipe/lp_state.h
@@ -119,6 +119,10 @@ struct lp_vertex_shader {
    struct draw_vertex_shader *draw_data;
 };
 
+struct lp_velems_state {
+   unsigned count;
+   struct pipe_vertex_element velem[PIPE_MAX_ATTRIBS];
+};
 
 
 void *
@@ -176,8 +180,14 @@ void *llvmpipe_create_vs_state(struct pipe_context *,
 void llvmpipe_bind_vs_state(struct pipe_context *, void *);
 void llvmpipe_delete_vs_state(struct pipe_context *, void *);
 
+void *llvmpipe_create_vertex_elements_state(struct pipe_context *,
+                                            unsigned count,
+                                            const struct pipe_vertex_element *);
+void llvmpipe_bind_vertex_elements_state(struct pipe_context *, void *);
+void llvmpipe_delete_vertex_elements_state(struct pipe_context *, void *);
+
 void llvmpipe_set_polygon_stipple( struct pipe_context *,
-				  const struct pipe_poly_stipple * );
+                                   const struct pipe_poly_stipple * );
 
 void llvmpipe_set_scissor_state( struct pipe_context *,
                                  const struct pipe_scissor_state * );
@@ -203,10 +213,6 @@ llvmpipe_sampler_view_destroy(struct pipe_context *pipe,
 void llvmpipe_set_viewport_state( struct pipe_context *,
                                   const struct pipe_viewport_state * );
 
-void llvmpipe_set_vertex_elements(struct pipe_context *,
-                                  unsigned count,
-                                  const struct pipe_vertex_element *);
-
 void llvmpipe_set_vertex_buffers(struct pipe_context *,
                                  unsigned count,
                                  const struct pipe_vertex_buffer *);
diff --git a/src/gallium/drivers/llvmpipe/lp_state_fs.c b/src/gallium/drivers/llvmpipe/lp_state_fs.c
index 2e018df05c0..3e681b561aa 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_fs.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_fs.c
@@ -974,6 +974,13 @@ llvmpipe_delete_fs_state(struct pipe_context *pipe, void *fs)
    assert(fs != llvmpipe->fs);
    (void) llvmpipe;
 
+   /*
+    * XXX: we need to flush the context until we have some sort of reference
+    * counting in fragment shaders as they may still be binned
+    */
+   draw_flush(llvmpipe->draw);
+   lp_setup_flush(llvmpipe->setup, 0);
+
    variant = shader->variants;
    while(variant) {
       struct lp_fragment_shader_variant *next = variant->next;
diff --git a/src/gallium/drivers/llvmpipe/lp_state_vertex.c b/src/gallium/drivers/llvmpipe/lp_state_vertex.c
index 57ac25ea0cb..f6427aa908e 100644
--- a/src/gallium/drivers/llvmpipe/lp_state_vertex.c
+++ b/src/gallium/drivers/llvmpipe/lp_state_vertex.c
@@ -35,24 +35,41 @@
 #include "draw/draw_context.h"
 
 
+void *
+llvmpipe_create_vertex_elements_state(struct pipe_context *pipe,
+                                      unsigned count,
+                                      const struct pipe_vertex_element *attribs)
+{
+   struct lp_velems_state *velems;
+   assert(count <= PIPE_MAX_ATTRIBS);
+   velems = (struct lp_velems_state *) MALLOC(sizeof(struct lp_velems_state));
+   if (velems) {
+      velems->count = count;
+      memcpy(velems->velem, attribs, sizeof(*attribs) * count);
+   }
+   return velems;
+}
+
 void
-llvmpipe_set_vertex_elements(struct pipe_context *pipe,
-                             unsigned count,
-                             const struct pipe_vertex_element *attribs)
+llvmpipe_bind_vertex_elements_state(struct pipe_context *pipe,
+                                    void *velems)
 {
    struct llvmpipe_context *llvmpipe = llvmpipe_context(pipe);
+   struct lp_velems_state *lp_velems = (struct lp_velems_state *) velems;
 
-   assert(count <= PIPE_MAX_ATTRIBS);
-
-   memcpy(llvmpipe->vertex_element, attribs,
-          count * sizeof(struct pipe_vertex_element));
-   llvmpipe->num_vertex_elements = count;
+   llvmpipe->velems = lp_velems;
 
    llvmpipe->dirty |= LP_NEW_VERTEX;
 
-   draw_set_vertex_elements(llvmpipe->draw, count, attribs);
+   if (velems)
+      draw_set_vertex_elements(llvmpipe->draw, lp_velems->count, lp_velems->velem);
 }
 
+void
+llvmpipe_delete_vertex_elements_state(struct pipe_context *pipe, void *velems)
+{
+   FREE( velems );
+}
 
 void
 llvmpipe_set_vertex_buffers(struct pipe_context *pipe,
diff --git a/src/gallium/drivers/llvmpipe/lp_test_format.c b/src/gallium/drivers/llvmpipe/lp_test_format.c
index 48828bd0a0f..d05157991bb 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_format.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_format.c
@@ -52,25 +52,25 @@ struct pixel_test_case
 
 struct pixel_test_case test_cases[] =
 {
-   {PIPE_FORMAT_R5G6B5_UNORM,   0x0000, {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R5G6B5_UNORM,   0x001f, {0.0, 0.0, 1.0, 1.0}},
-   {PIPE_FORMAT_R5G6B5_UNORM,   0x07e0, {0.0, 1.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R5G6B5_UNORM,   0xf800, {1.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_R5G6B5_UNORM,   0xffff, {1.0, 1.0, 1.0, 1.0}},
-
-   {PIPE_FORMAT_A1R5G5B5_UNORM, 0x0000, {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A1R5G5B5_UNORM, 0x001f, {0.0, 0.0, 1.0, 0.0}},
-   {PIPE_FORMAT_A1R5G5B5_UNORM, 0x03e0, {0.0, 1.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A1R5G5B5_UNORM, 0x7c00, {1.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A1R5G5B5_UNORM, 0x8000, {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_A1R5G5B5_UNORM, 0xffff, {1.0, 1.0, 1.0, 1.0}},
+   {PIPE_FORMAT_B5G6R5_UNORM,   0x0000, {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B5G6R5_UNORM,   0x001f, {0.0, 0.0, 1.0, 1.0}},
+   {PIPE_FORMAT_B5G6R5_UNORM,   0x07e0, {0.0, 1.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B5G6R5_UNORM,   0xf800, {1.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B5G6R5_UNORM,   0xffff, {1.0, 1.0, 1.0, 1.0}},
+
+   {PIPE_FORMAT_B5G5R5A1_UNORM, 0x0000, {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B5G5R5A1_UNORM, 0x001f, {0.0, 0.0, 1.0, 0.0}},
+   {PIPE_FORMAT_B5G5R5A1_UNORM, 0x03e0, {0.0, 1.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B5G5R5A1_UNORM, 0x7c00, {1.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B5G5R5A1_UNORM, 0x8000, {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B5G5R5A1_UNORM, 0xffff, {1.0, 1.0, 1.0, 1.0}},
 
-   {PIPE_FORMAT_A8R8G8B8_UNORM, 0x00000000, {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A8R8G8B8_UNORM, 0x000000ff, {0.0, 0.0, 1.0, 0.0}},
-   {PIPE_FORMAT_A8R8G8B8_UNORM, 0x0000ff00, {0.0, 1.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A8R8G8B8_UNORM, 0x00ff0000, {1.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_A8R8G8B8_UNORM, 0xff000000, {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_A8R8G8B8_UNORM, 0xffffffff, {1.0, 1.0, 1.0, 1.0}},
+   {PIPE_FORMAT_B8G8R8A8_UNORM, 0x00000000, {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B8G8R8A8_UNORM, 0x000000ff, {0.0, 0.0, 1.0, 0.0}},
+   {PIPE_FORMAT_B8G8R8A8_UNORM, 0x0000ff00, {0.0, 1.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B8G8R8A8_UNORM, 0x00ff0000, {1.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_B8G8R8A8_UNORM, 0xff000000, {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_B8G8R8A8_UNORM, 0xffffffff, {1.0, 1.0, 1.0, 1.0}},
 
 #if 0
    {PIPE_FORMAT_R8G8B8A8_UNORM, 0x00000000, {0.0, 0.0, 0.0, 0.0}},
@@ -81,12 +81,12 @@ struct pixel_test_case test_cases[] =
    {PIPE_FORMAT_R8G8B8A8_UNORM, 0xffffffff, {1.0, 1.0, 1.0, 1.0}},
 #endif
 
-   {PIPE_FORMAT_B8G8R8A8_UNORM, 0x00000000, {0.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_B8G8R8A8_UNORM, 0x000000ff, {0.0, 0.0, 0.0, 1.0}},
-   {PIPE_FORMAT_B8G8R8A8_UNORM, 0x0000ff00, {1.0, 0.0, 0.0, 0.0}},
-   {PIPE_FORMAT_B8G8R8A8_UNORM, 0x00ff0000, {0.0, 1.0, 0.0, 0.0}},
-   {PIPE_FORMAT_B8G8R8A8_UNORM, 0xff000000, {0.0, 0.0, 1.0, 0.0}},
-   {PIPE_FORMAT_B8G8R8A8_UNORM, 0xffffffff, {1.0, 1.0, 1.0, 1.0}},
+   {PIPE_FORMAT_A8R8G8B8_UNORM, 0x00000000, {0.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_A8R8G8B8_UNORM, 0x000000ff, {0.0, 0.0, 0.0, 1.0}},
+   {PIPE_FORMAT_A8R8G8B8_UNORM, 0x0000ff00, {1.0, 0.0, 0.0, 0.0}},
+   {PIPE_FORMAT_A8R8G8B8_UNORM, 0x00ff0000, {0.0, 1.0, 0.0, 0.0}},
+   {PIPE_FORMAT_A8R8G8B8_UNORM, 0xff000000, {0.0, 0.0, 1.0, 0.0}},
+   {PIPE_FORMAT_A8R8G8B8_UNORM, 0xffffffff, {1.0, 1.0, 1.0, 1.0}},
 };
 
 
diff --git a/src/gallium/drivers/llvmpipe/lp_test_main.c b/src/gallium/drivers/llvmpipe/lp_test_main.c
index 14ff00469b0..f9dce8b9c25 100644
--- a/src/gallium/drivers/llvmpipe/lp_test_main.c
+++ b/src/gallium/drivers/llvmpipe/lp_test_main.c
@@ -37,7 +37,7 @@
 #include "util/u_cpu_detect.h"
 
 #include "gallivm/lp_bld_const.h"
-#include "gallivm/lp_bld_misc.h"
+#include "gallivm/lp_bld_init.h"
 #include "lp_test.h"
 
 
@@ -380,8 +380,7 @@ int main(int argc, char **argv)
          n = atoi(argv[i]);
    }
 
-   LLVMLinkInJIT();
-   LLVMInitializeNativeTarget();
+   lp_build_init();
 
    util_cpu_detect();
 
diff --git a/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c b/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c
index 2533275dc18..5a3cf37d6d8 100644
--- a/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c
+++ b/src/gallium/drivers/llvmpipe/lp_tex_sample_llvm.c
@@ -51,10 +51,11 @@
 
 
 /**
- * This provides the bridge between the sampler state store in lp_jit_context
- * and lp_jit_texture and the sampler code generator. It provides the
- * texture layout information required by the texture sampler code generator
- * in terms of the state stored in lp_jit_context and lp_jit_texture in runtime.
+ * This provides the bridge between the sampler state store in
+ * lp_jit_context and lp_jit_texture and the sampler code
+ * generator. It provides the texture layout information required by
+ * the texture sampler code generator in terms of the state stored in
+ * lp_jit_context and lp_jit_texture in runtime.
  */
 struct llvmpipe_sampler_dynamic_state
 {
@@ -79,6 +80,9 @@ struct lp_llvm_sampler_soa
 
 /**
  * Fetch the specified member of the lp_jit_texture structure.
+ * \param emit_load  if TRUE, emit the LLVM load instruction to actually
+ *                   fetch the field's value.  Otherwise, just emit the
+ *                   GEP code to address the field.
  *
  * @sa http://llvm.org/docs/GetElementPtr.html
  */
@@ -87,9 +91,11 @@ lp_llvm_texture_member(struct lp_sampler_dynamic_state *base,
                        LLVMBuilderRef builder,
                        unsigned unit,
                        unsigned member_index,
-                       const char *member_name)
+                       const char *member_name,
+                       boolean emit_load)
 {
-   struct llvmpipe_sampler_dynamic_state *state = (struct llvmpipe_sampler_dynamic_state *)base;
+   struct llvmpipe_sampler_dynamic_state *state =
+      (struct llvmpipe_sampler_dynamic_state *)base;
    LLVMValueRef indices[4];
    LLVMValueRef ptr;
    LLVMValueRef res;
@@ -107,7 +113,10 @@ lp_llvm_texture_member(struct lp_sampler_dynamic_state *base,
 
    ptr = LLVMBuildGEP(builder, state->context_ptr, indices, Elements(indices), "");
 
-   res = LLVMBuildLoad(builder, ptr, "");
+   if (emit_load)
+      res = LLVMBuildLoad(builder, ptr, "");
+   else
+      res = ptr;
 
    lp_build_name(res, "context.texture%u.%s", unit, member_name);
 
@@ -116,26 +125,30 @@ lp_llvm_texture_member(struct lp_sampler_dynamic_state *base,
 
 
 /**
- * Helper macro to instantiate the functions that generate the code to fetch
- * the members of lp_jit_texture to fulfill the sampler code generator requests.
+ * Helper macro to instantiate the functions that generate the code to
+ * fetch the members of lp_jit_texture to fulfill the sampler code
+ * generator requests.
  *
- * This complexity is the price we have to pay to keep the texture sampler code
- * generator a reusable module without dependencies to llvmpipe internals.
+ * This complexity is the price we have to pay to keep the texture
+ * sampler code generator a reusable module without dependencies to
+ * llvmpipe internals.
  */
-#define LP_LLVM_TEXTURE_MEMBER(_name, _index) \
+#define LP_LLVM_TEXTURE_MEMBER(_name, _index, _emit_load)  \
    static LLVMValueRef \
    lp_llvm_texture_##_name( struct lp_sampler_dynamic_state *base, \
                             LLVMBuilderRef builder, \
                             unsigned unit) \
    { \
-      return lp_llvm_texture_member(base, builder, unit, _index, #_name ); \
+      return lp_llvm_texture_member(base, builder, unit, _index, #_name, _emit_load ); \
    }
 
 
-LP_LLVM_TEXTURE_MEMBER(width,    LP_JIT_TEXTURE_WIDTH)
-LP_LLVM_TEXTURE_MEMBER(height,   LP_JIT_TEXTURE_HEIGHT)
-LP_LLVM_TEXTURE_MEMBER(stride,   LP_JIT_TEXTURE_STRIDE)
-LP_LLVM_TEXTURE_MEMBER(data_ptr, LP_JIT_TEXTURE_DATA)
+LP_LLVM_TEXTURE_MEMBER(width,      LP_JIT_TEXTURE_WIDTH, TRUE)
+LP_LLVM_TEXTURE_MEMBER(height,     LP_JIT_TEXTURE_HEIGHT, TRUE)
+LP_LLVM_TEXTURE_MEMBER(depth,      LP_JIT_TEXTURE_DEPTH, TRUE)
+LP_LLVM_TEXTURE_MEMBER(last_level, LP_JIT_TEXTURE_LAST_LEVEL, TRUE)
+LP_LLVM_TEXTURE_MEMBER(stride,     LP_JIT_TEXTURE_STRIDE, TRUE)
+LP_LLVM_TEXTURE_MEMBER(data_ptr,   LP_JIT_TEXTURE_DATA, FALSE)
 
 
 static void
@@ -145,6 +158,10 @@ lp_llvm_sampler_soa_destroy(struct lp_build_sampler_soa *sampler)
 }
 
 
+/**
+ * Fetch filtered values from texture.
+ * The 'texel' parameter returns four vectors corresponding to R, G, B, A.
+ */
 static void
 lp_llvm_sampler_soa_emit_fetch_texel(struct lp_build_sampler_soa *base,
                                      LLVMBuilderRef builder,
@@ -185,6 +202,8 @@ lp_llvm_sampler_soa_create(const struct lp_sampler_static_state *static_state,
    sampler->base.emit_fetch_texel = lp_llvm_sampler_soa_emit_fetch_texel;
    sampler->dynamic_state.base.width = lp_llvm_texture_width;
    sampler->dynamic_state.base.height = lp_llvm_texture_height;
+   sampler->dynamic_state.base.depth = lp_llvm_texture_depth;
+   sampler->dynamic_state.base.last_level = lp_llvm_texture_last_level;
    sampler->dynamic_state.base.stride = lp_llvm_texture_stride;
    sampler->dynamic_state.base.data_ptr = lp_llvm_texture_data_ptr;
    sampler->dynamic_state.static_state = static_state;
diff --git a/src/gallium/drivers/llvmpipe/lp_texture.c b/src/gallium/drivers/llvmpipe/lp_texture.c
index 022bf92cb46..74b7b4ec5e1 100644
--- a/src/gallium/drivers/llvmpipe/lp_texture.c
+++ b/src/gallium/drivers/llvmpipe/lp_texture.c
@@ -42,7 +42,7 @@
 #include "lp_screen.h"
 #include "lp_texture.h"
 #include "lp_tile_size.h"
-#include "lp_winsys.h"
+#include "state_tracker/sw_winsys.h"
 
 
 /**
@@ -93,7 +93,7 @@ static boolean
 llvmpipe_displaytarget_layout(struct llvmpipe_screen *screen,
                               struct llvmpipe_texture *lpt)
 {
-   struct llvmpipe_winsys *winsys = screen->winsys;
+   struct sw_winsys *winsys = screen->winsys;
 
    /* Round up the surface size to a multiple of the tile size to
     * avoid tile clipping.
@@ -124,14 +124,9 @@ llvmpipe_texture_create(struct pipe_screen *_screen,
    pipe_reference_init(&lpt->base.reference, 1);
    lpt->base.screen = &screen->base;
 
-   /* XXX: The xlib state tracker is brain-dead and will request
-    * PIPE_FORMAT_Z16_UNORM no matter how much we tell it we don't support it.
-    */
-   if (lpt->base.format == PIPE_FORMAT_Z16_UNORM)
-      lpt->base.format = PIPE_FORMAT_Z32_UNORM;
-
    if (lpt->base.tex_usage & (PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
-                              PIPE_TEXTURE_USAGE_PRIMARY)) {
+                              PIPE_TEXTURE_USAGE_SCANOUT |
+                              PIPE_TEXTURE_USAGE_SHARED)) {
       if (!llvmpipe_displaytarget_layout(screen, lpt))
          goto fail;
    }
@@ -148,43 +143,6 @@ llvmpipe_texture_create(struct pipe_screen *_screen,
 }
 
 
-static struct pipe_texture *
-llvmpipe_texture_blanket(struct pipe_screen * screen,
-                         const struct pipe_texture *base,
-                         const unsigned *stride,
-                         struct pipe_buffer *buffer)
-{
-   /* FIXME */
-#if 0
-   struct llvmpipe_texture *lpt;
-   assert(screen);
-
-   /* Only supports one type */
-   if (base->target != PIPE_TEXTURE_2D ||
-       base->last_level != 0 ||
-       base->depth0 != 1) {
-      return NULL;
-   }
-
-   lpt = CALLOC_STRUCT(llvmpipe_texture);
-   if (!lpt)
-      return NULL;
-
-   lpt->base = *base;
-   pipe_reference_init(&lpt->base.reference, 1);
-   lpt->base.screen = screen;
-   lpt->stride[0] = stride[0];
-
-   pipe_buffer_reference(&lpt->buffer, buffer);
-
-   return &lpt->base;
-#else
-   debug_printf("llvmpipe_texture_blanket() not implemented!");
-   return NULL;
-#endif
-}
-
-
 static void
 llvmpipe_texture_destroy(struct pipe_texture *pt)
 {
@@ -193,7 +151,7 @@ llvmpipe_texture_destroy(struct pipe_texture *pt)
 
    if (lpt->dt) {
       /* display target */
-      struct llvmpipe_winsys *winsys = screen->winsys;
+      struct sw_winsys *winsys = screen->winsys;
       winsys->displaytarget_destroy(winsys, lpt->dt);
    }
    else {
@@ -363,7 +321,7 @@ llvmpipe_transfer_map( struct pipe_screen *_screen,
 
    if (lpt->dt) {
       /* display target */
-      struct llvmpipe_winsys *winsys = screen->winsys;
+      struct sw_winsys *winsys = screen->winsys;
 
       map = winsys->displaytarget_map(winsys, lpt->dt,
                                       pipe_transfer_buffer_flags(transfer));
@@ -404,7 +362,7 @@ llvmpipe_transfer_unmap(struct pipe_screen *screen,
 
    if (lpt->dt) {
       /* display target */
-      struct llvmpipe_winsys *winsys = lp_screen->winsys;
+      struct sw_winsys *winsys = lp_screen->winsys;
       winsys->displaytarget_unmap(winsys, lpt->dt);
    }
 }
@@ -414,7 +372,6 @@ void
 llvmpipe_init_screen_texture_funcs(struct pipe_screen *screen)
 {
    screen->texture_create = llvmpipe_texture_create;
-   screen->texture_blanket = llvmpipe_texture_blanket;
    screen->texture_destroy = llvmpipe_texture_destroy;
 
    screen->get_tex_surface = llvmpipe_get_tex_surface;
diff --git a/src/gallium/drivers/llvmpipe/lp_texture.h b/src/gallium/drivers/llvmpipe/lp_texture.h
index 87c905bc027..b23f929b167 100644
--- a/src/gallium/drivers/llvmpipe/lp_texture.h
+++ b/src/gallium/drivers/llvmpipe/lp_texture.h
@@ -32,24 +32,29 @@
 #include "pipe/p_state.h"
 
 
+#define LP_MAX_TEXTURE_2D_LEVELS 13  /* 4K x 4K for now */
+#define LP_MAX_TEXTURE_3D_LEVELS 10  /* 512 x 512 x 512 for now */
+
+
 struct pipe_context;
 struct pipe_screen;
 struct llvmpipe_context;
-struct llvmpipe_displaytarget;
+
+struct sw_displaytarget;
 
 
 struct llvmpipe_texture
 {
    struct pipe_texture base;
 
-   unsigned long level_offset[PIPE_MAX_TEXTURE_LEVELS];
-   unsigned stride[PIPE_MAX_TEXTURE_LEVELS];
+   unsigned long level_offset[LP_MAX_TEXTURE_2D_LEVELS];
+   unsigned stride[LP_MAX_TEXTURE_2D_LEVELS];
 
    /**
     * Display target, for textures with the PIPE_TEXTURE_USAGE_DISPLAY_TARGET
     * usage.
     */
-   struct llvmpipe_displaytarget *dt;
+   struct sw_displaytarget *dt;
 
    /**
     * Malloc'ed data for regular textures, or a mapping to dt above.
diff --git a/src/gallium/drivers/llvmpipe/lp_tile_image.c b/src/gallium/drivers/llvmpipe/lp_tile_image.c
new file mode 100644
index 00000000000..c1980b316d5
--- /dev/null
+++ b/src/gallium/drivers/llvmpipe/lp_tile_image.c
@@ -0,0 +1,126 @@
+/**************************************************************************
+ * 
+ * Copyright 2010 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 THE 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.
+ * 
+ **************************************************************************/
+
+
+#include "lp_tile_soa.h"
+#include "lp_tile_image.h"
+
+
+#define BYTES_PER_TILE (TILE_SIZE * TILE_SIZE * 4)
+
+
+/**
+ * Convert a tiled image into a linear image.
+ * \param src_stride  source row stride in bytes (bytes per row of tiles)
+ * \param dst_stride  dest row stride in bytes
+ */
+void
+lp_tiled_to_linear(const uint8_t *src,
+                   uint8_t *dst,
+                   unsigned width, unsigned height,
+                   enum pipe_format format,
+                   unsigned src_stride,
+                   unsigned dst_stride)
+{
+   const unsigned tiles_per_row = src_stride / BYTES_PER_TILE;
+   unsigned i, j;
+
+   for (j = 0; j < height; j += TILE_SIZE) {
+      for (i = 0; i < width; i += TILE_SIZE) {
+         unsigned tile_offset =
+            ((j / TILE_SIZE) * tiles_per_row + i / TILE_SIZE);
+         unsigned byte_offset = tile_offset * BYTES_PER_TILE;
+         const uint8_t *src_tile = src + byte_offset;
+
+         lp_tile_write_4ub(format,
+                           src_tile,
+                           dst,
+                           dst_stride,
+                           i, j, TILE_SIZE, TILE_SIZE);
+      }
+   }
+}
+
+
+/**
+ * Convert a linear image into a tiled image.
+ * \param src_stride  source row stride in bytes
+ * \param dst_stride  dest row stride in bytes (bytes per row of tiles)
+ */
+void
+lp_linear_to_tiled(const uint8_t *src,
+                   uint8_t *dst,
+                   unsigned width, unsigned height,
+                   enum pipe_format format,
+                   unsigned src_stride,
+                   unsigned dst_stride)
+{
+   const unsigned tiles_per_row = dst_stride / BYTES_PER_TILE;
+   unsigned i, j;
+
+   for (j = 0; j < height; j += TILE_SIZE) {
+      for (i = 0; i < width; i += TILE_SIZE) {
+         unsigned tile_offset =
+            ((j / TILE_SIZE) * tiles_per_row + i / TILE_SIZE);
+         unsigned byte_offset = tile_offset * BYTES_PER_TILE;
+         uint8_t *dst_tile = dst + byte_offset;
+
+         lp_tile_read_4ub(format,
+                          dst_tile,
+                          src,
+                          src_stride,
+                          i, j, TILE_SIZE, TILE_SIZE);
+      }
+   }
+}
+
+
+/**
+ * For testing only.
+ */
+void
+test_tiled_linear_conversion(uint8_t *data,
+                             enum pipe_format format,
+                             unsigned width, unsigned height,
+                             unsigned stride)
+{
+   /* size in tiles */
+   unsigned wt = (width + TILE_SIZE - 1) / TILE_SIZE;
+   unsigned ht = (height + TILE_SIZE - 1) / TILE_SIZE;
+
+   uint8_t *tiled = malloc(wt * ht * TILE_SIZE * TILE_SIZE * 4);
+
+   unsigned tiled_stride = wt * TILE_SIZE * TILE_SIZE * 4;
+
+   lp_linear_to_tiled(data, tiled, width, height, format,
+                      stride, tiled_stride);
+
+   lp_tiled_to_linear(tiled, data, width, height, format,
+                      tiled_stride, stride);
+
+   free(tiled);
+}
+
diff --git a/src/gallium/drivers/llvmpipe/lp_tile_image.h b/src/gallium/drivers/llvmpipe/lp_tile_image.h
new file mode 100644
index 00000000000..60d472e8c5b
--- /dev/null
+++ b/src/gallium/drivers/llvmpipe/lp_tile_image.h
@@ -0,0 +1,57 @@
+/**************************************************************************
+ * 
+ * Copyright 2010 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 THE 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.
+ * 
+ **************************************************************************/
+
+
+#ifndef LP_TILE_IMAGE_H
+#define LP_TILE_IMAGE_H
+
+
+void
+lp_tiled_to_linear(const uint8_t *src,
+                   uint8_t *dst,
+                   unsigned width, unsigned height,
+                   enum pipe_format format,
+                   unsigned src_stride,
+                   unsigned dst_stride);
+
+
+void
+lp_linear_to_tiled(const uint8_t *src,
+                   uint8_t *dst,
+                   unsigned width, unsigned height,
+                   enum pipe_format format,
+                   unsigned src_stride,
+                   unsigned dst_stride);
+
+
+void
+test_tiled_linear_conversion(uint8_t *data,
+                             enum pipe_format format,
+                             unsigned width, unsigned height,
+                             unsigned stride);
+
+
+#endif /* LP_TILE_IMAGE_H */
diff --git a/src/gallium/drivers/llvmpipe/lp_tile_soa.py b/src/gallium/drivers/llvmpipe/lp_tile_soa.py
index 5d53689a3db..00b8d4fc382 100644
--- a/src/gallium/drivers/llvmpipe/lp_tile_soa.py
+++ b/src/gallium/drivers/llvmpipe/lp_tile_soa.py
@@ -45,10 +45,10 @@ sys.path.insert(0, os.path.join(os.path.dirname(sys.argv[0]), '../../auxiliary/u
 from u_format_access import *
 
 
-def generate_format_read(format, dst_type, dst_native_type, dst_suffix):
+def generate_format_read(format, dst_channel, dst_native_type, dst_suffix):
     '''Generate the function to read pixels from a particular format'''
 
-    name = short_name(format)
+    name = format.short_name()
 
     src_native_type = native_type(format)
 
@@ -64,11 +64,11 @@ def generate_format_read(format, dst_type, dst_native_type, dst_suffix):
     names = ['']*4
     if format.colorspace == 'rgb':
         for i in range(4):
-            swizzle = format.out_swizzle[i]
+            swizzle = format.swizzles[i]
             if swizzle < 4:
                 names[swizzle] += 'rgba'[i]
     elif format.colorspace == 'zs':
-        swizzle = format.out_swizzle[0]
+        swizzle = format.swizzles[0]
         if swizzle < 4:
             names[swizzle] = 'z'
         else:
@@ -76,48 +76,49 @@ def generate_format_read(format, dst_type, dst_native_type, dst_suffix):
     else:
         assert False
 
-    if format.layout == ARITH:
-        print '         %s pixel = *src_pixel++;' % src_native_type
-        shift = 0;
-        for i in range(4):
-            src_type = format.in_types[i]
-            width = src_type.size
-            if names[i]:
-                value = 'pixel'
-                mask = (1 << width) - 1
-                if shift:
-                    value = '(%s >> %u)' % (value, shift)
-                if shift + width < format.block_size():
-                    value = '(%s & 0x%x)' % (value, mask)
-                value = conversion_expr(src_type, dst_type, dst_native_type, value)
-                print '         %s %s = %s;' % (dst_native_type, names[i], value)
-            shift += width
-    elif format.layout == ARRAY:
-        for i in range(4):
-            src_type = format.in_types[i]
-            if names[i]:
-                value = '(*src_pixel++)'
-                value = conversion_expr(src_type, dst_type, dst_native_type, value)
-                print '         %s %s = %s;' % (dst_native_type, names[i], value)
+    if format.layout == PLAIN:
+        if not format.is_array():
+            print '         %s pixel = *src_pixel++;' % src_native_type
+            shift = 0;
+            for i in range(4):
+                src_channel = format.channels[i]
+                width = src_channel.size
+                if names[i]:
+                    value = 'pixel'
+                    mask = (1 << width) - 1
+                    if shift:
+                        value = '(%s >> %u)' % (value, shift)
+                    if shift + width < format.block_size():
+                        value = '(%s & 0x%x)' % (value, mask)
+                    value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
+                    print '         %s %s = %s;' % (dst_native_type, names[i], value)
+                shift += width
+        else:
+            for i in range(4):
+                src_channel = format.channels[i]
+                if names[i]:
+                    value = '(*src_pixel++)'
+                    value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
+                    print '         %s %s = %s;' % (dst_native_type, names[i], value)
     else:
         assert False
 
     for i in range(4):
         if format.colorspace == 'rgb':
-            swizzle = format.out_swizzle[i]
+            swizzle = format.swizzles[i]
             if swizzle < 4:
                 value = names[swizzle]
             elif swizzle == SWIZZLE_0:
                 value = '0'
             elif swizzle == SWIZZLE_1:
-                value = '1'
+                value = get_one(dst_channel)
             else:
                 assert False
         elif format.colorspace == 'zs':
             if i < 3:
                 value = 'z'
             else:
-                value = '1'
+                value = get_one(dst_channel)
         else:
             assert False
         print '         TILE_PIXEL(dst, x, y, %u) = %s; /* %s */' % (i, value, 'rgba'[i])
@@ -129,31 +130,16 @@ def generate_format_read(format, dst_type, dst_native_type, dst_suffix):
     print
     
 
-def compute_inverse_swizzle(format):
-    '''Return an array[4] of inverse swizzle terms'''
-    inv_swizzle = [None]*4
-    if format.colorspace == 'rgb':
-        for i in range(4):
-            swizzle = format.out_swizzle[i]
-            if swizzle < 4:
-                inv_swizzle[swizzle] = i
-    elif format.colorspace == 'zs':
-        swizzle = format.out_swizzle[0]
-        if swizzle < 4:
-            inv_swizzle[swizzle] = 0
-    return inv_swizzle
-
-
-def pack_rgba(format, src_type, r, g, b, a):
+def pack_rgba(format, src_channel, r, g, b, a):
     """Return an expression for packing r, g, b, a into a pixel of the
     given format.  Ex: '(b << 24) | (g << 16) | (r << 8) | (a << 0)'
     """
     assert format.colorspace == 'rgb'
-    inv_swizzle = compute_inverse_swizzle(format)
+    inv_swizzle = format.inv_swizzles()
     shift = 0
     expr = None
     for i in range(4):
-		# choose r, g, b, or a depending on the inverse swizzle term
+        # choose r, g, b, or a depending on the inverse swizzle term
         if inv_swizzle[i] == 0:
             value = r
         elif inv_swizzle[i] == 1:
@@ -166,25 +152,25 @@ def pack_rgba(format, src_type, r, g, b, a):
             value = None
 
         if value:
-            dst_type = format.in_types[i]
+            dst_channel = format.channels[i]
             dst_native_type = native_type(format)
-            value = conversion_expr(src_type, dst_type, dst_native_type, value)
+            value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
             term = "((%s) << %d)" % (value, shift)
             if expr:
                 expr = expr + " | " + term
             else:
                 expr = term
 
-        width = format.in_types[i].size
+        width = format.channels[i].size
         shift = shift + width
     return expr
 
 
-def emit_unrolled_write_code(format, src_type):
+def emit_unrolled_write_code(format, src_channel):
     '''Emit code for writing a block based on unrolled loops.
     This is considerably faster than the TILE_PIXEL-based code below.
     '''
-    dst_native_type = native_type(format)
+    dst_native_type = 'uint%u_t' % format.block_size()
     print '   const unsigned dstpix_stride = dst_stride / %d;' % format.stride()
     print '   %s *dstpix = (%s *) dst;' % (dst_native_type, dst_native_type)
     print '   unsigned int qx, qy, i;'
@@ -199,8 +185,8 @@ def emit_unrolled_write_code(format, src_type):
     print '         const uint8_t *a = src + 3 * TILE_C_STRIDE;'
     print '         (void) r; (void) g; (void) b; (void) a; /* silence warnings */'
     print '         for (i = 0; i < TILE_C_STRIDE; i += 2) {'
-    print '            const uint32_t pixel0 = %s;' % pack_rgba(format, src_type, "r[i+0]", "g[i+0]", "b[i+0]", "a[i+0]")
-    print '            const uint32_t pixel1 = %s;' % pack_rgba(format, src_type, "r[i+1]", "g[i+1]", "b[i+1]", "a[i+1]")
+    print '            const uint32_t pixel0 = %s;' % pack_rgba(format, src_channel, "r[i+0]", "g[i+0]", "b[i+0]", "a[i+0]")
+    print '            const uint32_t pixel1 = %s;' % pack_rgba(format, src_channel, "r[i+1]", "g[i+1]", "b[i+1]", "a[i+1]")
     print '            const unsigned offset = (py + tile_y_offset[i]) * dstpix_stride + (px + tile_x_offset[i]);'
     print '            dstpix[offset + 0] = pixel0;'
     print '            dstpix[offset + 1] = pixel1;'
@@ -210,11 +196,11 @@ def emit_unrolled_write_code(format, src_type):
     print '   }'
 
 
-def emit_tile_pixel_write_code(format, src_type):
+def emit_tile_pixel_write_code(format, src_channel):
     '''Emit code for writing a block based on the TILE_PIXEL macro.'''
     dst_native_type = native_type(format)
 
-    inv_swizzle = compute_inverse_swizzle(format)
+    inv_swizzle = format.inv_swizzles()
 
     print '   unsigned x, y;'
     print '   uint8_t *dst_row = dst + y0*dst_stride;'
@@ -222,27 +208,28 @@ def emit_tile_pixel_write_code(format, src_type):
     print '      %s *dst_pixel = (%s *)(dst_row + x0*%u);' % (dst_native_type, dst_native_type, format.stride())
     print '      for (x = 0; x < w; ++x) {'
 
-    if format.layout == ARITH:
-        print '         %s pixel = 0;' % dst_native_type
-        shift = 0;
-        for i in range(4):
-            dst_type = format.in_types[i]
-            width = dst_type.size
-            if inv_swizzle[i] is not None:
-                value = 'TILE_PIXEL(src, x, y, %u)' % inv_swizzle[i]
-                value = conversion_expr(src_type, dst_type, dst_native_type, value)
-                if shift:
-                    value = '(%s << %u)' % (value, shift)
-                print '         pixel |= %s;' % value
-            shift += width
-        print '         *dst_pixel++ = pixel;'
-    elif format.layout == ARRAY:
-        for i in range(4):
-            dst_type = format.in_types[i]
-            if inv_swizzle[i] is not None:
-                value = 'TILE_PIXEL(src, x, y, %u)' % inv_swizzle[i]
-                value = conversion_expr(src_type, dst_type, dst_native_type, value)
-                print '         *dst_pixel++ = %s;' % value
+    if format.layout == PLAIN:
+        if not format.is_array():
+            print '         %s pixel = 0;' % dst_native_type
+            shift = 0;
+            for i in range(4):
+                dst_channel = format.channels[i]
+                width = dst_channel.size
+                if inv_swizzle[i] is not None:
+                    value = 'TILE_PIXEL(src, x, y, %u)' % inv_swizzle[i]
+                    value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
+                    if shift:
+                        value = '(%s << %u)' % (value, shift)
+                    print '         pixel |= %s;' % value
+                shift += width
+            print '         *dst_pixel++ = pixel;'
+        else:
+            for i in range(4):
+                dst_channel = format.channels[i]
+                if inv_swizzle[i] is not None:
+                    value = 'TILE_PIXEL(src, x, y, %u)' % inv_swizzle[i]
+                    value = conversion_expr(src_channel, dst_channel, dst_native_type, value, clamp=False)
+                    print '         *dst_pixel++ = %s;' % value
     else:
         assert False
 
@@ -251,28 +238,33 @@ def emit_tile_pixel_write_code(format, src_type):
     print '   }'
 
 
-def generate_format_write(format, src_type, src_native_type, src_suffix):
+def generate_format_write(format, src_channel, src_native_type, src_suffix):
     '''Generate the function to write pixels to a particular format'''
 
-    name = short_name(format)
+    name = format.short_name()
 
     print 'static void'
     print 'lp_tile_%s_write_%s(const %s *src, uint8_t *dst, unsigned dst_stride, unsigned x0, unsigned y0, unsigned w, unsigned h)' % (name, src_suffix, src_native_type)
     print '{'
-    if format.layout == ARITH and format.colorspace == 'rgb':
-        emit_unrolled_write_code(format, src_type)
+    if format.layout == PLAIN \
+        and format.colorspace == 'rgb' \
+        and format.block_size() <= 32 \
+        and format.is_pot() \
+        and not format.is_mixed() \
+        and format.channels[0].type == UNSIGNED:
+        emit_unrolled_write_code(format, src_channel)
     else:
-        emit_tile_pixel_write_code(format, src_type)
+        emit_tile_pixel_write_code(format, src_channel)
     print '}'
     print
     
 
-def generate_read(formats, dst_type, dst_native_type, dst_suffix):
+def generate_read(formats, dst_channel, dst_native_type, dst_suffix):
     '''Generate the dispatch function to read pixels from any format'''
 
     for format in formats:
         if is_format_supported(format):
-            generate_format_read(format, dst_type, dst_native_type, dst_suffix)
+            generate_format_read(format, dst_channel, dst_native_type, dst_suffix)
 
     print 'void'
     print 'lp_tile_read_%s(enum pipe_format format, %s *dst, const void *src, unsigned src_stride, unsigned x, unsigned y, unsigned w, unsigned h)' % (dst_suffix, dst_native_type)
@@ -282,7 +274,7 @@ def generate_read(formats, dst_type, dst_native_type, dst_suffix):
     for format in formats:
         if is_format_supported(format):
             print '   case %s:' % format.name
-            print '      func = &lp_tile_%s_read_%s;' % (short_name(format), dst_suffix)
+            print '      func = &lp_tile_%s_read_%s;' % (format.short_name(), dst_suffix)
             print '      break;'
     print '   default:'
     print '      debug_printf("unsupported format\\n");'
@@ -293,12 +285,12 @@ def generate_read(formats, dst_type, dst_native_type, dst_suffix):
     print
 
 
-def generate_write(formats, src_type, src_native_type, src_suffix):
+def generate_write(formats, src_channel, src_native_type, src_suffix):
     '''Generate the dispatch function to write pixels to any format'''
 
     for format in formats:
         if is_format_supported(format):
-            generate_format_write(format, src_type, src_native_type, src_suffix)
+            generate_format_write(format, src_channel, src_native_type, src_suffix)
 
     print 'void'
     print 'lp_tile_write_%s(enum pipe_format format, const %s *src, void *dst, unsigned dst_stride, unsigned x, unsigned y, unsigned w, unsigned h)' % (src_suffix, src_native_type)
@@ -309,7 +301,7 @@ def generate_write(formats, src_type, src_native_type, src_suffix):
     for format in formats:
         if is_format_supported(format):
             print '   case %s:' % format.name
-            print '      func = &lp_tile_%s_write_%s;' % (short_name(format), src_suffix)
+            print '      func = &lp_tile_%s_write_%s;' % (format.short_name(), src_suffix)
             print '      break;'
     print '   default:'
     print '      debug_printf("unsupported format\\n");'
@@ -359,12 +351,12 @@ def main():
 
     generate_clamp()
 
-    type = Type(UNSIGNED, True, 8)
+    channel = Channel(UNSIGNED, True, 8)
     native_type = 'uint8_t'
     suffix = '4ub'
 
-    generate_read(formats, type, native_type, suffix)
-    generate_write(formats, type, native_type, suffix)
+    generate_read(formats, channel, native_type, suffix)
+    generate_write(formats, channel, native_type, suffix)
 
 
 if __name__ == '__main__':
diff --git a/src/gallium/drivers/llvmpipe/lp_winsys.h b/src/gallium/drivers/llvmpipe/lp_winsys.h
deleted file mode 100644
index ce11fa93041..00000000000
--- a/src/gallium/drivers/llvmpipe/lp_winsys.h
+++ /dev/null
@@ -1,125 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2007-2009 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.
- * 
- **************************************************************************/
-
-/**
- * @file
- * llvmpipe public interface.
- */
-
-
-#ifndef LP_WINSYS_H
-#define LP_WINSYS_H
-
-
-#include "pipe/p_compiler.h" /* for boolean */
-#include "pipe/p_format.h"
-
-
-#ifdef __cplusplus
-extern "C" {
-#endif
-
-
-struct pipe_screen;
-struct pipe_context;
-
-
-/**
- * Opaque pointer.
- */
-struct llvmpipe_displaytarget;
-
-
-/**
- * This is the interface that llvmpipe expects any window system
- * hosting it to implement.
- * 
- * llvmpipe is for the most part a self sufficient driver. The only thing it
- * does not know is how to display a surface.
- */
-struct llvmpipe_winsys
-{
-   void 
-   (*destroy)( struct llvmpipe_winsys *ws );
-
-   boolean
-   (*is_displaytarget_format_supported)( struct llvmpipe_winsys *ws,
-                                         enum pipe_format format );
-   
-   /**
-    * Allocate storage for a render target.
-    * 
-    * Often surfaces which are meant to be blitted to the front screen (i.e.,
-    * display targets) must be allocated with special characteristics, memory 
-    * pools, or obtained directly from the windowing system.
-    *  
-    * This callback is invoked by the pipe_screen when creating a texture marked
-    * with the PIPE_TEXTURE_USAGE_DISPLAY_TARGET flag to get the underlying 
-    * storage.
-    */
-   struct llvmpipe_displaytarget *
-   (*displaytarget_create)( struct llvmpipe_winsys *ws,
-                            enum pipe_format format,
-                            unsigned width, unsigned height,
-                            unsigned alignment,
-                            unsigned *stride );
-
-   void *
-   (*displaytarget_map)( struct llvmpipe_winsys *ws, 
-                         struct llvmpipe_displaytarget *dt,
-                         unsigned flags );
-
-   void
-   (*displaytarget_unmap)( struct llvmpipe_winsys *ws,
-                           struct llvmpipe_displaytarget *dt );
-
-   /**
-    * @sa pipe_screen:flush_frontbuffer.
-    *
-    * This call will likely become asynchronous eventually.
-    */
-   void
-   (*displaytarget_display)( struct llvmpipe_winsys *ws, 
-                             struct llvmpipe_displaytarget *dt,
-                             void *context_private );
-
-   void 
-   (*displaytarget_destroy)( struct llvmpipe_winsys *ws, 
-                             struct llvmpipe_displaytarget *dt );
-};
-
-
-
-struct pipe_screen *
-llvmpipe_create_screen( struct llvmpipe_winsys * );
-
-
-#ifdef __cplusplus
-}
-#endif
-
-#endif /* LP_WINSYS_H */
diff --git a/src/gallium/drivers/nouveau/nouveau_context.c b/src/gallium/drivers/nouveau/nouveau_context.c
index 23443869e68..15174983e7f 100644
--- a/src/gallium/drivers/nouveau/nouveau_context.c
+++ b/src/gallium/drivers/nouveau/nouveau_context.c
@@ -1,5 +1,5 @@
-#include <pipe/p_defines.h>
-#include <pipe/p_context.h>
+#include "pipe/p_defines.h"
+#include "pipe/p_context.h"
 
 #include "nouveau/nouveau_screen.h"
 #include "nouveau/nouveau_context.h"
diff --git a/src/gallium/drivers/nouveau/nouveau_screen.c b/src/gallium/drivers/nouveau/nouveau_screen.c
index 81bc296ab44..b1ad686022a 100644
--- a/src/gallium/drivers/nouveau/nouveau_screen.c
+++ b/src/gallium/drivers/nouveau/nouveau_screen.c
@@ -1,9 +1,10 @@
-#include <pipe/p_defines.h>
-#include <pipe/p_screen.h>
-#include <pipe/p_state.h>
+#include "pipe/p_defines.h"
+#include "pipe/p_screen.h"
+#include "pipe/p_state.h"
 
-#include <util/u_memory.h>
-#include <util/u_inlines.h>
+#include "util/u_memory.h"
+#include "util/u_inlines.h"
+#include "util/u_format.h"
 
 #include <stdio.h>
 #include <errno.h>
@@ -12,6 +13,9 @@
 #include "nouveau_winsys.h"
 #include "nouveau_screen.h"
 
+/* XXX this should go away */
+#include "state_tracker/drm_api.h"
+
 static const char *
 nouveau_screen_get_name(struct pipe_screen *pscreen)
 {
@@ -120,7 +124,7 @@ nouveau_screen_map_flags(unsigned pipe)
 	if (pipe & PIPE_BUFFER_USAGE_DONTBLOCK)
 		flags |= NOUVEAU_BO_NOWAIT;
 	else
-	if (pipe & 0 /*PIPE_BUFFER_USAGE_UNSYNCHRONIZED*/)
+	if (pipe & PIPE_BUFFER_USAGE_UNSYNCHRONIZED)
 		flags |= NOUVEAU_BO_NOSYNC;
 
 	return flags;
@@ -231,6 +235,72 @@ nouveau_screen_fence_finish(struct pipe_screen *screen,
 	return 0;
 }
 
+
+/*
+ * Both texture_{from|get}_handle use drm api defines directly which they
+ * shouldn't do. The problem is that from|get are pipe functions and as
+ * such they should be defined in the pipe level. If nouveau had a propper
+ * winsys interface we would have added from|get to that interface using
+ * the winsys_handle struct as done with other drivers. However this code
+ * calls directly into the libdrm_nouveau.so functions (nouveau_bo_*). So
+ * we need to translate the handle into something they understand.
+ */
+static struct pipe_texture *
+nouveau_screen_texture_from_handle(struct pipe_screen *pscreen,
+				   const struct pipe_texture *templ,
+				   struct winsys_handle *whandle)
+{
+	struct nouveau_device *dev = nouveau_screen(pscreen)->device;
+	struct pipe_texture *pt;
+	struct pipe_buffer *pb;
+	int ret;
+
+	pb = CALLOC(1, sizeof(struct pipe_buffer) + sizeof(struct nouveau_bo*));
+	if (!pb)
+		return NULL;
+
+	ret = nouveau_bo_handle_ref(dev, whandle->handle, (struct nouveau_bo**)(pb+1));
+	if (ret) {
+		debug_printf("%s: ref name 0x%08x failed with %d\n",
+			     __func__, whandle->handle, ret);
+		FREE(pb);
+		return NULL;
+	}
+
+	pipe_reference_init(&pb->reference, 1);
+	pb->screen = pscreen;
+	pb->alignment = 0;
+	pb->usage = PIPE_BUFFER_USAGE_GPU_READ_WRITE |
+		    PIPE_BUFFER_USAGE_CPU_READ_WRITE;
+	pb->size = nouveau_bo(pb)->size;
+	pt = nouveau_screen(pscreen)->texture_blanket(pscreen, templ,
+                                                      &whandle->stride, pb);
+	pipe_buffer_reference(&pb, NULL);
+	return pt;
+}
+
+static boolean
+nouveau_screen_texture_get_handle(struct pipe_screen *pscreen,
+				  struct pipe_texture *pt,
+				  struct winsys_handle *whandle)
+{
+	struct nouveau_miptree *mt = nouveau_miptree(pt);
+
+	if (!mt || !mt->bo)
+		return false;
+
+	whandle->stride = util_format_get_stride(mt->base.format, mt->base.width0);
+
+	if (whandle->type == DRM_API_HANDLE_TYPE_SHARED) { 
+		return nouveau_bo_handle_get(mt->bo, &whandle->handle) == 0;
+	} else if (whandle->type == DRM_API_HANDLE_TYPE_KMS) {
+		whandle->handle = mt->bo->handle;
+		return TRUE;
+	} else {
+		return FALSE;
+	}
+}
+
 int
 nouveau_screen_init(struct nouveau_screen *screen, struct nouveau_device *dev)
 {
@@ -258,6 +328,9 @@ nouveau_screen_init(struct nouveau_screen *screen, struct nouveau_device *dev)
 	pscreen->fence_signalled = nouveau_screen_fence_signalled;
 	pscreen->fence_finish = nouveau_screen_fence_finish;
 
+	pscreen->texture_from_handle = nouveau_screen_texture_from_handle;
+	pscreen->texture_get_handle = nouveau_screen_texture_get_handle;
+
 	return 0;
 }
 
diff --git a/src/gallium/drivers/nouveau/nouveau_screen.h b/src/gallium/drivers/nouveau/nouveau_screen.h
index a7927d88dfc..f4a7a2bc234 100644
--- a/src/gallium/drivers/nouveau/nouveau_screen.h
+++ b/src/gallium/drivers/nouveau/nouveau_screen.h
@@ -6,6 +6,18 @@ struct nouveau_screen {
 	struct nouveau_device *device;
 	struct nouveau_channel *channel;
 
+        /**
+         * Create a new texture object, using the given template info, but on top of
+         * existing memory.
+         * 
+         * It is assumed that the buffer data is layed out according to the expected
+         * by the hardware. NULL will be returned if any inconsistency is found.  
+         */
+        struct pipe_texture * (*texture_blanket)(struct pipe_screen *,
+                                                 const struct pipe_texture *templat,
+                                                 const unsigned *stride,
+                                                 struct pipe_buffer *buffer);
+
 	int (*pre_pipebuffer_map_callback) (struct pipe_screen *pscreen,
 		struct pipe_buffer *pb, unsigned usage);
 };
diff --git a/src/gallium/drivers/nouveau/nouveau_util.h b/src/gallium/drivers/nouveau/nouveau_util.h
index a10114beab9..7f16e31c3f0 100644
--- a/src/gallium/drivers/nouveau/nouveau_util.h
+++ b/src/gallium/drivers/nouveau/nouveau_util.h
@@ -88,4 +88,104 @@ static INLINE unsigned log2i(unsigned i)
 	return r;
 }
 
+struct u_split_prim {
+   void *priv;
+   void (*emit)(void *priv, unsigned start, unsigned count);
+   void (*edge)(void *priv, boolean enabled);
+
+   unsigned mode;
+   unsigned start;
+   unsigned p_start;
+   unsigned p_end;
+
+   int repeat_first:1;
+   int close_first:1;
+   int edgeflag_off:1;
+};
+
+static inline void
+u_split_prim_init(struct u_split_prim *s,
+                  unsigned mode, unsigned start, unsigned count)
+{
+   if (mode == PIPE_PRIM_LINE_LOOP) {
+      s->mode = PIPE_PRIM_LINE_STRIP;
+      s->close_first = 1;
+   } else {
+      s->mode = mode;
+      s->close_first = 0;
+   }
+   s->start = start;
+   s->p_start = start;
+   s->p_end = start + count;
+   s->edgeflag_off = 0;
+   s->repeat_first = 0;
+}
+
+static INLINE boolean
+u_split_prim_next(struct u_split_prim *s, unsigned max_verts)
+{
+   int repeat = 0;
+
+   if (s->repeat_first) {
+      s->emit(s->priv, s->start, 1);
+      max_verts--;
+      if (s->edgeflag_off) {
+         s->edge(s->priv, TRUE);
+         s->edgeflag_off = FALSE;
+      }
+   }
+
+   if (s->p_start + s->close_first + max_verts >= s->p_end) {
+      s->emit(s->priv, s->p_start, s->p_end - s->p_start);
+      if (s->close_first)
+         s->emit(s->priv, s->start, 1);
+      return TRUE;
+   }
+
+   switch (s->mode) {
+   case PIPE_PRIM_LINES:
+      max_verts &= ~1;
+      break;
+   case PIPE_PRIM_LINE_STRIP:
+      repeat = 1;
+      break;
+   case PIPE_PRIM_POLYGON:
+      max_verts--;
+      s->emit(s->priv, s->p_start, max_verts);
+      s->edge(s->priv, FALSE);
+      s->emit(s->priv, s->p_start + max_verts, 1);
+      s->p_start += max_verts;
+      s->repeat_first = TRUE;
+      s->edgeflag_off = TRUE;
+      return FALSE;
+   case PIPE_PRIM_TRIANGLES:
+      max_verts = max_verts - (max_verts % 3);
+      break;
+   case PIPE_PRIM_TRIANGLE_STRIP:
+      /* to ensure winding stays correct, always split
+       * on an even number of generated triangles
+       */
+      max_verts = max_verts & ~1;
+      repeat = 2;
+      break;
+   case PIPE_PRIM_TRIANGLE_FAN:
+      s->repeat_first = TRUE;
+      repeat = 1;
+      break;
+   case PIPE_PRIM_QUADS:
+      max_verts &= ~3;
+      break;
+   case PIPE_PRIM_QUAD_STRIP:
+      max_verts &= ~1;
+      repeat = 2;
+      break;
+   default:
+      break;
+   }
+
+   s->emit (s->priv, s->p_start, max_verts);
+   s->p_start += (max_verts - repeat);
+   return FALSE;
+}
+
 #endif
diff --git a/src/gallium/drivers/nouveau/nv04_surface_2d.c b/src/gallium/drivers/nouveau/nv04_surface_2d.c
index 42c2ca932d8..b074547c4da 100644
--- a/src/gallium/drivers/nouveau/nv04_surface_2d.c
+++ b/src/gallium/drivers/nouveau/nv04_surface_2d.c
@@ -18,15 +18,15 @@ nv04_surface_format(enum pipe_format format)
 	case PIPE_FORMAT_I8_UNORM:
 		return NV04_CONTEXT_SURFACES_2D_FORMAT_Y8;
 	case PIPE_FORMAT_R16_SNORM:
-	case PIPE_FORMAT_R5G6B5_UNORM:
+	case PIPE_FORMAT_B5G6R5_UNORM:
 	case PIPE_FORMAT_Z16_UNORM:
-	case PIPE_FORMAT_A8L8_UNORM:
+	case PIPE_FORMAT_L8A8_UNORM:
 		return NV04_CONTEXT_SURFACES_2D_FORMAT_R5G6B5;
-	case PIPE_FORMAT_X8R8G8B8_UNORM:
-	case PIPE_FORMAT_A8R8G8B8_UNORM:
+	case PIPE_FORMAT_B8G8R8X8_UNORM:
+	case PIPE_FORMAT_B8G8R8A8_UNORM:
 		return NV04_CONTEXT_SURFACES_2D_FORMAT_A8R8G8B8;
-	case PIPE_FORMAT_Z24S8_UNORM:
-	case PIPE_FORMAT_Z24X8_UNORM:
+	case PIPE_FORMAT_S8Z24_UNORM:
+	case PIPE_FORMAT_X8Z24_UNORM:
 		return NV04_CONTEXT_SURFACES_2D_FORMAT_Y32;
 	default:
 		return -1;
@@ -39,14 +39,14 @@ nv04_rect_format(enum pipe_format format)
 	switch (format) {
 	case PIPE_FORMAT_A8_UNORM:
 		return NV04_GDI_RECTANGLE_TEXT_COLOR_FORMAT_A8R8G8B8;
-	case PIPE_FORMAT_R5G6B5_UNORM:
-	case PIPE_FORMAT_A8L8_UNORM:
+	case PIPE_FORMAT_B5G6R5_UNORM:
+	case PIPE_FORMAT_L8A8_UNORM:
 	case PIPE_FORMAT_Z16_UNORM:
 		return NV04_GDI_RECTANGLE_TEXT_COLOR_FORMAT_A16R5G6B5;
-	case PIPE_FORMAT_X8R8G8B8_UNORM:
-	case PIPE_FORMAT_A8R8G8B8_UNORM:
-	case PIPE_FORMAT_Z24S8_UNORM:
-	case PIPE_FORMAT_Z24X8_UNORM:
+	case PIPE_FORMAT_B8G8R8X8_UNORM:
+	case PIPE_FORMAT_B8G8R8A8_UNORM:
+	case PIPE_FORMAT_S8Z24_UNORM:
+	case PIPE_FORMAT_X8Z24_UNORM:
 		return NV04_GDI_RECTANGLE_TEXT_COLOR_FORMAT_A8R8G8B8;
 	default:
 		return -1;
@@ -61,15 +61,15 @@ nv04_scaled_image_format(enum pipe_format format)
 	case PIPE_FORMAT_L8_UNORM:
 	case PIPE_FORMAT_I8_UNORM:
 		return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_Y8;
-	case PIPE_FORMAT_A1R5G5B5_UNORM:
+	case PIPE_FORMAT_B5G5R5A1_UNORM:
 		return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_A1R5G5B5;
-	case PIPE_FORMAT_A8R8G8B8_UNORM:
+	case PIPE_FORMAT_B8G8R8A8_UNORM:
 		return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_A8R8G8B8;
-	case PIPE_FORMAT_X8R8G8B8_UNORM:
+	case PIPE_FORMAT_B8G8R8X8_UNORM:
 		return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_X8R8G8B8;
-	case PIPE_FORMAT_R5G6B5_UNORM:
+	case PIPE_FORMAT_B5G6R5_UNORM:
 	case PIPE_FORMAT_R16_SNORM:
-	case PIPE_FORMAT_A8L8_UNORM:
+	case PIPE_FORMAT_L8A8_UNORM:
 		return NV03_SCALED_IMAGE_FROM_MEMORY_COLOR_FORMAT_R5G6B5;
 	default:
 		return -1;
diff --git a/src/gallium/drivers/nv30/nv30_context.h b/src/gallium/drivers/nv30/nv30_context.h
index 04813526f40..34b5953ccf1 100644
--- a/src/gallium/drivers/nv30/nv30_context.h
+++ b/src/gallium/drivers/nv30/nv30_context.h
@@ -101,13 +101,17 @@ struct nv30_blend_state {
 struct nv30_state {
 	unsigned scissor_enabled;
 	unsigned stipple_enabled;
-	unsigned viewport_bypass;
 	unsigned fp_samplers;
 
 	uint64_t dirty;
 	struct nouveau_stateobj *hw[NV30_STATE_MAX];
 };
 
+struct nv30_vtxelt_state {
+	struct pipe_vertex_element pipe[16];
+	unsigned num_elements;
+};
+
 struct nv30_context {
 	struct pipe_context pipe;
 
@@ -144,8 +148,7 @@ struct nv30_context {
 	unsigned dirty_samplers;
 	struct pipe_vertex_buffer vtxbuf[PIPE_MAX_ATTRIBS];
 	unsigned vtxbuf_nr;
-	struct pipe_vertex_element vtxelt[PIPE_MAX_ATTRIBS];
-	unsigned vtxelt_nr;
+	struct nv30_vtxelt_state *vtxelt;
 };
 
 static INLINE struct nv30_context *
diff --git a/src/gallium/drivers/nv30/nv30_fragtex.c b/src/gallium/drivers/nv30/nv30_fragtex.c
index 9f4a104f673..f7d98f3f208 100644
--- a/src/gallium/drivers/nv30/nv30_fragtex.c
+++ b/src/gallium/drivers/nv30/nv30_fragtex.c
@@ -23,17 +23,17 @@ struct nv30_texture_format {
 
 static struct nv30_texture_format
 nv30_texture_formats[] = {
-	_(X8R8G8B8_UNORM, A8R8G8B8,   S1,   S1,   S1,  ONE, X, Y, Z, W),
-	_(A8R8G8B8_UNORM, A8R8G8B8,   S1,   S1,   S1,   S1, X, Y, Z, W),
-	_(A1R5G5B5_UNORM, A1R5G5B5,   S1,   S1,   S1,   S1, X, Y, Z, W),
-	_(A4R4G4B4_UNORM, A4R4G4B4,   S1,   S1,   S1,   S1, X, Y, Z, W),
-	_(R5G6B5_UNORM  , R5G6B5  ,   S1,   S1,   S1,  ONE, X, Y, Z, W),
+	_(B8G8R8X8_UNORM, A8R8G8B8,   S1,   S1,   S1,  ONE, X, Y, Z, W),
+	_(B8G8R8A8_UNORM, A8R8G8B8,   S1,   S1,   S1,   S1, X, Y, Z, W),
+	_(B5G5R5A1_UNORM, A1R5G5B5,   S1,   S1,   S1,   S1, X, Y, Z, W),
+	_(B4G4R4A4_UNORM, A4R4G4B4,   S1,   S1,   S1,   S1, X, Y, Z, W),
+	_(B5G6R5_UNORM  , R5G6B5  ,   S1,   S1,   S1,  ONE, X, Y, Z, W),
 	_(L8_UNORM      , L8      ,   S1,   S1,   S1,  ONE, X, X, X, X),
 	_(A8_UNORM      , L8      , ZERO, ZERO, ZERO,   S1, X, X, X, X),
 	_(I8_UNORM      , L8      ,   S1,   S1,   S1,   S1, X, X, X, X),
-	_(A8L8_UNORM    , A8L8    ,   S1,   S1,   S1,   S1, X, X, X, Y),
+	_(L8A8_UNORM    , A8L8    ,   S1,   S1,   S1,   S1, X, X, X, Y),
 	_(Z16_UNORM     , R5G6B5  ,   S1,   S1,   S1,  ONE, X, X, X, X),
-	_(Z24S8_UNORM   , A8R8G8B8,   S1,   S1,   S1,  ONE, X, X, X, X),
+	_(S8Z24_UNORM   , A8R8G8B8,   S1,   S1,   S1,  ONE, X, X, X, X),
 	_(DXT1_RGB      , DXT1    ,   S1,   S1,   S1,  ONE, X, Y, Z, W),
 	_(DXT1_RGBA     , DXT1    ,   S1,   S1,   S1,   S1, X, Y, Z, W),
 	_(DXT3_RGBA     , DXT3    ,   S1,   S1,   S1,   S1, X, Y, Z, W),
diff --git a/src/gallium/drivers/nv30/nv30_miptree.c b/src/gallium/drivers/nv30/nv30_miptree.c
index c29c36e20aa..bfa27b632f7 100644
--- a/src/gallium/drivers/nv30/nv30_miptree.c
+++ b/src/gallium/drivers/nv30/nv30_miptree.c
@@ -18,7 +18,7 @@ nv30_miptree_layout(struct nv30_miptree *nv30mt)
 		                           PIPE_TEXTURE_USAGE_DEPTH_STENCIL |
 		                           PIPE_TEXTURE_USAGE_RENDER_TARGET |
 		                           PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
-		                           PIPE_TEXTURE_USAGE_PRIMARY);
+		                           PIPE_TEXTURE_USAGE_SCANOUT);
 
 	if (pt->target == PIPE_TEXTURE_CUBE) {
 		nr_faces = 6;
@@ -78,7 +78,7 @@ nv30_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 	    pt->height0 & (pt->height0 - 1))
 		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
 	else
-	if (pt->tex_usage & (PIPE_TEXTURE_USAGE_PRIMARY |
+	if (pt->tex_usage & (PIPE_TEXTURE_USAGE_SCANOUT |
 	                     PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
 	                     PIPE_TEXTURE_USAGE_DEPTH_STENCIL))
 		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
@@ -88,11 +88,11 @@ nv30_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 	else {
 		switch (pt->format) {
 		/* TODO: Figure out which formats can be swizzled */
-		case PIPE_FORMAT_A8R8G8B8_UNORM:
-		case PIPE_FORMAT_X8R8G8B8_UNORM:
+		case PIPE_FORMAT_B8G8R8A8_UNORM:
+		case PIPE_FORMAT_B8G8R8X8_UNORM:
 		case PIPE_FORMAT_R16_SNORM:
-		case PIPE_FORMAT_R5G6B5_UNORM:
-		case PIPE_FORMAT_A8L8_UNORM:
+		case PIPE_FORMAT_B5G6R5_UNORM:
+		case PIPE_FORMAT_L8A8_UNORM:
 		case PIPE_FORMAT_A8_UNORM:
 		case PIPE_FORMAT_L8_UNORM:
 		case PIPE_FORMAT_I8_UNORM:
@@ -232,8 +232,9 @@ void
 nv30_screen_init_miptree_functions(struct pipe_screen *pscreen)
 {
 	pscreen->texture_create = nv30_miptree_create;
-	pscreen->texture_blanket = nv30_miptree_blanket;
 	pscreen->texture_destroy = nv30_miptree_destroy;
 	pscreen->get_tex_surface = nv30_miptree_surface_new;
 	pscreen->tex_surface_destroy = nv30_miptree_surface_del;
+
+	nouveau_screen(pscreen)->texture_blanket = nv30_miptree_blanket;
 }
diff --git a/src/gallium/drivers/nv30/nv30_screen.c b/src/gallium/drivers/nv30/nv30_screen.c
index aef37d303d8..85433d20953 100644
--- a/src/gallium/drivers/nv30/nv30_screen.c
+++ b/src/gallium/drivers/nv30/nv30_screen.c
@@ -112,8 +112,8 @@ nv30_screen_surface_format_supported(struct pipe_screen *pscreen,
 
 	if (tex_usage & PIPE_TEXTURE_USAGE_RENDER_TARGET) {
 		switch (format) {
-		case PIPE_FORMAT_A8R8G8B8_UNORM:
-		case PIPE_FORMAT_R5G6B5_UNORM:
+		case PIPE_FORMAT_B8G8R8A8_UNORM:
+		case PIPE_FORMAT_B5G6R5_UNORM:
 			return TRUE;
 		default:
 			break;
@@ -121,12 +121,12 @@ nv30_screen_surface_format_supported(struct pipe_screen *pscreen,
 	} else
 	if (tex_usage & PIPE_TEXTURE_USAGE_DEPTH_STENCIL) {
 		switch (format) {
-		case PIPE_FORMAT_Z24S8_UNORM:
-		case PIPE_FORMAT_Z24X8_UNORM:
+		case PIPE_FORMAT_S8Z24_UNORM:
+		case PIPE_FORMAT_X8Z24_UNORM:
 			return TRUE;
 		case PIPE_FORMAT_Z16_UNORM:
 			if (front) {
-				return (front->format == PIPE_FORMAT_R5G6B5_UNORM);
+				return (front->format == PIPE_FORMAT_B5G6R5_UNORM);
 			}
 			return TRUE;
 		default:
@@ -134,16 +134,16 @@ nv30_screen_surface_format_supported(struct pipe_screen *pscreen,
 		}
 	} else {
 		switch (format) {
-		case PIPE_FORMAT_A8R8G8B8_UNORM:
-		case PIPE_FORMAT_A1R5G5B5_UNORM:
-		case PIPE_FORMAT_A4R4G4B4_UNORM:
-		case PIPE_FORMAT_R5G6B5_UNORM:
+		case PIPE_FORMAT_B8G8R8A8_UNORM:
+		case PIPE_FORMAT_B5G5R5A1_UNORM:
+		case PIPE_FORMAT_B4G4R4A4_UNORM:
+		case PIPE_FORMAT_B5G6R5_UNORM:
 		case PIPE_FORMAT_L8_UNORM:
 		case PIPE_FORMAT_A8_UNORM:
 		case PIPE_FORMAT_I8_UNORM:
-		case PIPE_FORMAT_A8L8_UNORM:
+		case PIPE_FORMAT_L8A8_UNORM:
 		case PIPE_FORMAT_Z16_UNORM:
-		case PIPE_FORMAT_Z24S8_UNORM:
+		case PIPE_FORMAT_S8Z24_UNORM:
 			return TRUE;
 		default:
 			break;
diff --git a/src/gallium/drivers/nv30/nv30_state.c b/src/gallium/drivers/nv30/nv30_state.c
index fdfb51a9b34..321575da0a9 100644
--- a/src/gallium/drivers/nv30/nv30_state.c
+++ b/src/gallium/drivers/nv30/nv30_state.c
@@ -697,15 +697,34 @@ nv30_set_vertex_buffers(struct pipe_context *pipe, unsigned count,
 	/*nv30->draw_dirty |= NV30_NEW_ARRAYS;*/
 }
 
+static void *
+nv30_vtxelts_state_create(struct pipe_context *pipe,
+			  unsigned num_elements,
+			  const struct pipe_vertex_element *elements)
+{
+	struct nv30_vtxelt_state *cso = CALLOC_STRUCT(nv30_vtxelt_state);
+
+	assert(num_elements < 16); /* not doing fallbacks yet */
+	cso->num_elements = num_elements;
+	memcpy(cso->pipe, elements, num_elements * sizeof(*elements));
+
+/*	nv30_vtxelt_construct(cso);*/
+
+	return (void *)cso;
+}
+
 static void
-nv30_set_vertex_elements(struct pipe_context *pipe, unsigned count,
-			 const struct pipe_vertex_element *ve)
+nv30_vtxelts_state_delete(struct pipe_context *pipe, void *hwcso)
 {
-	struct nv30_context *nv30 = nv30_context(pipe);
+	FREE(hwcso);
+}
 
-	memcpy(nv30->vtxelt, ve, sizeof(*ve) * count);
-	nv30->vtxelt_nr = count;
+static void
+nv30_vtxelts_state_bind(struct pipe_context *pipe, void *hwcso)
+{
+	struct nv30_context *nv30 = nv30_context(pipe);
 
+	nv30->vtxelt = hwcso;
 	nv30->dirty |= NV30_NEW_ARRAYS;
 	/*nv30->draw_dirty |= NV30_NEW_ARRAYS;*/
 }
@@ -752,7 +771,10 @@ nv30_init_state_functions(struct nv30_context *nv30)
 	nv30->pipe.set_scissor_state = nv30_set_scissor_state;
 	nv30->pipe.set_viewport_state = nv30_set_viewport_state;
 
+	nv30->pipe.create_vertex_elements_state = nv30_vtxelts_state_create;
+	nv30->pipe.delete_vertex_elements_state = nv30_vtxelts_state_delete;
+	nv30->pipe.bind_vertex_elements_state = nv30_vtxelts_state_bind;
+
 	nv30->pipe.set_vertex_buffers = nv30_set_vertex_buffers;
-	nv30->pipe.set_vertex_elements = nv30_set_vertex_elements;
 }
 
diff --git a/src/gallium/drivers/nv30/nv30_state.h b/src/gallium/drivers/nv30/nv30_state.h
index e42e872de75..66c26360cb2 100644
--- a/src/gallium/drivers/nv30/nv30_state.h
+++ b/src/gallium/drivers/nv30/nv30_state.h
@@ -70,6 +70,8 @@ struct nv30_fragment_program {
 	struct nouveau_stateobj *so;
 };
 
+#define NV30_MAX_TEXTURE_LEVELS  16
+
 struct nv30_miptree {
 	struct pipe_texture base;
 	struct nouveau_bo *bo;
@@ -80,7 +82,7 @@ struct nv30_miptree {
 	struct {
 		uint pitch;
 		uint *image_offset;
-	} level[PIPE_MAX_TEXTURE_LEVELS];
+	} level[NV30_MAX_TEXTURE_LEVELS];
 };
 
 #endif
diff --git a/src/gallium/drivers/nv30/nv30_state_fb.c b/src/gallium/drivers/nv30/nv30_state_fb.c
index 2ed2ea55e84..f7fe9833c77 100644
--- a/src/gallium/drivers/nv30/nv30_state_fb.c
+++ b/src/gallium/drivers/nv30/nv30_state_fb.c
@@ -66,14 +66,14 @@ nv30_state_framebuffer_validate(struct nv30_context *nv30)
 	}
 
 	switch (colour_format) {
-	case PIPE_FORMAT_X8R8G8B8_UNORM:
+	case PIPE_FORMAT_B8G8R8X8_UNORM:
 		rt_format |= NV34TCL_RT_FORMAT_COLOR_X8R8G8B8;
 		break;
-	case PIPE_FORMAT_A8R8G8B8_UNORM:
+	case PIPE_FORMAT_B8G8R8A8_UNORM:
 	case 0:
 		rt_format |= NV34TCL_RT_FORMAT_COLOR_A8R8G8B8;
 		break;
-	case PIPE_FORMAT_R5G6B5_UNORM:
+	case PIPE_FORMAT_B5G6R5_UNORM:
 		rt_format |= NV34TCL_RT_FORMAT_COLOR_R5G6B5;
 		colour_bits = 16;
 		break;
@@ -86,8 +86,8 @@ nv30_state_framebuffer_validate(struct nv30_context *nv30)
 		rt_format |= NV34TCL_RT_FORMAT_ZETA_Z16;
 		zeta_bits = 16;
 		break;
-	case PIPE_FORMAT_Z24S8_UNORM:
-	case PIPE_FORMAT_Z24X8_UNORM:
+	case PIPE_FORMAT_S8Z24_UNORM:
+	case PIPE_FORMAT_X8Z24_UNORM:
 	case 0:
 		rt_format |= NV34TCL_RT_FORMAT_ZETA_Z24S8;
 		break;
diff --git a/src/gallium/drivers/nv30/nv30_state_viewport.c b/src/gallium/drivers/nv30/nv30_state_viewport.c
index 2d7781292bd..6fccd6b60e7 100644
--- a/src/gallium/drivers/nv30/nv30_state_viewport.c
+++ b/src/gallium/drivers/nv30/nv30_state_viewport.c
@@ -5,55 +5,25 @@ nv30_state_viewport_validate(struct nv30_context *nv30)
 {
 	struct pipe_viewport_state *vpt = &nv30->viewport;
 	struct nouveau_stateobj *so;
-	unsigned bypass;
-
-	if (/*nv30->render_mode == HW &&*/
-	    !nv30->rasterizer->pipe.bypass_vs_clip_and_viewport)
-		bypass = 0;
-	else
-		bypass = 1;
 
 	if (nv30->state.hw[NV30_STATE_VIEWPORT] &&
-	    (bypass || !(nv30->dirty & NV30_NEW_VIEWPORT)) &&
-	    nv30->state.viewport_bypass == bypass)
+	    !(nv30->dirty & NV30_NEW_VIEWPORT))
 		return FALSE;
-	nv30->state.viewport_bypass = bypass;
 
 	so = so_new(3, 10, 0);
-	if (!bypass) {
-		so_method(so, nv30->screen->rankine,
-			  NV34TCL_VIEWPORT_TRANSLATE_X, 8);
-		so_data  (so, fui(vpt->translate[0]));
-		so_data  (so, fui(vpt->translate[1]));
-		so_data  (so, fui(vpt->translate[2]));
-		so_data  (so, fui(vpt->translate[3]));
-		so_data  (so, fui(vpt->scale[0]));
-		so_data  (so, fui(vpt->scale[1]));
-		so_data  (so, fui(vpt->scale[2]));
-		so_data  (so, fui(vpt->scale[3]));
-/*		so_method(so, nv30->screen->rankine, 0x1d78, 1);
-		so_data  (so, 1);
-*/	} else {
-		so_method(so, nv30->screen->rankine,
-			  NV34TCL_VIEWPORT_TRANSLATE_X, 8);
-		so_data  (so, fui(0.0));
-		so_data  (so, fui(0.0));
-		so_data  (so, fui(0.0));
-		so_data  (so, fui(0.0));
-		so_data  (so, fui(1.0));
-		so_data  (so, fui(1.0));
-		so_data  (so, fui(1.0));
-		so_data  (so, fui(0.0));
-		/* Not entirely certain what this is yet.  The DDX uses this
-		 * value also as it fixes rendering when you pass
-		 * pre-transformed vertices to the GPU.  My best gusss is that
-		 * this bypasses some culling/clipping stage.  Might be worth
-		 * noting that points/lines are uneffected by whatever this
-		 * value fixes, only filled polygons are effected.
-		 */
-/*		so_method(so, nv30->screen->rankine, 0x1d78, 1);
-		so_data  (so, 0x110);
-*/	}
+	so_method(so, nv30->screen->rankine,
+		  NV34TCL_VIEWPORT_TRANSLATE_X, 8);
+	so_data  (so, fui(vpt->translate[0]));
+	so_data  (so, fui(vpt->translate[1]));
+	so_data  (so, fui(vpt->translate[2]));
+	so_data  (so, fui(vpt->translate[3]));
+	so_data  (so, fui(vpt->scale[0]));
+	so_data  (so, fui(vpt->scale[1]));
+	so_data  (so, fui(vpt->scale[2]));
+	so_data  (so, fui(vpt->scale[3]));
+/*	so_method(so, nv30->screen->rankine, 0x1d78, 1);
+	so_data  (so, 1);
+*/
 	/* TODO/FIXME: never saw value 0x0110 in renouveau dumps, only 0x0001 */
 	so_method(so, nv30->screen->rankine, 0x1d78, 1);
 	so_data  (so, 1);
diff --git a/src/gallium/drivers/nv30/nv30_transfer.c b/src/gallium/drivers/nv30/nv30_transfer.c
index 554bcbbdd0e..3aeda51ea19 100644
--- a/src/gallium/drivers/nv30/nv30_transfer.c
+++ b/src/gallium/drivers/nv30/nv30_transfer.c
@@ -1,10 +1,10 @@
-#include <pipe/p_state.h>
-#include <pipe/p_defines.h>
-#include <util/u_inlines.h>
-#include <util/u_format.h>
-#include <util/u_memory.h>
-#include <util/u_math.h>
-#include <nouveau/nouveau_winsys.h>
+#include "pipe/p_state.h"
+#include "pipe/p_defines.h"
+#include "util/u_inlines.h"
+#include "util/u_format.h"
+#include "util/u_memory.h"
+#include "util/u_math.h"
+#include "nouveau/nouveau_winsys.h"
 #include "nv30_context.h"
 #include "nv30_screen.h"
 #include "nv30_state.h"
diff --git a/src/gallium/drivers/nv30/nv30_vbo.c b/src/gallium/drivers/nv30/nv30_vbo.c
index e48823a9138..f3856bb5a5e 100644
--- a/src/gallium/drivers/nv30/nv30_vbo.c
+++ b/src/gallium/drivers/nv30/nv30_vbo.c
@@ -492,16 +492,16 @@ nv30_vbo_validate(struct nv30_context *nv30)
 	int hw;
 
 	vtxbuf = so_new(3, 17, 18);
-	so_method(vtxbuf, rankine, NV34TCL_VTXBUF_ADDRESS(0), nv30->vtxelt_nr);
+	so_method(vtxbuf, rankine, NV34TCL_VTXBUF_ADDRESS(0), nv30->vtxelt->num_elements);
 	vtxfmt = so_new(1, 16, 0);
-	so_method(vtxfmt, rankine, NV34TCL_VTXFMT(0), nv30->vtxelt_nr);
+	so_method(vtxfmt, rankine, NV34TCL_VTXFMT(0), nv30->vtxelt->num_elements);
 
-	for (hw = 0; hw < nv30->vtxelt_nr; hw++) {
+	for (hw = 0; hw < nv30->vtxelt->num_elements; hw++) {
 		struct pipe_vertex_element *ve;
 		struct pipe_vertex_buffer *vb;
 		unsigned type, ncomp;
 
-		ve = &nv30->vtxelt[hw];
+		ve = &nv30->vtxelt->pipe[hw];
 		vb = &nv30->vtxbuf[ve->vertex_buffer_index];
 
 		if (!vb->stride) {
diff --git a/src/gallium/drivers/nv40/nv40_context.h b/src/gallium/drivers/nv40/nv40_context.h
index 3d41042a771..4d2ffd9772d 100644
--- a/src/gallium/drivers/nv40/nv40_context.h
+++ b/src/gallium/drivers/nv40/nv40_context.h
@@ -101,13 +101,18 @@ struct nv40_blend_state {
 struct nv40_state {
 	unsigned scissor_enabled;
 	unsigned stipple_enabled;
-	unsigned viewport_bypass;
 	unsigned fp_samplers;
 
 	uint64_t dirty;
 	struct nouveau_stateobj *hw[NV40_STATE_MAX];
 };
 
+
+struct nv40_vtxelt_state {
+	struct pipe_vertex_element pipe[16];
+	unsigned num_elements;
+};
+
 struct nv40_context {
 	struct pipe_context pipe;
 
@@ -159,8 +164,7 @@ struct nv40_context {
 	unsigned dirty_samplers;
 	struct pipe_vertex_buffer vtxbuf[PIPE_MAX_ATTRIBS];
 	unsigned vtxbuf_nr;
-	struct pipe_vertex_element vtxelt[PIPE_MAX_ATTRIBS];
-	unsigned vtxelt_nr;
+	struct nv40_vtxelt_state *vtxelt;
 };
 
 static INLINE struct nv40_context *
diff --git a/src/gallium/drivers/nv40/nv40_fragtex.c b/src/gallium/drivers/nv40/nv40_fragtex.c
index 7a28d577b14..b60118922a2 100644
--- a/src/gallium/drivers/nv40/nv40_fragtex.c
+++ b/src/gallium/drivers/nv40/nv40_fragtex.c
@@ -25,18 +25,18 @@ struct nv40_texture_format {
 
 static struct nv40_texture_format
 nv40_texture_formats[] = {
-	_(X8R8G8B8_UNORM, A8R8G8B8,   S1,   S1,   S1,  ONE, X, Y, Z, W, 0, 0, 0, 0),
-	_(A8R8G8B8_UNORM, A8R8G8B8,   S1,   S1,   S1,   S1, X, Y, Z, W, 0, 0, 0, 0),
-	_(A1R5G5B5_UNORM, A1R5G5B5,   S1,   S1,   S1,   S1, X, Y, Z, W, 0, 0, 0, 0),
-	_(A4R4G4B4_UNORM, A4R4G4B4,   S1,   S1,   S1,   S1, X, Y, Z, W, 0, 0, 0, 0),
-	_(R5G6B5_UNORM  , R5G6B5  ,   S1,   S1,   S1,  ONE, X, Y, Z, W, 0, 0, 0, 0),
+	_(B8G8R8X8_UNORM, A8R8G8B8,   S1,   S1,   S1,  ONE, X, Y, Z, W, 0, 0, 0, 0),
+	_(B8G8R8A8_UNORM, A8R8G8B8,   S1,   S1,   S1,   S1, X, Y, Z, W, 0, 0, 0, 0),
+	_(B5G5R5A1_UNORM, A1R5G5B5,   S1,   S1,   S1,   S1, X, Y, Z, W, 0, 0, 0, 0),
+	_(B4G4R4A4_UNORM, A4R4G4B4,   S1,   S1,   S1,   S1, X, Y, Z, W, 0, 0, 0, 0),
+	_(B5G6R5_UNORM  , R5G6B5  ,   S1,   S1,   S1,  ONE, X, Y, Z, W, 0, 0, 0, 0),
 	_(L8_UNORM      , L8      ,   S1,   S1,   S1,  ONE, X, X, X, X, 0, 0, 0, 0),
 	_(A8_UNORM      , L8      , ZERO, ZERO, ZERO,   S1, X, X, X, X, 0, 0, 0, 0),
 	_(R16_SNORM     , A16     , ZERO, ZERO,   S1,  ONE, X, X, X, Y, 1, 1, 1, 1),
 	_(I8_UNORM      , L8      ,   S1,   S1,   S1,   S1, X, X, X, X, 0, 0, 0, 0),
-	_(A8L8_UNORM    , A8L8    ,   S1,   S1,   S1,   S1, X, X, X, Y, 0, 0, 0, 0),
+	_(L8A8_UNORM    , A8L8    ,   S1,   S1,   S1,   S1, X, X, X, Y, 0, 0, 0, 0),
 	_(Z16_UNORM     , Z16     ,   S1,   S1,   S1,  ONE, X, X, X, X, 0, 0, 0, 0),
-	_(Z24S8_UNORM   , Z24     ,   S1,   S1,   S1,  ONE, X, X, X, X, 0, 0, 0, 0),
+	_(S8Z24_UNORM   , Z24     ,   S1,   S1,   S1,  ONE, X, X, X, X, 0, 0, 0, 0),
 	_(DXT1_RGB      , DXT1    ,   S1,   S1,   S1,  ONE, X, Y, Z, W, 0, 0, 0, 0),
 	_(DXT1_RGBA     , DXT1    ,   S1,   S1,   S1,   S1, X, Y, Z, W, 0, 0, 0, 0),
 	_(DXT3_RGBA     , DXT3    ,   S1,   S1,   S1,   S1, X, Y, Z, W, 0, 0, 0, 0),
diff --git a/src/gallium/drivers/nv40/nv40_miptree.c b/src/gallium/drivers/nv40/nv40_miptree.c
index ad1a9a51952..62e97bcea48 100644
--- a/src/gallium/drivers/nv40/nv40_miptree.c
+++ b/src/gallium/drivers/nv40/nv40_miptree.c
@@ -20,7 +20,7 @@ nv40_miptree_layout(struct nv40_miptree *mt)
 		                           PIPE_TEXTURE_USAGE_DEPTH_STENCIL |
 		                           PIPE_TEXTURE_USAGE_RENDER_TARGET |
 		                           PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
-		                           PIPE_TEXTURE_USAGE_PRIMARY);
+		                           PIPE_TEXTURE_USAGE_SCANOUT);
 
 	if (pt->target == PIPE_TEXTURE_CUBE) {
 		nr_faces = 6;
@@ -80,7 +80,7 @@ nv40_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 	    pt->height0 & (pt->height0 - 1))
 		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
 	else
-	if (pt->tex_usage & (PIPE_TEXTURE_USAGE_PRIMARY |
+	if (pt->tex_usage & (PIPE_TEXTURE_USAGE_SCANOUT |
 	                     PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
 	                     PIPE_TEXTURE_USAGE_DEPTH_STENCIL))
 		mt->base.tex_usage |= NOUVEAU_TEXTURE_USAGE_LINEAR;
@@ -90,8 +90,8 @@ nv40_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *pt)
 	else {
 		switch (pt->format) {
 		/* TODO: Figure out which formats can be swizzled */
-		case PIPE_FORMAT_A8R8G8B8_UNORM:
-		case PIPE_FORMAT_X8R8G8B8_UNORM:
+		case PIPE_FORMAT_B8G8R8A8_UNORM:
+		case PIPE_FORMAT_B8G8R8X8_UNORM:
 		case PIPE_FORMAT_R16_SNORM:
 		{
 			if (debug_get_bool_option("NOUVEAU_NO_SWIZZLE", FALSE))
@@ -227,9 +227,10 @@ void
 nv40_screen_init_miptree_functions(struct pipe_screen *pscreen)
 {
 	pscreen->texture_create = nv40_miptree_create;
-	pscreen->texture_blanket = nv40_miptree_blanket;
 	pscreen->texture_destroy = nv40_miptree_destroy;
 	pscreen->get_tex_surface = nv40_miptree_surface_new;
 	pscreen->tex_surface_destroy = nv40_miptree_surface_del;
+
+	nouveau_screen(pscreen)->texture_blanket = nv40_miptree_blanket;
 }
 
diff --git a/src/gallium/drivers/nv40/nv40_screen.c b/src/gallium/drivers/nv40/nv40_screen.c
index edee4b9a3aa..b216c5e38c9 100644
--- a/src/gallium/drivers/nv40/nv40_screen.c
+++ b/src/gallium/drivers/nv40/nv40_screen.c
@@ -98,8 +98,8 @@ nv40_screen_surface_format_supported(struct pipe_screen *pscreen,
 {
 	if (tex_usage & PIPE_TEXTURE_USAGE_RENDER_TARGET) {
 		switch (format) {
-		case PIPE_FORMAT_A8R8G8B8_UNORM:
-		case PIPE_FORMAT_R5G6B5_UNORM: 
+		case PIPE_FORMAT_B8G8R8A8_UNORM:
+		case PIPE_FORMAT_B5G6R5_UNORM: 
 			return TRUE;
 		default:
 			break;
@@ -107,8 +107,8 @@ nv40_screen_surface_format_supported(struct pipe_screen *pscreen,
 	} else
 	if (tex_usage & PIPE_TEXTURE_USAGE_DEPTH_STENCIL) {
 		switch (format) {
-		case PIPE_FORMAT_Z24S8_UNORM:
-		case PIPE_FORMAT_Z24X8_UNORM:
+		case PIPE_FORMAT_S8Z24_UNORM:
+		case PIPE_FORMAT_X8Z24_UNORM:
 		case PIPE_FORMAT_Z16_UNORM:
 			return TRUE;
 		default:
@@ -116,17 +116,17 @@ nv40_screen_surface_format_supported(struct pipe_screen *pscreen,
 		}
 	} else {
 		switch (format) {
-		case PIPE_FORMAT_A8R8G8B8_UNORM:
-		case PIPE_FORMAT_A1R5G5B5_UNORM:
-		case PIPE_FORMAT_A4R4G4B4_UNORM:
-		case PIPE_FORMAT_R5G6B5_UNORM:
+		case PIPE_FORMAT_B8G8R8A8_UNORM:
+		case PIPE_FORMAT_B5G5R5A1_UNORM:
+		case PIPE_FORMAT_B4G4R4A4_UNORM:
+		case PIPE_FORMAT_B5G6R5_UNORM:
 		case PIPE_FORMAT_R16_SNORM:
 		case PIPE_FORMAT_L8_UNORM:
 		case PIPE_FORMAT_A8_UNORM:
 		case PIPE_FORMAT_I8_UNORM:
-		case PIPE_FORMAT_A8L8_UNORM:
+		case PIPE_FORMAT_L8A8_UNORM:
 		case PIPE_FORMAT_Z16_UNORM:
-		case PIPE_FORMAT_Z24S8_UNORM:
+		case PIPE_FORMAT_S8Z24_UNORM:
 		case PIPE_FORMAT_DXT1_RGB:
 		case PIPE_FORMAT_DXT1_RGBA:
 		case PIPE_FORMAT_DXT3_RGBA:
diff --git a/src/gallium/drivers/nv40/nv40_state.c b/src/gallium/drivers/nv40/nv40_state.c
index dbb52bf5bc9..120dc42f7b1 100644
--- a/src/gallium/drivers/nv40/nv40_state.c
+++ b/src/gallium/drivers/nv40/nv40_state.c
@@ -712,15 +712,34 @@ nv40_set_vertex_buffers(struct pipe_context *pipe, unsigned count,
 	nv40->draw_dirty |= NV40_NEW_ARRAYS;
 }
 
+static void *
+nv40_vtxelts_state_create(struct pipe_context *pipe,
+			  unsigned num_elements,
+			  const struct pipe_vertex_element *elements)
+{
+	struct nv40_vtxelt_state *cso = CALLOC_STRUCT(nv40_vtxelt_state);
+
+	assert(num_elements < 16); /* not doing fallbacks yet */
+	cso->num_elements = num_elements;
+	memcpy(cso->pipe, elements, num_elements * sizeof(*elements));
+
+/*	nv40_vtxelt_construct(cso);*/
+
+	return (void *)cso;
+}
+
 static void
-nv40_set_vertex_elements(struct pipe_context *pipe, unsigned count,
-			 const struct pipe_vertex_element *ve)
+nv40_vtxelts_state_delete(struct pipe_context *pipe, void *hwcso)
 {
-	struct nv40_context *nv40 = nv40_context(pipe);
+	FREE(hwcso);
+}
 
-	memcpy(nv40->vtxelt, ve, sizeof(*ve) * count);
-	nv40->vtxelt_nr = count;
+static void
+nv40_vtxelts_state_bind(struct pipe_context *pipe, void *hwcso)
+{
+	struct nv40_context *nv40 = nv40_context(pipe);
 
+	nv40->vtxelt = hwcso;
 	nv40->dirty |= NV40_NEW_ARRAYS;
 	nv40->draw_dirty |= NV40_NEW_ARRAYS;
 }
@@ -767,7 +786,10 @@ nv40_init_state_functions(struct nv40_context *nv40)
 	nv40->pipe.set_scissor_state = nv40_set_scissor_state;
 	nv40->pipe.set_viewport_state = nv40_set_viewport_state;
 
+	nv40->pipe.create_vertex_elements_state = nv40_vtxelts_state_create;
+	nv40->pipe.delete_vertex_elements_state = nv40_vtxelts_state_delete;
+	nv40->pipe.bind_vertex_elements_state = nv40_vtxelts_state_bind;
+
 	nv40->pipe.set_vertex_buffers = nv40_set_vertex_buffers;
-	nv40->pipe.set_vertex_elements = nv40_set_vertex_elements;
 }
 
diff --git a/src/gallium/drivers/nv40/nv40_state.h b/src/gallium/drivers/nv40/nv40_state.h
index 192074e7471..e2e69420eae 100644
--- a/src/gallium/drivers/nv40/nv40_state.h
+++ b/src/gallium/drivers/nv40/nv40_state.h
@@ -73,6 +73,8 @@ struct nv40_fragment_program {
 	struct nouveau_stateobj *so;
 };
 
+#define NV40_MAX_TEXTURE_LEVELS  16
+
 struct nv40_miptree {
 	struct pipe_texture base;
 	struct nouveau_bo *bo;
@@ -83,7 +85,7 @@ struct nv40_miptree {
 	struct {
 		uint pitch;
 		uint *image_offset;
-	} level[PIPE_MAX_TEXTURE_LEVELS];
+	} level[NV40_MAX_TEXTURE_LEVELS];
 };
 
 #endif
diff --git a/src/gallium/drivers/nv40/nv40_state_emit.c b/src/gallium/drivers/nv40/nv40_state_emit.c
index 8990f303ce4..297d71f4fac 100644
--- a/src/gallium/drivers/nv40/nv40_state_emit.c
+++ b/src/gallium/drivers/nv40/nv40_state_emit.c
@@ -174,7 +174,7 @@ nv40_state_validate_swtnl(struct nv40_context *nv40)
 
 	if (nv40->draw_dirty & NV40_NEW_ARRAYS) {
 		draw_set_vertex_buffers(draw, nv40->vtxbuf_nr, nv40->vtxbuf);
-		draw_set_vertex_elements(draw, nv40->vtxelt_nr, nv40->vtxelt);	
+		draw_set_vertex_elements(draw, nv40->vtxelt->num_elements, nv40->vtxelt->pipe);	
 	}
 
 	nv40_state_do_validate(nv40, swtnl_states);
diff --git a/src/gallium/drivers/nv40/nv40_state_fb.c b/src/gallium/drivers/nv40/nv40_state_fb.c
index a58fe9ddb19..fd3fdfddc09 100644
--- a/src/gallium/drivers/nv40/nv40_state_fb.c
+++ b/src/gallium/drivers/nv40/nv40_state_fb.c
@@ -57,14 +57,14 @@ nv40_state_framebuffer_validate(struct nv40_context *nv40)
 		rt_format = NV40TCL_RT_FORMAT_TYPE_LINEAR;
 
 	switch (colour_format) {
-	case PIPE_FORMAT_X8R8G8B8_UNORM:
+	case PIPE_FORMAT_B8G8R8X8_UNORM:
 		rt_format |= NV40TCL_RT_FORMAT_COLOR_X8R8G8B8;
 		break;
-	case PIPE_FORMAT_A8R8G8B8_UNORM:
+	case PIPE_FORMAT_B8G8R8A8_UNORM:
 	case 0:
 		rt_format |= NV40TCL_RT_FORMAT_COLOR_A8R8G8B8;
 		break;
-	case PIPE_FORMAT_R5G6B5_UNORM:
+	case PIPE_FORMAT_B5G6R5_UNORM:
 		rt_format |= NV40TCL_RT_FORMAT_COLOR_R5G6B5;
 		break;
 	default:
@@ -75,8 +75,8 @@ nv40_state_framebuffer_validate(struct nv40_context *nv40)
 	case PIPE_FORMAT_Z16_UNORM:
 		rt_format |= NV40TCL_RT_FORMAT_ZETA_Z16;
 		break;
-	case PIPE_FORMAT_Z24S8_UNORM:
-	case PIPE_FORMAT_Z24X8_UNORM:
+	case PIPE_FORMAT_S8Z24_UNORM:
+	case PIPE_FORMAT_X8Z24_UNORM:
 	case 0:
 		rt_format |= NV40TCL_RT_FORMAT_ZETA_Z24S8;
 		break;
diff --git a/src/gallium/drivers/nv40/nv40_state_viewport.c b/src/gallium/drivers/nv40/nv40_state_viewport.c
index 9919ba1d0b0..3aacb00f996 100644
--- a/src/gallium/drivers/nv40/nv40_state_viewport.c
+++ b/src/gallium/drivers/nv40/nv40_state_viewport.c
@@ -5,55 +5,24 @@ nv40_state_viewport_validate(struct nv40_context *nv40)
 {
 	struct pipe_viewport_state *vpt = &nv40->viewport;
 	struct nouveau_stateobj *so;
-	unsigned bypass;
-
-	if (nv40->render_mode == HW &&
-	    !nv40->rasterizer->pipe.bypass_vs_clip_and_viewport)
-		bypass = 0;
-	else
-		bypass = 1;
 
 	if (nv40->state.hw[NV40_STATE_VIEWPORT] &&
-	    (bypass || !(nv40->dirty & NV40_NEW_VIEWPORT)) &&
-	    nv40->state.viewport_bypass == bypass)
+	    !(nv40->dirty & NV40_NEW_VIEWPORT))
 		return FALSE;
-	nv40->state.viewport_bypass = bypass;
 
 	so = so_new(2, 9, 0);
-	if (!bypass) {
-		so_method(so, nv40->screen->curie,
-			  NV40TCL_VIEWPORT_TRANSLATE_X, 8);
-		so_data  (so, fui(vpt->translate[0]));
-		so_data  (so, fui(vpt->translate[1]));
-		so_data  (so, fui(vpt->translate[2]));
-		so_data  (so, fui(vpt->translate[3]));
-		so_data  (so, fui(vpt->scale[0]));
-		so_data  (so, fui(vpt->scale[1]));
-		so_data  (so, fui(vpt->scale[2]));
-		so_data  (so, fui(vpt->scale[3]));
-		so_method(so, nv40->screen->curie, 0x1d78, 1);
-		so_data  (so, 1);
-	} else {
-		so_method(so, nv40->screen->curie,
-			  NV40TCL_VIEWPORT_TRANSLATE_X, 8);
-		so_data  (so, fui(0.0));
-		so_data  (so, fui(0.0));
-		so_data  (so, fui(0.0));
-		so_data  (so, fui(0.0));
-		so_data  (so, fui(1.0));
-		so_data  (so, fui(1.0));
-		so_data  (so, fui(1.0));
-		so_data  (so, fui(0.0));
-		/* Not entirely certain what this is yet.  The DDX uses this
-		 * value also as it fixes rendering when you pass
-		 * pre-transformed vertices to the GPU.  My best gusss is that
-		 * this bypasses some culling/clipping stage.  Might be worth
-		 * noting that points/lines are uneffected by whatever this
-		 * value fixes, only filled polygons are effected.
-		 */
-		so_method(so, nv40->screen->curie, 0x1d78, 1);
-		so_data  (so, 0x110);
-	}
+	so_method(so, nv40->screen->curie,
+		  NV40TCL_VIEWPORT_TRANSLATE_X, 8);
+	so_data  (so, fui(vpt->translate[0]));
+	so_data  (so, fui(vpt->translate[1]));
+	so_data  (so, fui(vpt->translate[2]));
+	so_data  (so, fui(vpt->translate[3]));
+	so_data  (so, fui(vpt->scale[0]));
+	so_data  (so, fui(vpt->scale[1]));
+	so_data  (so, fui(vpt->scale[2]));
+	so_data  (so, fui(vpt->scale[3]));
+	so_method(so, nv40->screen->curie, 0x1d78, 1);
+	so_data  (so, 1);
 
 	so_ref(so, &nv40->state.hw[NV40_STATE_VIEWPORT]);
 	so_ref(NULL, &so);
diff --git a/src/gallium/drivers/nv40/nv40_transfer.c b/src/gallium/drivers/nv40/nv40_transfer.c
index ee266c6cfb1..0462a042c38 100644
--- a/src/gallium/drivers/nv40/nv40_transfer.c
+++ b/src/gallium/drivers/nv40/nv40_transfer.c
@@ -1,10 +1,10 @@
-#include <pipe/p_state.h>
-#include <pipe/p_defines.h>
-#include <util/u_inlines.h>
-#include <util/u_format.h>
-#include <util/u_memory.h>
-#include <util/u_math.h>
-#include <nouveau/nouveau_winsys.h>
+#include "pipe/p_state.h"
+#include "pipe/p_defines.h"
+#include "util/u_inlines.h"
+#include "util/u_format.h"
+#include "util/u_memory.h"
+#include "util/u_math.h"
+#include "nouveau/nouveau_winsys.h"
 #include "nv40_context.h"
 #include "nv40_screen.h"
 #include "nv40_state.h"
diff --git a/src/gallium/drivers/nv40/nv40_vbo.c b/src/gallium/drivers/nv40/nv40_vbo.c
index 7812460d2ed..fabdf4bf23b 100644
--- a/src/gallium/drivers/nv40/nv40_vbo.c
+++ b/src/gallium/drivers/nv40/nv40_vbo.c
@@ -493,16 +493,16 @@ nv40_vbo_validate(struct nv40_context *nv40)
 	int hw;
 
 	vtxbuf = so_new(3, 17, 18);
-	so_method(vtxbuf, curie, NV40TCL_VTXBUF_ADDRESS(0), nv40->vtxelt_nr);
+	so_method(vtxbuf, curie, NV40TCL_VTXBUF_ADDRESS(0), nv40->vtxelt->num_elements);
 	vtxfmt = so_new(1, 16, 0);
-	so_method(vtxfmt, curie, NV40TCL_VTXFMT(0), nv40->vtxelt_nr);
+	so_method(vtxfmt, curie, NV40TCL_VTXFMT(0), nv40->vtxelt->num_elements);
 
-	for (hw = 0; hw < nv40->vtxelt_nr; hw++) {
+	for (hw = 0; hw < nv40->vtxelt->num_elements; hw++) {
 		struct pipe_vertex_element *ve;
 		struct pipe_vertex_buffer *vb;
 		unsigned type, ncomp;
 
-		ve = &nv40->vtxelt[hw];
+		ve = &nv40->vtxelt->pipe[hw];
 		vb = &nv40->vtxbuf[ve->vertex_buffer_index];
 
 		if (!vb->stride) {
diff --git a/src/gallium/drivers/nv50/Makefile b/src/gallium/drivers/nv50/Makefile
index 612aea28a34..5d622e1c13c 100644
--- a/src/gallium/drivers/nv50/Makefile
+++ b/src/gallium/drivers/nv50/Makefile
@@ -16,6 +16,7 @@ C_SOURCES = \
 	nv50_surface.c \
 	nv50_tex.c \
 	nv50_transfer.c \
-	nv50_vbo.c
+	nv50_vbo.c \
+	nv50_push.c
 
 include ../../Makefile.template
diff --git a/src/gallium/drivers/nv50/nv50_clear.c b/src/gallium/drivers/nv50/nv50_clear.c
index e0b2d2880b0..8afc95c9fc6 100644
--- a/src/gallium/drivers/nv50/nv50_clear.c
+++ b/src/gallium/drivers/nv50/nv50_clear.c
@@ -36,7 +36,7 @@ nv50_clear(struct pipe_context *pipe, unsigned buffers,
 	struct pipe_framebuffer_state *fb = &nv50->framebuffer;
 	unsigned mode = 0, i;
 
-	if (!nv50_state_validate(nv50))
+	if (!nv50_state_validate(nv50, 64))
 		return;
 
 	if (buffers & PIPE_CLEAR_COLOR && fb->nr_cbufs) {
diff --git a/src/gallium/drivers/nv50/nv50_context.c b/src/gallium/drivers/nv50/nv50_context.c
index 7be12fcdef4..0eb42f323ff 100644
--- a/src/gallium/drivers/nv50/nv50_context.c
+++ b/src/gallium/drivers/nv50/nv50_context.c
@@ -46,43 +46,13 @@ static void
 nv50_destroy(struct pipe_context *pipe)
 {
 	struct nv50_context *nv50 = nv50_context(pipe);
+	int i;
 
-        if (nv50->state.fb)
-		so_ref(NULL, &nv50->state.fb);
-	if (nv50->state.blend)
-		so_ref(NULL, &nv50->state.blend);
-	if (nv50->state.blend_colour)
-		so_ref(NULL, &nv50->state.blend_colour);
-	if (nv50->state.zsa)
-		so_ref(NULL, &nv50->state.zsa);
-	if (nv50->state.rast)
-		so_ref(NULL, &nv50->state.rast);
-	if (nv50->state.stipple)
-		so_ref(NULL, &nv50->state.stipple);
-	if (nv50->state.scissor)
-		so_ref(NULL, &nv50->state.scissor);
-	if (nv50->state.viewport)
-		so_ref(NULL, &nv50->state.viewport);
-	if (nv50->state.tsc_upload)
-		so_ref(NULL, &nv50->state.tsc_upload);
-	if (nv50->state.tic_upload)
-		so_ref(NULL, &nv50->state.tic_upload);
-	if (nv50->state.vertprog)
-		so_ref(NULL, &nv50->state.vertprog);
-	if (nv50->state.fragprog)
-		so_ref(NULL, &nv50->state.fragprog);
-	if (nv50->state.geomprog)
-		so_ref(NULL, &nv50->state.geomprog);
-	if (nv50->state.fp_linkage)
-		so_ref(NULL, &nv50->state.fp_linkage);
-	if (nv50->state.gp_linkage)
-		so_ref(NULL, &nv50->state.gp_linkage);
-	if (nv50->state.vtxfmt)
-		so_ref(NULL, &nv50->state.vtxfmt);
-	if (nv50->state.vtxbuf)
-		so_ref(NULL, &nv50->state.vtxbuf);
-	if (nv50->state.vtxattr)
-		so_ref(NULL, &nv50->state.vtxattr);
+	for (i = 0; i < 64; i++) {
+		if (!nv50->state.hw[i])
+			continue;
+		so_ref(NULL, &nv50->state.hw[i]);
+	}
 
 	draw_destroy(nv50->draw);
 
@@ -123,7 +93,6 @@ nv50_create(struct pipe_screen *pscreen, void *priv)
 	nv50->pipe.is_buffer_referenced = nouveau_is_buffer_referenced;
 
 	screen->base.channel->user_private = nv50;
-	screen->base.channel->flush_notify = nv50_state_flush_notify;
 
 	nv50_init_surface_functions(nv50);
 	nv50_init_state_functions(nv50);
diff --git a/src/gallium/drivers/nv50/nv50_context.h b/src/gallium/drivers/nv50/nv50_context.h
index 42405c1a890..8a5f7cb2519 100644
--- a/src/gallium/drivers/nv50/nv50_context.h
+++ b/src/gallium/drivers/nv50/nv50_context.h
@@ -77,6 +77,12 @@ struct nv50_sampler_view {
 	uint32_t tic[8];
 };
 
+struct nv50_vtxelt_stateobj {
+	struct pipe_vertex_element pipe[16];
+	unsigned num_elements;
+	uint32_t hw[16];
+};
+
 static INLINE struct nv50_sampler_view *
 nv50_sampler_view(struct pipe_sampler_view *view)
 {
@@ -101,10 +107,12 @@ struct nv50_miptree_level {
 	unsigned tile_mode;
 };
 
+#define NV50_MAX_TEXTURE_LEVELS 16
+
 struct nv50_miptree {
 	struct nouveau_miptree base;
 
-	struct nv50_miptree_level level[PIPE_MAX_TEXTURE_LEVELS];
+	struct nv50_miptree_level level[NV50_MAX_TEXTURE_LEVELS];
 	int image_nr;
 	int total_size;
 };
@@ -126,31 +134,12 @@ nv50_surface(struct pipe_surface *pt)
 }
 
 struct nv50_state {
-	unsigned dirty;
+	struct nouveau_stateobj *hw[64];
+	uint64_t hw_dirty;
 
-	struct nouveau_stateobj *fb;
-	struct nouveau_stateobj *blend;
-	struct nouveau_stateobj *blend_colour;
-	struct nouveau_stateobj *zsa;
-	struct nouveau_stateobj *stencil_ref;
-	struct nouveau_stateobj *rast;
-	struct nouveau_stateobj *stipple;
-	struct nouveau_stateobj *scissor;
-	unsigned scissor_enabled;
-	struct nouveau_stateobj *viewport;
-	unsigned viewport_bypass;
-	struct nouveau_stateobj *tsc_upload;
-	struct nouveau_stateobj *tic_upload;
-	unsigned sampler_view_nr[3];
-	struct nouveau_stateobj *vertprog;
-	struct nouveau_stateobj *fragprog;
-	struct nouveau_stateobj *geomprog;
-	struct nouveau_stateobj *fp_linkage;
-	struct nouveau_stateobj *gp_linkage;
-	struct nouveau_stateobj *vtxfmt;
+	unsigned miptree_nr[PIPE_SHADER_TYPES];
 	struct nouveau_stateobj *vtxbuf;
 	struct nouveau_stateobj *vtxattr;
-	struct nouveau_stateobj *instbuf;
 	unsigned vtxelt_nr;
 };
 
@@ -179,14 +168,15 @@ struct nv50_context {
 	struct pipe_buffer *constbuf[PIPE_SHADER_TYPES];
 	struct pipe_vertex_buffer vtxbuf[PIPE_MAX_ATTRIBS];
 	unsigned vtxbuf_nr;
-	struct pipe_vertex_element vtxelt[PIPE_MAX_ATTRIBS];
-	unsigned vtxelt_nr;
-	struct nv50_sampler_stateobj *sampler[3][PIPE_MAX_SAMPLERS];
-	unsigned sampler_nr[3];
+	struct nv50_vtxelt_stateobj *vtxelt;
+	struct nv50_sampler_stateobj *sampler[PIPE_SHADER_TYPES][PIPE_MAX_SAMPLERS];
+	unsigned sampler_nr[PIPE_SHADER_TYPES];
 	struct pipe_sampler_view *sampler_views[3][PIPE_MAX_SAMPLERS];
 	unsigned sampler_view_nr[3];
+	struct nv50_miptree *miptree[PIPE_SHADER_TYPES][PIPE_MAX_SAMPLERS];
+	unsigned miptree_nr[PIPE_SHADER_TYPES];
 
-	uint16_t vbo_fifo;
+	unsigned vbo_fifo;
 };
 
 static INLINE struct nv50_context *
@@ -228,24 +218,36 @@ extern void nv50_draw_elements_instanced(struct pipe_context *pipe,
 					 unsigned count,
 					 unsigned startInstance,
 					 unsigned instanceCount);
-extern void nv50_vbo_validate(struct nv50_context *nv50);
+extern void nv50_vtxelt_construct(struct nv50_vtxelt_stateobj *cso);
+extern struct nouveau_stateobj *nv50_vbo_validate(struct nv50_context *nv50);
+
+/* nv50_push.c */
+extern void
+nv50_push_elements_instanced(struct pipe_context *, struct pipe_buffer *,
+			     unsigned idxsize, unsigned mode, unsigned start,
+			     unsigned count, unsigned i_start,
+			     unsigned i_count);
 
 /* nv50_clear.c */
 extern void nv50_clear(struct pipe_context *pipe, unsigned buffers,
 		       const float *rgba, double depth, unsigned stencil);
 
 /* nv50_program.c */
-extern void nv50_vertprog_validate(struct nv50_context *nv50);
-extern void nv50_fragprog_validate(struct nv50_context *nv50);
-extern void nv50_geomprog_validate(struct nv50_context *nv50);
-extern void nv50_fp_linkage_validate(struct nv50_context *nv50);
-extern void nv50_gp_linkage_validate(struct nv50_context *nv50);
+extern struct nouveau_stateobj *
+nv50_vertprog_validate(struct nv50_context *nv50);
+extern struct nouveau_stateobj *
+nv50_fragprog_validate(struct nv50_context *nv50);
+extern struct nouveau_stateobj *
+nv50_geomprog_validate(struct nv50_context *nv50);
+extern struct nouveau_stateobj *
+nv50_fp_linkage_validate(struct nv50_context *nv50);
+extern struct nouveau_stateobj *
+nv50_gp_linkage_validate(struct nv50_context *nv50);
 extern void nv50_program_destroy(struct nv50_context *nv50,
 				 struct nv50_program *p);
 
 /* nv50_state_validate.c */
-extern boolean nv50_state_validate(struct nv50_context *nv50);
-extern void nv50_state_flush_notify(struct nouveau_channel *chan);
+extern boolean nv50_state_validate(struct nv50_context *nv50, unsigned dwords);
 
 extern void nv50_so_init_sifc(struct nv50_context *nv50,
 			      struct nouveau_stateobj *so,
@@ -253,8 +255,9 @@ extern void nv50_so_init_sifc(struct nv50_context *nv50,
 			      unsigned offset, unsigned size);
 
 /* nv50_tex.c */
-extern void nv50_tex_validate(struct nv50_context *);
 extern boolean nv50_tex_construct(struct nv50_sampler_view *view);
+extern void nv50_tex_relocs(struct nv50_context *);
+extern struct nouveau_stateobj *nv50_tex_validate(struct nv50_context *);
 
 /* nv50_transfer.c */
 extern void
@@ -268,4 +271,35 @@ nv50_upload_sifc(struct nv50_context *nv50,
 struct pipe_context *
 nv50_create(struct pipe_screen *pscreen, void *priv);
 
+static INLINE unsigned
+nv50_prim(unsigned mode)
+{
+	switch (mode) {
+	case PIPE_PRIM_POINTS: return NV50TCL_VERTEX_BEGIN_POINTS;
+	case PIPE_PRIM_LINES: return NV50TCL_VERTEX_BEGIN_LINES;
+	case PIPE_PRIM_LINE_LOOP: return NV50TCL_VERTEX_BEGIN_LINE_LOOP;
+	case PIPE_PRIM_LINE_STRIP: return NV50TCL_VERTEX_BEGIN_LINE_STRIP;
+	case PIPE_PRIM_TRIANGLES: return NV50TCL_VERTEX_BEGIN_TRIANGLES;
+	case PIPE_PRIM_TRIANGLE_STRIP:
+		return NV50TCL_VERTEX_BEGIN_TRIANGLE_STRIP;
+	case PIPE_PRIM_TRIANGLE_FAN: return NV50TCL_VERTEX_BEGIN_TRIANGLE_FAN;
+	case PIPE_PRIM_QUADS: return NV50TCL_VERTEX_BEGIN_QUADS;
+	case PIPE_PRIM_QUAD_STRIP: return NV50TCL_VERTEX_BEGIN_QUAD_STRIP;
+	case PIPE_PRIM_POLYGON: return NV50TCL_VERTEX_BEGIN_POLYGON;
+	case PIPE_PRIM_LINES_ADJACENCY:
+		return NV50TCL_VERTEX_BEGIN_LINES_ADJACENCY;
+	case PIPE_PRIM_LINE_STRIP_ADJACENCY:
+		return NV50TCL_VERTEX_BEGIN_LINE_STRIP_ADJACENCY;
+	case PIPE_PRIM_TRIANGLES_ADJACENCY:
+		return NV50TCL_VERTEX_BEGIN_TRIANGLES_ADJACENCY;
+	case PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY:
+		return NV50TCL_VERTEX_BEGIN_TRIANGLE_STRIP_ADJACENCY;
+	default:
+		break;
+	}
+
+	NOUVEAU_ERR("invalid primitive type %d\n", mode);
+	return NV50TCL_VERTEX_BEGIN_POINTS;
+}
+
 #endif
diff --git a/src/gallium/drivers/nv50/nv50_miptree.c b/src/gallium/drivers/nv50/nv50_miptree.c
index 7297c74a83c..e091cae6024 100644
--- a/src/gallium/drivers/nv50/nv50_miptree.c
+++ b/src/gallium/drivers/nv50/nv50_miptree.c
@@ -89,14 +89,14 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp)
 	case PIPE_FORMAT_Z32_FLOAT:
 		tile_flags = 0x4800;
 		break;
-	case PIPE_FORMAT_Z24S8_UNORM:
+	case PIPE_FORMAT_S8Z24_UNORM:
 		tile_flags = 0x1800;
 		break;
 	case PIPE_FORMAT_Z16_UNORM:
 		tile_flags = 0x6c00;
 		break;
-	case PIPE_FORMAT_X8Z24_UNORM:
-	case PIPE_FORMAT_S8Z24_UNORM:
+	case PIPE_FORMAT_Z24X8_UNORM:
+	case PIPE_FORMAT_Z24S8_UNORM:
 		tile_flags = 0x2800;
 		break;
 	case PIPE_FORMAT_R32G32B32A32_FLOAT:
@@ -104,7 +104,7 @@ nv50_miptree_create(struct pipe_screen *pscreen, const struct pipe_texture *tmp)
 		tile_flags = 0x7400;
 		break;
 	default:
-		if ((pt->tex_usage & PIPE_TEXTURE_USAGE_PRIMARY) &&
+		if ((pt->tex_usage & PIPE_TEXTURE_USAGE_SCANOUT) &&
 		    util_format_get_blocksizebits(pt->format) == 32)
 			tile_flags = 0x7a00;
 		else
@@ -255,9 +255,10 @@ void
 nv50_screen_init_miptree_functions(struct pipe_screen *pscreen)
 {
 	pscreen->texture_create = nv50_miptree_create;
-	pscreen->texture_blanket = nv50_miptree_blanket;
 	pscreen->texture_destroy = nv50_miptree_destroy;
 	pscreen->get_tex_surface = nv50_miptree_surface_new;
 	pscreen->tex_surface_destroy = nv50_miptree_surface_del;
+
+	nouveau_screen(pscreen)->texture_blanket = nv50_miptree_blanket;
 }
 
diff --git a/src/gallium/drivers/nv50/nv50_program.c b/src/gallium/drivers/nv50/nv50_program.c
index 2372cbbef69..c857816b31a 100644
--- a/src/gallium/drivers/nv50/nv50_program.c
+++ b/src/gallium/drivers/nv50/nv50_program.c
@@ -4270,7 +4270,7 @@ nv50_program_validate_code(struct nv50_context *nv50, struct nv50_program *p)
 	FREE(up);
 }
 
-void
+struct nouveau_stateobj *
 nv50_vertprog_validate(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
@@ -4286,6 +4286,9 @@ nv50_vertprog_validate(struct nv50_context *nv50)
 	nv50_program_validate_data(nv50, p);
 	nv50_program_validate_code(nv50, p);
 
+	if (!(nv50->dirty & NV50_NEW_VERTPROG))
+		return NULL;
+
 	so = so_new(5, 7, 2);
 	so_method(so, tesla, NV50TCL_VP_ADDRESS_HIGH, 2);
 	so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD |
@@ -4301,11 +4304,10 @@ nv50_vertprog_validate(struct nv50_context *nv50)
 	so_data  (so, p->cfg.high_temp);
 	so_method(so, tesla, NV50TCL_VP_START_ID, 1);
 	so_data  (so, 0); /* program start offset */
-	so_ref(so, &nv50->state.vertprog);
-	so_ref(NULL, &so);
+	return so;
 }
 
-void
+struct nouveau_stateobj *
 nv50_fragprog_validate(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
@@ -4321,6 +4323,9 @@ nv50_fragprog_validate(struct nv50_context *nv50)
 	nv50_program_validate_data(nv50, p);
 	nv50_program_validate_code(nv50, p);
 
+	if (!(nv50->dirty & NV50_NEW_FRAGPROG))
+		return NULL;
+
 	so = so_new(6, 7, 2);
 	so_method(so, tesla, NV50TCL_FP_ADDRESS_HIGH, 2);
 	so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD |
@@ -4337,11 +4342,10 @@ nv50_fragprog_validate(struct nv50_context *nv50)
 	so_data  (so, p->cfg.regs[3]);
 	so_method(so, tesla, NV50TCL_FP_START_ID, 1);
 	so_data  (so, 0); /* program start offset */
-	so_ref(so, &nv50->state.fragprog);
-	so_ref(NULL, &so);
+	return so;
 }
 
-void
+struct nouveau_stateobj *
 nv50_geomprog_validate(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
@@ -4357,6 +4361,9 @@ nv50_geomprog_validate(struct nv50_context *nv50)
 	nv50_program_validate_data(nv50, p);
 	nv50_program_validate_code(nv50, p);
 
+	if (!(nv50->dirty & NV50_NEW_GEOMPROG))
+		return NULL;
+
 	so = so_new(6, 7, 2);
 	so_method(so, tesla, NV50TCL_GP_ADDRESS_HIGH, 2);
 	so_reloc (so, p->bo, 0, NOUVEAU_BO_VRAM | NOUVEAU_BO_RD |
@@ -4373,8 +4380,7 @@ nv50_geomprog_validate(struct nv50_context *nv50)
 	so_data  (so, p->cfg.vert_count);
 	so_method(so, tesla, NV50TCL_GP_START_ID, 1);
 	so_data  (so, 0);
-	so_ref(so, &nv50->state.geomprog);
-	so_ref(NULL, &so);
+	return so;
 }
 
 static uint32_t
@@ -4454,7 +4460,7 @@ nv50_vec4_map(uint32_t *map32, int mid, uint8_t zval, uint32_t lin[4],
 	return mid;
 }
 
-void
+struct nouveau_stateobj *
 nv50_fp_linkage_validate(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
@@ -4580,8 +4586,7 @@ nv50_fp_linkage_validate(struct nv50_context *nv50)
 	so_method(so, tesla, NV50TCL_GP_ENABLE, 1);
 	so_data  (so, (vp->type == PIPE_SHADER_GEOMETRY) ? 1 : 0);
 
-	so_ref(so, &nv50->state.fp_linkage);
-	so_ref(NULL, &so);
+	return so;
 }
 
 static int
@@ -4615,7 +4620,7 @@ construct_vp_gp_mapping(uint32_t *map32, int m,
 	return m;
 }
 
-void
+struct nouveau_stateobj *
 nv50_gp_linkage_validate(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
@@ -4625,10 +4630,8 @@ nv50_gp_linkage_validate(struct nv50_context *nv50)
 	uint32_t map[16];
 	int m = 0;
 
-	if (!gp) {
-		so_ref(NULL, &nv50->state.gp_linkage);
-		return;
-	}
+	if (!gp)
+		return NULL;
 	memset(map, 0, sizeof(map));
 
 	m = construct_vp_gp_mapping(map, m, vp, gp);
@@ -4646,8 +4649,7 @@ nv50_gp_linkage_validate(struct nv50_context *nv50)
 	so_method(so, tesla, NV50TCL_VP_RESULT_MAP(0), m);
 	so_datap (so, map, m);
 
-	so_ref(so, &nv50->state.gp_linkage);
-	so_ref(NULL, &so);
+	return so;
 }
 
 void
diff --git a/src/gallium/drivers/nv50/nv50_push.c b/src/gallium/drivers/nv50/nv50_push.c
new file mode 100644
index 00000000000..96a1f32d304
--- /dev/null
+++ b/src/gallium/drivers/nv50/nv50_push.c
@@ -0,0 +1,326 @@
+#include "pipe/p_context.h"
+#include "pipe/p_state.h"
+#include "util/u_inlines.h"
+#include "util/u_format.h"
+
+#include "nouveau/nouveau_util.h"
+#include "nv50_context.h"
+
+struct push_context {
+   struct nv50_context *nv50;
+
+   unsigned vtx_size;
+
+   void *idxbuf;
+   unsigned idxsize;
+
+   float edgeflag;
+   int edgeflag_attr;
+
+   struct {
+      void *map;
+      unsigned stride;
+      unsigned divisor;
+      unsigned step;
+      void (*push)(struct nouveau_channel *, void *);
+   } attr[16];
+   unsigned attr_nr;
+};
+
+static void
+emit_b32_1(struct nouveau_channel *chan, void *data)
+{
+   uint32_t *v = data;
+
+   OUT_RING(chan, v[0]);
+}
+
+static void
+emit_b32_2(struct nouveau_channel *chan, void *data)
+{
+   uint32_t *v = data;
+
+   OUT_RING(chan, v[0]);
+   OUT_RING(chan, v[1]);
+}
+
+static void
+emit_b32_3(struct nouveau_channel *chan, void *data)
+{
+   uint32_t *v = data;
+
+   OUT_RING(chan, v[0]);
+   OUT_RING(chan, v[1]);
+   OUT_RING(chan, v[2]);
+}
+
+static void
+emit_b32_4(struct nouveau_channel *chan, void *data)
+{
+   uint32_t *v = data;
+
+   OUT_RING(chan, v[0]);
+   OUT_RING(chan, v[1]);
+   OUT_RING(chan, v[2]);
+   OUT_RING(chan, v[3]);
+}
+
+static void
+emit_b16_1(struct nouveau_channel *chan, void *data)
+{
+   uint16_t *v = data;
+
+   OUT_RING(chan, v[0]);
+}
+
+static void
+emit_b16_3(struct nouveau_channel *chan, void *data)
+{
+   uint16_t *v = data;
+
+   OUT_RING(chan, (v[1] << 16) | v[0]);
+   OUT_RING(chan, v[2]);
+}
+
+static void
+emit_b08_1(struct nouveau_channel *chan, void *data)
+{
+   uint8_t *v = data;
+
+   OUT_RING(chan, v[0]);
+}
+
+static void
+emit_b08_3(struct nouveau_channel *chan, void *data)
+{
+   uint8_t *v = data;
+
+   OUT_RING(chan, (v[2] << 16) | (v[1] << 8) | v[0]);
+}
+
+static INLINE void
+emit_vertex(struct push_context *ctx, unsigned n)
+{
+   struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+   struct nouveau_channel *chan = tesla->channel;
+   int i;
+
+   if (ctx->edgeflag_attr < 16) {
+      float *edgeflag = ctx->attr[ctx->edgeflag_attr].map +
+                        ctx->attr[ctx->edgeflag_attr].stride * n;
+
+      if (*edgeflag != ctx->edgeflag) {
+         BEGIN_RING(chan, tesla, NV50TCL_EDGEFLAG_ENABLE, 1);
+         OUT_RING  (chan, *edgeflag ? 1 : 0);
+         ctx->edgeflag = *edgeflag;
+      }
+   }
+
+   BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, ctx->vtx_size);
+   for (i = 0; i < ctx->attr_nr; i++)
+      ctx->attr[i].push(chan, ctx->attr[i].map + ctx->attr[i].stride * n);
+}
+
+static void
+emit_edgeflag(void *priv, boolean enabled)
+{
+   struct push_context *ctx = priv;
+   struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+   struct nouveau_channel *chan = tesla->channel;
+
+   BEGIN_RING(chan, tesla, NV50TCL_EDGEFLAG_ENABLE, 1);
+   OUT_RING  (chan, enabled ? 1 : 0);
+}
+
+static void
+emit_elt08(void *priv, unsigned start, unsigned count)
+{
+   struct push_context *ctx = priv;
+   uint8_t *idxbuf = ctx->idxbuf;
+
+   while (count--)
+      emit_vertex(ctx, idxbuf[start++]);
+}
+
+static void
+emit_elt16(void *priv, unsigned start, unsigned count)
+{
+   struct push_context *ctx = priv;
+   uint16_t *idxbuf = ctx->idxbuf;
+
+   while (count--)
+      emit_vertex(ctx, idxbuf[start++]);
+}
+
+static void
+emit_elt32(void *priv, unsigned start, unsigned count)
+{
+   struct push_context *ctx = priv;
+   uint32_t *idxbuf = ctx->idxbuf;
+
+   while (count--)
+      emit_vertex(ctx, idxbuf[start++]);
+}
+
+static void
+emit_verts(void *priv, unsigned start, unsigned count)
+{
+   while (count--)
+      emit_vertex(priv, start++);
+}
+
+void
+nv50_push_elements_instanced(struct pipe_context *pipe,
+                             struct pipe_buffer *idxbuf, unsigned idxsize,
+                             unsigned mode, unsigned start, unsigned count,
+                             unsigned i_start, unsigned i_count)
+{
+   struct nv50_context *nv50 = nv50_context(pipe);
+   struct nouveau_grobj *tesla = nv50->screen->tesla;
+   struct nouveau_channel *chan = tesla->channel;
+   struct push_context ctx;
+   const unsigned p_overhead = 4 + /* begin/end */
+                               4; /* potential edgeflag enable/disable */
+   const unsigned v_overhead = 1 + /* VERTEX_DATA packet header */
+                               2; /* potential edgeflag modification */
+   struct u_split_prim s;
+   unsigned vtx_size;
+   boolean nzi = FALSE;
+   int i;
+
+   ctx.nv50 = nv50;
+   ctx.attr_nr = 0;
+   ctx.idxbuf = NULL;
+   ctx.vtx_size = 0;
+   ctx.edgeflag = 0.5f;
+   ctx.edgeflag_attr = nv50->vertprog->cfg.edgeflag_in;
+
+   /* map vertex buffers, determine vertex size */
+   for (i = 0; i < nv50->vtxelt->num_elements; i++) {
+      struct pipe_vertex_element *ve = &nv50->vtxelt->pipe[i];
+      struct pipe_vertex_buffer *vb = &nv50->vtxbuf[ve->vertex_buffer_index];
+      struct nouveau_bo *bo = nouveau_bo(vb->buffer);
+      unsigned size, nr_components, n;
+
+      if (!(nv50->vbo_fifo & (1 << i)))
+         continue;
+      n = ctx.attr_nr++;
+
+      if (nouveau_bo_map(bo, NOUVEAU_BO_RD)) {
+         assert(bo->map);
+         return;
+      }
+      ctx.attr[n].map = bo->map + vb->buffer_offset + ve->src_offset;
+      nouveau_bo_unmap(bo);
+
+      ctx.attr[n].stride = vb->stride;
+      ctx.attr[n].divisor = ve->instance_divisor;
+      if (ctx.attr[n].divisor) {
+         ctx.attr[n].step = i_start % ve->instance_divisor;
+         ctx.attr[n].map += i_start * vb->stride;
+      }
+
+      size = util_format_get_component_bits(ve->src_format,
+                                            UTIL_FORMAT_COLORSPACE_RGB, 0);
+      nr_components = util_format_get_nr_components(ve->src_format);
+      switch (size) {
+      case 8:
+         switch (nr_components) {
+         case 1: ctx.attr[n].push = emit_b08_1; break;
+         case 2: ctx.attr[n].push = emit_b16_1; break;
+         case 3: ctx.attr[n].push = emit_b08_3; break;
+         case 4: ctx.attr[n].push = emit_b32_1; break;
+         }
+         ctx.vtx_size++;
+         break;
+      case 16:
+         switch (nr_components) {
+         case 1: ctx.attr[n].push = emit_b16_1; break;
+         case 2: ctx.attr[n].push = emit_b32_1; break;
+         case 3: ctx.attr[n].push = emit_b16_3; break;
+         case 4: ctx.attr[n].push = emit_b32_2; break;
+         }
+         ctx.vtx_size += (nr_components + 1) >> 1;
+         break;
+      case 32:
+         switch (nr_components) {
+         case 1: ctx.attr[n].push = emit_b32_1; break;
+         case 2: ctx.attr[n].push = emit_b32_2; break;
+         case 3: ctx.attr[n].push = emit_b32_3; break;
+         case 4: ctx.attr[n].push = emit_b32_4; break;
+         }
+         ctx.vtx_size += nr_components;
+         break;
+      default:
+         assert(0);
+         return;
+      }
+   }
+   vtx_size = ctx.vtx_size + v_overhead;
+
+   /* map index buffer, if present */
+   if (idxbuf) {
+      struct nouveau_bo *bo = nouveau_bo(idxbuf);
+
+      if (nouveau_bo_map(bo, NOUVEAU_BO_RD)) {
+         assert(bo->map);
+         return;
+      }
+      ctx.idxbuf = bo->map;
+      ctx.idxsize = idxsize;
+      nouveau_bo_unmap(bo);
+   }
+
+   s.priv = &ctx;
+   s.edge = emit_edgeflag;
+   if (idxbuf) {
+      if (idxsize == 1)
+         s.emit = emit_elt08;
+      else
+      if (idxsize == 2)
+         s.emit = emit_elt16;
+      else
+         s.emit = emit_elt32;
+   } else
+      s.emit = emit_verts;
+
+   /* per-instance loop */
+   BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
+   OUT_RING  (chan, NV50_CB_AUX | (24 << 8));
+   OUT_RING  (chan, i_start);
+   while (i_count--) {
+      unsigned max_verts;
+      boolean done;
+
+      for (i = 0; i < ctx.attr_nr; i++) {
+         if (!ctx.attr[i].divisor ||
+              ctx.attr[i].divisor != ++ctx.attr[i].step)
+            continue;
+         ctx.attr[i].step = 0;
+         ctx.attr[i].map += ctx.attr[i].stride;
+      }
+
+      u_split_prim_init(&s, mode, start, count);
+      do {
+         if (AVAIL_RING(chan) < p_overhead + (6 * vtx_size)) {
+            FIRE_RING(chan);
+            if (!nv50_state_validate(nv50, p_overhead + (6 * vtx_size))) {
+               assert(0);
+               return;
+            }
+         }
+
+         max_verts  = AVAIL_RING(chan);
+         max_verts -= p_overhead;
+         max_verts /= vtx_size;
+
+         BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
+         OUT_RING  (chan, nv50_prim(s.mode) | (nzi ? (1 << 28) : 0));
+         done = u_split_prim_next(&s, max_verts);
+         BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
+         OUT_RING  (chan, 0);
+      } while (!done);
+
+      nzi = TRUE;
+   }
+}
diff --git a/src/gallium/drivers/nv50/nv50_screen.c b/src/gallium/drivers/nv50/nv50_screen.c
index ee54749443c..8283524b638 100644
--- a/src/gallium/drivers/nv50/nv50_screen.c
+++ b/src/gallium/drivers/nv50/nv50_screen.c
@@ -35,9 +35,9 @@ nv50_screen_is_format_supported(struct pipe_screen *pscreen,
 {
 	if (tex_usage & PIPE_TEXTURE_USAGE_RENDER_TARGET) {
 		switch (format) {
-		case PIPE_FORMAT_X8R8G8B8_UNORM:
-		case PIPE_FORMAT_A8R8G8B8_UNORM:
-		case PIPE_FORMAT_R5G6B5_UNORM:
+		case PIPE_FORMAT_B8G8R8X8_UNORM:
+		case PIPE_FORMAT_B8G8R8A8_UNORM:
+		case PIPE_FORMAT_B5G6R5_UNORM:
 		case PIPE_FORMAT_R16G16B16A16_SNORM:
 		case PIPE_FORMAT_R16G16B16A16_UNORM:
 		case PIPE_FORMAT_R32G32B32A32_FLOAT:
@@ -51,32 +51,32 @@ nv50_screen_is_format_supported(struct pipe_screen *pscreen,
 	if (tex_usage & PIPE_TEXTURE_USAGE_DEPTH_STENCIL) {
 		switch (format) {
 		case PIPE_FORMAT_Z32_FLOAT:
-		case PIPE_FORMAT_Z24S8_UNORM:
-		case PIPE_FORMAT_X8Z24_UNORM:
 		case PIPE_FORMAT_S8Z24_UNORM:
+		case PIPE_FORMAT_Z24X8_UNORM:
+		case PIPE_FORMAT_Z24S8_UNORM:
 			return TRUE;
 		default:
 			break;
 		}
 	} else {
 		switch (format) {
-		case PIPE_FORMAT_A8R8G8B8_UNORM:
-		case PIPE_FORMAT_X8R8G8B8_UNORM:
-		case PIPE_FORMAT_A8R8G8B8_SRGB:
-		case PIPE_FORMAT_X8R8G8B8_SRGB:
-		case PIPE_FORMAT_A1R5G5B5_UNORM:
-		case PIPE_FORMAT_A4R4G4B4_UNORM:
-		case PIPE_FORMAT_R5G6B5_UNORM:
+		case PIPE_FORMAT_B8G8R8A8_UNORM:
+		case PIPE_FORMAT_B8G8R8X8_UNORM:
+		case PIPE_FORMAT_B8G8R8A8_SRGB:
+		case PIPE_FORMAT_B8G8R8X8_SRGB:
+		case PIPE_FORMAT_B5G5R5A1_UNORM:
+		case PIPE_FORMAT_B4G4R4A4_UNORM:
+		case PIPE_FORMAT_B5G6R5_UNORM:
 		case PIPE_FORMAT_L8_UNORM:
 		case PIPE_FORMAT_A8_UNORM:
 		case PIPE_FORMAT_I8_UNORM:
-		case PIPE_FORMAT_A8L8_UNORM:
+		case PIPE_FORMAT_L8A8_UNORM:
 		case PIPE_FORMAT_DXT1_RGB:
 		case PIPE_FORMAT_DXT1_RGBA:
 		case PIPE_FORMAT_DXT3_RGBA:
 		case PIPE_FORMAT_DXT5_RGBA:
-		case PIPE_FORMAT_Z24S8_UNORM:
 		case PIPE_FORMAT_S8Z24_UNORM:
+		case PIPE_FORMAT_Z24S8_UNORM:
 		case PIPE_FORMAT_Z32_FLOAT:
 		case PIPE_FORMAT_R16G16B16A16_SNORM:
 		case PIPE_FORMAT_R16G16B16A16_UNORM:
@@ -95,6 +95,8 @@ nv50_screen_is_format_supported(struct pipe_screen *pscreen,
 static int
 nv50_screen_get_param(struct pipe_screen *pscreen, int param)
 {
+	struct nv50_screen *screen = nv50_screen(pscreen);
+
 	switch (param) {
 	case PIPE_CAP_MAX_TEXTURE_IMAGE_UNITS:
 		return 32;
@@ -132,9 +134,9 @@ nv50_screen_get_param(struct pipe_screen *pscreen, int param)
 	case PIPE_CAP_BLEND_EQUATION_SEPARATE:
 		return 1;
 	case NOUVEAU_CAP_HW_VTXBUF:
-		return 1;
+		return screen->force_push ? 0 : 1;
 	case NOUVEAU_CAP_HW_IDXBUF:
-		return 1;
+		return screen->force_push ? 0 : 1;
 	case PIPE_CAP_INDEP_BLEND_ENABLE:
 		return 1;
 	case PIPE_CAP_INDEP_BLEND_FUNC:
@@ -202,28 +204,6 @@ nv50_screen_destroy(struct pipe_screen *pscreen)
 	FREE(screen);
 }
 
-static int
-nv50_pre_pipebuffer_map(struct pipe_screen *pscreen, struct pipe_buffer *pb,
-	unsigned usage)
-{
-	struct nv50_screen *screen = nv50_screen(pscreen);
-	struct nv50_context *ctx = screen->cur_ctx;
-
-	if (!(pb->usage & PIPE_BUFFER_USAGE_VERTEX))
-		return 0;
-
-	/* Our vtxbuf got mapped, it can no longer be considered part of current
-	 * state, remove it to avoid emitting reloc markers.
-	 */
-	if (ctx && ctx->state.vtxbuf && so_bo_is_reloc(ctx->state.vtxbuf,
-			nouveau_bo(pb))) {
-		so_ref(NULL, &ctx->state.vtxbuf);
-		ctx->dirty |= NV50_NEW_ARRAYS;
-	}
-
-	return 0;
-}
-
 struct pipe_screen *
 nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 {
@@ -252,7 +232,6 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	pscreen->get_paramf = nv50_screen_get_paramf;
 	pscreen->is_format_supported = nv50_screen_is_format_supported;
 	pscreen->context_create = nv50_create;
-	screen->base.pre_pipebuffer_map_callback = nv50_pre_pipebuffer_map;
 
 	nv50_screen_init_miptree_functions(pscreen);
 	nv50_transfer_init_screen_functions(pscreen);
@@ -508,10 +487,6 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	so_method(so, screen->tesla, NV50TCL_LINKED_TSC, 1);
 	so_data  (so, 1);
 
-	/* activate first scissor rectangle */
-	so_method(so, screen->tesla, NV50TCL_SCISSOR_ENABLE(0), 1);
-	so_data  (so, 1);
-
 	so_method(so, screen->tesla, NV50TCL_EDGEFLAG_ENABLE, 1);
 	so_data  (so, 1); /* default edgeflag to TRUE */
 
@@ -520,6 +495,7 @@ nv50_screen_create(struct pipe_winsys *ws, struct nouveau_device *dev)
 	so_ref (NULL, &so);
 	nouveau_pushbuf_flush(chan, 0);
 
+	screen->force_push = debug_get_bool_option("NV50_ALWAYS_PUSH", FALSE);
 	return pscreen;
 }
 
diff --git a/src/gallium/drivers/nv50/nv50_screen.h b/src/gallium/drivers/nv50/nv50_screen.h
index 2687b721277..d1bc80cb9ed 100644
--- a/src/gallium/drivers/nv50/nv50_screen.h
+++ b/src/gallium/drivers/nv50/nv50_screen.h
@@ -28,6 +28,8 @@ struct nv50_screen {
 	struct nouveau_bo *tsc;
 
 	struct nouveau_stateobj *static_init;
+
+	boolean force_push;
 };
 
 static INLINE struct nv50_screen *
diff --git a/src/gallium/drivers/nv50/nv50_state.c b/src/gallium/drivers/nv50/nv50_state.c
index dd457c0f00f..c1628089285 100644
--- a/src/gallium/drivers/nv50/nv50_state.c
+++ b/src/gallium/drivers/nv50/nv50_state.c
@@ -339,7 +339,7 @@ static void *
 nv50_rasterizer_state_create(struct pipe_context *pipe,
 			     const struct pipe_rasterizer_state *cso)
 {
-	struct nouveau_stateobj *so = so_new(15, 21, 0);
+	struct nouveau_stateobj *so = so_new(16, 22, 0);
 	struct nouveau_grobj *tesla = nv50_context(pipe)->screen->tesla;
 	struct nv50_rasterizer_stateobj *rso =
 		CALLOC_STRUCT(nv50_rasterizer_stateobj);
@@ -351,6 +351,9 @@ nv50_rasterizer_state_create(struct pipe_context *pipe,
 	 * 	- point_sprite / sprite_coord_mode
 	 */
 
+	so_method(so, tesla, NV50TCL_SCISSOR_ENABLE(0), 1);
+	so_data  (so, cso->scissor);
+
 	so_method(so, tesla, NV50TCL_SHADE_MODEL, 1);
 	so_data  (so, cso->flatshade ? NV50TCL_SHADE_MODEL_FLAT :
 				       NV50TCL_SHADE_MODEL_SMOOTH);
@@ -757,15 +760,34 @@ nv50_set_vertex_buffers(struct pipe_context *pipe, unsigned count,
 	nv50->dirty |= NV50_NEW_ARRAYS;
 }
 
+static void *
+nv50_vtxelts_state_create(struct pipe_context *pipe,
+			  unsigned num_elements,
+			  const struct pipe_vertex_element *elements)
+{
+	struct nv50_vtxelt_stateobj *cso = CALLOC_STRUCT(nv50_vtxelt_stateobj);
+
+	assert(num_elements < 16); /* not doing fallbacks yet */
+	cso->num_elements = num_elements;
+	memcpy(cso->pipe, elements, num_elements * sizeof(*elements));
+
+	nv50_vtxelt_construct(cso);
+
+	return (void *)cso;
+}
+
 static void
-nv50_set_vertex_elements(struct pipe_context *pipe, unsigned count,
-			 const struct pipe_vertex_element *ve)
+nv50_vtxelts_state_delete(struct pipe_context *pipe, void *hwcso)
 {
-	struct nv50_context *nv50 = nv50_context(pipe);
+	FREE(hwcso);
+}
 
-	memcpy(nv50->vtxelt, ve, sizeof(*ve) * count);
-	nv50->vtxelt_nr = count;
+static void
+nv50_vtxelts_state_bind(struct pipe_context *pipe, void *hwcso)
+{
+	struct nv50_context *nv50 = nv50_context(pipe);
 
+	nv50->vtxelt = hwcso;
 	nv50->dirty |= NV50_NEW_ARRAYS;
 }
 
@@ -817,7 +839,10 @@ nv50_init_state_functions(struct nv50_context *nv50)
 	nv50->pipe.set_scissor_state = nv50_set_scissor_state;
 	nv50->pipe.set_viewport_state = nv50_set_viewport_state;
 
+	nv50->pipe.create_vertex_elements_state = nv50_vtxelts_state_create;
+	nv50->pipe.delete_vertex_elements_state = nv50_vtxelts_state_delete;
+	nv50->pipe.bind_vertex_elements_state = nv50_vtxelts_state_bind;
+
 	nv50->pipe.set_vertex_buffers = nv50_set_vertex_buffers;
-	nv50->pipe.set_vertex_elements = nv50_set_vertex_elements;
 }
 
diff --git a/src/gallium/drivers/nv50/nv50_state_validate.c b/src/gallium/drivers/nv50/nv50_state_validate.c
index 741f8e8f59d..2c8e7ca7982 100644
--- a/src/gallium/drivers/nv50/nv50_state_validate.c
+++ b/src/gallium/drivers/nv50/nv50_state_validate.c
@@ -25,14 +25,8 @@
 #include "nv50_context.h"
 #include "nouveau/nouveau_stateobj.h"
 
-#define NV50_CBUF_FORMAT_CASE(n) \
-	case PIPE_FORMAT_##n: so_data(so, NV50TCL_RT_FORMAT_##n); break
-
-#define NV50_ZETA_FORMAT_CASE(n) \
-	case PIPE_FORMAT_##n: so_data(so, NV50TCL_ZETA_FORMAT_##n); break
-
-static void
-nv50_state_validate_fb(struct nv50_context *nv50)
+static struct nouveau_stateobj *
+validate_fb(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
 	struct nouveau_stateobj *so = so_new(32, 79, 18);
@@ -71,14 +65,30 @@ nv50_state_validate_fb(struct nv50_context *nv50)
 		so_reloc (so, bo, fb->cbufs[i]->offset, NOUVEAU_BO_VRAM |
 			      NOUVEAU_BO_LOW | NOUVEAU_BO_RDWR, 0, 0);
 		switch (fb->cbufs[i]->format) {
-		NV50_CBUF_FORMAT_CASE(A8R8G8B8_UNORM);
-		NV50_CBUF_FORMAT_CASE(X8R8G8B8_UNORM);
-		NV50_CBUF_FORMAT_CASE(R5G6B5_UNORM);
-		NV50_CBUF_FORMAT_CASE(R16G16B16A16_SNORM);
-		NV50_CBUF_FORMAT_CASE(R16G16B16A16_UNORM);
-		NV50_CBUF_FORMAT_CASE(R32G32B32A32_FLOAT);
-		NV50_CBUF_FORMAT_CASE(R16G16_SNORM);
-		NV50_CBUF_FORMAT_CASE(R16G16_UNORM);
+		case PIPE_FORMAT_B8G8R8A8_UNORM:
+			so_data(so, NV50TCL_RT_FORMAT_A8R8G8B8_UNORM);
+			break;
+		case PIPE_FORMAT_B8G8R8X8_UNORM:
+			so_data(so, NV50TCL_RT_FORMAT_X8R8G8B8_UNORM);
+			break;
+		case PIPE_FORMAT_B5G6R5_UNORM:
+			so_data(so, NV50TCL_RT_FORMAT_R5G6B5_UNORM);
+			break;
+		case PIPE_FORMAT_R16G16B16A16_SNORM:
+			so_data(so, NV50TCL_RT_FORMAT_R16G16B16A16_SNORM);
+			break;
+		case PIPE_FORMAT_R16G16B16A16_UNORM:
+			so_data(so, NV50TCL_RT_FORMAT_R16G16B16A16_UNORM);
+			break;
+		case PIPE_FORMAT_R32G32B32A32_FLOAT:
+			so_data(so, NV50TCL_RT_FORMAT_R32G32B32A32_FLOAT);
+			break;
+		case PIPE_FORMAT_R16G16_SNORM:
+			so_data(so, NV50TCL_RT_FORMAT_R16G16_SNORM);
+			break;
+		case PIPE_FORMAT_R16G16_UNORM:
+			so_data(so, NV50TCL_RT_FORMAT_R16G16_UNORM);
+			break;
 		default:
 			NOUVEAU_ERR("AIIII unknown format %s\n",
 			            util_format_name(fb->cbufs[i]->format));
@@ -112,10 +122,18 @@ nv50_state_validate_fb(struct nv50_context *nv50)
 		so_reloc (so, bo, fb->zsbuf->offset, NOUVEAU_BO_VRAM |
 			      NOUVEAU_BO_LOW | NOUVEAU_BO_RDWR, 0, 0);
 		switch (fb->zsbuf->format) {
-		NV50_ZETA_FORMAT_CASE(S8Z24_UNORM);
-		NV50_ZETA_FORMAT_CASE(X8Z24_UNORM);
-		NV50_ZETA_FORMAT_CASE(Z24S8_UNORM);
-		NV50_ZETA_FORMAT_CASE(Z32_FLOAT);
+		case PIPE_FORMAT_Z24S8_UNORM:
+			so_data(so, NV50TCL_ZETA_FORMAT_S8Z24_UNORM);
+			break;
+		case PIPE_FORMAT_Z24X8_UNORM:
+			so_data(so, NV50TCL_ZETA_FORMAT_X8Z24_UNORM);
+			break;
+		case PIPE_FORMAT_S8Z24_UNORM:
+			so_data(so, NV50TCL_ZETA_FORMAT_Z24S8_UNORM);
+			break;
+		case PIPE_FORMAT_Z32_FLOAT:
+			so_data(so, NV50TCL_ZETA_FORMAT_Z32_FLOAT);
+			break;
 		default:
 			NOUVEAU_ERR("AIIII unknown format %s\n",
 			            util_format_name(fb->zsbuf->format));
@@ -149,12 +167,7 @@ nv50_state_validate_fb(struct nv50_context *nv50)
 	so_data  (so, w << 16);
 	so_data  (so, h << 16);
 
-	/* we set scissors to framebuffer size when they're 'turned off' */
-	nv50->dirty |= NV50_NEW_SCISSOR;
-	so_ref(NULL, &nv50->state.scissor);
-
-	so_ref(so, &nv50->state.fb);
-	so_ref(NULL, &so);
+	return so;
 }
 
 static void
@@ -181,281 +194,256 @@ nv50_validate_samplers(struct nv50_context *nv50, struct nouveau_stateobj *so,
 	}
 }
 
-static void
-nv50_state_emit(struct nv50_context *nv50)
+static struct nouveau_stateobj *
+validate_blend(struct nv50_context *nv50)
+{
+	struct nouveau_stateobj *so = NULL;
+	so_ref(nv50->blend->so, &so);
+	return so;
+}
+
+static struct nouveau_stateobj *
+validate_zsa(struct nv50_context *nv50)
 {
-	struct nv50_screen *screen = nv50->screen;
-	struct nouveau_channel *chan = screen->base.channel;
+	struct nouveau_stateobj *so = NULL;
+	so_ref(nv50->zsa->so, &so);
+	return so;
+}
 
-	/* XXX: this is racy for multiple contexts active on separate
-	 * threads.
-	 */
-	if (screen->cur_ctx != nv50) {
-		if (nv50->state.fb)
-			nv50->state.dirty |= NV50_NEW_FRAMEBUFFER;
-		if (nv50->state.blend)
-			nv50->state.dirty |= NV50_NEW_BLEND;
-		if (nv50->state.zsa)
-			nv50->state.dirty |= NV50_NEW_ZSA;
-		if (nv50->state.vertprog)
-			nv50->state.dirty |= NV50_NEW_VERTPROG;
-		if (nv50->state.fragprog)
-			nv50->state.dirty |= NV50_NEW_FRAGPROG;
-		if (nv50->state.geomprog)
-			nv50->state.dirty |= NV50_NEW_GEOMPROG;
-		if (nv50->state.rast)
-			nv50->state.dirty |= NV50_NEW_RASTERIZER;
-		if (nv50->state.blend_colour)
-			nv50->state.dirty |= NV50_NEW_BLEND_COLOUR;
-		if (nv50->state.stencil_ref)
-			nv50->state.dirty |= NV50_NEW_STENCIL_REF;
-		if (nv50->state.stipple)
-			nv50->state.dirty |= NV50_NEW_STIPPLE;
-		if (nv50->state.scissor)
-			nv50->state.dirty |= NV50_NEW_SCISSOR;
-		if (nv50->state.viewport)
-			nv50->state.dirty |= NV50_NEW_VIEWPORT;
-		if (nv50->state.tsc_upload)
-			nv50->state.dirty |= NV50_NEW_SAMPLER;
-		if (nv50->state.tic_upload)
-			nv50->state.dirty |= NV50_NEW_TEXTURE;
-		if (nv50->state.vtxfmt && nv50->state.vtxbuf)
-			nv50->state.dirty |= NV50_NEW_ARRAYS;
-		screen->cur_ctx = nv50;
-	}
+static struct nouveau_stateobj *
+validate_rast(struct nv50_context *nv50)
+{
+	struct nouveau_stateobj *so = NULL;
+	so_ref(nv50->rasterizer->so, &so);
+	return so;
+}
 
-	if (nv50->state.dirty & NV50_NEW_FRAMEBUFFER)
-		so_emit(chan, nv50->state.fb);
-	if (nv50->state.dirty & NV50_NEW_BLEND)
-		so_emit(chan, nv50->state.blend);
-	if (nv50->state.dirty & NV50_NEW_ZSA)
-		so_emit(chan, nv50->state.zsa);
-	if (nv50->state.dirty & NV50_NEW_VERTPROG)
-		so_emit(chan, nv50->state.vertprog);
-	if (nv50->state.dirty & NV50_NEW_FRAGPROG)
-		so_emit(chan, nv50->state.fragprog);
-	if (nv50->state.dirty & NV50_NEW_GEOMPROG && nv50->state.geomprog)
-		so_emit(chan, nv50->state.geomprog);
-	if (nv50->state.dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG |
-				 NV50_NEW_GEOMPROG | NV50_NEW_RASTERIZER))
-		so_emit(chan, nv50->state.fp_linkage);
-	if ((nv50->state.dirty & (NV50_NEW_VERTPROG | NV50_NEW_GEOMPROG))
-	    && nv50->state.gp_linkage)
-		so_emit(chan, nv50->state.gp_linkage);
-	if (nv50->state.dirty & NV50_NEW_RASTERIZER)
-		so_emit(chan, nv50->state.rast);
-	if (nv50->state.dirty & NV50_NEW_BLEND_COLOUR)
-		so_emit(chan, nv50->state.blend_colour);
-	if (nv50->state.dirty & NV50_NEW_STENCIL_REF)
-		so_emit(chan, nv50->state.stencil_ref);
-	if (nv50->state.dirty & NV50_NEW_STIPPLE)
-		so_emit(chan, nv50->state.stipple);
-	if (nv50->state.dirty & NV50_NEW_SCISSOR)
-		so_emit(chan, nv50->state.scissor);
-	if (nv50->state.dirty & NV50_NEW_VIEWPORT)
-		so_emit(chan, nv50->state.viewport);
-	if (nv50->state.dirty & NV50_NEW_SAMPLER)
-		so_emit(chan, nv50->state.tsc_upload);
-	if (nv50->state.dirty & NV50_NEW_TEXTURE)
-		so_emit(chan, nv50->state.tic_upload);
-	if (nv50->state.dirty & NV50_NEW_ARRAYS) {
-		so_emit(chan, nv50->state.vtxfmt);
-		so_emit(chan, nv50->state.vtxbuf);
-		if (nv50->state.vtxattr)
-			so_emit(chan, nv50->state.vtxattr);
-	}
-	nv50->state.dirty = 0;
+static struct nouveau_stateobj *
+validate_blend_colour(struct nv50_context *nv50)
+{
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nouveau_stateobj *so = so_new(1, 4, 0);
+
+	so_method(so, tesla, NV50TCL_BLEND_COLOR(0), 4);
+	so_data  (so, fui(nv50->blend_colour.color[0]));
+	so_data  (so, fui(nv50->blend_colour.color[1]));
+	so_data  (so, fui(nv50->blend_colour.color[2]));
+	so_data  (so, fui(nv50->blend_colour.color[3]));
+	return so;
 }
 
-void
-nv50_state_flush_notify(struct nouveau_channel *chan)
+static struct nouveau_stateobj *
+validate_stencil_ref(struct nv50_context *nv50)
 {
-	struct nv50_context *nv50 = chan->user_private;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nouveau_stateobj *so = so = so_new(2, 2, 0);
 
-	if (nv50->state.tic_upload && !(nv50->dirty & NV50_NEW_TEXTURE))
-		so_emit(chan, nv50->state.tic_upload);
+	so_method(so, tesla, NV50TCL_STENCIL_FRONT_FUNC_REF, 1);
+	so_data  (so, nv50->stencil_ref.ref_value[0]);
+	so_method(so, tesla, NV50TCL_STENCIL_BACK_FUNC_REF, 1);
+	so_data  (so, nv50->stencil_ref.ref_value[1]);
+	return so;
+}
 
-	so_emit_reloc_markers(chan, nv50->state.fb);
-	so_emit_reloc_markers(chan, nv50->state.vertprog);
-	so_emit_reloc_markers(chan, nv50->state.fragprog);
-	so_emit_reloc_markers(chan, nv50->state.vtxbuf);
-	so_emit_reloc_markers(chan, nv50->screen->static_init);
+static struct nouveau_stateobj *
+validate_stipple(struct nv50_context *nv50)
+{
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nouveau_stateobj *so = so_new(1, 32, 0);
+	int i;
 
-	if (nv50->state.instbuf)
-		so_emit_reloc_markers(chan, nv50->state.instbuf);
+	so_method(so, tesla, NV50TCL_POLYGON_STIPPLE_PATTERN(0), 32);
+	for (i = 0; i < 32; i++)
+		so_data(so, util_bswap32(nv50->stipple.stipple[i]));
+	return so;
 }
 
-boolean
-nv50_state_validate(struct nv50_context *nv50)
+static struct nouveau_stateobj *
+validate_scissor(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
+        struct pipe_scissor_state *s = &nv50->scissor;
 	struct nouveau_stateobj *so;
-	unsigned i;
 
-	if (nv50->dirty & NV50_NEW_FRAMEBUFFER)
-		nv50_state_validate_fb(nv50);
+	so = so_new(1, 2, 0);
+	so_method(so, tesla, NV50TCL_SCISSOR_HORIZ(0), 2);
+	so_data  (so, (s->maxx << 16) | s->minx);
+	so_data  (so, (s->maxy << 16) | s->miny);
+	return so;
+}
+
+static struct nouveau_stateobj *
+validate_viewport(struct nv50_context *nv50)
+{
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nouveau_stateobj *so = so_new(5, 9, 0);
+
+	so_method(so, tesla, NV50TCL_VIEWPORT_TRANSLATE_X(0), 3);
+	so_data  (so, fui(nv50->viewport.translate[0]));
+	so_data  (so, fui(nv50->viewport.translate[1]));
+	so_data  (so, fui(nv50->viewport.translate[2]));
+	so_method(so, tesla, NV50TCL_VIEWPORT_SCALE_X(0), 3);
+	so_data  (so, fui(nv50->viewport.scale[0]));
+	so_data  (so, fui(nv50->viewport.scale[1]));
+	so_data  (so, fui(nv50->viewport.scale[2]));
+
+	so_method(so, tesla, NV50TCL_VIEWPORT_TRANSFORM_EN, 1);
+	so_data  (so, 1);
+	/* 0x0000 = remove whole primitive only (xyz)
+	 * 0x1018 = remove whole primitive only (xy), clamp z
+	 * 0x1080 = clip primitive (xyz)
+	 * 0x1098 = clip primitive (xy), clamp z
+	 */
+	so_method(so, tesla, NV50TCL_VIEW_VOLUME_CLIP_CTRL, 1);
+	so_data  (so, 0x1080);
+	/* no idea what 0f90 does */
+	so_method(so, tesla, 0x0f90, 1);
+	so_data  (so, 0);
+
+	return so;
+}
 
-	if (nv50->dirty & NV50_NEW_BLEND)
-		so_ref(nv50->blend->so, &nv50->state.blend);
+static struct nouveau_stateobj *
+validate_sampler(struct nv50_context *nv50)
+{
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct nouveau_stateobj *so;
+	unsigned nr = 0, i;
 
-	if (nv50->dirty & NV50_NEW_ZSA)
-		so_ref(nv50->zsa->so, &nv50->state.zsa);
+	for (i = 0; i < PIPE_SHADER_TYPES; ++i)
+		nr += nv50->sampler_nr[i];
 
-	if (nv50->dirty & (NV50_NEW_VERTPROG | NV50_NEW_VERTPROG_CB))
-		nv50_vertprog_validate(nv50);
+	so = so_new(1 + 5 * PIPE_SHADER_TYPES,
+		    1 + 19 * PIPE_SHADER_TYPES + nr * 8,
+		    PIPE_SHADER_TYPES * 2);
 
-	if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_FRAGPROG_CB))
-		nv50_fragprog_validate(nv50);
+	nv50_validate_samplers(nv50, so, PIPE_SHADER_VERTEX);
+	nv50_validate_samplers(nv50, so, PIPE_SHADER_FRAGMENT);
 
-	if (nv50->dirty & (NV50_NEW_GEOMPROG | NV50_NEW_GEOMPROG_CB))
-		nv50_geomprog_validate(nv50);
+	so_method(so, tesla, 0x1334, 1); /* flush TSC */
+	so_data  (so, 0);
 
-	if (nv50->dirty & (NV50_NEW_FRAGPROG | NV50_NEW_VERTPROG |
-			   NV50_NEW_GEOMPROG | NV50_NEW_RASTERIZER))
-		nv50_fp_linkage_validate(nv50);
+	return so;
+}
 
-	if (nv50->dirty & (NV50_NEW_GEOMPROG | NV50_NEW_VERTPROG))
-		nv50_gp_linkage_validate(nv50);
+static struct nouveau_stateobj *
+validate_vtxbuf(struct nv50_context *nv50)
+{
+	struct nouveau_stateobj *so = NULL;
+	so_ref(nv50->state.vtxbuf, &so);
+	return so;
+}
 
-	if (nv50->dirty & NV50_NEW_RASTERIZER)
-		so_ref(nv50->rasterizer->so, &nv50->state.rast);
+static struct nouveau_stateobj *
+validate_vtxattr(struct nv50_context *nv50)
+{
+	struct nouveau_stateobj *so = NULL;
+	so_ref(nv50->state.vtxattr, &so);
+	return so;
+}
 
-	if (nv50->dirty & NV50_NEW_BLEND_COLOUR) {
-		so = so_new(1, 4, 0);
-		so_method(so, tesla, NV50TCL_BLEND_COLOR(0), 4);
-		so_data  (so, fui(nv50->blend_colour.color[0]));
-		so_data  (so, fui(nv50->blend_colour.color[1]));
-		so_data  (so, fui(nv50->blend_colour.color[2]));
-		so_data  (so, fui(nv50->blend_colour.color[3]));
-		so_ref(so, &nv50->state.blend_colour);
-		so_ref(NULL, &so);
-	}
+struct state_validate {
+	struct nouveau_stateobj *(*func)(struct nv50_context *nv50);
+	unsigned states;
+} validate_list[] = {
+	{ validate_fb             , NV50_NEW_FRAMEBUFFER                      },
+	{ validate_blend          , NV50_NEW_BLEND                            },
+	{ validate_zsa            , NV50_NEW_ZSA                              },
+	{ nv50_vertprog_validate  , NV50_NEW_VERTPROG | NV50_NEW_VERTPROG_CB  },
+	{ nv50_fragprog_validate  , NV50_NEW_FRAGPROG | NV50_NEW_FRAGPROG_CB  },
+	{ nv50_geomprog_validate  , NV50_NEW_GEOMPROG | NV50_NEW_GEOMPROG_CB  },
+	{ nv50_fp_linkage_validate, NV50_NEW_VERTPROG | NV50_NEW_GEOMPROG |
+				    NV50_NEW_FRAGPROG | NV50_NEW_RASTERIZER   },
+	{ nv50_gp_linkage_validate, NV50_NEW_VERTPROG | NV50_NEW_GEOMPROG     },
+	{ validate_rast           , NV50_NEW_RASTERIZER                       },
+	{ validate_blend_colour   , NV50_NEW_BLEND_COLOUR                     },
+	{ validate_stencil_ref    , NV50_NEW_STENCIL_REF                      },
+	{ validate_stipple        , NV50_NEW_STIPPLE                          },
+	{ validate_scissor        , NV50_NEW_SCISSOR                          },
+	{ validate_viewport       , NV50_NEW_VIEWPORT                         },
+	{ validate_sampler        , NV50_NEW_SAMPLER                          },
+	{ nv50_tex_validate       , NV50_NEW_TEXTURE | NV50_NEW_SAMPLER       },
+	{ nv50_vbo_validate       , NV50_NEW_ARRAYS                           },
+	{ validate_vtxbuf         , NV50_NEW_ARRAYS                           },
+	{ validate_vtxattr        , NV50_NEW_ARRAYS                           },
+	{}
+};
+#define validate_list_len (sizeof(validate_list) / sizeof(validate_list[0]))
 
-	if (nv50->dirty & NV50_NEW_STENCIL_REF) {
-		so = so_new(2, 2, 0);
-		so_method(so, tesla, NV50TCL_STENCIL_FRONT_FUNC_REF, 1);
-		so_data  (so, nv50->stencil_ref.ref_value[0]);
-		so_method(so, tesla, NV50TCL_STENCIL_BACK_FUNC_REF, 1);
-		so_data  (so, nv50->stencil_ref.ref_value[1]);
-		so_ref(so, &nv50->state.stencil_ref);
-		so_ref(NULL, &so);
-	}
+boolean
+nv50_state_validate(struct nv50_context *nv50, unsigned wait_dwords)
+{
+	struct nouveau_channel *chan = nv50->screen->base.channel;
+	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	unsigned nr_relocs = 128, nr_dwords = wait_dwords + 128 + 4;
+	int ret, i;
 
-	if (nv50->dirty & NV50_NEW_STIPPLE) {
-		so = so_new(1, 32, 0);
-		so_method(so, tesla, NV50TCL_POLYGON_STIPPLE_PATTERN(0), 32);
-		for (i = 0; i < 32; i++)
-			so_data(so, util_bswap32(nv50->stipple.stipple[i]));
-		so_ref(so, &nv50->state.stipple);
-		so_ref(NULL, &so);
-	}
+	for (i = 0; i < validate_list_len; i++) {
+		struct state_validate *validate = &validate_list[i];
+		struct nouveau_stateobj *so;
 
-	if (nv50->dirty & (NV50_NEW_SCISSOR | NV50_NEW_RASTERIZER)) {
-		struct pipe_rasterizer_state *rast = &nv50->rasterizer->pipe;
-		struct pipe_scissor_state *s = &nv50->scissor;
+		if (!(nv50->dirty & validate->states))
+			continue;
 
-		if (nv50->state.scissor &&
-		    (rast->scissor == 0 && nv50->state.scissor_enabled == 0))
-			goto scissor_uptodate;
-		nv50->state.scissor_enabled = rast->scissor;
+		so = validate->func(nv50);
+		if (!so)
+			continue;
 
-		so = so_new(1, 2, 0);
-		so_method(so, tesla, NV50TCL_SCISSOR_HORIZ(0), 2);
-		if (nv50->state.scissor_enabled) {
-			so_data(so, (s->maxx << 16) | s->minx);
-			so_data(so, (s->maxy << 16) | s->miny);
-		} else {
-			so_data(so, (nv50->framebuffer.width << 16));
-			so_data(so, (nv50->framebuffer.height << 16));
-		}
-		so_ref(so, &nv50->state.scissor);
-		so_ref(NULL, &so);
-		nv50->state.dirty |= NV50_NEW_SCISSOR;
-	}
-scissor_uptodate:
-
-	if (nv50->dirty & (NV50_NEW_VIEWPORT | NV50_NEW_RASTERIZER)) {
-		unsigned bypass;
-
-		if (!nv50->rasterizer->pipe.bypass_vs_clip_and_viewport)
-			bypass = 0;
-		else
-			bypass = 1;
-
-		if (nv50->state.viewport &&
-		    (bypass || !(nv50->dirty & NV50_NEW_VIEWPORT)) &&
-		    nv50->state.viewport_bypass == bypass)
-			goto viewport_uptodate;
-		nv50->state.viewport_bypass = bypass;
-
-		so = so_new(5, 9, 0);
-		if (!bypass) {
-			so_method(so, tesla, NV50TCL_VIEWPORT_TRANSLATE_X(0), 3);
-			so_data  (so, fui(nv50->viewport.translate[0]));
-			so_data  (so, fui(nv50->viewport.translate[1]));
-			so_data  (so, fui(nv50->viewport.translate[2]));
-			so_method(so, tesla, NV50TCL_VIEWPORT_SCALE_X(0), 3);
-			so_data  (so, fui(nv50->viewport.scale[0]));
-			so_data  (so, fui(nv50->viewport.scale[1]));
-			so_data  (so, fui(nv50->viewport.scale[2]));
-
-			so_method(so, tesla, NV50TCL_VIEWPORT_TRANSFORM_EN, 1);
-			so_data  (so, 1);
-			/* 0x0000 = remove whole primitive only (xyz)
-			 * 0x1018 = remove whole primitive only (xy), clamp z
-			 * 0x1080 = clip primitive (xyz)
-			 * 0x1098 = clip primitive (xy), clamp z
-			 */
-			so_method(so, tesla, NV50TCL_VIEW_VOLUME_CLIP_CTRL, 1);
-			so_data  (so, 0x1080);
-			/* no idea what 0f90 does */
-			so_method(so, tesla, 0x0f90, 1);
-			so_data  (so, 0);
-		} else {
-			so_method(so, tesla, NV50TCL_VIEWPORT_TRANSFORM_EN, 1);
-			so_data  (so, 0);
-			so_method(so, tesla, NV50TCL_VIEW_VOLUME_CLIP_CTRL, 1);
-			so_data  (so, 0x0000);
-			so_method(so, tesla, 0x0f90, 1);
-			so_data  (so, 1);
-		}
+		nr_dwords += (so->total + so->cur);
+		nr_relocs += so->cur_reloc;
 
-		so_ref(so, &nv50->state.viewport);
+		so_ref(so, &nv50->state.hw[i]);
 		so_ref(NULL, &so);
-		nv50->state.dirty |= NV50_NEW_VIEWPORT;
+		nv50->state.hw_dirty |= (1 << i);
 	}
-viewport_uptodate:
-
-	if (nv50->dirty & NV50_NEW_SAMPLER) {
-		unsigned nr = 0;
-
-		for (i = 0; i < PIPE_SHADER_TYPES; ++i)
-			nr += nv50->sampler_nr[i];
+	nv50->dirty = 0;
 
-		so = so_new(1 + 5 * PIPE_SHADER_TYPES,
-			    1 + 19 * PIPE_SHADER_TYPES + nr * 8,
-			    PIPE_SHADER_TYPES * 2);
+	if (nv50->screen->cur_ctx != nv50) {
+		for (i = 0; i < validate_list_len; i++) {
+			if (!nv50->state.hw[i] ||
+			    (nv50->state.hw_dirty & (1 << i)))
+				continue;
 
-		nv50_validate_samplers(nv50, so, 0); /* VP samplers */
-		nv50_validate_samplers(nv50, so, 2); /* FP samplers */
+			nr_dwords += (nv50->state.hw[i]->total +
+				      nv50->state.hw[i]->cur);
+			nr_relocs += nv50->state.hw[i]->cur_reloc;
+			nv50->state.hw_dirty |= (1 << i);
+		}
 
-		so_method(so, tesla, 0x1334, 1); /* flush TSC */
-		so_data  (so, 0);
+		nv50->screen->cur_ctx = nv50;
+	}
 
-		so_ref(so, &nv50->state.tsc_upload);
-		so_ref(NULL, &so);
+	ret = MARK_RING(chan, nr_dwords, nr_relocs);
+	if (ret) {
+		debug_printf("MARK_RING(%d, %d) failed: %d\n",
+			     nr_dwords, nr_relocs, ret);
+		return FALSE;
 	}
 
-	if (nv50->dirty & (NV50_NEW_TEXTURE | NV50_NEW_SAMPLER))
-		nv50_tex_validate(nv50);
+	while (nv50->state.hw_dirty) {
+		i = ffs(nv50->state.hw_dirty) - 1;
+		nv50->state.hw_dirty &= ~(1 << i);
 
-	if (nv50->dirty & NV50_NEW_ARRAYS)
-		nv50_vbo_validate(nv50);
+		so_emit(chan, nv50->state.hw[i]);
+	}
 
-	nv50->state.dirty |= nv50->dirty;
-	nv50->dirty = 0;
-	nv50_state_emit(nv50);
+	/* Yes, really, we need to do this.  If a buffer that is referenced
+	 * on the hardware isn't part of changed state above, without doing
+	 * this the kernel is given no clue that the buffer is being used
+	 * still.  This can cause all sorts of fun issues.
+	 */
+	nv50_tex_relocs(nv50);
+	so_emit_reloc_markers(chan, nv50->state.hw[0]); /* fb */
+	so_emit_reloc_markers(chan, nv50->state.hw[3]); /* vp */
+	so_emit_reloc_markers(chan, nv50->state.hw[4]); /* fp */
+	so_emit_reloc_markers(chan, nv50->state.hw[17]); /* vb */
+	so_emit_reloc_markers(chan, nv50->screen->static_init);
 
+	/* No idea.. */
+	BEGIN_RING(chan, tesla, 0x142c, 1);
+	OUT_RING  (chan, 0);
+	BEGIN_RING(chan, tesla, 0x142c, 1);
+	OUT_RING  (chan, 0);
 	return TRUE;
 }
 
diff --git a/src/gallium/drivers/nv50/nv50_surface.c b/src/gallium/drivers/nv50/nv50_surface.c
index 7405b67414a..cabd148bc5b 100644
--- a/src/gallium/drivers/nv50/nv50_surface.c
+++ b/src/gallium/drivers/nv50/nv50_surface.c
@@ -33,11 +33,11 @@ static INLINE int
 nv50_format(enum pipe_format format)
 {
 	switch (format) {
-	case PIPE_FORMAT_A8R8G8B8_UNORM:
+	case PIPE_FORMAT_B8G8R8A8_UNORM:
 		return NV50_2D_DST_FORMAT_A8R8G8B8_UNORM;
-	case PIPE_FORMAT_X8R8G8B8_UNORM:
+	case PIPE_FORMAT_B8G8R8X8_UNORM:
 		return NV50_2D_DST_FORMAT_X8R8G8B8_UNORM;
-	case PIPE_FORMAT_R5G6B5_UNORM:
+	case PIPE_FORMAT_B5G6R5_UNORM:
 		return NV50_2D_DST_FORMAT_R5G6B5_UNORM;
 	case PIPE_FORMAT_A8_UNORM:
 		return NV50_2D_DST_FORMAT_R8_UNORM;
diff --git a/src/gallium/drivers/nv50/nv50_tex.c b/src/gallium/drivers/nv50/nv50_tex.c
index 2f2ebc8ec43..c5029bad2d8 100644
--- a/src/gallium/drivers/nv50/nv50_tex.c
+++ b/src/gallium/drivers/nv50/nv50_tex.c
@@ -24,6 +24,7 @@
 #include "nv50_texture.h"
 
 #include "nouveau/nouveau_stateobj.h"
+#include "nouveau/nouveau_reloc.h"
 
 #include "util/u_format.h"
 
@@ -39,14 +40,14 @@
 
 static const uint32_t nv50_texture_formats[PIPE_FORMAT_COUNT] =
 {
-	_(A8R8G8B8_UNORM, UNORM, C2, C1, C0, C3,  8_8_8_8),
-	_(A8R8G8B8_SRGB,  UNORM, C2, C1, C0, C3,  8_8_8_8),
-	_(X8R8G8B8_UNORM, UNORM, C2, C1, C0, ONE, 8_8_8_8),
-	_(X8R8G8B8_SRGB,  UNORM, C2, C1, C0, ONE, 8_8_8_8),
-	_(A1R5G5B5_UNORM, UNORM, C2, C1, C0, C3,  1_5_5_5),
-	_(A4R4G4B4_UNORM, UNORM, C2, C1, C0, C3,  4_4_4_4),
+	_(B8G8R8A8_UNORM, UNORM, C2, C1, C0, C3,  8_8_8_8),
+	_(B8G8R8A8_SRGB,  UNORM, C2, C1, C0, C3,  8_8_8_8),
+	_(B8G8R8X8_UNORM, UNORM, C2, C1, C0, ONE, 8_8_8_8),
+	_(B8G8R8X8_SRGB,  UNORM, C2, C1, C0, ONE, 8_8_8_8),
+	_(B5G5R5A1_UNORM, UNORM, C2, C1, C0, C3,  1_5_5_5),
+	_(B4G4R4A4_UNORM, UNORM, C2, C1, C0, C3,  4_4_4_4),
 
-	_(R5G6B5_UNORM, UNORM, C2, C1, C0, ONE, 5_6_5),
+	_(B5G6R5_UNORM, UNORM, C2, C1, C0, ONE, 5_6_5),
 
 	_(L8_UNORM, UNORM, C0, C0, C0, ONE, 8),
 	_(L8_SRGB,  UNORM, C0, C0, C0, ONE, 8),
@@ -61,8 +62,8 @@ static const uint32_t nv50_texture_formats[PIPE_FORMAT_COUNT] =
 	_(DXT3_RGBA, UNORM, C0, C1, C2, C3, DXT3),
 	_(DXT5_RGBA, UNORM, C0, C1, C2, C3, DXT5),
 
-	_MIXED(Z24S8_UNORM, UINT, UNORM, UINT, UINT, C1, C1, C1, ONE, 24_8),
-	_MIXED(S8Z24_UNORM, UNORM, UINT, UINT, UINT, C0, C0, C0, ONE, 8_24),
+	_MIXED(S8Z24_UNORM, UINT, UNORM, UINT, UINT, C1, C1, C1, ONE, 24_8),
+	_MIXED(Z24S8_UNORM, UNORM, UINT, UINT, UINT, C0, C0, C0, ONE, 8_24),
 
 	_(R16G16B16A16_SNORM, UNORM, C0, C1, C2, C3, 16_16_16_16),
 	_(R16G16B16A16_UNORM, SNORM, C0, C1, C2, C3, 16_16_16_16),
@@ -214,6 +215,35 @@ nv50_validate_textures(struct nv50_context *nv50, struct nouveau_stateobj *so,
 }
 
 void
+nv50_tex_relocs(struct nv50_context *nv50)
+{
+	struct nouveau_channel *chan = nv50->screen->tesla->channel;
+	int p, unit;
+
+	p = PIPE_SHADER_FRAGMENT;
+	for (unit = 0; unit < nv50->miptree_nr[p]; unit++) {
+		if (!nv50->miptree[p][unit])
+			continue;
+		nouveau_reloc_emit(chan, nv50->screen->tic,
+				   ((p * 32) + unit) * 32, NULL,
+				   nv50->miptree[p][unit]->base.bo, 0, 0,
+				   NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW |
+				   NOUVEAU_BO_RD, 0, 0);
+	}
+
+	p = PIPE_SHADER_VERTEX;
+	for (unit = 0; unit < nv50->miptree_nr[p]; unit++) {
+		if (!nv50->miptree[p][unit])
+			continue;
+		nouveau_reloc_emit(chan, nv50->screen->tic,
+				   ((p * 32) + unit) * 32, NULL,
+				   nv50->miptree[p][unit]->base.bo, 0, 0,
+				   NOUVEAU_BO_VRAM | NOUVEAU_BO_LOW |
+				   NOUVEAU_BO_RD, 0, 0);
+	}
+}
+
+struct nouveau_stateobj *
 nv50_tex_validate(struct nv50_context *nv50)
 {
 	struct nouveau_stateobj *so;
@@ -238,12 +268,11 @@ nv50_tex_validate(struct nv50_context *nv50)
 		so_ref(NULL, &so);
 
 		NOUVEAU_ERR("failed tex validate\n");
-		return;
+		return NULL;
 	}
 
 	so_method(so, tesla, 0x1330, 1); /* flush TIC */
 	so_data  (so, 0);
 
-	so_ref(so, &nv50->state.tic_upload);
-	so_ref(NULL, &so);
+	return so;
 }
diff --git a/src/gallium/drivers/nv50/nv50_vbo.c b/src/gallium/drivers/nv50/nv50_vbo.c
index 1c8ee0b9adf..6b9c1ee231e 100644
--- a/src/gallium/drivers/nv50/nv50_vbo.c
+++ b/src/gallium/drivers/nv50/nv50_vbo.c
@@ -25,53 +25,9 @@
 #include "util/u_inlines.h"
 #include "util/u_format.h"
 
+#include "nouveau/nouveau_util.h"
 #include "nv50_context.h"
 
-static boolean
-nv50_push_elements_u08(struct nv50_context *, uint8_t *, unsigned);
-
-static boolean
-nv50_push_elements_u16(struct nv50_context *, uint16_t *, unsigned);
-
-static boolean
-nv50_push_elements_u32(struct nv50_context *, uint32_t *, unsigned);
-
-static boolean
-nv50_push_arrays(struct nv50_context *, unsigned, unsigned);
-
-#define NV50_USING_LOATHED_EDGEFLAG(ctx) ((ctx)->vertprog->cfg.edgeflag_in < 16)
-
-static INLINE unsigned
-nv50_prim(unsigned mode)
-{
-	switch (mode) {
-	case PIPE_PRIM_POINTS: return NV50TCL_VERTEX_BEGIN_POINTS;
-	case PIPE_PRIM_LINES: return NV50TCL_VERTEX_BEGIN_LINES;
-	case PIPE_PRIM_LINE_LOOP: return NV50TCL_VERTEX_BEGIN_LINE_LOOP;
-	case PIPE_PRIM_LINE_STRIP: return NV50TCL_VERTEX_BEGIN_LINE_STRIP;
-	case PIPE_PRIM_TRIANGLES: return NV50TCL_VERTEX_BEGIN_TRIANGLES;
-	case PIPE_PRIM_TRIANGLE_STRIP:
-		return NV50TCL_VERTEX_BEGIN_TRIANGLE_STRIP;
-	case PIPE_PRIM_TRIANGLE_FAN: return NV50TCL_VERTEX_BEGIN_TRIANGLE_FAN;
-	case PIPE_PRIM_QUADS: return NV50TCL_VERTEX_BEGIN_QUADS;
-	case PIPE_PRIM_QUAD_STRIP: return NV50TCL_VERTEX_BEGIN_QUAD_STRIP;
-	case PIPE_PRIM_POLYGON: return NV50TCL_VERTEX_BEGIN_POLYGON;
-	case PIPE_PRIM_LINES_ADJACENCY:
-		return NV50TCL_VERTEX_BEGIN_LINES_ADJACENCY;
-	case PIPE_PRIM_LINE_STRIP_ADJACENCY:
-		return NV50TCL_VERTEX_BEGIN_LINE_STRIP_ADJACENCY;
-	case PIPE_PRIM_TRIANGLES_ADJACENCY:
-		return NV50TCL_VERTEX_BEGIN_TRIANGLES_ADJACENCY;
-	case PIPE_PRIM_TRIANGLE_STRIP_ADJACENCY:
-		return NV50TCL_VERTEX_BEGIN_TRIANGLE_STRIP_ADJACENCY;
-	default:
-		break;
-	}
-
-	NOUVEAU_ERR("invalid primitive type %d\n", mode);
-	return NV50TCL_VERTEX_BEGIN_POINTS;
-}
-
 static INLINE uint32_t
 nv50_vbo_type_to_hw(enum pipe_format format)
 {
@@ -139,15 +95,16 @@ nv50_vbo_vtxelt_to_hw(struct pipe_vertex_element *ve)
 	uint32_t hw_type, hw_size;
 	enum pipe_format pf = ve->src_format;
 	const struct util_format_description *desc;
-	unsigned size;
+	unsigned size, nr_components;
 
 	desc = util_format_description(pf);
 	assert(desc);
 
 	size = util_format_get_component_bits(pf, UTIL_FORMAT_COLORSPACE_RGB, 0);
+	nr_components = util_format_get_nr_components(pf);
 
 	hw_type = nv50_vbo_type_to_hw(pf);
-	hw_size = nv50_vbo_size_to_hw(size, ve->nr_components);
+	hw_size = nv50_vbo_size_to_hw(size, nr_components);
 
 	if (!hw_type || !hw_size) {
 		NOUVEAU_ERR("unsupported vbo format: %s\n", util_format_name(pf));
@@ -161,250 +118,58 @@ nv50_vbo_vtxelt_to_hw(struct pipe_vertex_element *ve)
 	return (hw_type | hw_size);
 }
 
-/* For instanced drawing from user buffers, hitting the FIFO repeatedly
- * with the same vertex data is probably worse than uploading all data.
- */
-static boolean
-nv50_upload_vtxbuf(struct nv50_context *nv50, unsigned i)
-{
-	struct nv50_screen *nscreen = nv50->screen;
-	struct pipe_screen *pscreen = &nscreen->base.base;
-	struct pipe_buffer *buf = nscreen->strm_vbuf[i];
-	struct pipe_vertex_buffer *vb = &nv50->vtxbuf[i];
-	uint8_t *src;
-	unsigned size = align(vb->buffer->size, 4096);
-
-	if (buf && buf->size < size)
-		pipe_buffer_reference(&nscreen->strm_vbuf[i], NULL);
-
-	if (!nscreen->strm_vbuf[i]) {
-		nscreen->strm_vbuf[i] = pipe_buffer_create(
-			pscreen, 0, PIPE_BUFFER_USAGE_VERTEX, size);
-		buf = nscreen->strm_vbuf[i];
-	}
-
-	src = pipe_buffer_map(pscreen, vb->buffer, PIPE_BUFFER_USAGE_CPU_READ);
-	if (!src)
-		return FALSE;
-	src += vb->buffer_offset;
-
-	size = (vb->max_index + 1) * vb->stride + 16; /* + 16 is for stride 0 */
-	if (vb->buffer_offset + size > vb->buffer->size)
-		size = vb->buffer->size - vb->buffer_offset;
-
-	pipe_buffer_write(pscreen, buf, vb->buffer_offset, size, src);
-	pipe_buffer_unmap(pscreen, vb->buffer);
-
-	vb->buffer = buf; /* don't pipe_reference, this is a private copy */
-	return TRUE;
-}
-
-static void
-nv50_upload_user_vbufs(struct nv50_context *nv50)
-{
-	unsigned i;
-
-	if (nv50->vbo_fifo)
-		nv50->dirty |= NV50_NEW_ARRAYS;
-	if (!(nv50->dirty & NV50_NEW_ARRAYS))
-		return;
-
-	for (i = 0; i < nv50->vtxbuf_nr; ++i) {
-		if (nv50->vtxbuf[i].buffer->usage & PIPE_BUFFER_USAGE_VERTEX)
-			continue;
-		nv50_upload_vtxbuf(nv50, i);
-	}
-}
-
-static void
-nv50_set_static_vtxattr(struct nv50_context *nv50, unsigned i, void *data)
-{
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nouveau_channel *chan = tesla->channel;
-	float v[4];
-
-	util_format_read_4f(nv50->vtxelt[i].src_format,
-			    v, 0, data, 0, 0, 0, 1, 1);
-
-	switch (nv50->vtxelt[i].nr_components) {
-	case 4:
-		BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_4F_X(i), 4);
-		OUT_RINGf (chan, v[0]);
-		OUT_RINGf (chan, v[1]);
-		OUT_RINGf (chan, v[2]);
-		OUT_RINGf (chan, v[3]);
-		break;
-	case 3:
-		BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_3F_X(i), 3);
-		OUT_RINGf (chan, v[0]);
-		OUT_RINGf (chan, v[1]);
-		OUT_RINGf (chan, v[2]);
-		break;
-	case 2:
-		BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_2F_X(i), 2);
-		OUT_RINGf (chan, v[0]);
-		OUT_RINGf (chan, v[1]);
-		break;
-	case 1:
-		BEGIN_RING(chan, tesla, NV50TCL_VTX_ATTR_1F(i), 1);
-		OUT_RINGf (chan, v[0]);
-		break;
-	default:
-		assert(0);
-		break;
-	}
-}
-
-static unsigned
-init_per_instance_arrays_immd(struct nv50_context *nv50,
-			      unsigned startInstance,
-			      unsigned pos[16], unsigned step[16])
-{
-	struct nouveau_bo *bo;
-	unsigned i, b, count = 0;
-
-	for (i = 0; i < nv50->vtxelt_nr; ++i) {
-		if (!nv50->vtxelt[i].instance_divisor)
-			continue;
-		++count;
-		b = nv50->vtxelt[i].vertex_buffer_index;
-
-		pos[i] = nv50->vtxelt[i].src_offset +
-			nv50->vtxbuf[b].buffer_offset +
-			startInstance * nv50->vtxbuf[b].stride;
-		step[i] = startInstance % nv50->vtxelt[i].instance_divisor;
-
-		bo = nouveau_bo(nv50->vtxbuf[b].buffer);
-		if (!bo->map)
-			nouveau_bo_map(bo, NOUVEAU_BO_RD);
-
-		nv50_set_static_vtxattr(nv50, i, (uint8_t *)bo->map + pos[i]);
-	}
-
-	return count;
-}
-
-static unsigned
-init_per_instance_arrays(struct nv50_context *nv50,
-			 unsigned startInstance,
-			 unsigned pos[16], unsigned step[16])
-{
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nouveau_channel *chan = tesla->channel;
+struct instance {
 	struct nouveau_bo *bo;
-	struct nouveau_stateobj *so;
-	unsigned i, b, count = 0;
-	const uint32_t rl = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
-
-	if (nv50->vbo_fifo)
-		return init_per_instance_arrays_immd(nv50, startInstance,
-						     pos, step);
-
-	so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 2, nv50->vtxelt_nr * 2);
-
-	for (i = 0; i < nv50->vtxelt_nr; ++i) {
-		if (!nv50->vtxelt[i].instance_divisor)
-			continue;
-		++count;
-		b = nv50->vtxelt[i].vertex_buffer_index;
-
-		pos[i] = nv50->vtxelt[i].src_offset +
-			nv50->vtxbuf[b].buffer_offset +
-			startInstance * nv50->vtxbuf[b].stride;
-
-		if (!startInstance) {
-			step[i] = 0;
-			continue;
-		}
-		step[i] = startInstance % nv50->vtxelt[i].instance_divisor;
-
-		bo = nouveau_bo(nv50->vtxbuf[b].buffer);
-
-		so_method(so, tesla, NV50TCL_VERTEX_ARRAY_START_HIGH(i), 2);
-		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_HIGH, 0, 0);
-		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_LOW, 0, 0);
-	}
-
-	if (count && startInstance) {
-		so_ref (so, &nv50->state.instbuf); /* for flush notify */
-		so_emit(chan, nv50->state.instbuf);
-	}
-	so_ref (NULL, &so);
-
-	return count;
-}
+	unsigned delta;
+	unsigned stride;
+	unsigned step;
+	unsigned divisor;
+};
 
 static void
-step_per_instance_arrays_immd(struct nv50_context *nv50,
-			      unsigned pos[16], unsigned step[16])
+instance_init(struct nv50_context *nv50, struct instance *a, unsigned first)
 {
-	struct nouveau_bo *bo;
-	unsigned i, b;
+	int i;
 
-	for (i = 0; i < nv50->vtxelt_nr; ++i) {
-		if (!nv50->vtxelt[i].instance_divisor)
-			continue;
-		if (++step[i] != nv50->vtxelt[i].instance_divisor)
-			continue;
-		b = nv50->vtxelt[i].vertex_buffer_index;
-		bo = nouveau_bo(nv50->vtxbuf[b].buffer);
+	for (i = 0; i < nv50->vtxelt->num_elements; i++) {
+		struct pipe_vertex_element *ve = &nv50->vtxelt->pipe[i];
+		struct pipe_vertex_buffer *vb;
 
-		step[i] = 0;
-		pos[i] += nv50->vtxbuf[b].stride;
+		a[i].divisor = ve->instance_divisor;
+		if (a[i].divisor) {
+			vb = &nv50->vtxbuf[ve->vertex_buffer_index];
 
-		nv50_set_static_vtxattr(nv50, i, (uint8_t *)bo->map + pos[i]);
+			a[i].bo = nouveau_bo(vb->buffer);
+			a[i].stride = vb->stride;
+			a[i].step = first % a[i].divisor;
+			a[i].delta = vb->buffer_offset + ve->src_offset +
+				     (first * a[i].stride);
+		}
 	}
 }
 
 static void
-step_per_instance_arrays(struct nv50_context *nv50,
-			 unsigned pos[16], unsigned step[16])
+instance_step(struct nv50_context *nv50, struct instance *a)
 {
+	struct nouveau_channel *chan = nv50->screen->tesla->channel;
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nouveau_channel *chan = tesla->channel;
-	struct nouveau_bo *bo;
-	struct nouveau_stateobj *so;
-	unsigned i, b;
-	const uint32_t rl = NOUVEAU_BO_VRAM | NOUVEAU_BO_GART | NOUVEAU_BO_RD;
-
-	if (nv50->vbo_fifo) {
-		step_per_instance_arrays_immd(nv50, pos, step);
-		return;
-	}
-
-	so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 2, nv50->vtxelt_nr * 2);
+	int i;
 
-	for (i = 0; i < nv50->vtxelt_nr; ++i) {
-		if (!nv50->vtxelt[i].instance_divisor)
+	for (i = 0; i < nv50->vtxelt->num_elements; i++) {
+		if (!a[i].divisor)
 			continue;
-		b = nv50->vtxelt[i].vertex_buffer_index;
 
-		if (++step[i] == nv50->vtxelt[i].instance_divisor) {
-			step[i] = 0;
-			pos[i] += nv50->vtxbuf[b].stride;
+		BEGIN_RING(chan, tesla,
+			   NV50TCL_VERTEX_ARRAY_START_HIGH(i), 2);
+		OUT_RELOCh(chan, a[i].bo, a[i].delta, NOUVEAU_BO_RD |
+			   NOUVEAU_BO_VRAM | NOUVEAU_BO_GART);
+		OUT_RELOCl(chan, a[i].bo, a[i].delta, NOUVEAU_BO_RD |
+			   NOUVEAU_BO_VRAM | NOUVEAU_BO_GART);
+		if (++a[i].step == a[i].divisor) {
+			a[i].step = 0;
+			a[i].delta += a[i].stride;
 		}
-
-		bo = nouveau_bo(nv50->vtxbuf[b].buffer);
-
-		so_method(so, tesla, NV50TCL_VERTEX_ARRAY_START_HIGH(i), 2);
-		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_HIGH, 0, 0);
-		so_reloc (so, bo, pos[i], rl | NOUVEAU_BO_LOW, 0, 0);
 	}
-
-	so_ref (so, &nv50->state.instbuf); /* for flush notify */
-	so_ref (NULL, &so);
-
-	so_emit(chan, nv50->state.instbuf);
-}
-
-static INLINE void
-nv50_unmap_vbufs(struct nv50_context *nv50)
-{
-        unsigned i;
-
-        for (i = 0; i < nv50->vtxbuf_nr; ++i)
-                if (nouveau_bo(nv50->vtxbuf[i].buffer)->map)
-                        nouveau_bo_unmap(nouveau_bo(nv50->vtxbuf[i].buffer));
 }
 
 void
@@ -415,198 +180,207 @@ nv50_draw_arrays_instanced(struct pipe_context *pipe,
 	struct nv50_context *nv50 = nv50_context(pipe);
 	struct nouveau_channel *chan = nv50->screen->tesla->channel;
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	unsigned i, nz_divisors;
-	unsigned step[16], pos[16];
-
-	if (!NV50_USING_LOATHED_EDGEFLAG(nv50))
-		nv50_upload_user_vbufs(nv50);
+	struct instance a[16];
+	unsigned prim = nv50_prim(mode);
 
-	nv50_state_validate(nv50);
+	instance_init(nv50, a, startInstance);
+	if (!nv50_state_validate(nv50, 10 + 16*3))
+		return;
 
-	nz_divisors = init_per_instance_arrays(nv50, startInstance, pos, step);
+	if (nv50->vbo_fifo) {
+		nv50_push_elements_instanced(pipe, NULL, 0, mode, start,
+					     count, startInstance,
+					     instanceCount);
+		return;
+	}
 
 	BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
 	OUT_RING  (chan, NV50_CB_AUX | (24 << 8));
 	OUT_RING  (chan, startInstance);
+	while (instanceCount--) {
+		if (AVAIL_RING(chan) < (7 + 16*3)) {
+			FIRE_RING(chan);
+			if (!nv50_state_validate(nv50, 7 + 16*3)) {
+				assert(0);
+				return;
+			}
+		}
+		instance_step(nv50, a);
 
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
-	OUT_RING  (chan, nv50_prim(mode));
-
-	if (nv50->vbo_fifo)
-		nv50_push_arrays(nv50, start, count);
-	else {
+		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
+		OUT_RING  (chan, prim);
 		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
 		OUT_RING  (chan, start);
 		OUT_RING  (chan, count);
-	}
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
-	OUT_RING  (chan, 0);
-
-	for (i = 1; i < instanceCount; i++) {
-		if (nz_divisors) /* any non-zero array divisors ? */
-			step_per_instance_arrays(nv50, pos, step);
-
-		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
-		OUT_RING  (chan, nv50_prim(mode) | (1 << 28));
-
-		if (nv50->vbo_fifo)
-			nv50_push_arrays(nv50, start, count);
-		else {
-			BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
-			OUT_RING  (chan, start);
-			OUT_RING  (chan, count);
-		}
 		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
 		OUT_RING  (chan, 0);
-	}
-	nv50_unmap_vbufs(nv50);
 
-	so_ref(NULL, &nv50->state.instbuf);
+		prim |= (1 << 28);
+	}
 }
 
 void
 nv50_draw_arrays(struct pipe_context *pipe, unsigned mode, unsigned start,
 		 unsigned count)
 {
-	struct nv50_context *nv50 = nv50_context(pipe);
-	struct nouveau_channel *chan = nv50->screen->tesla->channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	boolean ret;
-
-	nv50_state_validate(nv50);
-
-	BEGIN_RING(chan, tesla, 0x142c, 1);
-	OUT_RING  (chan, 0);
-	BEGIN_RING(chan, tesla, 0x142c, 1);
-	OUT_RING  (chan, 0);
-
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
-	OUT_RING  (chan, nv50_prim(mode));
-
-	if (nv50->vbo_fifo)
-		ret = nv50_push_arrays(nv50, start, count);
-	else {
-		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BUFFER_FIRST, 2);
-		OUT_RING  (chan, start);
-		OUT_RING  (chan, count);
-		ret = TRUE;
-	}
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
-	OUT_RING  (chan, 0);
-
-	nv50_unmap_vbufs(nv50);
-
-        /* XXX: not sure what to do if ret != TRUE: flush and retry?
-         */
-        assert(ret);
+	nv50_draw_arrays_instanced(pipe, mode, start, count, 0, 1);
 }
 
-static INLINE boolean
-nv50_draw_elements_inline_u08(struct nv50_context *nv50, uint8_t *map,
-			      unsigned start, unsigned count)
-{
-	struct nouveau_channel *chan = nv50->screen->tesla->channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-
-	map += start;
+struct inline_ctx {
+	struct nv50_context *nv50;
+	void *map;
+};
 
-	if (nv50->vbo_fifo)
-		return nv50_push_elements_u08(nv50, map, count);
+static void
+inline_elt08(void *priv, unsigned start, unsigned count)
+{
+	struct inline_ctx *ctx = priv;
+	struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+	struct nouveau_channel *chan = tesla->channel;
+	uint8_t *map = (uint8_t *)ctx->map + start;
 
 	if (count & 1) {
 		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32, 1);
 		OUT_RING  (chan, map[0]);
 		map++;
-		count--;
+		count &= ~1;
 	}
 
-	while (count) {
-		unsigned nr = count > 2046 ? 2046 : count;
-		int i;
-
-		BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, nr >> 1);
-		for (i = 0; i < nr; i += 2)
-			OUT_RING  (chan, (map[i + 1] << 16) | map[i]);
+	count >>= 1;
+	if (!count)
+		return;
 
-		count -= nr;
-		map += nr;
+	BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, count);
+	while (count--) {
+		OUT_RING(chan, (map[1] << 16) | map[0]);
+		map += 2;
 	}
-	return TRUE;
 }
 
-static INLINE boolean
-nv50_draw_elements_inline_u16(struct nv50_context *nv50, uint16_t *map,
-			      unsigned start, unsigned count)
+static void
+inline_elt16(void *priv, unsigned start, unsigned count)
 {
-	struct nouveau_channel *chan = nv50->screen->tesla->channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-
-	map += start;
-
-	if (nv50->vbo_fifo)
-		return nv50_push_elements_u16(nv50, map, count);
+	struct inline_ctx *ctx = priv;
+	struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+	struct nouveau_channel *chan = tesla->channel;
+	uint16_t *map = (uint16_t *)ctx->map + start;
 
 	if (count & 1) {
 		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32, 1);
 		OUT_RING  (chan, map[0]);
+		count &= ~1;
 		map++;
-		count--;
 	}
 
-	while (count) {
-		unsigned nr = count > 2046 ? 2046 : count;
-		int i;
-
-		BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, nr >> 1);
-		for (i = 0; i < nr; i += 2)
-			OUT_RING  (chan, (map[i + 1] << 16) | map[i]);
+	count >>= 1;
+	if (!count)
+		return;
 
-		count -= nr;
-		map += nr;
+	BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U16, count);
+	while (count--) {
+		OUT_RING(chan, (map[1] << 16) | map[0]);
+		map += 2;
 	}
-	return TRUE;
 }
 
-static INLINE boolean
-nv50_draw_elements_inline_u32(struct nv50_context *nv50, uint32_t *map,
-			      unsigned start, unsigned count)
+static void
+inline_elt32(void *priv, unsigned start, unsigned count)
+{
+	struct inline_ctx *ctx = priv;
+	struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+	struct nouveau_channel *chan = tesla->channel;
+
+	BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U32, count);
+	OUT_RINGp    (chan, (uint32_t *)ctx->map + start, count);
+}
+
+static void
+inline_edgeflag(void *priv, boolean enabled)
 {
+	struct inline_ctx *ctx = priv;
+	struct nouveau_grobj *tesla = ctx->nv50->screen->tesla;
+	struct nouveau_channel *chan = tesla->channel;
+
+	BEGIN_RING(chan, tesla, NV50TCL_EDGEFLAG_ENABLE, 1);
+	OUT_RING  (chan, enabled ? 1 : 0);
+}
+
+static void
+nv50_draw_elements_inline(struct pipe_context *pipe,
+			  struct pipe_buffer *indexBuffer, unsigned indexSize,
+			  unsigned mode, unsigned start, unsigned count,
+			  unsigned startInstance, unsigned instanceCount)
+{
+	struct pipe_screen *pscreen = pipe->screen;
+	struct nv50_context *nv50 = nv50_context(pipe);
 	struct nouveau_channel *chan = nv50->screen->tesla->channel;
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
+	struct instance a[16];
+	struct inline_ctx ctx;
+	struct u_split_prim s;
+	boolean nzi = FALSE;
+	unsigned overhead;
+
+	overhead = 16*3; /* potential instance adjustments */
+	overhead += 4; /* Begin()/End() */
+	overhead += 4; /* potential edgeflag disable/reenable */
+	overhead += 3; /* potentially 3 VTX_ELT_U16/U32 packet headers */
+
+	s.priv = &ctx;
+	if (indexSize == 1)
+		s.emit = inline_elt08;
+	else
+	if (indexSize == 2)
+		s.emit = inline_elt16;
+	else
+		s.emit = inline_elt32;
+	s.edge = inline_edgeflag;
+
+	ctx.nv50 = nv50;
+	ctx.map = pipe_buffer_map(pscreen, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ);
+	assert(ctx.map);
+	if (!ctx.map)
+		return;
 
-	map += start;
+	instance_init(nv50, a, startInstance);
+	if (!nv50_state_validate(nv50, overhead + 6 + 3))
+		return;
 
-	if (nv50->vbo_fifo)
-		return nv50_push_elements_u32(nv50, map, count);
+	BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
+	OUT_RING  (chan, NV50_CB_AUX | (24 << 8));
+	OUT_RING  (chan, startInstance);
+	while (instanceCount--) {
+		unsigned max_verts;
+		boolean done;
+
+		u_split_prim_init(&s, mode, start, count);
+		do {
+			if (AVAIL_RING(chan) < (overhead + 6)) {
+				FIRE_RING(chan);
+				if (!nv50_state_validate(nv50, (overhead + 6))) {
+					assert(0);
+					return;
+				}
+			}
 
-	while (count) {
-		unsigned nr = count > 2047 ? 2047 : count;
+			max_verts = AVAIL_RING(chan) - overhead;
+			if (max_verts > 2047)
+				max_verts = 2047;
+			if (indexSize != 4)
+				max_verts <<= 1;
+			instance_step(nv50, a);
 
-		BEGIN_RING_NI(chan, tesla, NV50TCL_VB_ELEMENT_U32, nr);
-		OUT_RINGp (chan, map, nr);
+			BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
+			OUT_RING  (chan, nv50_prim(s.mode) | (nzi ? (1<<28) : 0));
+			done = u_split_prim_next(&s, max_verts);
+			BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
+			OUT_RING  (chan, 0);
+		} while (!done);
 
-		count -= nr;
-		map += nr;
+		nzi = TRUE;
 	}
-	return TRUE;
-}
 
-static INLINE void
-nv50_draw_elements_inline(struct nv50_context *nv50,
-			  void *map, unsigned indexSize,
-			  unsigned start, unsigned count)
-{
-	switch (indexSize) {
-	case 1:
-		nv50_draw_elements_inline_u08(nv50, map, start, count);
-		break;
-	case 2:
-		nv50_draw_elements_inline_u16(nv50, map, start, count);
-		break;
-	case 4:
-		nv50_draw_elements_inline_u32(nv50, map, start, count);
-		break;
-	}
+	pipe_buffer_unmap(pscreen, indexBuffer);
 }
 
 void
@@ -617,49 +391,68 @@ nv50_draw_elements_instanced(struct pipe_context *pipe,
 			     unsigned startInstance, unsigned instanceCount)
 {
 	struct nv50_context *nv50 = nv50_context(pipe);
+	struct nouveau_channel *chan = nv50->screen->tesla->channel;
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nouveau_channel *chan = tesla->channel;
-	struct pipe_screen *pscreen = pipe->screen;
-	void *map;
-	unsigned i, nz_divisors;
-	unsigned step[16], pos[16];
+	struct instance a[16];
+	unsigned prim = nv50_prim(mode);
 
-	map = pipe_buffer_map(pscreen, indexBuffer, PIPE_BUFFER_USAGE_CPU_READ);
-
-	if (!NV50_USING_LOATHED_EDGEFLAG(nv50))
-		nv50_upload_user_vbufs(nv50);
-
-	nv50_state_validate(nv50);
+	instance_init(nv50, a, startInstance);
+	if (!nv50_state_validate(nv50, 13 + 16*3))
+		return;
 
-	nz_divisors = init_per_instance_arrays(nv50, startInstance, pos, step);
+	if (nv50->vbo_fifo) {
+		nv50_push_elements_instanced(pipe, indexBuffer, indexSize,
+					     mode, start, count, startInstance,
+					     instanceCount);
+		return;
+	} else
+	if (!(indexBuffer->usage & PIPE_BUFFER_USAGE_INDEX) || indexSize == 1) {
+		nv50_draw_elements_inline(pipe, indexBuffer, indexSize,
+					  mode, start, count, startInstance,
+					  instanceCount);
+		return;
+	}
 
 	BEGIN_RING(chan, tesla, NV50TCL_CB_ADDR, 2);
 	OUT_RING  (chan, NV50_CB_AUX | (24 << 8));
 	OUT_RING  (chan, startInstance);
-
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
-	OUT_RING  (chan, nv50_prim(mode));
-
-	nv50_draw_elements_inline(nv50, map, indexSize, start, count);
-
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
-	OUT_RING  (chan, 0);
-
-	for (i = 1; i < instanceCount; ++i) {
-		if (nz_divisors) /* any non-zero array divisors ? */
-			step_per_instance_arrays(nv50, pos, step);
+	while (instanceCount--) {
+		if (AVAIL_RING(chan) < (7 + 16*3)) {
+			FIRE_RING(chan);
+			if (!nv50_state_validate(nv50, 10 + 16*3)) {
+				assert(0);
+				return;
+			}
+		}
+		instance_step(nv50, a);
 
 		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
-		OUT_RING  (chan, nv50_prim(mode) | (1 << 28));
-
-		nv50_draw_elements_inline(nv50, map, indexSize, start, count);
-
+		OUT_RING  (chan, prim);
+		if (indexSize == 4) {
+			BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32 | 0x30000, 0);
+			OUT_RING  (chan, count);
+			nouveau_pushbuf_submit(chan, nouveau_bo(indexBuffer),
+					       start << 2, count << 2);
+		} else
+		if (indexSize == 2) {
+			unsigned vb_start = (start & ~1);
+			unsigned vb_end = (start + count + 1) & ~1;
+			unsigned dwords = (vb_end - vb_start) >> 1;
+
+			BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16_SETUP, 1);
+			OUT_RING  (chan, ((start & 1) << 31) | count);
+			BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16 | 0x30000, 0);
+			OUT_RING  (chan, dwords);
+			nouveau_pushbuf_submit(chan, nouveau_bo(indexBuffer),
+					       vb_start << 1, dwords << 2);
+			BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16_SETUP, 1);
+			OUT_RING  (chan, 0);
+		}
 		BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
 		OUT_RING  (chan, 0);
-	}
-	nv50_unmap_vbufs(nv50);
 
-	so_ref(NULL, &nv50->state.instbuf);
+		prim |= (1 << 28);
+	}
 }
 
 void
@@ -667,51 +460,8 @@ nv50_draw_elements(struct pipe_context *pipe,
 		   struct pipe_buffer *indexBuffer, unsigned indexSize,
 		   unsigned mode, unsigned start, unsigned count)
 {
-	struct nv50_context *nv50 = nv50_context(pipe);
-	struct nouveau_channel *chan = nv50->screen->tesla->channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct pipe_screen *pscreen = pipe->screen;
-	void *map;
-	
-	nv50_state_validate(nv50);
-
-	BEGIN_RING(chan, tesla, 0x142c, 1);
-	OUT_RING  (chan, 0);
-	BEGIN_RING(chan, tesla, 0x142c, 1);
-	OUT_RING  (chan, 0);
-
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_BEGIN, 1);
-	OUT_RING  (chan, nv50_prim(mode));
-
-	if (!nv50->vbo_fifo && indexSize == 4) {
-		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U32 | 0x30000, 0);
-		OUT_RING  (chan, count);
-		nouveau_pushbuf_submit(chan, nouveau_bo(indexBuffer),
-				       start << 2, count << 2);
-	} else
-	if (!nv50->vbo_fifo && indexSize == 2) {
-		unsigned vb_start = (start & ~1);
-		unsigned vb_end = (start + count + 1) & ~1;
-		unsigned dwords = (vb_end - vb_start) >> 1;
-
-		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16_SETUP, 1);
-		OUT_RING  (chan, ((start & 1) << 31) | count);
-		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16 | 0x30000, 0);
-		OUT_RING  (chan, dwords);
-		nouveau_pushbuf_submit(chan, nouveau_bo(indexBuffer),
-				       vb_start << 1, dwords << 2);
-		BEGIN_RING(chan, tesla, NV50TCL_VB_ELEMENT_U16_SETUP, 1);
-		OUT_RING  (chan, 0);
-	} else {
-		map = pipe_buffer_map(pscreen, indexBuffer,
-				      PIPE_BUFFER_USAGE_CPU_READ);
-		nv50_draw_elements_inline(nv50, map, indexSize, start, count);
-		nv50_unmap_vbufs(nv50);
-		pipe_buffer_unmap(pscreen, indexBuffer);
-	}
-
-	BEGIN_RING(chan, tesla, NV50TCL_VERTEX_END, 1);
-	OUT_RING  (chan, 0);
+	nv50_draw_elements_instanced(pipe, indexBuffer, indexSize,
+				     mode, start, count, 0, 1);
 }
 
 static INLINE boolean
@@ -726,6 +476,7 @@ nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib,
 	struct nouveau_bo *bo = nouveau_bo(vb->buffer);
 	float v[4];
 	int ret;
+	unsigned nr_components = util_format_get_nr_components(ve->src_format);
 
 	ret = nouveau_bo_map(bo, NOUVEAU_BO_RD);
 	if (ret)
@@ -736,9 +487,10 @@ nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib,
 			    0, 0, 1, 1);
 	so = *pso;
 	if (!so)
-		*pso = so = so_new(nv50->vtxelt_nr, nv50->vtxelt_nr * 4, 0);
+		*pso = so = so_new(nv50->vtxelt->num_elements,
+				   nv50->vtxelt->num_elements * 4, 0);
 
-	switch (ve->nr_components) {
+	switch (nr_components) {
 	case 4:
 		so_method(so, tesla, NV50TCL_VTX_ATTR_4F_X(attrib), 4);
 		so_data  (so, fui(v[0]));
@@ -775,6 +527,18 @@ nv50_vbo_static_attrib(struct nv50_context *nv50, unsigned attrib,
 }
 
 void
+nv50_vtxelt_construct(struct nv50_vtxelt_stateobj *cso)
+{
+	unsigned i;
+
+	for (i = 0; i < cso->num_elements; ++i) {
+		struct pipe_vertex_element *ve = &cso->pipe[i];
+
+		cso->hw[i] = nv50_vbo_vtxelt_to_hw(ve);
+	}
+}
+
+struct nouveau_stateobj *
 nv50_vbo_validate(struct nv50_context *nv50)
 {
 	struct nouveau_grobj *tesla = nv50->screen->tesla;
@@ -783,30 +547,31 @@ nv50_vbo_validate(struct nv50_context *nv50)
 
 	/* don't validate if Gallium took away our buffers */
 	if (nv50->vtxbuf_nr == 0)
-		return;
-	nv50->vbo_fifo = 0;
+		return NULL;
+
+	if (nv50->screen->force_push ||
+	    nv50->vertprog->cfg.edgeflag_in < 16)
+		nv50->vbo_fifo = 0xffff;
 
-	for (i = 0; i < nv50->vtxbuf_nr; ++i)
+	for (i = 0; i < nv50->vtxbuf_nr; i++) {
 		if (nv50->vtxbuf[i].stride &&
 		    !(nv50->vtxbuf[i].buffer->usage & PIPE_BUFFER_USAGE_VERTEX))
 			nv50->vbo_fifo = 0xffff;
+	}
 
-	if (NV50_USING_LOATHED_EDGEFLAG(nv50))
-		nv50->vbo_fifo = 0xffff; /* vertprog can't set edgeflag */
-
-	n_ve = MAX2(nv50->vtxelt_nr, nv50->state.vtxelt_nr);
+	n_ve = MAX2(nv50->vtxelt->num_elements, nv50->state.vtxelt_nr);
 
 	vtxattr = NULL;
-	vtxbuf = so_new(n_ve * 2, n_ve * 5, nv50->vtxelt_nr * 4);
+	vtxbuf = so_new(n_ve * 2, n_ve * 5, nv50->vtxelt->num_elements * 4);
 	vtxfmt = so_new(1, n_ve, 0);
 	so_method(vtxfmt, tesla, NV50TCL_VERTEX_ARRAY_ATTRIB(0), n_ve);
 
-	for (i = 0; i < nv50->vtxelt_nr; i++) {
-		struct pipe_vertex_element *ve = &nv50->vtxelt[i];
+	for (i = 0; i < nv50->vtxelt->num_elements; i++) {
+		struct pipe_vertex_element *ve = &nv50->vtxelt->pipe[i];
 		struct pipe_vertex_buffer *vb =
 			&nv50->vtxbuf[ve->vertex_buffer_index];
 		struct nouveau_bo *bo = nouveau_bo(vb->buffer);
-		uint32_t hw = nv50_vbo_vtxelt_to_hw(ve);
+		uint32_t hw = nv50->vtxelt->hw[i];
 
 		if (!vb->stride &&
 		    nv50_vbo_static_attrib(nv50, i, &vtxattr, ve, vb)) {
@@ -821,13 +586,13 @@ nv50_vbo_validate(struct nv50_context *nv50)
 		}
 
 		if (nv50->vbo_fifo) {
-			so_data  (vtxfmt, hw |
-				  (ve->instance_divisor ? (1 << 4) : i));
+			so_data  (vtxfmt, hw | (ve->instance_divisor ? (1 << 4) : i));
 			so_method(vtxbuf, tesla,
 				  NV50TCL_VERTEX_ARRAY_FORMAT(i), 1);
 			so_data  (vtxbuf, 0);
 			continue;
 		}
+
 		so_data(vtxfmt, hw | i);
 
 		so_method(vtxbuf, tesla, NV50TCL_VERTEX_ARRAY_FORMAT(i), 3);
@@ -855,355 +620,13 @@ nv50_vbo_validate(struct nv50_context *nv50)
 		so_method(vtxbuf, tesla, NV50TCL_VERTEX_ARRAY_FORMAT(i), 1);
 		so_data  (vtxbuf, 0);
 	}
-	nv50->state.vtxelt_nr = nv50->vtxelt_nr;
+	nv50->state.vtxelt_nr = nv50->vtxelt->num_elements;
 
-	so_ref (vtxfmt, &nv50->state.vtxfmt);
 	so_ref (vtxbuf, &nv50->state.vtxbuf);
 	so_ref (vtxattr, &nv50->state.vtxattr);
 	so_ref (NULL, &vtxbuf);
-	so_ref (NULL, &vtxfmt);
 	so_ref (NULL, &vtxattr);
+	return vtxfmt;
 }
 
-typedef void (*pfn_push)(struct nouveau_channel *, void *);
-
-struct nv50_vbo_emitctx
-{
-	pfn_push push[16];
-	uint8_t *map[16];
-	unsigned stride[16];
-	unsigned nr_ve;
-	unsigned vtx_dwords;
-	unsigned vtx_max;
-
-	float edgeflag;
-	unsigned ve_edgeflag;
-};
-
-static INLINE void
-emit_vtx_next(struct nouveau_channel *chan, struct nv50_vbo_emitctx *emit)
-{
-	unsigned i;
-
-	for (i = 0; i < emit->nr_ve; ++i) {
-		emit->push[i](chan, emit->map[i]);
-		emit->map[i] += emit->stride[i];
-	}
-}
-
-static INLINE void
-emit_vtx(struct nouveau_channel *chan, struct nv50_vbo_emitctx *emit,
-	 uint32_t vi)
-{
-	unsigned i;
-
-	for (i = 0; i < emit->nr_ve; ++i)
-		emit->push[i](chan, emit->map[i] + emit->stride[i] * vi);
-}
-
-static INLINE boolean
-nv50_map_vbufs(struct nv50_context *nv50)
-{
-	int i;
-
-	for (i = 0; i < nv50->vtxbuf_nr; ++i) {
-		struct pipe_vertex_buffer *vb = &nv50->vtxbuf[i];
-		unsigned size = vb->stride * (vb->max_index + 1) + 16;
-
-		if (nouveau_bo(vb->buffer)->map)
-			continue;
-
-		size = vb->stride * (vb->max_index + 1) + 16;
-		size = MIN2(size, vb->buffer->size);
-		if (!size)
-			size = vb->buffer->size;
-
-		if (nouveau_bo_map_range(nouveau_bo(vb->buffer),
-					 0, size, NOUVEAU_BO_RD))
-			break;
-	}
-
-	if (i == nv50->vtxbuf_nr)
-		return TRUE;
-	for (; i >= 0; --i)
-		nouveau_bo_unmap(nouveau_bo(nv50->vtxbuf[i].buffer));
-	return FALSE;
-}
-
-static void
-emit_b32_1(struct nouveau_channel *chan, void *data)
-{
-	uint32_t *v = data;
-
-	OUT_RING(chan, v[0]);
-}
-
-static void
-emit_b32_2(struct nouveau_channel *chan, void *data)
-{
-	uint32_t *v = data;
-
-	OUT_RING(chan, v[0]);
-	OUT_RING(chan, v[1]);
-}
-
-static void
-emit_b32_3(struct nouveau_channel *chan, void *data)
-{
-	uint32_t *v = data;
-
-	OUT_RING(chan, v[0]);
-	OUT_RING(chan, v[1]);
-	OUT_RING(chan, v[2]);
-}
-
-static void
-emit_b32_4(struct nouveau_channel *chan, void *data)
-{
-	uint32_t *v = data;
-
-	OUT_RING(chan, v[0]);
-	OUT_RING(chan, v[1]);
-	OUT_RING(chan, v[2]);
-	OUT_RING(chan, v[3]);
-}
-
-static void
-emit_b16_1(struct nouveau_channel *chan, void *data)
-{
-	uint16_t *v = data;
-
-	OUT_RING(chan, v[0]);
-}
-
-static void
-emit_b16_3(struct nouveau_channel *chan, void *data)
-{
-	uint16_t *v = data;
-
-	OUT_RING(chan, (v[1] << 16) | v[0]);
-	OUT_RING(chan, v[2]);
-}
-
-static void
-emit_b08_1(struct nouveau_channel *chan, void *data)
-{
-	uint8_t *v = data;
-
-	OUT_RING(chan, v[0]);
-}
-
-static void
-emit_b08_3(struct nouveau_channel *chan, void *data)
-{
-	uint8_t *v = data;
-
-	OUT_RING(chan, (v[2] << 16) | (v[1] << 8) | v[0]);
-}
-
-static boolean
-emit_prepare(struct nv50_context *nv50, struct nv50_vbo_emitctx *emit,
-	     unsigned start)
-{
-	unsigned i;
-
-	if (nv50_map_vbufs(nv50) == FALSE)
-		return FALSE;
-
-	emit->ve_edgeflag = nv50->vertprog->cfg.edgeflag_in;
-
-	emit->edgeflag = 0.5f;
-	emit->nr_ve = 0;
-	emit->vtx_dwords = 0;
-
-	for (i = 0; i < nv50->vtxelt_nr; ++i) {
-		struct pipe_vertex_element *ve;
-		struct pipe_vertex_buffer *vb;
-		unsigned n, size;
-		const struct util_format_description *desc;
-
-		ve = &nv50->vtxelt[i];
-		vb = &nv50->vtxbuf[ve->vertex_buffer_index];
-		if (!(nv50->vbo_fifo & (1 << i)) || ve->instance_divisor)
-			continue;
-		n = emit->nr_ve++;
-
-		emit->stride[n] = vb->stride;
-		emit->map[n] = (uint8_t *)nouveau_bo(vb->buffer)->map +
-			vb->buffer_offset +
-			(start * vb->stride + ve->src_offset);
-
-		desc = util_format_description(ve->src_format);
-		assert(desc);
-
-		size = util_format_get_component_bits(
-			ve->src_format, UTIL_FORMAT_COLORSPACE_RGB, 0);
-
-		assert(ve->nr_components > 0 && ve->nr_components <= 4);
-
-		/* It shouldn't be necessary to push the implicit 1s
-		 * for case 3 and size 8 cases 1, 2, 3.
-		 */
-		switch (size) {
-		default:
-			NOUVEAU_ERR("unsupported vtxelt size: %u\n", size);
-			return FALSE;
-		case 32:
-			switch (ve->nr_components) {
-			case 1: emit->push[n] = emit_b32_1; break;
-			case 2: emit->push[n] = emit_b32_2; break;
-			case 3: emit->push[n] = emit_b32_3; break;
-			case 4: emit->push[n] = emit_b32_4; break;
-			}
-			emit->vtx_dwords += ve->nr_components;
-			break;
-		case 16:
-			switch (ve->nr_components) {
-			case 1: emit->push[n] = emit_b16_1; break;
-			case 2: emit->push[n] = emit_b32_1; break;
-			case 3: emit->push[n] = emit_b16_3; break;
-			case 4: emit->push[n] = emit_b32_2; break;
-			}
-			emit->vtx_dwords += (ve->nr_components + 1) >> 1;
-			break;
-		case 8:
-			switch (ve->nr_components) {
-			case 1: emit->push[n] = emit_b08_1; break;
-			case 2: emit->push[n] = emit_b16_1; break;
-			case 3: emit->push[n] = emit_b08_3; break;
-			case 4: emit->push[n] = emit_b32_1; break;
-			}
-			emit->vtx_dwords += 1;
-			break;
-		}
-	}
-
-	emit->vtx_max = 512 / emit->vtx_dwords;
-	if (emit->ve_edgeflag < 16)
-		emit->vtx_max = 1;
-
-	return TRUE;
-}
-
-static INLINE void
-set_edgeflag(struct nouveau_channel *chan,
-	     struct nouveau_grobj *tesla,
-	     struct nv50_vbo_emitctx *emit, uint32_t index)
-{
-	unsigned i = emit->ve_edgeflag;
-
-	if (i < 16) {
-		float f = *((float *)(emit->map[i] + index * emit->stride[i]));
-
-		if (emit->edgeflag != f) {
-			emit->edgeflag = f;
-
-			BEGIN_RING(chan, tesla, 0x15e4, 1);
-			OUT_RING  (chan, f ? 1 : 0);
-		}
-	}
-}
-
-static boolean
-nv50_push_arrays(struct nv50_context *nv50, unsigned start, unsigned count)
-{
-	struct nouveau_channel *chan = nv50->screen->base.channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nv50_vbo_emitctx emit;
 
-	if (emit_prepare(nv50, &emit, start) == FALSE)
-		return FALSE;
-
-	while (count) {
-		unsigned i, dw, nr = MIN2(count, emit.vtx_max);
-	        dw = nr * emit.vtx_dwords;
-
-		set_edgeflag(chan, tesla, &emit, 0); /* nr will be 1 */
-
-		BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
-		for (i = 0; i < nr; ++i)
-			emit_vtx_next(chan, &emit);
-
-		count -= nr;
-	}
-
-	return TRUE;
-}
-
-static boolean
-nv50_push_elements_u32(struct nv50_context *nv50, uint32_t *map, unsigned count)
-{
-	struct nouveau_channel *chan = nv50->screen->base.channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nv50_vbo_emitctx emit;
-
-	if (emit_prepare(nv50, &emit, 0) == FALSE)
-		return FALSE;
-
-	while (count) {
-		unsigned i, dw, nr = MIN2(count, emit.vtx_max);
-	        dw = nr * emit.vtx_dwords;
-
-		set_edgeflag(chan, tesla, &emit, *map);
-
-		BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
-		for (i = 0; i < nr; ++i)
-			emit_vtx(chan, &emit, *map++);
-
-		count -= nr;
-	}
-
-	return TRUE;
-}
-
-static boolean
-nv50_push_elements_u16(struct nv50_context *nv50, uint16_t *map, unsigned count)
-{
-	struct nouveau_channel *chan = nv50->screen->base.channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nv50_vbo_emitctx emit;
-
-	if (emit_prepare(nv50, &emit, 0) == FALSE)
-		return FALSE;
-
-	while (count) {
-		unsigned i, dw, nr = MIN2(count, emit.vtx_max);
-	        dw = nr * emit.vtx_dwords;
-
-		set_edgeflag(chan, tesla, &emit, *map);
-
-		BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
-		for (i = 0; i < nr; ++i)
-			emit_vtx(chan, &emit, *map++);
-
-		count -= nr;
-	}
-
-	return TRUE;
-}
-
-static boolean
-nv50_push_elements_u08(struct nv50_context *nv50, uint8_t *map, unsigned count)
-{
-	struct nouveau_channel *chan = nv50->screen->base.channel;
-	struct nouveau_grobj *tesla = nv50->screen->tesla;
-	struct nv50_vbo_emitctx emit;
-
-	if (emit_prepare(nv50, &emit, 0) == FALSE)
-		return FALSE;
-
-	while (count) {
-		unsigned i, dw, nr = MIN2(count, emit.vtx_max);
-	        dw = nr * emit.vtx_dwords;
-
-		set_edgeflag(chan, tesla, &emit, *map);
-
-		BEGIN_RING_NI(chan, tesla, NV50TCL_VERTEX_DATA, dw);
-		for (i = 0; i < nr; ++i)
-			emit_vtx(chan, &emit, *map++);
-
-		count -= nr;
-	}
-
-	return TRUE;
-}
diff --git a/src/gallium/drivers/r300/Makefile b/src/gallium/drivers/r300/Makefile
index afddcb161fa..61b54af4ddf 100644
--- a/src/gallium/drivers/r300/Makefile
+++ b/src/gallium/drivers/r300/Makefile
@@ -19,7 +19,8 @@ C_SOURCES = \
 	r300_state_invariant.c \
 	r300_vs.c \
 	r300_texture.c \
-	r300_tgsi_to_rc.c
+	r300_tgsi_to_rc.c \
+	r300_transfer.c
 
 LIBRARY_INCLUDES = \
 	-I$(TOP)/src/mesa/drivers/dri/r300/compiler \
@@ -32,7 +33,5 @@ EXTRA_OBJECTS = \
 
 include ../../Makefile.template
 
-.PHONY : $(COMPILER_ARCHIVE)
-
 $(COMPILER_ARCHIVE):
 	$(MAKE) -C $(TOP)/src/mesa/drivers/dri/r300/compiler
diff --git a/src/gallium/drivers/r300/SConscript b/src/gallium/drivers/r300/SConscript
index 183aa17f9b3..27b2e309932 100644
--- a/src/gallium/drivers/r300/SConscript
+++ b/src/gallium/drivers/r300/SConscript
@@ -30,6 +30,7 @@ r300 = env.ConvenienceLibrary(
         'r300_vs.c',
         'r300_texture.c',
         'r300_tgsi_to_rc.c',
+        'r300_transfer.c',
     ] + r300compiler) + r300compiler
 
 Export('r300')
diff --git a/src/gallium/drivers/r300/r300_blit.c b/src/gallium/drivers/r300/r300_blit.c
index c48684fe51f..a60b12844d6 100644
--- a/src/gallium/drivers/r300/r300_blit.c
+++ b/src/gallium/drivers/r300/r300_blit.c
@@ -33,7 +33,10 @@ static void r300_blitter_save_states(struct r300_context* r300)
     util_blitter_save_stencil_ref(r300->blitter, &(r300->stencil_ref));
     util_blitter_save_rasterizer(r300->blitter, r300->rs_state.state);
     util_blitter_save_fragment_shader(r300->blitter, r300->fs);
-    util_blitter_save_vertex_shader(r300->blitter, r300->vs);
+    util_blitter_save_vertex_shader(r300->blitter, r300->vs_state.state);
+    util_blitter_save_viewport(r300->blitter, &r300->viewport);
+    util_blitter_save_clip(r300->blitter, &r300->clip);
+    util_blitter_save_vertex_elements(r300->blitter, r300->velems);
 }
 
 /* Clear currently bound buffers. */
@@ -98,6 +101,8 @@ static void r300_hw_copy(struct pipe_context* pipe,
                          unsigned width, unsigned height)
 {
     struct r300_context* r300 = r300_context(pipe);
+    struct r300_textures_state* state =
+        (struct r300_textures_state*)r300->textures_state.state;
 
     /* Yeah we have to save all those states to ensure this blitter operation
      * is really transparent. The states will be restored by the blitter once
@@ -106,7 +111,7 @@ static void r300_hw_copy(struct pipe_context* pipe,
     util_blitter_save_framebuffer(r300->blitter, r300->fb_state.state);
 
     util_blitter_save_fragment_sampler_states(
-        r300->blitter, r300->sampler_count, (void**)r300->sampler_states);
+        r300->blitter, state->sampler_count, (void**)state->sampler_states);
 
     util_blitter_save_fragment_sampler_views(
         r300->blitter, r300->fragment_sampler_view_count,
@@ -139,10 +144,10 @@ void r300_surface_copy(struct pipe_context* pipe,
                 new_format = PIPE_FORMAT_I8_UNORM;
                 break;
             case 2:
-                new_format = PIPE_FORMAT_A4R4G4B4_UNORM;
+                new_format = PIPE_FORMAT_B4G4R4A4_UNORM;
                 break;
             case 4:
-                new_format = PIPE_FORMAT_A8R8G8B8_UNORM;
+                new_format = PIPE_FORMAT_B8G8R8A8_UNORM;
                 break;
             default:
                 debug_printf("r300: surface_copy: Unhandled format: %s. Falling back to software.\n"
diff --git a/src/gallium/drivers/r300/r300_context.c b/src/gallium/drivers/r300/r300_context.c
index f631b4ed27f..923e1e541ff 100644
--- a/src/gallium/drivers/r300/r300_context.c
+++ b/src/gallium/drivers/r300/r300_context.c
@@ -59,7 +59,8 @@ static void r300_destroy_context(struct pipe_context* context)
     FREE(r300->fb_state.state);
     FREE(r300->rs_block_state.state);
     FREE(r300->scissor_state.state);
-    FREE(r300->vertex_format_state.state);
+    FREE(r300->textures_state.state);
+    FREE(r300->vap_output_state.state);
     FREE(r300->viewport_state.state);
     FREE(r300->ztop_state.state);
     FREE(r300);
@@ -70,11 +71,8 @@ r300_is_texture_referenced(struct pipe_context *pipe,
                            struct pipe_texture *texture,
                            unsigned face, unsigned level)
 {
-    struct pipe_buffer* buf = 0;
-
-    r300_get_texture_buffer(pipe->screen, texture, &buf, NULL);
-
-    return pipe->is_buffer_referenced(pipe, buf);
+    return pipe->is_buffer_referenced(pipe,
+                                      ((struct r300_texture *)texture)->buffer);
 }
 
 static unsigned int
@@ -84,7 +82,14 @@ r300_is_buffer_referenced(struct pipe_context *pipe,
     /* This only checks to see whether actual hardware buffers are
      * referenced. Since we use managed BOs and transfers, it's actually not
      * possible for pipe_buffers to ever reference the actual hardware, so
-     * buffers are never referenced. */
+     * buffers are never referenced. 
+     */
+
+    /* XXX: that doesn't make sense given that
+     * r300_is_texture_referenced is implemented on top of this
+     * function and hardware can certainly refer to textures
+     * directly...
+     */
     return 0;
 }
 
@@ -96,39 +101,54 @@ static void r300_flush_cb(void *data)
 }
 
 #define R300_INIT_ATOM(atomname, atomsize) \
-    r300->atomname##_state.name = #atomname; \
-    r300->atomname##_state.state = NULL; \
-    r300->atomname##_state.size = atomsize; \
-    r300->atomname##_state.emit = r300_emit_##atomname##_state; \
-    r300->atomname##_state.dirty = FALSE; \
-    insert_at_tail(&r300->atom_list, &r300->atomname##_state);
+    r300->atomname.name = #atomname; \
+    r300->atomname.state = NULL; \
+    r300->atomname.size = atomsize; \
+    r300->atomname.emit = r300_emit_##atomname; \
+    r300->atomname.dirty = FALSE; \
+    insert_at_tail(&r300->atom_list, &r300->atomname);
 
 static void r300_setup_atoms(struct r300_context* r300)
 {
+    boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500;
+    boolean has_tcl = r300_screen(r300->context.screen)->caps->has_tcl;
+
     /* Create the actual atom list.
      *
      * Each atom is examined and emitted in the order it appears here, which
      * can affect performance and conformance if not handled with care.
      *
-     * Some atoms never change size, others change every emit. This is just
-     * an upper bound on each atom, to keep the emission machinery from
-     * underallocating space. */
+     * Some atoms never change size, others change every emit - those have
+     * the size of 0 here. */
     make_empty_list(&r300->atom_list);
-    R300_INIT_ATOM(invariant, 71);
-    R300_INIT_ATOM(ztop, 2);
-    R300_INIT_ATOM(blend, 8);
-    R300_INIT_ATOM(blend_color, 3);
-    R300_INIT_ATOM(clip, 29);
-    R300_INIT_ATOM(dsa, 8);
-    R300_INIT_ATOM(fb, 56);
-    R300_INIT_ATOM(rs, 25);
-    R300_INIT_ATOM(scissor, 3);
-    R300_INIT_ATOM(viewport, 9);
-    R300_INIT_ATOM(rs_block, 21);
-    R300_INIT_ATOM(vertex_format, 26);
+    R300_INIT_ATOM(invariant_state, 71);
+    R300_INIT_ATOM(ztop_state, 2);
+    R300_INIT_ATOM(blend_state, 8);
+    R300_INIT_ATOM(blend_color_state, is_r500 ? 3 : 2);
+    R300_INIT_ATOM(clip_state, has_tcl ? 5 + (6 * 4) : 2);
+    R300_INIT_ATOM(dsa_state, is_r500 ? 8 : 6);
+    R300_INIT_ATOM(fb_state, 0);
+    R300_INIT_ATOM(rs_state, 0);
+    R300_INIT_ATOM(scissor_state, 3);
+    R300_INIT_ATOM(viewport_state, 9);
+    R300_INIT_ATOM(rs_block_state, 0);
+    R300_INIT_ATOM(vertex_stream_state, 0);
+    R300_INIT_ATOM(vap_output_state, 6);
+    R300_INIT_ATOM(pvs_flush, 2);
+    R300_INIT_ATOM(vs_state, 0);
+    R300_INIT_ATOM(texture_cache_inval, 2);
+    R300_INIT_ATOM(textures_state, 0);
 
     /* Some non-CSO atoms need explicit space to store the state locally. */
+    r300->blend_color_state.state = CALLOC_STRUCT(r300_blend_color_state);
+    r300->clip_state.state = CALLOC_STRUCT(pipe_clip_state);
     r300->fb_state.state = CALLOC_STRUCT(pipe_framebuffer_state);
+    r300->rs_block_state.state = CALLOC_STRUCT(r300_rs_block);
+    r300->scissor_state.state = CALLOC_STRUCT(pipe_scissor_state);
+    r300->textures_state.state = CALLOC_STRUCT(r300_textures_state);
+    r300->vap_output_state.state = CALLOC_STRUCT(r300_vap_output_state);
+    r300->viewport_state.state = CALLOC_STRUCT(r300_viewport_state);
+    r300->ztop_state.state = CALLOC_STRUCT(r300_ztop_state);
 }
 
 struct pipe_context* r300_create_context(struct pipe_screen* screen,
@@ -141,6 +161,8 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
     if (!r300)
         return NULL;
 
+    r300screen->ctx = (struct pipe_context*)r300;
+
     r300->winsys = radeon_winsys;
 
     r300->context.winsys = (struct pipe_winsys*)radeon_winsys;
@@ -178,14 +200,6 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
 
     r300_setup_atoms(r300);
 
-    r300->blend_color_state.state = CALLOC_STRUCT(r300_blend_color_state);
-    r300->clip_state.state = CALLOC_STRUCT(pipe_clip_state);
-    r300->rs_block_state.state = CALLOC_STRUCT(r300_rs_block);
-    r300->scissor_state.state = CALLOC_STRUCT(pipe_scissor_state);
-    r300->vertex_format_state.state = CALLOC_STRUCT(r300_vertex_info);
-    r300->viewport_state.state = CALLOC_STRUCT(r300_viewport_state);
-    r300->ztop_state.state = CALLOC_STRUCT(r300_ztop_state);
-
     /* Open up the OQ BO. */
     r300->oqbo = screen->buffer_create(screen, 4096,
             PIPE_BUFFER_USAGE_VERTEX, 4096);
@@ -202,7 +216,6 @@ struct pipe_context* r300_create_context(struct pipe_screen* screen,
     r300->invariant_state.dirty = TRUE;
 
     r300->winsys->set_flush_cb(r300->winsys, r300_flush_cb, r300);
-    r300->dirty_state = R300_NEW_KITCHEN_SINK;
     r300->dirty_hw++;
 
     r300->blitter = util_blitter_create(&r300->context);
diff --git a/src/gallium/drivers/r300/r300_context.h b/src/gallium/drivers/r300/r300_context.h
index 2edf65797a2..3b70bcda827 100644
--- a/src/gallium/drivers/r300/r300_context.h
+++ b/src/gallium/drivers/r300/r300_context.h
@@ -45,7 +45,7 @@ struct r300_atom {
     /* Opaque state. */
     void* state;
     /* Emit the state to the context. */
-    void (*emit)(struct r300_context*, void*);
+    void (*emit)(struct r300_context*, unsigned, void*);
     /* Upper bound on number of dwords to emit. */
     unsigned size;
     /* Whether this atom should be emitted. */
@@ -86,7 +86,6 @@ struct r300_rs_state {
     uint32_t vap_control_status;    /* R300_VAP_CNTL_STATUS: 0x2140 */
     uint32_t antialiasing_config;   /* R300_GB_AA_CONFIG: 0x4020 */
     uint32_t point_size;            /* R300_GA_POINT_SIZE: 0x421c */
-    uint32_t point_minmax;          /* R300_GA_POINT_MINMAX: 0x4230 */
     uint32_t line_control;          /* R300_GA_LINE_CNTL: 0x4234 */
     float depth_scale;            /* R300_SU_POLY_OFFSET_FRONT_SCALE: 0x42a4 */
                                   /* R300_SU_POLY_OFFSET_BACK_SCALE: 0x42ac */
@@ -119,12 +118,58 @@ struct r300_sampler_state {
     unsigned min_lod, max_lod;
 };
 
-struct r300_texture_state {
+struct r300_texture_format_state {
     uint32_t format0; /* R300_TX_FORMAT0: 0x4480 */
     uint32_t format1; /* R300_TX_FORMAT1: 0x44c0 */
     uint32_t format2; /* R300_TX_FORMAT2: 0x4500 */
 };
 
+#define R300_MAX_TEXTURE_LEVELS 13
+
+struct r300_texture_fb_state {
+    /* Colorbuffer. */
+    uint32_t colorpitch[R300_MAX_TEXTURE_LEVELS]; /* R300_RB3D_COLORPITCH[0-3]*/
+    uint32_t us_out_fmt; /* R300_US_OUT_FMT[0-3] */
+
+    /* Zbuffer. */
+    uint32_t depthpitch[R300_MAX_TEXTURE_LEVELS]; /* R300_RB3D_DEPTHPITCH */
+    uint32_t zb_format; /* R300_ZB_FORMAT */
+};
+
+struct r300_textures_state {
+    /* Textures. */
+    struct r300_texture *textures[8];
+    int texture_count;
+    /* Sampler states. */
+    struct r300_sampler_state *sampler_states[8];
+    int sampler_count;
+
+    /* These is the merge of the texture and sampler states. */
+    unsigned count;
+    uint32_t tx_enable;         /* R300_TX_ENABLE: 0x4101 */
+    struct r300_texture_sampler_state {
+        uint32_t format[3];     /* R300_TX_FORMAT[0-2] */
+        uint32_t filter[2];     /* R300_TX_FILTER[0-1] */
+        uint32_t border_color;  /* R300_TX_BORDER_COLOR: 0x45c0 */
+        uint32_t tile_config;   /* R300_TX_OFFSET (subset thereof) */
+    } regs[8];
+};
+
+struct r300_vertex_stream_state {
+    /* R300_VAP_PROG_STREAK_CNTL_[0-7] */
+    uint32_t vap_prog_stream_cntl[8];
+    /* R300_VAP_PROG_STREAK_CNTL_EXT_[0-7] */
+    uint32_t vap_prog_stream_cntl_ext[8];
+
+    unsigned count;
+};
+
+struct r300_vap_output_state {
+    uint32_t vap_vtx_state_cntl;  /* R300_VAP_VTX_STATE_CNTL: 0x2180 */
+    uint32_t vap_vsm_vtx_assm;    /* R300_VAP_VSM_VTX_ASSM: 0x2184 */
+    uint32_t vap_out_vtx_fmt[2];  /* R300_VAP_OUTPUT_VTX_FMT_[0-1]: 0x2090 */
+};
+
 struct r300_viewport_state {
     float xscale;         /* R300_VAP_VPORT_XSCALE:  0x2098 */
     float xoffset;        /* R300_VAP_VPORT_XOFFSET: 0x209c */
@@ -141,11 +186,6 @@ struct r300_ztop_state {
 
 #define R300_NEW_FRAGMENT_SHADER 0x00000020
 #define R300_NEW_FRAGMENT_SHADER_CONSTANTS    0x00000040
-#define R300_NEW_SAMPLER         0x00000200
-#define R300_ANY_NEW_SAMPLERS    0x0001fe00
-#define R300_NEW_TEXTURE         0x00040000
-#define R300_ANY_NEW_TEXTURES    0x03fc0000
-#define R300_NEW_VERTEX_SHADER   0x08000000
 #define R300_NEW_VERTEX_SHADER_CONSTANTS    0x10000000
 #define R300_NEW_QUERY           0x40000000
 #define R300_NEW_KITCHEN_SINK    0x7fffffff
@@ -198,16 +238,16 @@ struct r300_texture {
     struct pipe_texture tex;
 
     /* Offsets into the buffer. */
-    unsigned offset[PIPE_MAX_TEXTURE_LEVELS];
+    unsigned offset[R300_MAX_TEXTURE_LEVELS];
 
     /* A pitch for each mip-level */
-    unsigned pitch[PIPE_MAX_TEXTURE_LEVELS];
+    unsigned pitch[R300_MAX_TEXTURE_LEVELS];
 
     /* Size of one zslice or face based on the texture target */
-    unsigned layer_size[PIPE_MAX_TEXTURE_LEVELS];
+    unsigned layer_size[R300_MAX_TEXTURE_LEVELS];
 
     /* Whether the mipmap level is macrotiled. */
-    enum r300_buffer_tiling mip_macrotile[PIPE_MAX_TEXTURE_LEVELS];
+    enum r300_buffer_tiling mip_macrotile[R300_MAX_TEXTURE_LEVELS];
 
     /**
      * If non-zero, override the natural texture layout with
@@ -231,7 +271,8 @@ struct r300_texture {
     struct pipe_buffer* buffer;
 
     /* Registers carrying texture format data. */
-    struct r300_texture_state state;
+    struct r300_texture_format_state state;
+    struct r300_texture_fb_state fb_state;
 
     /* Buffer tiling */
     enum r300_buffer_tiling microtile, macrotile;
@@ -247,6 +288,13 @@ struct r300_vertex_info {
     uint32_t vap_prog_stream_cntl_ext[8];
 };
 
+struct r300_vertex_element_state {
+    unsigned count;
+    struct pipe_vertex_element velem[PIPE_MAX_ATTRIBS];
+
+    struct r300_vertex_stream_state vertex_stream;
+};
+
 extern struct pipe_viewport_state r300_viewport_identity;
 
 struct r300_context {
@@ -271,9 +319,6 @@ struct r300_context {
     struct r300_query *query_current;
     struct r300_query query_list;
 
-    /* Vertex formatting information. */
-    struct r300_atom vertex_format_state;
-
     /* Various CSO state objects. */
     /* Beginning of atom list. */
     struct r300_atom atom_list;
@@ -295,20 +340,25 @@ struct r300_context {
     struct r300_atom rs_state;
     /* RS block state. */
     struct r300_atom rs_block_state;
-    /* Sampler states. */
-    struct r300_sampler_state* sampler_states[8];
-    int sampler_count;
     /* Scissor state. */
     struct r300_atom scissor_state;
     /* Sampler view states. */
     struct pipe_sampler_view* fragment_sampler_views[8];
     int fragment_sampler_view_count;
+    /* Vertex stream formatting state. */
+    struct r300_atom vertex_stream_state;
+    /* VAP (vertex shader) output mapping state. */
+    struct r300_atom vap_output_state;
     /* Vertex shader. */
-    struct r300_vertex_shader* vs;
+    struct r300_atom vs_state;
     /* Viewport state. */
     struct r300_atom viewport_state;
     /* ZTOP state. */
     struct r300_atom ztop_state;
+    /* PVS flush. */
+    struct r300_atom pvs_flush;
+    /* Texture cache invalidate. */
+    struct r300_atom texture_cache_inval;
 
     /* Invariant state. This must be emitted to get the engine started. */
     struct r300_atom invariant_state;
@@ -316,22 +366,29 @@ struct r300_context {
     /* Vertex buffers for Gallium. */
     struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
     int vertex_buffer_count;
+    int vertex_buffer_max_index;
     /* Vertex elements for Gallium. */
-    struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS];
-    int vertex_element_count;
+    struct r300_vertex_element_state *velems;
+
+    /* Vertex info for Draw. */
+    struct vertex_info vertex_info;
 
     struct pipe_stencil_ref stencil_ref;
 
+    struct pipe_clip_state clip;
+
+    struct pipe_viewport_state viewport;
+
     /* Bitmask of dirty state objects. */
     uint32_t dirty_state;
     /* Flag indicating whether or not the HW is dirty. */
     uint32_t dirty_hw;
-    /* Whether the TCL engine should be in bypass mode. */
-    boolean tcl_bypass;
     /* Whether polygon offset is enabled. */
     boolean polygon_offset_enabled;
     /* Z buffer bit depth. */
     uint32_t zbuffer_bpp;
+    /* Whether scissor is enabled. */
+    boolean scissor_enabled;
 };
 
 /* Convenience cast wrapper. */
diff --git a/src/gallium/drivers/r300/r300_emit.c b/src/gallium/drivers/r300/r300_emit.c
index 4310ff1411d..fc8a8a27738 100644
--- a/src/gallium/drivers/r300/r300_emit.c
+++ b/src/gallium/drivers/r300/r300_emit.c
@@ -32,17 +32,17 @@
 #include "r300_emit.h"
 #include "r300_fs.h"
 #include "r300_screen.h"
-#include "r300_state_inlines.h"
 #include "r300_vs.h"
 
-void r300_emit_blend_state(struct r300_context* r300, void* state)
+void r300_emit_blend_state(struct r300_context* r300,
+                           unsigned size, void* state)
 {
     struct r300_blend_state* blend = (struct r300_blend_state*)state;
     struct pipe_framebuffer_state* fb =
         (struct pipe_framebuffer_state*)r300->fb_state.state;
     CS_LOCALS(r300);
 
-    BEGIN_CS(8);
+    BEGIN_CS(size);
     OUT_CS_REG(R300_RB3D_ROPCNTL, blend->rop);
     OUT_CS_REG_SEQ(R300_RB3D_CBLEND, 3);
     if (fb->nr_cbufs) {
@@ -59,26 +59,28 @@ void r300_emit_blend_state(struct r300_context* r300, void* state)
     END_CS;
 }
 
-void r300_emit_blend_color_state(struct r300_context* r300, void* state)
+void r300_emit_blend_color_state(struct r300_context* r300,
+                                 unsigned size, void* state)
 {
     struct r300_blend_color_state* bc = (struct r300_blend_color_state*)state;
     struct r300_screen* r300screen = r300_screen(r300->context.screen);
     CS_LOCALS(r300);
 
     if (r300screen->caps->is_r500) {
-        BEGIN_CS(3);
+        BEGIN_CS(size);
         OUT_CS_REG_SEQ(R500_RB3D_CONSTANT_COLOR_AR, 2);
         OUT_CS(bc->blend_color_red_alpha);
         OUT_CS(bc->blend_color_green_blue);
         END_CS;
     } else {
-        BEGIN_CS(2);
+        BEGIN_CS(size);
         OUT_CS_REG(R300_RB3D_BLEND_COLOR, bc->blend_color);
         END_CS;
     }
 }
 
-void r300_emit_clip_state(struct r300_context* r300, void* state)
+void r300_emit_clip_state(struct r300_context* r300,
+                          unsigned size, void* state)
 {
     struct pipe_clip_state* clip = (struct pipe_clip_state*)state;
     int i;
@@ -86,7 +88,7 @@ void r300_emit_clip_state(struct r300_context* r300, void* state)
     CS_LOCALS(r300);
 
     if (r300screen->caps->has_tcl) {
-        BEGIN_CS(5 + (6 * 4));
+        BEGIN_CS(size);
         OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG,
                 (r300screen->caps->is_r500 ?
                  R500_PVS_UCP_START : R300_PVS_UCP_START));
@@ -101,14 +103,14 @@ void r300_emit_clip_state(struct r300_context* r300, void* state)
                 R300_PS_UCP_MODE_CLIP_AS_TRIFAN);
         END_CS;
     } else {
-        BEGIN_CS(2);
+        BEGIN_CS(size);
         OUT_CS_REG(R300_VAP_CLIP_CNTL, R300_CLIP_DISABLE);
         END_CS;
     }
 
 }
 
-void r300_emit_dsa_state(struct r300_context* r300, void* state)
+void r300_emit_dsa_state(struct r300_context* r300, unsigned size, void* state)
 {
     struct r300_dsa_state* dsa = (struct r300_dsa_state*)state;
     struct r300_screen* r300screen = r300_screen(r300->context.screen);
@@ -117,7 +119,7 @@ void r300_emit_dsa_state(struct r300_context* r300, void* state)
     struct pipe_stencil_ref stencil_ref = r300->stencil_ref;
     CS_LOCALS(r300);
 
-    BEGIN_CS(r300screen->caps->is_r500 ? 8 : 6);
+    BEGIN_CS(size);
     OUT_CS_REG(R300_FG_ALPHA_FUNC, dsa->alpha_function);
     OUT_CS_REG_SEQ(R300_ZB_CNTL, 3);
 
@@ -144,6 +146,8 @@ static const float * get_shader_constant(
 {
     struct r300_viewport_state* viewport =
         (struct r300_viewport_state*)r300->viewport_state.state;
+    struct r300_textures_state* texstate =
+        (struct r300_textures_state*)r300->textures_state.state;
     static float vec[4] = { 0.0, 0.0, 0.0, 1.0 };
     struct pipe_texture *tex;
 
@@ -171,23 +175,15 @@ static const float * get_shader_constant(
                     break;
 
                 case RC_STATE_R300_VIEWPORT_SCALE:
-                    if (r300->tcl_bypass) {
-                        vec[0] = 1;
-                        vec[1] = 1;
-                        vec[2] = 1;
-                    } else {
-                        vec[0] = viewport->xscale;
-                        vec[1] = viewport->yscale;
-                        vec[2] = viewport->zscale;
-                    }
+                    vec[0] = viewport->xscale;
+                    vec[1] = viewport->yscale;
+                    vec[2] = viewport->zscale;
                     break;
 
                 case RC_STATE_R300_VIEWPORT_OFFSET:
-                    if (!r300->tcl_bypass) {
-                        vec[0] = viewport->xoffset;
-                        vec[1] = viewport->yoffset;
-                        vec[2] = viewport->zoffset;
-                    }
+                    vec[0] = viewport->xoffset;
+                    vec[1] = viewport->yoffset;
+                    vec[2] = viewport->zoffset;
                     break;
 
                 default:
@@ -379,7 +375,7 @@ void r500_emit_fs_constant_buffer(struct r300_context* r300,
     END_CS;
 }
 
-void r300_emit_fb_state(struct r300_context* r300, void* state)
+void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state)
 {
     struct pipe_framebuffer_state* fb = (struct pipe_framebuffer_state*)state;
     struct r300_screen* r300screen = r300_screen(r300->context.screen);
@@ -388,7 +384,7 @@ void r300_emit_fb_state(struct r300_context* r300, void* state)
     int i;
     CS_LOCALS(r300);
 
-    BEGIN_CS((10 * fb->nr_cbufs) + (fb->zsbuf ? 10 : 0) + 6);
+    BEGIN_CS(size);
 
     /* Flush and free renderbuffer caches. */
     OUT_CS_REG(R300_RB3D_DSTCACHE_CTLSTAT,
@@ -422,14 +418,13 @@ void r300_emit_fb_state(struct r300_context* r300, void* state)
         OUT_CS_RELOC(tex->buffer, surf->offset, 0, RADEON_GEM_DOMAIN_VRAM, 0);
 
         OUT_CS_REG_SEQ(R300_RB3D_COLORPITCH0 + (4 * i), 1);
-        OUT_CS_RELOC(tex->buffer, tex->pitch[surf->level] |
-                     r300_translate_colorformat(tex->tex.format) |
-                     R300_COLOR_TILE(tex->mip_macrotile[surf->level]) |
-                     R300_COLOR_MICROTILE(tex->microtile),
+        OUT_CS_RELOC(tex->buffer, tex->fb_state.colorpitch[surf->level],
                      0, RADEON_GEM_DOMAIN_VRAM, 0);
 
-        OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i),
-            r300_translate_out_fmt(surf->format));
+        OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i), tex->fb_state.us_out_fmt);
+    }
+    for (; i < 4; i++) {
+        OUT_CS_REG(R300_US_OUT_FMT_0 + (4 * i), R300_US_OUT_FMT_UNUSED);
     }
 
     /* Set up a zbuffer. */
@@ -441,15 +436,15 @@ void r300_emit_fb_state(struct r300_context* r300, void* state)
         OUT_CS_REG_SEQ(R300_ZB_DEPTHOFFSET, 1);
         OUT_CS_RELOC(tex->buffer, surf->offset, 0, RADEON_GEM_DOMAIN_VRAM, 0);
 
-        OUT_CS_REG(R300_ZB_FORMAT, r300_translate_zsformat(tex->tex.format));
+        OUT_CS_REG(R300_ZB_FORMAT, tex->fb_state.zb_format);
 
         OUT_CS_REG_SEQ(R300_ZB_DEPTHPITCH, 1);
-        OUT_CS_RELOC(tex->buffer, tex->pitch[surf->level] |
-                     R300_DEPTHMACROTILE(tex->mip_macrotile[surf->level]) |
-                     R300_DEPTHMICROTILE(tex->microtile),
+        OUT_CS_RELOC(tex->buffer, tex->fb_state.depthpitch[surf->level],
                      0, RADEON_GEM_DOMAIN_VRAM, 0);
     }
 
+    OUT_CS_REG(R300_GA_POINT_MINMAX,
+        (MAX2(fb->width, fb->height) * 6) << R300_GA_POINT_MINMAX_MAX_SHIFT);
     END_CS;
 }
 
@@ -579,21 +574,19 @@ void r300_emit_query_end(struct r300_context* r300)
         r300_emit_query_finish(r300, query);
 }
 
-void r300_emit_rs_state(struct r300_context* r300, void* state)
+void r300_emit_rs_state(struct r300_context* r300, unsigned size, void* state)
 {
     struct r300_rs_state* rs = (struct r300_rs_state*)state;
     float scale, offset;
     CS_LOCALS(r300);
 
-    BEGIN_CS(18 + (rs->polygon_offset_enable ? 5 : 0));
+    BEGIN_CS(size);
     OUT_CS_REG(R300_VAP_CNTL_STATUS, rs->vap_control_status);
 
     OUT_CS_REG(R300_GB_AA_CONFIG, rs->antialiasing_config);
 
     OUT_CS_REG(R300_GA_POINT_SIZE, rs->point_size);
-    OUT_CS_REG_SEQ(R300_GA_POINT_MINMAX, 2);
-    OUT_CS(rs->point_minmax);
-    OUT_CS(rs->line_control);
+    OUT_CS_REG(R300_GA_LINE_CNTL, rs->line_control);
 
     if (rs->polygon_offset_enable) {
         scale = rs->depth_scale * 12;
@@ -624,7 +617,8 @@ void r300_emit_rs_state(struct r300_context* r300, void* state)
     END_CS;
 }
 
-void r300_emit_rs_block_state(struct r300_context* r300, void* state)
+void r300_emit_rs_block_state(struct r300_context* r300,
+                              unsigned size, void* state)
 {
     struct r300_rs_block* rs = (struct r300_rs_block*)state;
     unsigned i;
@@ -635,7 +629,7 @@ void r300_emit_rs_block_state(struct r300_context* r300, void* state)
 
     DBG(r300, DBG_DRAW, "r300: RS emit:\n");
 
-    BEGIN_CS(5 + count*2);
+    BEGIN_CS(size);
     if (r300screen->caps->is_r500) {
         OUT_CS_REG_SEQ(R500_RS_IP_0, count);
     } else {
@@ -666,7 +660,8 @@ void r300_emit_rs_block_state(struct r300_context* r300, void* state)
     END_CS;
 }
 
-void r300_emit_scissor_state(struct r300_context* r300, void* state)
+void r300_emit_scissor_state(struct r300_context* r300,
+                             unsigned size, void* state)
 {
     unsigned minx, miny, maxx, maxy;
     uint32_t top_left, bottom_right;
@@ -680,7 +675,7 @@ void r300_emit_scissor_state(struct r300_context* r300, void* state)
     maxx = fb->width;
     maxy = fb->height;
 
-    if (((struct r300_rs_state*)r300->rs_state.state)->rs.scissor) {
+    if (r300->scissor_enabled) {
         minx = MAX2(minx, scissor->minx);
         miny = MAX2(miny, scissor->miny);
         maxx = MIN2(maxx, scissor->maxx);
@@ -720,65 +715,51 @@ void r300_emit_scissor_state(struct r300_context* r300, void* state)
             (((maxy - 1) + 1440) << R300_SCISSORS_Y_SHIFT);
     }
 
-    BEGIN_CS(3);
+    BEGIN_CS(size);
     OUT_CS_REG_SEQ(R300_SC_SCISSORS_TL, 2);
     OUT_CS(top_left);
     OUT_CS(bottom_right);
     END_CS;
 }
 
-void r300_emit_texture(struct r300_context* r300,
-                       struct r300_sampler_state* sampler,
-                       struct r300_texture* tex,
-                       unsigned offset)
+void r300_emit_textures_state(struct r300_context *r300,
+                              unsigned size, void *state)
 {
-    uint32_t filter0 = sampler->filter0;
-    uint32_t format0 = tex->state.format0;
-    unsigned min_level, max_level;
+    struct r300_textures_state *allstate = (struct r300_textures_state*)state;
+    struct r300_texture_sampler_state *texstate;
+    unsigned i;
     CS_LOCALS(r300);
 
-    /* to emulate 1D textures through 2D ones correctly */
-    if (tex->tex.target == PIPE_TEXTURE_1D) {
-        filter0 &= ~R300_TX_WRAP_T_MASK;
-        filter0 |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE);
-    }
+    BEGIN_CS(size);
+    OUT_CS_REG(R300_TX_ENABLE, allstate->tx_enable);
 
-    if (tex->is_npot) {
-        /* NPOT textures don't support mip filter, unfortunately.
-         * This prevents incorrect rendering. */
-        filter0 &= ~R300_TX_MIN_FILTER_MIP_MASK;
-    } else {
-        /* determine min/max levels */
-        /* the MAX_MIP level is the largest (finest) one */
-        max_level = MIN2(sampler->max_lod, tex->tex.last_level);
-        min_level = MIN2(sampler->min_lod, max_level);
-        format0 |= R300_TX_NUM_LEVELS(max_level);
-        filter0 |= R300_TX_MAX_MIP_LEVEL(min_level);
-    }
+    for (i = 0; i < allstate->count; i++) {
+        if ((1 << i) & allstate->tx_enable) {
+            texstate = &allstate->regs[i];
+
+            OUT_CS_REG(R300_TX_FILTER0_0 + (i * 4), texstate->filter[0]);
+            OUT_CS_REG(R300_TX_FILTER1_0 + (i * 4), texstate->filter[1]);
+            OUT_CS_REG(R300_TX_BORDER_COLOR_0 + (i * 4),
+                       texstate->border_color);
 
-    BEGIN_CS(16);
-    OUT_CS_REG(R300_TX_FILTER0_0 + (offset * 4), filter0 |
-        (offset << 28));
-    OUT_CS_REG(R300_TX_FILTER1_0 + (offset * 4), sampler->filter1);
-    OUT_CS_REG(R300_TX_BORDER_COLOR_0 + (offset * 4), sampler->border_color);
-
-    OUT_CS_REG(R300_TX_FORMAT0_0 + (offset * 4), format0);
-    OUT_CS_REG(R300_TX_FORMAT1_0 + (offset * 4), tex->state.format1);
-    OUT_CS_REG(R300_TX_FORMAT2_0 + (offset * 4), tex->state.format2);
-    OUT_CS_REG_SEQ(R300_TX_OFFSET_0 + (offset * 4), 1);
-    OUT_CS_RELOC(tex->buffer,
-                 R300_TXO_MACRO_TILE(tex->macrotile) |
-                 R300_TXO_MICRO_TILE(tex->microtile),
-                 RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0, 0);
+            OUT_CS_REG(R300_TX_FORMAT0_0 + (i * 4), texstate->format[0]);
+            OUT_CS_REG(R300_TX_FORMAT1_0 + (i * 4), texstate->format[1]);
+            OUT_CS_REG(R300_TX_FORMAT2_0 + (i * 4), texstate->format[2]);
+
+            OUT_CS_REG_SEQ(R300_TX_OFFSET_0 + (i * 4), 1);
+            OUT_CS_RELOC(allstate->textures[i]->buffer, texstate->tile_config,
+                         RADEON_GEM_DOMAIN_GTT | RADEON_GEM_DOMAIN_VRAM, 0, 0);
+        }
+    }
     END_CS;
 }
 
 void r300_emit_aos(struct r300_context* r300, unsigned offset)
 {
     struct pipe_vertex_buffer *vb1, *vb2, *vbuf = r300->vertex_buffer;
-    struct pipe_vertex_element *velem = r300->vertex_element;
+    struct pipe_vertex_element *velem = r300->velems->velem;
     int i;
-    unsigned size1, size2, aos_count = r300->vertex_element_count;
+    unsigned size1, size2, aos_count = r300->velems->count;
     unsigned packet_size = (aos_count * 3 + 1) / 2;
     CS_LOCALS(r300);
 
@@ -813,68 +794,110 @@ void r300_emit_aos(struct r300_context* r300, unsigned offset)
     END_CS;
 }
 
-void r300_emit_vertex_format_state(struct r300_context* r300, void* state)
+void r300_emit_vertex_buffer(struct r300_context* r300)
 {
-    struct r300_vertex_info* vertex_info = (struct r300_vertex_info*)state;
-    unsigned i;
     CS_LOCALS(r300);
 
-    DBG(r300, DBG_DRAW, "r300: VAP/PSC emit:\n");
+    DBG(r300, DBG_DRAW, "r300: Preparing vertex buffer %p for render, "
+            "vertex size %d\n", r300->vbo,
+            r300->vertex_info.size);
+    /* Set the pointer to our vertex buffer. The emitted values are this:
+     * PACKET3 [3D_LOAD_VBPNTR]
+     * COUNT   [1]
+     * FORMAT  [size | stride << 8]
+     * OFFSET  [offset into BO]
+     * VBPNTR  [relocated BO]
+     */
+    BEGIN_CS(7);
+    OUT_CS_PKT3(R300_PACKET3_3D_LOAD_VBPNTR, 3);
+    OUT_CS(1);
+    OUT_CS(r300->vertex_info.size |
+            (r300->vertex_info.size << 8));
+    OUT_CS(r300->vbo_offset);
+    OUT_CS_RELOC(r300->vbo, 0, RADEON_GEM_DOMAIN_GTT, 0, 0);
+    END_CS;
+}
 
-    BEGIN_CS(26);
-    OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_info->vinfo.size);
+void r300_emit_vertex_stream_state(struct r300_context* r300,
+                                   unsigned size, void* state)
+{
+    struct r300_vertex_stream_state *streams =
+        (struct r300_vertex_stream_state*)state;
+    unsigned i;
+    CS_LOCALS(r300);
 
-    OUT_CS_REG_SEQ(R300_VAP_VTX_STATE_CNTL, 2);
-    OUT_CS(vertex_info->vinfo.hwfmt[0]);
-    OUT_CS(vertex_info->vinfo.hwfmt[1]);
-    OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2);
-    OUT_CS(vertex_info->vinfo.hwfmt[2]);
-    OUT_CS(vertex_info->vinfo.hwfmt[3]);
-    for (i = 0; i < 4; i++) {
-       DBG(r300, DBG_DRAW, "    : hwfmt%d: 0x%08x\n", i,
-               vertex_info->vinfo.hwfmt[i]);
-    }
+    DBG(r300, DBG_DRAW, "r300: PSC emit:\n");
 
-    OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_0, 8);
-    for (i = 0; i < 8; i++) {
-        OUT_CS(vertex_info->vap_prog_stream_cntl[i]);
+    BEGIN_CS(size);
+    OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_0, streams->count);
+    for (i = 0; i < streams->count; i++) {
+        OUT_CS(streams->vap_prog_stream_cntl[i]);
         DBG(r300, DBG_DRAW, "    : prog_stream_cntl%d: 0x%08x\n", i,
-               vertex_info->vap_prog_stream_cntl[i]);
+               streams->vap_prog_stream_cntl[i]);
     }
-    OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_EXT_0, 8);
-    for (i = 0; i < 8; i++) {
-        OUT_CS(vertex_info->vap_prog_stream_cntl_ext[i]);
+    OUT_CS_REG_SEQ(R300_VAP_PROG_STREAM_CNTL_EXT_0, streams->count);
+    for (i = 0; i < streams->count; i++) {
+        OUT_CS(streams->vap_prog_stream_cntl_ext[i]);
         DBG(r300, DBG_DRAW, "    : prog_stream_cntl_ext%d: 0x%08x\n", i,
-               vertex_info->vap_prog_stream_cntl_ext[i]);
+               streams->vap_prog_stream_cntl_ext[i]);
     }
     END_CS;
 }
 
+void r300_emit_vap_output_state(struct r300_context* r300,
+                               unsigned size, void* state)
+{
+    struct r300_vap_output_state *vap_out_state =
+        (struct r300_vap_output_state*)state;
+    CS_LOCALS(r300);
+
+    DBG(r300, DBG_DRAW, "r300: VAP emit:\n");
 
-void r300_emit_vertex_program_code(struct r300_context* r300,
-                                   struct r300_vertex_program_code* code)
+    BEGIN_CS(size);
+    OUT_CS_REG_SEQ(R300_VAP_VTX_STATE_CNTL, 2);
+    OUT_CS(vap_out_state->vap_vtx_state_cntl);
+    OUT_CS(vap_out_state->vap_vsm_vtx_assm);
+    OUT_CS_REG_SEQ(R300_VAP_OUTPUT_VTX_FMT_0, 2);
+    OUT_CS(vap_out_state->vap_out_vtx_fmt[0]);
+    OUT_CS(vap_out_state->vap_out_vtx_fmt[1]);
+    END_CS;
+}
+
+void r300_emit_pvs_flush(struct r300_context* r300, unsigned size, void* state)
 {
-    int i;
+    CS_LOCALS(r300);
+
+    BEGIN_CS(size);
+    OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0);
+    END_CS;
+}
+
+void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state)
+{
+    struct r300_vertex_shader* vs = (struct r300_vertex_shader*)state;
+    struct r300_vertex_program_code* code = &vs->code;
     struct r300_screen* r300screen = r300_screen(r300->context.screen);
     unsigned instruction_count = code->length / 4;
+    unsigned i;
+
+    unsigned vtx_mem_size = r300screen->caps->is_r500 ? 128 : 72;
+    unsigned input_count = MAX2(util_bitcount(code->InputsRead), 1);
+    unsigned output_count = MAX2(util_bitcount(code->OutputsWritten), 1);
+    unsigned temp_count = MAX2(code->num_temporaries, 1);
 
-    int vtx_mem_size = r300screen->caps->is_r500 ? 128 : 72;
-    int input_count = MAX2(util_bitcount(code->InputsRead), 1);
-    int output_count = MAX2(util_bitcount(code->OutputsWritten), 1);
-    int temp_count = MAX2(code->num_temporaries, 1);
-    int pvs_num_slots = MIN3(vtx_mem_size / input_count,
-                             vtx_mem_size / output_count, 10);
-    int pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 6);
+    unsigned pvs_num_slots = MIN3(vtx_mem_size / input_count,
+                                  vtx_mem_size / output_count, 10);
+    unsigned pvs_num_controllers = MIN2(vtx_mem_size / temp_count, 6);
 
     CS_LOCALS(r300);
 
     if (!r300screen->caps->has_tcl) {
-        debug_printf("r300: Implementation error: emit_vertex_shader called,"
+        debug_printf("r300: Implementation error: emit_vs_state called,"
                 " but has_tcl is FALSE!\n");
         return;
     }
 
-    BEGIN_CS(9 + code->length);
+    BEGIN_CS(size);
     /* R300_VAP_PVS_CODE_CNTL_0
      * R300_VAP_PVS_CONST_CNTL
      * R300_VAP_PVS_CODE_CNTL_1
@@ -888,8 +911,9 @@ void r300_emit_vertex_program_code(struct r300_context* r300,
 
     OUT_CS_REG(R300_VAP_PVS_VECTOR_INDX_REG, 0);
     OUT_CS_ONE_REG(R300_VAP_PVS_UPLOAD_DATA, code->length);
-    for (i = 0; i < code->length; i++)
+    for (i = 0; i < code->length; i++) {
         OUT_CS(code->body.d[i]);
+    }
 
     OUT_CS_REG(R300_VAP_CNTL, R300_PVS_NUM_SLOTS(pvs_num_slots) |
             R300_PVS_NUM_CNTLRS(pvs_num_controllers) |
@@ -899,12 +923,6 @@ void r300_emit_vertex_program_code(struct r300_context* r300,
     END_CS;
 }
 
-void r300_emit_vertex_shader(struct r300_context* r300,
-                             struct r300_vertex_shader* vs)
-{
-    r300_emit_vertex_program_code(r300, &vs->code);
-}
-
 void r300_emit_vs_constant_buffer(struct r300_context* r300,
                                   struct rc_constant_list* constants)
 {
@@ -913,7 +931,7 @@ void r300_emit_vs_constant_buffer(struct r300_context* r300,
     CS_LOCALS(r300);
 
     if (!r300screen->caps->has_tcl) {
-        debug_printf("r300: Implementation error: emit_vertex_shader called,"
+        debug_printf("r300: Implementation error: emit_vs_constant_buffer called,"
         " but has_tcl is FALSE!\n");
         return;
     }
@@ -938,83 +956,56 @@ void r300_emit_vs_constant_buffer(struct r300_context* r300,
     END_CS;
 }
 
-void r300_emit_viewport_state(struct r300_context* r300, void* state)
+void r300_emit_viewport_state(struct r300_context* r300,
+                              unsigned size, void* state)
 {
     struct r300_viewport_state* viewport = (struct r300_viewport_state*)state;
     CS_LOCALS(r300);
 
-    if (r300->tcl_bypass) {
-        BEGIN_CS(2);
-        OUT_CS_REG(R300_VAP_VTE_CNTL, 0);
-        END_CS;
-    } else {
-        BEGIN_CS(9);
-        OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6);
-        OUT_CS_32F(viewport->xscale);
-        OUT_CS_32F(viewport->xoffset);
-        OUT_CS_32F(viewport->yscale);
-        OUT_CS_32F(viewport->yoffset);
-        OUT_CS_32F(viewport->zscale);
-        OUT_CS_32F(viewport->zoffset);
-        OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control);
-        END_CS;
-    }
-}
-
-void r300_emit_texture_count(struct r300_context* r300)
-{
-    uint32_t tx_enable = 0;
-    int i;
-    CS_LOCALS(r300);
-
-    /* Notice that fragment_sampler_view_count and sampler_count are just sizes
-     * of the respective arrays. We still have to check for the individual
-     * elements. */
-    for (i = 0; i < MIN2(r300->sampler_count, r300->fragment_sampler_view_count); i++) {
-        if (r300->fragment_sampler_views[i]) {
-            tx_enable |= 1 << i;
-        }
-    }
-
-    BEGIN_CS(2);
-    OUT_CS_REG(R300_TX_ENABLE, tx_enable);
-    END_CS;
-
+     BEGIN_CS(size);
+     OUT_CS_REG_SEQ(R300_SE_VPORT_XSCALE, 6);
+     OUT_CS_32F(viewport->xscale);
+     OUT_CS_32F(viewport->xoffset);
+     OUT_CS_32F(viewport->yscale);
+     OUT_CS_32F(viewport->yoffset);
+     OUT_CS_32F(viewport->zscale);
+     OUT_CS_32F(viewport->zoffset);
+     OUT_CS_REG(R300_VAP_VTE_CNTL, viewport->vte_control);
+     END_CS;
 }
 
-void r300_emit_ztop_state(struct r300_context* r300, void* state)
+void r300_emit_ztop_state(struct r300_context* r300,
+                          unsigned size, void* state)
 {
     struct r300_ztop_state* ztop = (struct r300_ztop_state*)state;
     CS_LOCALS(r300);
 
-    BEGIN_CS(2);
+    BEGIN_CS(size);
     OUT_CS_REG(R300_ZB_ZTOP, ztop->z_buffer_top);
     END_CS;
 }
 
-void r300_flush_textures(struct r300_context* r300)
+void r300_emit_texture_cache_inval(struct r300_context* r300, unsigned size, void* state)
 {
     CS_LOCALS(r300);
 
-    BEGIN_CS(2);
+    BEGIN_CS(size);
     OUT_CS_REG(R300_TX_INVALTAGS, 0);
     END_CS;
 }
 
-static void r300_flush_pvs(struct r300_context* r300)
-{
-    CS_LOCALS(r300);
-
-    BEGIN_CS(2);
-    OUT_CS_REG(R300_VAP_PVS_STATE_FLUSH_REG, 0x0);
-    END_CS;
-}
-
-void r300_emit_buffer_validate(struct r300_context *r300)
+void r300_emit_buffer_validate(struct r300_context *r300,
+                               boolean do_validate_vertex_buffers,
+                               struct pipe_buffer *index_buffer)
 {
     struct pipe_framebuffer_state* fb =
         (struct pipe_framebuffer_state*)r300->fb_state.state;
+    struct r300_textures_state *texstate =
+        (struct r300_textures_state*)r300->textures_state.state;
     struct r300_texture* tex;
+    struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
+    struct pipe_vertex_element *velem = r300->velems->velem;
+    struct pipe_buffer *pbuf;
     unsigned i;
     boolean invalid = FALSE;
 
@@ -1061,16 +1052,35 @@ validate:
             goto validate;
         }
     }
-    /* ...and vertex buffer. */
+    /* ...vertex buffer for SWTCL path... */
     if (r300->vbo) {
         if (!r300->winsys->add_buffer(r300->winsys, r300->vbo,
                     RADEON_GEM_DOMAIN_GTT, 0)) {
             r300->context.flush(&r300->context, 0, NULL);
             goto validate;
         }
-    } else {
-        /* debug_printf("No VBO while emitting dirty state!\n"); */
     }
+    /* ...vertex buffers for HWTCL path... */
+    if (do_validate_vertex_buffers) {
+        for (i = 0; i < r300->velems->count; i++) {
+            pbuf = vbuf[velem[i].vertex_buffer_index].buffer;
+
+            if (!r300->winsys->add_buffer(r300->winsys, pbuf,
+                                          RADEON_GEM_DOMAIN_GTT, 0)) {
+                r300->context.flush(&r300->context, 0, NULL);
+                goto validate;
+            }
+        }
+    }
+    /* ...and index buffer for HWTCL path. */
+    if (index_buffer) {
+        if (!r300->winsys->add_buffer(r300->winsys, index_buffer,
+                                      RADEON_GEM_DOMAIN_GTT, 0)) {
+            r300->context.flush(&r300->context, 0, NULL);
+            goto validate;
+        }
+    }
+
     if (!r300->winsys->validate(r300->winsys)) {
         r300->context.flush(&r300->context, 0, NULL);
         if (invalid) {
@@ -1083,16 +1093,10 @@ validate:
     }
 }
 
-/* Emit all dirty state. */
-void r300_emit_dirty_state(struct r300_context* r300)
+unsigned r300_get_num_dirty_dwords(struct r300_context *r300)
 {
-    struct r300_screen* r300screen = r300_screen(r300->context.screen);
     struct r300_atom* atom;
-    unsigned i, dwords = 1024;
-    int dirty_tex = 0;
-
-    /* Check the required number of dwords against the space remaining in the
-     * current CS object. If we need more, then flush. */
+    unsigned dwords = 0;
 
     foreach(atom, &r300->atom_list) {
         if (atom->dirty || atom->always_dirty) {
@@ -1100,12 +1104,17 @@ void r300_emit_dirty_state(struct r300_context* r300)
         }
     }
 
-    /* Make sure we have at least 2*1024 spare dwords. */
-    /* XXX It would be nice to know the number of dwords we really need to
-     * XXX emit. */
-    while (!r300->winsys->check_cs(r300->winsys, dwords)) {
-        r300->context.flush(&r300->context, 0, NULL);
-    }
+    /* XXX This is the compensation for the non-atomized states. */
+    dwords += 1024;
+
+    return dwords;
+}
+
+/* Emit all dirty state. */
+void r300_emit_dirty_state(struct r300_context* r300)
+{
+    struct r300_screen* r300screen = r300_screen(r300->context.screen);
+    struct r300_atom* atom;
 
     if (r300->dirty_state & R300_NEW_QUERY) {
         r300_emit_query_start(r300);
@@ -1114,7 +1123,7 @@ void r300_emit_dirty_state(struct r300_context* r300)
 
     foreach(atom, &r300->atom_list) {
         if (atom->dirty || atom->always_dirty) {
-            atom->emit(r300, atom->state);
+            atom->emit(r300, atom->size, atom->state);
             atom->dirty = FALSE;
         }
     }
@@ -1140,43 +1149,9 @@ void r300_emit_dirty_state(struct r300_context* r300)
         r300->dirty_state &= ~R300_NEW_FRAGMENT_SHADER_CONSTANTS;
     }
 
-    /* Samplers and textures are tracked separately but emitted together. */
-    if (r300->dirty_state &
-            (R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES)) {
-        r300_emit_texture_count(r300);
-
-        for (i = 0; i < MIN2(r300->sampler_count, r300->fragment_sampler_view_count); i++) {
-  	    if (r300->dirty_state &
-		((R300_NEW_SAMPLER << i) | (R300_NEW_TEXTURE << i))) {
-		if (r300->fragment_sampler_views[i]) {
-		    r300_emit_texture(r300,
-				      r300->sampler_states[i],
-				      (struct r300_texture *)r300->fragment_sampler_views[i]->texture,
-				      i);
-                    dirty_tex |= r300->dirty_state & (R300_NEW_TEXTURE << i);
-                }
-                r300->dirty_state &=
-                    ~((R300_NEW_SAMPLER << i) | (R300_NEW_TEXTURE << i));
-            }
-        }
-        r300->dirty_state &= ~(R300_ANY_NEW_SAMPLERS | R300_ANY_NEW_TEXTURES);
-    }
-
-    if (dirty_tex) {
-        r300_flush_textures(r300);
-    }
-
-    if (r300->dirty_state & (R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS)) {
-        r300_flush_pvs(r300);
-    }
-
-    if (r300->dirty_state & R300_NEW_VERTEX_SHADER) {
-        r300_emit_vertex_shader(r300, r300->vs);
-        r300->dirty_state &= ~R300_NEW_VERTEX_SHADER;
-    }
-
     if (r300->dirty_state & R300_NEW_VERTEX_SHADER_CONSTANTS) {
-        r300_emit_vs_constant_buffer(r300, &r300->vs->code.constants);
+        struct r300_vertex_shader* vs = r300->vs_state.state;
+        r300_emit_vs_constant_buffer(r300, &vs->code.constants);
         r300->dirty_state &= ~R300_NEW_VERTEX_SHADER_CONSTANTS;
     }
 
@@ -1184,8 +1159,10 @@ void r300_emit_dirty_state(struct r300_context* r300)
     assert(r300->dirty_state == 0);
     */
 
-    /* Finally, emit the VBO. */
-    /* r300_emit_vertex_buffer(r300); */
+    /* Emit the VBO for SWTCL. */
+    if (!r300screen->caps->has_tcl) {
+        r300_emit_vertex_buffer(r300);
+    }
 
     r300->dirty_hw++;
 }
diff --git a/src/gallium/drivers/r300/r300_emit.h b/src/gallium/drivers/r300/r300_emit.h
index 6b96d9b57c0..449e640a884 100644
--- a/src/gallium/drivers/r300/r300_emit.h
+++ b/src/gallium/drivers/r300/r300_emit.h
@@ -31,13 +31,17 @@ struct r300_vertex_program_code;
 
 void r300_emit_aos(struct r300_context* r300, unsigned offset);
 
-void r300_emit_blend_state(struct r300_context* r300, void* state);
+void r300_emit_blend_state(struct r300_context* r300,
+                           unsigned size, void* state);
 
-void r300_emit_blend_color_state(struct r300_context* r300, void* state);
+void r300_emit_blend_color_state(struct r300_context* r300,
+                                 unsigned size, void* state);
 
-void r300_emit_clip_state(struct r300_context* r300, void* state);
+void r300_emit_clip_state(struct r300_context* r300,
+                          unsigned size, void* state);
 
-void r300_emit_dsa_state(struct r300_context* r300, void* state);
+void r300_emit_dsa_state(struct r300_context* r300,
+                         unsigned size, void* state);
 
 void r300_emit_fragment_program_code(struct r300_context* r300,
                                      struct rX00_fragment_program_code* generic_code);
@@ -51,48 +55,54 @@ void r500_emit_fragment_program_code(struct r300_context* r300,
 void r500_emit_fs_constant_buffer(struct r300_context* r300,
                                   struct rc_constant_list* constants);
 
-void r300_emit_fb_state(struct r300_context* r300, void* state);
+void r300_emit_fb_state(struct r300_context* r300, unsigned size, void* state);
 
 void r300_emit_query_begin(struct r300_context* r300,
                            struct r300_query* query);
 
 void r300_emit_query_end(struct r300_context* r300);
 
-void r300_emit_rs_state(struct r300_context* r300, void* state);
+void r300_emit_rs_state(struct r300_context* r300, unsigned size, void* state);
 
-void r300_emit_rs_block_state(struct r300_context* r300, void* state);
+void r300_emit_rs_block_state(struct r300_context* r300,
+                              unsigned size, void* state);
 
-void r300_emit_scissor_state(struct r300_context* r300, void* state);
+void r300_emit_scissor_state(struct r300_context* r300,
+                             unsigned size, void* state);
 
-void r300_emit_texture(struct r300_context* r300,
-                       struct r300_sampler_state* sampler,
-                       struct r300_texture* tex,
-                       unsigned offset);
+void r300_emit_textures_state(struct r300_context *r300,
+                              unsigned size, void *state);
 
 void r300_emit_vertex_buffer(struct r300_context* r300);
 
-void r300_emit_vertex_format_state(struct r300_context* r300, void* state);
+void r300_emit_vertex_stream_state(struct r300_context* r300,
+                                   unsigned size, void* state);
 
-void r300_emit_vertex_program_code(struct r300_context* r300,
-                                   struct r300_vertex_program_code* code);
+void r300_emit_vap_output_state(struct r300_context* r300,
+                               unsigned size, void* state);
 
 void r300_emit_vs_constant_buffer(struct r300_context* r300,
                                   struct rc_constant_list* constants);
 
-void r300_emit_vertex_shader(struct r300_context* r300,
-                             struct r300_vertex_shader* vs);
+void r300_emit_vs_state(struct r300_context* r300, unsigned size, void* state);
 
-void r300_emit_viewport_state(struct r300_context* r300, void* state);
+void r300_emit_viewport_state(struct r300_context* r300,
+                              unsigned size, void* state);
 
-void r300_emit_texture_count(struct r300_context* r300);
+void r300_emit_ztop_state(struct r300_context* r300,
+                          unsigned size, void* state);
 
-void r300_emit_ztop_state(struct r300_context* r300, void* state);
+void r300_emit_pvs_flush(struct r300_context* r300, unsigned size, void* state);
 
-void r300_flush_textures(struct r300_context* r300);
+void r300_emit_texture_cache_inval(struct r300_context* r300, unsigned size, void* state);
+
+unsigned r300_get_num_dirty_dwords(struct r300_context *r300);
 
 /* Emit all dirty state. */
 void r300_emit_dirty_state(struct r300_context* r300);
 
-void r300_emit_buffer_validate(struct r300_context *r300);
+void r300_emit_buffer_validate(struct r300_context *r300,
+                               boolean do_validate_vertex_buffers,
+                               struct pipe_buffer *index_buffer);
 
 #endif /* R300_EMIT_H */
diff --git a/src/gallium/drivers/r300/r300_flush.c b/src/gallium/drivers/r300/r300_flush.c
index e37d3092703..70de152713d 100644
--- a/src/gallium/drivers/r300/r300_flush.c
+++ b/src/gallium/drivers/r300/r300_flush.c
@@ -61,6 +61,12 @@ static void r300_flush(struct pipe_context* pipe,
                 atom->dirty = TRUE;
             }
         }
+
+        /* Unmark HWTCL state for SWTCL. */
+        if (!r300_screen(pipe->screen)->caps->has_tcl) {
+            r300->vs_state.dirty = FALSE;
+            r300->dirty_state &= ~R300_NEW_VERTEX_SHADER_CONSTANTS;
+        }
     }
 
     /* reset flushed query */
diff --git a/src/gallium/drivers/r300/r300_fs.c b/src/gallium/drivers/r300/r300_fs.c
index ae4c62b2f1d..9e71e61c303 100644
--- a/src/gallium/drivers/r300/r300_fs.c
+++ b/src/gallium/drivers/r300/r300_fs.c
@@ -133,10 +133,13 @@ static void get_compare_state(
     struct r300_fragment_program_external_state* state,
     unsigned shadow_samplers)
 {
+    struct r300_textures_state *texstate =
+        (struct r300_textures_state*)r300->textures_state.state;
+
     memset(state, 0, sizeof(*state));
 
-    for (int i = 0; i < r300->sampler_count; i++) {
-        struct r300_sampler_state* s = r300->sampler_states[i];
+    for (int i = 0; i < texstate->sampler_count; i++) {
+        struct r300_sampler_state* s = texstate->sampler_states[i];
 
         if (s && s->state.compare_mode == PIPE_TEX_COMPARE_R_TO_TEXTURE) {
             /* XXX Gallium doesn't provide us with any information regarding
@@ -204,6 +207,7 @@ static void r300_translate_fragment_shader(
         DBG(r300, DBG_FP, "r300: Error compiling fragment program: %s\n",
             compiler.Base.ErrorMsg);
         assert(0);
+        abort();
     }
 
     /* And, finally... */
diff --git a/src/gallium/drivers/r300/r300_reg.h b/src/gallium/drivers/r300/r300_reg.h
index a049da69e2e..a249e8b36be 100644
--- a/src/gallium/drivers/r300/r300_reg.h
+++ b/src/gallium/drivers/r300/r300_reg.h
@@ -113,11 +113,14 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 
 	/* index size - when not set the indices are assumed to be 16 bit */
 #	define	R300_VAP_VF_CNTL__INDEX_SIZE_32bit              (1<<11)
+#       define R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS          (1<<14)
 	/* number of vertices */
 #	define	R300_VAP_VF_CNTL__NUM_VERTICES__SHIFT           16
 
 #define R500_VAP_INDEX_OFFSET		    0x208c
 
+#define R500_VAP_ALT_NUM_VERTICES                           0x2088
+
 #define R300_VAP_OUTPUT_VTX_FMT_0           0x2090
 #       define R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT     (1<<0)
 #       define R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT (1<<1)
diff --git a/src/gallium/drivers/r300/r300_render.c b/src/gallium/drivers/r300/r300_render.c
index cd4971ae136..9c001ae186d 100644
--- a/src/gallium/drivers/r300/r300_render.c
+++ b/src/gallium/drivers/r300/r300_render.c
@@ -42,6 +42,9 @@
 /* r300_render: Vertex and index buffer primitive emission. */
 #define R300_MAX_VBO_SIZE  (1024 * 1024)
 
+/* XXX The DRM rejects VAP_ALT_NUM_VERTICES.. */
+//#define ENABLE_ALT_NUM_VERTS
+
 uint32_t r300_translate_primitive(unsigned prim)
 {
     switch (prim) {
@@ -115,6 +118,18 @@ static uint32_t r300_provoking_vertex_fixes(struct r300_context *r300,
     return color_control;
 }
 
+/* Check if the requested number of dwords is available in the CS and
+ * if not, flush. Return TRUE if the flush occured. */
+static boolean r300_reserve_cs_space(struct r300_context *r300,
+                                     unsigned dwords)
+{
+    if (!r300->winsys->check_cs(r300->winsys, dwords)) {
+        r300->context.flush(&r300->context, 0, NULL);
+        return TRUE;
+    }
+    return FALSE;
+}
+
 static boolean immd_is_good_idea(struct r300_context *r300,
                                       unsigned count)
 {
@@ -128,8 +143,8 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
 {
     struct pipe_vertex_element* velem;
     struct pipe_vertex_buffer* vbuf;
-    unsigned vertex_element_count = r300->vertex_element_count;
-    unsigned i, v, vbi, dw, elem_offset;
+    unsigned vertex_element_count = r300->velems->count;
+    unsigned i, v, vbi, dw, elem_offset, dwords;
 
     /* Size of the vertex, in dwords. */
     unsigned vertex_size = 0;
@@ -151,7 +166,7 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
 
     /* Calculate the vertex size, offsets, strides etc. and map the buffers. */
     for (i = 0; i < vertex_element_count; i++) {
-        velem = &r300->vertex_element[i];
+        velem = &r300->velems->velem[i];
         offset[i] = velem->src_offset / 4;
         size[i] = util_format_get_blocksize(velem->src_format) / 4;
         vertex_size += size[i];
@@ -168,14 +183,19 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
         }
     }
 
+    dwords = 9 + count * vertex_size;
+
+    r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords);
+    r300_emit_buffer_validate(r300, FALSE, NULL);
     r300_emit_dirty_state(r300);
 
-    BEGIN_CS(10 + count * vertex_size);
+    BEGIN_CS(dwords);
     OUT_CS_REG(R300_GA_COLOR_CONTROL,
             r300_provoking_vertex_fixes(r300, mode));
     OUT_CS_REG(R300_VAP_VTX_SIZE, vertex_size);
-    OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0);
-    OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
+    OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
+    OUT_CS(count - 1);
+    OUT_CS(0);
     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_IMMD_2, count * vertex_size);
     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_EMBEDDED | (count << 16) |
             r300_translate_primitive(mode));
@@ -183,7 +203,7 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
     /* Emit vertices. */
     for (v = 0; v < count; v++) {
         for (i = 0; i < vertex_element_count; i++) {
-            velem = &r300->vertex_element[i];
+            velem = &r300->velems->velem[i];
             vbi = velem->vertex_buffer_index;
             elem_offset = offset[i] + stride[vbi] * (v + start);
 
@@ -196,7 +216,7 @@ static void r300_emit_draw_arrays_immediate(struct r300_context *r300,
 
     /* Unmap buffers. */
     for (i = 0; i < vertex_element_count; i++) {
-        vbi = r300->vertex_element[i].vertex_buffer_index;
+        vbi = r300->velems->velem[i].vertex_buffer_index;
 
         if (map[vbi]) {
             vbuf = &r300->vertex_buffer[vbi];
@@ -210,16 +230,29 @@ static void r300_emit_draw_arrays(struct r300_context *r300,
                                   unsigned mode,
                                   unsigned count)
 {
+#if defined(ENABLE_ALT_NUM_VERTS)
+    boolean alt_num_verts = count > 65535;
+#else
+    boolean alt_num_verts = FALSE;
+#endif
     CS_LOCALS(r300);
 
-    BEGIN_CS(8);
+    if (alt_num_verts) {
+        assert(count < (1 << 24));
+        BEGIN_CS(9);
+        OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
+    } else {
+        BEGIN_CS(7);
+    }
     OUT_CS_REG(R300_GA_COLOR_CONTROL,
             r300_provoking_vertex_fixes(r300, mode));
-    OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, 0);
-    OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, count - 1);
+    OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
+    OUT_CS(count - 1);
+    OUT_CS(0);
     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_VBUF_2, 0);
     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_VERTEX_LIST | (count << 16) |
-           r300_translate_primitive(mode));
+           r300_translate_primitive(mode) |
+           (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
     END_CS;
 }
 
@@ -234,72 +267,61 @@ static void r300_emit_draw_elements(struct r300_context *r300,
 {
     uint32_t count_dwords;
     uint32_t offset_dwords = indexSize * start / sizeof(uint32_t);
+#if defined(ENABLE_ALT_NUM_VERTS)
+    boolean alt_num_verts = count > 65535;
+#else
+    boolean alt_num_verts = FALSE;
+#endif
     CS_LOCALS(r300);
 
-    /* XXX most of these are stupid */
-    assert(indexSize == 4 || indexSize == 2);
     assert((start * indexSize)  % 4 == 0);
-    assert(offset_dwords == 0);
+    assert(count < (1 << 24));
+
+    DBG(r300, DBG_DRAW, "r300: Indexbuf of %u indices, min %u max %u\n",
+        count, minIndex, maxIndex);
 
-    BEGIN_CS(14);
+    maxIndex = MIN2(maxIndex, r300->vertex_buffer_max_index);
+
+    if (alt_num_verts) {
+        BEGIN_CS(15);
+        OUT_CS_REG(R500_VAP_ALT_NUM_VERTICES, count);
+    } else {
+        BEGIN_CS(13);
+    }
     OUT_CS_REG(R300_GA_COLOR_CONTROL,
             r300_provoking_vertex_fixes(r300, mode));
-    OUT_CS_REG(R300_VAP_VF_MIN_VTX_INDX, minIndex);
-    OUT_CS_REG(R300_VAP_VF_MAX_VTX_INDX, maxIndex);
+    OUT_CS_REG_SEQ(R300_VAP_VF_MAX_VTX_INDX, 2);
+    OUT_CS(maxIndex);
+    OUT_CS(minIndex);
     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, 0);
     if (indexSize == 4) {
-        count_dwords = count + start;
+        count_dwords = count;
         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
                R300_VAP_VF_CNTL__INDEX_SIZE_32bit |
-               r300_translate_primitive(mode));
+               r300_translate_primitive(mode) |
+               (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
     } else {
-        count_dwords = (count + start + 1) / 2;
+        count_dwords = (count + 1) / 2;
         OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
-               r300_translate_primitive(mode));
+               r300_translate_primitive(mode) |
+               (alt_num_verts ? R500_VAP_VF_CNTL__USE_ALT_NUM_VERTS : 0));
     }
 
     /* INDX_BUFFER is a truly special packet3.
      * Unlike most other packet3, where the offset is after the count,
      * the order is reversed, so the relocation ends up carrying the
      * size of the indexbuf instead of the offset.
-     *
-     * XXX Fix offset
      */
     OUT_CS_PKT3(R300_PACKET3_INDX_BUFFER, 2);
     OUT_CS(R300_INDX_BUFFER_ONE_REG_WR | (R300_VAP_PORT_IDX0 >> 2) |
            (0 << R300_INDX_BUFFER_SKIP_SHIFT));
-    OUT_CS(offset_dwords);
+    OUT_CS(offset_dwords << 2);
     OUT_CS_RELOC(indexBuffer, count_dwords,
         RADEON_GEM_DOMAIN_GTT, 0, 0);
 
     END_CS;
 }
 
-static boolean r300_setup_vertex_buffers(struct r300_context *r300)
-{
-    struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
-    struct pipe_vertex_element *velem = r300->vertex_element;
-    struct pipe_buffer *pbuf;
-
-validate:
-    for (int i = 0; i < r300->vertex_element_count; i++) {
-        pbuf = vbuf[velem[i].vertex_buffer_index].buffer;
-
-        if (!r300->winsys->add_buffer(r300->winsys, pbuf,
-                                      RADEON_GEM_DOMAIN_GTT, 0)) {
-            r300->context.flush(&r300->context, 0, NULL);
-            goto validate;
-        }
-    }
-
-    if (!r300->winsys->validate(r300->winsys)) {
-        r300->context.flush(&r300->context, 0, NULL);
-        return r300->winsys->validate(r300->winsys);
-    }
-
-    return TRUE;
-}
-
 static void r300_shorten_ubyte_elts(struct r300_context* r300,
                                     struct pipe_buffer** elts,
                                     unsigned count)
@@ -343,48 +365,52 @@ void r300_draw_range_elements(struct pipe_context* pipe,
 {
     struct r300_context* r300 = r300_context(pipe);
     struct pipe_buffer* orgIndexBuffer = indexBuffer;
+#if defined(ENABLE_ALT_NUM_VERTS)
+    boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 &&
+                            count > 65536;
+#else
+    boolean alt_num_verts = FALSE;
+#endif
+    unsigned short_count;
 
     if (!u_trim_pipe_prim(mode, &count)) {
         return;
     }
 
-    if (count > 65535) {
-       /* XXX: use aux/indices functions to split this into smaller
-        * primitives.
-        */
-        return;
-    }
-
-    r300_update_derived_state(r300);
-
-    r300_emit_buffer_validate(r300);
-
-    if (!r300_setup_vertex_buffers(r300)) {
-        return;
-    }
-
     if (indexSize == 1) {
         r300_shorten_ubyte_elts(r300, &indexBuffer, count);
         indexSize = 2;
     }
 
-    if (!r300->winsys->add_buffer(r300->winsys, indexBuffer,
-                                  RADEON_GEM_DOMAIN_GTT, 0)) {
-        goto cleanup;
-    }
-
-    if (!r300->winsys->validate(r300->winsys)) {
-        goto cleanup;
-    }
+    r300_update_derived_state(r300);
 
+    /* 128 dwords for emit_aos and emit_draw_elements */
+    r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128);
+    r300_emit_buffer_validate(r300, TRUE, indexBuffer);
     r300_emit_dirty_state(r300);
-
     r300_emit_aos(r300, 0);
 
-    r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex, maxIndex,
-                            mode, start, count);
+    if (alt_num_verts || count <= 65535) {
+        r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
+                                maxIndex, mode, start, count);
+    } else {
+        do {
+            short_count = MIN2(count, 65534);
+            r300_emit_draw_elements(r300, indexBuffer, indexSize, minIndex,
+                                    maxIndex, mode, start, short_count);
+
+            start += short_count;
+            count -= short_count;
+
+            /* 16 spare dwords are enough for emit_draw_elements. */
+            if (count && r300_reserve_cs_space(r300, 16)) {
+                r300_emit_buffer_validate(r300, TRUE, indexBuffer);
+                r300_emit_dirty_state(r300);
+                r300_emit_aos(r300, 0);
+            }
+        } while (count);
+    }
 
-cleanup:
     if (indexBuffer != orgIndexBuffer) {
         pipe->screen->buffer_destroy(indexBuffer);
     }
@@ -396,40 +422,60 @@ void r300_draw_elements(struct pipe_context* pipe,
                         unsigned indexSize, unsigned mode,
                         unsigned start, unsigned count)
 {
-   pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0, ~0,
-                             mode, start, count);
+    struct r300_context *r300 = r300_context(pipe);
+
+    pipe->draw_range_elements(pipe, indexBuffer, indexSize, 0,
+                              r300->vertex_buffer_max_index,
+                              mode, start, count);
 }
 
 void r300_draw_arrays(struct pipe_context* pipe, unsigned mode,
                       unsigned start, unsigned count)
 {
     struct r300_context* r300 = r300_context(pipe);
+#if defined(ENABLE_ALT_NUM_VERTS)
+    boolean alt_num_verts = r300_screen(pipe->screen)->caps->is_r500 &&
+                            count > 65536;
+#else
+    boolean alt_num_verts = FALSE;
+#endif
+    unsigned short_count;
 
     if (!u_trim_pipe_prim(mode, &count)) {
         return;
     }
 
-    if (count > 65535) {
-        /* XXX: driver needs to handle this -- use the functions in
-         * aux/indices to split this into several smaller primitives.
-         */
-        return;
-    }
-
     r300_update_derived_state(r300);
 
-    r300_emit_buffer_validate(r300);
-
     if (immd_is_good_idea(r300, count)) {
         r300_emit_draw_arrays_immediate(r300, mode, start, count);
     } else {
-        if (!r300_setup_vertex_buffers(r300)) {
-            return;
-        }
-
+        /* Make sure there are at least 128 spare dwords in the command buffer.
+         * (most of it being consumed by emit_aos) */
+        r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 128);
+        r300_emit_buffer_validate(r300, TRUE, NULL);
         r300_emit_dirty_state(r300);
-        r300_emit_aos(r300, start);
-        r300_emit_draw_arrays(r300, mode, count);
+
+        if (alt_num_verts || count <= 65535) {
+            r300_emit_aos(r300, start);
+            r300_emit_draw_arrays(r300, mode, count);
+        } else {
+            do {
+                short_count = MIN2(count, 65535);
+                r300_emit_aos(r300, start);
+                r300_emit_draw_arrays(r300, mode, short_count);
+
+                start += short_count;
+                count -= short_count;
+
+                /* Again, we emit both AOS and draw_arrays so there should be
+                 * at least 128 spare dwords. */
+                if (count && r300_reserve_cs_space(r300, 128)) {
+                    r300_emit_buffer_validate(r300, TRUE, NULL);
+                    r300_emit_dirty_state(r300);
+                }
+            } while (count);
+        }
     }
 }
 
@@ -559,7 +605,7 @@ r300_render_get_vertex_info(struct vbuf_render* render)
 
     r300_update_derived_state(r300);
 
-    return (struct vertex_info*)r300->vertex_format_state.state;
+    return &r300->vertex_info;
 }
 
 static boolean r300_render_allocate_vertices(struct vbuf_render* render,
@@ -644,6 +690,8 @@ static void r300_render_draw_arrays(struct vbuf_render* render,
 
     CS_LOCALS(r300);
 
+    r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + 2);
+    r300_emit_buffer_validate(r300, FALSE, NULL);
     r300_emit_dirty_state(r300);
 
     DBG(r300, DBG_DRAW, "r300: Doing vbuf render, count %d\n", count);
@@ -662,12 +710,15 @@ static void r300_render_draw(struct vbuf_render* render,
     struct r300_render* r300render = r300_render(render);
     struct r300_context* r300 = r300render->r300;
     int i;
+    unsigned dwords = 2 + (count+1)/2;
 
     CS_LOCALS(r300);
 
+    r300_reserve_cs_space(r300, r300_get_num_dirty_dwords(r300) + dwords);
+    r300_emit_buffer_validate(r300, FALSE, NULL);
     r300_emit_dirty_state(r300);
 
-    BEGIN_CS(2 + (count+1)/2);
+    BEGIN_CS(dwords);
     OUT_CS_PKT3(R300_PACKET3_3D_DRAW_INDX_2, (count+1)/2);
     OUT_CS(R300_VAP_VF_CNTL__PRIM_WALK_INDICES | (count << 16) |
            r300render->hwprim);
diff --git a/src/gallium/drivers/r300/r300_screen.c b/src/gallium/drivers/r300/r300_screen.c
index 5fa1ea0850e..5880eecd5fe 100644
--- a/src/gallium/drivers/r300/r300_screen.c
+++ b/src/gallium/drivers/r300/r300_screen.c
@@ -1,5 +1,6 @@
 /*
  * Copyright 2008 Corbin Simpson <[email protected]>
+ * Copyright 2010 Marek Olšák <[email protected]>
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -20,14 +21,13 @@
  * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
  * USE OR OTHER DEALINGS IN THE SOFTWARE. */
 
-#include "util/u_inlines.h"
 #include "util/u_format.h"
 #include "util/u_memory.h"
 #include "util/u_simple_screen.h"
 
 #include "r300_context.h"
-#include "r300_screen.h"
 #include "r300_texture.h"
+#include "r300_transfer.h"
 
 #include "radeon_winsys.h"
 #include "r300_winsys.h"
@@ -210,8 +210,9 @@ static boolean r300_is_format_supported(struct pipe_screen* screen,
 {
     uint32_t retval = 0;
     boolean is_r500 = r300_screen(screen)->caps->is_r500;
-    boolean is_z24 = format == PIPE_FORMAT_Z24X8_UNORM ||
-                     format == PIPE_FORMAT_Z24S8_UNORM;
+    boolean is_z24 = format == PIPE_FORMAT_X8Z24_UNORM ||
+                     format == PIPE_FORMAT_S8Z24_UNORM;
+    boolean is_color2101010 = format == PIPE_FORMAT_R10G10B10A2_UNORM;
 
     if (target >= PIPE_MAX_TEXTURE_TYPES) {
         debug_printf("r300: Implementation error: Received bogus texture "
@@ -221,114 +222,34 @@ static boolean r300_is_format_supported(struct pipe_screen* screen,
 
     /* Check sampler format support. */
     if ((usage & PIPE_TEXTURE_USAGE_SAMPLER) &&
-        (is_r500 || !is_z24) && /* Z24 cannot be sampled from on non-r5xx. */
-        r300_translate_texformat(format) != ~0) {
+        /* Z24 cannot be sampled from on non-r5xx. */
+        (is_r500 || !is_z24) &&
+        r300_is_sampler_format_supported(format)) {
         retval |= PIPE_TEXTURE_USAGE_SAMPLER;
     }
 
-    switch (format) {
-        /* Supported formats. */
-        /* Colorbuffer */
-        case PIPE_FORMAT_A8_UNORM:
-        case PIPE_FORMAT_L8_UNORM:
-        case PIPE_FORMAT_R5G6B5_UNORM:
-        case PIPE_FORMAT_A1R5G5B5_UNORM:
-        case PIPE_FORMAT_A4R4G4B4_UNORM:
-        case PIPE_FORMAT_A8R8G8B8_UNORM:
-        case PIPE_FORMAT_X8R8G8B8_UNORM:
-        case PIPE_FORMAT_R8G8B8A8_UNORM:
-        case PIPE_FORMAT_R8G8B8X8_UNORM:
-        case PIPE_FORMAT_I8_UNORM:
-            retval |= usage &
-                (PIPE_TEXTURE_USAGE_RENDER_TARGET |
-                 PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
-                 PIPE_TEXTURE_USAGE_PRIMARY);
-            break;
-
-        /* ZS buffer */
-        case PIPE_FORMAT_Z16_UNORM:
-        case PIPE_FORMAT_Z24X8_UNORM:
-        case PIPE_FORMAT_Z24S8_UNORM:
-            retval = usage & PIPE_TEXTURE_USAGE_DEPTH_STENCIL;
-            break;
-
-        /* XXX Add all remaining gallium-supported formats,
-         * see util/u_format.csv. */
-
-        default:;
+    /* Check colorbuffer format support. */
+    if ((usage & (PIPE_TEXTURE_USAGE_RENDER_TARGET |
+                  PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
+                  PIPE_TEXTURE_USAGE_SCANOUT |
+                  PIPE_TEXTURE_USAGE_SHARED)) &&
+        /* 2101010 cannot be rendered to on non-r5xx. */
+        (is_r500 || !is_color2101010) &&
+        r300_is_colorbuffer_format_supported(format)) {
+        retval |= usage &
+            (PIPE_TEXTURE_USAGE_RENDER_TARGET |
+             PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
+             PIPE_TEXTURE_USAGE_SCANOUT |
+             PIPE_TEXTURE_USAGE_SHARED);
     }
 
-    /* If usage was a mask that contained multiple bits, and not all of them
-     * are supported, this will catch that and return FALSE.
-     * e.g. usage = 2 | 4; retval = 4; (retval >= usage) == FALSE
-     *
-     * This also returns FALSE for any unknown formats.
-     */
-    return (retval >= usage);
-}
-
-static struct pipe_transfer*
-r300_get_tex_transfer(struct pipe_screen *screen,
-                      struct pipe_texture *texture,
-                      unsigned face, unsigned level, unsigned zslice,
-                      enum pipe_transfer_usage usage, unsigned x, unsigned y,
-                      unsigned w, unsigned h)
-{
-    struct r300_texture *tex = (struct r300_texture *)texture;
-    struct r300_transfer *trans;
-    struct r300_screen *rscreen = r300_screen(screen);
-    unsigned offset;
-
-    offset = r300_texture_get_offset(tex, level, zslice, face);  /* in bytes */
-
-    trans = CALLOC_STRUCT(r300_transfer);
-    if (trans) {
-        pipe_texture_reference(&trans->transfer.texture, texture);
-        trans->transfer.x = x;
-        trans->transfer.y = y;
-        trans->transfer.width = w;
-        trans->transfer.height = h;
-        trans->transfer.stride = r300_texture_get_stride(rscreen, tex, level);
-        trans->transfer.usage = usage;
-        trans->transfer.zslice = zslice;
-        trans->transfer.face = face;
-
-        trans->offset = offset;
+    /* Check depth-stencil format support. */
+    if (usage & PIPE_TEXTURE_USAGE_DEPTH_STENCIL &&
+        r300_is_zs_format_supported(format)) {
+        retval |= PIPE_TEXTURE_USAGE_DEPTH_STENCIL;
     }
-    return &trans->transfer;
-}
 
-static void
-r300_tex_transfer_destroy(struct pipe_transfer *trans)
-{
-   pipe_texture_reference(&trans->texture, NULL);
-   FREE(trans);
-}
-
-static void* r300_transfer_map(struct pipe_screen* screen,
-                              struct pipe_transfer* transfer)
-{
-    struct r300_texture* tex = (struct r300_texture*)transfer->texture;
-    char* map;
-    enum pipe_format format = tex->tex.format;
-
-    map = pipe_buffer_map(screen, tex->buffer,
-                          pipe_transfer_buffer_flags(transfer));
-
-    if (!map) {
-        return NULL;
-    }
-
-    return map + r300_transfer(transfer)->offset +
-        transfer->y / util_format_get_blockheight(format) * transfer->stride +
-        transfer->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
-}
-
-static void r300_transfer_unmap(struct pipe_screen* screen,
-                                struct pipe_transfer* transfer)
-{
-    struct r300_texture* tex = (struct r300_texture*)transfer->texture;
-    pipe_buffer_unmap(screen, tex->buffer);
+    return retval == usage;
 }
 
 static void r300_destroy_screen(struct pipe_screen* pscreen)
@@ -367,13 +288,11 @@ struct pipe_screen* r300_create_screen(struct radeon_winsys* radeon_winsys)
     r300screen->screen.get_paramf = r300_get_paramf;
     r300screen->screen.is_format_supported = r300_is_format_supported;
     r300screen->screen.context_create = r300_create_context;
-    r300screen->screen.get_tex_transfer = r300_get_tex_transfer;
-    r300screen->screen.tex_transfer_destroy = r300_tex_transfer_destroy;
-    r300screen->screen.transfer_map = r300_transfer_map;
-    r300screen->screen.transfer_unmap = r300_transfer_unmap;
 
     r300_init_screen_texture_functions(&r300screen->screen);
+    r300_init_screen_transfer_functions(&r300screen->screen);
     u_simple_screen_init(&r300screen->screen);
 
     return &r300screen->screen;
 }
+
diff --git a/src/gallium/drivers/r300/r300_screen.h b/src/gallium/drivers/r300/r300_screen.h
index 502fbfa5a24..484bde6a6be 100644
--- a/src/gallium/drivers/r300/r300_screen.h
+++ b/src/gallium/drivers/r300/r300_screen.h
@@ -1,5 +1,6 @@
 /*
  * Copyright 2008 Corbin Simpson <[email protected]>
+ * Copyright 2010 Marek Olšák <[email protected]>
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -27,6 +28,8 @@
 
 #include "r300_chipset.h"
 
+#define R300_TEXTURE_USAGE_TRANSFER PIPE_TEXTURE_USAGE_CUSTOM
+
 struct radeon_winsys;
 
 struct r300_screen {
@@ -35,6 +38,10 @@ struct r300_screen {
 
     struct radeon_winsys* radeon_winsys;
 
+    /* XXX This hack will be removed once texture transfers become part of
+     * pipe_context. */
+    struct pipe_context* ctx;
+
     /* Chipset capabilities */
     struct r300_capabilities* caps;
 
@@ -42,25 +49,14 @@ struct r300_screen {
     unsigned debug;
 };
 
-struct r300_transfer {
-    /* Parent class */
-    struct pipe_transfer transfer;
-
-    /* Offset from start of buffer. */
-    unsigned offset;
-};
 
 /* Convenience cast wrapper. */
 static INLINE struct r300_screen* r300_screen(struct pipe_screen* screen) {
     return (struct r300_screen*)screen;
 }
 
-/* Convenience cast wrapper. */
-static INLINE struct r300_transfer*
-r300_transfer(struct pipe_transfer* transfer)
-{
-    return (struct r300_transfer*)transfer;
-}
+/* Creates a new r300 screen. */
+struct pipe_screen* r300_create_screen(struct radeon_winsys* radeon_winsys);
 
 /* Debug functionality. */
 
diff --git a/src/gallium/drivers/r300/r300_state.c b/src/gallium/drivers/r300/r300_state.c
index e5ec2701c81..09bbf6c60e2 100644
--- a/src/gallium/drivers/r300/r300_state.c
+++ b/src/gallium/drivers/r300/r300_state.c
@@ -43,6 +43,12 @@
 /* r300_state: Functions used to intialize state context by translating
  * Gallium state objects into semi-native r300 state objects. */
 
+#define UPDATE_STATE(cso, atom) \
+    if (cso != atom.state) { \
+        atom.state = cso;    \
+        atom.dirty = TRUE;   \
+    }
+
 static boolean blend_discard_if_src_alpha_0(unsigned srcRGB, unsigned srcA,
                                             unsigned dstRGB, unsigned dstA)
 {
@@ -328,8 +334,7 @@ static void r300_bind_blend_state(struct pipe_context* pipe,
 {
     struct r300_context* r300 = r300_context(pipe);
 
-    r300->blend_state.state = state;
-    r300->blend_state.dirty = TRUE;
+    UPDATE_STATE(state, r300->blend_state);
 }
 
 /* Free blend state. */
@@ -356,7 +361,7 @@ static void r300_set_blend_color(struct pipe_context* pipe,
         (struct r300_blend_color_state*)r300->blend_color_state.state;
     union util_color uc;
 
-    util_pack_color(color->color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc);
+    util_pack_color(color->color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
     state->blend_color = uc.ui;
 
     /* XXX if FP16 blending is enabled, we should use the FP16 format */
@@ -376,6 +381,8 @@ static void r300_set_clip_state(struct pipe_context* pipe,
 {
     struct r300_context* r300 = r300_context(pipe);
 
+    r300->clip = *state;
+
     if (r300_screen(pipe->screen)->caps->has_tcl) {
         memcpy(r300->clip_state.state, state, sizeof(struct pipe_clip_state));
         r300->clip_state.size = 29;
@@ -476,11 +483,8 @@ static void r300_bind_dsa_state(struct pipe_context* pipe,
                                 void* state)
 {
     struct r300_context* r300 = r300_context(pipe);
-    struct r300_screen* r300screen = r300_screen(pipe->screen);
 
-    r300->dsa_state.state = state;
-    r300->dsa_state.size = r300screen->caps->is_r500 ? 8 : 6;
-    r300->dsa_state.dirty = TRUE;
+    UPDATE_STATE(state, r300->dsa_state);
 }
 
 /* Free DSA state. */
@@ -567,6 +571,7 @@ static void
 {
     struct r300_context* r300 = r300_context(pipe);
     struct r300_screen* r300screen = r300_screen(pipe->screen);
+    struct pipe_framebuffer_state *old_state = r300->fb_state.state;
     unsigned max_width, max_height;
     uint32_t zbuffer_bpp = 0;
 
@@ -591,22 +596,30 @@ static void
         return;
     }
 
-
     if (r300->draw) {
         draw_flush(r300->draw);
     }
 
-    memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state));
+    r300->fb_state.dirty = TRUE;
 
-    r300->fb_state.size = (10 * state->nr_cbufs) + (state->zsbuf ? 10 : 0) + 6;
+    /* If nr_cbufs is changed from zero to non-zero or vice versa... */
+    if (!!old_state->nr_cbufs != !!state->nr_cbufs) {
+        r300->blend_state.dirty = TRUE;
+    }
+    /* If zsbuf is set from NULL to non-NULL or vice versa.. */
+    if (!!old_state->zsbuf != !!state->zsbuf) {
+        r300->dsa_state.dirty = TRUE;
+    }
+    if (!r300->scissor_enabled) {
+        r300->scissor_state.dirty = TRUE;
+    }
 
     r300_fb_update_tiling_flags(r300, r300->fb_state.state, state);
 
-    /* XXX wait what */
-    r300->blend_state.dirty = TRUE;
-    r300->dsa_state.dirty = TRUE;
-    r300->fb_state.dirty = TRUE;
-    r300->scissor_state.dirty = TRUE;
+    memcpy(r300->fb_state.state, state, sizeof(struct pipe_framebuffer_state));
+
+    r300->fb_state.size = (10 * state->nr_cbufs) + (2 * (4 - state->nr_cbufs)) +
+                          (state->zsbuf ? 10 : 0) + 8;
 
     /* Polygon offset depends on the zbuffer bit depth. */
     if (state->zsbuf && r300->polygon_offset_enabled) {
@@ -658,8 +671,10 @@ static void r300_bind_fs_state(struct pipe_context* pipe, void* shader)
     r300->fs = fs;
     r300_pick_fragment_shader(r300);
 
-    if (r300->vs && r300_vertex_shader_setup_wpos(r300)) {
-        r300->vertex_format_state.dirty = TRUE;
+    r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
+
+    if (r300->vs_state.state && r300_vertex_shader_setup_wpos(r300)) {
+        r300->vap_output_state.dirty = TRUE;
     }
 
     r300->dirty_state |= R300_NEW_FRAGMENT_SHADER | R300_NEW_FRAGMENT_SHADER_CONSTANTS;
@@ -709,32 +724,14 @@ static void* r300_create_rs_state(struct pipe_context* pipe,
     rs->vap_control_status = R300_VC_32BIT_SWAP;
 #endif
 
-    /* If bypassing TCL, or if no TCL engine is present, turn off the HW TCL.
-     * Else, enable HW TCL and force Draw's TCL off. */
-    if (state->bypass_vs_clip_and_viewport ||
-            !r300screen->caps->has_tcl) {
+    /* If no TCL engine is present, turn off the HW TCL. */
+    if (!r300screen->caps->has_tcl) {
         rs->vap_control_status |= R300_VAP_TCL_BYPASS;
     }
 
     rs->point_size = pack_float_16_6x(state->point_size) |
         (pack_float_16_6x(state->point_size) << R300_POINTSIZE_X_SHIFT);
 
-        /* Point minimum and maximum sizes. This register has to be emitted,
-         * and it'd be a step backwards to put it in invariant state. */
-        if (r300screen->caps->is_r500) {
-            rs->point_minmax =
-            ((int)(0.0 * 6.0) << R300_GA_POINT_MINMAX_MIN_SHIFT) |
-            ((int)(4096.0 * 6.0) << R300_GA_POINT_MINMAX_MAX_SHIFT);
-        } else if (r300screen->caps->is_r400) {
-            rs->point_minmax =
-            ((int)(0.0 * 6.0) << R300_GA_POINT_MINMAX_MIN_SHIFT) |
-            ((int)(4021.0 * 6.0) << R300_GA_POINT_MINMAX_MAX_SHIFT);
-        } else {
-            rs->point_minmax =
-            ((int)(0.0 * 6.0) << R300_GA_POINT_MINMAX_MIN_SHIFT) |
-            ((int)(2560.0 * 6.0) << R300_GA_POINT_MINMAX_MAX_SHIFT);
-        }
-
     rs->line_control = pack_float_16_6x(state->line_width) |
         R300_GA_LINE_CNTL_END_TYPE_COMP;
 
@@ -817,6 +814,7 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
 {
     struct r300_context* r300 = r300_context(pipe);
     struct r300_rs_state* rs = (struct r300_rs_state*)state;
+    boolean scissor_was_enabled = r300->scissor_enabled;
 
     if (r300->draw) {
         draw_flush(r300->draw);
@@ -824,22 +822,18 @@ static void r300_bind_rs_state(struct pipe_context* pipe, void* state)
     }
 
     if (rs) {
-        r300->tcl_bypass = rs->rs.bypass_vs_clip_and_viewport;
         r300->polygon_offset_enabled = rs->rs.offset_cw || rs->rs.offset_ccw;
+        r300->scissor_enabled = rs->rs.scissor;
     } else {
-        r300->tcl_bypass = FALSE;
         r300->polygon_offset_enabled = FALSE;
+        r300->scissor_enabled = FALSE;
     }
 
-    r300->rs_state.state = rs;
-    r300->rs_state.dirty = TRUE;
-    /* XXX Why is this still needed, dammit!? */
-    r300->scissor_state.dirty = TRUE;
-    r300->viewport_state.dirty = TRUE;
+    UPDATE_STATE(state, r300->rs_state);
+    r300->rs_state.size = 17 + (r300->polygon_offset_enabled ? 5 : 0);
 
-    /* XXX Clean these up when we move to atom emits */
-    if (r300->fs && r300->fs->inputs.wpos != ATTR_UNUSED) {
-        r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
+    if (scissor_was_enabled != r300->scissor_enabled) {
+        r300->scissor_state.dirty = TRUE;
     }
 }
 
@@ -871,7 +865,7 @@ static void*
                                                    state->max_anisotropy > 0);
 
     /* Unfortunately, r300-r500 don't support floating-point mipmap lods. */
-    /* We must pass these to the emit function to clamp them properly. */
+    /* We must pass these to the merge function to clamp them properly. */
     sampler->min_lod = MAX2((unsigned)state->min_lod, 0);
     sampler->max_lod = MAX2((unsigned)ceilf(state->max_lod), 0);
 
@@ -881,7 +875,7 @@ static void*
 
     sampler->filter1 |= r300_anisotropy(state->max_anisotropy);
 
-    util_pack_color(state->border_color, PIPE_FORMAT_A8R8G8B8_UNORM, &uc);
+    util_pack_color(state->border_color, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
     sampler->border_color = uc.ui;
 
     /* R500-specific fixups and optimizations */
@@ -897,23 +891,20 @@ static void r300_bind_sampler_states(struct pipe_context* pipe,
                                      void** states)
 {
     struct r300_context* r300 = r300_context(pipe);
-    int i;
+    struct r300_textures_state* state =
+        (struct r300_textures_state*)r300->textures_state.state;
 
     if (count > 8) {
         return;
     }
 
-    for (i = 0; i < count; i++) {
-        if (r300->sampler_states[i] != states[i]) {
-            r300->sampler_states[i] = (struct r300_sampler_state*)states[i];
-            r300->dirty_state |= (R300_NEW_SAMPLER << i);
-        }
-    }
+    memcpy(state->sampler_states, states, sizeof(void*) * count);
+    state->sampler_count = count;
 
-    r300->sampler_count = count;
+    r300->textures_state.dirty = TRUE;
 
     /* Pick a fragment shader based on the texture compare state. */
-    if (r300->fs && (r300->dirty_state & R300_ANY_NEW_SAMPLERS)) {
+    if (r300->fs && count) {
         if (r300_pick_fragment_shader(r300)) {
             r300->dirty_state |= R300_NEW_FRAGMENT_SHADER |
                                  R300_NEW_FRAGMENT_SHADER_CONSTANTS;
@@ -937,21 +928,22 @@ static void r300_set_fragment_sampler_views(struct pipe_context* pipe,
                                             struct pipe_sampler_view** views)
 {
     struct r300_context* r300 = r300_context(pipe);
+    unsigned i;
     boolean is_r500 = r300_screen(r300->context.screen)->caps->is_r500;
-    int i;
+    boolean dirty_tex = FALSE;
 
     /* XXX magic num */
     if (count > 8) {
         return;
     }
-    
+
     for (i = 0; i < count; i++) {
         if (r300->fragment_sampler_views[i] != views[i]) {
             struct r300_texture *texture;
 
             pipe_sampler_view_reference(&r300->fragment_sampler_views[i],
                 views[i]);
-            r300->dirty_state |= (R300_NEW_TEXTURE << i);
+            dirty_tex = TRUE;
 
             texture = (struct r300_texture *)views[i]->texture;
 
@@ -968,11 +960,16 @@ static void r300_set_fragment_sampler_views(struct pipe_context* pipe,
         if (r300->fragment_sampler_views[i]) {
             pipe_sampler_view_reference(&r300->fragment_sampler_views[i],
                 NULL);
-            r300->dirty_state |= (R300_NEW_TEXTURE << i);
         }
     }
 
     r300->fragment_sampler_view_count = count;
+
+    r300->textures_state.dirty = TRUE;
+
+    if (dirty_tex) {
+        r300->texture_cache_inval.dirty = TRUE;
+    }
 }
 
 static struct pipe_sampler_view *
@@ -1009,7 +1006,9 @@ static void r300_set_scissor_state(struct pipe_context* pipe,
     memcpy(r300->scissor_state.state, state,
         sizeof(struct pipe_scissor_state));
 
-    r300->scissor_state.dirty = TRUE;
+    if (r300->scissor_enabled) {
+        r300->scissor_state.dirty = TRUE;
+    }
 }
 
 static void r300_set_viewport_state(struct pipe_context* pipe,
@@ -1019,6 +1018,8 @@ static void r300_set_viewport_state(struct pipe_context* pipe,
     struct r300_viewport_state* viewport =
         (struct r300_viewport_state*)r300->viewport_state.state;
 
+    r300->viewport = *state;
+
     /* Do the transform in HW. */
     viewport->vte_control = R300_VTX_W0_FMT;
 
@@ -1058,27 +1059,32 @@ static void r300_set_vertex_buffers(struct pipe_context* pipe,
                                     const struct pipe_vertex_buffer* buffers)
 {
     struct r300_context* r300 = r300_context(pipe);
+    unsigned i, max_index = ~0;
 
     memcpy(r300->vertex_buffer, buffers,
         sizeof(struct pipe_vertex_buffer) * count);
+
+    for (i = 0; i < count; i++) {
+        max_index = MIN2(buffers[i].max_index, max_index);
+    }
+
     r300->vertex_buffer_count = count;
+    r300->vertex_buffer_max_index = max_index;
 
     if (r300->draw) {
         draw_flush(r300->draw);
         draw_set_vertex_buffers(r300->draw, count, buffers);
     }
-
-    r300->vertex_format_state.dirty = TRUE;
 }
 
 static boolean r300_validate_aos(struct r300_context *r300)
 {
     struct pipe_vertex_buffer *vbuf = r300->vertex_buffer;
-    struct pipe_vertex_element *velem = r300->vertex_element;
+    struct pipe_vertex_element *velem = r300->velems->velem;
     int i;
 
     /* Check if formats and strides are aligned to the size of DWORD. */
-    for (i = 0; i < r300->vertex_element_count; i++) {
+    for (i = 0; i < r300->velems->count; i++) {
         if (vbuf[velem[i].vertex_buffer_index].stride % 4 != 0 ||
             util_format_get_blocksize(velem[i].src_format) % 4 != 0) {
             return FALSE;
@@ -1087,20 +1093,209 @@ static boolean r300_validate_aos(struct r300_context *r300)
     return TRUE;
 }
 
-static void r300_set_vertex_elements(struct pipe_context* pipe,
-                                    unsigned count,
-                                    const struct pipe_vertex_element* elements)
+static void r300_draw_emit_attrib(struct r300_context* r300,
+                                  enum attrib_emit emit,
+                                  enum interp_mode interp,
+                                  int index)
 {
-    struct r300_context* r300 = r300_context(pipe);
+    struct r300_vertex_shader* vs = r300->vs_state.state;
+    struct tgsi_shader_info* info = &vs->info;
+    int output;
+
+    output = draw_find_shader_output(r300->draw,
+                                     info->output_semantic_name[index],
+                                     info->output_semantic_index[index]);
+    draw_emit_vertex_attr(&r300->vertex_info, emit, interp, output);
+}
+
+static void r300_draw_emit_all_attribs(struct r300_context* r300)
+{
+    struct r300_vertex_shader* vs = r300->vs_state.state;
+    struct r300_shader_semantics* vs_outputs = &vs->outputs;
+    int i, gen_count;
+
+    /* Position. */
+    if (vs_outputs->pos != ATTR_UNUSED) {
+        r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
+                              vs_outputs->pos);
+    } else {
+        assert(0);
+    }
 
-    memcpy(r300->vertex_element,
-           elements,
-           sizeof(struct pipe_vertex_element) * count);
-    r300->vertex_element_count = count;
+    /* Point size. */
+    if (vs_outputs->psize != ATTR_UNUSED) {
+        r300_draw_emit_attrib(r300, EMIT_1F_PSIZE, INTERP_POS,
+                              vs_outputs->psize);
+    }
+
+    /* Colors. */
+    for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+        if (vs_outputs->color[i] != ATTR_UNUSED) {
+            r300_draw_emit_attrib(r300, EMIT_4F, INTERP_LINEAR,
+                                  vs_outputs->color[i]);
+        }
+    }
+
+    /* XXX Back-face colors. */
+
+    /* Texture coordinates. */
+    gen_count = 0;
+    for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
+        if (vs_outputs->generic[i] != ATTR_UNUSED) {
+            r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
+                                  vs_outputs->generic[i]);
+            gen_count++;
+        }
+    }
+
+    /* Fog coordinates. */
+    if (vs_outputs->fog != ATTR_UNUSED) {
+        r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
+                              vs_outputs->fog);
+        gen_count++;
+    }
+
+    /* XXX magic */
+    assert(gen_count <= 8);
+}
+
+/* Update the PSC tables. */
+static void r300_vertex_psc(struct r300_vertex_element_state *velems)
+{
+    struct r300_vertex_stream_state *vstream = &velems->vertex_stream;
+    uint16_t type, swizzle;
+    enum pipe_format format;
+    unsigned i;
+
+    assert(velems->count <= 16);
+
+    /* Vertex shaders have no semantics on their inputs,
+     * so PSC should just route stuff based on the vertex elements,
+     * and not on attrib information. */
+    for (i = 0; i < velems->count; i++) {
+        format = velems->velem[i].src_format;
+
+        type = r300_translate_vertex_data_type(format) |
+            (i << R300_DST_VEC_LOC_SHIFT);
+        swizzle = r300_translate_vertex_data_swizzle(format);
+
+        if (i & 1) {
+            vstream->vap_prog_stream_cntl[i >> 1] |= type << 16;
+            vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16;
+        } else {
+            vstream->vap_prog_stream_cntl[i >> 1] |= type;
+            vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle;
+        }
+    }
+
+    /* Set the last vector in the PSC. */
+    if (i) {
+        i -= 1;
+    }
+    vstream->vap_prog_stream_cntl[i >> 1] |=
+        (R300_LAST_VEC << (i & 1 ? 16 : 0));
+
+    vstream->count = (i >> 1) + 1;
+}
+
+/* Update the PSC tables for SW TCL, using Draw. */
+static void r300_swtcl_vertex_psc(struct r300_context *r300,
+                                  struct r300_vertex_element_state *velems)
+{
+    struct r300_vertex_stream_state *vstream = &velems->vertex_stream;
+    struct r300_vertex_shader* vs = r300->vs_state.state;
+    struct vertex_info* vinfo = &r300->vertex_info;
+    uint16_t type, swizzle;
+    enum pipe_format format;
+    unsigned i, attrib_count;
+    int* vs_output_tab = vs->stream_loc_notcl;
+
+    /* For each Draw attribute, route it to the fragment shader according
+     * to the vs_output_tab. */
+    attrib_count = vinfo->num_attribs;
+    DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count);
+    for (i = 0; i < attrib_count; i++) {
+        DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d,"
+               " vs_output_tab %d\n", vinfo->attrib[i].src_index,
+               vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit,
+               vs_output_tab[i]);
+    }
+
+    for (i = 0; i < attrib_count; i++) {
+        /* Make sure we have a proper destination for our attribute. */
+        assert(vs_output_tab[i] != -1);
+
+        format = draw_translate_vinfo_format(vinfo->attrib[i].emit);
+
+        /* Obtain the type of data in this attribute. */
+        type = r300_translate_vertex_data_type(format) |
+            vs_output_tab[i] << R300_DST_VEC_LOC_SHIFT;
+
+        /* Obtain the swizzle for this attribute. Note that the default
+         * swizzle in the hardware is not XYZW! */
+        swizzle = r300_translate_vertex_data_swizzle(format);
+
+        /* Add the attribute to the PSC table. */
+        if (i & 1) {
+            vstream->vap_prog_stream_cntl[i >> 1] |= type << 16;
+            vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16;
+        } else {
+            vstream->vap_prog_stream_cntl[i >> 1] |= type;
+            vstream->vap_prog_stream_cntl_ext[i >> 1] |= swizzle;
+        }
+    }
+
+    /* Set the last vector in the PSC. */
+    if (i) {
+        i -= 1;
+    }
+    vstream->vap_prog_stream_cntl[i >> 1] |=
+        (R300_LAST_VEC << (i & 1 ? 16 : 0));
+
+    vstream->count = (i >> 1) + 1;
+}
+
+static void* r300_create_vertex_elements_state(struct pipe_context* pipe,
+                                               unsigned count,
+                                               const struct pipe_vertex_element* attribs)
+{
+    struct r300_context *r300 = r300_context(pipe);
+    struct r300_screen* r300screen = r300_screen(pipe->screen);
+    struct r300_vertex_element_state *velems;
+
+    assert(count <= PIPE_MAX_ATTRIBS);
+    velems = CALLOC_STRUCT(r300_vertex_element_state);
+    if (velems != NULL) {
+        velems->count = count;
+        memcpy(velems->velem, attribs, sizeof(struct pipe_vertex_element) * count);
+
+        if (r300screen->caps->has_tcl) {
+            r300_vertex_psc(velems);
+        } else {
+            memset(&r300->vertex_info, 0, sizeof(struct vertex_info));
+            r300_draw_emit_all_attribs(r300);
+            draw_compute_vertex_size(&r300->vertex_info);
+            r300_swtcl_vertex_psc(r300, velems);
+        }
+    }
+    return velems;
+}
+
+static void r300_bind_vertex_elements_state(struct pipe_context *pipe,
+                                            void *state)
+{
+    struct r300_context *r300 = r300_context(pipe);
+    struct r300_vertex_element_state *velems = state;
+
+    if (velems == NULL) {
+        return;
+    }
+
+    r300->velems = velems;
 
     if (r300->draw) {
         draw_flush(r300->draw);
-        draw_set_vertex_elements(r300->draw, count, elements);
+        draw_set_vertex_elements(r300->draw, velems->count, velems->velem);
     }
 
     if (!r300_validate_aos(r300)) {
@@ -1108,6 +1303,14 @@ static void r300_set_vertex_elements(struct pipe_context* pipe,
         assert(0);
         abort();
     }
+
+    UPDATE_STATE(&velems->vertex_stream, r300->vertex_stream_state);
+    r300->vertex_stream_state.size = (1 + velems->vertex_stream.count) * 2;
+}
+
+static void r300_delete_vertex_elements_state(struct pipe_context *pipe, void *state)
+{
+   FREE(state);
 }
 
 static void* r300_create_vs_state(struct pipe_context* pipe,
@@ -1115,64 +1318,71 @@ static void* r300_create_vs_state(struct pipe_context* pipe,
 {
     struct r300_context* r300 = r300_context(pipe);
 
-    if (r300_screen(pipe->screen)->caps->has_tcl) {
-        struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader);
-        /* Copy state directly into shader. */
-        vs->state = *shader;
-        vs->state.tokens = tgsi_dup_tokens(shader->tokens);
-
-        tgsi_scan_shader(shader->tokens, &vs->info);
+    struct r300_vertex_shader* vs = CALLOC_STRUCT(r300_vertex_shader);
+    r300_vertex_shader_common_init(vs, shader);
 
-        return (void*)vs;
+    if (r300_screen(pipe->screen)->caps->has_tcl) {
+        r300_translate_vertex_shader(r300, vs);
     } else {
-        return draw_create_vertex_shader(r300->draw, shader);
+        vs->draw_vs = draw_create_vertex_shader(r300->draw, shader);
     }
+
+    return vs;
 }
 
 static void r300_bind_vs_state(struct pipe_context* pipe, void* shader)
 {
     struct r300_context* r300 = r300_context(pipe);
+    struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
 
-    if (r300_screen(pipe->screen)->caps->has_tcl) {
-        struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
+    if (vs == NULL) {
+        r300->vs_state.state = NULL;
+        return;
+    }
+    if (vs == r300->vs_state.state) {
+        return;
+    }
+    r300->vs_state.state = vs;
 
-        if (vs == NULL) {
-            r300->vs = NULL;
-            return;
-        } else if (!vs->translated) {
-            r300_translate_vertex_shader(r300, vs);
-        }
+    // VS output mapping for HWTCL or stream mapping for SWTCL to the RS block
+    if (r300->fs) {
+        r300_vertex_shader_setup_wpos(r300);
+    }
+    memcpy(r300->vap_output_state.state, &vs->vap_out,
+           sizeof(struct r300_vap_output_state));
+    r300->vap_output_state.dirty = TRUE;
 
-        r300->vs = vs;
-        if (r300->fs) {
-            r300_vertex_shader_setup_wpos(r300);
-        }
+    /* The majority of the RS block bits is dependent on the vertex shader. */
+    r300->rs_block_state.dirty = TRUE; /* Will be updated before the emission. */
+
+    if (r300_screen(pipe->screen)->caps->has_tcl) {
+        r300->vs_state.dirty = TRUE;
+        r300->vs_state.size = vs->code.length + 9;
 
-        r300->vertex_format_state.dirty = TRUE;
+        r300->pvs_flush.dirty = TRUE;
 
-        r300->dirty_state |=
-            R300_NEW_VERTEX_SHADER | R300_NEW_VERTEX_SHADER_CONSTANTS;
+        r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS;
     } else {
         draw_flush(r300->draw);
         draw_bind_vertex_shader(r300->draw,
-                (struct draw_vertex_shader*)shader);
+                (struct draw_vertex_shader*)vs->draw_vs);
     }
 }
 
 static void r300_delete_vs_state(struct pipe_context* pipe, void* shader)
 {
     struct r300_context* r300 = r300_context(pipe);
+    struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
 
     if (r300_screen(pipe->screen)->caps->has_tcl) {
-        struct r300_vertex_shader* vs = (struct r300_vertex_shader*)shader;
-
         rc_constants_destroy(&vs->code.constants);
-        FREE((void*)vs->state.tokens);
-        FREE(shader);
     } else {
         draw_delete_vertex_shader(r300->draw,
-                (struct draw_vertex_shader*)shader);
+                (struct draw_vertex_shader*)vs->draw_vs);
     }
+
+    FREE((void*)vs->state.tokens);
+    FREE(shader);
 }
 
 static void r300_set_constant_buffer(struct pipe_context *pipe,
@@ -1223,8 +1433,12 @@ static void r300_set_constant_buffer(struct pipe_context *pipe,
     r300->shader_constants[shader].count = buf->size / (4 * sizeof(float));
     pipe_buffer_unmap(pipe->screen, buf);
 
-    if (shader == PIPE_SHADER_VERTEX)
-        r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS;
+    if (shader == PIPE_SHADER_VERTEX) {
+        if (r300screen->caps->has_tcl) {
+            r300->dirty_state |= R300_NEW_VERTEX_SHADER_CONSTANTS;
+            r300->pvs_flush.dirty = TRUE;
+        }
+    }
     else if (shader == PIPE_SHADER_FRAGMENT)
         r300->dirty_state |= R300_NEW_FRAGMENT_SHADER_CONSTANTS;
 }
@@ -1273,7 +1487,10 @@ void r300_init_state_functions(struct r300_context* r300)
     r300->context.set_viewport_state = r300_set_viewport_state;
 
     r300->context.set_vertex_buffers = r300_set_vertex_buffers;
-    r300->context.set_vertex_elements = r300_set_vertex_elements;
+
+    r300->context.create_vertex_elements_state = r300_create_vertex_elements_state;
+    r300->context.bind_vertex_elements_state = r300_bind_vertex_elements_state;
+    r300->context.delete_vertex_elements_state = r300_delete_vertex_elements_state;
 
     r300->context.create_vs_state = r300_create_vs_state;
     r300->context.bind_vs_state = r300_bind_vs_state;
diff --git a/src/gallium/drivers/r300/r300_state_derived.c b/src/gallium/drivers/r300/r300_state_derived.c
index 2cbce9210a7..6b9f61acd7b 100644
--- a/src/gallium/drivers/r300/r300_state_derived.c
+++ b/src/gallium/drivers/r300/r300_state_derived.c
@@ -37,179 +37,6 @@
 /* r300_state_derived: Various bits of state which are dependent upon
  * currently bound CSO data. */
 
-static void r300_draw_emit_attrib(struct r300_context* r300,
-                                  enum attrib_emit emit,
-                                  enum interp_mode interp,
-                                  int index)
-{
-    struct tgsi_shader_info* info = &r300->vs->info;
-    int output;
-
-    output = draw_find_shader_output(r300->draw,
-                                     info->output_semantic_name[index],
-                                     info->output_semantic_index[index]);
-    draw_emit_vertex_attr(
-        (struct vertex_info*)r300->vertex_format_state.state,
-        emit, interp, output);
-}
-
-static void r300_draw_emit_all_attribs(struct r300_context* r300)
-{
-    struct r300_shader_semantics* vs_outputs = &r300->vs->outputs;
-    int i, gen_count;
-
-    /* Position. */
-    if (vs_outputs->pos != ATTR_UNUSED) {
-        r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
-                              vs_outputs->pos);
-    } else {
-        assert(0);
-    }
-
-    /* Point size. */
-    if (vs_outputs->psize != ATTR_UNUSED) {
-        r300_draw_emit_attrib(r300, EMIT_1F_PSIZE, INTERP_POS,
-                              vs_outputs->psize);
-    }
-
-    /* Colors. */
-    for (i = 0; i < ATTR_COLOR_COUNT; i++) {
-        if (vs_outputs->color[i] != ATTR_UNUSED) {
-            r300_draw_emit_attrib(r300, EMIT_4F, INTERP_LINEAR,
-                                  vs_outputs->color[i]);
-        }
-    }
-
-    /* XXX Back-face colors. */
-
-    /* Texture coordinates. */
-    gen_count = 0;
-    for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
-        if (vs_outputs->generic[i] != ATTR_UNUSED) {
-            r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
-                                  vs_outputs->generic[i]);
-            gen_count++;
-        }
-    }
-
-    /* Fog coordinates. */
-    if (vs_outputs->fog != ATTR_UNUSED) {
-        r300_draw_emit_attrib(r300, EMIT_4F, INTERP_PERSPECTIVE,
-                              vs_outputs->fog);
-        gen_count++;
-    }
-
-    /* XXX magic */
-    assert(gen_count <= 8);
-}
-
-/* Update the PSC tables. */
-static void r300_vertex_psc(struct r300_context* r300)
-{
-    struct r300_vertex_info *vformat =
-        (struct r300_vertex_info*)r300->vertex_format_state.state;
-    uint16_t type, swizzle;
-    enum pipe_format format;
-    unsigned i;
-    int identity[16] = {0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 10, 11, 12, 13, 14, 15};
-    int* stream_tab;
-
-    /* If TCL is bypassed, map vertex streams to equivalent VS output
-     * locations. */
-    if (r300->tcl_bypass) {
-        stream_tab = r300->vs->stream_loc_notcl;
-    } else {
-        stream_tab = identity;
-    }
-
-    /* Vertex shaders have no semantics on their inputs,
-     * so PSC should just route stuff based on the vertex elements,
-     * and not on attrib information. */
-    DBG(r300, DBG_DRAW, "r300: vs expects %d attribs, routing %d elements"
-            " in psc\n",
-            r300->vs->info.num_inputs,
-            r300->vertex_element_count);
-
-    for (i = 0; i < r300->vertex_element_count; i++) {
-        format = r300->vertex_element[i].src_format;
-
-        type = r300_translate_vertex_data_type(format) |
-            (stream_tab[i] << R300_DST_VEC_LOC_SHIFT);
-        swizzle = r300_translate_vertex_data_swizzle(format);
-
-        if (i & 1) {
-            vformat->vap_prog_stream_cntl[i >> 1] |= type << 16;
-            vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16;
-        } else {
-            vformat->vap_prog_stream_cntl[i >> 1] |= type;
-            vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle;
-        }
-    }
-
-    assert(i <= 15);
-
-    /* Set the last vector in the PSC. */
-    if (i) {
-        i -= 1;
-    }
-    vformat->vap_prog_stream_cntl[i >> 1] |=
-        (R300_LAST_VEC << (i & 1 ? 16 : 0));
-}
-
-/* Update the PSC tables for SW TCL, using Draw. */
-static void r300_swtcl_vertex_psc(struct r300_context* r300)
-{
-    struct r300_vertex_info *vformat =
-        (struct r300_vertex_info*)r300->vertex_format_state.state;
-    struct vertex_info* vinfo = &vformat->vinfo;
-    uint16_t type, swizzle;
-    enum pipe_format format;
-    unsigned i, attrib_count;
-    int* vs_output_tab = r300->vs->stream_loc_notcl;
-
-    /* For each Draw attribute, route it to the fragment shader according
-     * to the vs_output_tab. */
-    attrib_count = vinfo->num_attribs;
-    DBG(r300, DBG_DRAW, "r300: attrib count: %d\n", attrib_count);
-    for (i = 0; i < attrib_count; i++) {
-        DBG(r300, DBG_DRAW, "r300: attrib: offset %d, interp %d, size %d,"
-               " vs_output_tab %d\n", vinfo->attrib[i].src_index,
-               vinfo->attrib[i].interp_mode, vinfo->attrib[i].emit,
-               vs_output_tab[i]);
-    }
-
-    for (i = 0; i < attrib_count; i++) {
-        /* Make sure we have a proper destination for our attribute. */
-        assert(vs_output_tab[i] != -1);
-
-        format = draw_translate_vinfo_format(vinfo->attrib[i].emit);
-
-        /* Obtain the type of data in this attribute. */
-        type = r300_translate_vertex_data_type(format) |
-            vs_output_tab[i] << R300_DST_VEC_LOC_SHIFT;
-
-        /* Obtain the swizzle for this attribute. Note that the default
-         * swizzle in the hardware is not XYZW! */
-        swizzle = r300_translate_vertex_data_swizzle(format);
-
-        /* Add the attribute to the PSC table. */
-        if (i & 1) {
-            vformat->vap_prog_stream_cntl[i >> 1] |= type << 16;
-            vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle << 16;
-        } else {
-            vformat->vap_prog_stream_cntl[i >> 1] |= type;
-            vformat->vap_prog_stream_cntl_ext[i >> 1] |= swizzle;
-        }
-    }
-
-    /* Set the last vector in the PSC. */
-    if (i) {
-        i -= 1;
-    }
-    vformat->vap_prog_stream_cntl[i >> 1] |=
-        (R300_LAST_VEC << (i & 1 ? 16 : 0));
-}
-
 static void r300_rs_col(struct r300_rs_block* rs, int id, int ptr,
                         boolean swizzle_0001)
 {
@@ -416,33 +243,16 @@ static void r300_update_rs_block(struct r300_context* r300,
     /* Now, after all that, see if we actually need to update the state. */
     if (memcmp(r300->rs_block_state.state, &rs, sizeof(struct r300_rs_block))) {
         memcpy(r300->rs_block_state.state, &rs, sizeof(struct r300_rs_block));
-        r300->rs_block_state.size = 5 + count;
-        r300->rs_block_state.dirty = TRUE;
+        r300->rs_block_state.size = 5 + count*2;
     }
 }
 
 /* Update the shader-dependant states. */
 static void r300_update_derived_shader_state(struct r300_context* r300)
 {
-    struct r300_screen* r300screen = r300_screen(r300->context.screen);
-    struct r300_vertex_info *vformat =
-        (struct r300_vertex_info*)r300->vertex_format_state.state;
-    struct vertex_info* vinfo = &vformat->vinfo;
+    struct r300_vertex_shader* vs = r300->vs_state.state;
 
-    /* Mmm, delicious hax */
-    memset(r300->vertex_format_state.state, 0, sizeof(struct r300_vertex_info));
-    memcpy(vinfo->hwfmt, r300->vs->hwfmt, sizeof(uint)*4);
-
-    r300_update_rs_block(r300, &r300->vs->outputs, &r300->fs->inputs);
-
-    if (r300screen->caps->has_tcl) {
-        r300_vertex_psc(r300);
-    } else {
-        r300_draw_emit_all_attribs(r300);
-        draw_compute_vertex_size(
-            (struct vertex_info*)r300->vertex_format_state.state);
-        r300_swtcl_vertex_psc(r300);
-    }
+    r300_update_rs_block(r300, &vs->outputs, &r300->fs->inputs);
 }
 
 static boolean r300_dsa_writes_depth_stencil(struct r300_dsa_state* dsa)
@@ -516,14 +326,72 @@ static void r300_update_ztop(struct r300_context* r300)
     r300->ztop_state.dirty = TRUE;
 }
 
+static void r300_merge_textures_and_samplers(struct r300_context* r300)
+{
+    struct r300_textures_state *state =
+        (struct r300_textures_state*)r300->textures_state.state;
+    struct r300_texture_sampler_state *texstate;
+    struct r300_sampler_state *sampler;
+    struct r300_texture *tex;
+    unsigned min_level, max_level, i, size;
+    unsigned count = MIN2(state->texture_count, state->sampler_count);
+
+    state->tx_enable = 0;
+    size = 2;
+
+    for (i = 0; i < count; i++) {
+        if (state->textures[i] && state->sampler_states[i]) {
+            state->tx_enable |= 1 << i;
+
+            tex = state->textures[i];
+            sampler = state->sampler_states[i];
+
+            texstate = &state->regs[i];
+            memcpy(texstate->format, &tex->state, sizeof(uint32_t)*3);
+            texstate->filter[0] = sampler->filter0;
+            texstate->filter[1] = sampler->filter1;
+            texstate->border_color = sampler->border_color;
+            texstate->tile_config = R300_TXO_MACRO_TILE(tex->macrotile) |
+                                    R300_TXO_MICRO_TILE(tex->microtile);
+
+            /* to emulate 1D textures through 2D ones correctly */
+            if (tex->tex.target == PIPE_TEXTURE_1D) {
+                texstate->filter[0] &= ~R300_TX_WRAP_T_MASK;
+                texstate->filter[0] |= R300_TX_WRAP_T(R300_TX_CLAMP_TO_EDGE);
+            }
+
+            if (tex->is_npot) {
+                /* NPOT textures don't support mip filter, unfortunately.
+                 * This prevents incorrect rendering. */
+                texstate->filter[0] &= ~R300_TX_MIN_FILTER_MIP_MASK;
+            } else {
+                /* determine min/max levels */
+                /* the MAX_MIP level is the largest (finest) one */
+                max_level = MIN2(sampler->max_lod, tex->tex.last_level);
+                min_level = MIN2(sampler->min_lod, max_level);
+                texstate->format[0] |= R300_TX_NUM_LEVELS(max_level);
+                texstate->filter[0] |= R300_TX_MAX_MIP_LEVEL(min_level);
+            }
+
+            texstate->filter[0] |= i << 28;
+
+            size += 16;
+            state->count = i+1;
+        }
+    }
+
+    r300->textures_state.size = size;
+}
+
 void r300_update_derived_state(struct r300_context* r300)
 {
-    /* XXX */
-    if (r300->dirty_state &
-        (R300_NEW_FRAGMENT_SHADER | R300_NEW_VERTEX_SHADER) ||
-        r300->vertex_format_state.dirty || r300->rs_state.dirty) {
+    if (r300->rs_block_state.dirty) {
         r300_update_derived_shader_state(r300);
     }
 
+    if (r300->textures_state.dirty) {
+        r300_merge_textures_and_samplers(r300);
+    }
+
     r300_update_ztop(r300);
 }
diff --git a/src/gallium/drivers/r300/r300_state_inlines.h b/src/gallium/drivers/r300/r300_state_inlines.h
index a6083925326..a32924ed0a3 100644
--- a/src/gallium/drivers/r300/r300_state_inlines.h
+++ b/src/gallium/drivers/r300/r300_state_inlines.h
@@ -327,117 +327,6 @@ static INLINE uint32_t r300_anisotropy(unsigned max_aniso)
     }
 }
 
-/* Buffer formats. */
-
-/* Colorbuffer formats. This is the unswizzled format of the RB3D block's
- * output. For the swizzling of the targets, check the shader's format. */
-static INLINE uint32_t r300_translate_colorformat(enum pipe_format format)
-{
-    switch (format) {
-        /* 8-bit buffers */
-        case PIPE_FORMAT_A8_UNORM:
-        case PIPE_FORMAT_I8_UNORM:
-        case PIPE_FORMAT_L8_UNORM:
-            return R300_COLOR_FORMAT_I8;
-        /* 16-bit buffers */
-        case PIPE_FORMAT_R5G6B5_UNORM:
-            return R300_COLOR_FORMAT_RGB565;
-        case PIPE_FORMAT_A1R5G5B5_UNORM:
-            return R300_COLOR_FORMAT_ARGB1555;
-        case PIPE_FORMAT_A4R4G4B4_UNORM:
-            return R300_COLOR_FORMAT_ARGB4444;
-        /* 32-bit buffers */
-        case PIPE_FORMAT_A8R8G8B8_UNORM:
-        case PIPE_FORMAT_X8R8G8B8_UNORM:
-        case PIPE_FORMAT_R8G8B8A8_UNORM:
-        case PIPE_FORMAT_R8G8B8X8_UNORM:
-            return R300_COLOR_FORMAT_ARGB8888;
-        /* XXX Not in pipe_format
-        case PIPE_FORMAT_A32R32G32B32:
-            return R300_COLOR_FORMAT_ARGB32323232;
-        case PIPE_FORMAT_A16R16G16B16:
-            return R300_COLOR_FORMAT_ARGB16161616;
-        case PIPE_FORMAT_A10R10G10B10_UNORM:
-            return R500_COLOR_FORMAT_ARGB10101010;
-        case PIPE_FORMAT_A2R10G10B10_UNORM:
-            return R500_COLOR_FORMAT_ARGB2101010;
-        case PIPE_FORMAT_I10_UNORM:
-            return R500_COLOR_FORMAT_I10; */
-        default:
-            debug_printf("r300: Implementation error: "
-                "Got unsupported color format %s in %s\n",
-                util_format_name(format), __FUNCTION__);
-            assert(0);
-            break;
-    }
-    return 0;
-}
-
-/* Depthbuffer and stencilbuffer. Thankfully, we only support two flavors. */
-static INLINE uint32_t r300_translate_zsformat(enum pipe_format format)
-{
-    switch (format) {
-        /* 16-bit depth, no stencil */
-        case PIPE_FORMAT_Z16_UNORM:
-            return R300_DEPTHFORMAT_16BIT_INT_Z;
-        /* 24-bit depth, ignored stencil */
-        case PIPE_FORMAT_Z24X8_UNORM:
-        /* 24-bit depth, 8-bit stencil */
-        case PIPE_FORMAT_Z24S8_UNORM:
-            return R300_DEPTHFORMAT_24BIT_INT_Z_8BIT_STENCIL;
-        default:
-            debug_printf("r300: Implementation error: "
-                "Got unsupported ZS format %s in %s\n",
-                util_format_name(format), __FUNCTION__);
-            assert(0);
-            break;
-    }
-    return 0;
-}
-
-/* Shader output formats. This is essentially the swizzle from the shader
- * to the RB3D block.
- *
- * Note that formats are stored from C3 to C0. */
-static INLINE uint32_t r300_translate_out_fmt(enum pipe_format format)
-{
-    switch (format) {
-        case PIPE_FORMAT_R5G6B5_UNORM:
-            /* C_5_6_5 is missing in US_OUT_FMT, but C4_8 works just fine. */
-        case PIPE_FORMAT_A1R5G5B5_UNORM:
-            /* C_1_5_5_5 is missing in US_OUT_FMT, but C4_8 works just fine. */
-        case PIPE_FORMAT_A4R4G4B4_UNORM:
-            /* C4_4 is missing in US_OUT_FMT, but C4_8 works just fine. */
-        case PIPE_FORMAT_A8R8G8B8_UNORM:
-        case PIPE_FORMAT_X8R8G8B8_UNORM:
-            return R300_US_OUT_FMT_C4_8 |
-                R300_C0_SEL_B | R300_C1_SEL_G |
-                R300_C2_SEL_R | R300_C3_SEL_A;
-        case PIPE_FORMAT_R8G8B8A8_UNORM:
-        case PIPE_FORMAT_R8G8B8X8_UNORM:
-            return R300_US_OUT_FMT_C4_8 |
-                R300_C0_SEL_A | R300_C1_SEL_B |
-                R300_C2_SEL_G | R300_C3_SEL_R;
-
-        /* 8-bit outputs */
-        case PIPE_FORMAT_A8_UNORM:
-            return R300_US_OUT_FMT_C4_8 |
-                R300_C0_SEL_A;
-        case PIPE_FORMAT_I8_UNORM:
-        case PIPE_FORMAT_L8_UNORM:
-            return R300_US_OUT_FMT_C4_8 |
-                R300_C0_SEL_R;
- /* R300_OUT_SIGN(x) */
-        default:
-            debug_printf("r300: Implementation error: "
-                "Got unsupported output format %s in %s\n",
-                util_format_name(format), __FUNCTION__);
-            assert(0);
-            return R300_US_OUT_FMT_UNUSED;
-    }
-    return 0;
-}
-
 /* Non-CSO state. (For now.) */
 
 static INLINE uint32_t r300_translate_gb_pipes(int pipe_count)
@@ -459,44 +348,16 @@ static INLINE uint32_t r300_translate_gb_pipes(int pipe_count)
     return 0;
 }
 
-/* Utility function to count the number of components in RGBAZS formats.
- * XXX should go to util or p_format.h */
-static INLINE unsigned pf_component_count(enum pipe_format format) {
-    unsigned count = 0;
-
-    if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 0)) {
-        count++;
-    }
-    if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 1)) {
-        count++;
-    }
-    if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 2)) {
-        count++;
-    }
-    if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_RGB, 3)) {
-        count++;
-    }
-    if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_ZS, 0)) {
-        count++;
-    }
-    if (util_format_get_component_bits(format, UTIL_FORMAT_COLORSPACE_ZS, 1)) {
-        count++;
-    }
-
-    return count;
-}
-
 /* Translate pipe_formats into PSC vertex types. */
 static INLINE uint16_t
 r300_translate_vertex_data_type(enum pipe_format format) {
     uint32_t result = 0;
     const struct util_format_description *desc;
-    unsigned components = pf_component_count(format);
+    unsigned components = util_format_get_nr_components(format);
 
     desc = util_format_description(format);
 
-    if (desc->layout != UTIL_FORMAT_LAYOUT_ARITH &&
-        desc->layout != UTIL_FORMAT_LAYOUT_ARRAY) {
+    if (desc->layout != UTIL_FORMAT_LAYOUT_PLAIN) {
         debug_printf("r300: Bad format %s in %s:%d\n", util_format_name(format),
             __FUNCTION__, __LINE__);
         assert(0);
@@ -565,36 +426,19 @@ r300_translate_vertex_data_type(enum pipe_format format) {
 static INLINE uint16_t
 r300_translate_vertex_data_swizzle(enum pipe_format format) {
     const struct util_format_description *desc = util_format_description(format);
-    unsigned swizzle[4], i;
 
     assert(format);
 
-    if (desc->layout != UTIL_FORMAT_LAYOUT_ARITH &&
-        desc->layout != UTIL_FORMAT_LAYOUT_ARRAY) {
+    if (desc->layout != UTIL_FORMAT_LAYOUT_PLAIN) {
         debug_printf("r300: Bad format %s in %s:%d\n",
             util_format_name(format), __FUNCTION__, __LINE__);
         return 0;
     }
 
-    /* Swizzles for 8bits formats are in the reversed order, not sure why. */
-    if (desc->channel[0].size == 8) {
-        for (i = 0; i < 4; i++) {
-            if (desc->swizzle[i] <= 3) {
-                swizzle[i] = 3 - desc->swizzle[i];
-            } else {
-                swizzle[i] = desc->swizzle[i];
-            }
-        }
-    } else {
-        for (i = 0; i < 4; i++) {
-            swizzle[i] = desc->swizzle[i];
-        }
-    }
-
-    return ((swizzle[0] << R300_SWIZZLE_SELECT_X_SHIFT) |
-            (swizzle[1] << R300_SWIZZLE_SELECT_Y_SHIFT) |
-            (swizzle[2] << R300_SWIZZLE_SELECT_Z_SHIFT) |
-            (swizzle[3] << R300_SWIZZLE_SELECT_W_SHIFT) |
+    return ((desc->swizzle[0] << R300_SWIZZLE_SELECT_X_SHIFT) |
+            (desc->swizzle[1] << R300_SWIZZLE_SELECT_Y_SHIFT) |
+            (desc->swizzle[2] << R300_SWIZZLE_SELECT_Z_SHIFT) |
+            (desc->swizzle[3] << R300_SWIZZLE_SELECT_W_SHIFT) |
             (0xf << R300_WRITE_ENA_SHIFT));
 }
 
diff --git a/src/gallium/drivers/r300/r300_state_invariant.c b/src/gallium/drivers/r300/r300_state_invariant.c
index 97927acf1b4..4a2c68269b1 100644
--- a/src/gallium/drivers/r300/r300_state_invariant.c
+++ b/src/gallium/drivers/r300/r300_state_invariant.c
@@ -38,7 +38,8 @@ struct pipe_viewport_state r300_viewport_identity = {
  *
  * Note that eventually this should be empty, but it's useful for development
  * and general unduplication of code. */
-void r300_emit_invariant_state(struct r300_context* r300, void* state)
+void r300_emit_invariant_state(struct r300_context* r300,
+                               unsigned size, void* state)
 {
     struct r300_capabilities* caps = r300_screen(r300->context.screen)->caps;
     CS_LOCALS(r300);
diff --git a/src/gallium/drivers/r300/r300_state_invariant.h b/src/gallium/drivers/r300/r300_state_invariant.h
index 5d1a9636545..83d031c7fe9 100644
--- a/src/gallium/drivers/r300/r300_state_invariant.h
+++ b/src/gallium/drivers/r300/r300_state_invariant.h
@@ -25,6 +25,7 @@
 
 struct r300_context;
 
-void r300_emit_invariant_state(struct r300_context* r300, void* state);
+void r300_emit_invariant_state(struct r300_context* r300,
+                               unsigned size, void* state);
 
 #endif /* R300_STATE_INVARIANT_H */
diff --git a/src/gallium/drivers/r300/r300_texture.c b/src/gallium/drivers/r300/r300_texture.c
index 34a49bec34c..04124afd4dc 100644
--- a/src/gallium/drivers/r300/r300_texture.c
+++ b/src/gallium/drivers/r300/r300_texture.c
@@ -1,5 +1,6 @@
 /*
  * Copyright 2008 Corbin Simpson <[email protected]>
+ * Copyright 2010 Marek Olšák <[email protected]>
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -29,6 +30,7 @@
 #include "r300_context.h"
 #include "r300_texture.h"
 #include "r300_screen.h"
+#include "r300_state_inlines.h"
 
 #include "radeon_winsys.h"
 
@@ -44,12 +46,465 @@ static const unsigned microblock_table[5][3][2] = {
     {{ 2, 1}, {0, 0}, {0, 0}}  /* 128 bits per pixel */
 };
 
+/* Translate a pipe_format into a useful texture format for sampling.
+ *
+ * Some special formats are translated directly using R300_EASY_TX_FORMAT,
+ * but the majority of them is translated in a generic way, automatically
+ * supporting all the formats hw can support.
+ *
+ * R300_EASY_TX_FORMAT swizzles the texture.
+ * Note the signature of R300_EASY_TX_FORMAT:
+ *   R300_EASY_TX_FORMAT(B, G, R, A, FORMAT);
+ *
+ * The FORMAT specifies how the texture sampler will treat the texture, and
+ * makes available X, Y, Z, W, ZERO, and ONE for swizzling. */
+static uint32_t r300_translate_texformat(enum pipe_format format)
+{
+    uint32_t result = 0;
+    const struct util_format_description *desc;
+    unsigned components = 0, i;
+    boolean uniform = TRUE;
+    const uint32_t swizzle_shift[4] = {
+        R300_TX_FORMAT_R_SHIFT,
+        R300_TX_FORMAT_G_SHIFT,
+        R300_TX_FORMAT_B_SHIFT,
+        R300_TX_FORMAT_A_SHIFT
+    };
+    const uint32_t swizzle[4] = {
+        R300_TX_FORMAT_X,
+        R300_TX_FORMAT_Y,
+        R300_TX_FORMAT_Z,
+        R300_TX_FORMAT_W
+    };
+    const uint32_t sign_bit[4] = {
+        R300_TX_FORMAT_SIGNED_X,
+        R300_TX_FORMAT_SIGNED_Y,
+        R300_TX_FORMAT_SIGNED_Z,
+        R300_TX_FORMAT_SIGNED_W,
+    };
+
+    desc = util_format_description(format);
+
+    /* Colorspace (return non-RGB formats directly). */
+    switch (desc->colorspace) {
+        /* Depth stencil formats. */
+        case UTIL_FORMAT_COLORSPACE_ZS:
+            switch (format) {
+                case PIPE_FORMAT_Z16_UNORM:
+                    return R300_EASY_TX_FORMAT(X, X, X, X, X16);
+                case PIPE_FORMAT_X8Z24_UNORM:
+                case PIPE_FORMAT_S8Z24_UNORM:
+                    return R300_EASY_TX_FORMAT(X, X, X, X, W24_FP);
+                default:
+                    return ~0; /* Unsupported. */
+            }
+
+        /* YUV formats. */
+        case UTIL_FORMAT_COLORSPACE_YUV:
+            result |= R300_TX_FORMAT_YUV_TO_RGB;
+
+            switch (format) {
+                case PIPE_FORMAT_UYVY:
+                    return R300_EASY_TX_FORMAT(X, Y, Z, ONE, YVYU422) | result;
+                case PIPE_FORMAT_YUYV:
+                    return R300_EASY_TX_FORMAT(X, Y, Z, ONE, VYUY422) | result;
+                default:
+                    return ~0; /* Unsupported/unknown. */
+            }
+
+        /* Add gamma correction. */
+        case UTIL_FORMAT_COLORSPACE_SRGB:
+            result |= R300_TX_FORMAT_GAMMA;
+            break;
+
+        default:;
+    }
+
+    /* Add swizzle. */
+    for (i = 0; i < 4; i++) {
+        switch (desc->swizzle[i]) {
+            case UTIL_FORMAT_SWIZZLE_X:
+            case UTIL_FORMAT_SWIZZLE_NONE:
+                result |= swizzle[0] << swizzle_shift[i];
+                break;
+            case UTIL_FORMAT_SWIZZLE_Y:
+                result |= swizzle[1] << swizzle_shift[i];
+                break;
+            case UTIL_FORMAT_SWIZZLE_Z:
+                result |= swizzle[2] << swizzle_shift[i];
+                break;
+            case UTIL_FORMAT_SWIZZLE_W:
+                result |= swizzle[3] << swizzle_shift[i];
+                break;
+            case UTIL_FORMAT_SWIZZLE_0:
+                result |= R300_TX_FORMAT_ZERO << swizzle_shift[i];
+                break;
+            case UTIL_FORMAT_SWIZZLE_1:
+                result |= R300_TX_FORMAT_ONE << swizzle_shift[i];
+                break;
+            default:
+                return ~0; /* Unsupported. */
+        }
+    }
+
+    /* Compressed formats. */
+    if (desc->layout == UTIL_FORMAT_LAYOUT_COMPRESSED) {
+        switch (format) {
+            case PIPE_FORMAT_DXT1_RGB:
+            case PIPE_FORMAT_DXT1_RGBA:
+            case PIPE_FORMAT_DXT1_SRGB:
+            case PIPE_FORMAT_DXT1_SRGBA:
+                return R300_TX_FORMAT_DXT1 | result;
+            case PIPE_FORMAT_DXT3_RGBA:
+            case PIPE_FORMAT_DXT3_SRGBA:
+                return R300_TX_FORMAT_DXT3 | result;
+            case PIPE_FORMAT_DXT5_RGBA:
+            case PIPE_FORMAT_DXT5_SRGBA:
+                return R300_TX_FORMAT_DXT5 | result;
+            default:
+                return ~0; /* Unsupported/unknown. */
+        }
+    }
+
+    /* Get the number of components. */
+    for (i = 0; i < 4; i++) {
+        if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) {
+            ++components;
+        }
+    }
+
+    /* Add sign. */
+    for (i = 0; i < components; i++) {
+        if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
+            result |= sign_bit[i];
+        }
+    }
+
+    /* See whether the components are of the same size. */
+    for (i = 1; i < components; i++) {
+        uniform = uniform && desc->channel[0].size == desc->channel[i].size;
+    }
+
+    /* Non-uniform formats. */
+    if (!uniform) {
+        switch (components) {
+            case 3:
+                if (desc->channel[0].size == 5 &&
+                    desc->channel[1].size == 6 &&
+                    desc->channel[2].size == 5) {
+                    return R300_TX_FORMAT_Z5Y6X5 | result;
+                }
+                if (desc->channel[0].size == 5 &&
+                    desc->channel[1].size == 5 &&
+                    desc->channel[2].size == 6) {
+                    return R300_TX_FORMAT_Z6Y5X5 | result;
+                }
+                return ~0; /* Unsupported/unknown. */
+
+            case 4:
+                if (desc->channel[0].size == 5 &&
+                    desc->channel[1].size == 5 &&
+                    desc->channel[2].size == 5 &&
+                    desc->channel[3].size == 1) {
+                    return R300_TX_FORMAT_W1Z5Y5X5 | result;
+                }
+                if (desc->channel[0].size == 10 &&
+                    desc->channel[1].size == 10 &&
+                    desc->channel[2].size == 10 &&
+                    desc->channel[3].size == 2) {
+                    return R300_TX_FORMAT_W2Z10Y10X10 | result;
+                }
+        }
+        return ~0; /* Unsupported/unknown. */
+    }
+
+    /* And finally, uniform formats. */
+    switch (desc->channel[0].type) {
+        case UTIL_FORMAT_TYPE_UNSIGNED:
+        case UTIL_FORMAT_TYPE_SIGNED:
+            if (!desc->channel[0].normalized &&
+                desc->colorspace != UTIL_FORMAT_COLORSPACE_SRGB) {
+                return ~0;
+            }
+
+            switch (desc->channel[0].size) {
+                case 4:
+                    switch (components) {
+                        case 2:
+                            return R300_TX_FORMAT_Y4X4 | result;
+                        case 4:
+                            return R300_TX_FORMAT_W4Z4Y4X4 | result;
+                    }
+                    return ~0;
+
+                case 8:
+                    switch (components) {
+                        case 1:
+                            return R300_TX_FORMAT_X8 | result;
+                        case 2:
+                            return R300_TX_FORMAT_Y8X8 | result;
+                        case 4:
+                            return R300_TX_FORMAT_W8Z8Y8X8 | result;
+                    }
+                    return ~0;
+
+                case 16:
+                    switch (components) {
+                        case 1:
+                            return R300_TX_FORMAT_X16 | result;
+                        case 2:
+                            return R300_TX_FORMAT_Y16X16 | result;
+                        case 4:
+                            return R300_TX_FORMAT_W16Z16Y16X16 | result;
+                    }
+            }
+            return ~0;
+
+/* XXX Enable float textures here. */
+#if 0
+        case UTIL_FORMAT_TYPE_FLOAT:
+            switch (desc->channel[0].size) {
+                case 16:
+                    switch (components) {
+                        case 1:
+                            return R300_TX_FORMAT_16F | result;
+                        case 2:
+                            return R300_TX_FORMAT_16F_16F | result;
+                        case 4:
+                            return R300_TX_FORMAT_16F_16F_16F_16F | result;
+                    }
+                    return ~0;
+
+                case 32:
+                    switch (components) {
+                        case 1:
+                            return R300_TX_FORMAT_32F | result;
+                        case 2:
+                            return R300_TX_FORMAT_32F_32F | result;
+                        case 4:
+                            return R300_TX_FORMAT_32F_32F_32F_32F | result;
+                    }
+            }
+#endif
+    }
+
+    return ~0; /* Unsupported/unknown. */
+}
+
+/* Buffer formats. */
+
+/* Colorbuffer formats. This is the unswizzled format of the RB3D block's
+ * output. For the swizzling of the targets, check the shader's format. */
+static uint32_t r300_translate_colorformat(enum pipe_format format)
+{
+    switch (format) {
+        /* 8-bit buffers. */
+        case PIPE_FORMAT_A8_UNORM:
+        case PIPE_FORMAT_I8_UNORM:
+        case PIPE_FORMAT_L8_UNORM:
+        case PIPE_FORMAT_L8_SRGB:
+        case PIPE_FORMAT_R8_UNORM:
+        case PIPE_FORMAT_R8_SNORM:
+            return R300_COLOR_FORMAT_I8;
+
+        /* 16-bit buffers. */
+        case PIPE_FORMAT_B5G6R5_UNORM:
+            return R300_COLOR_FORMAT_RGB565;
+        case PIPE_FORMAT_B5G5R5A1_UNORM:
+            return R300_COLOR_FORMAT_ARGB1555;
+        case PIPE_FORMAT_B4G4R4A4_UNORM:
+            return R300_COLOR_FORMAT_ARGB4444;
+
+        /* 32-bit buffers. */
+        case PIPE_FORMAT_B8G8R8A8_UNORM:
+        case PIPE_FORMAT_B8G8R8A8_SRGB:
+        case PIPE_FORMAT_B8G8R8X8_UNORM:
+        case PIPE_FORMAT_B8G8R8X8_SRGB:
+        case PIPE_FORMAT_A8R8G8B8_UNORM:
+        case PIPE_FORMAT_A8R8G8B8_SRGB:
+        case PIPE_FORMAT_X8R8G8B8_UNORM:
+        case PIPE_FORMAT_X8R8G8B8_SRGB:
+        case PIPE_FORMAT_A8B8G8R8_UNORM:
+        case PIPE_FORMAT_R8G8B8A8_SNORM:
+        case PIPE_FORMAT_A8B8G8R8_SRGB:
+        case PIPE_FORMAT_X8B8G8R8_UNORM:
+        case PIPE_FORMAT_X8B8G8R8_SRGB:
+        case PIPE_FORMAT_R8SG8SB8UX8U_NORM:
+            return R300_COLOR_FORMAT_ARGB8888;
+        case PIPE_FORMAT_R10G10B10A2_UNORM:
+            return R500_COLOR_FORMAT_ARGB2101010;  /* R5xx-only? */
+
+        /* 64-bit buffers. */
+        case PIPE_FORMAT_R16G16B16A16_UNORM:
+        case PIPE_FORMAT_R16G16B16A16_SNORM:
+        //case PIPE_FORMAT_R16G16B16A16_FLOAT: /* not in pipe_format */
+            return R300_COLOR_FORMAT_ARGB16161616;
+
+/* XXX Enable float textures here. */
+#if 0
+        /* 128-bit buffers. */
+        case PIPE_FORMAT_R32G32B32A32_FLOAT:
+            return R300_COLOR_FORMAT_ARGB32323232;
+#endif
+
+        /* YUV buffers. */
+        case PIPE_FORMAT_UYVY:
+            return R300_COLOR_FORMAT_YVYU;
+        case PIPE_FORMAT_YUYV:
+            return R300_COLOR_FORMAT_VYUY;
+        default:
+            return ~0; /* Unsupported. */
+    }
+}
+
+/* Depthbuffer and stencilbuffer. Thankfully, we only support two flavors. */
+static uint32_t r300_translate_zsformat(enum pipe_format format)
+{
+    switch (format) {
+        /* 16-bit depth, no stencil */
+        case PIPE_FORMAT_Z16_UNORM:
+            return R300_DEPTHFORMAT_16BIT_INT_Z;
+        /* 24-bit depth, ignored stencil */
+        case PIPE_FORMAT_X8Z24_UNORM:
+        /* 24-bit depth, 8-bit stencil */
+        case PIPE_FORMAT_S8Z24_UNORM:
+            return R300_DEPTHFORMAT_24BIT_INT_Z_8BIT_STENCIL;
+        default:
+            return ~0; /* Unsupported. */
+    }
+}
+
+/* Shader output formats. This is essentially the swizzle from the shader
+ * to the RB3D block.
+ *
+ * Note that formats are stored from C3 to C0. */
+static uint32_t r300_translate_out_fmt(enum pipe_format format)
+{
+    uint32_t modifier = 0;
+    unsigned i;
+    const struct util_format_description *desc;
+    static const uint32_t sign_bit[4] = {
+        R300_OUT_SIGN(0x1),
+        R300_OUT_SIGN(0x2),
+        R300_OUT_SIGN(0x4),
+        R300_OUT_SIGN(0x8),
+    };
+
+    desc = util_format_description(format);
+
+    /* Specifies how the shader output is written to the fog unit. */
+    if (desc->colorspace == UTIL_FORMAT_COLORSPACE_SRGB) {
+        /* The gamma correction causes precision loss so we need
+         * higher precision to maintain reasonable quality.
+         * It has nothing to do with the colorbuffer format. */
+        modifier |= R300_US_OUT_FMT_C4_10_GAMMA;
+    } else if (desc->channel[0].type == UTIL_FORMAT_TYPE_FLOAT) {
+        if (desc->channel[0].size == 32) {
+            modifier |= R300_US_OUT_FMT_C4_32_FP;
+        } else {
+            modifier |= R300_US_OUT_FMT_C4_16_FP;
+        }
+    } else {
+        if (desc->channel[0].size == 16) {
+            modifier |= R300_US_OUT_FMT_C4_16;
+        } else {
+            /* C4_8 seems to be used for the formats whose pixel size
+             * is <= 32 bits. */
+            modifier |= R300_US_OUT_FMT_C4_8;
+        }
+    }
+
+    /* Add sign. */
+    for (i = 0; i < 4; i++)
+        if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
+            modifier |= sign_bit[i];
+        }
+
+    /* Add swizzles and return. */
+    switch (format) {
+        /* 8-bit outputs.
+         * COLORFORMAT_I8 stores the C2 component. */
+        case PIPE_FORMAT_A8_UNORM:
+            return modifier | R300_C2_SEL_A;
+        case PIPE_FORMAT_I8_UNORM:
+        case PIPE_FORMAT_L8_UNORM:
+        case PIPE_FORMAT_L8_SRGB:
+        case PIPE_FORMAT_R8_UNORM:
+        case PIPE_FORMAT_R8_SNORM:
+            return modifier | R300_C2_SEL_R;
+
+        /* ARGB 32-bit outputs. */
+        case PIPE_FORMAT_B5G6R5_UNORM:
+        case PIPE_FORMAT_B5G5R5A1_UNORM:
+        case PIPE_FORMAT_B4G4R4A4_UNORM:
+        case PIPE_FORMAT_B8G8R8A8_UNORM:
+        case PIPE_FORMAT_B8G8R8A8_SRGB:
+        case PIPE_FORMAT_B8G8R8X8_UNORM:
+        case PIPE_FORMAT_B8G8R8X8_SRGB:
+            return modifier |
+                R300_C0_SEL_B | R300_C1_SEL_G |
+                R300_C2_SEL_R | R300_C3_SEL_A;
+
+        /* BGRA 32-bit outputs. */
+        case PIPE_FORMAT_A8R8G8B8_UNORM:
+        case PIPE_FORMAT_A8R8G8B8_SRGB:
+        case PIPE_FORMAT_X8R8G8B8_UNORM:
+        case PIPE_FORMAT_X8R8G8B8_SRGB:
+            return modifier |
+                R300_C0_SEL_A | R300_C1_SEL_R |
+                R300_C2_SEL_G | R300_C3_SEL_B;
+
+        /* RGBA 32-bit outputs. */
+        case PIPE_FORMAT_A8B8G8R8_UNORM:
+        case PIPE_FORMAT_R8G8B8A8_SNORM:
+        case PIPE_FORMAT_A8B8G8R8_SRGB:
+        case PIPE_FORMAT_X8B8G8R8_UNORM:
+        case PIPE_FORMAT_X8B8G8R8_SRGB:
+            return modifier |
+                R300_C0_SEL_A | R300_C1_SEL_B |
+                R300_C2_SEL_G | R300_C3_SEL_R;
+
+        /* ABGR 32-bit outputs. */
+        case PIPE_FORMAT_R8SG8SB8UX8U_NORM:
+        case PIPE_FORMAT_R10G10B10A2_UNORM:
+        /* RGBA high precision outputs (same swizzles as ABGR low precision) */
+        case PIPE_FORMAT_R16G16B16A16_UNORM:
+        case PIPE_FORMAT_R16G16B16A16_SNORM:
+        //case PIPE_FORMAT_R16G16B16A16_FLOAT: /* not in pipe_format */
+        case PIPE_FORMAT_R32G32B32A32_FLOAT:
+            return modifier |
+                R300_C0_SEL_R | R300_C1_SEL_G |
+                R300_C2_SEL_B | R300_C3_SEL_A;
+
+        default:
+            return ~0; /* Unsupported. */
+    }
+}
+
+boolean r300_is_colorbuffer_format_supported(enum pipe_format format)
+{
+    return r300_translate_colorformat(format) != ~0 &&
+           r300_translate_out_fmt(format) != ~0;
+}
+
+boolean r300_is_zs_format_supported(enum pipe_format format)
+{
+    return r300_translate_zsformat(format) != ~0;
+}
+
+boolean r300_is_sampler_format_supported(enum pipe_format format)
+{
+    return r300_translate_texformat(format) != ~0;
+}
+
 static void r300_setup_texture_state(struct r300_screen* screen, struct r300_texture* tex)
 {
-    struct r300_texture_state* state = &tex->state;
+    struct r300_texture_format_state* state = &tex->state;
     struct pipe_texture *pt = &tex->tex;
+    unsigned i;
     boolean is_r500 = screen->caps->is_r500;
 
+    /* Set sampler state. */
     state->format0 = R300_TX_WIDTH((pt->width0 - 1) & 0x7ff) |
                      R300_TX_HEIGHT((pt->height0 - 1) & 0x7ff);
 
@@ -83,6 +538,26 @@ static void r300_setup_texture_state(struct r300_screen* screen, struct r300_tex
 
     SCREEN_DBG(screen, DBG_TEX, "r300: Set texture state (%dx%d, %d levels)\n",
                pt->width0, pt->height0, pt->last_level);
+
+    /* Set framebuffer state. */
+    if (util_format_is_depth_or_stencil(tex->tex.format)) {
+        for (i = 0; i <= tex->tex.last_level; i++) {
+            tex->fb_state.depthpitch[i] =
+                tex->pitch[i] |
+                R300_DEPTHMACROTILE(tex->mip_macrotile[i]) |
+                R300_DEPTHMICROTILE(tex->microtile);
+        }
+        tex->fb_state.zb_format = r300_translate_zsformat(tex->tex.format);
+    } else {
+        for (i = 0; i <= tex->tex.last_level; i++) {
+            tex->fb_state.colorpitch[i] =
+                tex->pitch[i] |
+                r300_translate_colorformat(tex->tex.format) |
+                R300_COLOR_TILE(tex->mip_macrotile[i]) |
+                R300_COLOR_MICROTILE(tex->microtile);
+        }
+        tex->fb_state.us_out_fmt = r300_translate_out_fmt(tex->tex.format);
+    }
 }
 
 void r300_texture_reinterpret_format(struct pipe_screen *screen,
@@ -142,18 +617,23 @@ static unsigned r300_texture_get_tile_size(struct r300_texture* tex,
 /* Return true if macrotiling should be enabled on the miplevel. */
 static boolean r300_texture_macro_switch(struct r300_texture *tex,
                                          unsigned level,
-                                         boolean rv350_mode)
+                                         boolean rv350_mode,
+                                         int dim)
 {
-    unsigned tile_width, width;
+    unsigned tile, texdim;
 
-    tile_width = r300_texture_get_tile_size(tex, TILE_WIDTH, TRUE);
-    width = u_minify(tex->tex.width0, level);
+    tile = r300_texture_get_tile_size(tex, dim, TRUE);
+    if (dim == TILE_WIDTH) {
+        texdim = u_minify(tex->tex.width0, level);
+    } else {
+        texdim = u_minify(tex->tex.height0, level);
+    }
 
     /* See TX_FILTER1_n.MACRO_SWITCH. */
     if (rv350_mode) {
-        return width >= tile_width;
+        return texdim >= tile;
     } else {
-        return width > tile_width;
+        return texdim > tile;
     }
 }
 
@@ -217,9 +697,10 @@ static void r300_setup_miptree(struct r300_screen* screen,
 
     for (i = 0; i <= base->last_level; i++) {
         /* Let's see if this miplevel can be macrotiled. */
-        tex->mip_macrotile[i] = (tex->macrotile == R300_BUFFER_TILED &&
-                                 r300_texture_macro_switch(tex, i, rv350_mode)) ?
-                                 R300_BUFFER_TILED : R300_BUFFER_LINEAR;
+        tex->mip_macrotile[i] =
+            (tex->macrotile == R300_BUFFER_TILED &&
+             r300_texture_macro_switch(tex, i, rv350_mode, TILE_WIDTH)) ?
+             R300_BUFFER_TILED : R300_BUFFER_LINEAR;
 
         stride = r300_texture_get_stride(screen, tex, i);
         nblocksy = r300_texture_get_nblocksy(tex, i);
@@ -249,10 +730,46 @@ static void r300_setup_flags(struct r300_texture* tex)
                    !util_is_power_of_two(tex->tex.height0);
 }
 
+static void r300_setup_tiling(struct pipe_screen *screen,
+                              struct r300_texture *tex)
+{
+    enum pipe_format format = tex->tex.format;
+    boolean rv350_mode = r300_screen(screen)->caps->family >= CHIP_FAMILY_RV350;
+
+    if (util_format_is_compressed(format)) {
+        return;
+    }
+
+    if (tex->tex.width0 == 1 ||
+        tex->tex.height0 == 1) {
+        return;
+    }
+
+    /* Set microtiling. */
+    switch (util_format_get_blocksize(format)) {
+        case 1:
+        case 4:
+            tex->microtile = R300_BUFFER_TILED;
+            break;
+
+        /* XXX Square-tiling doesn't work with kernel older than 2.6.34,
+         * XXX need to check the DRM version */
+        /*case 2:
+        case 8:
+            tex->microtile = R300_BUFFER_SQUARETILED;
+            break;*/
+    }
+
+    /* Set macrotiling. */
+    if (r300_texture_macro_switch(tex, 0, rv350_mode, TILE_WIDTH) &&
+        r300_texture_macro_switch(tex, 0, rv350_mode, TILE_HEIGHT)) {
+        tex->macrotile = R300_BUFFER_TILED;
+    }
+}
+
 /* Create a new texture. */
-static struct pipe_texture*
-    r300_texture_create(struct pipe_screen* screen,
-                        const struct pipe_texture* template)
+static struct pipe_texture* r300_texture_create(struct pipe_screen* screen,
+                                         const struct pipe_texture* template)
 {
     struct r300_texture* tex = CALLOC_STRUCT(r300_texture);
     struct r300_screen* rscreen = r300_screen(screen);
@@ -267,6 +784,9 @@ static struct pipe_texture*
     tex->tex.screen = screen;
 
     r300_setup_flags(tex);
+    if (!(template->tex_usage & R300_TEXTURE_USAGE_TRANSFER)) {
+        r300_setup_tiling(screen, tex);
+    }
     r300_setup_miptree(rscreen, tex);
     r300_setup_texture_state(rscreen, tex);
 
@@ -331,14 +851,17 @@ static void r300_tex_surface_destroy(struct pipe_surface* s)
     FREE(s);
 }
 
+
 static struct pipe_texture*
-    r300_texture_blanket(struct pipe_screen* screen,
-                         const struct pipe_texture* base,
-                         const unsigned* stride,
-                         struct pipe_buffer* buffer)
+    r300_texture_from_handle(struct pipe_screen* screen,
+                             const struct pipe_texture* base,
+                             struct winsys_handle *whandle)
 {
-    struct r300_texture* tex;
+    struct radeon_winsys* winsys = (struct radeon_winsys*)screen->winsys;
     struct r300_screen* rscreen = r300_screen(screen);
+    struct pipe_buffer *buffer;
+    struct r300_texture* tex;
+    unsigned stride;
 
     /* Support only 2D textures without mipmaps */
     if (base->target != PIPE_TEXTURE_2D ||
@@ -347,6 +870,11 @@ static struct pipe_texture*
         return NULL;
     }
 
+    buffer = winsys->buffer_from_handle(winsys, screen, whandle, &stride);
+    if (!buffer) {
+        return NULL;
+    }
+
     tex = CALLOC_STRUCT(r300_texture);
     if (!tex) {
         return NULL;
@@ -356,17 +884,38 @@ static struct pipe_texture*
     pipe_reference_init(&tex->tex.reference, 1);
     tex->tex.screen = screen;
 
-    tex->stride_override = *stride;
-    tex->pitch[0] = *stride / util_format_get_blocksize(base->format);
+    tex->stride_override = stride;
+    tex->pitch[0] = stride / util_format_get_blocksize(base->format);
 
     r300_setup_flags(tex);
     r300_setup_texture_state(rscreen, tex);
 
-    pipe_buffer_reference(&tex->buffer, buffer);
+    /* one ref already taken */
+    tex->buffer = buffer;
 
     return (struct pipe_texture*)tex;
 }
 
+static boolean
+    r300_texture_get_handle(struct pipe_screen* screen,
+                            struct pipe_texture *texture,
+                            struct winsys_handle *whandle)
+{
+    struct radeon_winsys* winsys = (struct radeon_winsys*)screen->winsys;
+    struct r300_texture* tex = (struct r300_texture*)texture;
+    unsigned stride;
+
+    if (!tex) {
+        return FALSE;
+    }
+
+    stride = r300_texture_get_stride(r300_screen(screen), tex, 0);
+
+    winsys->buffer_get_handle(winsys, tex->buffer, stride, whandle);
+
+    return TRUE;
+}
+
 static struct pipe_video_surface *
 r300_video_surface_create(struct pipe_screen *screen,
                           enum pipe_video_chroma_format chroma_format,
@@ -390,7 +939,7 @@ r300_video_surface_create(struct pipe_screen *screen,
 
     memset(&template, 0, sizeof(struct pipe_texture));
     template.target = PIPE_TEXTURE_2D;
-    template.format = PIPE_FORMAT_X8R8G8B8_UNORM;
+    template.format = PIPE_FORMAT_B8G8R8X8_UNORM;
     template.last_level = 0;
     template.width0 = util_next_power_of_two(width);
     template.height0 = util_next_power_of_two(height);
@@ -418,30 +967,13 @@ static void r300_video_surface_destroy(struct pipe_video_surface *vsfc)
 void r300_init_screen_texture_functions(struct pipe_screen* screen)
 {
     screen->texture_create = r300_texture_create;
+    screen->texture_from_handle = r300_texture_from_handle;
+    screen->texture_get_handle = r300_texture_get_handle;
     screen->texture_destroy = r300_texture_destroy;
     screen->get_tex_surface = r300_get_tex_surface;
     screen->tex_surface_destroy = r300_tex_surface_destroy;
-    screen->texture_blanket = r300_texture_blanket;
 
     screen->video_surface_create = r300_video_surface_create;
     screen->video_surface_destroy= r300_video_surface_destroy;
 }
 
-boolean r300_get_texture_buffer(struct pipe_screen* screen,
-                                struct pipe_texture* texture,
-                                struct pipe_buffer** buffer,
-                                unsigned* stride)
-{
-    struct r300_texture* tex = (struct r300_texture*)texture;
-    if (!tex) {
-        return FALSE;
-    }
-
-    pipe_buffer_reference(buffer, tex->buffer);
-
-    if (stride) {
-        *stride = r300_texture_get_stride(r300_screen(screen), tex, 0);
-    }
-
-    return TRUE;
-}
diff --git a/src/gallium/drivers/r300/r300_texture.h b/src/gallium/drivers/r300/r300_texture.h
index 153f4eebdaf..138b62784e6 100644
--- a/src/gallium/drivers/r300/r300_texture.h
+++ b/src/gallium/drivers/r300/r300_texture.h
@@ -42,239 +42,11 @@ void r300_texture_reinterpret_format(struct pipe_screen *screen,
                                      struct pipe_texture *tex,
                                      enum pipe_format new_format);
 
-/* Translate a pipe_format into a useful texture format for sampling.
- *
- * R300_EASY_TX_FORMAT swizzles the texture.
- * Note the signature of R300_EASY_TX_FORMAT:
- *   R300_EASY_TX_FORMAT(B, G, R, A, FORMAT);
- *
- * The FORMAT specifies how the texture sampler will treat the texture, and
- * makes available X, Y, Z, W, ZERO, and ONE for swizzling. */
-static INLINE uint32_t r300_translate_texformat(enum pipe_format format)
-{
-    uint32_t result = 0;
-    const struct util_format_description *desc;
-    unsigned components = 0, i;
-    boolean uniform = TRUE;
-    const uint32_t swizzle_shift[4] = {
-        R300_TX_FORMAT_R_SHIFT,
-        R300_TX_FORMAT_G_SHIFT,
-        R300_TX_FORMAT_B_SHIFT,
-        R300_TX_FORMAT_A_SHIFT
-    };
-    const uint32_t sign_bit[4] = {
-        R300_TX_FORMAT_SIGNED_X,
-        R300_TX_FORMAT_SIGNED_Y,
-        R300_TX_FORMAT_SIGNED_Z,
-        R300_TX_FORMAT_SIGNED_W,
-    };
-
-    desc = util_format_description(format);
-
-    /* Colorspace (return non-RGB formats directly). */
-    switch (desc->colorspace) {
-        /* Depth stencil formats. */
-        case UTIL_FORMAT_COLORSPACE_ZS:
-            switch (format) {
-                case PIPE_FORMAT_Z16_UNORM:
-                    return R300_EASY_TX_FORMAT(X, X, X, X, X16);
-                case PIPE_FORMAT_Z24X8_UNORM:
-                case PIPE_FORMAT_Z24S8_UNORM:
-                    return R300_EASY_TX_FORMAT(X, X, X, X, W24_FP);
-                default:
-                    return ~0; /* Unsupported. */
-            }
-
-        /* YUV formats. */
-        case UTIL_FORMAT_COLORSPACE_YUV:
-            result |= R300_TX_FORMAT_YUV_TO_RGB;
-
-            switch (format) {
-                case PIPE_FORMAT_YCBCR:
-                    return R300_EASY_TX_FORMAT(X, Y, Z, ONE, YVYU422) | result;
-                case PIPE_FORMAT_YCBCR_REV:
-                    return R300_EASY_TX_FORMAT(X, Y, Z, ONE, VYUY422) | result;
-                default:
-                    return ~0; /* Unsupported/unknown. */
-            }
-
-        /* Add gamma correction. */
-        case UTIL_FORMAT_COLORSPACE_SRGB:
-            result |= R300_TX_FORMAT_GAMMA;
-            break;
-
-        default:;
-    }
-
-    /* Add swizzle. */
-    for (i = 0; i < 4; i++) {
-        switch (desc->swizzle[i]) {
-            case UTIL_FORMAT_SWIZZLE_X:
-            case UTIL_FORMAT_SWIZZLE_NONE:
-                result |= R300_TX_FORMAT_X << swizzle_shift[i];
-                break;
-            case UTIL_FORMAT_SWIZZLE_Y:
-                result |= R300_TX_FORMAT_Y << swizzle_shift[i];
-                break;
-            case UTIL_FORMAT_SWIZZLE_Z:
-                result |= R300_TX_FORMAT_Z << swizzle_shift[i];
-                break;
-            case UTIL_FORMAT_SWIZZLE_W:
-                result |= R300_TX_FORMAT_W << swizzle_shift[i];
-                break;
-            case UTIL_FORMAT_SWIZZLE_0:
-                result |= R300_TX_FORMAT_ZERO << swizzle_shift[i];
-                break;
-            case UTIL_FORMAT_SWIZZLE_1:
-                result |= R300_TX_FORMAT_ONE << swizzle_shift[i];
-                break;
-            default:
-                return ~0; /* Unsupported. */
-        }
-    }
-
-    /* Compressed formats. */
-    if (desc->layout == UTIL_FORMAT_LAYOUT_DXT) {
-        switch (format) {
-            case PIPE_FORMAT_DXT1_RGB:
-            case PIPE_FORMAT_DXT1_RGBA:
-            case PIPE_FORMAT_DXT1_SRGB:
-            case PIPE_FORMAT_DXT1_SRGBA:
-                return R300_TX_FORMAT_DXT1 | result;
-            case PIPE_FORMAT_DXT3_RGBA:
-            case PIPE_FORMAT_DXT3_SRGBA:
-                return R300_TX_FORMAT_DXT3 | result;
-            case PIPE_FORMAT_DXT5_RGBA:
-            case PIPE_FORMAT_DXT5_SRGBA:
-                return R300_TX_FORMAT_DXT5 | result;
-            default:
-                return ~0; /* Unsupported/unknown. */
-        }
-    }
-
-    /* Get the number of components. */
-    for (i = 0; i < 4; i++) {
-        if (desc->channel[i].type != UTIL_FORMAT_TYPE_VOID) {
-            ++components;
-        }
-    }
+boolean r300_is_colorbuffer_format_supported(enum pipe_format format);
 
-    /* Add sign. */
-    for (i = 0; i < components; i++) {
-        if (desc->channel[i].type == UTIL_FORMAT_TYPE_SIGNED) {
-            result |= sign_bit[i];
-        }
-    }
+boolean r300_is_zs_format_supported(enum pipe_format format);
 
-    /* See whether the components are of the same size. */
-    for (i = 1; i < components; i++) {
-        uniform = uniform && desc->channel[0].size == desc->channel[i].size;
-    }
-
-    /* Non-uniform formats. */
-    if (!uniform) {
-        switch (components) {
-            case 3:
-                if (desc->channel[0].size == 5 &&
-                    desc->channel[1].size == 6 &&
-                    desc->channel[2].size == 5) {
-                    return R300_TX_FORMAT_Z5Y6X5 | result;
-                }
-                if (desc->channel[0].size == 5 &&
-                    desc->channel[1].size == 5 &&
-                    desc->channel[2].size == 6) {
-                    return R300_TX_FORMAT_Z6Y5X5 | result;
-                }
-                return ~0; /* Unsupported/unknown. */
-
-            case 4:
-                if (desc->channel[0].size == 5 &&
-                    desc->channel[1].size == 5 &&
-                    desc->channel[2].size == 5 &&
-                    desc->channel[3].size == 1) {
-                    return R300_TX_FORMAT_W1Z5Y5X5 | result;
-                }
-                if (desc->channel[0].size == 10 &&
-                    desc->channel[1].size == 10 &&
-                    desc->channel[2].size == 10 &&
-                    desc->channel[3].size == 2) {
-                    return R300_TX_FORMAT_W2Z10Y10X10 | result;
-                }
-        }
-        return ~0; /* Unsupported/unknown. */
-    }
-
-    /* And finally, uniform formats. */
-    switch (desc->channel[0].type) {
-        case UTIL_FORMAT_TYPE_UNSIGNED:
-        case UTIL_FORMAT_TYPE_SIGNED:
-            if (!desc->channel[0].normalized) {
-                return ~0;
-            }
-
-            switch (desc->channel[0].size) {
-                case 4:
-                    switch (components) {
-                        case 2:
-                            return R300_TX_FORMAT_Y4X4 | result;
-                        case 4:
-                            return R300_TX_FORMAT_W4Z4Y4X4 | result;
-                    }
-                    return ~0;
-
-                case 8:
-                    switch (components) {
-                        case 1:
-                            return R300_TX_FORMAT_X8 | result;
-                        case 2:
-                            return R300_TX_FORMAT_Y8X8 | result;
-                        case 4:
-                            return R300_TX_FORMAT_W8Z8Y8X8 | result;
-                    }
-                    return ~0;
-
-                case 16:
-                    switch (components) {
-                        case 1:
-                            return R300_TX_FORMAT_X16 | result;
-                        case 2:
-                            return R300_TX_FORMAT_Y16X16 | result;
-                        case 4:
-                            return R300_TX_FORMAT_W16Z16Y16X16 | result;
-                    }
-            }
-            return ~0;
-
-/* XXX Enable float textures here. */
-#if 0
-        case UTIL_FORMAT_TYPE_FLOAT:
-            switch (desc->channel[0].size) {
-                case 16:
-                    switch (components) {
-                        case 1:
-                            return R300_TX_FORMAT_16F | result;
-                        case 2:
-                            return R300_TX_FORMAT_16F_16F | result;
-                        case 4:
-                            return R300_TX_FORMAT_16F_16F_16F_16F | result;
-                    }
-                    return ~0;
-
-                case 32:
-                    switch (components) {
-                        case 1:
-                            return R300_TX_FORMAT_32F | result;
-                        case 2:
-                            return R300_TX_FORMAT_32F_32F | result;
-                        case 4:
-                            return R300_TX_FORMAT_32F_32F_32F_32F | result;
-                    }
-            }
-#endif
-    }
-
-    return ~0; /* Unsupported/unknown. */
-}
+boolean r300_is_sampler_format_supported(enum pipe_format format);
 
 struct r300_video_surface
 {
@@ -288,13 +60,11 @@ r300_video_surface(struct pipe_video_surface *pvs)
     return (struct r300_video_surface *)pvs;
 }
 
-#ifndef R300_WINSYS_H
-
+/* Used internally for texture_is_referenced()
+ */
 boolean r300_get_texture_buffer(struct pipe_screen* screen,
                                 struct pipe_texture* texture,
                                 struct pipe_buffer** buffer,
                                 unsigned* stride);
 
-#endif /* R300_WINSYS_H */
-
 #endif /* R300_TEXTURE_H */
diff --git a/src/gallium/drivers/r300/r300_transfer.c b/src/gallium/drivers/r300/r300_transfer.c
new file mode 100644
index 00000000000..ec89681a3c0
--- /dev/null
+++ b/src/gallium/drivers/r300/r300_transfer.c
@@ -0,0 +1,270 @@
+/*
+ * Copyright 2008 Corbin Simpson <[email protected]>
+ * Copyright 2010 Marek Olšák <[email protected]>
+ *
+ * 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
+ * THE AUTHOR(S) 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 "r300_context.h"
+#include "r300_transfer.h"
+#include "r300_texture.h"
+#include "r300_screen.h"
+
+#include "util/u_memory.h"
+#include "util/u_format.h"
+
+struct r300_transfer {
+    /* Parent class */
+    struct pipe_transfer transfer;
+
+    /* Pipe context. */
+    struct pipe_context *ctx;
+
+    /* Parameters of get_tex_transfer. */
+    unsigned x, y, level, zslice, face;
+
+    /* Offset from start of buffer. */
+    unsigned offset;
+
+    /* Detiled texture. */
+    struct r300_texture *detiled_texture;
+
+    /* Transfer and format flags. */
+    unsigned buffer_usage, render_target_usage;
+};
+
+/* Convenience cast wrapper. */
+static INLINE struct r300_transfer*
+r300_transfer(struct pipe_transfer* transfer)
+{
+    return (struct r300_transfer*)transfer;
+}
+
+/* Copy from a tiled texture to a detiled one. */
+static void r300_copy_from_tiled_texture(struct pipe_context *ctx,
+                                         struct r300_transfer *r300transfer)
+{
+    struct pipe_screen *screen = ctx->screen;
+    struct pipe_transfer *transfer = (struct pipe_transfer*)r300transfer;
+    struct pipe_texture *tex = transfer->texture;
+    struct pipe_surface *src, *dst;
+
+    src = screen->get_tex_surface(screen, tex, r300transfer->face,
+                                  r300transfer->level, r300transfer->zslice,
+                                  PIPE_BUFFER_USAGE_GPU_READ |
+                                  PIPE_BUFFER_USAGE_PIXEL);
+
+    dst = screen->get_tex_surface(screen, &r300transfer->detiled_texture->tex,
+                                  0, 0, 0,
+                                  PIPE_BUFFER_USAGE_GPU_WRITE |
+                                  PIPE_BUFFER_USAGE_PIXEL |
+                                  r300transfer->buffer_usage);
+
+    ctx->surface_copy(ctx, dst, 0, 0, src, r300transfer->x, r300transfer->y,
+                      transfer->width, transfer->height);
+
+    pipe_surface_reference(&src, NULL);
+    pipe_surface_reference(&dst, NULL);
+}
+
+/* Copy a detiled texture to a tiled one. */
+static void r300_copy_into_tiled_texture(struct pipe_context *ctx,
+                                         struct r300_transfer *r300transfer)
+{
+    struct pipe_screen *screen = ctx->screen;
+    struct pipe_transfer *transfer = (struct pipe_transfer*)r300transfer;
+    struct pipe_texture *tex = transfer->texture;
+    struct pipe_surface *src, *dst;
+
+    src = screen->get_tex_surface(screen, &r300transfer->detiled_texture->tex,
+                                  0, 0, 0,
+                                  PIPE_BUFFER_USAGE_GPU_READ |
+                                  PIPE_BUFFER_USAGE_PIXEL);
+
+    dst = screen->get_tex_surface(screen, tex, r300transfer->face,
+                                  r300transfer->level, r300transfer->zslice,
+                                  PIPE_BUFFER_USAGE_GPU_WRITE |
+                                  PIPE_BUFFER_USAGE_PIXEL);
+
+    /* XXX this flush prevents the following DRM error from occuring:
+     * [drm:radeon_cs_ioctl] *ERROR* Failed to parse relocation !
+     * Reproducible with perf/copytex. */
+    ctx->flush(ctx, 0, NULL);
+
+    ctx->surface_copy(ctx, dst, r300transfer->x, r300transfer->y, src, 0, 0,
+                      transfer->width, transfer->height);
+
+    /* XXX this flush fixes a few piglit tests (e.g. glean/pixelFormats). */
+    ctx->flush(ctx, 0, NULL);
+
+    pipe_surface_reference(&src, NULL);
+    pipe_surface_reference(&dst, NULL);
+}
+
+static struct pipe_transfer*
+r300_get_tex_transfer(struct pipe_screen *screen,
+                      struct pipe_texture *texture,
+                      unsigned face, unsigned level, unsigned zslice,
+                      enum pipe_transfer_usage usage, unsigned x, unsigned y,
+                      unsigned w, unsigned h)
+{
+    struct r300_texture *tex = (struct r300_texture *)texture;
+    struct r300_transfer *trans;
+    struct r300_screen *r300screen = r300_screen(screen);
+    struct pipe_texture template;
+
+    trans = CALLOC_STRUCT(r300_transfer);
+    if (trans) {
+        /* Initialize the transfer object. */
+        pipe_texture_reference(&trans->transfer.texture, texture);
+        trans->transfer.usage = usage;
+        trans->transfer.width = w;
+        trans->transfer.height = h;
+        trans->ctx = r300screen->ctx;
+        trans->x = x;
+        trans->y = y;
+        trans->level = level;
+        trans->zslice = zslice;
+        trans->face = face;
+
+        /* If the texture is tiled, we must create a temporary detiled texture
+         * for this transfer. */
+        if (tex->microtile || tex->macrotile) {
+            trans->buffer_usage = pipe_transfer_buffer_flags(&trans->transfer);
+            trans->render_target_usage =
+                util_format_is_depth_or_stencil(texture->format) ?
+                PIPE_TEXTURE_USAGE_DEPTH_STENCIL :
+                PIPE_TEXTURE_USAGE_RENDER_TARGET;
+
+            template.target = PIPE_TEXTURE_2D;
+            template.format = texture->format;
+            template.width0 = w;
+            template.height0 = h;
+            template.depth0 = 0;
+            template.last_level = 0;
+            template.nr_samples = 0;
+            template.tex_usage = PIPE_TEXTURE_USAGE_DYNAMIC |
+                                 R300_TEXTURE_USAGE_TRANSFER;
+
+            /* For texture reading, the temporary (detiled) texture is used as
+             * a render target when blitting from a tiled texture. */
+            if (usage & PIPE_TRANSFER_READ) {
+                template.tex_usage |= trans->render_target_usage;
+            }
+            /* For texture writing, the temporary texture is used as a sampler
+             * when blitting into a tiled texture. */
+            if (usage & PIPE_TRANSFER_WRITE) {
+                template.tex_usage |= PIPE_TEXTURE_USAGE_SAMPLER;
+            }
+
+            /* Create the temporary texture. */
+            trans->detiled_texture =
+                (struct r300_texture*)screen->texture_create(screen, &template);
+            assert(!trans->detiled_texture->microtile &&
+                   !trans->detiled_texture->macrotile);
+
+            /* Set the stride.
+             * Parameters x, y, level, zslice, and face remain zero. */
+            trans->transfer.stride =
+                r300_texture_get_stride(r300screen, trans->detiled_texture, 0);
+
+            if (usage & PIPE_TRANSFER_READ) {
+                /* We cannot map a tiled texture directly because the data is
+                 * in a different order, therefore we do detiling using a blit. */
+                r300_copy_from_tiled_texture(r300screen->ctx, trans);
+            }
+        } else {
+            trans->transfer.x = x;
+            trans->transfer.y = y;
+            trans->transfer.stride =
+                r300_texture_get_stride(r300screen, tex, level);
+            trans->transfer.level = level;
+            trans->transfer.zslice = zslice;
+            trans->transfer.face = face;
+            trans->offset = r300_texture_get_offset(tex, level, zslice, face);
+        }
+    }
+    return &trans->transfer;
+}
+
+static void r300_tex_transfer_destroy(struct pipe_transfer *trans)
+{
+    struct r300_transfer *r300transfer = r300_transfer(trans);
+
+    if (r300transfer->detiled_texture) {
+        if (trans->usage & PIPE_TRANSFER_WRITE) {
+            r300_copy_into_tiled_texture(r300transfer->ctx, r300transfer);
+        }
+
+        pipe_texture_reference(
+            (struct pipe_texture**)&r300transfer->detiled_texture, NULL);
+    }
+    pipe_texture_reference(&trans->texture, NULL);
+    FREE(trans);
+}
+
+static void* r300_transfer_map(struct pipe_screen *screen,
+                               struct pipe_transfer *transfer)
+{
+    struct r300_transfer *r300transfer = r300_transfer(transfer);
+    struct r300_texture *tex = (struct r300_texture*)transfer->texture;
+    char *map;
+    enum pipe_format format = tex->tex.format;
+
+    if (r300transfer->detiled_texture) {
+        /* The detiled texture is of the same size as the region being mapped
+         * (no offset needed). */
+        return pipe_buffer_map(screen,
+                               r300transfer->detiled_texture->buffer,
+                               pipe_transfer_buffer_flags(transfer));
+    } else {
+        /* Tiling is disabled. */
+        map = pipe_buffer_map(screen, tex->buffer,
+                              pipe_transfer_buffer_flags(transfer));
+
+        if (!map) {
+            return NULL;
+        }
+
+        return map + r300_transfer(transfer)->offset +
+            transfer->y / util_format_get_blockheight(format) * transfer->stride +
+            transfer->x / util_format_get_blockwidth(format) * util_format_get_blocksize(format);
+    }
+}
+
+static void r300_transfer_unmap(struct pipe_screen *screen,
+                                struct pipe_transfer *transfer)
+{
+    struct r300_transfer *r300transfer = r300_transfer(transfer);
+    struct r300_texture *tex = (struct r300_texture*)transfer->texture;
+
+    if (r300transfer->detiled_texture) {
+        pipe_buffer_unmap(screen, r300transfer->detiled_texture->buffer);
+    } else {
+        pipe_buffer_unmap(screen, tex->buffer);
+    }
+}
+
+void r300_init_screen_transfer_functions(struct pipe_screen *screen)
+{
+    screen->get_tex_transfer = r300_get_tex_transfer;
+    screen->tex_transfer_destroy = r300_tex_transfer_destroy;
+    screen->transfer_map = r300_transfer_map;
+    screen->transfer_unmap = r300_transfer_unmap;
+}
diff --git a/src/gallium/drivers/r300/r300_transfer.h b/src/gallium/drivers/r300/r300_transfer.h
new file mode 100644
index 00000000000..60d1d3dc85c
--- /dev/null
+++ b/src/gallium/drivers/r300/r300_transfer.h
@@ -0,0 +1,31 @@
+/*
+ * Copyright 2008 Corbin Simpson <[email protected]>
+ * Copyright 2010 Marek Olšák <[email protected]>
+ *
+ * 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
+ * THE AUTHOR(S) 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 R300_TRANSFER
+#define R300_TRANSFER
+
+#include "pipe/p_screen.h"
+
+void r300_init_screen_transfer_functions(struct pipe_screen *screen);
+
+#endif
diff --git a/src/gallium/drivers/r300/r300_vs.c b/src/gallium/drivers/r300/r300_vs.c
index a6786c321c6..379939ac750 100644
--- a/src/gallium/drivers/r300/r300_vs.c
+++ b/src/gallium/drivers/r300/r300_vs.c
@@ -89,95 +89,41 @@ static void r300_shader_read_vs_outputs(
                 assert(0);
         }
     }
+
+    /* WPOS is a straight copy of POSITION and it's always emitted. */
+    vs_outputs->wpos = i;
 }
 
-static void r300_shader_vap_output_fmt(struct r300_vertex_shader* vs)
+/* This function sets up:
+ * - VAP mapping, which maps VS registers to output semantics and
+ *   at the same time it indicates which attributes are enabled and should
+ *   be rasterized.
+ * - Stream mapping to VS outputs if TCL is not present. */
+static void r300_init_vs_output_mapping(struct r300_vertex_shader* vs)
 {
     struct r300_shader_semantics* vs_outputs = &vs->outputs;
-    uint32_t* hwfmt = vs->hwfmt;
-    int i, gen_count;
+    struct r300_vap_output_state *vap_out = &vs->vap_out;
+    int *stream_loc = vs->stream_loc_notcl;
+    int i, gen_count, tabi = 0;
     boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED ||
                               vs_outputs->bcolor[1] != ATTR_UNUSED;
 
-    /* Do the actual vertex_info setup.
-     *
-     * vertex_info has four uints of hardware-specific data in it.
-     * vinfo.hwfmt[0] is R300_VAP_VTX_STATE_CNTL
-     * vinfo.hwfmt[1] is R300_VAP_VSM_VTX_ASSM
-     * vinfo.hwfmt[2] is R300_VAP_OUTPUT_VTX_FMT_0
-     * vinfo.hwfmt[3] is R300_VAP_OUTPUT_VTX_FMT_1 */
-
-    hwfmt[0] = 0x5555; /* XXX this is classic Mesa bonghits */
+    vap_out->vap_vtx_state_cntl = 0x5555; /* XXX this is classic Mesa bonghits */
 
     /* Position. */
     if (vs_outputs->pos != ATTR_UNUSED) {
-        hwfmt[1] |= R300_INPUT_CNTL_POS;
-        hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
+        vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_POS;
+        vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT;
+
+        stream_loc[tabi++] = 0;
     } else {
         assert(0);
     }
 
     /* Point size. */
     if (vs_outputs->psize != ATTR_UNUSED) {
-        hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
-    }
-
-    /* Colors. */
-    for (i = 0; i < ATTR_COLOR_COUNT; i++) {
-        if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used ||
-            vs_outputs->color[1] != ATTR_UNUSED) {
-            hwfmt[1] |= R300_INPUT_CNTL_COLOR;
-            hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i;
-        }
-    }
-
-    /* Back-face colors. */
-    if (any_bcolor_used) {
-        for (i = 0; i < ATTR_COLOR_COUNT; i++) {
-            hwfmt[1] |= R300_INPUT_CNTL_COLOR;
-            hwfmt[2] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << (2+i);
-        }
-    }
+        vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__PT_SIZE_PRESENT;
 
-    /* Texture coordinates. */
-    gen_count = 0;
-    for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
-        if (vs_outputs->generic[i] != ATTR_UNUSED) {
-            hwfmt[1] |= (R300_INPUT_CNTL_TC0 << gen_count);
-            hwfmt[3] |= (4 << (3 * gen_count));
-            gen_count++;
-        }
-    }
-
-    /* Fog coordinates. */
-    if (vs_outputs->fog != ATTR_UNUSED) {
-        hwfmt[1] |= (R300_INPUT_CNTL_TC0 << gen_count);
-        hwfmt[3] |= (4 << (3 * gen_count));
-        gen_count++;
-    }
-
-    /* XXX magic */
-    assert(gen_count <= 8);
-
-    /* WPOS. */
-    vs->wpos_tex_output = gen_count;
-}
-
-/* Sets up stream mapping to equivalent VS outputs if TCL is bypassed
- * or isn't present. */
-static void r300_stream_locations_notcl(
-    struct r300_shader_semantics* vs_outputs,
-    int* stream_loc)
-{
-    int i, tabi = 0, gen_count;
-    boolean any_bcolor_used = vs_outputs->bcolor[0] != ATTR_UNUSED ||
-                              vs_outputs->bcolor[1] != ATTR_UNUSED;
-
-    /* Position. */
-    stream_loc[tabi++] = 0;
-
-    /* Point size. */
-    if (vs_outputs->psize != ATTR_UNUSED) {
         stream_loc[tabi++] = 1;
     }
 
@@ -185,6 +131,9 @@ static void r300_stream_locations_notcl(
     for (i = 0; i < ATTR_COLOR_COUNT; i++) {
         if (vs_outputs->color[i] != ATTR_UNUSED || any_bcolor_used ||
             vs_outputs->color[1] != ATTR_UNUSED) {
+            vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_COLOR;
+            vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << i;
+
             stream_loc[tabi++] = 2 + i;
         }
     }
@@ -192,6 +141,9 @@ static void r300_stream_locations_notcl(
     /* Back-face colors. */
     if (any_bcolor_used) {
         for (i = 0; i < ATTR_COLOR_COUNT; i++) {
+            vap_out->vap_vsm_vtx_assm |= R300_INPUT_CNTL_COLOR;
+            vap_out->vap_out_vtx_fmt[0] |= R300_VAP_OUTPUT_VTX_FMT_0__COLOR_0_PRESENT << (2+i);
+
             stream_loc[tabi++] = 4 + i;
         }
     }
@@ -200,6 +152,9 @@ static void r300_stream_locations_notcl(
     gen_count = 0;
     for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
         if (vs_outputs->generic[i] != ATTR_UNUSED) {
+            vap_out->vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << gen_count);
+            vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * gen_count));
+
             assert(tabi < 16);
             stream_loc[tabi++] = 6 + gen_count;
             gen_count++;
@@ -208,17 +163,22 @@ static void r300_stream_locations_notcl(
 
     /* Fog coordinates. */
     if (vs_outputs->fog != ATTR_UNUSED) {
+        vap_out->vap_vsm_vtx_assm |= (R300_INPUT_CNTL_TC0 << gen_count);
+        vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * gen_count));
+
         assert(tabi < 16);
         stream_loc[tabi++] = 6 + gen_count;
         gen_count++;
     }
 
+    /* XXX magic */
+    assert(gen_count <= 8);
+
     /* WPOS. */
-    if (vs_outputs->wpos != ATTR_UNUSED) {
-        assert(tabi < 16);
-        stream_loc[tabi++] = 6 + gen_count;
-        gen_count++;
-    }
+    vs->wpos_tex_output = gen_count;
+
+    assert(tabi < 16);
+    stream_loc[tabi++] = 6 + gen_count;
 
     for (; tabi < 16;) {
         stream_loc[tabi++] = -1;
@@ -294,26 +254,16 @@ static void set_vertex_inputs_outputs(struct r300_vertex_program_compiler * c)
     }
 }
 
-static void r300_insert_wpos(struct r300_vertex_program_compiler* c,
-                             struct r300_shader_semantics* outputs)
+void r300_vertex_shader_common_init(struct r300_vertex_shader *vs,
+                                    const struct pipe_shader_state *shader)
 {
-    int i, lastOutput = 0;
-
-    /* Find the max output index. */
-    lastOutput = MAX2(lastOutput, outputs->psize);
-    for (i = 0; i < ATTR_COLOR_COUNT; i++) {
-        lastOutput = MAX2(lastOutput, outputs->color[i]);
-        lastOutput = MAX2(lastOutput, outputs->bcolor[i]);
-    }
-    for (i = 0; i < ATTR_GENERIC_COUNT; i++) {
-        lastOutput = MAX2(lastOutput, outputs->generic[i]);
-    }
-    lastOutput = MAX2(lastOutput, outputs->fog);
+    /* Copy state directly into shader. */
+    vs->state = *shader;
+    vs->state.tokens = tgsi_dup_tokens(shader->tokens);
+    tgsi_scan_shader(shader->tokens, &vs->info);
 
-    /* Set WPOS after the last output. */
-    lastOutput++;
-    rc_copy_output(&c->Base, 0, lastOutput); /* out[lastOutput] = out[0]; */
-    outputs->wpos = lastOutput;
+    r300_shader_read_vs_outputs(&vs->info, &vs->outputs);
+    r300_init_vs_output_mapping(vs);
 }
 
 void r300_translate_vertex_shader(struct r300_context* r300,
@@ -322,9 +272,6 @@ void r300_translate_vertex_shader(struct r300_context* r300,
     struct r300_vertex_program_compiler compiler;
     struct tgsi_to_rc ttr;
 
-    /* Initialize. */
-    r300_shader_read_vs_outputs(&vs->info, &vs->outputs);
-
     /* Setup the compiler */
     rc_init(&compiler.Base);
 
@@ -348,10 +295,7 @@ void r300_translate_vertex_shader(struct r300_context* r300,
     compiler.SetHwInputOutput = &set_vertex_inputs_outputs;
 
     /* Insert the WPOS output. */
-    r300_insert_wpos(&compiler, &vs->outputs);
-
-    r300_shader_vap_output_fmt(vs);
-    r300_stream_locations_notcl(&vs->outputs, vs->stream_loc_notcl);
+    rc_copy_output(&compiler.Base, 0, vs->outputs.wpos);
 
     /* Invoke the compiler */
     r3xx_compile_vertex_program(&compiler);
@@ -363,30 +307,29 @@ void r300_translate_vertex_shader(struct r300_context* r300,
 
     /* And, finally... */
     rc_destroy(&compiler.Base);
-    vs->translated = TRUE;
 }
 
 boolean r300_vertex_shader_setup_wpos(struct r300_context* r300)
 {
-    struct r300_vertex_shader* vs = r300->vs;
-    int tex_output = r300->vs->wpos_tex_output;
+    struct r300_vertex_shader* vs = r300->vs_state.state;
+    struct r300_vap_output_state *vap_out = &vs->vap_out;
+    int tex_output = vs->wpos_tex_output;
     uint32_t tex_fmt = R300_INPUT_CNTL_TC0 << tex_output;
-    uint32_t* hwfmt = vs->hwfmt;
 
     if (r300->fs->inputs.wpos != ATTR_UNUSED) {
         /* Enable WPOS in VAP. */
-        if (!(hwfmt[1] & tex_fmt)) {
-            hwfmt[1] |= tex_fmt;
-            hwfmt[3] |= (4 << (3 * tex_output));
+        if (!(vap_out->vap_vsm_vtx_assm & tex_fmt)) {
+            vap_out->vap_vsm_vtx_assm |= tex_fmt;
+            vap_out->vap_out_vtx_fmt[1] |= (4 << (3 * tex_output));
 
             assert(tex_output < 8);
             return TRUE;
         }
     } else {
         /* Disable WPOS in VAP. */
-        if (hwfmt[1] & tex_fmt) {
-            hwfmt[1] &= ~tex_fmt;
-            hwfmt[3] &= ~(4 << (3 * tex_output));
+        if (vap_out->vap_vsm_vtx_assm & tex_fmt) {
+            vap_out->vap_vsm_vtx_assm &= ~tex_fmt;
+            vap_out->vap_out_vtx_fmt[1] &= ~(4 << (3 * tex_output));
             return TRUE;
         }
     }
diff --git a/src/gallium/drivers/r300/r300_vs.h b/src/gallium/drivers/r300/r300_vs.h
index 18cfeee3cd4..f6f0b86b683 100644
--- a/src/gallium/drivers/r300/r300_vs.h
+++ b/src/gallium/drivers/r300/r300_vs.h
@@ -28,6 +28,7 @@
 #include "tgsi/tgsi_scan.h"
 #include "radeon_code.h"
 
+#include "r300_context.h"
 #include "r300_shader_semantics.h"
 
 struct r300_context;
@@ -38,7 +39,7 @@ struct r300_vertex_shader {
 
     struct tgsi_shader_info info;
     struct r300_shader_semantics outputs;
-    uint hwfmt[4];
+    struct r300_vap_output_state vap_out;
 
     /* Stream locations for SWTCL or if TCL is bypassed. */
     int stream_loc_notcl[16];
@@ -46,13 +47,17 @@ struct r300_vertex_shader {
     /* Output stream location for WPOS. */
     int wpos_tex_output;
 
-    /* Has this shader been translated yet? */
-    boolean translated;
-
+    /* HWTCL-specific.  */
     /* Machine code (if translated) */
     struct r300_vertex_program_code code;
+
+    /* SWTCL-specific. */
+    void *draw_vs;
 };
 
+void r300_vertex_shader_common_init(struct r300_vertex_shader *vs,
+                                    const struct pipe_shader_state *shader);
+
 void r300_translate_vertex_shader(struct r300_context* r300,
                                   struct r300_vertex_shader* vs);
 
diff --git a/src/gallium/drivers/r300/r300_winsys.h b/src/gallium/drivers/r300/r300_winsys.h
index 40fb8a95ca5..ddf2b790039 100644
--- a/src/gallium/drivers/r300/r300_winsys.h
+++ b/src/gallium/drivers/r300/r300_winsys.h
@@ -40,11 +40,6 @@ struct radeon_winsys;
 struct pipe_screen* r300_create_screen(struct radeon_winsys* radeon_winsys);
 
 
-boolean r300_get_texture_buffer(struct pipe_screen* screen,
-                                struct pipe_texture* texture,
-                                struct pipe_buffer** buffer,
-                                unsigned* stride);
-
 #ifdef __cplusplus
 }
 #endif
diff --git a/src/gallium/drivers/softpipe/Makefile b/src/gallium/drivers/softpipe/Makefile
index e4ac49fa85f..1c6e4ae0768 100644
--- a/src/gallium/drivers/softpipe/Makefile
+++ b/src/gallium/drivers/softpipe/Makefile
@@ -6,6 +6,7 @@ LIBNAME = softpipe
 C_SOURCES = \
 	sp_fs_exec.c \
 	sp_fs_sse.c \
+	sp_buffer.c \
 	sp_clear.c \
 	sp_flush.c \
 	sp_query.c \
@@ -32,7 +33,6 @@ C_SOURCES = \
 	sp_tex_tile_cache.c \
 	sp_tile_cache.c \
 	sp_surface.c \
-	sp_video_context.c \
-	sp_winsys.c
+	sp_video_context.c
 
 include ../../Makefile.template
diff --git a/src/gallium/drivers/softpipe/SConscript b/src/gallium/drivers/softpipe/SConscript
index 3042e556c64..27ab00b036e 100644
--- a/src/gallium/drivers/softpipe/SConscript
+++ b/src/gallium/drivers/softpipe/SConscript
@@ -7,6 +7,7 @@ softpipe = env.ConvenienceLibrary(
 	source = [
 		'sp_fs_exec.c',
 		'sp_fs_sse.c',
+		'sp_buffer.c',
 		'sp_clear.c',
 		'sp_context.c',
 		'sp_draw_arrays.c',
@@ -33,8 +34,7 @@ softpipe = env.ConvenienceLibrary(
 		'sp_tex_tile_cache.c',
 		'sp_texture.c',
 		'sp_tile_cache.c',
-		'sp_video_context.c',
-		'sp_winsys.c'
+		'sp_video_context.c'
 	])
 
 Export('softpipe')
diff --git a/src/gallium/drivers/softpipe/sp_buffer.c b/src/gallium/drivers/softpipe/sp_buffer.c
new file mode 100644
index 00000000000..8f390250869
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_buffer.c
@@ -0,0 +1,118 @@
+/**************************************************************************
+ *
+ * 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 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.
+ *
+ **************************************************************************/
+
+
+#include "util/u_inlines.h"
+#include "util/u_memory.h"
+#include "util/u_math.h"
+
+#include "sp_screen.h"
+#include "sp_buffer.h"
+
+
+static void *
+softpipe_buffer_map(struct pipe_screen *screen,
+                    struct pipe_buffer *buf,
+                    unsigned flags)
+{
+   struct softpipe_buffer *softpipe_buf = softpipe_buffer(buf);
+   return softpipe_buf->data;
+}
+
+
+static void
+softpipe_buffer_unmap(struct pipe_screen *screen,
+                      struct pipe_buffer *buf)
+{
+}
+
+
+static void
+softpipe_buffer_destroy(struct pipe_buffer *buf)
+{
+   struct softpipe_buffer *sbuf = softpipe_buffer(buf);
+
+   if (!sbuf->userBuffer)
+      align_free(sbuf->data);
+      
+   FREE(sbuf);
+}
+
+
+static struct pipe_buffer *
+softpipe_buffer_create(struct pipe_screen *screen,
+                       unsigned alignment,
+                       unsigned usage,
+                       unsigned size)
+{
+   struct softpipe_buffer *buffer = CALLOC_STRUCT(softpipe_buffer);
+
+   pipe_reference_init(&buffer->base.reference, 1);
+   buffer->base.screen = screen;
+   buffer->base.alignment = MAX2(alignment, 16);
+   buffer->base.usage = usage;
+   buffer->base.size = size;
+
+   buffer->data = align_malloc(size, alignment);
+
+   return &buffer->base;
+}
+
+
+/**
+ * Create buffer which wraps user-space data.
+ */
+static struct pipe_buffer *
+softpipe_user_buffer_create(struct pipe_screen *screen,
+                            void *ptr,
+                            unsigned bytes)
+{
+   struct softpipe_buffer *buffer;
+
+   buffer = CALLOC_STRUCT(softpipe_buffer);
+   if(!buffer)
+      return NULL;
+
+   pipe_reference_init(&buffer->base.reference, 1);
+   buffer->base.screen = screen;
+   buffer->base.size = bytes;
+   buffer->userBuffer = TRUE;
+   buffer->data = ptr;
+
+   return &buffer->base;
+}
+
+
+void
+softpipe_init_screen_buffer_funcs(struct pipe_screen *screen)
+{
+   screen->buffer_create = softpipe_buffer_create;
+   screen->user_buffer_create = softpipe_user_buffer_create;
+   screen->buffer_map = softpipe_buffer_map;
+   screen->buffer_unmap = softpipe_buffer_unmap;
+   screen->buffer_destroy = softpipe_buffer_destroy;
+}
diff --git a/src/gallium/drivers/softpipe/sp_winsys.h b/src/gallium/drivers/softpipe/sp_buffer.h
index 6e3920c49b2..9d8e56a176b 100644
--- a/src/gallium/drivers/softpipe/sp_winsys.h
+++ b/src/gallium/drivers/softpipe/sp_buffer.h
@@ -1,8 +1,8 @@
 /**************************************************************************
- * 
- * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ *
+ * 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
@@ -10,64 +10,46 @@
  * 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
+ * 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.
- * 
+ *
  **************************************************************************/
 
-/* This is the interface that softpipe requires any window system
- * hosting it to implement.  This is the only include file in softpipe
- * which is public.
- */
+#ifndef SP_BUFFER_H
+#define SP_BUFFER_H
 
+#include "pipe/p_compiler.h"
+#include "pipe/p_state.h"
 
-#ifndef SP_WINSYS_H
-#define SP_WINSYS_H
 
-#ifdef __cplusplus
-extern "C" {
-#endif
+struct softpipe_buffer
+{
+   struct pipe_buffer base;
+   boolean userBuffer;  /** Is this a user-space buffer? */
+   void *data;
+};
 
-#include "pipe/p_defines.h"
 
-struct pipe_screen;
-struct pipe_winsys;
-struct pipe_context;
-struct pipe_texture;
-struct pipe_buffer;
-
-
-
-/**
- * Create a softpipe screen that uses the
- * given winsys for allocating buffers.
- */
-struct pipe_screen *softpipe_create_screen( struct pipe_winsys * );
-
-/**
- * Create a softpipe screen that uses
- * regular malloc to create all its buffers.
- */
-struct pipe_screen *softpipe_create_screen_malloc(void);
+/** Cast wrapper */
+static INLINE struct softpipe_buffer *
+softpipe_buffer( struct pipe_buffer *buf )
+{
+   return (struct softpipe_buffer *)buf;
+}
 
-boolean
-softpipe_get_texture_buffer( struct pipe_texture *texture,
-                             struct pipe_buffer **buf,
-                             unsigned *stride );
 
+void
+softpipe_init_screen_buffer_funcs(struct pipe_screen *screen);
 
-#ifdef __cplusplus
-}
-#endif
 
-#endif /* SP_WINSYS_H */
+#endif /* SP_BUFFER_H */
diff --git a/src/gallium/drivers/softpipe/sp_context.c b/src/gallium/drivers/softpipe/sp_context.c
index 0c74b490fea..891a903e4f1 100644
--- a/src/gallium/drivers/softpipe/sp_context.c
+++ b/src/gallium/drivers/softpipe/sp_context.c
@@ -210,7 +210,7 @@ softpipe_create_context( struct pipe_screen *screen,
    softpipe->dump_fs = debug_get_bool_option( "GALLIUM_DUMP_FS", FALSE );
    softpipe->dump_gs = debug_get_bool_option( "SOFTPIPE_DUMP_GS", FALSE );
 
-   softpipe->pipe.winsys = screen->winsys;
+   softpipe->pipe.winsys = NULL;
    softpipe->pipe.screen = screen;
    softpipe->pipe.destroy = softpipe_destroy;
    softpipe->pipe.priv = priv;
@@ -245,6 +245,10 @@ softpipe_create_context( struct pipe_screen *screen,
    softpipe->pipe.bind_gs_state   = softpipe_bind_gs_state;
    softpipe->pipe.delete_gs_state = softpipe_delete_gs_state;
 
+   softpipe->pipe.create_vertex_elements_state = softpipe_create_vertex_elements_state;
+   softpipe->pipe.bind_vertex_elements_state = softpipe_bind_vertex_elements_state;
+   softpipe->pipe.delete_vertex_elements_state = softpipe_delete_vertex_elements_state;
+
    softpipe->pipe.set_blend_color = softpipe_set_blend_color;
    softpipe->pipe.set_stencil_ref = softpipe_set_stencil_ref;
    softpipe->pipe.set_clip_state = softpipe_set_clip_state;
@@ -259,7 +263,6 @@ softpipe_create_context( struct pipe_screen *screen,
    softpipe->pipe.set_viewport_state = softpipe_set_viewport_state;
 
    softpipe->pipe.set_vertex_buffers = softpipe_set_vertex_buffers;
-   softpipe->pipe.set_vertex_elements = softpipe_set_vertex_elements;
 
    softpipe->pipe.draw_arrays = softpipe_draw_arrays;
    softpipe->pipe.draw_elements = softpipe_draw_elements;
diff --git a/src/gallium/drivers/softpipe/sp_context.h b/src/gallium/drivers/softpipe/sp_context.h
index 2f0f51de2c4..75e03c8ae6b 100644
--- a/src/gallium/drivers/softpipe/sp_context.h
+++ b/src/gallium/drivers/softpipe/sp_context.h
@@ -45,6 +45,7 @@ struct softpipe_tile_cache;
 struct softpipe_tex_tile_cache;
 struct sp_fragment_shader;
 struct sp_vertex_shader;
+struct sp_velems_state;
 
 
 struct softpipe_context {
@@ -59,6 +60,7 @@ struct softpipe_context {
    struct sp_fragment_shader *fs;
    struct sp_vertex_shader *vs;
    struct sp_geometry_shader *gs;
+   struct sp_velems_state *velems;
 
    /** Other rendering state */
    struct pipe_blend_color blend_color;
@@ -72,13 +74,11 @@ struct softpipe_context {
    struct pipe_sampler_view *vertex_sampler_views[PIPE_MAX_VERTEX_SAMPLERS];
    struct pipe_viewport_state viewport;
    struct pipe_vertex_buffer vertex_buffer[PIPE_MAX_ATTRIBS];
-   struct pipe_vertex_element vertex_element[PIPE_MAX_ATTRIBS];
 
    unsigned num_samplers;
    unsigned num_sampler_views;
    unsigned num_vertex_samplers;
    unsigned num_vertex_sampler_views;
-   unsigned num_vertex_elements;
    unsigned num_vertex_buffers;
 
    unsigned dirty; /**< Mask of SP_NEW_x flags */
@@ -93,7 +93,7 @@ struct softpipe_context {
    ubyte *mapped_vbuffer[PIPE_MAX_ATTRIBS];
 
    /** Mapped constant buffers */
-   void *mapped_constants[PIPE_SHADER_TYPES][PIPE_MAX_CONSTANT_BUFFERS];
+   const void *mapped_constants[PIPE_SHADER_TYPES][PIPE_MAX_CONSTANT_BUFFERS];
 
    /** Vertex format */
    struct vertex_info vertex_info;
diff --git a/src/gallium/drivers/softpipe/sp_draw_arrays.c b/src/gallium/drivers/softpipe/sp_draw_arrays.c
index b2acc36bf7a..7b77eb239fd 100644
--- a/src/gallium/drivers/softpipe/sp_draw_arrays.c
+++ b/src/gallium/drivers/softpipe/sp_draw_arrays.c
@@ -33,90 +33,19 @@
 
 #include "pipe/p_defines.h"
 #include "pipe/p_context.h"
-#include "util/u_simple_screen.h"
 #include "util/u_inlines.h"
 #include "util/u_prim.h"
 
 #include "sp_context.h"
 #include "sp_query.h"
 #include "sp_state.h"
+#include "sp_buffer.h"
 
 #include "draw/draw_context.h"
 
 
 
-static void
-softpipe_map_constant_buffers(struct softpipe_context *sp)
-{
-   struct pipe_winsys *ws = sp->pipe.winsys;
-   uint i;
-
-   for (i = 0; i < PIPE_SHADER_TYPES; i++) {
-      uint j;
-
-      for (j = 0; j < PIPE_MAX_CONSTANT_BUFFERS; j++) {
-         if (sp->constants[i][j] && sp->constants[i][j]->size) {
-            sp->mapped_constants[i][j] = ws->buffer_map(ws,
-                                                        sp->constants[i][j],
-                                                        PIPE_BUFFER_USAGE_CPU_READ);
-         }
-      }
-   }
-
-   for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
-      if (sp->constants[PIPE_SHADER_VERTEX][i]) {
-         draw_set_mapped_constant_buffer(sp->draw,
-                                         PIPE_SHADER_VERTEX,
-                                         i,
-                                         sp->mapped_constants[PIPE_SHADER_VERTEX][i],
-                                         sp->constants[PIPE_SHADER_VERTEX][i]->size);
-      }
-      if (sp->constants[PIPE_SHADER_GEOMETRY][i]) {
-         draw_set_mapped_constant_buffer(sp->draw,
-                                         PIPE_SHADER_GEOMETRY,
-                                         i,
-                                         sp->mapped_constants[PIPE_SHADER_GEOMETRY][i],
-                                         sp->constants[PIPE_SHADER_GEOMETRY][i]->size);
-      }
-   }
-}
-
-
-static void
-softpipe_unmap_constant_buffers(struct softpipe_context *sp)
-{
-   struct pipe_winsys *ws = sp->pipe.winsys;
-   uint i;
 
-   /* really need to flush all prims since the vert/frag shaders const buffers
-    * are going away now.
-    */
-   draw_flush(sp->draw);
-
-   for (i = 0; i < PIPE_MAX_CONSTANT_BUFFERS; i++) {
-      draw_set_mapped_constant_buffer(sp->draw,
-                                      PIPE_SHADER_VERTEX,
-                                      i,
-                                      NULL,
-                                      0);
-      draw_set_mapped_constant_buffer(sp->draw,
-                                      PIPE_SHADER_GEOMETRY,
-                                      i,
-                                      NULL,
-                                      0);
-   }
-
-   for (i = 0; i < PIPE_SHADER_TYPES; i++) {
-      uint j;
-
-      for (j = 0; j < PIPE_MAX_CONSTANT_BUFFERS; j++) {
-         if (sp->constants[i][j] && sp->constants[i][j]->size) {
-            ws->buffer_unmap(ws, sp->constants[i][j]);
-         }
-         sp->mapped_constants[i][j] = NULL;
-      }
-   }
-}
 
 
 /**
@@ -261,25 +190,16 @@ softpipe_draw_range_elements_instanced(struct pipe_context *pipe,
    }
 
    softpipe_map_transfers(sp);
-   softpipe_map_constant_buffers(sp);
 
    /* Map vertex buffers */
    for (i = 0; i < sp->num_vertex_buffers; i++) {
-      void *buf;
-
-      buf = pipe_buffer_map(pipe->screen,
-                            sp->vertex_buffer[i].buffer,
-                            PIPE_BUFFER_USAGE_CPU_READ);
+      void *buf = softpipe_buffer(sp->vertex_buffer[i].buffer)->data;
       draw_set_mapped_vertex_buffer(draw, i, buf);
    }
 
    /* Map index buffer, if present */
    if (indexBuffer) {
-      void *mapped_indexes;
-
-      mapped_indexes = pipe_buffer_map(pipe->screen,
-                                       indexBuffer,
-                                       PIPE_BUFFER_USAGE_CPU_READ);
+      void *mapped_indexes = softpipe_buffer(indexBuffer)->data;
       draw_set_mapped_element_buffer_range(draw,
                                            indexSize,
                                            minIndex,
@@ -300,15 +220,18 @@ softpipe_draw_range_elements_instanced(struct pipe_context *pipe,
    /* unmap vertex/index buffers - will cause draw module to flush */
    for (i = 0; i < sp->num_vertex_buffers; i++) {
       draw_set_mapped_vertex_buffer(draw, i, NULL);
-      pipe_buffer_unmap(pipe->screen, sp->vertex_buffer[i].buffer);
    }
    if (indexBuffer) {
       draw_set_mapped_element_buffer(draw, 0, NULL);
-      pipe_buffer_unmap(pipe->screen, indexBuffer);
    }
 
-   /* Note: leave drawing surfaces mapped */
-   softpipe_unmap_constant_buffers(sp);
+   /*
+    * TODO: Flush only when a user vertex/index buffer is present
+    * (or even better, modify draw module to do this
+    * internally when this condition is seen?)
+    */
+   draw_flush(draw);
 
+   /* Note: leave drawing surfaces mapped */
    sp->dirty_render_cache = TRUE;
 }
diff --git a/src/gallium/drivers/softpipe/sp_prim_vbuf.c b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
index 98c08eaffaf..6749243ab41 100644
--- a/src/gallium/drivers/softpipe/sp_prim_vbuf.c
+++ b/src/gallium/drivers/softpipe/sp_prim_vbuf.c
@@ -264,57 +264,29 @@ sp_vbuf_draw(struct vbuf_render *vbr, const ushort *indices, uint nr)
       break;
 
    case PIPE_PRIM_QUADS:
-      if (softpipe->rasterizer->flatshade_first) {
-         for (i = 3; i < nr; i += 4) {
-            sp_setup_tri( setup_ctx,
-                       get_vert(vertex_buffer, indices[i-2], stride),
-                       get_vert(vertex_buffer, indices[i-1], stride),
-                       get_vert(vertex_buffer, indices[i-3], stride) );
-            sp_setup_tri( setup_ctx,
-                       get_vert(vertex_buffer, indices[i-1], stride),
-                       get_vert(vertex_buffer, indices[i-0], stride),
-                       get_vert(vertex_buffer, indices[i-3], stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 4) {
-            sp_setup_tri( setup_ctx,
+      for (i = 3; i < nr; i += 4) {
+         sp_setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[i-3], stride),
                        get_vert(vertex_buffer, indices[i-2], stride),
                        get_vert(vertex_buffer, indices[i-0], stride) );
 
-            sp_setup_tri( setup_ctx,
+         sp_setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[i-2], stride),
                        get_vert(vertex_buffer, indices[i-1], stride),
                        get_vert(vertex_buffer, indices[i-0], stride) );
-         }
       }
       break;
 
    case PIPE_PRIM_QUAD_STRIP:
-      if (softpipe->rasterizer->flatshade_first) {
-         for (i = 3; i < nr; i += 2) {
-            sp_setup_tri( setup_ctx,
-                       get_vert(vertex_buffer, indices[i-0], stride),
-                       get_vert(vertex_buffer, indices[i-1], stride),
-                       get_vert(vertex_buffer, indices[i-3], stride));
-            sp_setup_tri( setup_ctx,
-                       get_vert(vertex_buffer, indices[i-2], stride),
-                       get_vert(vertex_buffer, indices[i-0], stride),
-                       get_vert(vertex_buffer, indices[i-3], stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 2) {
-            sp_setup_tri( setup_ctx,
+      for (i = 3; i < nr; i += 2) {
+         sp_setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[i-3], stride),
                        get_vert(vertex_buffer, indices[i-2], stride),
                        get_vert(vertex_buffer, indices[i-0], stride) );
-            sp_setup_tri( setup_ctx,
+         sp_setup_tri( setup_ctx,
                        get_vert(vertex_buffer, indices[i-1], stride),
                        get_vert(vertex_buffer, indices[i-3], stride),
                        get_vert(vertex_buffer, indices[i-0], stride) );
-         }
       }
       break;
 
@@ -448,56 +420,28 @@ sp_vbuf_draw_arrays(struct vbuf_render *vbr, uint start, uint nr)
       break;
 
    case PIPE_PRIM_QUADS:
-      if (softpipe->rasterizer->flatshade_first) {
-         for (i = 3; i < nr; i += 4) {
-            sp_setup_tri( setup_ctx,
-                       get_vert(vertex_buffer, i-2, stride),
-                       get_vert(vertex_buffer, i-1, stride),
-                       get_vert(vertex_buffer, i-3, stride) );
-            sp_setup_tri( setup_ctx,
-                       get_vert(vertex_buffer, i-1, stride),
-                       get_vert(vertex_buffer, i-0, stride),
-                       get_vert(vertex_buffer, i-3, stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 4) {
-            sp_setup_tri( setup_ctx,
+      for (i = 3; i < nr; i += 4) {
+         sp_setup_tri( setup_ctx,
                        get_vert(vertex_buffer, i-3, stride),
                        get_vert(vertex_buffer, i-2, stride),
                        get_vert(vertex_buffer, i-0, stride) );
-            sp_setup_tri( setup_ctx,
+         sp_setup_tri( setup_ctx,
                        get_vert(vertex_buffer, i-2, stride),
                        get_vert(vertex_buffer, i-1, stride),
                        get_vert(vertex_buffer, i-0, stride) );
-         }
       }
       break;
 
    case PIPE_PRIM_QUAD_STRIP:
-      if (softpipe->rasterizer->flatshade_first) {
-         for (i = 3; i < nr; i += 2) {
-            sp_setup_tri( setup_ctx,
-                       get_vert(vertex_buffer, i-0, stride),
-                       get_vert(vertex_buffer, i-1, stride),
-                       get_vert(vertex_buffer, i-3, stride) );
-            sp_setup_tri( setup_ctx,
-                       get_vert(vertex_buffer, i-2, stride),
-                       get_vert(vertex_buffer, i-0, stride),
-                       get_vert(vertex_buffer, i-3, stride) );
-         }
-      }
-      else {
-         for (i = 3; i < nr; i += 2) {
-            sp_setup_tri( setup_ctx,
+      for (i = 3; i < nr; i += 2) {
+         sp_setup_tri( setup_ctx,
                        get_vert(vertex_buffer, i-3, stride),
                        get_vert(vertex_buffer, i-2, stride),
                        get_vert(vertex_buffer, i-0, stride) );
-            sp_setup_tri( setup_ctx,
+         sp_setup_tri( setup_ctx,
                        get_vert(vertex_buffer, i-1, stride),
                        get_vert(vertex_buffer, i-3, stride),
                        get_vert(vertex_buffer, i-0, stride) );
-         }
       }
       break;
 
diff --git a/src/gallium/drivers/softpipe/sp_public.h b/src/gallium/drivers/softpipe/sp_public.h
new file mode 100644
index 00000000000..62d0903d87a
--- /dev/null
+++ b/src/gallium/drivers/softpipe/sp_public.h
@@ -0,0 +1,10 @@
+#ifndef SP_PUBLIC_H
+#define SP_PUBLIC_H
+
+struct pipe_screen;
+struct sw_winsys;
+
+struct pipe_screen *
+softpipe_create_screen(struct sw_winsys *winsys);
+
+#endif
diff --git a/src/gallium/drivers/softpipe/sp_quad_depth_test.c b/src/gallium/drivers/softpipe/sp_quad_depth_test.c
index 499eebd671b..4815a0d49f1 100644
--- a/src/gallium/drivers/softpipe/sp_quad_depth_test.c
+++ b/src/gallium/drivers/softpipe/sp_quad_depth_test.c
@@ -73,8 +73,8 @@ get_depth_stencil_values( struct depth_data *data,
          data->bzzzz[j] = tile->data.depth32[y][x];
       }
       break;
-   case PIPE_FORMAT_X8Z24_UNORM:
-   case PIPE_FORMAT_S8Z24_UNORM:
+   case PIPE_FORMAT_Z24X8_UNORM:
+   case PIPE_FORMAT_Z24S8_UNORM:
       for (j = 0; j < QUAD_SIZE; j++) {
          int x = quad->input.x0 % TILE_SIZE + (j & 1);
          int y = quad->input.y0 % TILE_SIZE + (j >> 1);
@@ -82,8 +82,8 @@ get_depth_stencil_values( struct depth_data *data,
          data->stencilVals[j] = tile->data.depth32[y][x] >> 24;
       }
    break;
-   case PIPE_FORMAT_Z24X8_UNORM:
-   case PIPE_FORMAT_Z24S8_UNORM:
+   case PIPE_FORMAT_X8Z24_UNORM:
+   case PIPE_FORMAT_S8Z24_UNORM:
       for (j = 0; j < QUAD_SIZE; j++) {
          int x = quad->input.x0 % TILE_SIZE + (j & 1);
          int y = quad->input.y0 % TILE_SIZE + (j >> 1);
@@ -146,8 +146,8 @@ convert_quad_depth( struct depth_data *data,
          }
       }
       break;
-   case PIPE_FORMAT_X8Z24_UNORM:
-   case PIPE_FORMAT_S8Z24_UNORM:
+   case PIPE_FORMAT_Z24X8_UNORM:
+   case PIPE_FORMAT_Z24S8_UNORM:
       {
          float scale = (float) ((1 << 24) - 1);
 
@@ -156,8 +156,8 @@ convert_quad_depth( struct depth_data *data,
          }
       }
       break;
-   case PIPE_FORMAT_Z24X8_UNORM:
-   case PIPE_FORMAT_Z24S8_UNORM:
+   case PIPE_FORMAT_X8Z24_UNORM:
+   case PIPE_FORMAT_S8Z24_UNORM:
       {
          float scale = (float) ((1 << 24) - 1);
 
@@ -189,7 +189,7 @@ write_depth_stencil_values( struct depth_data *data,
          tile->data.depth16[y][x] = (ushort) data->bzzzz[j];
       }
       break;
-   case PIPE_FORMAT_X8Z24_UNORM:
+   case PIPE_FORMAT_Z24X8_UNORM:
    case PIPE_FORMAT_Z32_UNORM:
       for (j = 0; j < QUAD_SIZE; j++) {
          int x = quad->input.x0 % TILE_SIZE + (j & 1);
@@ -197,21 +197,21 @@ write_depth_stencil_values( struct depth_data *data,
          tile->data.depth32[y][x] = data->bzzzz[j];
       }
       break;
-   case PIPE_FORMAT_S8Z24_UNORM:
+   case PIPE_FORMAT_Z24S8_UNORM:
       for (j = 0; j < QUAD_SIZE; j++) {
          int x = quad->input.x0 % TILE_SIZE + (j & 1);
          int y = quad->input.y0 % TILE_SIZE + (j >> 1);
          tile->data.depth32[y][x] = (data->stencilVals[j] << 24) | data->bzzzz[j];
       }
       break;
-   case PIPE_FORMAT_Z24S8_UNORM:
+   case PIPE_FORMAT_S8Z24_UNORM:
       for (j = 0; j < QUAD_SIZE; j++) {
          int x = quad->input.x0 % TILE_SIZE + (j & 1);
          int y = quad->input.y0 % TILE_SIZE + (j >> 1);
          tile->data.depth32[y][x] = (data->bzzzz[j] << 8) | data->stencilVals[j];
       }
       break;
-   case PIPE_FORMAT_Z24X8_UNORM:
+   case PIPE_FORMAT_X8Z24_UNORM:
       for (j = 0; j < QUAD_SIZE; j++) {
          int x = quad->input.x0 % TILE_SIZE + (j & 1);
          int y = quad->input.y0 % TILE_SIZE + (j >> 1);
@@ -728,9 +728,14 @@ depth_test_quads_fallback(struct quad_stage *qs,
       qs->next->run(qs->next, quads, nr);
 }
 
-/* XXX: this function assumes setup function actually emits linear
- * spans of quads.  It seems a lot more natural to do (early)
- * depth-testing on spans rather than quads.
+
+/**
+ * Special-case Z testing for 16-bit Zbuffer, PIPE_FUNC_LESS and
+ * Z buffer writes enabled.
+ *
+ * NOTE: there's no guarantee that the quads are sequentially side by
+ * side.  The fragment shader may have culled some quads, etc.  Sliver
+ * triangles may generate non-sequential quads.
  */
 static void
 depth_interp_z16_less_write(struct quad_stage *qs, 
@@ -747,25 +752,33 @@ depth_interp_z16_less_write(struct quad_stage *qs,
    const float z0 = quads[0]->posCoef->a0[2] + dzdx * fx + dzdy * fy;
    struct softpipe_cached_tile *tile;
    ushort (*depth16)[TILE_SIZE];
-   ushort idepth[4], depth_step;
+   ushort init_idepth[4], idepth[4], depth_step;
    const float scale = 65535.0;
 
-   idepth[0] = (ushort)((z0) * scale);
-   idepth[1] = (ushort)((z0 + dzdx) * scale);
-   idepth[2] = (ushort)((z0 + dzdy) * scale);
-   idepth[3] = (ushort)((z0 + dzdx + dzdy) * scale);
+   /* compute scaled depth of the four pixels in first quad */
+   init_idepth[0] = (ushort)((z0) * scale);
+   init_idepth[1] = (ushort)((z0 + dzdx) * scale);
+   init_idepth[2] = (ushort)((z0 + dzdy) * scale);
+   init_idepth[3] = (ushort)((z0 + dzdx + dzdy) * scale);
 
-   depth_step = (ushort)(dzdx * 2 * scale);
+   depth_step = (ushort)(dzdx * scale);
 
    tile = sp_get_cached_tile(qs->softpipe->zsbuf_cache, ix, iy);
 
-   depth16 = (ushort (*)[TILE_SIZE])
-      &tile->data.depth16[iy % TILE_SIZE][ix % TILE_SIZE];
-
    for (i = 0; i < nr; i++) {
-      unsigned outmask = quads[i]->inout.mask;
+      const unsigned outmask = quads[i]->inout.mask;
+      const int dx = quads[i]->input.x0 - ix;
       unsigned mask = 0;
-      
+
+      /* compute depth for this quad */
+      idepth[0] = init_idepth[0] + dx * depth_step;
+      idepth[1] = init_idepth[1] + dx * depth_step;
+      idepth[2] = init_idepth[2] + dx * depth_step;
+      idepth[3] = init_idepth[3] + dx * depth_step;
+
+      depth16 = (ushort (*)[TILE_SIZE])
+         &tile->data.depth16[iy % TILE_SIZE][(ix + dx)% TILE_SIZE];
+
       if ((outmask & 1) && idepth[0] < depth16[0][0]) {
          depth16[0][0] = idepth[0];
          mask |= (1 << 0);
@@ -786,13 +799,6 @@ depth_interp_z16_less_write(struct quad_stage *qs,
          mask |= (1 << 3);
       }
 
-      idepth[0] += depth_step;
-      idepth[1] += depth_step;
-      idepth[2] += depth_step;
-      idepth[3] += depth_step;
-
-      depth16 = (ushort (*)[TILE_SIZE]) &depth16[0][2];
-
       quads[i]->inout.mask = mask;
       if (quads[i]->inout.mask)
          quads[pass++] = quads[i];
@@ -804,6 +810,14 @@ depth_interp_z16_less_write(struct quad_stage *qs,
 }
 
 
+/**
+ * Special-case Z testing for 16-bit Zbuffer, PIPE_FUNC_LEQUAL and
+ * Z buffer writes enabled.
+ *
+ * NOTE: there's no guarantee that the quads are sequentially side by
+ * side.  The fragment shader may have culled some quads, etc.  Sliver
+ * triangles may generate non-sequential quads.
+ */
 static void
 depth_interp_z16_lequal_write(struct quad_stage *qs, 
                             struct quad_header *quads[],
@@ -819,25 +833,33 @@ depth_interp_z16_lequal_write(struct quad_stage *qs,
    const float z0 = quads[0]->posCoef->a0[2] + dzdx * fx + dzdy * fy;
    struct softpipe_cached_tile *tile;
    ushort (*depth16)[TILE_SIZE];
-   ushort idepth[4], depth_step;
+   ushort init_idepth[4], idepth[4], depth_step;
    const float scale = 65535.0;
 
-   idepth[0] = (ushort)((z0) * scale);
-   idepth[1] = (ushort)((z0 + dzdx) * scale);
-   idepth[2] = (ushort)((z0 + dzdy) * scale);
-   idepth[3] = (ushort)((z0 + dzdx + dzdy) * scale);
+   /* compute scaled depth of the four pixels in first quad */
+   init_idepth[0] = (ushort)((z0) * scale);
+   init_idepth[1] = (ushort)((z0 + dzdx) * scale);
+   init_idepth[2] = (ushort)((z0 + dzdy) * scale);
+   init_idepth[3] = (ushort)((z0 + dzdx + dzdy) * scale);
 
-   depth_step = (ushort)(dzdx * 2 * scale);
+   depth_step = (ushort)(dzdx * scale);
 
    tile = sp_get_cached_tile(qs->softpipe->zsbuf_cache, ix, iy);
 
-   depth16 = (ushort (*)[TILE_SIZE])
-      &tile->data.depth16[iy % TILE_SIZE][ix % TILE_SIZE];
-
    for (i = 0; i < nr; i++) {
-      unsigned outmask = quads[i]->inout.mask;
+      const unsigned outmask = quads[i]->inout.mask;
+      const int dx = quads[i]->input.x0 - ix;
       unsigned mask = 0;
       
+      /* compute depth for this quad */
+      idepth[0] = init_idepth[0] + dx * depth_step;
+      idepth[1] = init_idepth[1] + dx * depth_step;
+      idepth[2] = init_idepth[2] + dx * depth_step;
+      idepth[3] = init_idepth[3] + dx * depth_step;
+
+      depth16 = (ushort (*)[TILE_SIZE])
+         &tile->data.depth16[iy % TILE_SIZE][(ix + dx)% TILE_SIZE];
+
       if ((outmask & 1) && idepth[0] <= depth16[0][0]) {
          depth16[0][0] = idepth[0];
          mask |= (1 << 0);
@@ -858,11 +880,6 @@ depth_interp_z16_lequal_write(struct quad_stage *qs,
          mask |= (1 << 3);
       }
 
-      idepth[0] += depth_step;
-      idepth[1] += depth_step;
-      idepth[2] += depth_step;
-      idepth[3] += depth_step;
-
       depth16 = (ushort (*)[TILE_SIZE]) &depth16[0][2];
 
       quads[i]->inout.mask = mask;
diff --git a/src/gallium/drivers/softpipe/sp_quad_fs.c b/src/gallium/drivers/softpipe/sp_quad_fs.c
index ad04dc2afc3..8ae5a7f028b 100644
--- a/src/gallium/drivers/softpipe/sp_quad_fs.c
+++ b/src/gallium/drivers/softpipe/sp_quad_fs.c
@@ -65,6 +65,7 @@ quad_shade_stage(struct quad_stage *qs)
 
 /**
  * Execute fragment shader for the four fragments in the quad.
+ * \return TRUE if quad is alive, FALSE if all four pixels are killed
  */
 static INLINE boolean
 shade_quad(struct quad_stage *qs, struct quad_header *quad)
@@ -98,11 +99,14 @@ coverage_quad(struct quad_stage *qs, struct quad_header *quad)
 }
 
 
-
+/**
+ * Shade/write an array of quads
+ * Called via quad_stage::run()
+ */
 static void
 shade_quads(struct quad_stage *qs, 
-                 struct quad_header *quads[],
-                 unsigned nr)
+            struct quad_header *quads[],
+            unsigned nr)
 {
    struct quad_shade_stage *qss = quad_shade_stage( qs );
    struct softpipe_context *softpipe = qs->softpipe;
@@ -116,7 +120,7 @@ shade_quads(struct quad_stage *qs,
 
    for (i = 0; i < nr; i++) {
       if (!shade_quad(qs, quads[i]))
-         continue;
+         continue; /* quad totally culled/killed */
 
       if (/*do_coverage*/ 0)
          coverage_quad( qs, quads[i] );
@@ -129,9 +133,6 @@ shade_quads(struct quad_stage *qs,
 }
    
 
-
-
-
 /**
  * Per-primitive (or per-begin?) setup
  */
diff --git a/src/gallium/drivers/softpipe/sp_screen.c b/src/gallium/drivers/softpipe/sp_screen.c
index 87415f43404..e0a2ef604e9 100644
--- a/src/gallium/drivers/softpipe/sp_screen.c
+++ b/src/gallium/drivers/softpipe/sp_screen.c
@@ -27,15 +27,16 @@
 
 
 #include "util/u_memory.h"
-#include "util/u_simple_screen.h"
-#include "util/u_simple_screen.h"
 #include "pipe/p_defines.h"
 #include "pipe/p_screen.h"
 
+#include "state_tracker/sw_winsys.h"
+
 #include "sp_texture.h"
-#include "sp_winsys.h"
 #include "sp_screen.h"
 #include "sp_context.h"
+#include "sp_buffer.h"
+#include "sp_public.h"
 
 
 static const char *
@@ -83,11 +84,11 @@ softpipe_get_param(struct pipe_screen *screen, int param)
    case PIPE_CAP_TEXTURE_SHADOW_MAP:
       return 1;
    case PIPE_CAP_MAX_TEXTURE_2D_LEVELS:
-      return 13; /* max 4Kx4K */
+      return SP_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_3D_LEVELS:
-      return 9;  /* max 256x256x256 */
+      return SP_MAX_TEXTURE_3D_LEVELS;
    case PIPE_CAP_MAX_TEXTURE_CUBE_LEVELS:
-      return 13; /* max 4Kx4K */
+      return SP_MAX_TEXTURE_2D_LEVELS;
    case PIPE_CAP_TGSI_CONT_SUPPORTED:
       return 1;
    case PIPE_CAP_BLEND_EQUATION_SEPARATE:
@@ -145,6 +146,8 @@ softpipe_is_format_supported( struct pipe_screen *screen,
                               unsigned tex_usage, 
                               unsigned geom_flags )
 {
+   struct sw_winsys *winsys = softpipe_screen(screen)->winsys;
+
    assert(target == PIPE_TEXTURE_1D ||
           target == PIPE_TEXTURE_2D ||
           target == PIPE_TEXTURE_3D ||
@@ -152,29 +155,39 @@ softpipe_is_format_supported( struct pipe_screen *screen,
 
    switch(format) {
    case PIPE_FORMAT_L16_UNORM:
-   case PIPE_FORMAT_YCBCR_REV:
-   case PIPE_FORMAT_YCBCR:
+   case PIPE_FORMAT_YUYV:
+   case PIPE_FORMAT_UYVY:
    case PIPE_FORMAT_DXT1_RGB:
    case PIPE_FORMAT_DXT1_RGBA:
    case PIPE_FORMAT_DXT3_RGBA:
    case PIPE_FORMAT_DXT5_RGBA:
    case PIPE_FORMAT_Z32_FLOAT:
    case PIPE_FORMAT_R8G8_SNORM:
-   case PIPE_FORMAT_B6UG5SR5S_NORM:
-   case PIPE_FORMAT_X8UB8UG8SR8S_NORM:
-   case PIPE_FORMAT_A8B8G8R8_SNORM:
+   case PIPE_FORMAT_R5SG5SB6U_NORM:
+   case PIPE_FORMAT_R8SG8SB8UX8U_NORM:
+   case PIPE_FORMAT_R8G8B8A8_SNORM:
    case PIPE_FORMAT_NONE:
       return FALSE;
    default:
-      return TRUE;
+      break;
    }
+
+   if(tex_usage & PIPE_TEXTURE_USAGE_DISPLAY_TARGET) {
+      if(!winsys->is_displaytarget_format_supported(winsys, format))
+         return FALSE;
+   }
+
+   /* XXX: this is often a lie.  Pull in logic from llvmpipe to fix.
+    */
+   return TRUE;
 }
 
 
 static void
 softpipe_destroy_screen( struct pipe_screen *screen )
 {
-   struct pipe_winsys *winsys = screen->winsys;
+   struct softpipe_screen *sp_screen = softpipe_screen(screen);
+   struct sw_winsys *winsys = sp_screen->winsys;
 
    if(winsys->destroy)
       winsys->destroy(winsys);
@@ -183,21 +196,37 @@ softpipe_destroy_screen( struct pipe_screen *screen )
 }
 
 
+/* This is often overriden by the co-state tracker.
+ */
+static void
+softpipe_flush_frontbuffer(struct pipe_screen *_screen,
+                           struct pipe_surface *surface,
+                           void *context_private)
+{
+   struct softpipe_screen *screen = softpipe_screen(_screen);
+   struct sw_winsys *winsys = screen->winsys;
+   struct softpipe_texture *texture = softpipe_texture(surface->texture);
+
+   assert(texture->dt);
+   if (texture->dt)
+      winsys->displaytarget_display(winsys, texture->dt, context_private);
+}
 
 /**
  * Create a new pipe_screen object
  * Note: we're not presently subclassing pipe_screen (no softpipe_screen).
  */
 struct pipe_screen *
-softpipe_create_screen(struct pipe_winsys *winsys)
+softpipe_create_screen(struct sw_winsys *winsys)
 {
    struct softpipe_screen *screen = CALLOC_STRUCT(softpipe_screen);
 
    if (!screen)
       return NULL;
 
-   screen->base.winsys = winsys;
+   screen->winsys = winsys;
 
+   screen->base.winsys = NULL;
    screen->base.destroy = softpipe_destroy_screen;
 
    screen->base.get_name = softpipe_get_name;
@@ -206,9 +235,10 @@ softpipe_create_screen(struct pipe_winsys *winsys)
    screen->base.get_paramf = softpipe_get_paramf;
    screen->base.is_format_supported = softpipe_is_format_supported;
    screen->base.context_create = softpipe_create_context;
+   screen->base.flush_frontbuffer = softpipe_flush_frontbuffer;
 
    softpipe_init_screen_texture_funcs(&screen->base);
-   u_simple_screen_init(&screen->base);
+   softpipe_init_screen_buffer_funcs(&screen->base);
 
    return &screen->base;
 }
diff --git a/src/gallium/drivers/softpipe/sp_screen.h b/src/gallium/drivers/softpipe/sp_screen.h
index 3d4bfd3e840..f741454c9e5 100644
--- a/src/gallium/drivers/softpipe/sp_screen.h
+++ b/src/gallium/drivers/softpipe/sp_screen.h
@@ -35,10 +35,13 @@
 #include "pipe/p_defines.h"
 
 
+struct sw_winsys;
 
 struct softpipe_screen {
    struct pipe_screen base;
 
+   struct sw_winsys *winsys;
+
    /* Increments whenever textures are modified.  Contexts can track
     * this.
     */
@@ -55,4 +58,5 @@ softpipe_screen( struct pipe_screen *pipe )
 }
 
 
+
 #endif /* SP_SCREEN_H */
diff --git a/src/gallium/drivers/softpipe/sp_setup.c b/src/gallium/drivers/softpipe/sp_setup.c
index b8590a8cc2c..85966bc5e49 100644
--- a/src/gallium/drivers/softpipe/sp_setup.c
+++ b/src/gallium/drivers/softpipe/sp_setup.c
@@ -211,11 +211,11 @@ static void flush_spans( struct setup_context *setup )
    const int xright1 = setup->span.right[1];
    struct quad_stage *pipe = setup->softpipe->quad.first;
 
-
-   int minleft = block_x(MIN2(xleft0, xleft1));
-   int maxright = MAX2(xright0, xright1);
+   const int minleft = block_x(MIN2(xleft0, xleft1));
+   const int maxright = MAX2(xright0, xright1);
    int x;
 
+   /* process quads in horizontal chunks of 16 */
    for (x = minleft; x < maxright; x += step) {
       unsigned skip_left0 = CLAMP(xleft0 - x, 0, step);
       unsigned skip_left1 = CLAMP(xleft1 - x, 0, step);
diff --git a/src/gallium/drivers/softpipe/sp_state.h b/src/gallium/drivers/softpipe/sp_state.h
index b37bb951520..ade96b0fd4c 100644
--- a/src/gallium/drivers/softpipe/sp_state.h
+++ b/src/gallium/drivers/softpipe/sp_state.h
@@ -100,6 +100,11 @@ struct sp_geometry_shader {
    struct draw_geometry_shader *draw_data;
 };
 
+struct sp_velems_state {
+   unsigned count;
+   struct pipe_vertex_element velem[PIPE_MAX_ATTRIBS];
+};
+
 
 void *
 softpipe_create_blend_state(struct pipe_context *,
@@ -160,8 +165,14 @@ void *softpipe_create_gs_state(struct pipe_context *,
 void softpipe_bind_gs_state(struct pipe_context *, void *);
 void softpipe_delete_gs_state(struct pipe_context *, void *);
 
+void *softpipe_create_vertex_elements_state(struct pipe_context *,
+                                            unsigned count,
+                                            const struct pipe_vertex_element *);
+void softpipe_bind_vertex_elements_state(struct pipe_context *, void *);
+void softpipe_delete_vertex_elements_state(struct pipe_context *, void *);
+
 void softpipe_set_polygon_stipple( struct pipe_context *,
-				  const struct pipe_poly_stipple * );
+                                   const struct pipe_poly_stipple * );
 
 void softpipe_set_scissor_state( struct pipe_context *,
                                  const struct pipe_scissor_state * );
@@ -187,10 +198,6 @@ softpipe_sampler_view_destroy(struct pipe_context *pipe,
 void softpipe_set_viewport_state( struct pipe_context *,
                                   const struct pipe_viewport_state * );
 
-void softpipe_set_vertex_elements(struct pipe_context *,
-                                  unsigned count,
-                                  const struct pipe_vertex_element *);
-
 void softpipe_set_vertex_buffers(struct pipe_context *,
                                  unsigned count,
                                  const struct pipe_vertex_buffer *);
diff --git a/src/gallium/drivers/softpipe/sp_state_fs.c b/src/gallium/drivers/softpipe/sp_state_fs.c
index c88e2137510..2b089c28316 100644
--- a/src/gallium/drivers/softpipe/sp_state_fs.c
+++ b/src/gallium/drivers/softpipe/sp_state_fs.c
@@ -28,6 +28,7 @@
 #include "sp_context.h"
 #include "sp_state.h"
 #include "sp_fs.h"
+#include "sp_buffer.h"
 
 #include "pipe/p_defines.h"
 #include "util/u_memory.h"
@@ -163,26 +164,33 @@ softpipe_delete_vs_state(struct pipe_context *pipe, void *vs)
    FREE( state );
 }
 
-
-
 void
 softpipe_set_constant_buffer(struct pipe_context *pipe,
                              uint shader, uint index,
-                             struct pipe_buffer *buf)
+                             struct pipe_buffer *constants)
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
+   unsigned size = constants ? constants->size : 0;
+   const void *data = constants ? softpipe_buffer(constants)->data : NULL;
 
    assert(shader < PIPE_SHADER_TYPES);
-   assert(index < PIPE_MAX_CONSTANT_BUFFERS);
+   assert(index == 0);
 
    draw_flush(softpipe->draw);
 
    /* note: reference counting */
-   pipe_buffer_reference(&softpipe->constants[shader][index], buf);
+   pipe_buffer_reference(&softpipe->constants[shader][index], constants);
 
+   if(shader == PIPE_SHADER_VERTEX) {
+      draw_set_mapped_constant_buffer(softpipe->draw, PIPE_SHADER_VERTEX, index,
+                                      data, size);
+   }
+
+   softpipe->mapped_constants[shader][index] = data;
    softpipe->dirty |= SP_NEW_CONSTANTS;
 }
 
+
 void *
 softpipe_create_gs_state(struct pipe_context *pipe,
                          const struct pipe_shader_state *templ)
diff --git a/src/gallium/drivers/softpipe/sp_state_vertex.c b/src/gallium/drivers/softpipe/sp_state_vertex.c
index b491d92ed15..462f4d2655e 100644
--- a/src/gallium/drivers/softpipe/sp_state_vertex.c
+++ b/src/gallium/drivers/softpipe/sp_state_vertex.c
@@ -32,27 +32,45 @@
 #include "sp_context.h"
 #include "sp_state.h"
 
+#include "util/u_memory.h"
 #include "draw/draw_context.h"
 
 
+void *
+softpipe_create_vertex_elements_state(struct pipe_context *pipe,
+                                      unsigned count,
+                                      const struct pipe_vertex_element *attribs)
+{
+   struct sp_velems_state *velems;
+   assert(count <= PIPE_MAX_ATTRIBS);
+   velems = (struct sp_velems_state *) MALLOC(sizeof(struct sp_velems_state));
+   if (velems) {
+      velems->count = count;
+      memcpy(velems->velem, attribs, sizeof(*attribs) * count);
+   }
+   return velems;
+}
+
 void
-softpipe_set_vertex_elements(struct pipe_context *pipe,
-                             unsigned count,
-                             const struct pipe_vertex_element *attribs)
+softpipe_bind_vertex_elements_state(struct pipe_context *pipe,
+                                    void *velems)
 {
    struct softpipe_context *softpipe = softpipe_context(pipe);
+   struct sp_velems_state *sp_velems = (struct sp_velems_state *) velems;
 
-   assert(count <= PIPE_MAX_ATTRIBS);
-
-   memcpy(softpipe->vertex_element, attribs,
-          count * sizeof(struct pipe_vertex_element));
-   softpipe->num_vertex_elements = count;
+   softpipe->velems = sp_velems;
 
    softpipe->dirty |= SP_NEW_VERTEX;
 
-   draw_set_vertex_elements(softpipe->draw, count, attribs);
+   if (sp_velems)
+      draw_set_vertex_elements(softpipe->draw, sp_velems->count, sp_velems->velem);
 }
 
+void
+softpipe_delete_vertex_elements_state(struct pipe_context *pipe, void *velems)
+{
+   FREE( velems );
+}
 
 void
 softpipe_set_vertex_buffers(struct pipe_context *pipe,
diff --git a/src/gallium/drivers/softpipe/sp_tex_sample.c b/src/gallium/drivers/softpipe/sp_tex_sample.c
index ecd6b39863c..ef7ccf41898 100644
--- a/src/gallium/drivers/softpipe/sp_tex_sample.c
+++ b/src/gallium/drivers/softpipe/sp_tex_sample.c
@@ -55,7 +55,7 @@
 static INLINE float
 frac(float f)
 {
-   return f - util_ifloor(f);
+   return f - floorf(f);
 }
 
 
@@ -201,11 +201,9 @@ wrap_nearest_mirror_repeat(const float s[4], unsigned size, int icoord[4])
    const float max = 1.0F - min;
    for (ch = 0; ch < 4; ch++) {
       const int flr = util_ifloor(s[ch]);
-      float u;
+      float u = frac(s[ch]);
       if (flr & 1)
-         u = 1.0F - (s[ch] - (float) flr);
-      else
-         u = s[ch] - (float) flr;
+         u = 1.0F - u;
       if (u < min)
          icoord[ch] = 0;
       else if (u > max)
@@ -358,11 +356,9 @@ wrap_linear_mirror_repeat(const float s[4], unsigned size,
    uint ch;
    for (ch = 0; ch < 4; ch++) {
       const int flr = util_ifloor(s[ch]);
-      float u;
+      float u = frac(s[ch]);
       if (flr & 1)
-         u = 1.0F - (s[ch] - (float) flr);
-      else
-         u = s[ch] - (float) flr;
+         u = 1.0F - u;
       u = u * size - 0.5F;
       icoord0[ch] = util_ifloor(u);
       icoord1[ch] = icoord0[ch] + 1;
@@ -441,8 +437,7 @@ wrap_linear_mirror_clamp_to_border(const float s[4], unsigned size,
 
 
 /**
- * For RECT textures / unnormalized texcoords
- * Only a subset of wrap modes supported.
+ * PIPE_TEX_WRAP_CLAMP for nearest sampling, unnormalized coords.
  */
 static void
 wrap_nearest_unorm_clamp(const float s[4], unsigned size, int icoord[4])
@@ -456,7 +451,7 @@ wrap_nearest_unorm_clamp(const float s[4], unsigned size, int icoord[4])
 
 
 /**
- * Handles clamp_to_edge and clamp_to_border:
+ * PIPE_TEX_WRAP_CLAMP_TO_BORDER for nearest sampling, unnormalized coords.
  */
 static void
 wrap_nearest_unorm_clamp_to_border(const float s[4], unsigned size,
@@ -464,14 +459,27 @@ wrap_nearest_unorm_clamp_to_border(const float s[4], unsigned size,
 {
    uint ch;
    for (ch = 0; ch < 4; ch++) {
+      icoord[ch]= util_ifloor( CLAMP(s[ch], -0.5F, (float) size + 0.5F) );
+   }
+}
+
+
+/**
+ * PIPE_TEX_WRAP_CLAMP_TO_EDGE for nearest sampling, unnormalized coords.
+ */
+static void
+wrap_nearest_unorm_clamp_to_edge(const float s[4], unsigned size,
+                                 int icoord[4])
+{
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
       icoord[ch]= util_ifloor( CLAMP(s[ch], 0.5F, (float) size - 0.5F) );
    }
 }
 
 
 /**
- * For RECT textures / unnormalized texcoords.
- * Only a subset of wrap modes supported.
+ * PIPE_TEX_WRAP_CLAMP for linear sampling, unnormalized coords.
  */
 static void
 wrap_linear_unorm_clamp(const float s[4], unsigned size,
@@ -488,13 +496,36 @@ wrap_linear_unorm_clamp(const float s[4], unsigned size,
 }
 
 
+/**
+ * PIPE_TEX_WRAP_CLAMP_TO_BORDER for linear sampling, unnormalized coords.
+ */
 static void
 wrap_linear_unorm_clamp_to_border(const float s[4], unsigned size,
                                   int icoord0[4], int icoord1[4], float w[4])
 {
    uint ch;
    for (ch = 0; ch < 4; ch++) {
-      float u = CLAMP(s[ch], 0.5F, (float) size - 0.5F);
+      float u = CLAMP(s[ch], -0.5F, (float) size + 0.5F);
+      u -= 0.5F;
+      icoord0[ch] = util_ifloor(u);
+      icoord1[ch] = icoord0[ch] + 1;
+      if (icoord1[ch] > (int) size - 1)
+         icoord1[ch] = size - 1;
+      w[ch] = frac(u);
+   }
+}
+
+
+/**
+ * PIPE_TEX_WRAP_CLAMP_TO_EDGE for linear sampling, unnormalized coords.
+ */
+static void
+wrap_linear_unorm_clamp_to_edge(const float s[4], unsigned size,
+                                int icoord0[4], int icoord1[4], float w[4])
+{
+   uint ch;
+   for (ch = 0; ch < 4; ch++) {
+      float u = CLAMP(s[ch], +0.5F, (float) size - 0.5F);
       u -= 0.5F;
       icoord0[ch] = util_ifloor(u);
       icoord1[ch] = icoord0[ch] + 1;
@@ -1711,6 +1742,7 @@ get_nearest_unorm_wrap(unsigned mode)
    case PIPE_TEX_WRAP_CLAMP:
       return wrap_nearest_unorm_clamp;
    case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+      return wrap_nearest_unorm_clamp_to_edge;
    case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
       return wrap_nearest_unorm_clamp_to_border;
    default:
@@ -1754,6 +1786,7 @@ get_linear_unorm_wrap(unsigned mode)
    case PIPE_TEX_WRAP_CLAMP:
       return wrap_linear_unorm_clamp;
    case PIPE_TEX_WRAP_CLAMP_TO_EDGE:
+      return wrap_linear_unorm_clamp_to_edge;
    case PIPE_TEX_WRAP_CLAMP_TO_BORDER:
       return wrap_linear_unorm_clamp_to_border;
    default:
diff --git a/src/gallium/drivers/softpipe/sp_texture.c b/src/gallium/drivers/softpipe/sp_texture.c
index 371c4e20251..11d184effb9 100644
--- a/src/gallium/drivers/softpipe/sp_texture.c
+++ b/src/gallium/drivers/softpipe/sp_texture.c
@@ -36,11 +36,13 @@
 #include "util/u_format.h"
 #include "util/u_math.h"
 #include "util/u_memory.h"
+#include "util/u_simple_screen.h"
 
 #include "sp_context.h"
 #include "sp_texture.h"
 #include "sp_screen.h"
-#include "sp_winsys.h"
+
+#include "state_tracker/sw_winsys.h"
 
 
 /**
@@ -72,11 +74,9 @@ softpipe_texture_layout(struct pipe_screen *screen,
       depth = u_minify(depth, 1);
    }
 
-   spt->buffer = screen->buffer_create(screen, 32,
-                                       PIPE_BUFFER_USAGE_PIXEL,
-                                       buffer_size);
+   spt->data = align_malloc(buffer_size, 16);
 
-   return spt->buffer != NULL;
+   return spt->data != NULL;
 }
 
 
@@ -87,19 +87,18 @@ static boolean
 softpipe_displaytarget_layout(struct pipe_screen *screen,
                               struct softpipe_texture * spt)
 {
-   unsigned usage = (PIPE_BUFFER_USAGE_CPU_READ_WRITE |
-                     PIPE_BUFFER_USAGE_GPU_READ_WRITE);
-   unsigned tex_usage = spt->base.tex_usage;
-
-   spt->buffer = screen->surface_buffer_create( screen, 
-                                                spt->base.width0, 
-                                                spt->base.height0,
-                                                spt->base.format,
-                                                usage,
-                                                tex_usage,
-                                                &spt->stride[0]);
-
-   return spt->buffer != NULL;
+   struct sw_winsys *winsys = softpipe_screen(screen)->winsys;
+
+   /* Round up the surface size to a multiple of the tile size?
+    */
+   spt->dt = winsys->displaytarget_create(winsys,
+                                          spt->base.format,
+                                          spt->base.width0, 
+                                          spt->base.height0,
+                                          16,
+                                          &spt->stride[0] );
+
+   return spt->dt != NULL;
 }
 
 
@@ -123,7 +122,8 @@ softpipe_texture_create(struct pipe_screen *screen,
                util_is_power_of_two(template->depth0));
 
    if (spt->base.tex_usage & (PIPE_TEXTURE_USAGE_DISPLAY_TARGET |
-                              PIPE_TEXTURE_USAGE_PRIMARY)) {
+                              PIPE_TEXTURE_USAGE_SCANOUT |
+                              PIPE_TEXTURE_USAGE_SHARED)) {
       if (!softpipe_displaytarget_layout(screen, spt))
          goto fail;
    }
@@ -140,46 +140,24 @@ softpipe_texture_create(struct pipe_screen *screen,
 }
 
 
-/**
- * Create a new pipe_texture which wraps an existing buffer.
- */
-static struct pipe_texture *
-softpipe_texture_blanket(struct pipe_screen * screen,
-                         const struct pipe_texture *base,
-                         const unsigned *stride,
-                         struct pipe_buffer *buffer)
-{
-   struct softpipe_texture *spt;
-   assert(screen);
-
-   /* Only supports one type */
-   if (base->target != PIPE_TEXTURE_2D ||
-       base->last_level != 0 ||
-       base->depth0 != 1) {
-      return NULL;
-   }
-
-   spt = CALLOC_STRUCT(softpipe_texture);
-   if (!spt)
-      return NULL;
-
-   spt->base = *base;
-   pipe_reference_init(&spt->base.reference, 1);
-   spt->base.screen = screen;
-   spt->stride[0] = stride[0];
-
-   pipe_buffer_reference(&spt->buffer, buffer);
-
-   return &spt->base;
-}
 
 
 static void
 softpipe_texture_destroy(struct pipe_texture *pt)
 {
+   struct softpipe_screen *screen = softpipe_screen(pt->screen);
    struct softpipe_texture *spt = softpipe_texture(pt);
 
-   pipe_buffer_reference(&spt->buffer, NULL);
+   if (spt->dt) {
+      /* display target */
+      struct sw_winsys *winsys = screen->winsys;
+      winsys->displaytarget_destroy(winsys, spt->dt);
+   }
+   else {
+      /* regular texture */
+      align_free(spt->data);
+   }
+
    FREE(spt);
 }
 
@@ -359,9 +337,20 @@ softpipe_transfer_map( struct pipe_screen *screen,
    spt = softpipe_texture(transfer->texture);
    format = transfer->texture->format;
 
-   map = pipe_buffer_map(screen, spt->buffer, pipe_transfer_buffer_flags(transfer));
-   if (map == NULL)
-      return NULL;
+   if (spt->dt) {
+      /* display target */
+      struct sw_winsys *winsys = softpipe_screen(screen)->winsys;
+
+      map = winsys->displaytarget_map(winsys, spt->dt,
+                                      pipe_transfer_buffer_flags(transfer));
+      if (map == NULL)
+         return NULL;
+   }
+   else {
+      map = spt->data;
+      if (map == NULL)
+         return NULL;
+   }
 
    /* May want to different things here depending on read/write nature
     * of the map:
@@ -393,7 +382,11 @@ softpipe_transfer_unmap(struct pipe_screen *screen,
    assert(transfer->texture);
    spt = softpipe_texture(transfer->texture);
 
-   pipe_buffer_unmap( screen, spt->buffer );
+   if (spt->dt) {
+      /* display target */
+      struct sw_winsys *winsys = softpipe_screen(screen)->winsys;
+      winsys->displaytarget_unmap(winsys, spt->dt);
+   }
 
    if (transfer->usage & PIPE_TRANSFER_WRITE) {
       /* Mark the texture as dirty to expire the tile caches. */
@@ -426,7 +419,7 @@ softpipe_video_surface_create(struct pipe_screen *screen,
 
    memset(&template, 0, sizeof(struct pipe_texture));
    template.target = PIPE_TEXTURE_2D;
-   template.format = PIPE_FORMAT_X8R8G8B8_UNORM;
+   template.format = PIPE_FORMAT_B8G8R8X8_UNORM;
    template.last_level = 0;
    /* vl_mpeg12_mc_renderer expects this when it's initialized with pot_buffers=true */
    template.width0 = util_next_power_of_two(width);
@@ -458,7 +451,6 @@ void
 softpipe_init_screen_texture_funcs(struct pipe_screen *screen)
 {
    screen->texture_create = softpipe_texture_create;
-   screen->texture_blanket = softpipe_texture_blanket;
    screen->texture_destroy = softpipe_texture_destroy;
 
    screen->get_tex_surface = softpipe_get_tex_surface;
@@ -474,24 +466,4 @@ softpipe_init_screen_texture_funcs(struct pipe_screen *screen)
 }
 
 
-/**
- * Return pipe_buffer handle and stride for given texture object.
- * XXX used for???
- */
-boolean
-softpipe_get_texture_buffer( struct pipe_texture *texture,
-                             struct pipe_buffer **buf,
-                             unsigned *stride )
-{
-   struct softpipe_texture *tex = (struct softpipe_texture *) texture;
 
-   if (!tex)
-      return FALSE;
-
-   pipe_buffer_reference(buf, tex->buffer);
-
-   if (stride)
-      *stride = tex->stride[0];
-
-   return TRUE;
-}
diff --git a/src/gallium/drivers/softpipe/sp_texture.h b/src/gallium/drivers/softpipe/sp_texture.h
index 2ef64e1e7c3..1c8636d1d56 100644
--- a/src/gallium/drivers/softpipe/sp_texture.h
+++ b/src/gallium/drivers/softpipe/sp_texture.h
@@ -33,6 +33,10 @@
 #include "pipe/p_video_state.h"
 
 
+#define SP_MAX_TEXTURE_2D_LEVELS 13  /* 4K x 4K */
+#define SP_MAX_TEXTURE_3D_LEVELS 9   /* 512 x 512 x 512 */
+
+
 struct pipe_context;
 struct pipe_screen;
 struct softpipe_context;
@@ -42,12 +46,19 @@ struct softpipe_texture
 {
    struct pipe_texture base;
 
-   unsigned long level_offset[PIPE_MAX_TEXTURE_LEVELS];
-   unsigned stride[PIPE_MAX_TEXTURE_LEVELS];
+   unsigned long level_offset[SP_MAX_TEXTURE_2D_LEVELS];
+   unsigned stride[SP_MAX_TEXTURE_2D_LEVELS];
 
-   /* The data is held here:
+   /**
+    * Display target, for textures with the PIPE_TEXTURE_USAGE_DISPLAY_TARGET
+    * usage.
+    */
+   struct sw_displaytarget *dt;
+
+   /**
+    * Malloc'ed data for regular textures, or a mapping to dt above.
     */
-   struct pipe_buffer *buffer;
+   void *data;
 
    /* True if texture images are power-of-two in all dimensions:
     */
diff --git a/src/gallium/drivers/softpipe/sp_tile_cache.c b/src/gallium/drivers/softpipe/sp_tile_cache.c
index 1b50bd7ffe5..aedfdf1b469 100644
--- a/src/gallium/drivers/softpipe/sp_tile_cache.c
+++ b/src/gallium/drivers/softpipe/sp_tile_cache.c
@@ -162,10 +162,10 @@ sp_tile_cache_set_surface(struct softpipe_tile_cache *tc,
                                               PIPE_TRANSFER_READ_WRITE,
                                               0, 0, ps->width, ps->height);
 
-      tc->depth_stencil = (ps->format == PIPE_FORMAT_S8Z24_UNORM ||
-                           ps->format == PIPE_FORMAT_X8Z24_UNORM ||
-                           ps->format == PIPE_FORMAT_Z24S8_UNORM ||
+      tc->depth_stencil = (ps->format == PIPE_FORMAT_Z24S8_UNORM ||
                            ps->format == PIPE_FORMAT_Z24X8_UNORM ||
+                           ps->format == PIPE_FORMAT_S8Z24_UNORM ||
+                           ps->format == PIPE_FORMAT_X8Z24_UNORM ||
                            ps->format == PIPE_FORMAT_Z16_UNORM ||
                            ps->format == PIPE_FORMAT_Z32_UNORM ||
                            ps->format == PIPE_FORMAT_S8_UNORM);
@@ -299,13 +299,14 @@ sp_tile_cache_flush_clear(struct softpipe_tile_cache *tc)
                               x, y, TILE_SIZE, TILE_SIZE,
                               tc->tile.data.color32, 0/*STRIDE*/);
 
-            /* do this? */
-            clear_clear_flag(tc->clear_flags, addr);
-
             numCleared++;
          }
       }
    }
+
+   /* reset all clear flags to zero */
+   memset(tc->clear_flags, 0, sizeof(tc->clear_flags));
+
 #if 0
    debug_printf("num cleared: %u\n", numCleared);
 #endif
diff --git a/src/gallium/drivers/softpipe/sp_video_context.c b/src/gallium/drivers/softpipe/sp_video_context.c
index 7a3a6361670..242aaac4665 100644
--- a/src/gallium/drivers/softpipe/sp_video_context.c
+++ b/src/gallium/drivers/softpipe/sp_video_context.c
@@ -25,10 +25,12 @@
  * 
  **************************************************************************/
 
+#include "util/u_inlines.h"
+#include "util/u_memory.h"
+
 #include "sp_video_context.h"
-#include <util/u_inlines.h>
-#include <util/u_memory.h>
-#include "softpipe/sp_texture.h"
+#include "sp_texture.h"
+
 
 static void
 sp_mpeg12_destroy(struct pipe_video_context *vpipe)
@@ -174,7 +176,6 @@ init_pipe_state(struct sp_mpeg12_context *ctx)
    rast.line_stipple_factor = 0;
    rast.line_stipple_pattern = 0;
    rast.line_last_pixel = 0;
-   rast.bypass_vs_clip_and_viewport = 0;
    rast.line_width = 1;
    rast.point_smooth = 0;
    rast.point_quad_rasterization = 0;
diff --git a/src/gallium/drivers/softpipe/sp_winsys.c b/src/gallium/drivers/softpipe/sp_winsys.c
deleted file mode 100644
index f6598927d35..00000000000
--- a/src/gallium/drivers/softpipe/sp_winsys.c
+++ /dev/null
@@ -1,245 +0,0 @@
-/**************************************************************************
- * 
- * Copyright 2007 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
- * Malloc softpipe winsys. Uses malloc for all memory allocations.
- * 
- * @author Keith Whitwell
- * @author Brian Paul
- * @author Jose Fonseca
- */
-
-
-#include "util/u_simple_screen.h"/* port to just p_screen */
-#include "pipe/p_format.h"
-#include "pipe/p_context.h"
-#include "util/u_format.h"
-#include "util/u_math.h"
-#include "util/u_memory.h"
-#include "util/u_inlines.h"
-#include "softpipe/sp_winsys.h"
-
-
-struct st_softpipe_buffer
-{
-   struct pipe_buffer base;
-   boolean userBuffer;  /** Is this a user-space buffer? */
-   void *data;
-   void *mapped;
-};
-
-
-/** Cast wrapper */
-static INLINE struct st_softpipe_buffer *
-st_softpipe_buffer( struct pipe_buffer *buf )
-{
-   return (struct st_softpipe_buffer *)buf;
-}
-
-
-static void *
-st_softpipe_buffer_map(struct pipe_winsys *winsys, 
-                       struct pipe_buffer *buf,
-                       unsigned flags)
-{
-   struct st_softpipe_buffer *st_softpipe_buf = st_softpipe_buffer(buf);
-   st_softpipe_buf->mapped = st_softpipe_buf->data;
-   return st_softpipe_buf->mapped;
-}
-
-
-static void
-st_softpipe_buffer_unmap(struct pipe_winsys *winsys, 
-                         struct pipe_buffer *buf)
-{
-   struct st_softpipe_buffer *st_softpipe_buf = st_softpipe_buffer(buf);
-   st_softpipe_buf->mapped = NULL;
-}
-
-
-static void
-st_softpipe_buffer_destroy(struct pipe_buffer *buf)
-{
-   struct st_softpipe_buffer *oldBuf = st_softpipe_buffer(buf);
-
-   if (oldBuf->data) {
-      if (!oldBuf->userBuffer)
-         align_free(oldBuf->data);
-
-      oldBuf->data = NULL;
-   }
-
-   FREE(oldBuf);
-}
-
-
-static void
-st_softpipe_flush_frontbuffer(struct pipe_winsys *winsys,
-                              struct pipe_surface *surf,
-                              void *context_private)
-{
-}
-
-
-
-static const char *
-st_softpipe_get_name(struct pipe_winsys *winsys)
-{
-   return "softpipe";
-}
-
-
-static struct pipe_buffer *
-st_softpipe_buffer_create(struct pipe_winsys *winsys, 
-                          unsigned alignment, 
-                          unsigned usage,
-                          unsigned size)
-{
-   struct st_softpipe_buffer *buffer = CALLOC_STRUCT(st_softpipe_buffer);
-
-   pipe_reference_init(&buffer->base.reference, 1);
-   buffer->base.alignment = alignment;
-   buffer->base.usage = usage;
-   buffer->base.size = size;
-
-   buffer->data = align_malloc(size, alignment);
-
-   return &buffer->base;
-}
-
-
-/**
- * Create buffer which wraps user-space data.
- */
-static struct pipe_buffer *
-st_softpipe_user_buffer_create(struct pipe_winsys *winsys, 
-                               void *ptr, 
-                               unsigned bytes)
-{
-   struct st_softpipe_buffer *buffer;
-   
-   buffer = CALLOC_STRUCT(st_softpipe_buffer);
-   if(!buffer)
-      return NULL;
-   
-   pipe_reference_init(&buffer->base.reference, 1);
-   buffer->base.size = bytes;
-   buffer->userBuffer = TRUE;
-   buffer->data = ptr;
-
-   return &buffer->base;
-}
-
-
-static struct pipe_buffer *
-st_softpipe_surface_buffer_create(struct pipe_winsys *winsys,
-                                  unsigned width, unsigned height,
-                                  enum pipe_format format,
-                                  unsigned usage,
-                                  unsigned tex_usage,
-                                  unsigned *stride)
-{
-   const unsigned alignment = 64;
-   unsigned nblocksy;
-
-   nblocksy = util_format_get_nblocksy(format, height);
-   *stride = align(util_format_get_stride(format, width), alignment);
-
-   return winsys->buffer_create(winsys, alignment,
-                                usage,
-                                *stride * nblocksy);
-}
-
-
-static void
-st_softpipe_fence_reference(struct pipe_winsys *winsys, 
-                            struct pipe_fence_handle **ptr,
-                            struct pipe_fence_handle *fence)
-{
-}
-
-
-static int
-st_softpipe_fence_signalled(struct pipe_winsys *winsys, 
-                            struct pipe_fence_handle *fence,
-                            unsigned flag)
-{
-   return 0;
-}
-
-
-static int
-st_softpipe_fence_finish(struct pipe_winsys *winsys, 
-                         struct pipe_fence_handle *fence,
-                         unsigned flag)
-{
-   return 0;
-}
-
-
-static void
-st_softpipe_destroy(struct pipe_winsys *winsys)
-{
-   FREE(winsys);
-}
-
-
-struct pipe_screen *
-softpipe_create_screen_malloc(void)
-{
-   static struct pipe_winsys *winsys;
-   struct pipe_screen *screen;
-
-   winsys = CALLOC_STRUCT(pipe_winsys);
-   if(!winsys)
-      return NULL;
-
-   winsys->destroy = st_softpipe_destroy;
-   
-   winsys->buffer_create = st_softpipe_buffer_create;
-   winsys->user_buffer_create = st_softpipe_user_buffer_create;
-   winsys->buffer_map = st_softpipe_buffer_map;
-   winsys->buffer_unmap = st_softpipe_buffer_unmap;
-   winsys->buffer_destroy = st_softpipe_buffer_destroy;
-
-   winsys->surface_buffer_create = st_softpipe_surface_buffer_create;
-
-   winsys->fence_reference = st_softpipe_fence_reference;
-   winsys->fence_signalled = st_softpipe_fence_signalled;
-   winsys->fence_finish = st_softpipe_fence_finish;
-
-   winsys->flush_frontbuffer = st_softpipe_flush_frontbuffer;
-   winsys->get_name = st_softpipe_get_name;
-
-   screen = softpipe_create_screen(winsys);
-   if(!screen)
-      st_softpipe_destroy(winsys);
-
-   return screen;
-}
diff --git a/src/gallium/drivers/svga/svga_cmd.c b/src/gallium/drivers/svga/svga_cmd.c
index a0da7d7e5d5..04307d17fe0 100644
--- a/src/gallium/drivers/svga/svga_cmd.c
+++ b/src/gallium/drivers/svga/svga_cmd.c
@@ -478,7 +478,8 @@ SVGA3D_BufferDMA(struct svga_winsys_context *swc,
                  struct svga_winsys_surface *host,
                  SVGA3dTransferType transfer,      // IN
                  uint32 size,                      // IN
-                 uint32 offset,                    // IN
+                 uint32 guest_offset,              // IN
+                 uint32 host_offset,               // IN
                  SVGA3dSurfaceDMAFlags flags)      // IN
 {
    SVGA3dCmdSurfaceDMA *cmd;
@@ -517,19 +518,19 @@ SVGA3D_BufferDMA(struct svga_winsys_context *swc,
    cmd->transfer = transfer;
 
    box = (SVGA3dCopyBox *)&cmd[1];
-   box->x = offset;
+   box->x = host_offset;
    box->y = 0;
    box->z = 0;
    box->w = size;
    box->h = 1;
    box->d = 1;
-   box->srcx = offset;
+   box->srcx = guest_offset;
    box->srcy = 0;
    box->srcz = 0;
    
    pSuffix = (SVGA3dCmdSurfaceDMASuffix *)((uint8_t*)cmd + sizeof *cmd + sizeof *box);
    pSuffix->suffixSize = sizeof *pSuffix;
-   pSuffix->maximumOffset = offset + size;
+   pSuffix->maximumOffset = guest_offset + size;
    pSuffix->flags = flags;
 
    swc->commit(swc);
diff --git a/src/gallium/drivers/svga/svga_cmd.h b/src/gallium/drivers/svga/svga_cmd.h
index 80410547690..da9fc4355fa 100644
--- a/src/gallium/drivers/svga/svga_cmd.h
+++ b/src/gallium/drivers/svga/svga_cmd.h
@@ -111,7 +111,8 @@ SVGA3D_BufferDMA(struct svga_winsys_context *swc,
                  struct svga_winsys_surface *host,
                  SVGA3dTransferType transfer,
                  uint32 size,
-                 uint32 offset,
+                 uint32 guest_offset,
+                 uint32 host_offset,
                  SVGA3dSurfaceDMAFlags flags);
 
 /*
diff --git a/src/gallium/drivers/svga/svga_context.h b/src/gallium/drivers/svga/svga_context.h
index b1022c3c996..1f66437dfe1 100644
--- a/src/gallium/drivers/svga/svga_context.h
+++ b/src/gallium/drivers/svga/svga_context.h
@@ -169,6 +169,11 @@ struct svga_sampler_state {
    unsigned view_max_lod;
 };
 
+struct svga_velems_state {
+   unsigned count;
+   struct pipe_vertex_element velem[PIPE_MAX_ATTRIBS];
+};
+
 /* Use to calculate differences between state emitted to hardware and
  * current driver-calculated state.  
  */
@@ -178,13 +183,13 @@ struct svga_state
    const struct svga_depth_stencil_state *depth;
    const struct svga_rasterizer_state *rast;
    const struct svga_sampler_state *sampler[PIPE_MAX_SAMPLERS];
+   const struct svga_velems_state *velems;
 
    struct pipe_sampler_view *sampler_views[PIPE_MAX_SAMPLERS]; /* or texture ID's? */
    struct svga_fragment_shader *fs;
    struct svga_vertex_shader *vs;
 
    struct pipe_vertex_buffer vb[PIPE_MAX_ATTRIBS];
-   struct pipe_vertex_element ve[PIPE_MAX_ATTRIBS];
    struct pipe_buffer *cb[PIPE_SHADER_TYPES];
 
    struct pipe_framebuffer_state framebuffer;
@@ -204,7 +209,6 @@ struct svga_state
 
    unsigned num_samplers;
    unsigned num_sampler_views;
-   unsigned num_vertex_elements;
    unsigned num_vertex_buffers;
    unsigned reduced_prim;
 
diff --git a/src/gallium/drivers/svga/svga_pipe_blend.c b/src/gallium/drivers/svga/svga_pipe_blend.c
index b60117f090e..594eec7166e 100644
--- a/src/gallium/drivers/svga/svga_pipe_blend.c
+++ b/src/gallium/drivers/svga/svga_pipe_blend.c
@@ -92,6 +92,7 @@ svga_create_blend_state(struct pipe_context *pipe,
       if (templ->logicop_enable) {
          switch (templ->logicop_func) {
          case PIPE_LOGICOP_XOR:
+         case PIPE_LOGICOP_INVERT:
             blend->need_white_fragments = TRUE;
             blend->rt[i].blend_enable = TRUE;
             blend->rt[i].srcblend       = SVGA3D_BLENDOP_ONE;
@@ -125,12 +126,6 @@ svga_create_blend_state(struct pipe_context *pipe,
             blend->rt[i].dstblend       = SVGA3D_BLENDOP_ONE;
             blend->rt[i].blendeq        = SVGA3D_BLENDEQ_MAXIMUM;
             break;
-         case PIPE_LOGICOP_INVERT:
-            blend->rt[i].blend_enable = TRUE;
-            blend->rt[i].srcblend       = SVGA3D_BLENDOP_INVSRCCOLOR;
-            blend->rt[i].dstblend       = SVGA3D_BLENDOP_ZERO;
-            blend->rt[i].blendeq        = SVGA3D_BLENDEQ_ADD;
-            break;
          case PIPE_LOGICOP_AND:
             /* Approximate with minimum - works for the 0 & anything case: */
             blend->rt[i].blend_enable = TRUE;
diff --git a/src/gallium/drivers/svga/svga_pipe_clear.c b/src/gallium/drivers/svga/svga_pipe_clear.c
index 409b3b41cbc..8483a3fad74 100644
--- a/src/gallium/drivers/svga/svga_pipe_clear.c
+++ b/src/gallium/drivers/svga/svga_pipe_clear.c
@@ -54,7 +54,7 @@ try_clear(struct svga_context *svga,
 
    if ((buffers & PIPE_CLEAR_COLOR) && fb->cbufs[0]) {
       flags |= SVGA3D_CLEAR_COLOR;
-      util_pack_color(rgba, PIPE_FORMAT_A8R8G8B8_UNORM, &uc);
+      util_pack_color(rgba, PIPE_FORMAT_B8G8R8A8_UNORM, &uc);
 
       rect.w = fb->cbufs[0]->width;
       rect.h = fb->cbufs[0]->height;
@@ -63,7 +63,7 @@ try_clear(struct svga_context *svga,
    if ((buffers & PIPE_CLEAR_DEPTHSTENCIL) && fb->zsbuf) {
       flags |= SVGA3D_CLEAR_DEPTH;
 
-      if (svga->curr.framebuffer.zsbuf->format == PIPE_FORMAT_Z24S8_UNORM)
+      if (svga->curr.framebuffer.zsbuf->format == PIPE_FORMAT_S8Z24_UNORM)
          flags |= SVGA3D_CLEAR_STENCIL;
 
       rect.w = MAX2(rect.w, fb->zsbuf->width);
diff --git a/src/gallium/drivers/svga/svga_pipe_misc.c b/src/gallium/drivers/svga/svga_pipe_misc.c
index 49b43bebc29..95bf0e6f91b 100644
--- a/src/gallium/drivers/svga/svga_pipe_misc.c
+++ b/src/gallium/drivers/svga/svga_pipe_misc.c
@@ -118,10 +118,10 @@ static void svga_set_framebuffer_state(struct pipe_context *pipe,
       case PIPE_FORMAT_Z16_UNORM:
          svga->curr.depthscale = 1.0f / DEPTH_BIAS_SCALE_FACTOR_D16;
          break;
-      case PIPE_FORMAT_S8Z24_UNORM:
-      case PIPE_FORMAT_X8Z24_UNORM:
       case PIPE_FORMAT_Z24S8_UNORM:
       case PIPE_FORMAT_Z24X8_UNORM:
+      case PIPE_FORMAT_S8Z24_UNORM:
+      case PIPE_FORMAT_X8Z24_UNORM:
          svga->curr.depthscale = 1.0f / DEPTH_BIAS_SCALE_FACTOR_D24S8;
          break;
       case PIPE_FORMAT_Z32_UNORM:
diff --git a/src/gallium/drivers/svga/svga_pipe_sampler.c b/src/gallium/drivers/svga/svga_pipe_sampler.c
index 50fe962b28d..ebd1b949972 100644
--- a/src/gallium/drivers/svga/svga_pipe_sampler.c
+++ b/src/gallium/drivers/svga/svga_pipe_sampler.c
@@ -155,7 +155,7 @@ static void svga_bind_sampler_states(struct pipe_context *pipe,
    /* Check for no-op */
    if (num == svga->curr.num_samplers &&
        !memcmp(svga->curr.sampler, sampler, num * sizeof(void *))) {
-      debug_printf("sampler noop\n");
+      if (0) debug_printf("sampler noop\n");
       return;
    }
 
@@ -226,7 +226,7 @@ static void svga_set_sampler_views(struct pipe_context *pipe,
       if (!views[i])
          continue;
 
-      if (views[i]->texture->format == PIPE_FORMAT_A8R8G8B8_SRGB)
+      if (views[i]->texture->format == PIPE_FORMAT_B8G8R8A8_SRGB)
          flag_srgb |= 1 << i;
 
       if (views[i]->texture->target == PIPE_TEXTURE_1D)
diff --git a/src/gallium/drivers/svga/svga_pipe_vertex.c b/src/gallium/drivers/svga/svga_pipe_vertex.c
index ffc0f99565b..1715a47fc62 100644
--- a/src/gallium/drivers/svga/svga_pipe_vertex.c
+++ b/src/gallium/drivers/svga/svga_pipe_vertex.c
@@ -26,6 +26,7 @@
 #include "util/u_inlines.h"
 #include "pipe/p_defines.h"
 #include "util/u_math.h"
+#include "util/u_memory.h"
 #include "tgsi/tgsi_parse.h"
 
 #include "svga_screen.h"
@@ -49,7 +50,7 @@ static void svga_set_vertex_buffers(struct pipe_context *pipe,
    /* Adjust refcounts */
    for (i = 0; i < count; i++) {
       pipe_buffer_reference(&svga->curr.vb[i].buffer, buffers[i].buffer);
-      if (svga_buffer(buffers[i].buffer)->user)
+      if (svga_buffer_is_user_buffer(buffers[i].buffer))
          any_user_buffer = TRUE;
    }
 
@@ -64,20 +65,37 @@ static void svga_set_vertex_buffers(struct pipe_context *pipe,
    svga->dirty |= SVGA_NEW_VBUFFER;
 }
 
-static void svga_set_vertex_elements(struct pipe_context *pipe,
-                                     unsigned count,
-                                     const struct pipe_vertex_element *elements)
+
+static void *
+svga_create_vertex_elements_state(struct pipe_context *pipe,
+                                  unsigned count,
+                                  const struct pipe_vertex_element *attribs)
 {
-   struct svga_context *svga = svga_context(pipe);
-   unsigned i;
+   struct svga_velems_state *velems;
+   assert(count <= PIPE_MAX_ATTRIBS);
+   velems = (struct svga_velems_state *) MALLOC(sizeof(struct svga_velems_state));
+   if (velems) {
+      velems->count = count;
+      memcpy(velems->velem, attribs, sizeof(*attribs) * count);
+   }
+   return velems;
+}
 
-   for (i = 0; i < count; i++)
-      svga->curr.ve[i] = elements[i];
+static void svga_bind_vertex_elements_state(struct pipe_context *pipe,
+                                            void *velems)
+{
+   struct svga_context *svga = svga_context(pipe);
+   struct svga_velems_state *svga_velems = (struct svga_velems_state *) velems;
 
-   svga->curr.num_vertex_elements = count;
+   svga->curr.velems = svga_velems;
    svga->dirty |= SVGA_NEW_VELEMENT;
 }
 
+static void svga_delete_vertex_elements_state(struct pipe_context *pipe,
+                                              void *velems)
+{
+   FREE(velems);
+}
 
 void svga_cleanup_vertex_state( struct svga_context *svga )
 {
@@ -91,7 +109,9 @@ void svga_cleanup_vertex_state( struct svga_context *svga )
 void svga_init_vertex_functions( struct svga_context *svga )
 {
    svga->pipe.set_vertex_buffers = svga_set_vertex_buffers;
-   svga->pipe.set_vertex_elements = svga_set_vertex_elements;
+   svga->pipe.create_vertex_elements_state = svga_create_vertex_elements_state;
+   svga->pipe.bind_vertex_elements_state = svga_bind_vertex_elements_state;
+   svga->pipe.delete_vertex_elements_state = svga_delete_vertex_elements_state;
 }
 
 
diff --git a/src/gallium/drivers/svga/svga_screen.c b/src/gallium/drivers/svga/svga_screen.c
index 414ac52e1f6..6022c38cfca 100644
--- a/src/gallium/drivers/svga/svga_screen.c
+++ b/src/gallium/drivers/svga/svga_screen.c
@@ -196,23 +196,23 @@ svga_translate_format_cap(enum pipe_format format)
 {
    switch(format) {
    
-   case PIPE_FORMAT_A8R8G8B8_UNORM:
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
       return SVGA3D_DEVCAP_SURFACEFMT_A8R8G8B8;
-   case PIPE_FORMAT_X8R8G8B8_UNORM:
+   case PIPE_FORMAT_B8G8R8X8_UNORM:
       return SVGA3D_DEVCAP_SURFACEFMT_X8R8G8B8;
 
-   case PIPE_FORMAT_R5G6B5_UNORM:
+   case PIPE_FORMAT_B5G6R5_UNORM:
       return SVGA3D_DEVCAP_SURFACEFMT_R5G6B5;
-   case PIPE_FORMAT_A1R5G5B5_UNORM:
+   case PIPE_FORMAT_B5G5R5A1_UNORM:
       return SVGA3D_DEVCAP_SURFACEFMT_A1R5G5B5;
-   case PIPE_FORMAT_A4R4G4B4_UNORM:
+   case PIPE_FORMAT_B4G4R4A4_UNORM:
       return SVGA3D_DEVCAP_SURFACEFMT_A4R4G4B4;
 
    case PIPE_FORMAT_Z16_UNORM:
       return SVGA3D_DEVCAP_SURFACEFMT_Z_D16;
-   case PIPE_FORMAT_Z24S8_UNORM:
+   case PIPE_FORMAT_S8Z24_UNORM:
       return SVGA3D_DEVCAP_SURFACEFMT_Z_D24S8;
-   case PIPE_FORMAT_Z24X8_UNORM:
+   case PIPE_FORMAT_X8Z24_UNORM:
       return SVGA3D_DEVCAP_SURFACEFMT_Z_D24X8;
 
    case PIPE_FORMAT_A8_UNORM:
@@ -254,8 +254,8 @@ svga_is_format_supported( struct pipe_screen *screen,
       /* Often unsupported/problematic. This means we end up with the same
        * visuals for all virtual hardware implementations.
        */
-      case PIPE_FORMAT_A4R4G4B4_UNORM:
-      case PIPE_FORMAT_A1R5G5B5_UNORM:
+      case PIPE_FORMAT_B4G4R4A4_UNORM:
+      case PIPE_FORMAT_B5G5R5A1_UNORM:
          return FALSE;
          
       /* Simulate ability to render into compressed textures */
diff --git a/src/gallium/drivers/svga/svga_screen_buffer.c b/src/gallium/drivers/svga/svga_screen_buffer.c
index c9e9bef5406..1ff6a3a5b31 100644
--- a/src/gallium/drivers/svga/svga_screen_buffer.c
+++ b/src/gallium/drivers/svga/svga_screen_buffer.c
@@ -83,7 +83,7 @@ svga_buffer_create_host_surface(struct svga_screen *ss,
        * as svga_screen_surface_create might have passed a recycled host
        * buffer.
        */
-      sbuf->hw.flags.discard = TRUE;
+      sbuf->dma.flags.discard = TRUE;
 
       SVGA_DBG(DEBUG_DMA, "   --> got sid %p sz %d (buffer)\n", sbuf->handle, sbuf->base.size);
    }
@@ -109,10 +109,10 @@ svga_buffer_destroy_hw_storage(struct svga_screen *ss, struct svga_buffer *sbuf)
    struct svga_winsys_screen *sws = ss->sws;
 
    assert(!sbuf->map.count);
-   assert(sbuf->hw.buf);
-   if(sbuf->hw.buf) {
-      sws->buffer_destroy(sws, sbuf->hw.buf);
-      sbuf->hw.buf = NULL;
+   assert(sbuf->hwbuf);
+   if(sbuf->hwbuf) {
+      sws->buffer_destroy(sws, sbuf->hwbuf);
+      sbuf->hwbuf = NULL;
    }
 }
 
@@ -151,16 +151,18 @@ static INLINE enum pipe_error
 svga_buffer_create_hw_storage(struct svga_screen *ss,
                               struct svga_buffer *sbuf)
 {
-   if(!sbuf->hw.buf) {
+   assert(!sbuf->user);
+
+   if(!sbuf->hwbuf) {
       unsigned alignment = sbuf->base.alignment;
       unsigned usage = 0;
       unsigned size = sbuf->base.size;
       
-      sbuf->hw.buf = svga_winsys_buffer_create(ss, alignment, usage, size);
-      if(!sbuf->hw.buf)
+      sbuf->hwbuf = svga_winsys_buffer_create(ss, alignment, usage, size);
+      if(!sbuf->hwbuf)
          return PIPE_ERROR_OUT_OF_MEMORY;
       
-      assert(!sbuf->needs_flush);
+      assert(!sbuf->dma.pending);
    }
    
    return PIPE_OK;
@@ -175,12 +177,11 @@ svga_buffer_upload_command(struct svga_context *svga,
                            struct svga_buffer *sbuf)
 {
    struct svga_winsys_context *swc = svga->swc;
-   struct svga_winsys_buffer *guest = sbuf->hw.buf;
+   struct svga_winsys_buffer *guest = sbuf->hwbuf;
    struct svga_winsys_surface *host = sbuf->handle;
    SVGA3dTransferType transfer = SVGA3D_WRITE_HOST_VRAM;
-   SVGA3dSurfaceDMAFlags flags = sbuf->hw.flags;
    SVGA3dCmdSurfaceDMA *cmd;
-   uint32 numBoxes = sbuf->hw.num_ranges;
+   uint32 numBoxes = sbuf->map.num_ranges;
    SVGA3dCopyBox *boxes;
    SVGA3dCmdSurfaceDMASuffix *pSuffix;
    unsigned region_flags;
@@ -218,8 +219,8 @@ svga_buffer_upload_command(struct svga_context *svga,
 
    cmd->transfer = transfer;
 
-   sbuf->hw.boxes = (SVGA3dCopyBox *)&cmd[1];
-   sbuf->hw.svga = svga;
+   sbuf->dma.boxes = (SVGA3dCopyBox *)&cmd[1];
+   sbuf->dma.svga = svga;
 
    /* Increment reference count */
    dummy = NULL;
@@ -228,9 +229,11 @@ svga_buffer_upload_command(struct svga_context *svga,
    pSuffix = (SVGA3dCmdSurfaceDMASuffix *)((uint8_t*)cmd + sizeof *cmd + numBoxes * sizeof *boxes);
    pSuffix->suffixSize = sizeof *pSuffix;
    pSuffix->maximumOffset = sbuf->base.size;
-   pSuffix->flags = flags;
+   pSuffix->flags = sbuf->dma.flags;
+
+   SVGA_FIFOCommitAll(swc);
 
-   swc->commit(swc);
+   sbuf->dma.flags.discard = FALSE;
 
    return PIPE_OK;
 }
@@ -248,10 +251,10 @@ svga_buffer_upload_flush(struct svga_context *svga,
    unsigned i;
 
    assert(sbuf->handle); 
-   assert(sbuf->hw.buf);
-   assert(sbuf->hw.num_ranges);
-   assert(sbuf->hw.svga == svga);
-   assert(sbuf->hw.boxes);
+   assert(sbuf->hwbuf);
+   assert(sbuf->map.num_ranges);
+   assert(sbuf->dma.svga == svga);
+   assert(sbuf->dma.boxes);
    
    /*
     * Patch the DMA command with the final copy box.
@@ -259,36 +262,33 @@ svga_buffer_upload_flush(struct svga_context *svga,
 
    SVGA_DBG(DEBUG_DMA, "dma to sid %p\n", sbuf->handle);
 
-   boxes = sbuf->hw.boxes;
-   for(i = 0; i < sbuf->hw.num_ranges; ++i) {
+   boxes = sbuf->dma.boxes;
+   for(i = 0; i < sbuf->map.num_ranges; ++i) {
       SVGA_DBG(DEBUG_DMA, "  bytes %u - %u\n",
-               sbuf->hw.ranges[i].start, sbuf->hw.ranges[i].end);
+               sbuf->map.ranges[i].start, sbuf->map.ranges[i].end);
 
-      boxes[i].x = sbuf->hw.ranges[i].start;
+      boxes[i].x = sbuf->map.ranges[i].start;
       boxes[i].y = 0;
       boxes[i].z = 0;
-      boxes[i].w = sbuf->hw.ranges[i].end - sbuf->hw.ranges[i].start;
+      boxes[i].w = sbuf->map.ranges[i].end - sbuf->map.ranges[i].start;
       boxes[i].h = 1;
       boxes[i].d = 1;
-      boxes[i].srcx = sbuf->hw.ranges[i].start;
+      boxes[i].srcx = sbuf->map.ranges[i].start;
       boxes[i].srcy = 0;
       boxes[i].srcz = 0;
    }
 
-   sbuf->hw.num_ranges = 0;
-   memset(&sbuf->hw.flags, 0, sizeof sbuf->hw.flags);
+   sbuf->map.num_ranges = 0;
 
    assert(sbuf->head.prev && sbuf->head.next);
    LIST_DEL(&sbuf->head);
 #ifdef DEBUG
    sbuf->head.next = sbuf->head.prev = NULL; 
 #endif
-   sbuf->needs_flush = FALSE;
-
-   sbuf->hw.svga = NULL;
-   sbuf->hw.boxes = NULL;
+   sbuf->dma.pending = FALSE;
 
-   sbuf->host_written = TRUE;
+   sbuf->dma.svga = NULL;
+   sbuf->dma.boxes = NULL;
 
    /* Decrement reference count */
    pipe_reference(&(sbuf->base.reference), NULL);
@@ -297,7 +297,7 @@ svga_buffer_upload_flush(struct svga_context *svga,
 
 
 /**
- * Queue a DMA upload of a range of this buffer to the host.
+ * Note a dirty range.
  *
  * This function only notes the range down. It doesn't actually emit a DMA
  * upload command. That only happens when a context tries to refer to this
@@ -306,15 +306,24 @@ svga_buffer_upload_flush(struct svga_context *svga,
  * We try to lump as many contiguous DMA transfers together as possible.
  */
 static void
-svga_buffer_upload_queue(struct svga_buffer *sbuf,
-                         unsigned start,
-                         unsigned end)
+svga_buffer_add_range(struct svga_buffer *sbuf,
+                      unsigned start,
+                      unsigned end)
 {
    unsigned i;
+   unsigned nearest_range;
+   unsigned nearest_dist;
 
-   assert(sbuf->hw.buf);
    assert(end > start);
    
+   if (sbuf->map.num_ranges < SVGA_BUFFER_MAX_RANGES) {
+      nearest_range = sbuf->map.num_ranges;
+      nearest_dist = ~0;
+   } else {
+      nearest_range = SVGA_BUFFER_MAX_RANGES - 1;
+      nearest_dist = 0;
+   }
+
    /*
     * Try to grow one of the ranges.
     *
@@ -325,12 +334,34 @@ svga_buffer_upload_queue(struct svga_buffer *sbuf,
     * buffer should be flushed.
     */
 
-   for(i = 0; i < sbuf->hw.num_ranges; ++i) {
-      if(start <= sbuf->hw.ranges[i].end && sbuf->hw.ranges[i].start <= end) {
-         sbuf->hw.ranges[i].start = MIN2(sbuf->hw.ranges[i].start, start);
-         sbuf->hw.ranges[i].end   = MAX2(sbuf->hw.ranges[i].end,    end);
+   for(i = 0; i < sbuf->map.num_ranges; ++i) {
+      int left_dist;
+      int right_dist;
+      int dist;
+
+      left_dist = start - sbuf->map.ranges[i].end;
+      right_dist = sbuf->map.ranges[i].start - end;
+      dist = MAX2(left_dist, right_dist);
+
+      if (dist <= 0) {
+         /*
+          * Ranges are contiguous or overlapping -- extend this one and return.
+          */
+
+         sbuf->map.ranges[i].start = MIN2(sbuf->map.ranges[i].start, start);
+         sbuf->map.ranges[i].end   = MAX2(sbuf->map.ranges[i].end,   end);
          return;
       }
+      else {
+         /*
+          * Discontiguous ranges -- keep track of the nearest range.
+          */
+
+         if (dist < nearest_dist) {
+            nearest_range = i;
+            nearest_dist = dist;
+         }
+      }
    }
 
    /*
@@ -338,20 +369,34 @@ svga_buffer_upload_queue(struct svga_buffer *sbuf,
     * pending DMA upload and start clean.
     */
 
-   if(sbuf->needs_flush)
-      svga_buffer_upload_flush(sbuf->hw.svga, sbuf);
+   if(sbuf->dma.pending)
+      svga_buffer_upload_flush(sbuf->dma.svga, sbuf);
 
-   assert(!sbuf->needs_flush);
-   assert(!sbuf->hw.svga);
-   assert(!sbuf->hw.boxes);
+   assert(!sbuf->dma.pending);
+   assert(!sbuf->dma.svga);
+   assert(!sbuf->dma.boxes);
 
-   /*
-    * Add a new range.
-    */
+   if (sbuf->map.num_ranges < SVGA_BUFFER_MAX_RANGES) {
+      /*
+       * Add a new range.
+       */
+
+      sbuf->map.ranges[sbuf->map.num_ranges].start = start;
+      sbuf->map.ranges[sbuf->map.num_ranges].end = end;
+      ++sbuf->map.num_ranges;
+   } else {
+      /*
+       * Everything else failed, so just extend the nearest range.
+       *
+       * It is OK to do this because we always keep a local copy of the
+       * host buffer data, for SW TNL, and the host never modifies the buffer.
+       */
 
-   sbuf->hw.ranges[sbuf->hw.num_ranges].start = start;
-   sbuf->hw.ranges[sbuf->hw.num_ranges].end = end;
-   ++sbuf->hw.num_ranges;
+      assert(nearest_range < SVGA_BUFFER_MAX_RANGES);
+      assert(nearest_range < sbuf->map.num_ranges);
+      sbuf->map.ranges[nearest_range].start = MIN2(sbuf->map.ranges[nearest_range].start, start);
+      sbuf->map.ranges[nearest_range].end   = MAX2(sbuf->map.ranges[nearest_range].end,   end);
+   }
 }
 
 
@@ -366,55 +411,30 @@ svga_buffer_map_range( struct pipe_screen *screen,
    struct svga_buffer *sbuf = svga_buffer( buf );
    void *map;
 
-   if(sbuf->swbuf) {
+   if (!sbuf->swbuf && !sbuf->hwbuf) {
+      if (svga_buffer_create_hw_storage(ss, sbuf) != PIPE_OK) {
+         /*
+          * We can't create a hardware buffer big enough, so create a malloc
+          * buffer instead.
+          */
+
+         debug_printf("%s: failed to allocate %u KB of DMA, splitting DMA transfers\n",
+                      __FUNCTION__,
+                      (sbuf->base.size + 1023)/1024);
+
+         sbuf->swbuf = align_malloc(sbuf->base.size, sbuf->base.alignment);
+      }
+   }
+
+   if (sbuf->swbuf) {
       /* User/malloc buffer */
       map = sbuf->swbuf;
    }
+   else if (sbuf->hwbuf) {
+      map = sws->buffer_map(sws, sbuf->hwbuf, usage);
+   }
    else {
-      if(!sbuf->hw.buf) {
-         if(svga_buffer_create_hw_storage(ss, sbuf) != PIPE_OK)
-            return NULL;
-         
-         /* Populate the hardware storage if the host surface pre-existed */
-         if(sbuf->host_written) {
-            SVGA3dSurfaceDMAFlags flags;
-            enum pipe_error ret;
-            struct pipe_fence_handle *fence = NULL;
-            
-            assert(sbuf->handle);
-
-            SVGA_DBG(DEBUG_DMA|DEBUG_PERF, "dma from sid %p (buffer), bytes %u - %u\n", 
-                     sbuf->handle, 0, sbuf->base.size);
-
-            memset(&flags, 0, sizeof flags);
-            
-            ret = SVGA3D_BufferDMA(ss->swc,
-                                   sbuf->hw.buf,
-                                   sbuf->handle,
-                                   SVGA3D_READ_HOST_VRAM,
-                                   sbuf->base.size,
-                                   0,
-                                   flags);
-            if(ret != PIPE_OK) {
-               ss->swc->flush(ss->swc, NULL);
-               
-               ret = SVGA3D_BufferDMA(ss->swc,
-                                      sbuf->hw.buf,
-                                      sbuf->handle,
-                                      SVGA3D_READ_HOST_VRAM,
-                                      sbuf->base.size,
-                                      0,
-                                      flags);
-               assert(ret == PIPE_OK);
-            }
-            
-            ss->swc->flush(ss->swc, &fence);
-            sws->fence_finish(sws, fence, 0);
-            sws->fence_reference(sws, &fence, NULL);
-         }
-      }
-         
-      map = sws->buffer_map(sws, sbuf->hw.buf, usage);
+      map = NULL;
    }
 
    if(map) {
@@ -447,8 +467,7 @@ svga_buffer_flush_mapped_range( struct pipe_screen *screen,
    assert(sbuf->map.writing);
    if(sbuf->map.writing) {
       assert(sbuf->map.flush_explicit);
-      if(sbuf->hw.buf)
-         svga_buffer_upload_queue(sbuf, offset, offset + length);
+      svga_buffer_add_range(sbuf, offset, offset + length);
    }
    pipe_mutex_unlock(ss->swc_mutex);
 }
@@ -467,16 +486,15 @@ svga_buffer_unmap( struct pipe_screen *screen,
    if(sbuf->map.count)
       --sbuf->map.count;
 
-   if(sbuf->hw.buf)
-      sws->buffer_unmap(sws, sbuf->hw.buf);
+   if(sbuf->hwbuf)
+      sws->buffer_unmap(sws, sbuf->hwbuf);
 
    if(sbuf->map.writing) {
       if(!sbuf->map.flush_explicit) {
          /* No mapped range was flushed -- flush the whole buffer */
          SVGA_DBG(DEBUG_DMA, "flushing the whole buffer\n");
    
-         if(sbuf->hw.buf)
-            svga_buffer_upload_queue(sbuf, 0, sbuf->base.size);
+         svga_buffer_add_range(sbuf, 0, sbuf->base.size);
       }
       
       sbuf->map.writing = FALSE;
@@ -494,12 +512,15 @@ svga_buffer_destroy( struct pipe_buffer *buf )
 
    assert(!p_atomic_read(&buf->reference.count));
    
-   assert(!sbuf->needs_flush);
+   assert(!sbuf->dma.pending);
 
    if(sbuf->handle)
       svga_buffer_destroy_host_surface(ss, sbuf);
    
-   if(sbuf->hw.buf)
+   if(sbuf->uploaded.buffer)
+      pipe_buffer_reference(&sbuf->uploaded.buffer, NULL);
+
+   if(sbuf->hwbuf)
       svga_buffer_destroy_hw_storage(ss, sbuf);
    
    if(sbuf->swbuf && !sbuf->user)
@@ -596,13 +617,14 @@ svga_screen_init_buffer_functions(struct pipe_screen *screen)
 }
 
 
-/** 
- * Copy the contents of the user buffer / malloc buffer to a hardware buffer.
+/**
+ * Copy the contents of the malloc buffer to a hardware buffer.
  */
 static INLINE enum pipe_error
 svga_buffer_update_hw(struct svga_screen *ss, struct svga_buffer *sbuf)
 {
-   if(!sbuf->hw.buf) {
+   assert(!sbuf->user);
+   if(!sbuf->hwbuf) {
       enum pipe_error ret;
       void *map;
       
@@ -611,20 +633,20 @@ svga_buffer_update_hw(struct svga_screen *ss, struct svga_buffer *sbuf)
          return PIPE_ERROR;
       
       ret = svga_buffer_create_hw_storage(ss, sbuf);
-      assert(ret == PIPE_OK);
       if(ret != PIPE_OK)
          return ret;
 
       pipe_mutex_lock(ss->swc_mutex);
-      map = ss->sws->buffer_map(ss->sws, sbuf->hw.buf, PIPE_BUFFER_USAGE_CPU_WRITE);
+      map = ss->sws->buffer_map(ss->sws, sbuf->hwbuf, PIPE_BUFFER_USAGE_CPU_WRITE);
       assert(map);
       if(!map) {
 	 pipe_mutex_unlock(ss->swc_mutex);
-         return PIPE_ERROR_OUT_OF_MEMORY;
+         svga_buffer_destroy_hw_storage(ss, sbuf);
+         return PIPE_ERROR;
       }
 
       memcpy(map, sbuf->swbuf, sbuf->base.size);
-      ss->sws->buffer_unmap(ss->sws, sbuf->hw.buf);
+      ss->sws->buffer_unmap(ss->sws, sbuf->hwbuf);
 
       /* This user/malloc buffer is now indistinguishable from a gpu buffer */
       assert(!sbuf->map.count);
@@ -636,10 +658,89 @@ svga_buffer_update_hw(struct svga_screen *ss, struct svga_buffer *sbuf)
          sbuf->swbuf = NULL;
       }
       
-      svga_buffer_upload_queue(sbuf, 0, sbuf->base.size);
+      pipe_mutex_unlock(ss->swc_mutex);
    }
    
-   pipe_mutex_unlock(ss->swc_mutex);
+   return PIPE_OK;
+}
+
+
+/**
+ * Upload the buffer to the host in a piecewise fashion.
+ *
+ * Used when the buffer is too big to fit in the GMR aperture.
+ */
+static INLINE enum pipe_error
+svga_buffer_upload_piecewise(struct svga_screen *ss,
+                             struct svga_context *svga,
+                             struct svga_buffer *sbuf)
+{
+   struct svga_winsys_screen *sws = ss->sws;
+   const unsigned alignment = sizeof(void *);
+   const unsigned usage = 0;
+   unsigned i;
+
+   assert(sbuf->map.num_ranges);
+   assert(!sbuf->dma.pending);
+
+   SVGA_DBG(DEBUG_DMA, "dma to sid %p\n", sbuf->handle);
+
+   for (i = 0; i < sbuf->map.num_ranges; ++i) {
+      struct svga_buffer_range *range = &sbuf->map.ranges[i];
+      unsigned offset = range->start;
+      unsigned size = range->end - range->start;
+
+      while (offset < range->end) {
+         struct svga_winsys_buffer *hwbuf;
+         uint8_t *map;
+         enum pipe_error ret;
+
+         if (offset + size > range->end)
+            size = range->end - offset;
+
+         hwbuf = svga_winsys_buffer_create(ss, alignment, usage, size);
+         while (!hwbuf) {
+            size /= 2;
+            if (!size)
+               return PIPE_ERROR_OUT_OF_MEMORY;
+            hwbuf = svga_winsys_buffer_create(ss, alignment, usage, size);
+         }
+
+         SVGA_DBG(DEBUG_DMA, "  bytes %u - %u\n",
+                  offset, offset + size);
+
+         map = sws->buffer_map(sws, hwbuf,
+                               PIPE_BUFFER_USAGE_CPU_WRITE |
+                               PIPE_BUFFER_USAGE_DISCARD);
+         assert(map);
+         if (map) {
+            memcpy(map, sbuf->swbuf, size);
+            sws->buffer_unmap(sws, hwbuf);
+         }
+
+         ret = SVGA3D_BufferDMA(svga->swc,
+                                hwbuf, sbuf->handle,
+                                SVGA3D_WRITE_HOST_VRAM,
+                                size, 0, offset, sbuf->dma.flags);
+         if(ret != PIPE_OK) {
+            svga_context_flush(svga, NULL);
+            ret =  SVGA3D_BufferDMA(svga->swc,
+                                    hwbuf, sbuf->handle,
+                                    SVGA3D_WRITE_HOST_VRAM,
+                                    size, 0, offset, sbuf->dma.flags);
+            assert(ret == PIPE_OK);
+         }
+
+         sbuf->dma.flags.discard = FALSE;
+
+         sws->buffer_destroy(sws, hwbuf);
+
+         offset += size;
+      }
+   }
+
+   sbuf->map.num_ranges = 0;
+
    return PIPE_OK;
 }
 
@@ -659,34 +760,74 @@ svga_buffer_handle(struct svga_context *svga,
    sbuf = svga_buffer(buf);
    
    assert(!sbuf->map.count);
+   assert(!sbuf->user);
    
    if(!sbuf->handle) {
       ret = svga_buffer_create_host_surface(ss, sbuf);
       if(ret != PIPE_OK)
 	 return NULL;
-
-      ret = svga_buffer_update_hw(ss, sbuf);
-      if(ret != PIPE_OK)
-	 return NULL;
    }
 
-   if(!sbuf->needs_flush && sbuf->hw.num_ranges) {
-      /* Queue the buffer for flushing */
-      ret = svga_buffer_upload_command(svga, sbuf);
-      if(ret != PIPE_OK)
-         /* XXX: Should probably have a richer return value */
-         return NULL;
-
-      assert(sbuf->hw.svga == svga);
+   assert(sbuf->handle);
+
+   if (sbuf->map.num_ranges) {
+      if (!sbuf->dma.pending) {
+         /*
+          * No pending DMA upload yet, so insert a DMA upload command now.
+          */
+
+         /*
+          * Migrate the data from swbuf -> hwbuf if necessary.
+          */
+         ret = svga_buffer_update_hw(ss, sbuf);
+         if (ret == PIPE_OK) {
+            /*
+             * Queue a dma command.
+             */
+
+            ret = svga_buffer_upload_command(svga, sbuf);
+            if (ret == PIPE_ERROR_OUT_OF_MEMORY) {
+               svga_context_flush(svga, NULL);
+               ret = svga_buffer_upload_command(svga, sbuf);
+               assert(ret == PIPE_OK);
+            }
+            if (ret == PIPE_OK) {
+               sbuf->dma.pending = TRUE;
+               assert(!sbuf->head.prev && !sbuf->head.next);
+               LIST_ADDTAIL(&sbuf->head, &svga->dirty_buffers);
+            }
+         }
+         else if (ret == PIPE_ERROR_OUT_OF_MEMORY) {
+            /*
+             * The buffer is too big to fit in the GMR aperture, so break it in
+             * smaller pieces.
+             */
+            ret = svga_buffer_upload_piecewise(ss, svga, sbuf);
+         }
 
-      sbuf->needs_flush = TRUE;
-      assert(!sbuf->head.prev && !sbuf->head.next);
-      LIST_ADDTAIL(&sbuf->head, &svga->dirty_buffers);
+         if (ret != PIPE_OK) {
+            /*
+             * Something unexpected happened above. There is very little that
+             * we can do other than proceeding while ignoring the dirty ranges.
+             */
+            assert(0);
+            sbuf->map.num_ranges = 0;
+         }
+      }
+      else {
+         /*
+          * There a pending dma already. Make sure it is from this context.
+          */
+         assert(sbuf->dma.svga == svga);
+      }
    }
 
+   assert(!sbuf->map.num_ranges || sbuf->dma.pending);
+
    return sbuf->handle;
 }
 
+
 struct pipe_buffer *
 svga_screen_buffer_wrap_surface(struct pipe_screen *screen,
 				enum SVGA3dSurfaceFormat format,
@@ -739,7 +880,7 @@ svga_context_flush_buffers(struct svga_context *svga)
       sbuf = LIST_ENTRY(struct svga_buffer, curr, head);
 
       assert(p_atomic_read(&sbuf->base.reference.count) != 0);
-      assert(sbuf->needs_flush);
+      assert(sbuf->dma.pending);
       
       svga_buffer_upload_flush(svga, sbuf);
 
diff --git a/src/gallium/drivers/svga/svga_screen_buffer.h b/src/gallium/drivers/svga/svga_screen_buffer.h
index 448ac107c7f..8c862fa62d6 100644
--- a/src/gallium/drivers/svga/svga_screen_buffer.h
+++ b/src/gallium/drivers/svga/svga_screen_buffer.h
@@ -57,35 +57,6 @@ struct svga_buffer_range
 
 
 /**
- * Describe a
- *
- * This holds the information to emit a SVGA3dCmdSurfaceDMA.
- */
-struct svga_buffer_upload
-{
-   /**
-    * Guest memory region.
-    */
-   struct svga_winsys_buffer *buf;
-
-   struct svga_buffer_range ranges[SVGA_BUFFER_MAX_RANGES];
-   unsigned num_ranges;
-
-   SVGA3dSurfaceDMAFlags flags;
-
-   /**
-    * Pointer to the DMA copy box *inside* the command buffer.
-    */
-   SVGA3dCopyBox *boxes;
-
-   /**
-    * Context that has the pending DMA to this buffer.
-    */
-   struct svga_context *svga;
-};
-
-
-/**
  * SVGA pipe buffer.
  */
 struct svga_buffer 
@@ -111,14 +82,6 @@ struct svga_buffer
    boolean user;
    
    /**
-    * DMA'ble memory.
-    * 
-    * A piece of GMR memory. It is created when mapping the buffer, and will be
-    * used to upload/download vertex data from the host.
-    */
-   struct svga_buffer_upload hw;
-
-   /**
     * Creation key for the host surface handle.
     * 
     * This structure describes all the host surface characteristics so that it 
@@ -134,19 +97,94 @@ struct svga_buffer
     * trying to bind
     */
    struct svga_winsys_surface *handle;
-   
+
    /**
-    * Whether the host has been ever written.
+    * Information about ongoing and past map operations.
     */
-   boolean host_written;
-
    struct {
+      /**
+       * Number of concurrent mappings.
+       *
+       * XXX: It is impossible to guarantee concurrent maps work in all
+       * circumstances -- pipe_buffers really need transfer objects too.
+       */
       unsigned count;
+
+      /**
+       * Whether this buffer is currently mapped for writing.
+       */
       boolean writing;
+
+      /**
+       * Whether the application will tell us explicity which ranges it touched
+       * or not.
+       */
       boolean flush_explicit;
+
+      /**
+       * Dirty ranges.
+       *
+       * Ranges that were touched by the application and need to be uploaded to
+       * the host.
+       *
+       * This information will be copied into dma.boxes, when emiting the
+       * SVGA3dCmdSurfaceDMA command.
+       */
+      struct svga_buffer_range ranges[SVGA_BUFFER_MAX_RANGES];
+      unsigned num_ranges;
    } map;
-   
-   boolean needs_flush;
+
+   /**
+    * Information about uploaded version of user buffers.
+    */
+   struct {
+      struct pipe_buffer *buffer;
+
+      /**
+       * We combine multiple user buffers into the same hardware buffer. This
+       * is the relative offset within that buffer.
+       */
+      unsigned offset;
+   } uploaded;
+
+   /**
+    * DMA'ble memory.
+    *
+    * A piece of GMR memory, with the same size of the buffer. It is created
+    * when mapping the buffer, and will be used to upload vertex data to the
+    * host.
+    */
+   struct svga_winsys_buffer *hwbuf;
+
+   /**
+    * Information about pending DMA uploads.
+    *
+    */
+   struct {
+      /**
+       * Whether this buffer has an unfinished DMA upload command.
+       *
+       * If not set then the rest of the information is null.
+       */
+      boolean pending;
+
+      SVGA3dSurfaceDMAFlags flags;
+
+      /**
+       * Pointer to the DMA copy box *inside* the command buffer.
+       */
+      SVGA3dCopyBox *boxes;
+
+      /**
+       * Context that has the pending DMA to this buffer.
+       */
+      struct svga_context *svga;
+   } dma;
+
+   /**
+    * Linked list head, used to gather all buffers with pending dma uploads on
+    * a context. It is only valid if the dma.pending is set above.
+    */
    struct list_head head;
 };
 
@@ -176,6 +214,16 @@ svga_buffer_is_user_buffer( struct pipe_buffer *buffer )
 void
 svga_screen_init_buffer_functions(struct pipe_screen *screen);
 
+
+/**
+ * Get the host surface handle for this buffer.
+ *
+ * This will ensure the host surface is updated, issuing DMAs as needed.
+ *
+ * NOTE: This may insert new commands in the context, so it *must* be called
+ * before reserving command buffer space. And, in order to insert commands
+ * it may need to call svga_context_flush().
+ */
 struct svga_winsys_surface *
 svga_buffer_handle(struct svga_context *svga,
                    struct pipe_buffer *buf);
diff --git a/src/gallium/drivers/svga/svga_screen_texture.c b/src/gallium/drivers/svga/svga_screen_texture.c
index 12f3531a1df..5b581debfc7 100644
--- a/src/gallium/drivers/svga/svga_screen_texture.c
+++ b/src/gallium/drivers/svga/svga_screen_texture.c
@@ -59,21 +59,21 @@ svga_translate_format(enum pipe_format format)
 {
    switch(format) {
    
-   case PIPE_FORMAT_A8R8G8B8_UNORM:
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
       return SVGA3D_A8R8G8B8;
-   case PIPE_FORMAT_X8R8G8B8_UNORM:
+   case PIPE_FORMAT_B8G8R8X8_UNORM:
       return SVGA3D_X8R8G8B8;
 
       /* Required for GL2.1:
        */
-   case PIPE_FORMAT_A8R8G8B8_SRGB:
+   case PIPE_FORMAT_B8G8R8A8_SRGB:
       return SVGA3D_A8R8G8B8;
 
-   case PIPE_FORMAT_R5G6B5_UNORM:
+   case PIPE_FORMAT_B5G6R5_UNORM:
       return SVGA3D_R5G6B5;
-   case PIPE_FORMAT_A1R5G5B5_UNORM:
+   case PIPE_FORMAT_B5G5R5A1_UNORM:
       return SVGA3D_A1R5G5B5;
-   case PIPE_FORMAT_A4R4G4B4_UNORM:
+   case PIPE_FORMAT_B4G4R4A4_UNORM:
       return SVGA3D_A4R4G4B4;
 
       
@@ -83,9 +83,9 @@ svga_translate_format(enum pipe_format format)
     */
    case PIPE_FORMAT_Z16_UNORM:
       return SVGA3D_Z_D16;
-   case PIPE_FORMAT_Z24S8_UNORM:
+   case PIPE_FORMAT_S8Z24_UNORM:
       return SVGA3D_Z_D24S8;
-   case PIPE_FORMAT_Z24X8_UNORM:
+   case PIPE_FORMAT_X8Z24_UNORM:
       return SVGA3D_Z_D24X8;
 
    case PIPE_FORMAT_A8_UNORM:
@@ -111,13 +111,13 @@ SVGA3dSurfaceFormat
 svga_translate_format_render(enum pipe_format format)
 {
    switch(format) { 
-   case PIPE_FORMAT_A8R8G8B8_UNORM:
-   case PIPE_FORMAT_X8R8G8B8_UNORM:
-   case PIPE_FORMAT_A1R5G5B5_UNORM:
-   case PIPE_FORMAT_A4R4G4B4_UNORM:
-   case PIPE_FORMAT_R5G6B5_UNORM:
-   case PIPE_FORMAT_Z24S8_UNORM:
-   case PIPE_FORMAT_Z24X8_UNORM:
+   case PIPE_FORMAT_B8G8R8A8_UNORM:
+   case PIPE_FORMAT_B8G8R8X8_UNORM:
+   case PIPE_FORMAT_B5G5R5A1_UNORM:
+   case PIPE_FORMAT_B4G4R4A4_UNORM:
+   case PIPE_FORMAT_B5G6R5_UNORM:
+   case PIPE_FORMAT_S8Z24_UNORM:
+   case PIPE_FORMAT_X8Z24_UNORM:
    case PIPE_FORMAT_Z32_UNORM:
    case PIPE_FORMAT_Z16_UNORM:
    case PIPE_FORMAT_L8_UNORM:
@@ -315,7 +315,11 @@ svga_texture_create(struct pipe_screen *screen,
       tex->key.cachable = 0;
    }
 
-   if(templat->tex_usage & PIPE_TEXTURE_USAGE_PRIMARY) {
+   if(templat->tex_usage & PIPE_TEXTURE_USAGE_SHARED) {
+      tex->key.cachable = 0;
+   }
+
+   if(templat->tex_usage & PIPE_TEXTURE_USAGE_SCANOUT) {
       tex->key.flags |= SVGA3D_SURFACE_HINT_SCANOUT;
       tex->key.cachable = 0;
    }
@@ -355,80 +359,18 @@ error1:
 }
 
 
-static struct pipe_texture *
-svga_texture_blanket(struct pipe_screen * screen,
-                     const struct pipe_texture *base,
-                     const unsigned *stride,
-                     struct pipe_buffer *buffer)
-{
-   struct svga_texture *tex;
-   struct svga_buffer *sbuf = svga_buffer(buffer);
-   struct svga_winsys_screen *sws = svga_winsys_screen(screen);
-   assert(screen);
-
-   /* Only supports one type */
-   if (base->target != PIPE_TEXTURE_2D ||
-       base->last_level != 0 ||
-       base->depth0 != 1) {
-      return NULL;
-   }
-
-   /**
-    * We currently can't do texture blanket on
-    * SVGA3D_BUFFER. Need to blit to a temporary surface?
-    */
-
-   assert(sbuf->handle);
-   if (!sbuf->handle)
-      return NULL;
-
-   if (svga_translate_format(base->format) != sbuf->key.format) {
-      unsigned f1 = svga_translate_format(base->format);
-      unsigned f2 = sbuf->key.format;
-
-      /* It's okay for XRGB and ARGB or depth with/out stencil to get mixed up */
-      if ( !( (f1 == SVGA3D_X8R8G8B8 && f2 == SVGA3D_A8R8G8B8) ||
-              (f1 == SVGA3D_A8R8G8B8 && f2 == SVGA3D_X8R8G8B8) ||
-              (f1 == SVGA3D_Z_D24X8 && f2 == SVGA3D_Z_D24S8) ) ) {
-         debug_printf("%s wrong format %u != %u\n", __FUNCTION__, f1, f2);
-         return NULL;
-      }
-   }
-
-   tex = CALLOC_STRUCT(svga_texture);
-   if (!tex)
-      return NULL;
-
-   tex->base = *base;
-   
-
-   if (sbuf->key.format == 1)
-      tex->base.format = PIPE_FORMAT_X8R8G8B8_UNORM;
-   else if (sbuf->key.format == 2)
-      tex->base.format = PIPE_FORMAT_A8R8G8B8_UNORM;
-
-   pipe_reference_init(&tex->base.reference, 1);
-   tex->base.screen = screen;
-
-   SVGA_DBG(DEBUG_DMA, "blanket sid %p\n", sbuf->handle);
-
-   /* We don't own this storage, so don't try to cache it.
-    */
-   assert(sbuf->key.cachable == 0);
-   tex->key.cachable = 0;
-   sws->surface_reference(sws, &tex->handle, sbuf->handle);
 
-   return &tex->base;
-}
 
 
-struct pipe_texture *
-svga_screen_texture_wrap_surface(struct pipe_screen *screen,
-				 struct pipe_texture *base,
-				 enum SVGA3dSurfaceFormat format,
-				 struct svga_winsys_surface *srf)
+static struct pipe_texture *
+svga_screen_texture_from_handle(struct pipe_screen *screen,
+                                const struct pipe_texture *base,
+                                struct winsys_handle *whandle)
 {
+   struct svga_winsys_screen *sws = svga_winsys_screen(screen);
+   struct svga_winsys_surface *srf;
    struct svga_texture *tex;
+   enum SVGA3dSurfaceFormat format = 0;
    assert(screen);
 
    /* Only supports one type */
@@ -438,6 +380,8 @@ svga_screen_texture_wrap_surface(struct pipe_screen *screen,
       return NULL;
    }
 
+   srf = sws->surface_from_handle(sws, whandle, &format);
+
    if (!srf)
       return NULL;
 
@@ -462,9 +406,9 @@ svga_screen_texture_wrap_surface(struct pipe_screen *screen,
    
 
    if (format == 1)
-      tex->base.format = PIPE_FORMAT_X8R8G8B8_UNORM;
+      tex->base.format = PIPE_FORMAT_B8G8R8X8_UNORM;
    else if (format == 2)
-      tex->base.format = PIPE_FORMAT_A8R8G8B8_UNORM;
+      tex->base.format = PIPE_FORMAT_B8G8R8A8_UNORM;
 
    pipe_reference_init(&tex->base.reference, 1);
    tex->base.screen = screen;
@@ -478,6 +422,22 @@ svga_screen_texture_wrap_surface(struct pipe_screen *screen,
 }
 
 
+static boolean 
+svga_screen_texture_get_handle(struct pipe_screen *screen,
+                               struct pipe_texture *texture,
+                               struct winsys_handle *whandle)
+{
+   struct svga_winsys_screen *sws = svga_winsys_screen(texture->screen);
+   unsigned stride;
+
+   assert(svga_texture(texture)->key.cachable == 0);
+   svga_texture(texture)->key.cachable = 0;
+   stride = util_format_get_nblocksx(texture->format, texture->width0) *
+            util_format_get_blocksize(texture->format);
+   return sws->surface_get_handle(sws, svga_texture(texture)->handle, stride, whandle);
+}
+
+
 static void
 svga_texture_destroy(struct pipe_texture *pt)
 {
@@ -955,10 +915,11 @@ void
 svga_screen_init_texture_functions(struct pipe_screen *screen)
 {
    screen->texture_create = svga_texture_create;
+   screen->texture_from_handle = svga_screen_texture_from_handle;
+   screen->texture_get_handle = svga_screen_texture_get_handle;
    screen->texture_destroy = svga_texture_destroy;
    screen->get_tex_surface = svga_get_tex_surface;
    screen->tex_surface_destroy = svga_tex_surface_destroy;
-   screen->texture_blanket = svga_texture_blanket;
    screen->get_tex_transfer = svga_get_tex_transfer;
    screen->transfer_map = svga_transfer_map;
    screen->transfer_unmap = svga_transfer_unmap;
@@ -1120,33 +1081,3 @@ svga_destroy_sampler_view_priv(struct svga_sampler_view *v)
    pipe_texture_reference(&v->texture, NULL);
    FREE(v);
 }
-
-boolean
-svga_screen_buffer_from_texture(struct pipe_texture *texture,
-				struct pipe_buffer **buffer,
-				unsigned *stride)
-{
-   struct svga_texture *stex = svga_texture(texture);
-
-   *buffer = svga_screen_buffer_wrap_surface
-      (texture->screen,
-       svga_translate_format(texture->format),
-       stex->handle);
-
-   *stride = util_format_get_stride(texture->format, texture->width0);
-
-   return *buffer != NULL;
-}
-
-
-struct svga_winsys_surface *
-svga_screen_texture_get_winsys_surface(struct pipe_texture *texture)
-{
-   struct svga_winsys_screen *sws = svga_winsys_screen(texture->screen);
-   struct svga_winsys_surface *vsurf = NULL;
-
-   assert(svga_texture(texture)->key.cachable == 0);
-   svga_texture(texture)->key.cachable = 0;
-   sws->surface_reference(sws, &vsurf, svga_texture(texture)->handle);
-   return vsurf;
-}
diff --git a/src/gallium/drivers/svga/svga_screen_texture.h b/src/gallium/drivers/svga/svga_screen_texture.h
index 24c1f78ca55..ca6602b4369 100644
--- a/src/gallium/drivers/svga/svga_screen_texture.h
+++ b/src/gallium/drivers/svga/svga_screen_texture.h
@@ -78,7 +78,7 @@ struct svga_texture
 {
    struct pipe_texture base;
 
-   boolean defined[6][PIPE_MAX_TEXTURE_LEVELS];
+   boolean defined[6][SVGA_MAX_TEXTURE_LEVELS];
    
    struct svga_sampler_view *cached_view;
 
diff --git a/src/gallium/drivers/svga/svga_state_framebuffer.c b/src/gallium/drivers/svga/svga_state_framebuffer.c
index b4cafb8f219..b710914acda 100644
--- a/src/gallium/drivers/svga/svga_state_framebuffer.c
+++ b/src/gallium/drivers/svga/svga_state_framebuffer.c
@@ -70,7 +70,7 @@ static int emit_framebuffer( struct svga_context *svga,
          return ret;
 
       if (curr->zsbuf &&
-          curr->zsbuf->format == PIPE_FORMAT_Z24S8_UNORM) {
+          curr->zsbuf->format == PIPE_FORMAT_S8Z24_UNORM) {
          ret = SVGA3D_SetRenderTarget(svga->swc, SVGA3D_RT_STENCIL, curr->zsbuf);
          if (ret != PIPE_OK)
             return ret;
@@ -120,174 +120,153 @@ static int emit_viewport( struct svga_context *svga,
    float fb_width = svga->curr.framebuffer.width;
    float fb_height = svga->curr.framebuffer.height;
 
-   memset( &prescale, 0, sizeof(prescale) );
-
-   if (svga->curr.rast->templ.bypass_vs_clip_and_viewport) {
-
-      /* Avoid POSITIONT as it has a non trivial implementation outside the D3D
-       * API. Always generate a vertex shader.
-       */
-      rect.x = 0;
-      rect.y = 0;
-      rect.w = svga->curr.framebuffer.width;
-      rect.h = svga->curr.framebuffer.height;
-
-      prescale.scale[0] = 2.0 / (float)rect.w;
-      prescale.scale[1] = - 2.0 / (float)rect.h;
-      prescale.scale[2] = 1.0;
-      prescale.scale[3] = 1.0;
-      prescale.translate[0] = -1.0f;
-      prescale.translate[1] = 1.0f;
-      prescale.translate[2] = 0;
-      prescale.translate[3] = 0;
-      prescale.enabled = TRUE;
-   } else {
-
-      /* Examine gallium viewport transformation and produce a screen
-       * rectangle and possibly vertex shader pre-transformation to
-       * get the same results.
-       */
-      float fx =        viewport->scale[0] * -1.0 + viewport->translate[0];
-      float fy = flip * viewport->scale[1] * -1.0 + viewport->translate[1];
-      float fw =        viewport->scale[0] * 2; 
-      float fh = flip * viewport->scale[1] * 2; 
-
-      SVGA_DBG(DEBUG_VIEWPORT,
-               "\ninitial %f,%f %fx%f\n",
-               fx,
-               fy,
-               fw,
-               fh);
-
-      prescale.scale[0] = 1.0;
-      prescale.scale[1] = 1.0;
-      prescale.scale[2] = 1.0;
-      prescale.scale[3] = 1.0;
-      prescale.translate[0] = 0;
-      prescale.translate[1] = 0;
-      prescale.translate[2] = 0;
-      prescale.translate[3] = 0;
-      prescale.enabled = TRUE;
-
-
-
-      if (fw < 0) {
-         prescale.scale[0] *= -1.0;
-         prescale.translate[0] += -fw;
-         fw = -fw;
-         fx =        viewport->scale[0] * 1.0 + viewport->translate[0];
-      }
+   float fx =        viewport->scale[0] * -1.0 + viewport->translate[0];
+   float fy = flip * viewport->scale[1] * -1.0 + viewport->translate[1];
+   float fw =        viewport->scale[0] * 2; 
+   float fh = flip * viewport->scale[1] * 2; 
 
-      if (fh < 0) {
-         prescale.scale[1] *= -1.0;
-         prescale.translate[1] += -fh;
-         fh = -fh;
-         fy = flip * viewport->scale[1] * 1.0 + viewport->translate[1];
-      }
+   memset( &prescale, 0, sizeof(prescale) );
 
-      if (fx < 0) {
-         prescale.translate[0] += fx;
-         prescale.scale[0] *= fw / (fw + fx); 
-         fw += fx;
-         fx = 0;
-      }
+   /* Examine gallium viewport transformation and produce a screen
+    * rectangle and possibly vertex shader pre-transformation to
+    * get the same results.
+    */
 
-      if (fy < 0) {
-         prescale.translate[1] += fy;
-         prescale.scale[1] *= fh / (fh + fy); 
-         fh += fy;
-         fy = 0;
-      }
+   SVGA_DBG(DEBUG_VIEWPORT,
+            "\ninitial %f,%f %fx%f\n",
+            fx,
+            fy,
+            fw,
+            fh);
+
+   prescale.scale[0] = 1.0;
+   prescale.scale[1] = 1.0;
+   prescale.scale[2] = 1.0;
+   prescale.scale[3] = 1.0;
+   prescale.translate[0] = 0;
+   prescale.translate[1] = 0;
+   prescale.translate[2] = 0;
+   prescale.translate[3] = 0;
+   prescale.enabled = TRUE;
+
+
+
+   if (fw < 0) {
+      prescale.scale[0] *= -1.0;
+      prescale.translate[0] += -fw;
+      fw = -fw;
+      fx =        viewport->scale[0] * 1.0 + viewport->translate[0];
+   }
 
-      if (fx + fw > fb_width) {
-         prescale.scale[0] *= fw / (fb_width - fx); 
-         prescale.translate[0] -= fx * (fw / (fb_width - fx));
-         prescale.translate[0] += fx;
-         fw = fb_width - fx;
-         
-      }
+   if (fh < 0) {
+      prescale.scale[1] *= -1.0;
+      prescale.translate[1] += -fh;
+      fh = -fh;
+      fy = flip * viewport->scale[1] * 1.0 + viewport->translate[1];
+   }
 
-      if (fy + fh > fb_height) {
-         prescale.scale[1] *= fh / (fb_height - fy);
-         prescale.translate[1] -= fy * (fh / (fb_height - fy));
-         prescale.translate[1] += fy;
-         fh = fb_height - fy;
-      }
+   if (fx < 0) {
+      prescale.translate[0] += fx;
+      prescale.scale[0] *= fw / (fw + fx); 
+      fw += fx;
+      fx = 0;
+   }
 
-      if (fw < 0 || fh < 0) {
-         fw = fh = fx = fy = 0;
-         degenerate = TRUE;
-         goto out;
-      }
+   if (fy < 0) {
+      prescale.translate[1] += fy;
+      prescale.scale[1] *= fh / (fh + fy); 
+      fh += fy;
+      fy = 0;
+   }
 
+   if (fx + fw > fb_width) {
+      prescale.scale[0] *= fw / (fb_width - fx); 
+      prescale.translate[0] -= fx * (fw / (fb_width - fx));
+      prescale.translate[0] += fx;
+      fw = fb_width - fx;
+      
+   }
 
-      /* D3D viewport is integer space.  Convert fx,fy,etc. to
-       * integers.
-       *
-       * TODO: adjust pretranslate correct for any subpixel error
-       * introduced converting to integers.
-       */
-      rect.x = fx;
-      rect.y = fy;
-      rect.w = fw;
-      rect.h = fh;
+   if (fy + fh > fb_height) {
+      prescale.scale[1] *= fh / (fb_height - fy);
+      prescale.translate[1] -= fy * (fh / (fb_height - fy));
+      prescale.translate[1] += fy;
+      fh = fb_height - fy;
+   }
 
-      SVGA_DBG(DEBUG_VIEWPORT,
-               "viewport error %f,%f %fx%f\n",
-               fabs((float)rect.x - fx),
-               fabs((float)rect.y - fy),
-               fabs((float)rect.w - fw),
-               fabs((float)rect.h - fh));
+   if (fw < 0 || fh < 0) {
+      fw = fh = fx = fy = 0;
+      degenerate = TRUE;
+      goto out;
+   }
 
-      SVGA_DBG(DEBUG_VIEWPORT,
-               "viewport %d,%d %dx%d\n",
-               rect.x,
-               rect.y,
-               rect.w,
-               rect.h);
-      
 
-      /* Finally, to get GL rasterization rules, need to tweak the
-       * screen-space coordinates slightly relative to D3D which is
-       * what hardware implements natively.
-       */
-      if (svga->curr.rast->templ.gl_rasterization_rules) {
-         float adjust_x = 0.0;
-         float adjust_y = 0.0;
-
-         switch (svga->curr.reduced_prim) {
-         case PIPE_PRIM_LINES:
-            adjust_x = -0.5;
-            adjust_y = 0;
-            break;
-         case PIPE_PRIM_POINTS:
-         case PIPE_PRIM_TRIANGLES:
-            adjust_x = -0.375;
-            adjust_y = -0.5;
-            break;
-         }
-
-         prescale.translate[0] += adjust_x;
-         prescale.translate[1] += adjust_y;
-         prescale.translate[2] = 0.5; /* D3D clip space */
-         prescale.scale[2]     = 0.5; /* D3D clip space */
+   /* D3D viewport is integer space.  Convert fx,fy,etc. to
+    * integers.
+    *
+    * TODO: adjust pretranslate correct for any subpixel error
+    * introduced converting to integers.
+    */
+   rect.x = fx;
+   rect.y = fy;
+   rect.w = fw;
+   rect.h = fh;
+
+   SVGA_DBG(DEBUG_VIEWPORT,
+            "viewport error %f,%f %fx%f\n",
+            fabs((float)rect.x - fx),
+            fabs((float)rect.y - fy),
+            fabs((float)rect.w - fw),
+            fabs((float)rect.h - fh));
+
+   SVGA_DBG(DEBUG_VIEWPORT,
+            "viewport %d,%d %dx%d\n",
+            rect.x,
+            rect.y,
+            rect.w,
+            rect.h);
+
+
+   /* Finally, to get GL rasterization rules, need to tweak the
+    * screen-space coordinates slightly relative to D3D which is
+    * what hardware implements natively.
+    */
+   if (svga->curr.rast->templ.gl_rasterization_rules) {
+      float adjust_x = 0.0;
+      float adjust_y = 0.0;
+
+      switch (svga->curr.reduced_prim) {
+      case PIPE_PRIM_LINES:
+         adjust_x = -0.5;
+         adjust_y = 0;
+         break;
+      case PIPE_PRIM_POINTS:
+      case PIPE_PRIM_TRIANGLES:
+         adjust_x = -0.375;
+         adjust_y = -0.5;
+         break;
       }
 
+      prescale.translate[0] += adjust_x;
+      prescale.translate[1] += adjust_y;
+      prescale.translate[2] = 0.5; /* D3D clip space */
+      prescale.scale[2]     = 0.5; /* D3D clip space */
+   }
 
-      range_min = viewport->scale[2] * -1.0 + viewport->translate[2];
-      range_max = viewport->scale[2] *  1.0 + viewport->translate[2];
 
-      /* D3D (and by implication SVGA) doesn't like dealing with zmax
-       * less than zmin.  Detect that case, flip the depth range and
-       * invert our z-scale factor to achieve the same effect.
-       */
-      if (range_min > range_max) {
-         float range_tmp;
-         range_tmp = range_min; 
-         range_min = range_max; 
-         range_max = range_tmp;
-         prescale.scale[2]     = -prescale.scale[2];
-      }
+   range_min = viewport->scale[2] * -1.0 + viewport->translate[2];
+   range_max = viewport->scale[2] *  1.0 + viewport->translate[2];
+
+   /* D3D (and by implication SVGA) doesn't like dealing with zmax
+    * less than zmin.  Detect that case, flip the depth range and
+    * invert our z-scale factor to achieve the same effect.
+    */
+   if (range_min > range_max) {
+      float range_tmp;
+      range_tmp = range_min; 
+      range_min = range_max; 
+      range_max = range_tmp;
+      prescale.scale[2]     = -prescale.scale[2];
    }
 
    if (prescale.enabled) {
diff --git a/src/gallium/drivers/svga/svga_state_need_swtnl.c b/src/gallium/drivers/svga/svga_state_need_swtnl.c
index dd13a89d24d..dfaab53aef4 100644
--- a/src/gallium/drivers/svga/svga_state_need_swtnl.c
+++ b/src/gallium/drivers/svga/svga_state_need_swtnl.c
@@ -43,7 +43,7 @@ svga_translate_vertex_format(enum pipe_format format)
    case PIPE_FORMAT_R32G32_FLOAT:         return SVGA3D_DECLTYPE_FLOAT2;
    case PIPE_FORMAT_R32G32B32_FLOAT:      return SVGA3D_DECLTYPE_FLOAT3;
    case PIPE_FORMAT_R32G32B32A32_FLOAT:   return SVGA3D_DECLTYPE_FLOAT4;
-   case PIPE_FORMAT_B8G8R8A8_UNORM:       return SVGA3D_DECLTYPE_D3DCOLOR;
+   case PIPE_FORMAT_A8R8G8B8_UNORM:       return SVGA3D_DECLTYPE_D3DCOLOR;
    case PIPE_FORMAT_R8G8B8A8_USCALED:     return SVGA3D_DECLTYPE_UBYTE4;
    case PIPE_FORMAT_R16G16_SSCALED:       return SVGA3D_DECLTYPE_SHORT2;
    case PIPE_FORMAT_R16G16B16A16_SSCALED: return SVGA3D_DECLTYPE_SHORT4;
@@ -76,8 +76,13 @@ static int update_need_swvfetch( struct svga_context *svga,
    unsigned i;
    boolean need_swvfetch = FALSE;
 
-   for (i = 0; i < svga->curr.num_vertex_elements; i++) {
-      svga->state.sw.ve_format[i] = svga_translate_vertex_format(svga->curr.ve[i].src_format);
+   if (!svga->curr.velems) {
+      /* No vertex elements bound. */
+      return 0;
+   }
+
+   for (i = 0; i < svga->curr.velems->count; i++) {
+      svga->state.sw.ve_format[i] = svga_translate_vertex_format(svga->curr.velems->velem[i].src_format);
       if (svga->state.sw.ve_format[i] == SVGA3D_DECLTYPE_MAX) {
          need_swvfetch = TRUE;
          break;
@@ -129,8 +134,7 @@ static int update_need_pipeline( struct svga_context *svga,
 
    /* SVGA_NEW_CLIP 
     */
-   if (!svga->curr.rast->templ.bypass_vs_clip_and_viewport &&
-       svga->curr.clip.nr) {
+   if (svga->curr.clip.nr) {
       SVGA_DBG(DEBUG_SWTNL, "%s: userclip\n", __FUNCTION__);
       need_pipeline = TRUE;
    }
diff --git a/src/gallium/drivers/svga/svga_state_rss.c b/src/gallium/drivers/svga/svga_state_rss.c
index 107cc403b4d..b7195d246bc 100644
--- a/src/gallium/drivers/svga/svga_state_rss.c
+++ b/src/gallium/drivers/svga/svga_state_rss.c
@@ -191,15 +191,24 @@ static int emit_rss( struct svga_context *svga,
       EMIT_RS( svga, svga->curr.stencil_ref.ref_value[0], STENCILREF, fail );
    }
 
-   if (dirty & SVGA_NEW_RAST)
+   if (dirty & (SVGA_NEW_RAST | SVGA_NEW_NEED_PIPELINE))
    {
       const struct svga_rasterizer_state *curr = svga->curr.rast; 
+      unsigned cullmode = curr->cullmode;
 
       /* Shademode: still need to rearrange index list to move
        * flat-shading PV first vertex.
        */
       EMIT_RS( svga, curr->shademode, SHADEMODE, fail );
-      EMIT_RS( svga, curr->cullmode, CULLMODE, fail );
+
+      /* Don't do culling while the software pipeline is active.  It
+       * does it for us, and additionally introduces potentially
+       * back-facing triangles.
+       */
+      if (svga->state.sw.need_pipeline)
+         cullmode = SVGA3D_FACE_NONE;
+
+      EMIT_RS( svga, cullmode, CULLMODE, fail );
       EMIT_RS( svga, curr->scissortestenable, SCISSORTESTENABLE, fail );
       EMIT_RS( svga, curr->multisampleantialias, MULTISAMPLEANTIALIAS, fail );
       EMIT_RS( svga, curr->lastpixel, LASTPIXEL, fail );
diff --git a/src/gallium/drivers/svga/svga_state_vdecl.c b/src/gallium/drivers/svga/svga_state_vdecl.c
index d1066ce13b0..f531e223048 100644
--- a/src/gallium/drivers/svga/svga_state_vdecl.c
+++ b/src/gallium/drivers/svga/svga_state_vdecl.c
@@ -54,33 +54,30 @@ upload_user_buffers( struct svga_context *svga )
    {
       if (svga_buffer_is_user_buffer(svga->curr.vb[i].buffer))
       {
-         struct pipe_buffer *upload_buffer = NULL;
-         unsigned offset = /*svga->curr.vb[i].buffer_offset*/ 0;
-         unsigned size = svga->curr.vb[i].buffer->size /*- offset*/;
-         unsigned upload_offset;
-
-         ret = u_upload_buffer( svga->upload_vb,
-                                offset,
-                                size,
-                                svga->curr.vb[i].buffer,
-                                &upload_offset,
-                                &upload_buffer );
-         if (ret)
-            return ret;
-
-         if (0)
-            debug_printf("%s: %d: orig buf %p upl buf %p ofs %d sz %d\n", 
-                         __FUNCTION__, 
-                         i,
-                         svga->curr.vb[i].buffer,
-                         upload_buffer, upload_offset, size);
-
-         /* Make sure we release the old buffer and end up with the
-          * correct refcount on the uploaded buffer.
-          */
-         pipe_buffer_reference( &svga->curr.vb[i].buffer, NULL );
-         svga->curr.vb[i].buffer = upload_buffer;
-         svga->curr.vb[i].buffer_offset = upload_offset;
+         struct svga_buffer *buffer = svga_buffer(svga->curr.vb[i].buffer);
+
+         if (!buffer->uploaded.buffer) {
+            ret = u_upload_buffer( svga->upload_vb,
+                                   0,
+                                   buffer->base.size,
+                                   &buffer->base,
+                                   &buffer->uploaded.offset,
+                                   &buffer->uploaded.buffer );
+            if (ret)
+               return ret;
+
+            if (0)
+               debug_printf("%s: %d: orig buf %p upl buf %p ofs %d sz %d\n",
+                            __FUNCTION__,
+                            i,
+                            buffer,
+                            buffer->uploaded.buffer,
+                            buffer->uploaded.offset,
+                            buffer->base.size);
+         }
+
+         pipe_buffer_reference( &svga->curr.vb[i].buffer, buffer->uploaded.buffer );
+         svga->curr.vb[i].buffer_offset = buffer->uploaded.offset;
       }
    }
 
@@ -98,17 +95,17 @@ upload_user_buffers( struct svga_context *svga )
 static int emit_hw_vs_vdecl( struct svga_context *svga,
                              unsigned dirty )
 {
-   const struct pipe_vertex_element *ve = svga->curr.ve;
+   const struct pipe_vertex_element *ve = svga->curr.velems->velem;
    SVGA3dVertexDecl decl;
    unsigned i;
 
-   assert(svga->curr.num_vertex_elements >=
+   assert(svga->curr.velems->count >=
           svga->curr.vs->base.info.file_count[TGSI_FILE_INPUT]);
 
    svga_hwtnl_reset_vdecl( svga->hwtnl, 
-                           svga->curr.num_vertex_elements );
+                           svga->curr.velems->count );
 
-   for (i = 0; i < svga->curr.num_vertex_elements; i++) {
+   for (i = 0; i < svga->curr.velems->count; i++) {
       const struct pipe_vertex_buffer *vb = &svga->curr.vb[ve[i].vertex_buffer_index];
       unsigned usage, index;
 
diff --git a/src/gallium/drivers/svga/svga_state_vs.c b/src/gallium/drivers/svga/svga_state_vs.c
index d7999fe53d2..781f7bf5339 100644
--- a/src/gallium/drivers/svga/svga_state_vs.c
+++ b/src/gallium/drivers/svga/svga_state_vs.c
@@ -186,8 +186,8 @@ static int update_zero_stride( struct svga_context *svga,
    svga->curr.zero_stride_vertex_elements = 0;
    svga->curr.num_zero_stride_vertex_elements = 0;
 
-   for (i = 0; i < svga->curr.num_vertex_elements; i++) {
-      const struct pipe_vertex_element *vel = &svga->curr.ve[i];
+   for (i = 0; i < svga->curr.velems->count; i++) {
+      const struct pipe_vertex_element *vel = &svga->curr.velems->velem[i];
       const struct pipe_vertex_buffer *vbuffer = &svga->curr.vb[
          vel->vertex_buffer_index];
       if (vbuffer->stride == 0) {
diff --git a/src/gallium/drivers/svga/svga_swtnl_state.c b/src/gallium/drivers/svga/svga_swtnl_state.c
index 35f36a828fd..246d34e649e 100644
--- a/src/gallium/drivers/svga/svga_swtnl_state.c
+++ b/src/gallium/drivers/svga/svga_swtnl_state.c
@@ -99,8 +99,8 @@ static int update_swtnl_draw( struct svga_context *svga,
 
    if (dirty & SVGA_NEW_VELEMENT)
       draw_set_vertex_elements(svga->swtnl.draw, 
-                               svga->curr.num_vertex_elements, 
-                               svga->curr.ve );
+                               svga->curr.velems->count, 
+                               svga->curr.velems->velem );
 
    if (dirty & SVGA_NEW_CLIP)
       draw_set_clip_state(svga->swtnl.draw, 
diff --git a/src/gallium/drivers/svga/svga_tgsi_emit.h b/src/gallium/drivers/svga/svga_tgsi_emit.h
index e8f75485d55..48eced2ecea 100644
--- a/src/gallium/drivers/svga/svga_tgsi_emit.h
+++ b/src/gallium/drivers/svga/svga_tgsi_emit.h
@@ -138,6 +138,7 @@ static INLINE boolean emit_dst( struct svga_shader_emitter *emit,
                          SVGA3dShaderDestToken dest )
 {
    assert(dest.reserved0);
+   assert(dest.mask);
    return svga_shader_emit_dword( emit, dest.value );
 }
 
@@ -267,6 +268,7 @@ static INLINE SVGA3dShaderDestToken
 writemask( SVGA3dShaderDestToken dest,
            unsigned mask )
 {
+   assert(dest.mask & mask);
    dest.mask &= mask;
    return dest;
 }
diff --git a/src/gallium/drivers/svga/svga_tgsi_insn.c b/src/gallium/drivers/svga/svga_tgsi_insn.c
index 87aed39f78f..3d4f56a67bd 100644
--- a/src/gallium/drivers/svga/svga_tgsi_insn.c
+++ b/src/gallium/drivers/svga/svga_tgsi_insn.c
@@ -112,6 +112,7 @@ translate_dst_register( struct svga_shader_emitter *emit,
    }
 
    dest.mask = reg->Register.WriteMask;
+   assert(dest.mask);
 
    if (insn->Instruction.Saturate) 
       dest.dstMod = SVGA3DDSTMOD_SATURATE;
@@ -1410,34 +1411,42 @@ static boolean emit_tex(struct svga_shader_emitter *emit,
 
 
    if (compare) {
-      SVGA3dShaderDestToken src0_zdivw = get_temp( emit );
-      struct src_register tex_src_x = scalar(src(tex_result), TGSI_SWIZZLE_Y);
-      struct src_register one =
-         scalar( get_zero_immediate( emit ), TGSI_SWIZZLE_W );
-
-      /* Divide texcoord R by Q */
-      if (!submit_op1( emit, inst_token( SVGA3DOP_RCP ),
-                       src0_zdivw,
-                       scalar(src0, TGSI_SWIZZLE_W) ))
-         return FALSE;
+      if (dst.mask & TGSI_WRITEMASK_XYZ) {
+         SVGA3dShaderDestToken src0_zdivw = get_temp( emit );
+         struct src_register tex_src_x = scalar(src(tex_result), TGSI_SWIZZLE_Y);
+
+         /* Divide texcoord R by Q */
+         if (!submit_op1( emit, inst_token( SVGA3DOP_RCP ),
+                          writemask(src0_zdivw, TGSI_WRITEMASK_X),
+                          scalar(src0, TGSI_SWIZZLE_W) ))
+            return FALSE;
 
-      if (!submit_op2( emit, inst_token( SVGA3DOP_MUL ),
-                       src0_zdivw,
-                       scalar(src0, TGSI_SWIZZLE_Z),
-                       src(src0_zdivw) ))
-         return FALSE;
+         if (!submit_op2( emit, inst_token( SVGA3DOP_MUL ),
+                          writemask(src0_zdivw, TGSI_WRITEMASK_X),
+                          scalar(src0, TGSI_SWIZZLE_Z),
+                          scalar(src(src0_zdivw), TGSI_SWIZZLE_X) ))
+            return FALSE;
 
-      if (!emit_select(
-             emit,
-             emit->key.fkey.tex[src1.base.num].compare_func,
-             dst,
-             src(src0_zdivw),
-             tex_src_x))
-         return FALSE;
+         if (!emit_select(
+                emit,
+                emit->key.fkey.tex[src1.base.num].compare_func,
+                writemask( dst, TGSI_WRITEMASK_XYZ ),
+                scalar(src(src0_zdivw), TGSI_SWIZZLE_X),
+                tex_src_x))
+            return FALSE;
+      }
 
-      return submit_op1( emit, inst_token( SVGA3DOP_MOV ),
-                         writemask( dst, TGSI_WRITEMASK_W),
-                         one );
+      if (dst.mask & TGSI_WRITEMASK_W) {
+         struct src_register one =
+            scalar( get_zero_immediate( emit ), TGSI_SWIZZLE_W );
+
+        if (!submit_op1( emit, inst_token( SVGA3DOP_MOV ),
+                         writemask( dst, TGSI_WRITEMASK_W ),
+                         one ))
+           return FALSE;
+      }
+
+      return TRUE;
    }
    else if (!emit->use_sm30 && dst.mask != TGSI_WRITEMASK_XYZW) 
    {
@@ -1827,13 +1836,13 @@ static boolean emit_exp(struct svga_shader_emitter *emit,
     */
    if (dst.mask & TGSI_WRITEMASK_X) {
       if (!submit_op2( emit, inst_token( SVGA3DOP_ADD ),
-                       writemask( dst, dst.mask & TGSI_WRITEMASK_X ),
+                       writemask( dst, TGSI_WRITEMASK_X ),
                        src0,
                        scalar( negate( src( fraction ) ), TGSI_SWIZZLE_Y ) ) )
          return FALSE;
 
       if (!submit_op1( emit, inst_token( SVGA3DOP_EXP ),
-                       writemask( dst, dst.mask & TGSI_WRITEMASK_X ),
+                       writemask( dst, TGSI_WRITEMASK_X ),
                        scalar( src( dst ), TGSI_SWIZZLE_X ) ) )
          return FALSE;
 
@@ -1845,7 +1854,7 @@ static boolean emit_exp(struct svga_shader_emitter *emit,
     */
    if (dst.mask & TGSI_WRITEMASK_Z) {
       if (!submit_op1( emit, inst_token( SVGA3DOP_EXPP ),
-                       writemask( dst, dst.mask & TGSI_WRITEMASK_Z ),
+                       writemask( dst, TGSI_WRITEMASK_Z ),
                        src0 ) )
          return FALSE;
    }
diff --git a/src/gallium/drivers/svga/svga_winsys.h b/src/gallium/drivers/svga/svga_winsys.h
index b4e3af0eafc..d4bb176f9a8 100644
--- a/src/gallium/drivers/svga/svga_winsys.h
+++ b/src/gallium/drivers/svga/svga_winsys.h
@@ -51,6 +51,7 @@ struct pipe_context;
 struct pipe_fence_handle;
 struct pipe_texture;
 struct svga_region;
+struct winsys_handle;
 
 
 #define SVGA_BUFFER_USAGE_PINNED  (PIPE_BUFFER_USAGE_CUSTOM << 0)
@@ -187,6 +188,25 @@ struct svga_winsys_screen
                      uint32 numMipLevels);
 
    /**
+    * Creates a surface from a winsys handle.
+    * Used to implement pipe_screen::texture_from_handle.
+    */
+   struct svga_winsys_surface *
+   (*surface_from_handle)(struct svga_winsys_screen *sws,
+                          struct winsys_handle *whandle,
+                          SVGA3dSurfaceFormat *format);
+
+   /**
+    * Get a winsys_handle from a surface.
+    * Used to implement pipe_screen::texture_get_handle.
+    */
+   boolean
+   (*surface_get_handle)(struct svga_winsys_screen *sws,
+                         struct svga_winsys_surface *surface,
+                         unsigned stride,
+                         struct winsys_handle *whandle);
+
+   /**
     * Whether this surface is sitting in a validate list
     */
    boolean
@@ -284,19 +304,6 @@ svga_screen_buffer_wrap_surface(struct pipe_screen *screen,
 				struct svga_winsys_surface *srf);
 
 struct svga_winsys_surface *
-svga_screen_texture_get_winsys_surface(struct pipe_texture *texture);
-struct svga_winsys_surface *
 svga_screen_buffer_get_winsys_surface(struct pipe_buffer *buffer);
 
-boolean
-svga_screen_buffer_from_texture(struct pipe_texture *texture,
-				struct pipe_buffer **buffer,
-				unsigned *stride);
-
-struct pipe_texture *
-svga_screen_texture_wrap_surface(struct pipe_screen *screen,
-				 struct pipe_texture *base,
-				 enum SVGA3dSurfaceFormat format,
-				 struct svga_winsys_surface *srf);
-
 #endif /* SVGA_WINSYS_H_ */
diff --git a/src/gallium/drivers/trace/tr_context.c b/src/gallium/drivers/trace/tr_context.c
index c84cbd0e521..f9555fb922c 100644
--- a/src/gallium/drivers/trace/tr_context.c
+++ b/src/gallium/drivers/trace/tr_context.c
@@ -774,6 +774,70 @@ trace_context_delete_vs_state(struct pipe_context *_pipe,
 }
 
 
+static INLINE void *
+trace_context_create_vertex_elements_state(struct pipe_context *_pipe,
+                                           unsigned num_elements,
+                                           const struct  pipe_vertex_element *elements)
+{
+   struct trace_context *tr_ctx = trace_context(_pipe);
+   struct pipe_context *pipe = tr_ctx->pipe;
+   void * result;
+
+   trace_dump_call_begin("pipe_context", "create_vertex_elements_state");
+
+   trace_dump_arg(ptr, pipe);
+   trace_dump_arg(uint, num_elements);
+
+   trace_dump_arg_begin("elements");
+   trace_dump_struct_array(vertex_element, elements, num_elements);
+   trace_dump_arg_end();
+
+   result = pipe->create_vertex_elements_state(pipe, num_elements, elements);
+
+   trace_dump_ret(ptr, result);
+
+   trace_dump_call_end();
+
+   return result;
+}
+
+
+static INLINE void
+trace_context_bind_vertex_elements_state(struct pipe_context *_pipe,
+                                         void *state)
+{
+   struct trace_context *tr_ctx = trace_context(_pipe);
+   struct pipe_context *pipe = tr_ctx->pipe;
+
+   trace_dump_call_begin("pipe_context", "bind_vertex_elements_state");
+
+   trace_dump_arg(ptr, pipe);
+   trace_dump_arg(ptr, state);
+
+   pipe->bind_vertex_elements_state(pipe, state);
+
+   trace_dump_call_end();
+}
+
+
+static INLINE void
+trace_context_delete_vertex_elements_state(struct pipe_context *_pipe,
+                                           void *state)
+{
+   struct trace_context *tr_ctx = trace_context(_pipe);
+   struct pipe_context *pipe = tr_ctx->pipe;
+
+   trace_dump_call_begin("pipe_context", "delete_verte_elements_state");
+
+   trace_dump_arg(ptr, pipe);
+   trace_dump_arg(ptr, state);
+
+   pipe->delete_vertex_elements_state(pipe, state);
+
+   trace_dump_call_end();
+}
+
+
 static INLINE void
 trace_context_set_blend_color(struct pipe_context *_pipe,
                               const struct pipe_blend_color *state)
@@ -1105,29 +1169,6 @@ trace_context_set_vertex_buffers(struct pipe_context *_pipe,
 
 
 static INLINE void
-trace_context_set_vertex_elements(struct pipe_context *_pipe,
-                                  unsigned num_elements,
-                                  const struct pipe_vertex_element *elements)
-{
-   struct trace_context *tr_ctx = trace_context(_pipe);
-   struct pipe_context *pipe = tr_ctx->pipe;
-
-   trace_dump_call_begin("pipe_context", "set_vertex_elements");
-
-   trace_dump_arg(ptr, pipe);
-   trace_dump_arg(uint, num_elements);
-
-   trace_dump_arg_begin("elements");
-   trace_dump_struct_array(vertex_element, elements, num_elements);
-   trace_dump_arg_end();
-
-   pipe->set_vertex_elements(pipe, num_elements, elements);
-
-   trace_dump_call_end();
-}
-
-
-static INLINE void
 trace_context_surface_copy(struct pipe_context *_pipe,
                            struct pipe_surface *dest,
                            unsigned destx, unsigned desty,
@@ -1360,6 +1401,9 @@ trace_context_create(struct trace_screen *tr_scr,
    tr_ctx->base.create_vs_state = trace_context_create_vs_state;
    tr_ctx->base.bind_vs_state = trace_context_bind_vs_state;
    tr_ctx->base.delete_vs_state = trace_context_delete_vs_state;
+   tr_ctx->base.create_vertex_elements_state = trace_context_create_vertex_elements_state;
+   tr_ctx->base.bind_vertex_elements_state = trace_context_bind_vertex_elements_state;
+   tr_ctx->base.delete_vertex_elements_state = trace_context_delete_vertex_elements_state;
    tr_ctx->base.set_blend_color = trace_context_set_blend_color;
    tr_ctx->base.set_stencil_ref = trace_context_set_stencil_ref;
    tr_ctx->base.set_clip_state = trace_context_set_clip_state;
@@ -1373,7 +1417,6 @@ trace_context_create(struct trace_screen *tr_scr,
    tr_ctx->base.create_sampler_view = trace_create_sampler_view;
    tr_ctx->base.sampler_view_destroy = trace_sampler_view_destroy;
    tr_ctx->base.set_vertex_buffers = trace_context_set_vertex_buffers;
-   tr_ctx->base.set_vertex_elements = trace_context_set_vertex_elements;
    if (pipe->surface_copy)
       tr_ctx->base.surface_copy = trace_context_surface_copy;
    if (pipe->surface_fill)
diff --git a/src/gallium/drivers/trace/tr_drm.c b/src/gallium/drivers/trace/tr_drm.c
index 919dc1b309f..906b3262e4b 100644
--- a/src/gallium/drivers/trace/tr_drm.c
+++ b/src/gallium/drivers/trace/tr_drm.c
@@ -28,11 +28,12 @@
 #include "state_tracker/drm_api.h"
 
 #include "util/u_memory.h"
-#include "trace/tr_drm.h"
-#include "trace/tr_screen.h"
-#include "trace/tr_context.h"
-#include "trace/tr_buffer.h"
-#include "trace/tr_texture.h"
+#include "tr_drm.h"
+#include "tr_screen.h"
+#include "tr_context.h"
+#include "tr_buffer.h"
+#include "tr_texture.h"
+#include "tr_public.h"
 
 struct trace_drm_api
 {
@@ -62,69 +63,8 @@ trace_drm_create_screen(struct drm_api *_api, int fd,
 
    screen = api->create_screen(api, fd, arg);
 
-   return trace_screen_create(screen);
-}
-
-
-static struct pipe_texture *
-trace_drm_texture_from_shared_handle(struct drm_api *_api,
-                                     struct pipe_screen *_screen,
-                                     struct pipe_texture *templ,
-                                     const char *name,
-                                     unsigned stride,
-                                     unsigned handle)
-{
-   struct trace_screen *tr_screen = trace_screen(_screen);
-   struct trace_drm_api *tr_api = trace_drm_api(_api);
-   struct pipe_screen *screen = tr_screen->screen;
-   struct drm_api *api = tr_api->api;
-   struct pipe_texture *result;
-
-   /* TODO trace call */
-
-   result = api->texture_from_shared_handle(api, screen, templ, name, stride, handle);
-
-   result = trace_texture_create(trace_screen(_screen), result);
-
-   return result;
-}
-
-static boolean
-trace_drm_shared_handle_from_texture(struct drm_api *_api,
-                                     struct pipe_screen *_screen,
-                                     struct pipe_texture *_texture,
-                                     unsigned *stride,
-                                     unsigned *handle)
-{
-   struct trace_screen *tr_screen = trace_screen(_screen);
-   struct trace_texture *tr_texture = trace_texture(_texture);
-   struct trace_drm_api *tr_api = trace_drm_api(_api);
-   struct pipe_screen *screen = tr_screen->screen;
-   struct pipe_texture *texture = tr_texture->texture;
-   struct drm_api *api = tr_api->api;
-
-   /* TODO trace call */
-
-   return api->shared_handle_from_texture(api, screen, texture, stride, handle);
-}
 
-static boolean
-trace_drm_local_handle_from_texture(struct drm_api *_api,
-                                    struct pipe_screen *_screen,
-                                    struct pipe_texture *_texture,
-                                    unsigned *stride,
-                                    unsigned *handle)
-{
-   struct trace_screen *tr_screen = trace_screen(_screen);
-   struct trace_texture *tr_texture = trace_texture(_texture);
-   struct trace_drm_api *tr_api = trace_drm_api(_api);
-   struct pipe_screen *screen = tr_screen->screen;
-   struct pipe_texture *texture = tr_texture->texture;
-   struct drm_api *api = tr_api->api;
-
-   /* TODO trace call */
-
-   return api->local_handle_from_texture(api, screen, texture, stride, handle);
+   return trace_screen_create(screen);
 }
 
 static void
@@ -155,11 +95,9 @@ trace_drm_create(struct drm_api *api)
    if (!tr_api)
       goto error;
 
+   tr_api->base.name = api->name;
    tr_api->base.driver_name = api->driver_name;
    tr_api->base.create_screen = trace_drm_create_screen;
-   tr_api->base.texture_from_shared_handle = trace_drm_texture_from_shared_handle;
-   tr_api->base.shared_handle_from_texture = trace_drm_shared_handle_from_texture;
-   tr_api->base.local_handle_from_texture = trace_drm_local_handle_from_texture;
    tr_api->base.destroy = trace_drm_destroy;
    tr_api->api = api;
 
diff --git a/src/gallium/drivers/trace/tr_dump_state.c b/src/gallium/drivers/trace/tr_dump_state.c
index 6da186a6555..f82dd01c697 100644
--- a/src/gallium/drivers/trace/tr_dump_state.c
+++ b/src/gallium/drivers/trace/tr_dump_state.c
@@ -112,7 +112,6 @@ void trace_dump_rasterizer_state(const struct pipe_rasterizer_state *state)
    trace_dump_member(uint, state, line_stipple_factor);
    trace_dump_member(uint, state, line_stipple_pattern);
    trace_dump_member(bool, state, line_last_pixel);
-   trace_dump_member(bool, state, bypass_vs_clip_and_viewport);
    trace_dump_member(bool, state, flatshade_first);
    trace_dump_member(bool, state, gl_rasterization_rules);
 
@@ -480,7 +479,6 @@ void trace_dump_vertex_element(const struct pipe_vertex_element *state)
    trace_dump_member(uint, state, src_offset);
 
    trace_dump_member(uint, state, vertex_buffer_index);
-   trace_dump_member(uint, state, nr_components);
 
    trace_dump_member(format, state, src_format);
 
diff --git a/src/gallium/drivers/trace/tr_public.h b/src/gallium/drivers/trace/tr_public.h
new file mode 100644
index 00000000000..62e217097d0
--- /dev/null
+++ b/src/gallium/drivers/trace/tr_public.h
@@ -0,0 +1,45 @@
+/**************************************************************************
+ *
+ * Copyright 2010 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 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.
+ *
+ **************************************************************************/
+
+#ifndef TR_PUBLIC_H
+#define TR_PUBLIC_H
+
+#ifdef __cplusplus
+extern "C" {
+#endif
+
+struct pipe_screen;
+struct pipe_context;
+
+struct pipe_screen *
+trace_screen_create(struct pipe_screen *screen);
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif /* TR_PUBLIC_H */
diff --git a/src/gallium/drivers/trace/tr_screen.c b/src/gallium/drivers/trace/tr_screen.c
index 388d83eb5c2..86ddb995405 100644
--- a/src/gallium/drivers/trace/tr_screen.c
+++ b/src/gallium/drivers/trace/tr_screen.c
@@ -35,6 +35,7 @@
 #include "tr_texture.h"
 #include "tr_context.h"
 #include "tr_screen.h"
+#include "tr_public.h"
 
 #include "util/u_inlines.h"
 #include "pipe/p_format.h"
@@ -236,39 +237,41 @@ trace_screen_texture_create(struct pipe_screen *_screen,
    return result;
 }
 
-
 static struct pipe_texture *
-trace_screen_texture_blanket(struct pipe_screen *_screen,
-                             const struct pipe_texture *templat,
-                             const unsigned *ppitch,
-                             struct pipe_buffer *_buffer)
+trace_screen_texture_from_handle(struct pipe_screen *_screen,
+                                 const struct pipe_texture *templ,
+                                 struct winsys_handle *handle)
 {
-   struct trace_screen *tr_scr = trace_screen(_screen);
-   struct trace_buffer *tr_buf = trace_buffer(_buffer);
-   struct pipe_screen *screen = tr_scr->screen;
-   struct pipe_buffer *buffer = tr_buf->buffer;
-   unsigned pitch = *ppitch;
+   struct trace_screen *tr_screen = trace_screen(_screen);
+   struct pipe_screen *screen = tr_screen->screen;
    struct pipe_texture *result;
 
-   trace_dump_call_begin("pipe_screen", "texture_blanket");
+   /* TODO trace call */
 
-   trace_dump_arg(ptr, screen);
-   trace_dump_arg(template, templat);
-   trace_dump_arg(uint, pitch);
-   trace_dump_arg(ptr, buffer);
+   result = screen->texture_from_handle(screen, templ, handle);
 
-   result = screen->texture_blanket(screen, templat, ppitch, buffer);
+   result = trace_texture_create(trace_screen(_screen), result);
 
-   trace_dump_ret(ptr, result);
+   return result;
+}
 
-   trace_dump_call_end();
+static boolean
+trace_screen_texture_get_handle(struct pipe_screen *_screen,
+                                struct pipe_texture *_texture,
+                                struct winsys_handle *handle)
+{
+   struct trace_screen *tr_screen = trace_screen(_screen);
+   struct trace_texture *tr_texture = trace_texture(_texture);
+   struct pipe_screen *screen = tr_screen->screen;
+   struct pipe_texture *texture = tr_texture->texture;
 
-   result = trace_texture_create(tr_scr, result);
+   /* TODO trace call */
 
-   return result;
+   return screen->texture_get_handle(screen, texture, handle);
 }
 
 
+
 static void
 trace_screen_texture_destroy(struct pipe_texture *_texture)
 {
@@ -484,45 +487,7 @@ trace_screen_transfer_unmap(struct pipe_screen *_screen,
  */
 
 
-static struct pipe_buffer *
-trace_screen_surface_buffer_create(struct pipe_screen *_screen,
-                                   unsigned width, unsigned height,
-                                   enum pipe_format format,
-                                   unsigned usage,
-                                   unsigned tex_usage,
-                                   unsigned *pstride)
-{
-   struct trace_screen *tr_scr = trace_screen(_screen);
-   struct pipe_screen *screen = tr_scr->screen;
-   unsigned stride;
-   struct pipe_buffer *result;
-
-   trace_dump_call_begin("pipe_screen", "surface_buffer_create");
-
-   trace_dump_arg(ptr, screen);
-   trace_dump_arg(uint, width);
-   trace_dump_arg(uint, height);
-   trace_dump_arg(format, format);
-   trace_dump_arg(uint, usage);
-   trace_dump_arg(uint, tex_usage);
 
-   result = screen->surface_buffer_create(screen,
-                                          width, height,
-                                          format,
-                                          usage,
-                                          tex_usage,
-                                          pstride);
-
-   stride = *pstride;
-
-   trace_dump_arg(uint, stride);
-
-   trace_dump_ret(ptr, result);
-
-   trace_dump_call_end();
-
-   return trace_buffer_create(tr_scr, result);
-}
 
 
 static struct pipe_buffer *
@@ -931,7 +896,8 @@ trace_screen_create(struct pipe_screen *screen)
    assert(screen->context_create);
    tr_scr->base.context_create = trace_screen_context_create;
    tr_scr->base.texture_create = trace_screen_texture_create;
-   tr_scr->base.texture_blanket = trace_screen_texture_blanket;
+   tr_scr->base.texture_from_handle = trace_screen_texture_from_handle;
+   tr_scr->base.texture_get_handle = trace_screen_texture_get_handle;
    tr_scr->base.texture_destroy = trace_screen_texture_destroy;
    tr_scr->base.get_tex_surface = trace_screen_get_tex_surface;
    tr_scr->base.tex_surface_destroy = trace_screen_tex_surface_destroy;
@@ -941,7 +907,6 @@ trace_screen_create(struct pipe_screen *screen)
    tr_scr->base.transfer_unmap = trace_screen_transfer_unmap;
    tr_scr->base.buffer_create = trace_screen_buffer_create;
    tr_scr->base.user_buffer_create = trace_screen_user_buffer_create;
-   tr_scr->base.surface_buffer_create = trace_screen_surface_buffer_create;
    if (screen->buffer_map)
       tr_scr->base.buffer_map = trace_screen_buffer_map;
    if (screen->buffer_map_range)
diff --git a/src/gallium/drivers/trace/tr_screen.h b/src/gallium/drivers/trace/tr_screen.h
index fe5a0fa1909..597e2fc2650 100644
--- a/src/gallium/drivers/trace/tr_screen.h
+++ b/src/gallium/drivers/trace/tr_screen.h
@@ -99,9 +99,6 @@ trace_enabled(void);
 struct trace_screen *
 trace_screen(struct pipe_screen *screen);
 
-struct pipe_screen *
-trace_screen_create(struct pipe_screen *screen);
-
 void
 trace_screen_user_buffer_update(struct pipe_screen *screen,
                                 struct pipe_buffer *buffer);