draw: subclass vertex shaders according to execution method

Create new files for shaders compiled/executed with llvm, sse, exec
respectively

7 files changed, 763 insertions, 232 deletions

diff --git a/src/gallium/auxiliary/draw/Makefile b/src/gallium/auxiliary/draw/Makefile
index 451911a354..fe9b150f30 100644
--- a/src/gallium/auxiliary/draw/Makefile
+++ b/src/gallium/auxiliary/draw/Makefile
@@ -1,2 +1,2 @@
 default:
-	cd .. ; make
+	cd ../../../mesa ; make
diff --git a/src/gallium/auxiliary/draw/draw_private.h b/src/gallium/auxiliary/draw/draw_private.h
index 3d09aef87c..bc11259cb2 100644
--- a/src/gallium/auxiliary/draw/draw_private.h
+++ b/src/gallium/auxiliary/draw/draw_private.h
@@ -128,13 +128,25 @@ struct draw_stage
  * Private version of the compiled vertex_shader
  */
 struct draw_vertex_shader {
+
+   /* This member will disappear shortly:
+    */
    const struct pipe_shader_state   *state;
-#if defined(__i386__) || defined(__386__)
-   struct x86_function              sse2_program;
-#endif
-#ifdef MESA_LLVM
-   struct gallivm_prog *llvm_prog;
-#endif
+
+   void (*prepare)( struct draw_vertex_shader *shader,
+		    struct draw_context *draw );
+
+   /* Run the shader - this interface will get cleaned up in the
+    * future:
+    */
+   void (*run)( struct draw_vertex_shader *shader,
+		struct draw_context *draw,
+		const unsigned *elts, 
+		unsigned count,
+		struct vertex_header *vOut[] );
+		    
+
+   void (*delete)( struct draw_vertex_shader * );
 };
 
 
@@ -176,7 +188,7 @@ struct draw_context
    struct pipe_viewport_state viewport;
    struct pipe_vertex_buffer vertex_buffer[PIPE_ATTRIB_MAX];
    struct pipe_vertex_element vertex_element[PIPE_ATTRIB_MAX];
-   const struct draw_vertex_shader *vertex_shader;
+   struct draw_vertex_shader *vertex_shader;
 
    uint num_vs_outputs;  /**< convenience, from vertex_shader */
 
@@ -201,6 +213,7 @@ struct draw_context
 
    boolean convert_wide_points; /**< convert wide points to tris? */
    boolean convert_wide_lines;  /**< convert side lines to tris? */
+   boolean use_sse;
 
    unsigned reduced_prim;
 
@@ -255,11 +268,10 @@ struct draw_context
       unsigned queue_nr;
    } pq;
 
-   int use_sse : 1;
-#ifdef MESA_LLVM
-   struct gallivm_cpu_engine *engine;
-#endif
-   
+
+   /* This (and the tgsi_exec_machine struct) probably need to be moved somewhere private.
+    */
+   struct gallivm_cpu_engine *engine;   
    void *driver_private;
 };
 
@@ -290,11 +302,7 @@ extern void draw_vertex_cache_invalidate( struct draw_context *draw );
 extern void draw_vertex_cache_unreference( struct draw_context *draw );
 extern void draw_vertex_cache_reset_vertex_ids( struct draw_context *draw );
 
-
 extern void draw_vertex_shader_queue_flush( struct draw_context *draw );
-#ifdef MESA_LLVM
-extern void draw_vertex_shader_queue_flush_llvm( struct draw_context *draw );
-#endif
 
 struct tgsi_exec_machine;
 
diff --git a/src/gallium/auxiliary/draw/draw_vertex_shader.c b/src/gallium/auxiliary/draw/draw_vertex_shader.c
index 9413f8b43a..f68f6e3244 100644
--- a/src/gallium/auxiliary/draw/draw_vertex_shader.c
+++ b/src/gallium/auxiliary/draw/draw_vertex_shader.c
@@ -33,177 +33,10 @@
 
 #include "pipe/p_util.h"
 #include "pipe/p_shader_tokens.h"
-#if defined(__i386__) || defined(__386__)
-#include "tgsi/exec/tgsi_sse2.h"
-#endif
 #include "draw_private.h"
 #include "draw_context.h"
+#include "draw_vs.h"
 
-#include "x86/rtasm/x86sse.h"
-#include "llvm/gallivm.h"
-
-
-#define DBG_VS 0
-
-
-static INLINE unsigned
-compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr)
-{
-   unsigned mask = 0;
-   unsigned i;
-
-   /* Do the hardwired planes first:
-    */
-   if (-clip[0] + clip[3] < 0) mask |= CLIP_RIGHT_BIT;
-   if ( clip[0] + clip[3] < 0) mask |= CLIP_LEFT_BIT;
-   if (-clip[1] + clip[3] < 0) mask |= CLIP_TOP_BIT;
-   if ( clip[1] + clip[3] < 0) mask |= CLIP_BOTTOM_BIT;
-   if (-clip[2] + clip[3] < 0) mask |= CLIP_FAR_BIT;
-   if ( clip[2] + clip[3] < 0) mask |= CLIP_NEAR_BIT;
-
-   /* Followed by any remaining ones:
-    */
-   for (i = 6; i < nr; i++) {
-      if (dot4(clip, plane[i]) < 0) 
-         mask |= (1<<i);
-   }
-
-   return mask;
-}
-
-
-typedef void (XSTDCALL *codegen_function) (
-   const struct tgsi_exec_vector *input,
-   struct tgsi_exec_vector *output,
-   float (*constant)[4],
-   struct tgsi_exec_vector *temporary );
-
-
-/**
- * Transform vertices with the current vertex program/shader
- * Up to four vertices can be shaded at a time.
- * \param vbuffer  the input vertex data
- * \param elts  indexes of four input vertices
- * \param count  number of vertices to shade [1..4]
- * \param vOut  array of pointers to four output vertices
- */
-static void
-run_vertex_program(struct draw_context *draw,
-                   unsigned elts[4], unsigned count,
-                   struct vertex_header *vOut[])
-{
-   struct tgsi_exec_machine *machine = &draw->machine;
-   unsigned int j;
-
-   ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_ATTRIB_MAX);
-   ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_ATTRIB_MAX);
-   const float *scale = draw->viewport.scale;
-   const float *trans = draw->viewport.translate;
-
-   assert(count <= 4);
-   assert(draw->vertex_shader->state->output_semantic_name[0]
-          == TGSI_SEMANTIC_POSITION);
-
-   /* Consts does not require 16 byte alignment. */
-   machine->Consts = (float (*)[4]) draw->user.constants;
-
-   machine->Inputs = ALIGN16_ASSIGN(inputs);
-   machine->Outputs = ALIGN16_ASSIGN(outputs);
-
-   draw->vertex_fetch.fetch_func( draw, machine, elts, count );
-
-   /* run shader */
-#ifdef MESA_LLVM
-   if (1) {
-   struct gallivm_prog  *prog  = draw->vertex_shader->llvm_prog;
-   gallivm_cpu_vs_exec(prog,
-                       machine->Inputs,
-                       machine->Outputs,
-                       machine->Consts,
-                       machine->Temps);
-   } else
-#elif defined(__i386__) || defined(__386__)
-   if (draw->use_sse) {
-      /* SSE */
-      /* cast away const */
-      struct draw_vertex_shader *shader
-         = (struct draw_vertex_shader *)draw->vertex_shader;
-      codegen_function func
-         = (codegen_function) x86_get_func( &shader->sse2_program );
-
-      if (func)
-         func(
-            machine->Inputs,
-            machine->Outputs,
-            machine->Consts,
-            machine->Temps );
-      else
-         /* interpreter */
-         tgsi_exec_machine_run( machine );
-   }
-   else
-#endif
-   {
-      /* interpreter */
-      tgsi_exec_machine_run( machine );
-   }
-
-   /* store machine results */
-   for (j = 0; j < count; j++) {
-      unsigned slot;
-      float x, y, z, w;
-
-      /* Handle attr[0] (position) specially:
-       *
-       * XXX: Computing the clipmask should be done in the vertex
-       * program as a set of DP4 instructions appended to the
-       * user-provided code.
-       */
-      x = vOut[j]->clip[0] = machine->Outputs[0].xyzw[0].f[j];
-      y = vOut[j]->clip[1] = machine->Outputs[0].xyzw[1].f[j];
-      z = vOut[j]->clip[2] = machine->Outputs[0].xyzw[2].f[j];
-      w = vOut[j]->clip[3] = machine->Outputs[0].xyzw[3].f[j];
-
-      vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, draw->nr_planes);
-      vOut[j]->edgeflag = 1;
-
-      /* divide by w */
-      w = 1.0f / w;
-      x *= w;
-      y *= w;
-      z *= w;
-
-      /* Viewport mapping */
-      vOut[j]->data[0][0] = x * scale[0] + trans[0];
-      vOut[j]->data[0][1] = y * scale[1] + trans[1];
-      vOut[j]->data[0][2] = z * scale[2] + trans[2];
-      vOut[j]->data[0][3] = w;
-
-#if DBG_VS
-      debug_printf("output[%d]win: %f %f %f %f\n", j,
-             vOut[j]->data[0][0],
-             vOut[j]->data[0][1],
-             vOut[j]->data[0][2],
-             vOut[j]->data[0][3]);
-#endif
-      /* Remaining attributes are packed into sequential post-transform
-       * vertex attrib slots.
-       */
-      for (slot = 1; slot < draw->num_vs_outputs; slot++) {
-         vOut[j]->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
-         vOut[j]->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
-         vOut[j]->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
-         vOut[j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
-#if DBG_VS
-         debug_printf("output[%d][%d]: %f %f %f %f\n", j, slot,
-                vOut[j]->data[slot][0],
-                vOut[j]->data[slot][1],
-                vOut[j]->data[slot][2],
-                vOut[j]->data[slot][3]);
-#endif
-      }
-   } /* loop over vertices */
-}
 
 
 /**
@@ -213,13 +46,14 @@ run_vertex_program(struct draw_context *draw,
 void
 draw_vertex_shader_queue_flush(struct draw_context *draw)
 {
+   struct draw_vertex_shader *shader = draw->vertex_shader;
    unsigned i;
 
    assert(draw->vs.queue_nr != 0);
 
    /* XXX: do this on statechange: 
     */
-   draw_update_vertex_fetch( draw );
+   shader->prepare( shader, draw );
 
 //   fprintf(stderr, " q(%d) ", draw->vs.queue_nr );
 
@@ -242,7 +76,7 @@ draw_vertex_shader_queue_flush(struct draw_context *draw)
       assert(n > 0);
       assert(n <= 4);
 
-      run_vertex_program(draw, elts, n, dests);
+      shader->run(shader, draw, elts, n, dests);
    }
 
    draw->vs.queue_nr = 0;
@@ -255,43 +89,16 @@ draw_create_vertex_shader(struct draw_context *draw,
 {
    struct draw_vertex_shader *vs;
 
-   vs = CALLOC_STRUCT( draw_vertex_shader );
-   if (vs == NULL) {
-      return NULL;
-   }
-
-   vs->state = shader;
+   vs = draw_create_vs_llvm( draw, shader );
+   if (vs)
+      return vs;
 
-#ifdef MESA_LLVM
-   struct gallivm_ir *ir = gallivm_ir_new(GALLIVM_VS);
-   gallivm_ir_set_layout(ir, GALLIVM_SOA);
-   gallivm_ir_set_components(ir, 4);
-   gallivm_ir_fill_from_tgsi(ir, shader->tokens);
-   vs->llvm_prog = gallivm_ir_compile(ir);
-   gallivm_ir_delete(ir);
-
-   draw->engine = gallivm_global_cpu_engine();
-   if (!draw->engine) {
-      draw->engine = gallivm_cpu_engine_create(vs->llvm_prog);
-   }
-   else {
-      gallivm_cpu_jit_compile(draw->engine, vs->llvm_prog);
-   }
-#elif defined(__i386__) || defined(__386__)
-   if (draw->use_sse) {
-      /* cast-away const */
-      struct pipe_shader_state *sh = (struct pipe_shader_state *) shader;
-
-      x86_init_func( &vs->sse2_program );
-      if (!tgsi_emit_sse2( (struct tgsi_token *) sh->tokens,
-                           &vs->sse2_program )) {
-         x86_release_func( (struct x86_function *) &vs->sse2_program );
-	 fprintf(stdout /*err*/,
-		 "tgsi_emit_sse2() failed, falling back to interpreter\n");
-      }
-   }
-#endif
+   vs = draw_create_vs_sse( draw, shader );
+   if (vs)
+      return vs;
 
+   vs = draw_create_vs_exec( draw, shader );
+   assert(vs);
    return vs;
 }
 
@@ -307,11 +114,7 @@ draw_bind_vertex_shader(struct draw_context *draw,
 
    tgsi_exec_machine_init(&draw->machine);
 
-   /* specify the vertex program to interpret/execute */
-   tgsi_exec_machine_bind_shader(&draw->machine,
-				 draw->vertex_shader->state->tokens,
-				 PIPE_MAX_SAMPLERS,
-				 NULL /*samplers*/ );
+   dvs->prepare( dvs, draw );
 }
 
 
@@ -319,9 +122,5 @@ void
 draw_delete_vertex_shader(struct draw_context *draw,
                           struct draw_vertex_shader *dvs)
 {
-#if defined(__i386__) || defined(__386__)
-   x86_release_func( (struct x86_function *) &dvs->sse2_program );
-#endif
-
-   FREE( dvs );
+   dvs->delete( dvs );
 }
diff --git a/src/gallium/auxiliary/draw/draw_vs.h b/src/gallium/auxiliary/draw/draw_vs.h
new file mode 100644
index 0000000000..4ee7e705e9
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs.h
@@ -0,0 +1,50 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+/* Authors:  Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#ifndef DRAW_VS_H
+#define DRAW_VS_H
+
+struct draw_vertex_shader;
+struct draw_context;
+struct pipe_shader_state;
+
+struct draw_vertex_shader *
+draw_create_vs_exec(struct draw_context *draw,
+		    const struct pipe_shader_state *templ);
+
+struct draw_vertex_shader *
+draw_create_vs_sse(struct draw_context *draw,
+		   const struct pipe_shader_state *templ);
+
+struct draw_vertex_shader *
+draw_create_vs_llvm(struct draw_context *draw,
+		    const struct pipe_shader_state *templ);
+
+#endif
diff --git a/src/gallium/auxiliary/draw/draw_vs_exec.c b/src/gallium/auxiliary/draw/draw_vs_exec.c
new file mode 100644
index 0000000000..8588879400
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs_exec.c
@@ -0,0 +1,186 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  *   Brian Paul
+  */
+
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "draw_private.h"
+#include "draw_context.h"
+#include "draw_vs.h"
+
+
+static INLINE unsigned
+compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr)
+{
+   unsigned mask = 0;
+   unsigned i;
+
+   /* Do the hardwired planes first:
+    */
+   if (-clip[0] + clip[3] < 0) mask |= CLIP_RIGHT_BIT;
+   if ( clip[0] + clip[3] < 0) mask |= CLIP_LEFT_BIT;
+   if (-clip[1] + clip[3] < 0) mask |= CLIP_TOP_BIT;
+   if ( clip[1] + clip[3] < 0) mask |= CLIP_BOTTOM_BIT;
+   if (-clip[2] + clip[3] < 0) mask |= CLIP_FAR_BIT;
+   if ( clip[2] + clip[3] < 0) mask |= CLIP_NEAR_BIT;
+
+   /* Followed by any remaining ones:
+    */
+   for (i = 6; i < nr; i++) {
+      if (dot4(clip, plane[i]) < 0) 
+         mask |= (1<<i);
+   }
+
+   return mask;
+}
+
+
+static void
+vs_exec_prepare( struct draw_vertex_shader *shader,
+		 struct draw_context *draw )
+{
+   /* specify the vertex program to interpret/execute */
+   tgsi_exec_machine_bind_shader(&draw->machine,
+				 shader->state->tokens,
+				 PIPE_MAX_SAMPLERS,
+				 NULL /*samplers*/ );
+
+   draw_update_vertex_fetch( draw );
+}
+
+
+/**
+ * Transform vertices with the current vertex program/shader
+ * Up to four vertices can be shaded at a time.
+ * \param vbuffer  the input vertex data
+ * \param elts  indexes of four input vertices
+ * \param count  number of vertices to shade [1..4]
+ * \param vOut  array of pointers to four output vertices
+ */
+static void
+vs_exec_run( struct draw_vertex_shader *shader,
+	     struct draw_context *draw,
+	     const unsigned *elts, 
+	     unsigned count,
+	     struct vertex_header *vOut[] )
+{
+   struct tgsi_exec_machine *machine = &draw->machine;
+   unsigned int j;
+
+   ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_ATTRIB_MAX);
+   ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_ATTRIB_MAX);
+   const float *scale = draw->viewport.scale;
+   const float *trans = draw->viewport.translate;
+
+   assert(count <= 4);
+   assert(draw->vertex_shader->state->output_semantic_name[0]
+          == TGSI_SEMANTIC_POSITION);
+
+   machine->Consts = (float (*)[4]) draw->user.constants;
+   machine->Inputs = ALIGN16_ASSIGN(inputs);
+   machine->Outputs = ALIGN16_ASSIGN(outputs);
+
+   draw->vertex_fetch.fetch_func( draw, machine, elts, count );
+
+   /* run interpreter */
+   tgsi_exec_machine_run( machine );
+
+
+   /* store machine results */
+   for (j = 0; j < count; j++) {
+      unsigned slot;
+      float x, y, z, w;
+
+      /* Handle attr[0] (position) specially:
+       *
+       * XXX: Computing the clipmask should be done in the vertex
+       * program as a set of DP4 instructions appended to the
+       * user-provided code.
+       */
+      x = vOut[j]->clip[0] = machine->Outputs[0].xyzw[0].f[j];
+      y = vOut[j]->clip[1] = machine->Outputs[0].xyzw[1].f[j];
+      z = vOut[j]->clip[2] = machine->Outputs[0].xyzw[2].f[j];
+      w = vOut[j]->clip[3] = machine->Outputs[0].xyzw[3].f[j];
+
+      vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, draw->nr_planes);
+      vOut[j]->edgeflag = 1;
+
+      /* divide by w */
+      w = 1.0f / w;
+      x *= w;
+      y *= w;
+      z *= w;
+
+      /* Viewport mapping */
+      vOut[j]->data[0][0] = x * scale[0] + trans[0];
+      vOut[j]->data[0][1] = y * scale[1] + trans[1];
+      vOut[j]->data[0][2] = z * scale[2] + trans[2];
+      vOut[j]->data[0][3] = w;
+
+      /* Remaining attributes are packed into sequential post-transform
+       * vertex attrib slots.
+       */
+      for (slot = 1; slot < draw->num_vs_outputs; slot++) {
+         vOut[j]->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
+         vOut[j]->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
+         vOut[j]->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
+         vOut[j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
+      }
+   } /* loop over vertices */
+}
+
+
+
+static void
+vs_exec_delete( struct draw_vertex_shader *dvs )
+{
+   FREE( dvs );
+}
+
+
+struct draw_vertex_shader *
+draw_create_vs_exec(struct draw_context *draw,
+		    const struct pipe_shader_state *state)
+{
+   struct draw_vertex_shader *vs = CALLOC_STRUCT( draw_vertex_shader );
+
+   if (vs == NULL) 
+      return NULL;
+
+   vs->state = state;
+   vs->prepare = vs_exec_prepare;
+   vs->run = vs_exec_run;
+   vs->delete = vs_exec_delete;
+
+   return vs;
+}
diff --git a/src/gallium/auxiliary/draw/draw_vs_llvm.c b/src/gallium/auxiliary/draw/draw_vs_llvm.c
new file mode 100644
index 0000000000..44022b6e07
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs_llvm.c
@@ -0,0 +1,237 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Zack Rusin
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  *   Brian Paul
+  */
+
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+#include "draw_private.h"
+#include "draw_context.h"
+#include "draw_vs.h"
+
+#ifdef MESA_LLVM
+
+#include "llvm/gallivm.h"
+
+struct draw_llvm_vertex_shader {
+   struct draw_vertex_shader base;
+   struct gallivm_prog *llvm_prog;
+};
+
+
+static INLINE unsigned
+compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr)
+{
+   unsigned mask = 0;
+   unsigned i;
+
+   /* Do the hardwired planes first:
+    */
+   if (-clip[0] + clip[3] < 0) mask |= CLIP_RIGHT_BIT;
+   if ( clip[0] + clip[3] < 0) mask |= CLIP_LEFT_BIT;
+   if (-clip[1] + clip[3] < 0) mask |= CLIP_TOP_BIT;
+   if ( clip[1] + clip[3] < 0) mask |= CLIP_BOTTOM_BIT;
+   if (-clip[2] + clip[3] < 0) mask |= CLIP_FAR_BIT;
+   if ( clip[2] + clip[3] < 0) mask |= CLIP_NEAR_BIT;
+
+   /* Followed by any remaining ones:
+    */
+   for (i = 6; i < nr; i++) {
+      if (dot4(clip, plane[i]) < 0) 
+         mask |= (1<<i);
+   }
+
+   return mask;
+}
+
+
+
+static void
+vs_llvm_prepare( struct draw_vertex_shader *base,
+		 struct draw_context *draw )
+{
+   draw_update_vertex_fetch( draw );
+}
+
+
+
+/**
+ * Transform vertices with the current vertex program/shader
+ * Up to four vertices can be shaded at a time.
+ * \param vbuffer  the input vertex data
+ * \param elts  indexes of four input vertices
+ * \param count  number of vertices to shade [1..4]
+ * \param vOut  array of pointers to four output vertices
+ */
+static void
+vs_llvm_run( struct draw_vertex_shader *base,
+	     struct draw_context *draw, 
+	     const unsigned *elts, 
+	     unsigned count,
+	     struct vertex_header *vOut[] )
+{
+   struct draw_llvm_vertex_shader *shader = 
+      (struct draw_llvm_vertex_shader *)base;
+
+   struct tgsi_exec_machine *machine = &draw->machine;
+   unsigned int j;
+
+   ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_ATTRIB_MAX);
+   ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_ATTRIB_MAX);
+   const float *scale = draw->viewport.scale;
+   const float *trans = draw->viewport.translate;
+
+
+   assert(count <= 4);
+   assert(draw->vertex_shader->state->output_semantic_name[0]
+          == TGSI_SEMANTIC_POSITION);
+
+   /* Consts does not require 16 byte alignment. */
+   machine->Consts = (float (*)[4]) draw->user.constants;
+
+   machine->Inputs = ALIGN16_ASSIGN(inputs);
+   machine->Outputs = ALIGN16_ASSIGN(outputs);
+
+   draw->vertex_fetch.fetch_func( draw, machine, elts, count );
+
+   /* run shader */
+   gallivm_cpu_vs_exec(shader->llvm_prog,
+                       machine->Inputs,
+                       machine->Outputs,
+                       machine->Consts,
+                       machine->Temps);
+
+   /* store machine results */
+   for (j = 0; j < count; j++) {
+      unsigned slot;
+      float x, y, z, w;
+
+      x = vOut[j]->clip[0] = machine->Outputs[0].xyzw[0].f[j];
+      y = vOut[j]->clip[1] = machine->Outputs[0].xyzw[1].f[j];
+      z = vOut[j]->clip[2] = machine->Outputs[0].xyzw[2].f[j];
+      w = vOut[j]->clip[3] = machine->Outputs[0].xyzw[3].f[j];
+
+      vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, draw->nr_planes);
+      vOut[j]->edgeflag = 1;
+
+      /* divide by w */
+      w = 1.0f / w;
+      x *= w;
+      y *= w;
+      z *= w;
+
+      /* Viewport mapping */
+      vOut[j]->data[0][0] = x * scale[0] + trans[0];
+      vOut[j]->data[0][1] = y * scale[1] + trans[1];
+      vOut[j]->data[0][2] = z * scale[2] + trans[2];
+      vOut[j]->data[0][3] = w;
+
+      /* Remaining attributes are packed into sequential post-transform
+       * vertex attrib slots.
+       */
+      for (slot = 1; slot < draw->num_vs_outputs; slot++) {
+         vOut[j]->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
+         vOut[j]->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
+         vOut[j]->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
+         vOut[j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
+      }
+   } /* loop over vertices */
+}
+
+static void
+vs_llvm_delete( struct draw_vertex_shader *base )
+{
+   struct draw_llvm_vertex_shader *shader = 
+      (struct draw_llvm_vertex_shader *)base;
+
+   /* Do something to free compiled shader:
+    */
+
+   FREE( shader );
+}
+
+
+
+
+struct draw_vertex_shader *
+draw_create_vs_llvm(struct draw_context *draw,
+		    const struct pipe_shader_state *templ)
+{
+   struct draw_llvm_vertex_shader *vs;
+
+   vs = CALLOC_STRUCT( draw_llvm_vertex_shader );
+   if (vs == NULL) 
+      return NULL;
+
+   vs->base.state = templ;
+   vs->base.prepare = vs_llvm_prepare;
+   vs->base.run = vs_llvm_run;
+   vs->base.delete = vs_llvm_delete;
+
+   {
+      struct gallivm_ir *ir = gallivm_ir_new(GALLIVM_VS);
+      gallivm_ir_set_layout(ir, GALLIVM_SOA);
+      gallivm_ir_set_components(ir, 4);
+      gallivm_ir_fill_from_tgsi(ir, vs->base.state->tokens);
+      vs->llvm_prog = gallivm_ir_compile(ir);
+      gallivm_ir_delete(ir);
+   }
+
+   draw->engine = gallivm_global_cpu_engine();
+
+   /* XXX: Why are there two versions of this?  Shouldn't creating the
+    *      engine be a separate operation to compiling a shader?
+    */
+   if (!draw->engine) {
+      draw->engine = gallivm_cpu_engine_create(vs->llvm_prog);
+   }
+   else {
+      gallivm_cpu_jit_compile(draw->engine, vs->llvm_prog);
+   }
+
+   return &vs->base;
+}
+
+
+
+
+
+#else
+
+struct draw_vertex_shader *
+draw_create_vs_llvm(struct draw_context *draw,
+                          const struct pipe_shader_state *shader)
+{
+   return NULL;
+}
+
+#endif
diff --git a/src/gallium/auxiliary/draw/draw_vs_sse.c b/src/gallium/auxiliary/draw/draw_vs_sse.c
new file mode 100644
index 0000000000..04349cb404
--- /dev/null
+++ b/src/gallium/auxiliary/draw/draw_vs_sse.c
@@ -0,0 +1,251 @@
+/**************************************************************************
+ * 
+ * Copyright 2007 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  *   Brian Paul
+  */
+
+#include "draw_vs.h"
+
+#if defined(__i386__) || defined(__386__)
+
+#include "pipe/p_util.h"
+#include "pipe/p_shader_tokens.h"
+
+#include "draw_private.h"
+#include "draw_context.h"
+
+#include "x86/rtasm/x86sse.h"
+#include "tgsi/exec/tgsi_sse2.h"
+
+
+typedef void (XSTDCALL *codegen_function) (
+   const struct tgsi_exec_vector *input,
+   struct tgsi_exec_vector *output,
+   float (*constant)[4],
+   struct tgsi_exec_vector *temporary );
+
+
+struct draw_sse_vertex_shader {
+   struct draw_vertex_shader base;
+   struct x86_function sse2_program;
+   codegen_function func;
+};
+
+
+/* Should be part of the generated shader:
+ */
+static INLINE unsigned
+compute_clipmask(const float *clip, /*const*/ float plane[][4], unsigned nr)
+{
+   unsigned mask = 0;
+   unsigned i;
+
+   /* Do the hardwired planes first:
+    */
+   if (-clip[0] + clip[3] < 0) mask |= CLIP_RIGHT_BIT;
+   if ( clip[0] + clip[3] < 0) mask |= CLIP_LEFT_BIT;
+   if (-clip[1] + clip[3] < 0) mask |= CLIP_TOP_BIT;
+   if ( clip[1] + clip[3] < 0) mask |= CLIP_BOTTOM_BIT;
+   if (-clip[2] + clip[3] < 0) mask |= CLIP_FAR_BIT;
+   if ( clip[2] + clip[3] < 0) mask |= CLIP_NEAR_BIT;
+
+   /* Followed by any remaining ones:
+    */
+   for (i = 6; i < nr; i++) {
+      if (dot4(clip, plane[i]) < 0) 
+         mask |= (1<<i);
+   }
+
+   return mask;
+}
+
+
+static void
+vs_sse_prepare( struct draw_vertex_shader *base,
+		struct draw_context *draw )
+{
+   draw_update_vertex_fetch( draw );
+}
+
+/**
+ * Transform vertices with the current vertex program/shader
+ * Up to four vertices can be shaded at a time.
+ * \param vbuffer  the input vertex data
+ * \param elts  indexes of four input vertices
+ * \param count  number of vertices to shade [1..4]
+ * \param vOut  array of pointers to four output vertices
+ */
+static void
+vs_sse_run( struct draw_vertex_shader *base,
+	    struct draw_context *draw, 
+	    const unsigned *elts, 
+	    unsigned count,
+	    struct vertex_header *vOut[] )
+{
+   struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base;
+   struct tgsi_exec_machine *machine = &draw->machine;
+   unsigned int j;
+
+   ALIGN16_DECL(struct tgsi_exec_vector, inputs, PIPE_ATTRIB_MAX);
+   ALIGN16_DECL(struct tgsi_exec_vector, outputs, PIPE_ATTRIB_MAX);
+   const float *scale = draw->viewport.scale;
+   const float *trans = draw->viewport.translate;
+
+   assert(count <= 4);
+   assert(draw->vertex_shader->state->output_semantic_name[0]
+          == TGSI_SEMANTIC_POSITION);
+
+   /* Consts does not require 16 byte alignment. */
+   machine->Consts = (float (*)[4]) draw->user.constants;
+   machine->Inputs = ALIGN16_ASSIGN(inputs);
+   machine->Outputs = ALIGN16_ASSIGN(outputs);
+
+
+   /* Fetch vertices.  This may at some point be integrated into the
+    * compiled shader -- that would require a reorganization where
+    * multiple versions of the compiled shader might exist,
+    * specialized for each fetch state.
+    */
+   draw->vertex_fetch.fetch_func( draw, machine, elts, count );
+
+
+   /* run compiled shader
+    */   
+   shader->func(
+      machine->Inputs,
+      machine->Outputs,
+      machine->Consts,
+      machine->Temps );
+
+
+   /* XXX: Computing the clipmask and emitting results should be done
+    *      in the vertex program as a set of instructions appended to
+    *      the user-provided code.
+    */
+   for (j = 0; j < count; j++) {
+      unsigned slot;
+      float x, y, z, w;
+
+      x = vOut[j]->clip[0] = machine->Outputs[0].xyzw[0].f[j];
+      y = vOut[j]->clip[1] = machine->Outputs[0].xyzw[1].f[j];
+      z = vOut[j]->clip[2] = machine->Outputs[0].xyzw[2].f[j];
+      w = vOut[j]->clip[3] = machine->Outputs[0].xyzw[3].f[j];
+
+      vOut[j]->clipmask = compute_clipmask(vOut[j]->clip, draw->plane, draw->nr_planes);
+      vOut[j]->edgeflag = 1;
+
+      /* divide by w */
+      w = 1.0f / w;
+      x *= w;
+      y *= w;
+      z *= w;
+
+      /* Viewport mapping */
+      vOut[j]->data[0][0] = x * scale[0] + trans[0];
+      vOut[j]->data[0][1] = y * scale[1] + trans[1];
+      vOut[j]->data[0][2] = z * scale[2] + trans[2];
+      vOut[j]->data[0][3] = w;
+
+      /* Remaining attributes are packed into sequential post-transform
+       * vertex attrib slots.
+       */
+      for (slot = 1; slot < draw->num_vs_outputs; slot++) {
+         vOut[j]->data[slot][0] = machine->Outputs[slot].xyzw[0].f[j];
+         vOut[j]->data[slot][1] = machine->Outputs[slot].xyzw[1].f[j];
+         vOut[j]->data[slot][2] = machine->Outputs[slot].xyzw[2].f[j];
+         vOut[j]->data[slot][3] = machine->Outputs[slot].xyzw[3].f[j];
+      }
+   } 
+}
+
+
+
+static void
+vs_sse_delete( struct draw_vertex_shader *base )
+{
+   struct draw_sse_vertex_shader *shader = (struct draw_sse_vertex_shader *)base;
+   
+   x86_release_func( &shader->sse2_program );
+
+   FREE( shader );
+}
+
+
+struct draw_vertex_shader *
+draw_create_vs_sse(struct draw_context *draw,
+                          const struct pipe_shader_state *templ)
+{
+   struct draw_sse_vertex_shader *vs;
+
+   if (!draw->use_sse) 
+      return NULL;
+
+   vs = CALLOC_STRUCT( draw_sse_vertex_shader );
+   if (vs == NULL) 
+      return NULL;
+
+   vs->base.state = templ;
+   vs->base.prepare = vs_sse_prepare;
+   vs->base.run = vs_sse_run;
+   vs->base.delete = vs_sse_delete;
+   
+   x86_init_func( &vs->sse2_program );
+
+   if (!tgsi_emit_sse2( (struct tgsi_token *) vs->base.state->tokens,
+			&vs->sse2_program )) 
+      goto fail;
+      
+   vs->func = (codegen_function) x86_get_func( &vs->sse2_program );
+   
+   return &vs->base;
+
+fail:
+   fprintf(stderr, "tgsi_emit_sse2() failed, falling back to interpreter\n");
+
+   x86_release_func( &vs->sse2_program );
+   
+   FREE(vs);
+   return NULL;
+}
+
+
+
+#else
+
+struct draw_vertex_shader *
+draw_create_vs_sse( struct draw_context *draw,
+		    const struct pipe_shader_state *templ )
+{
+   return NULL;
+}
+
+
+#endif
+