Merge branch 'gallium-tex-surfaces' into gallium-0.1

Conflicts:

	src/gallium/drivers/i915simple/i915_context.h

30 files changed, 1033 insertions, 415 deletions

diff --git a/src/gallium/auxiliary/draw/Makefile b/src/gallium/auxiliary/draw/Makefile
index 9a88ecc0708..f2e36a89e90 100644
--- a/src/gallium/auxiliary/draw/Makefile
+++ b/src/gallium/auxiliary/draw/Makefile
@@ -37,6 +37,7 @@ C_SOURCES = \
 	draw_vs_varient.c \
 	draw_vs_aos.c \
 	draw_vs_aos_io.c \
+	draw_vs_aos_machine.c \
 	draw_vs_exec.c \
 	draw_vs_llvm.c \
 	draw_vs_sse.c 
diff --git a/src/gallium/auxiliary/draw/SConscript b/src/gallium/auxiliary/draw/SConscript
index 3b5d5ed4921..544a04918b6 100644
--- a/src/gallium/auxiliary/draw/SConscript
+++ b/src/gallium/auxiliary/draw/SConscript
@@ -15,7 +15,7 @@ draw = env.ConvenienceLibrary(
 		'draw_pipe_stipple.c',
 		'draw_pipe_twoside.c',
 		'draw_pipe_unfilled.c',
-        	'draw_pipe_util.c',
+		'draw_pipe_util.c',
 		'draw_pipe_validate.c',
 		'draw_pipe_vbuf.c',
 		'draw_pipe_wide_line.c',
@@ -25,15 +25,21 @@ draw = env.ConvenienceLibrary(
 		'draw_pt_emit.c',
 		'draw_pt_fetch.c',
 		'draw_pt_fetch_emit.c',
+		'draw_pt_fetch_shade_emit.c',
 		'draw_pt_fetch_shade_pipeline.c',
 		'draw_pt_post_vs.c',
+		'draw_pt_util.c',
 		'draw_pt_varray.c',
 		'draw_pt_vcache.c',
 		'draw_vertex.c',
 		'draw_vs.c',
+		'draw_vs_aos.c',
+		'draw_vs_aos_io.c',
+		'draw_vs_aos_machine.c',
 		'draw_vs_exec.c',
 		'draw_vs_llvm.c',
 		'draw_vs_sse.c',
+		'draw_vs_varient.c'
 	])
 
 auxiliaries.insert(0, draw)
diff --git a/src/gallium/auxiliary/draw/draw_context.c b/src/gallium/auxiliary/draw/draw_context.c
index 22420749656..2f263cf06a9 100644
--- a/src/gallium/auxiliary/draw/draw_context.c
+++ b/src/gallium/auxiliary/draw/draw_context.c
@@ -174,6 +174,8 @@ void draw_set_viewport_state( struct draw_context *draw,
                               viewport->translate[1] == 0.0f &&
                               viewport->translate[2] == 0.0f &&
                               viewport->translate[3] == 0.0f);
+
+   draw_vs_set_viewport( draw, viewport );
 }
 
 
@@ -215,9 +217,11 @@ draw_set_mapped_vertex_buffer(struct draw_context *draw,
 
 void
 draw_set_mapped_constant_buffer(struct draw_context *draw,
-                                const void *buffer)
+                                const void *buffer, 
+                                unsigned size )
 {
    draw->pt.user.constants = buffer;
+   draw_vs_set_constants( draw, (const float (*)[4])buffer, size );
 }
 
 
@@ -345,14 +349,30 @@ void draw_set_edgeflags( struct draw_context *draw,
  * \param elements  the element buffer ptr
  */
 void
-draw_set_mapped_element_buffer( struct draw_context *draw,
-                                unsigned eltSize, void *elements )
+draw_set_mapped_element_buffer_range( struct draw_context *draw,
+                                      unsigned eltSize,
+                                      unsigned min_index,
+                                      unsigned max_index,
+                                      void *elements )
 {
    draw->pt.user.elts = elements;
    draw->pt.user.eltSize = eltSize;
+   draw->pt.user.min_index = min_index;
+   draw->pt.user.max_index = max_index;
 }
 
 
+void
+draw_set_mapped_element_buffer( struct draw_context *draw,
+                                unsigned eltSize,
+                                void *elements )
+{
+   draw->pt.user.elts = elements;
+   draw->pt.user.eltSize = eltSize;
+   draw->pt.user.min_index = 0;
+   draw->pt.user.max_index = 0xffffffff;
+}
+
  
 /* Revamp me please:
  */
diff --git a/src/gallium/auxiliary/draw/draw_context.h b/src/gallium/auxiliary/draw/draw_context.h
index c5c3d3b09e0..b8f2bfa3327 100644
--- a/src/gallium/auxiliary/draw/draw_context.h
+++ b/src/gallium/auxiliary/draw/draw_context.h
@@ -118,14 +118,23 @@ void draw_set_vertex_elements(struct draw_context *draw,
 			      unsigned count,
                               const struct pipe_vertex_element *elements);
 
+void
+draw_set_mapped_element_buffer_range( struct draw_context *draw,
+                                      unsigned eltSize,
+                                      unsigned min_index,
+                                      unsigned max_index,
+                                      void *elements );
+
 void draw_set_mapped_element_buffer( struct draw_context *draw,
-                                     unsigned eltSize, void *elements );
+                                     unsigned eltSize, 
+                                     void *elements );
 
 void draw_set_mapped_vertex_buffer(struct draw_context *draw,
                                    unsigned attr, const void *buffer);
 
 void draw_set_mapped_constant_buffer(struct draw_context *draw,
-                                     const void *buffer);
+                                     const void *buffer,
+                                     unsigned size );
 
 void draw_set_edgeflags( struct draw_context *draw,
                          const unsigned *edgeflag );
diff --git a/src/gallium/auxiliary/draw/draw_pipe_stipple.c b/src/gallium/auxiliary/draw/draw_pipe_stipple.c
index 4673d5dcbae..9522b79582e 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_stipple.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_stipple.c
@@ -175,6 +175,22 @@ reset_stipple_counter(struct draw_stage *stage)
    stage->next->reset_stipple_counter( stage->next );
 }
 
+static void
+stipple_reset_point(struct draw_stage *stage, struct prim_header *header)
+{
+   struct stipple_stage *stipple = stipple_stage(stage);
+   stipple->counter = 0;
+   stage->next->point(stage->next, header);
+}
+
+static void
+stipple_reset_tri(struct draw_stage *stage, struct prim_header *header)
+{
+   struct stipple_stage *stipple = stipple_stage(stage);
+   stipple->counter = 0;
+   stage->next->tri(stage->next, header);
+}
+
 
 static void
 stipple_first_line(struct draw_stage *stage, 
@@ -220,9 +236,9 @@ struct draw_stage *draw_stipple_stage( struct draw_context *draw )
 
    stipple->stage.draw = draw;
    stipple->stage.next = NULL;
-   stipple->stage.point = draw_pipe_passthrough_point;
+   stipple->stage.point = stipple_reset_point;
    stipple->stage.line = stipple_first_line;
-   stipple->stage.tri = draw_pipe_passthrough_tri;
+   stipple->stage.tri = stipple_reset_tri;
    stipple->stage.reset_stipple_counter = reset_stipple_counter;
    stipple->stage.flush = stipple_flush;
    stipple->stage.destroy = stipple_destroy;
diff --git a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
index 67b9a9503d2..d514e28b77f 100644
--- a/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
+++ b/src/gallium/auxiliary/draw/draw_pipe_vbuf.c
@@ -145,7 +145,7 @@ emit_vertex( struct vbuf_stage *vbuf,
       vertex->vertex_id = vbuf->nr_vertices++;
    }
 
-   return vertex->vertex_id;
+   return (ushort)vertex->vertex_id;
 }
 
 
diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index c095bf3d7b9..4f8cceee1e8 100644
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -147,6 +147,8 @@ struct draw_context
          const void *elts;
          /** bytes per index (0, 1, 2 or 4) */
          unsigned eltSize;
+         unsigned min_index;
+         unsigned max_index;
          
          /** vertex arrays */
          const void *vbuffer[PIPE_MAX_ATTRIBS];
@@ -155,7 +157,8 @@ struct draw_context
          const void *constants;
       } user;
 
-      boolean test_fse;
+      boolean test_fse;         /* enable FSE even though its not correct (eg for softpipe) */
+      boolean no_fse;           /* disable FSE even when it is correct */
    } pt;
 
    struct {
@@ -183,6 +186,16 @@ struct draw_context
        */
       struct gallivm_cpu_engine *engine;   
 
+      /* Here's another one:
+       */
+      struct aos_machine *aos_machine; 
+
+
+      const float (*aligned_constants)[4];
+
+      const float (*aligned_constant_storage)[4];
+      unsigned const_storage_size;
+
 
       struct translate *fetch;
       struct translate_cache *fetch_cache;
@@ -215,6 +228,12 @@ struct draw_context
 boolean draw_vs_init( struct draw_context *draw );
 void draw_vs_destroy( struct draw_context *draw );
 
+void draw_vs_set_viewport( struct draw_context *, 
+                           const struct pipe_viewport_state * );
+
+void draw_vs_set_constants( struct draw_context *,
+                            const float (*constants)[4],
+                            unsigned size );
 
 
 
diff --git a/src/gallium/auxiliary/draw/draw_pt.c b/src/gallium/auxiliary/draw/draw_pt.c
index 75f44d503eb..723077159b5 100644
--- a/src/gallium/auxiliary/draw/draw_pt.c
+++ b/src/gallium/auxiliary/draw/draw_pt.c
@@ -75,7 +75,7 @@ draw_pt_arrays(struct draw_context *draw,
 
    if (opt == 0) 
       middle = draw->pt.middle.fetch_emit;
-   else if (opt == PT_SHADE && draw->pt.test_fse)
+   else if (opt == PT_SHADE && !draw->pt.no_fse)
       middle = draw->pt.middle.fetch_shade_emit;
    else
       middle = draw->pt.middle.general;
@@ -105,6 +105,7 @@ draw_pt_arrays(struct draw_context *draw,
 boolean draw_pt_init( struct draw_context *draw )
 {
    draw->pt.test_fse = GETENV("DRAW_FSE") != NULL;
+   draw->pt.no_fse = GETENV("DRAW_NO_FSE") != NULL;
 
    draw->pt.front.vcache = draw_pt_vcache( draw );
    if (!draw->pt.front.vcache)
@@ -118,12 +119,9 @@ boolean draw_pt_init( struct draw_context *draw )
    if (!draw->pt.middle.fetch_emit)
       return FALSE;
 
-   if (draw->pt.test_fse) {
-      draw->pt.middle.fetch_shade_emit = draw_pt_middle_fse( draw );
-      if (!draw->pt.middle.fetch_shade_emit)
-         return FALSE;
-   }
-
+   draw->pt.middle.fetch_shade_emit = draw_pt_middle_fse( draw );
+   if (!draw->pt.middle.fetch_shade_emit)
+      return FALSE;
 
    draw->pt.middle.general = draw_pt_fetch_pipeline_or_emit( draw );
    if (!draw->pt.middle.general)
diff --git a/src/gallium/auxiliary/draw/draw_pt.h b/src/gallium/auxiliary/draw/draw_pt.h
index e03816ebbc7..6b8ba1d171b 100644
--- a/src/gallium/auxiliary/draw/draw_pt.h
+++ b/src/gallium/auxiliary/draw/draw_pt.h
@@ -96,6 +96,15 @@ struct draw_pt_middle_end {
                       unsigned start,
                       unsigned count);
 
+   /* Transform all vertices in a linear range and then draw them with
+    * the supplied element list.
+    */
+   void (*run_linear_elts)( struct draw_pt_middle_end *,
+                            unsigned fetch_start,
+                            unsigned fetch_count,
+                            const ushort *draw_elts,
+                            unsigned draw_count );
+
    void (*finish)( struct draw_pt_middle_end * );
    void (*destroy)( struct draw_pt_middle_end * );
 };
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
index a1d041a74f5..09bdc5fb5e1 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_emit.c
@@ -311,6 +311,53 @@ static void fetch_emit_run_linear( struct draw_pt_middle_end *middle,
 }
 
 
+static void fetch_emit_run_linear_elts( struct draw_pt_middle_end *middle,
+                                        unsigned start,
+                                        unsigned count,
+                                        const ushort *draw_elts,
+                                        unsigned draw_count )
+{
+   struct fetch_emit_middle_end *feme = (struct fetch_emit_middle_end *)middle;
+   struct draw_context *draw = feme->draw;
+   void *hw_verts;
+
+   /* XXX: need to flush to get prim_vbuf.c to release its allocation??
+    */
+   draw_do_flush( draw, DRAW_FLUSH_BACKEND );
+
+   hw_verts = draw->render->allocate_vertices( draw->render,
+                                               (ushort)feme->translate->key.output_stride,
+                                               (ushort)count );
+   if (!hw_verts) {
+      assert(0);
+      return;
+   }
+
+   /* Single routine to fetch vertices and emit HW verts.
+    */
+   feme->translate->run( feme->translate,
+                         start,
+                         count,
+                         hw_verts );
+
+   /* XXX: Draw arrays path to avoid re-emitting index list again and
+    * again.
+    */
+   draw->render->draw( draw->render, 
+                       draw_elts, 
+                       draw_count );
+
+   /* Done -- that was easy, wasn't it:
+    */
+   draw->render->release_vertices( draw->render,
+                                   hw_verts,
+                                   feme->translate->key.output_stride,
+                                   count );
+
+}
+
+
+
 
 static void fetch_emit_finish( struct draw_pt_middle_end *middle )
 {
@@ -343,6 +390,7 @@ struct draw_pt_middle_end *draw_pt_fetch_emit( struct draw_context *draw )
    fetch_emit->base.prepare    = fetch_emit_prepare;
    fetch_emit->base.run        = fetch_emit_run;
    fetch_emit->base.run_linear = fetch_emit_run_linear;
+   fetch_emit->base.run_linear_elts = fetch_emit_run_linear_elts;
    fetch_emit->base.finish     = fetch_emit_finish;
    fetch_emit->base.destroy    = fetch_emit_destroy;
 
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
index 729c7db9999..efa6dddbda8 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_emit.c
@@ -189,12 +189,6 @@ static void fse_prepare( struct draw_pt_middle_end *middle,
                               draw->pt.vertex_buffer[buf].pitch );
    }
 
-   fse->active->set_constants( fse->active,
-                               (const float (*)[4])draw->pt.user.constants );
-
-   fse->active->set_viewport( fse->active,
-                              &draw->viewport );
-
    //return TRUE;
 }
 
@@ -316,6 +310,54 @@ fse_run(struct draw_pt_middle_end *middle,
 }
 
 
+
+static void fse_run_linear_elts( struct draw_pt_middle_end *middle, 
+                                 unsigned start, 
+                                 unsigned count,
+                                 const ushort *draw_elts,
+                                 unsigned draw_count )
+{
+   struct fetch_shade_emit *fse = (struct fetch_shade_emit *)middle;
+   struct draw_context *draw = fse->draw;
+   unsigned alloc_count = align(count, 4);
+   char *hw_verts;
+
+   /* XXX: need to flush to get prim_vbuf.c to release its allocation??
+    */
+   draw_do_flush( draw, DRAW_FLUSH_BACKEND );
+
+   hw_verts = draw->render->allocate_vertices( draw->render,
+                                               (ushort)fse->key.output_stride,
+                                               (ushort)alloc_count );
+
+   if (!hw_verts) {
+      assert(0);
+      return;
+   }
+
+   /* Single routine to fetch vertices, run shader and emit HW verts.
+    * Clipping is done elsewhere -- either by the API or on hardware,
+    * or for some other reason not required...
+    */
+   fse->active->run_linear( fse->active, 
+                            start, count,
+                            hw_verts );
+
+
+   draw->render->draw( draw->render, 
+                       draw_elts, 
+                       draw_count );
+   
+
+
+   draw->render->release_vertices( draw->render, 
+				   hw_verts, 
+				   fse->key.output_stride, 
+				   count );
+}
+
+
+
 static void fse_finish( struct draw_pt_middle_end *middle )
 {
 }
@@ -336,6 +378,7 @@ struct draw_pt_middle_end *draw_pt_middle_fse( struct draw_context *draw )
    fse->base.prepare = fse_prepare;
    fse->base.run = fse_run;
    fse->base.run_linear = fse_run_linear;
+   fse->base.run_linear_elts = fse_run_linear_elts;
    fse->base.finish = fse_finish;
    fse->base.destroy = fse_destroy;
    fse->draw = draw;
diff --git a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c
index 06718779a5b..eb6988ff03b 100644
--- a/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c
+++ b/src/gallium/auxiliary/draw/draw_pt_fetch_shade_pipeline.c
@@ -81,9 +81,9 @@ static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle,
     * but gl vs dx9 clip spaces.
     */
    draw_pt_post_vs_prepare( fpme->post_vs,
-			    draw->bypass_clipping,
-			    draw->identity_viewport,
-			    draw->rasterizer->gl_rasterization_rules );
+			    (boolean)draw->bypass_clipping,
+			    (boolean)draw->identity_viewport,
+			    (boolean)draw->rasterizer->gl_rasterization_rules );
 			    
 
    if (!(opt & PT_PIPELINE)) 
@@ -98,7 +98,6 @@ static void fetch_pipeline_prepare( struct draw_pt_middle_end *middle,
 
 
 
-
 static void fetch_pipeline_run( struct draw_pt_middle_end *middle,
                                 const unsigned *fetch_elts,
                                 unsigned fetch_count,
@@ -251,6 +250,84 @@ static void fetch_pipeline_linear_run( struct draw_pt_middle_end *middle,
 
 
 
+static void fetch_pipeline_linear_run_elts( struct draw_pt_middle_end *middle,
+                                            unsigned start,
+                                            unsigned count,
+                                            const ushort *draw_elts,
+                                            unsigned draw_count )
+{
+   struct fetch_pipeline_middle_end *fpme = (struct fetch_pipeline_middle_end *)middle;
+   struct draw_context *draw = fpme->draw;
+   struct draw_vertex_shader *shader = draw->vs.vertex_shader;
+   unsigned opt = fpme->opt;
+   unsigned alloc_count = align_int( count, 4 );
+
+   struct vertex_header *pipeline_verts =
+      (struct vertex_header *)MALLOC(fpme->vertex_size * alloc_count);
+
+   if (!pipeline_verts) {
+      /* Not much we can do here - just skip the rendering.
+       */
+      assert(0);
+      return;
+   }
+
+   /* Fetch into our vertex buffer
+    */
+   draw_pt_fetch_run_linear( fpme->fetch,
+                             start,
+                             count,
+                             (char *)pipeline_verts );
+
+   /* Run the shader, note that this overwrites the data[] parts of
+    * the pipeline verts.  If there is no shader, ie a bypass shader,
+    * then the inputs == outputs, and are already in the correct
+    * place.
+    */
+   if (opt & PT_SHADE)
+   {
+      shader->run_linear(shader,
+			 (const float (*)[4])pipeline_verts->data,
+			 (      float (*)[4])pipeline_verts->data,
+			 (const float (*)[4])draw->pt.user.constants,
+			 count,
+			 fpme->vertex_size,
+			 fpme->vertex_size);
+   }
+
+   if (draw_pt_post_vs_run( fpme->post_vs,
+			    pipeline_verts,
+			    count,
+			    fpme->vertex_size ))
+   {
+      opt |= PT_PIPELINE;
+   }
+
+   /* Do we need to run the pipeline?
+    */
+   if (opt & PT_PIPELINE) {
+      draw_pipeline_run( fpme->draw,
+                         fpme->prim,
+                         pipeline_verts,
+                         count,
+                         fpme->vertex_size,
+                         draw_elts,
+                         draw_count );
+   }
+   else {
+      draw_pt_emit( fpme->emit,
+		    (const float (*)[4])pipeline_verts->data,
+		    count,
+		    fpme->vertex_size,
+		    draw_elts,
+		    draw_count );
+   }
+
+   FREE(pipeline_verts);
+}
+
+
+
 static void fetch_pipeline_finish( struct draw_pt_middle_end *middle )
 {
    /* nothing to do */
@@ -282,6 +359,7 @@ struct draw_pt_middle_end *draw_pt_fetch_pipeline_or_emit( struct draw_context *
    fpme->base.prepare        = fetch_pipeline_prepare;
    fpme->base.run            = fetch_pipeline_run;
    fpme->base.run_linear     = fetch_pipeline_linear_run;
+   fpme->base.run_linear_elts = fetch_pipeline_linear_run_elts;
    fpme->base.finish         = fetch_pipeline_finish;
    fpme->base.destroy        = fetch_pipeline_destroy;
 
diff --git a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h
index 6979f6b544f..7c722457c3c 100644
--- a/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h
+++ b/src/gallium/auxiliary/draw/draw_pt_varray_tmp.h
@@ -10,7 +10,8 @@ static void FUNC(struct draw_pt_front_end *frontend,
 
    boolean flatfirst = (draw->rasterizer->flatshade &&
                         draw->rasterizer->flatshade_first);
-   unsigned i, j, flags;
+   unsigned i, j;
+   ushort flags;
    unsigned first, incr;
 
    varray->fetch_start = start;
@@ -200,9 +201,9 @@ static void FUNC(struct draw_pt_front_end *frontend,
       /* These bitflags look a little odd because we submit the
        * vertices as (1,2,0) to satisfy flatshade requirements.
        */
-      const unsigned edge_first  = DRAW_PIPE_EDGE_FLAG_2;
-      const unsigned edge_middle = DRAW_PIPE_EDGE_FLAG_0;
-      const unsigned edge_last   = DRAW_PIPE_EDGE_FLAG_1;
+      const ushort edge_first  = DRAW_PIPE_EDGE_FLAG_2;
+      const ushort edge_middle = DRAW_PIPE_EDGE_FLAG_0;
+      const ushort edge_last   = DRAW_PIPE_EDGE_FLAG_1;
 
       flags = DRAW_PIPE_RESET_STIPPLE | edge_first | edge_middle;
       for (j = 0; j + first <= count; j += i) {
diff --git a/src/gallium/auxiliary/draw/draw_pt_vcache.c b/src/gallium/auxiliary/draw/draw_pt_vcache.c
index 96e02fbf3a9..9ffc99abf97 100644
--- a/src/gallium/auxiliary/draw/draw_pt_vcache.c
+++ b/src/gallium/auxiliary/draw/draw_pt_vcache.c
@@ -36,8 +36,8 @@
 #include "draw/draw_pt.h"
 
 
-#define CACHE_MAX 32
-#define FETCH_MAX 128
+#define CACHE_MAX 1024
+#define FETCH_MAX 4096
 #define DRAW_MAX (16*1024)
 
 struct vcache_frontend {
@@ -201,7 +201,128 @@ static void vcache_ef_quad( struct vcache_frontend *vcache,
 #define FUNC vcache_run
 #include "draw_pt_vcache_tmp.h"
 
+static void translate_uint_elts( const unsigned *src,
+                                 unsigned count,
+                                 int delta,
+                                 ushort *dest )
+{
+   unsigned i;
+
+   for (i = 0; i < count; i++) 
+      dest[i] = (ushort)(src[i] + delta);
+}
+
+static void translate_ushort_elts( const ushort *src,
+                                   unsigned count,
+                                   int delta,
+                                   ushort *dest )
+{
+   unsigned i;
+
+   for (i = 0; i < count; i++) 
+      dest[i] = (ushort)(src[i] + delta);
+}
 
+static void translate_ubyte_elts( const ubyte *src,
+                                  unsigned count,
+                                  int delta,
+                                  ushort *dest )
+{
+   unsigned i;
+
+   for (i = 0; i < count; i++) 
+      dest[i] = (ushort)(src[i] + delta);
+}
+
+#if 0
+static enum pipe_format format_from_get_elt( pt_elt_func get_elt )
+{
+   switch (draw->pt.user.eltSize) {
+   case 1: return PIPE_FORMAT_R8_UNORM;
+   case 2: return PIPE_FORMAT_R16_UNORM;
+   case 4: return PIPE_FORMAT_R32_UNORM;
+   default: return PIPE_FORMAT_NONE;
+   }
+}
+#endif
+
+static void vcache_check_run( struct draw_pt_front_end *frontend, 
+                              pt_elt_func get_elt,
+                              const void *elts,
+                              unsigned draw_count )
+{
+   struct vcache_frontend *vcache = (struct vcache_frontend *)frontend; 
+   struct draw_context *draw = vcache->draw;
+   unsigned min_index = draw->pt.user.min_index;
+   unsigned max_index = draw->pt.user.max_index;
+   unsigned index_size = draw->pt.user.eltSize;
+   unsigned fetch_count = max_index + 1 - min_index;
+   const ushort *transformed_elts;
+   ushort *storage = NULL;
+
+
+   if (0) debug_printf("fetch_count %d draw_count %d\n", fetch_count, draw_count);
+      
+   if (max_index == 0xffffffff ||
+       fetch_count >= FETCH_MAX ||
+       fetch_count > draw_count) {
+      if (0) debug_printf("fail\n");
+      goto fail;
+   }
+      
+
+   if (min_index == 0 &&
+       index_size == 2) 
+   {
+      transformed_elts = (const ushort *)elts;
+   }
+   else 
+   {
+      storage = MALLOC( draw_count * sizeof(ushort) );
+      if (!storage)
+         goto fail;
+      
+      switch(index_size) {
+      case 1:
+         translate_ubyte_elts( (const ubyte *)elts,
+                               draw_count,
+                               0 - (int)min_index,
+                               storage );
+         break;
+
+      case 2:
+         translate_ushort_elts( (const ushort *)elts,
+                                draw_count,
+                                0 - (int)min_index,
+                                storage );
+         break;
+
+      case 4:
+         translate_uint_elts( (const uint *)elts,
+                              draw_count,
+                              0 - (int)min_index,
+                              storage );
+         break;
+
+      default:
+         assert(0);
+         return;
+      }
+      transformed_elts = storage;
+   }
+
+   vcache->middle->run_linear_elts( vcache->middle,
+                                    min_index, /* start */
+                                    fetch_count,
+                                    transformed_elts,
+                                    draw_count );
+
+   FREE(storage);
+   return;
+
+ fail:
+   vcache_run( frontend, get_elt, elts, draw_count );
+}
 
 
 
@@ -219,7 +340,7 @@ static void vcache_prepare( struct draw_pt_front_end *frontend,
    }
    else 
    {
-      vcache->base.run = vcache_run;
+      vcache->base.run = vcache_check_run;
    }
 
    vcache->input_prim = prim;
diff --git a/src/gallium/auxiliary/draw/draw_pt_vcache_tmp.h b/src/gallium/auxiliary/draw/draw_pt_vcache_tmp.h
index cf9f394aa3b..5c99a47e49a 100644
--- a/src/gallium/auxiliary/draw/draw_pt_vcache_tmp.h
+++ b/src/gallium/auxiliary/draw/draw_pt_vcache_tmp.h
@@ -10,7 +10,8 @@ static void FUNC( struct draw_pt_front_end *frontend,
 
    boolean flatfirst = (draw->rasterizer->flatshade && 
                         draw->rasterizer->flatshade_first);
-   unsigned i, flags;
+   unsigned i;
+   ushort flags;
 
 
    switch (vcache->input_prim) {
@@ -138,9 +139,9 @@ static void FUNC( struct draw_pt_front_end *frontend,
          /* These bitflags look a little odd because we submit the
           * vertices as (1,2,0) to satisfy flatshade requirements.  
           */
-         const unsigned edge_first  = DRAW_PIPE_EDGE_FLAG_2;
-         const unsigned edge_middle = DRAW_PIPE_EDGE_FLAG_0;
-         const unsigned edge_last   = DRAW_PIPE_EDGE_FLAG_1;
+         const ushort edge_first  = DRAW_PIPE_EDGE_FLAG_2;
+         const ushort edge_middle = DRAW_PIPE_EDGE_FLAG_0;
+         const ushort edge_last   = DRAW_PIPE_EDGE_FLAG_1;
 
          flags = DRAW_PIPE_RESET_STIPPLE | edge_first | edge_middle;
 
diff --git a/src/gallium/auxiliary/draw/draw_vs.c b/src/gallium/auxiliary/draw/draw_vs.c
index 9b899d404e1..ce35112fc10 100644
--- a/src/gallium/auxiliary/draw/draw_vs.c
+++ b/src/gallium/auxiliary/draw/draw_vs.c
@@ -41,6 +41,36 @@
 
 
 
+
+void draw_vs_set_constants( struct draw_context *draw,
+                            const float (*constants)[4],
+                            unsigned size )
+{
+   if (((unsigned)constants) & 0xf) {
+      if (size > draw->vs.const_storage_size) {
+         if (draw->vs.aligned_constant_storage)
+            align_free(draw->vs.aligned_constant_storage);
+         draw->vs.aligned_constant_storage = align_malloc( size, 16 );
+      }
+      memcpy( draw->vs.aligned_constant_storage,
+              constants, 
+              size );
+      constants = draw->vs.aligned_constant_storage;
+   }
+      
+   draw->vs.aligned_constants = constants;
+   draw_vs_aos_machine_constants( draw->vs.aos_machine, constants );
+}
+
+
+void draw_vs_set_viewport( struct draw_context *draw,
+                           const struct pipe_viewport_state *viewport )
+{
+   draw_vs_aos_machine_viewport( draw->vs.aos_machine, viewport );
+}
+
+
+
 struct draw_vertex_shader *
 draw_create_vertex_shader(struct draw_context *draw,
                           const struct pipe_shader_state *shader)
@@ -83,6 +113,13 @@ void
 draw_delete_vertex_shader(struct draw_context *draw,
                           struct draw_vertex_shader *dvs)
 {
+   unsigned i;
+
+   for (i = 0; i < dvs->nr_varients; i++) 
+      dvs->varient[i]->destroy( dvs->varient[i] );
+
+   dvs->nr_varients = 0;
+
    dvs->delete( dvs );
 }
 
@@ -110,6 +147,10 @@ draw_vs_init( struct draw_context *draw )
    draw->vs.fetch_cache = translate_cache_create();
    if (!draw->vs.fetch_cache) 
       return FALSE;
+
+   draw->vs.aos_machine = draw_vs_aos_machine();
+   if (!draw->vs.aos_machine)
+      return FALSE;
       
    return TRUE;
 }
@@ -129,6 +170,12 @@ draw_vs_destroy( struct draw_context *draw )
    if (draw->vs.emit_cache)
       translate_cache_destroy(draw->vs.emit_cache);
 
+   if (draw->vs.aos_machine)
+      draw_vs_aos_machine_destroy(draw->vs.aos_machine);
+
+   if (draw->vs.aligned_constant_storage)
+      align_free(draw->vs.aligned_constant_storage);
+
    tgsi_exec_machine_free_data(&draw->vs.machine);
 
 }
@@ -153,10 +200,17 @@ draw_vs_lookup_varient( struct draw_vertex_shader *vs,
    if (varient == NULL)
       return NULL;
 
-   /* Add it to our list: 
+   /* Add it to our list, could be smarter: 
     */
-   assert(vs->nr_varients < Elements(vs->varient));
-   vs->varient[vs->nr_varients++] = varient;
+   if (vs->nr_varients < Elements(vs->varient)) {
+      vs->varient[vs->nr_varients++] = varient;
+   }
+   else {
+      vs->last_varient++;
+      vs->last_varient %= Elements(vs->varient);
+      vs->varient[vs->last_varient]->destroy(vs->varient[vs->last_varient]);
+      vs->varient[vs->last_varient] = varient;
+   }
 
    /* Done 
     */
diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h
index 01171bc23d5..08c6de8ba86 100644
--- a/src/gallium/auxiliary/draw/draw_vs.h
+++ b/src/gallium/auxiliary/draw/draw_vs.h
@@ -70,16 +70,6 @@ struct draw_vs_varient_key {
 
 struct draw_vs_varient;
 
-typedef void (PIPE_CDECL *vsv_run_elts_func)( struct draw_vs_varient *,
-                                              const unsigned *elts,
-                                              unsigned count,
-                                              void *output_buffer);
-
-typedef void (PIPE_CDECL *vsv_run_linear_func)( struct draw_vs_varient *,
-                                                unsigned start,
-                                                unsigned count,
-                                                void *output_buffer);
-
 
 struct draw_vs_varient {
    struct draw_vs_varient_key key;
@@ -91,12 +81,6 @@ struct draw_vs_varient {
                       const void *ptr,
                       unsigned stride );
 
-   void (*set_constants)( struct draw_vs_varient *,
-                          const float (*constants)[4] );
-
-   void (*set_viewport)( struct draw_vs_varient *,
-                         const struct pipe_viewport_state * );
-
    void (PIPE_CDECL *run_linear)( struct draw_vs_varient *shader,
                                   unsigned start,
                                   unsigned count,
@@ -123,10 +107,15 @@ struct draw_vertex_shader {
 
    struct tgsi_shader_info info;
 
+   /* Extracted from shader:
+    */
+   const float (*immediates)[4];
+
    /* 
     */
    struct draw_vs_varient *varient[16];
    unsigned nr_varients;
+   unsigned last_varient;
    struct draw_vs_varient *(*create_varient)( struct draw_vertex_shader *shader,
                                               const struct draw_vs_varient_key *key );
 
@@ -213,7 +202,14 @@ static INLINE int draw_vs_varient_key_compare( const struct draw_vs_varient_key
 }
 
 
+struct aos_machine *draw_vs_aos_machine( void );
+void draw_vs_aos_machine_destroy( struct aos_machine *machine );
+
+void draw_vs_aos_machine_constants( struct aos_machine *machine,
+                                    const float (*constants)[4] );
 
+void draw_vs_aos_machine_viewport( struct aos_machine *machine,
+                                   const struct pipe_viewport_state *viewport );
 
 
 #define MAX_TGSI_VERTICES 4
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.c b/src/gallium/auxiliary/draw/draw_vs_aos.c
index d3770b2c53b..725f36b5025 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.c
@@ -66,6 +66,37 @@ static INLINE boolean eq( struct x86_reg a,
 	   a.disp == b.disp);
 }
       
+struct x86_reg aos_get_x86( struct aos_compilation *cp,
+                            unsigned value )
+{
+   if (cp->ebp != value) {
+      unsigned offset;
+
+      switch (value) {
+      case X86_IMMEDIATES:
+         offset = Offset(struct aos_machine, immediates);
+         break;
+      case X86_CONSTANTS:
+         offset = Offset(struct aos_machine, constants);
+         break;
+      case X86_ATTRIBS:
+         offset = Offset(struct aos_machine, attrib);
+         break;
+      default:
+         assert(0);
+         offset = 0;
+      }
+
+      x86_mov(cp->func, cp->temp_EBP, 
+              x86_make_disp(cp->machine_EDX, offset));
+      /* x86_deref(x86_make_disp(cp->machine_EDX, offset))); */
+
+      cp->ebp = value;
+   }
+
+   return cp->temp_EBP;
+}
+
 
 static struct x86_reg get_reg_ptr(struct aos_compilation *cp,
                                   unsigned file,
@@ -83,15 +114,15 @@ static struct x86_reg get_reg_ptr(struct aos_compilation *cp,
    case TGSI_FILE_TEMPORARY:
       return x86_make_disp(ptr, Offset(struct aos_machine, temp[idx]));
 
-   case TGSI_FILE_IMMEDIATE:
-      return x86_make_disp(ptr, Offset(struct aos_machine, immediate[idx]));
-
-   case TGSI_FILE_CONSTANT:       
-      return x86_make_disp(ptr, Offset(struct aos_machine, constant[idx]));
-
    case AOS_FILE_INTERNAL:
       return x86_make_disp(ptr, Offset(struct aos_machine, internal[idx]));
 
+   case TGSI_FILE_IMMEDIATE: 
+      return x86_make_disp(aos_get_x86(cp, X86_IMMEDIATES), idx * 4 * sizeof(float));
+
+   case TGSI_FILE_CONSTANT: 
+      return x86_make_disp(aos_get_x86(cp, X86_CONSTANTS), idx * 4 * sizeof(float));
+
    default:
       ERROR(cp, "unknown reg file");
       return x86_make_reg(0,0);
@@ -118,70 +149,7 @@ static struct x86_reg get_reg_ptr(struct aos_compilation *cp,
 #define X87_CW_ROUND_MASK             (3<<10)
 #define X87_CW_INFINITY               (1<<12)
 
-static void do_populate_lut( struct shine_tab *tab,
-                             float unclamped_exponent )
-{
-   const float epsilon = 1.0F / 256.0F;    
-   float exponent = CLAMP(unclamped_exponent, -(128.0F - epsilon), (128.0F - epsilon));
-   unsigned i;
 
-   tab->exponent = unclamped_exponent; /* for later comparison */
-   
-   tab->values[0] = 0;
-   if (exponent == 0) {
-      for (i = 1; i < 258; i++) {
-         tab->values[i] = 1.0;
-      }      
-   }
-   else {
-      for (i = 1; i < 258; i++) {
-         tab->values[i] = powf((float)i * epsilon, exponent);
-      }
-   }
-}
-
-static void init_internals( struct aos_machine *machine )
-{
-   unsigned i;
-   float inv = 1.0f/255.0f;
-   float f255 = 255.0f;
-
-   ASSIGN_4V(machine->internal[IMM_SWZ],       1.0f,  -1.0f,  0.0f, 1.0f);
-   *(unsigned *)&machine->internal[IMM_SWZ][3] = 0xffffffff;
-
-   ASSIGN_4V(machine->internal[IMM_ONES],      1.0f,  1.0f,  1.0f,  1.0f);
-   ASSIGN_4V(machine->internal[IMM_NEGS],     -1.0f, -1.0f, -1.0f, -1.0f);
-   ASSIGN_4V(machine->internal[IMM_IDENTITY],  0.0f,  0.0f,  0.0f,  1.0f);
-   ASSIGN_4V(machine->internal[IMM_INV_255],   inv,   inv,   inv,   inv);
-   ASSIGN_4V(machine->internal[IMM_255],       f255,  f255,  f255,  f255);
-   ASSIGN_4V(machine->internal[IMM_RSQ],       -.5f,  1.5f,  0.0f,  0.0f);
-
-
-   machine->fpu_rnd_nearest = (X87_CW_EXCEPTION_INV_OP |
-                               X87_CW_EXCEPTION_DENORM_OP |
-                               X87_CW_EXCEPTION_ZERO_DIVIDE |
-                               X87_CW_EXCEPTION_OVERFLOW |
-                               X87_CW_EXCEPTION_UNDERFLOW |
-                               X87_CW_EXCEPTION_PRECISION |
-                               (1<<6) |
-                               X87_CW_ROUND_NEAREST |
-                               X87_CW_PRECISION_DOUBLE_EXT);
-
-   assert(machine->fpu_rnd_nearest == 0x37f);
-                               
-   machine->fpu_rnd_neg_inf = (X87_CW_EXCEPTION_INV_OP |
-                               X87_CW_EXCEPTION_DENORM_OP |
-                               X87_CW_EXCEPTION_ZERO_DIVIDE |
-                               X87_CW_EXCEPTION_OVERFLOW |
-                               X87_CW_EXCEPTION_UNDERFLOW |
-                               X87_CW_EXCEPTION_PRECISION |
-                               (1<<6) |
-                               X87_CW_ROUND_DOWN |
-                               X87_CW_PRECISION_DOUBLE_EXT);
-
-   for (i = 0; i < MAX_SHINE_TAB; i++)
-      do_populate_lut( &machine->shine_tab[i], 1.0f );
-}
 
 
 static void spill( struct aos_compilation *cp, unsigned idx )
@@ -1189,136 +1157,6 @@ static boolean emit_FRC( struct aos_compilation *cp, const struct tgsi_full_inst
    return TRUE;
 }
 
-static PIPE_CDECL void do_lit( struct aos_machine *machine,
-                               float *result,
-                               const float *in,
-                               unsigned count )
-{
-   if (in[0] > 0) 
-   {
-      if (in[1] <= 0.0) 
-      {
-         result[0] = 1.0F;
-         result[1] = in[0];
-         result[2] = 1.0;
-         result[3] = 1.0F;
-      }
-      else
-      {
-         const float epsilon = 1.0F / 256.0F;    
-         float exponent = CLAMP(in[3], -(128.0F - epsilon), (128.0F - epsilon));
-         result[0] = 1.0F;
-         result[1] = in[0];
-         result[2] = powf(in[1], exponent);
-         result[3] = 1.0;
-      }
-   }
-   else 
-   {
-      result[0] = 1.0F;
-      result[1] = 0.0;
-      result[2] = 0.0;
-      result[3] = 1.0F;
-   }
-}
-
-
-static PIPE_CDECL void do_lit_lut( struct aos_machine *machine,
-                                   float *result,
-                                   const float *in,
-                                   unsigned count )
-{
-   if (in[0] > 0) 
-   {
-      if (in[1] <= 0.0) 
-      {
-         result[0] = 1.0F;
-         result[1] = in[0];
-         result[2] = 1.0;
-         result[3] = 1.0F;
-         return;
-      }
-      
-      if (machine->lit_info[count].shine_tab->exponent != in[3]) {
-         machine->lit_info[count].func = do_lit;
-         goto no_luck;
-      }
-
-      if (in[1] <= 1.0)
-      {
-         const float *tab = machine->lit_info[count].shine_tab->values;
-         float f = in[1] * 256;
-         int k = (int)f;
-         float frac = f - (float)k;
-         
-         result[0] = 1.0F;
-         result[1] = in[0];
-         result[2] = tab[k] + frac*(tab[k+1]-tab[k]);
-         result[3] = 1.0;
-         return;
-      }
-      
-   no_luck:
-      {
-         const float epsilon = 1.0F / 256.0F;    
-         float exponent = CLAMP(in[3], -(128.0F - epsilon), (128.0F - epsilon));
-         result[0] = 1.0F;
-         result[1] = in[0];
-         result[2] = powf(in[1], exponent);
-         result[3] = 1.0;
-      }
-   }
-   else 
-   {
-      result[0] = 1.0F;
-      result[1] = 0.0;
-      result[2] = 0.0;
-      result[3] = 1.0F;
-   }
-}
-
-
-
-static void PIPE_CDECL populate_lut( struct aos_machine *machine,
-                                     float *result,
-                                     const float *in,
-                                     unsigned count )
-{
-   unsigned i, tab;
-
-   /* Search for an existing table for this value.  Note that without
-    * static analysis we don't really know if in[3] will be constant,
-    * but it usually is...
-    */
-   for (tab = 0; tab < 4; tab++) {
-      if (machine->shine_tab[tab].exponent == in[3]) {
-         goto found;
-      }
-   }
-
-   for (tab = 0, i = 1; i < 4; i++) {
-      if (machine->shine_tab[i].last_used < machine->shine_tab[tab].last_used)
-         tab = i;
-   }
-
-   if (machine->shine_tab[tab].last_used == machine->now) {
-      /* No unused tables (this is not a ffvertex program...).  Just
-       * call pow each time:
-       */
-      machine->lit_info[count].func = do_lit;
-      machine->lit_info[count].func( machine, result, in, count );
-      return;
-   }
-   else {
-      do_populate_lut( &machine->shine_tab[tab], in[3] );
-   }
-
- found:
-   machine->shine_tab[tab].last_used = machine->now;
-   machine->lit_info[count].shine_tab = &machine->shine_tab[tab];
-   machine->lit_info[count].func = do_lit_lut;
-   machine->lit_info[count].func( machine, result, in, count );
-}
 
 
 
@@ -1382,7 +1220,7 @@ static boolean emit_LIT( struct aos_compilation *cp, const struct tgsi_full_inst
                                              Offset(struct lit_info, func)));
    }
    else {
-      x86_mov_reg_imm( cp->func, ecx, (int)do_lit );
+      x86_mov_reg_imm( cp->func, ecx, (int)aos_do_lit );
    }
 
    x86_call( cp->func, ecx );
@@ -1403,7 +1241,7 @@ static boolean emit_LIT( struct aos_compilation *cp, const struct tgsi_full_inst
    return TRUE;
 }
 
-   
+#if 0   
 static boolean emit_inline_LIT( struct aos_compilation *cp, const struct tgsi_full_instruction *op )
 {
    struct x86_reg dst = get_dst_ptr(cp, &op->FullDstRegisters[0]); 
@@ -1464,6 +1302,7 @@ static boolean emit_inline_LIT( struct aos_compilation *cp, const struct tgsi_fu
 
    return TRUE;
 }
+#endif
 
 
 
@@ -1533,11 +1372,20 @@ static boolean emit_MAD( struct aos_compilation *cp, const struct tgsi_full_inst
    return TRUE;
 }
 
+/* A wrapper for powf().
+ * Makes sure it is cdecl and operates on floats.
+ */
+static float PIPE_CDECL _powerf( float x, float y )
+{
+   return powf( x, y );
+}
+
 /* Really not sufficient -- need to check for conditions that could
  * generate inf/nan values, which will slow things down hugely.
  */
 static boolean emit_POW( struct aos_compilation *cp, const struct tgsi_full_instruction *op ) 
 {
+#if 0
    x87_fld_src(cp, &op->FullSrcRegisters[1], 0);  /* a1.x */
    x87_fld_src(cp, &op->FullSrcRegisters[0], 0);	/* a0.x a1.x */
    x87_fyl2x(cp->func);	                                /* a1*log2(a0) */
@@ -1545,6 +1393,42 @@ static boolean emit_POW( struct aos_compilation *cp, const struct tgsi_full_inst
    x87_emit_ex2( cp );		/* 2^(a1*log2(a0)) */
 
    x87_fstp_dest4(cp, &op->FullDstRegisters[0]);
+#else
+   uint i;
+
+   /* For absolute correctness, need to spill/invalidate all XMM regs
+    * too.  
+    */
+   for (i = 0; i < 8; i++) {
+      if (cp->xmm[i].dirty) 
+         spill(cp, i);
+      aos_release_xmm_reg(cp, i);
+   }
+
+   /* Push caller-save (ie scratch) regs.  
+    */
+   x86_cdecl_caller_push_regs( cp->func );
+
+   x86_lea( cp->func, cp->stack_ESP, x86_make_disp(cp->stack_ESP, -8) );
+
+   x87_fld_src( cp, &op->FullSrcRegisters[1], 0 );
+   x87_fstp( cp->func, x86_make_disp( cp->stack_ESP, 4 ) );
+   x87_fld_src( cp, &op->FullSrcRegisters[0], 0 );
+   x87_fstp( cp->func, x86_make_disp( cp->stack_ESP, 0 ) );
+
+   x86_mov_reg_imm( cp->func, cp->tmp_EAX, (unsigned long) _powerf );
+   x86_call( cp->func, cp->tmp_EAX );
+
+   x86_lea( cp->func, cp->stack_ESP, x86_make_disp(cp->stack_ESP, 8) );
+
+   x86_cdecl_caller_pop_regs( cp->func );
+
+   /* Note retval on x87 stack:
+    */
+   cp->func->x87_stack++;
+
+   x87_fstp_dest4( cp, &op->FullDstRegisters[0] );
+#endif
    return TRUE;
 }
 
@@ -1865,6 +1749,7 @@ static boolean emit_rhw_viewport( struct aos_compilation *cp )
 }
 
 
+#if 0
 static boolean note_immediate( struct aos_compilation *cp,
                                struct tgsi_full_immediate *imm )
 {
@@ -1877,6 +1762,7 @@ static boolean note_immediate( struct aos_compilation *cp,
 
    return TRUE;
 }
+#endif
 
 
 
@@ -1912,7 +1798,7 @@ static void find_last_write_outputs( struct aos_compilation *cp )
 }
 
 
-#define ARG_VARIENT    1
+#define ARG_MACHINE    1
 #define ARG_START_ELTS 2
 #define ARG_COUNT      3
 #define ARG_OUTBUF     4
@@ -1939,6 +1825,8 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
    cp.outbuf_ECX    = x86_make_reg(file_REG32, reg_CX);
    cp.machine_EDX   = x86_make_reg(file_REG32, reg_DX);
    cp.count_ESI     = x86_make_reg(file_REG32, reg_SI);
+   cp.temp_EBP     = x86_make_reg(file_REG32, reg_BP);
+   cp.stack_ESP = x86_make_reg( file_REG32, reg_SP );
 
    x86_init_func(cp.func);
 
@@ -1946,11 +1834,12 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
 
    x86_push(cp.func, cp.idx_EBX);
    x86_push(cp.func, cp.count_ESI);
+   x86_push(cp.func, cp.temp_EBP);
 
 
    /* Load arguments into regs:
     */
-   x86_mov(cp.func, cp.machine_EDX, x86_fn_arg(cp.func, ARG_VARIENT));
+   x86_mov(cp.func, cp.machine_EDX, x86_fn_arg(cp.func, ARG_MACHINE));
    x86_mov(cp.func, cp.idx_EBX, x86_fn_arg(cp.func, ARG_START_ELTS));
    x86_mov(cp.func, cp.count_ESI, x86_fn_arg(cp.func, ARG_COUNT));
    x86_mov(cp.func, cp.outbuf_ECX, x86_fn_arg(cp.func, ARG_OUTBUF));
@@ -1962,11 +1851,6 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
    x86_cmp(cp.func, cp.count_ESI, cp.tmp_EAX);
    fixup = x86_jcc_forward(cp.func, cc_E);
 
-   /* Dig out the machine pointer from inside the varient arg 
-    */
-   x86_mov(cp.func, cp.machine_EDX, 
-           x86_make_disp(cp.machine_EDX,
-                         Offset( struct draw_vs_varient_aos_sse, machine )));
 
    save_fpu_state( &cp );
    set_fpu_round_nearest( &cp );
@@ -1988,8 +1872,10 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
 
          switch (parse.FullToken.Token.Type) {
          case TGSI_TOKEN_TYPE_IMMEDIATE:
+#if 0
             if (!note_immediate( &cp, &parse.FullToken.FullImmediate ))
                goto fail;
+#endif
             break;
 
          case TGSI_TOKEN_TYPE_INSTRUCTION:
@@ -2072,6 +1958,7 @@ static boolean build_vertex_program( struct draw_vs_varient_aos_sse *varient,
    if (cp.func->need_emms)
       mmx_emms(cp.func);
 
+   x86_pop(cp.func, cp.temp_EBP);
    x86_pop(cp.func, cp.count_ESI);
    x86_pop(cp.func, cp.idx_EBX);
 
@@ -2098,93 +1985,65 @@ static void vaos_set_buffer( struct draw_vs_varient *varient,
 
    for (i = 0; i < vaos->base.key.nr_inputs; i++) {
       if (vaos->base.key.element[i].in.buffer == buf) {
-         vaos->machine->attrib[i].input_ptr = ((char *)ptr +
-                                               vaos->base.key.element[i].in.offset);
-         vaos->machine->attrib[i].input_stride = stride;
+         vaos->attrib[i].input_ptr = ((char *)ptr +
+                                      vaos->base.key.element[i].in.offset);
+         vaos->attrib[i].input_stride = stride;
       }
    }
 }
 
 
-static void vaos_destroy( struct draw_vs_varient *varient )
-{
-   struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
-
-   if (vaos->machine)
-      align_free( vaos->machine );
-
-   x86_release_func( &vaos->func[0] );
-   x86_release_func( &vaos->func[1] );
 
-   FREE(vaos);
-}
-
-static void vaos_run_elts( struct draw_vs_varient *varient,
-                           const unsigned *elts,
-                           unsigned count,
-                           void *output_buffer )
+static void PIPE_CDECL vaos_run_elts( struct draw_vs_varient *varient,
+                                      const unsigned *elts,
+                                      unsigned count,
+                                      void *output_buffer )
 {
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
+   struct aos_machine *machine = vaos->draw->vs.aos_machine;
+
+   machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
+   machine->constants = vaos->draw->vs.aligned_constants;
+   machine->immediates = vaos->base.vs->immediates;
+   machine->attrib = vaos->attrib;
 
-   vaos->machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
-   vaos->gen_run_elts( varient,
+   vaos->gen_run_elts( machine,
                        elts,
                        count,
                        output_buffer );
 }
 
-static void vaos_run_linear( struct draw_vs_varient *varient,
-                             unsigned start,
-                             unsigned count,
-                             void *output_buffer )
+static void PIPE_CDECL vaos_run_linear( struct draw_vs_varient *varient,
+                                        unsigned start,
+                                        unsigned count,
+                                        void *output_buffer )
 {
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
+   struct aos_machine *machine = vaos->draw->vs.aos_machine;
 
-   vaos->machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
-   vaos->gen_run_linear( varient,
+   machine->internal[IMM_PSIZE][0] = vaos->draw->rasterizer->point_size;
+   machine->constants = vaos->draw->vs.aligned_constants;
+   machine->immediates = vaos->base.vs->immediates;
+   machine->attrib = vaos->attrib;
+
+   vaos->gen_run_linear( machine,
                          start,
                          count,
                          output_buffer );
 }
 
 
-static void vaos_set_constants( struct draw_vs_varient *varient,
-                                const float (*constants)[4] )
+
+static void vaos_destroy( struct draw_vs_varient *varient )
 {
    struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
 
-   memcpy(vaos->machine->constant,
-          constants,
-          (vaos->base.vs->info.file_max[TGSI_FILE_CONSTANT] + 1) * 4 * sizeof(float));
-
-#if 0
-   unsigned i;
-   for (i =0; i < vaos->base.vs->info.file_max[TGSI_FILE_CONSTANT] + 1; i++)
-      debug_printf("state %d: %f %f %f %f\n",
-                   i, 
-                   constants[i][0],
-                   constants[i][1],
-                   constants[i][2],
-                   constants[i][3]);
-#endif
-
-   {
-      unsigned i;
-      for (i = 0; i < MAX_LIT_INFO; i++) {
-         vaos->machine->lit_info[i].func = populate_lut;
-         vaos->machine->now++;
-      }
-   }
-}
-
+   FREE( vaos->attrib );
 
-static void vaos_set_viewport( struct draw_vs_varient *varient,
-                               const struct pipe_viewport_state *viewport )
-{
-   struct draw_vs_varient_aos_sse *vaos = (struct draw_vs_varient_aos_sse *)varient;
+   x86_release_func( &vaos->func[0] );
+   x86_release_func( &vaos->func[1] );
 
-   memcpy(vaos->machine->scale, viewport->scale, 4 * sizeof(float));
-   memcpy(vaos->machine->translate, viewport->translate, 4 * sizeof(float));
+   FREE(vaos);
 }
 
 
@@ -2200,19 +2059,15 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
    vaos->base.key = *key;
    vaos->base.vs = vs;
    vaos->base.set_input = vaos_set_buffer;
-   vaos->base.set_constants = vaos_set_constants;
-   vaos->base.set_viewport = vaos_set_viewport;
    vaos->base.destroy = vaos_destroy;
    vaos->base.run_linear = vaos_run_linear;
    vaos->base.run_elts = vaos_run_elts;
 
    vaos->draw = vs->draw;
-   vaos->machine = align_malloc( sizeof(struct aos_machine), 16 );
-   if (!vaos->machine)
+
+   vaos->attrib = MALLOC( key->nr_inputs * sizeof(vaos->attrib[0]) );
+   if (!vaos->attrib)
       goto fail;
-   
-   memset(vaos->machine, 0, sizeof(struct aos_machine));
-   init_internals(vaos->machine);
 
    tgsi_dump(vs->state.tokens, 0);
 
@@ -2222,19 +2077,19 @@ static struct draw_vs_varient *varient_aos_sse( struct draw_vertex_shader *vs,
    if (!build_vertex_program( vaos, FALSE ))
       goto fail;
 
-   vaos->gen_run_linear = (vsv_run_linear_func)x86_get_func(&vaos->func[0]);
+   vaos->gen_run_linear = (vaos_run_linear_func)x86_get_func(&vaos->func[0]);
    if (!vaos->gen_run_linear)
       goto fail;
 
-   vaos->gen_run_elts = (vsv_run_elts_func)x86_get_func(&vaos->func[1]);
+   vaos->gen_run_elts = (vaos_run_elts_func)x86_get_func(&vaos->func[1]);
    if (!vaos->gen_run_elts)
       goto fail;
 
    return &vaos->base;
 
  fail:
-   if (vaos->machine)
-      align_free( vaos->machine );
+   if (vaos && vaos->attrib)
+      FREE(vaos->attrib);
 
    if (vaos)
       x86_release_func( &vaos->func[0] );
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos.h b/src/gallium/auxiliary/draw/draw_vs_aos.h
index b47413ff43c..fb6d43d32e2 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos.h
+++ b/src/gallium/auxiliary/draw/draw_vs_aos.h
@@ -60,10 +60,16 @@ struct x86_function;
 #define FPU_RND_NEAREST 2
 
 struct aos_machine;
-typedef void PIPE_CDECL (*lit_func)( struct aos_machine *,
+typedef void (PIPE_CDECL *lit_func)( struct aos_machine *,
                                     float *result,
                                     const float *in,
                                     unsigned count );
+
+void PIPE_CDECL aos_do_lit( struct aos_machine *machine,
+                            float *result,
+                            const float *in,
+                            unsigned count );
+
 struct shine_tab {
    float exponent;
    float values[258];
@@ -78,6 +84,14 @@ struct lit_info {
 #define MAX_SHINE_TAB    4
 #define MAX_LIT_INFO     16
 
+struct aos_attrib {
+   const void *input_ptr;
+   unsigned input_stride;
+};
+
+
+
+
 /* This is the temporary storage used by all the aos_sse vs varients.
  * Create one per context and reuse by passing a pointer in at
  * vs_varient creation??
@@ -86,8 +100,6 @@ struct aos_machine {
    float input    [MAX_INPUTS    ][4];
    float output   [MAX_OUTPUTS   ][4];
    float temp     [MAX_TEMPS     ][4];
-   float constant [MAX_CONSTANTS ][4]; /* fixme -- should just be a pointer */
-   float immediate[MAX_IMMEDIATES][4]; /* fixme -- should just be a pointer */
    float internal [MAX_INTERNALS ][4];
 
    float scale[4];              /* viewport */
@@ -105,12 +117,10 @@ struct aos_machine {
    ushort fpu_restore;
    ushort fpucntl;              /* one of FPU_* above */
 
-   struct {
-      const void *input_ptr;
-      unsigned input_stride;
+   const float (*immediates)[4];     /* points to shader data */
+   const float (*constants)[4];      /* points to draw data */
 
-      unsigned output_offset;
-   } attrib[PIPE_MAX_ATTRIBS];
+   const struct aos_attrib *attrib; /* points to ? */
 };
 
 
@@ -132,6 +142,7 @@ struct aos_compilation {
       unsigned last_used;
    } xmm[8];
 
+   unsigned ebp;                /* one of X86_* */
 
    boolean input_fetched[PIPE_MAX_ATTRIBS];
    unsigned output_last_write[PIPE_MAX_ATTRIBS];
@@ -148,6 +159,8 @@ struct aos_compilation {
    struct x86_reg outbuf_ECX;
    struct x86_reg machine_EDX;
    struct x86_reg count_ESI;    /* decrements to zero */
+   struct x86_reg temp_EBP;
+   struct x86_reg stack_ESP;
 };
 
 struct x86_reg aos_get_xmm_reg( struct aos_compilation *cp );
@@ -192,25 +205,34 @@ do {                                                                    \
 } while (0)
 
 
+#define X86_NULL       0
+#define X86_IMMEDIATES 1
+#define X86_CONSTANTS  2
+#define X86_ATTRIBS    3
+
+struct x86_reg aos_get_x86( struct aos_compilation *cp,
+                            unsigned value );
 
 
+typedef void (PIPE_CDECL *vaos_run_elts_func)( struct aos_machine *,
+                                               const unsigned *elts,
+                                               unsigned count,
+                                               void *output_buffer);
+
+typedef void (PIPE_CDECL *vaos_run_linear_func)( struct aos_machine *,
+                                                unsigned start,
+                                                unsigned count,
+                                                void *output_buffer);
 
 
 struct draw_vs_varient_aos_sse {
    struct draw_vs_varient base;
    struct draw_context *draw;
 
-#if 0
-   struct {
-      const void *ptr;
-      unsigned stride;
-   } attrib[PIPE_MAX_ATTRIBS];
-#endif
-
-   struct aos_machine *machine; /* XXX: temporarily unshared */
+   struct aos_attrib *attrib;
 
-   vsv_run_linear_func gen_run_linear;
-   vsv_run_elts_func gen_run_elts;
+   vaos_run_linear_func gen_run_linear;
+   vaos_run_elts_func gen_run_elts;
 
 
    struct x86_function func[2];
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_io.c b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
index 836110f382d..45e20922092 100644
--- a/src/gallium/auxiliary/draw/draw_vs_aos_io.c
+++ b/src/gallium/auxiliary/draw/draw_vs_aos_io.c
@@ -91,25 +91,25 @@ static void emit_load_R8G8B8A8_UNORM( struct aos_compilation *cp,
 
 
 
-static void get_src_ptr( struct x86_function *func,
+static void get_src_ptr( struct aos_compilation *cp,
                          struct x86_reg src,
-                         struct x86_reg machine,
                          struct x86_reg elt,
                          unsigned a )
 {
-   struct x86_reg input_ptr = 
-      x86_make_disp(machine, 
-		    Offset(struct aos_machine, attrib[a].input_ptr));
+   struct x86_reg attrib = x86_make_disp(aos_get_x86( cp, X86_ATTRIBS ), 
+                                         a * sizeof(struct aos_attrib));
 
-   struct x86_reg input_stride = 
-      x86_make_disp(machine, 
-		    Offset(struct aos_machine, attrib[a].input_stride));
+   struct x86_reg input_ptr = x86_make_disp(attrib, 
+                                            Offset(struct aos_attrib, input_ptr));
+
+   struct x86_reg input_stride = x86_make_disp(attrib, 
+                                               Offset(struct aos_attrib, input_stride));
 
    /* Calculate pointer to current attrib:
     */
-   x86_mov(func, src, input_stride);
-   x86_imul(func, src, elt);
-   x86_add(func, src, input_ptr);
+   x86_mov(cp->func, src, input_stride);
+   x86_imul(cp->func, src, elt);
+   x86_add(cp->func, src, input_ptr);
 }
 
 
@@ -134,9 +134,8 @@ static boolean load_input( struct aos_compilation *cp,
 
    /* Figure out source pointer address:
     */
-   get_src_ptr(cp->func, 
+   get_src_ptr(cp, 
                src, 
-               cp->machine_EDX, 
                linear ? cp->idx_EBX : x86_deref(cp->idx_EBX),
                idx);
 
diff --git a/src/gallium/auxiliary/draw/draw_vs_aos_machine.c b/src/gallium/auxiliary/draw/draw_vs_aos_machine.c
new file mode 100644
index 00000000000..b7864e9f2f0
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs_aos_machine.c
@@ -0,0 +1,297 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "tgsi/util/tgsi_parse.h"
+#include "tgsi/util/tgsi_util.h"
+#include "tgsi/exec/tgsi_exec.h"
+#include "draw_vs.h"
+#include "draw_vs_aos.h"
+#include "draw_vertex.h"
+
+#include "rtasm/rtasm_x86sse.h"
+
+
+#define X87_CW_EXCEPTION_INV_OP       (1<<0)
+#define X87_CW_EXCEPTION_DENORM_OP    (1<<1)
+#define X87_CW_EXCEPTION_ZERO_DIVIDE  (1<<2)
+#define X87_CW_EXCEPTION_OVERFLOW     (1<<3)
+#define X87_CW_EXCEPTION_UNDERFLOW    (1<<4)
+#define X87_CW_EXCEPTION_PRECISION    (1<<5)
+#define X87_CW_PRECISION_SINGLE       (0<<8)
+#define X87_CW_PRECISION_RESERVED     (1<<8)
+#define X87_CW_PRECISION_DOUBLE       (2<<8)
+#define X87_CW_PRECISION_DOUBLE_EXT   (3<<8)
+#define X87_CW_PRECISION_MASK         (3<<8)
+#define X87_CW_ROUND_NEAREST          (0<<10)
+#define X87_CW_ROUND_DOWN             (1<<10)
+#define X87_CW_ROUND_UP               (2<<10)
+#define X87_CW_ROUND_ZERO             (3<<10)
+#define X87_CW_ROUND_MASK             (3<<10)
+#define X87_CW_INFINITY               (1<<12)
+
+
+void PIPE_CDECL aos_do_lit( struct aos_machine *machine,
+                            float *result,
+                            const float *in,
+                            unsigned count )
+{
+   if (in[0] > 0) 
+   {
+      if (in[1] <= 0.0) 
+      {
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = 1.0;
+         result[3] = 1.0F;
+      }
+      else
+      {
+         const float epsilon = 1.0F / 256.0F;    
+         float exponent = CLAMP(in[3], -(128.0F - epsilon), (128.0F - epsilon));
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = powf(in[1], exponent);
+         result[3] = 1.0;
+      }
+   }
+   else 
+   {
+      result[0] = 1.0F;
+      result[1] = 0.0;
+      result[2] = 0.0;
+      result[3] = 1.0F;
+   }
+}
+
+
+static void PIPE_CDECL do_lit_lut( struct aos_machine *machine,
+                                   float *result,
+                                   const float *in,
+                                   unsigned count )
+{
+   if (in[0] > 0) 
+   {
+      if (in[1] <= 0.0) 
+      {
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = 1.0;
+         result[3] = 1.0F;
+         return;
+      }
+      
+      if (machine->lit_info[count].shine_tab->exponent != in[3]) {
+         machine->lit_info[count].func = aos_do_lit;
+         goto no_luck;
+      }
+
+      if (in[1] <= 1.0)
+      {
+         const float *tab = machine->lit_info[count].shine_tab->values;
+         float f = in[1] * 256;
+         int k = (int)f;
+         float frac = f - (float)k;
+         
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = tab[k] + frac*(tab[k+1]-tab[k]);
+         result[3] = 1.0;
+         return;
+      }
+      
+   no_luck:
+      {
+         const float epsilon = 1.0F / 256.0F;    
+         float exponent = CLAMP(in[3], -(128.0F - epsilon), (128.0F - epsilon));
+         result[0] = 1.0F;
+         result[1] = in[0];
+         result[2] = powf(in[1], exponent);
+         result[3] = 1.0;
+      }
+   }
+   else 
+   {
+      result[0] = 1.0F;
+      result[1] = 0.0;
+      result[2] = 0.0;
+      result[3] = 1.0F;
+   }
+}
+
+
+static void do_populate_lut( struct shine_tab *tab,
+                             float unclamped_exponent )
+{
+   const float epsilon = 1.0F / 256.0F;    
+   float exponent = CLAMP(unclamped_exponent, -(128.0F - epsilon), (128.0F - epsilon));
+   unsigned i;
+
+   tab->exponent = unclamped_exponent; /* for later comparison */
+   
+   tab->values[0] = 0;
+   if (exponent == 0) {
+      for (i = 1; i < 258; i++) {
+         tab->values[i] = 1.0;
+      }      
+   }
+   else {
+      for (i = 1; i < 258; i++) {
+         tab->values[i] = powf((float)i * epsilon, exponent);
+      }
+   }
+}
+
+
+
+
+static void PIPE_CDECL populate_lut( struct aos_machine *machine,
+                                     float *result,
+                                     const float *in,
+                                     unsigned count )
+{
+   unsigned i, tab;
+
+   /* Search for an existing table for this value.  Note that without
+    * static analysis we don't really know if in[3] will be constant,
+    * but it usually is...
+    */
+   for (tab = 0; tab < 4; tab++) {
+      if (machine->shine_tab[tab].exponent == in[3]) {
+         goto found;
+      }
+   }
+
+   for (tab = 0, i = 1; i < 4; i++) {
+      if (machine->shine_tab[i].last_used < machine->shine_tab[tab].last_used)
+         tab = i;
+   }
+
+   if (machine->shine_tab[tab].last_used == machine->now) {
+      /* No unused tables (this is not a ffvertex program...).  Just
+       * call pow each time:
+       */
+      machine->lit_info[count].func = aos_do_lit;
+      machine->lit_info[count].func( machine, result, in, count );
+      return;
+   }
+   else {
+      do_populate_lut( &machine->shine_tab[tab], in[3] );
+   }
+
+ found:
+   machine->shine_tab[tab].last_used = machine->now;
+   machine->lit_info[count].shine_tab = &machine->shine_tab[tab];
+   machine->lit_info[count].func = do_lit_lut;
+   machine->lit_info[count].func( machine, result, in, count );
+}
+
+
+void draw_vs_aos_machine_constants( struct aos_machine *machine,
+                                    const float (*constants)[4] )
+{
+   machine->constants = constants;
+
+   {
+      unsigned i;
+      for (i = 0; i < MAX_LIT_INFO; i++) {
+         machine->lit_info[i].func = populate_lut;
+         machine->now++;
+      }
+   }
+}
+
+
+void draw_vs_aos_machine_viewport( struct aos_machine *machine,
+                                   const struct pipe_viewport_state *viewport )
+{
+   memcpy(machine->scale, viewport->scale, 4 * sizeof(float));
+   memcpy(machine->translate, viewport->translate, 4 * sizeof(float));
+}
+
+
+
+void draw_vs_aos_machine_destroy( struct aos_machine *machine )
+{
+   align_free(machine);
+}
+
+struct aos_machine *draw_vs_aos_machine( void )
+{
+   struct aos_machine *machine;
+   unsigned i;
+   float inv = 1.0f/255.0f;
+   float f255 = 255.0f;
+
+   machine = align_malloc(sizeof(struct aos_machine), 16);
+   if (!machine)
+      return NULL;
+
+   memset(machine, 0, sizeof(*machine));
+
+   ASSIGN_4V(machine->internal[IMM_SWZ],       1.0f,  -1.0f,  0.0f, 1.0f);
+   *(unsigned *)&machine->internal[IMM_SWZ][3] = 0xffffffff;
+
+   ASSIGN_4V(machine->internal[IMM_ONES],      1.0f,  1.0f,  1.0f,  1.0f);
+   ASSIGN_4V(machine->internal[IMM_NEGS],     -1.0f, -1.0f, -1.0f, -1.0f);
+   ASSIGN_4V(machine->internal[IMM_IDENTITY],  0.0f,  0.0f,  0.0f,  1.0f);
+   ASSIGN_4V(machine->internal[IMM_INV_255],   inv,   inv,   inv,   inv);
+   ASSIGN_4V(machine->internal[IMM_255],       f255,  f255,  f255,  f255);
+   ASSIGN_4V(machine->internal[IMM_RSQ],       -.5f,  1.5f,  0.0f,  0.0f);
+
+
+   machine->fpu_rnd_nearest = (X87_CW_EXCEPTION_INV_OP |
+                               X87_CW_EXCEPTION_DENORM_OP |
+                               X87_CW_EXCEPTION_ZERO_DIVIDE |
+                               X87_CW_EXCEPTION_OVERFLOW |
+                               X87_CW_EXCEPTION_UNDERFLOW |
+                               X87_CW_EXCEPTION_PRECISION |
+                               (1<<6) |
+                               X87_CW_ROUND_NEAREST |
+                               X87_CW_PRECISION_DOUBLE_EXT);
+
+   assert(machine->fpu_rnd_nearest == 0x37f);
+                               
+   machine->fpu_rnd_neg_inf = (X87_CW_EXCEPTION_INV_OP |
+                               X87_CW_EXCEPTION_DENORM_OP |
+                               X87_CW_EXCEPTION_ZERO_DIVIDE |
+                               X87_CW_EXCEPTION_OVERFLOW |
+                               X87_CW_EXCEPTION_UNDERFLOW |
+                               X87_CW_EXCEPTION_PRECISION |
+                               (1<<6) |
+                               X87_CW_ROUND_DOWN |
+                               X87_CW_PRECISION_DOUBLE_EXT);
+
+   for (i = 0; i < MAX_SHINE_TAB; i++)
+      do_populate_lut( &machine->shine_tab[i], 1.0f );
+
+   return machine;
+}
+
+
diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c
index 2dc9352c6f9..37cbe9bf9e3 100644
--- a/src/gallium/auxiliary/draw/draw_vs_sse.c
+++ b/src/gallium/auxiliary/draw/draw_vs_sse.c
@@ -70,8 +70,6 @@ struct draw_sse_vertex_shader {
    codegen_function func;
    
    struct tgsi_exec_machine *machine;
-
-   float immediates[TGSI_EXEC_NUM_IMMEDIATES][4];
 };
 
 
@@ -109,7 +107,7 @@ vs_sse_run_linear( struct draw_vertex_shader *base,
 		   machine->Outputs,
 		   (float (*)[4])constants,
 		   machine->Temps,
-		   shader->immediates,
+		   (float (*)[4])shader->base.immediates,
                    input,
                    base->info.num_inputs,
                    input_stride,
@@ -132,6 +130,8 @@ vs_sse_delete( struct draw_vertex_shader *base )
    
    x86_release_func( &shader->sse2_program );
 
+   align_free(shader->base.immediates);
+
    FREE( (void*) shader->base.state.tokens );
    FREE( shader );
 }
@@ -163,12 +163,18 @@ draw_create_vs_sse(struct draw_context *draw,
    vs->base.prepare = vs_sse_prepare;
    vs->base.run_linear = vs_sse_run_linear;
    vs->base.delete = vs_sse_delete;
+   
+   vs->base.immediates = align_malloc(TGSI_EXEC_NUM_IMMEDIATES * 4 *
+                                      sizeof(float), 16);
+
    vs->machine = &draw->vs.machine;
    
    x86_init_func( &vs->sse2_program );
 
    if (!tgsi_emit_sse2( (struct tgsi_token *) vs->base.state.tokens,
-			&vs->sse2_program, vs->immediates, TRUE )) 
+			&vs->sse2_program, 
+                        (float (*)[4])vs->base.immediates, 
+                        TRUE )) 
       goto fail;
       
    vs->func = (codegen_function) x86_get_func( &vs->sse2_program );
diff --git a/src/gallium/auxiliary/draw/draw_vs_varient.c b/src/gallium/auxiliary/draw/draw_vs_varient.c
index 119a3a04b56..4a155251b53 100644
--- a/src/gallium/auxiliary/draw/draw_vs_varient.c
+++ b/src/gallium/auxiliary/draw/draw_vs_varient.c
@@ -44,8 +44,6 @@
 struct draw_vs_varient_generic {
    struct draw_vs_varient base;
 
-   struct pipe_viewport_state viewport;
-
    struct draw_vertex_shader *shader;
    struct draw_context *draw;
    
@@ -57,21 +55,11 @@ struct draw_vs_varient_generic {
     */
    struct translate *fetch;
    struct translate *emit;
-
-   const float (*constants)[4];
 };
 
 
 
 
-static void vsvg_set_constants( struct draw_vs_varient *varient,
-                                const float (*constants)[4] )
-{
-   struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
-
-   vsvg->constants = constants;
-}
-
 
 static void vsvg_set_input( struct draw_vs_varient *varient,
                             unsigned buffer,
@@ -94,8 +82,8 @@ static void do_rhw_viewport( struct draw_vs_varient_generic *vsvg,
                              void *output_buffer )
 {
    char *ptr = (char *)output_buffer;
-   const float *scale = vsvg->viewport.scale;
-   const float *trans = vsvg->viewport.translate;
+   const float *scale = vsvg->base.vs->draw->viewport.scale;
+   const float *trans = vsvg->base.vs->draw->viewport.translate;
    unsigned stride = vsvg->base.key.output_stride;
    unsigned j;
 
@@ -115,8 +103,8 @@ static void do_viewport( struct draw_vs_varient_generic *vsvg,
                              void *output_buffer )
 {
    char *ptr = (char *)output_buffer;
-   const float *scale = vsvg->viewport.scale;
-   const float *trans = vsvg->viewport.translate;
+   const float *scale = vsvg->base.vs->draw->viewport.scale;
+   const float *trans = vsvg->base.vs->draw->viewport.translate;
    unsigned stride = vsvg->base.key.output_stride;
    unsigned j;
 
@@ -130,10 +118,10 @@ static void do_viewport( struct draw_vs_varient_generic *vsvg,
 }
                          
 
-static void vsvg_run_elts( struct draw_vs_varient *varient,
-                           const unsigned *elts,
-                           unsigned count,
-                           void *output_buffer)
+static void PIPE_CDECL vsvg_run_elts( struct draw_vs_varient *varient,
+                                      const unsigned *elts,
+                                      unsigned count,
+                                      void *output_buffer)
 {
    struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
 			
@@ -150,7 +138,7 @@ static void vsvg_run_elts( struct draw_vs_varient *varient,
       vsvg->base.vs->run_linear( vsvg->base.vs, 
                                  output_buffer,
                                  output_buffer,
-                                 vsvg->constants,
+                                 (const float (*)[4])vsvg->base.vs->draw->pt.user.constants,
                                  count,
                                  vsvg->base.key.output_stride, 
                                  vsvg->base.key.output_stride);
@@ -186,10 +174,10 @@ static void vsvg_run_elts( struct draw_vs_varient *varient,
 }
 
 
-static void vsvg_run_linear( struct draw_vs_varient *varient,
-                                   unsigned start,
-                                   unsigned count,
-                                   void *output_buffer )
+static void PIPE_CDECL vsvg_run_linear( struct draw_vs_varient *varient,
+                                        unsigned start,
+                                        unsigned count,
+                                        void *output_buffer )
 {
    struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
 	
@@ -206,7 +194,7 @@ static void vsvg_run_linear( struct draw_vs_varient *varient,
       vsvg->base.vs->run_linear( vsvg->base.vs, 
                                  output_buffer,
                                  output_buffer,
-                                 vsvg->constants,
+                                 (const float (*)[4])vsvg->base.vs->draw->pt.user.constants,
                                  count,
                                  vsvg->base.key.output_stride, 
                                  vsvg->base.key.output_stride);
@@ -245,13 +233,6 @@ static void vsvg_run_linear( struct draw_vs_varient *varient,
 
 
 
-static void vsvg_set_viewport( struct draw_vs_varient *varient,
-                               const struct pipe_viewport_state *viewport )
-{
-   struct draw_vs_varient_generic *vsvg = (struct draw_vs_varient_generic *)varient;
-
-   vsvg->viewport = *viewport;
-}
 
 static void vsvg_destroy( struct draw_vs_varient *varient )
 {
@@ -272,8 +253,6 @@ struct draw_vs_varient *draw_vs_varient_generic( struct draw_vertex_shader *vs,
    vsvg->base.key = *key;
    vsvg->base.vs = vs;
    vsvg->base.set_input     = vsvg_set_input;
-   vsvg->base.set_constants = vsvg_set_constants;
-   vsvg->base.set_viewport  = vsvg_set_viewport;
    vsvg->base.run_elts      = vsvg_run_elts;
    vsvg->base.run_linear    = vsvg_run_linear;
    vsvg->base.destroy       = vsvg_destroy;
diff --git a/src/gallium/auxiliary/pipebuffer/pb_buffer_malloc.c b/src/gallium/auxiliary/pipebuffer/pb_buffer_malloc.c
index 9e8244f909b..6c3502eea7f 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_buffer_malloc.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_buffer_malloc.c
@@ -30,13 +30,14 @@
  * Implementation of malloc-based buffers to store data that can't be processed
  * by the hardware. 
  * 
- * \author Jos� Fonseca <[email protected]>
+ * \author Jose Fonseca <[email protected]>
  */
 
 
 #include "pipe/p_debug.h"
 #include "pipe/p_util.h"
 #include "pb_buffer.h"
+#include "pb_bufmgr.h"
 
 
 struct malloc_buffer 
@@ -125,3 +126,33 @@ pb_malloc_buffer_create(size_t size,
 
    return &buf->base;
 }
+
+
+static struct pb_buffer *
+pb_malloc_buffer_create_buffer(struct pb_manager *mgr, 
+                               size_t size,
+                               const struct pb_desc *desc) 
+{
+   return pb_malloc_buffer_create(size, desc);
+}
+
+
+static void
+pb_malloc_bufmgr_destroy(struct pb_manager *mgr) 
+{
+   /* No-op */
+}
+
+
+static struct pb_manager 
+pb_malloc_bufmgr = {
+   pb_malloc_buffer_create_buffer,
+   pb_malloc_bufmgr_destroy
+};
+
+
+struct pb_manager *
+pb_malloc_bufmgr_create(void) 
+{
+  return &pb_malloc_bufmgr;
+}
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h b/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h
index f6cc7a525b7..4a922d16c18 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr.h
@@ -79,6 +79,15 @@ struct pb_manager
 };
 
 
+/**
+ * Malloc buffer provider.
+ * 
+ * Simple wrapper around pb_malloc_buffer_create for convenience.
+ */
+struct pb_manager *
+pb_malloc_bufmgr_create(void);
+
+
 /** 
  * Static buffer pool sub-allocator.
  * 
diff --git a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c
index 45ba158a4d9..b9dff09804c 100644
--- a/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c
+++ b/src/gallium/auxiliary/pipebuffer/pb_bufmgr_slab.c
@@ -312,8 +312,8 @@ pb_slab_manager_create_buffer(struct pb_manager *_mgr,
    struct list_head *list;
 
    /* check size */
-   assert(size == mgr->bufSize);
-   if(size != mgr->bufSize)
+   assert(size <= mgr->bufSize);
+   if(size > mgr->bufSize)
       return NULL;
    
    /* check if we can provide the requested alignment */
diff --git a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
index 698389ffb7b..664a69a5371 100644
--- a/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
+++ b/src/gallium/auxiliary/rtasm/rtasm_x86sse.c
@@ -330,7 +330,7 @@ struct x86_reg x86_make_disp( struct x86_reg reg,
    else
       reg.disp += disp;
 
-   if (reg.disp == 0)
+   if (reg.disp == 0 && reg.idx != reg_BP)
       reg.mod = mod_INDIRECT;
    else if (reg.disp <= 127 && reg.disp >= -128)
       reg.mod = mod_DISP8;
diff --git a/src/gallium/auxiliary/tgsi/exec/tgsi_sse2.c b/src/gallium/auxiliary/tgsi/exec/tgsi_sse2.c
index 8018bd7fa45..189dc6024d1 100755
--- a/src/gallium/auxiliary/tgsi/exec/tgsi_sse2.c
+++ b/src/gallium/auxiliary/tgsi/exec/tgsi_sse2.c
@@ -198,7 +198,7 @@ get_coef(
 static void
 emit_retw(
    struct x86_function  *func,
-   unsigned             size )
+   unsigned short        size )
 {
    x86_retw( func, size );
 }
diff --git a/src/gallium/auxiliary/tgsi/util/tgsi_dump.c b/src/gallium/auxiliary/tgsi/util/tgsi_dump.c
index 648afa2a51b..b018ea9fa1c 100644
--- a/src/gallium/auxiliary/tgsi/util/tgsi_dump.c
+++ b/src/gallium/auxiliary/tgsi/util/tgsi_dump.c
@@ -754,7 +754,7 @@ tgsi_dump_instruction(
    }
 
    for( i = 0; i < inst->Instruction.NumDstRegs; i++ ) {
-      struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i];
+      const struct tgsi_full_dst_register *dst = &inst->FullDstRegisters[i];
 
       if( !first_reg ) {
          CHR( ',' );
@@ -812,7 +812,7 @@ tgsi_dump_instruction(
    }
 
    for( i = 0; i < inst->Instruction.NumSrcRegs; i++ ) {
-      struct tgsi_full_src_register *src = &inst->FullSrcRegisters[i];
+      const struct tgsi_full_src_register *src = &inst->FullSrcRegisters[i];
 
       if( !first_reg ) {
          CHR( ',' );
diff --git a/src/gallium/auxiliary/translate/translate_sse.c b/src/gallium/auxiliary/translate/translate_sse.c
index af86260725a..18a212ac1cf 100644
--- a/src/gallium/auxiliary/translate/translate_sse.c
+++ b/src/gallium/auxiliary/translate/translate_sse.c
@@ -310,7 +310,7 @@ static void get_src_ptr( struct translate_sse *p,
 static void emit_swizzle( struct translate_sse *p,
 			  struct x86_reg dest,
 			  struct x86_reg src,
-			  unsigned shuffle )
+			  unsigned char shuffle )
 {
    sse_shufps(p->func, dest, src, shuffle);
 }
@@ -535,7 +535,7 @@ static void translate_sse_release( struct translate *translate )
    FREE(p);
 }
 
-static void translate_sse_run_elts( struct translate *translate,
+static void PIPE_CDECL translate_sse_run_elts( struct translate *translate,
 			      const unsigned *elts,
 			      unsigned count,
 			      void *output_buffer )
@@ -548,7 +548,7 @@ static void translate_sse_run_elts( struct translate *translate,
 		    output_buffer );
 }
 
-static void translate_sse_run( struct translate *translate,
+static void PIPE_CDECL translate_sse_run( struct translate *translate,
 			 unsigned start,
 			 unsigned count,
 			 void *output_buffer )