merge of glsl-compiler-1 branch

19 files changed, 2237 insertions, 501 deletions

diff --git a/src/mesa/tnl/t_context.c b/src/mesa/tnl/t_context.c
index d9458b74eca..fa42a3df988 100644
--- a/src/mesa/tnl/t_context.c
+++ b/src/mesa/tnl/t_context.c
@@ -60,7 +60,7 @@ _tnl_CreateContext( GLcontext *ctx )
 
    /* Initialize tnl state.
     */
-   if (ctx->_MaintainTnlProgram) {
+   if (ctx->VertexProgram._MaintainTnlProgram) {
       _tnl_ProgramCacheInit( ctx );
       _tnl_install_pipeline( ctx, _tnl_vp_pipeline );
    } else {
@@ -90,7 +90,7 @@ _tnl_DestroyContext( GLcontext *ctx )
 
    _tnl_destroy_pipeline( ctx );
 
-   if (ctx->_MaintainTnlProgram)
+   if (ctx->VertexProgram._MaintainTnlProgram)
       _tnl_ProgramCacheDestroy( ctx );
 
    FREE(tnl);
@@ -102,6 +102,8 @@ void
 _tnl_InvalidateState( GLcontext *ctx, GLuint new_state )
 {
    TNLcontext *tnl = TNL_CONTEXT(ctx);
+   const struct gl_vertex_program *vp = ctx->VertexProgram._Current;
+   const struct gl_fragment_program *fp = ctx->FragmentProgram._Current;
 
    if (new_state & (_NEW_HINT)) {
       ASSERT(tnl->AllowVertexFog || tnl->AllowPixelFog);
@@ -118,7 +120,9 @@ _tnl_InvalidateState( GLcontext *ctx, GLuint new_state )
 
       RENDERINPUTS_ZERO( tnl->render_inputs_bitset );
       RENDERINPUTS_SET( tnl->render_inputs_bitset, _TNL_ATTRIB_POS );
-      RENDERINPUTS_SET( tnl->render_inputs_bitset, _TNL_ATTRIB_COLOR0 );
+      if (!fp || fp->Base.InputsRead & FRAG_BIT_COL0) {
+         RENDERINPUTS_SET( tnl->render_inputs_bitset, _TNL_ATTRIB_COLOR0 );
+      }
       for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
          if (ctx->Texture._EnabledCoordUnits & (1 << i)) {
             RENDERINPUTS_SET( tnl->render_inputs_bitset, _TNL_ATTRIB_TEX(i) );
@@ -134,7 +138,7 @@ _tnl_InvalidateState( GLcontext *ctx, GLuint new_state )
    }
 
    if (ctx->Fog.Enabled ||
-       (ctx->FragmentProgram._Active &&
+       (ctx->FragmentProgram._Current &&
         (ctx->FragmentProgram._Current->FogOption != GL_NONE ||
          ctx->FragmentProgram._Current->Base.InputsRead & FRAG_BIT_FOGC)))
       RENDERINPUTS_SET( tnl->render_inputs_bitset, _TNL_ATTRIB_FOG );
@@ -150,8 +154,16 @@ _tnl_InvalidateState( GLcontext *ctx, GLuint new_state )
        (ctx->VertexProgram._Enabled && ctx->VertexProgram.PointSizeEnabled))
       RENDERINPUTS_SET( tnl->render_inputs_bitset, _TNL_ATTRIB_POINTSIZE );
 
-   if (ctx->ShaderObjects._VertexShaderPresent || ctx->ShaderObjects._FragmentShaderPresent)
-      RENDERINPUTS_SET_RANGE( tnl->render_inputs_bitset, _TNL_FIRST_GENERIC, _TNL_LAST_GENERIC );
+   /* check for varying vars which are written by the vertex program */
+   if (vp) {
+      GLuint i;
+      for (i = 0; i < MAX_VARYING; i++) {
+         if (vp->Base.OutputsWritten & (1 << (VERT_RESULT_VAR0 + i))) {
+            RENDERINPUTS_SET(tnl->render_inputs_bitset,
+                             _TNL_ATTRIB_GENERIC(i));
+         }
+      }
+   }
 }
 
 
diff --git a/src/mesa/tnl/t_context.h b/src/mesa/tnl/t_context.h
index a872f261775..31b89aca41f 100644
--- a/src/mesa/tnl/t_context.h
+++ b/src/mesa/tnl/t_context.h
@@ -217,7 +217,6 @@ struct vertex_buffer
    GLvector4f  *ColorPtr[2];	                /* _TNL_BIT_COLOR0 */
    GLvector4f  *SecondaryColorPtr[2];           /* _TNL_BIT_COLOR1 */
    GLvector4f  *FogCoordPtr;	                /* _TNL_BIT_FOG */
-   GLvector4f  *VaryingPtr[MAX_VARYING_VECTORS];
 
    const struct _mesa_prim  *Primitive;	              
    GLuint      PrimitiveCount;	      
diff --git a/src/mesa/tnl/t_pipeline.c b/src/mesa/tnl/t_pipeline.c
index 2efe701a802..a50a3f0f2f2 100644
--- a/src/mesa/tnl/t_pipeline.c
+++ b/src/mesa/tnl/t_pipeline.c
@@ -132,7 +132,7 @@ void _tnl_run_pipeline( GLcontext *ctx )
     * (ie const or non-const).
     */
    if (check_input_changes( ctx ) || tnl->pipeline.new_state) {
-      if (ctx->_MaintainTnlProgram)
+      if (ctx->VertexProgram._MaintainTnlProgram)
 	 _tnl_UpdateFixedFunctionProgram( ctx );
 
       for (i = 0; i < tnl->pipeline.nr_stages ; i++) {
@@ -208,9 +208,6 @@ const struct tnl_pipeline_stage *_tnl_default_pipeline[] = {
    &_tnl_arb_vertex_program_stage,
    &_tnl_vertex_program_stage, 
 #endif
-#if FEATURE_ARB_vertex_shader
-   &_tnl_arb_vertex_shader_stage,
-#endif
    &_tnl_render_stage,
    NULL 
 };
diff --git a/src/mesa/tnl/t_save_api.c b/src/mesa/tnl/t_save_api.c
new file mode 100644
index 00000000000..b08f05374e4
--- /dev/null
+++ b/src/mesa/tnl/t_save_api.c
@@ -0,0 +1,1834 @@
+/**************************************************************************
+
+Copyright 2002 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
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell <[email protected]>
+ */
+
+
+
+/**
+ * The display list compiler attempts to store lists of vertices with the
+ * same vertex layout.  Additionally it attempts to minimize the need
+ * for execute-time fixup of these vertex lists, allowing them to be
+ * cached on hardware.
+ *
+ * There are still some circumstances where this can be thwarted, for
+ * example by building a list that consists of one very long primitive
+ * (eg Begin(Triangles), 1000 vertices, End), and calling that list
+ * from inside a different begin/end object (Begin(Lines), CallList,
+ * End).  
+ *
+ * In that case the code will have to replay the list as individual
+ * commands through the Exec dispatch table, or fix up the copied
+ * vertices at execute-time.
+ *
+ * The other case where fixup is required is when a vertex attribute
+ * is introduced in the middle of a primitive.  Eg:
+ *  Begin(Lines)
+ *  TexCoord1f()           Vertex2f()
+ *  TexCoord1f() Color3f() Vertex2f()
+ *  End()
+ *
+ *  If the current value of Color isn't known at compile-time, this
+ *  primitive will require fixup.
+ *
+ *
+ * The list compiler currently doesn't attempt to compile lists
+ * containing EvalCoord or EvalPoint commands.  On encountering one of
+ * these, compilation falls back to opcodes.  
+ *
+ * This could be improved to fallback only when a mix of EvalCoord and
+ * Vertex commands are issued within a single primitive.
+ */
+
+
+#include "glheader.h"
+#include "context.h"
+#include "dlist.h"
+#include "enums.h"
+#include "macros.h"
+#include "api_validate.h"
+#include "api_arrayelt.h"
+#include "vtxfmt.h"
+#include "t_save_api.h"
+#include "dispatch.h"
+
+/*
+ * NOTE: Old 'parity' issue is gone, but copying can still be
+ * wrong-footed on replay.
+ */
+static GLuint _save_copy_vertices( GLcontext *ctx, 
+				   const struct tnl_vertex_list *node )
+{
+   TNLcontext *tnl = TNL_CONTEXT( ctx );
+   const struct tnl_prim *prim = &node->prim[node->prim_count-1];
+   GLuint nr = prim->count;
+   GLuint sz = tnl->save.vertex_size;
+   const GLfloat *src = node->buffer + prim->start * sz;
+   GLfloat *dst = tnl->save.copied.buffer;
+   GLuint ovf, i;
+
+   if (prim->mode & PRIM_END)
+      return 0;
+	 
+   switch( prim->mode & PRIM_MODE_MASK )
+   {
+   case GL_POINTS:
+      return 0;
+   case GL_LINES:
+      ovf = nr&1;
+      for (i = 0 ; i < ovf ; i++)
+	 _mesa_memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz*sizeof(GLfloat) );
+      return i;
+   case GL_TRIANGLES:
+      ovf = nr%3;
+      for (i = 0 ; i < ovf ; i++)
+	 _mesa_memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz*sizeof(GLfloat) );
+      return i;
+   case GL_QUADS:
+      ovf = nr&3;
+      for (i = 0 ; i < ovf ; i++)
+	 _mesa_memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz*sizeof(GLfloat) );
+      return i;
+   case GL_LINE_STRIP:
+      if (nr == 0) 
+	 return 0;
+      else {
+	 _mesa_memcpy( dst, src+(nr-1)*sz, sz*sizeof(GLfloat) );
+	 return 1;
+      }
+   case GL_LINE_LOOP:
+   case GL_TRIANGLE_FAN:
+   case GL_POLYGON:
+      if (nr == 0) 
+	 return 0;
+      else if (nr == 1) {
+	 _mesa_memcpy( dst, src+0, sz*sizeof(GLfloat) );
+	 return 1;
+      } else {
+	 _mesa_memcpy( dst, src+0, sz*sizeof(GLfloat) );
+	 _mesa_memcpy( dst+sz, src+(nr-1)*sz, sz*sizeof(GLfloat) );
+	 return 2;
+      }
+   case GL_TRIANGLE_STRIP:
+   case GL_QUAD_STRIP:
+      switch (nr) {
+      case 0: ovf = 0; break;
+      case 1: ovf = 1; break;
+      default: ovf = 2 + (nr&1); break;
+      }
+      for (i = 0 ; i < ovf ; i++)
+	 _mesa_memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz*sizeof(GLfloat) );
+      return i;
+   default:
+      assert(0);
+      return 0;
+   }
+}
+
+
+static void
+build_normal_lengths( struct tnl_vertex_list *node )
+{
+   GLuint i;
+   GLfloat *len;
+   GLfloat *n = node->buffer;
+   GLuint stride = node->vertex_size;
+   GLuint count = node->count;
+
+   len = node->normal_lengths = (GLfloat *) MALLOC( count * sizeof(GLfloat) );
+   if (!len)
+      return;
+
+   /* Find the normal of the first vertex:
+    */
+   for (i = 0 ; i < _TNL_ATTRIB_NORMAL ; i++) 
+      n += node->attrsz[i];
+
+   for (i = 0 ; i < count ; i++, n += stride) {
+      len[i] = LEN_3FV( n );
+      if (len[i] > 0.0F) len[i] = 1.0F / len[i];
+   } 
+}
+
+static struct tnl_vertex_store *alloc_vertex_store( GLcontext *ctx )
+{
+   struct tnl_vertex_store *store = MALLOC_STRUCT(tnl_vertex_store);
+   (void) ctx;
+   store->used = 0;
+   store->refcount = 1;
+   return store;
+}
+
+static struct tnl_primitive_store *alloc_prim_store( GLcontext *ctx )
+{
+   struct tnl_primitive_store *store = MALLOC_STRUCT(tnl_primitive_store);
+   (void) ctx;
+   store->used = 0;
+   store->refcount = 1;
+   return store;
+}
+
+static void _save_reset_counters( GLcontext *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+
+   tnl->save.prim = tnl->save.prim_store->buffer + tnl->save.prim_store->used;
+   tnl->save.buffer = (tnl->save.vertex_store->buffer + 
+		       tnl->save.vertex_store->used);
+
+   if (tnl->save.vertex_size)
+      tnl->save.initial_counter = ((SAVE_BUFFER_SIZE - 
+				    tnl->save.vertex_store->used) / 
+				   tnl->save.vertex_size);
+   else
+      tnl->save.initial_counter = 0;
+
+   if (tnl->save.initial_counter > ctx->Const.MaxArrayLockSize )
+      tnl->save.initial_counter = ctx->Const.MaxArrayLockSize;
+
+   tnl->save.counter = tnl->save.initial_counter;
+   tnl->save.prim_count = 0;
+   tnl->save.prim_max = SAVE_PRIM_SIZE - tnl->save.prim_store->used;
+   tnl->save.copied.nr = 0;
+   tnl->save.dangling_attr_ref = 0;
+}
+
+
+/* Insert the active immediate struct onto the display list currently
+ * being built.
+ */
+static void _save_compile_vertex_list( GLcontext *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   struct tnl_vertex_list *node;
+
+   /* Allocate space for this structure in the display list currently
+    * being compiled.
+    */
+   node = (struct tnl_vertex_list *)
+      _mesa_alloc_instruction(ctx, tnl->save.opcode_vertex_list, sizeof(*node));
+
+   if (!node)
+      return;
+
+   /* Duplicate our template, increment refcounts to the storage structs:
+    */
+   _mesa_memcpy(node->attrsz, tnl->save.attrsz, sizeof(node->attrsz));
+   node->vertex_size = tnl->save.vertex_size;
+   node->buffer = tnl->save.buffer;
+   node->count = tnl->save.initial_counter - tnl->save.counter;
+   node->wrap_count = tnl->save.copied.nr;
+   node->have_materials = tnl->save.have_materials;
+   node->dangling_attr_ref = tnl->save.dangling_attr_ref;
+   node->normal_lengths = NULL;
+   node->prim = tnl->save.prim;
+   node->prim_count = tnl->save.prim_count;
+   node->vertex_store = tnl->save.vertex_store;
+   node->prim_store = tnl->save.prim_store;
+
+   node->vertex_store->refcount++;
+   node->prim_store->refcount++;
+
+   assert(node->attrsz[_TNL_ATTRIB_POS] != 0 ||
+	  node->count == 0);
+
+   if (tnl->save.dangling_attr_ref)
+      ctx->ListState.CurrentList->flags |= MESA_DLIST_DANGLING_REFS;
+
+   /* Maybe calculate normal lengths:
+    */
+   if (tnl->CalcDListNormalLengths && 
+       node->attrsz[_TNL_ATTRIB_NORMAL] == 3 &&
+       !(ctx->ListState.CurrentList->flags & MESA_DLIST_DANGLING_REFS))
+      build_normal_lengths( node );
+
+
+   tnl->save.vertex_store->used += tnl->save.vertex_size * node->count;
+   tnl->save.prim_store->used += node->prim_count;
+
+   /* Decide whether the storage structs are full, or can be used for
+    * the next vertex lists as well.
+    */
+   if (tnl->save.vertex_store->used > 
+       SAVE_BUFFER_SIZE - 16 * (tnl->save.vertex_size + 4)) {
+
+      tnl->save.vertex_store->refcount--; 
+      assert(tnl->save.vertex_store->refcount != 0);
+      tnl->save.vertex_store = alloc_vertex_store( ctx );
+      tnl->save.vbptr = tnl->save.vertex_store->buffer;
+   } 
+
+   if (tnl->save.prim_store->used > SAVE_PRIM_SIZE - 6) {
+      tnl->save.prim_store->refcount--; 
+      assert(tnl->save.prim_store->refcount != 0);
+      tnl->save.prim_store = alloc_prim_store( ctx );
+   } 
+
+   /* Reset our structures for the next run of vertices:
+    */
+   _save_reset_counters( ctx );
+
+   /* Copy duplicated vertices 
+    */
+   tnl->save.copied.nr = _save_copy_vertices( ctx, node );
+
+
+   /* Deal with GL_COMPILE_AND_EXECUTE:
+    */
+   if (ctx->ExecuteFlag) {
+      _tnl_playback_vertex_list( ctx, (void *) node );
+   }
+}
+
+
+/* TODO -- If no new vertices have been stored, don't bother saving
+ * it.
+ */
+static void _save_wrap_buffers( GLcontext *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   GLint i = tnl->save.prim_count - 1;
+   GLenum mode;
+
+   assert(i < (GLint) tnl->save.prim_max);
+   assert(i >= 0);
+
+   /* Close off in-progress primitive.
+    */
+   tnl->save.prim[i].count = ((tnl->save.initial_counter - tnl->save.counter) - 
+			     tnl->save.prim[i].start);
+   mode = tnl->save.prim[i].mode & ~(PRIM_BEGIN|PRIM_END);
+   
+   /* store the copied vertices, and allocate a new list.
+    */
+   _save_compile_vertex_list( ctx );
+
+   /* Restart interrupted primitive
+    */
+   tnl->save.prim[0].mode = mode;
+   tnl->save.prim[0].start = 0;
+   tnl->save.prim[0].count = 0;
+   tnl->save.prim_count = 1;
+}
+
+
+
+/* Called only when buffers are wrapped as the result of filling the
+ * vertex_store struct.  
+ */
+static void _save_wrap_filled_vertex( GLcontext *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   GLfloat *data = tnl->save.copied.buffer;
+   GLuint i;
+
+   /* Emit a glEnd to close off the last vertex list.
+    */
+   _save_wrap_buffers( ctx );
+   
+    /* Copy stored stored vertices to start of new list.
+    */
+   assert(tnl->save.counter > tnl->save.copied.nr);
+
+   for (i = 0 ; i < tnl->save.copied.nr ; i++) {
+      _mesa_memcpy( tnl->save.vbptr, data, tnl->save.vertex_size * sizeof(GLfloat));
+      data += tnl->save.vertex_size;
+      tnl->save.vbptr += tnl->save.vertex_size;
+      tnl->save.counter--;
+   }
+}
+
+
+static void _save_copy_to_current( GLcontext *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx); 
+   GLuint i;
+
+   /* XXX Use _TNL_FIRST_* and _TNL_LAST_* values instead? */
+   for (i = _TNL_ATTRIB_POS+1 ; i <= _TNL_ATTRIB_EDGEFLAG ; i++) {
+      if (tnl->save.attrsz[i]) {
+	 tnl->save.currentsz[i][0] = tnl->save.attrsz[i];
+	 COPY_CLEAN_4V(tnl->save.current[i], 
+		    tnl->save.attrsz[i], 
+		    tnl->save.attrptr[i]);
+      }
+   }
+
+   /* Edgeflag requires special treatment: 
+    *
+    * TODO: change edgeflag to GLfloat in Mesa.
+    */
+   if (tnl->save.attrsz[_TNL_ATTRIB_EDGEFLAG]) {
+      ctx->ListState.ActiveEdgeFlag = 1;
+      tnl->save.CurrentFloatEdgeFlag = 
+	 tnl->save.attrptr[_TNL_ATTRIB_EDGEFLAG][0];
+      ctx->ListState.CurrentEdgeFlag = 
+	 (tnl->save.CurrentFloatEdgeFlag == 1.0);
+   }
+}
+
+
+static void _save_copy_from_current( GLcontext *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx); 
+   GLint i;
+
+   for (i = _TNL_ATTRIB_POS+1 ; i <= _TNL_ATTRIB_EDGEFLAG ; i++) 
+      switch (tnl->save.attrsz[i]) {
+      case 4: tnl->save.attrptr[i][3] = tnl->save.current[i][3];
+      case 3: tnl->save.attrptr[i][2] = tnl->save.current[i][2];
+      case 2: tnl->save.attrptr[i][1] = tnl->save.current[i][1];
+      case 1: tnl->save.attrptr[i][0] = tnl->save.current[i][0];
+      case 0: break;
+      }
+
+   /* Edgeflag requires special treatment:
+    */
+   if (tnl->save.attrsz[_TNL_ATTRIB_EDGEFLAG]) {
+      tnl->save.CurrentFloatEdgeFlag = (GLfloat)ctx->ListState.CurrentEdgeFlag;
+      tnl->save.attrptr[_TNL_ATTRIB_EDGEFLAG][0] = tnl->save.CurrentFloatEdgeFlag;
+   }
+}
+
+
+
+
+/* Flush existing data, set new attrib size, replay copied vertices.
+ */ 
+static void _save_upgrade_vertex( GLcontext *ctx, 
+				 GLuint attr,
+				 GLuint newsz )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   GLuint oldsz;
+   GLuint i;
+   GLfloat *tmp;
+
+   /* Store the current run of vertices, and emit a GL_END.  Emit a
+    * BEGIN in the new buffer.
+    */
+   if (tnl->save.initial_counter != tnl->save.counter) 
+      _save_wrap_buffers( ctx );
+   else
+      assert( tnl->save.copied.nr == 0 );
+
+   /* Do a COPY_TO_CURRENT to ensure back-copying works for the case
+    * when the attribute already exists in the vertex and is having
+    * its size increased.  
+    */
+   _save_copy_to_current( ctx );
+
+   /* Fix up sizes:
+    */
+   oldsz = tnl->save.attrsz[attr];
+   tnl->save.attrsz[attr] = newsz;
+
+   tnl->save.vertex_size += newsz - oldsz;
+   tnl->save.counter = ((SAVE_BUFFER_SIZE - tnl->save.vertex_store->used) / 
+			tnl->save.vertex_size);
+   if (tnl->save.counter > ctx->Const.MaxArrayLockSize )
+      tnl->save.counter = ctx->Const.MaxArrayLockSize;
+   tnl->save.initial_counter = tnl->save.counter;
+
+   /* Recalculate all the attrptr[] values:
+    */
+   for (i = 0, tmp = tnl->save.vertex ; i < _TNL_ATTRIB_MAX ; i++) {
+      if (tnl->save.attrsz[i]) {
+	 tnl->save.attrptr[i] = tmp;
+	 tmp += tnl->save.attrsz[i];
+      }
+      else 
+	 tnl->save.attrptr[i] = NULL; /* will not be dereferenced. */
+   }
+
+   /* Copy from current to repopulate the vertex with correct values.
+    */
+   _save_copy_from_current( ctx );
+
+   /* Replay stored vertices to translate them to new format here.
+    *
+    * If there are copied vertices and the new (upgraded) attribute
+    * has not been defined before, this list is somewhat degenerate,
+    * and will need fixup at runtime.
+    */
+   if (tnl->save.copied.nr)
+   {
+      GLfloat *data = tnl->save.copied.buffer;
+      GLfloat *dest = tnl->save.buffer;
+      GLuint j;
+
+      /* Need to note this and fix up at runtime (or loopback):
+       */
+      if (tnl->save.currentsz[attr][0] == 0) {
+	 assert(oldsz == 0);
+	 tnl->save.dangling_attr_ref = GL_TRUE;
+
+/* 	 _mesa_debug(NULL, "_save_upgrade_vertex: dangling reference attr %d\n",  */
+/* 		     attr);  */
+
+#if 0
+	 /* The current strategy is to punt these degenerate cases
+	  * through _tnl_loopback_vertex_list(), a lower-performance
+	  * option.  To minimize the impact of this, artificially
+	  * reduce the size of this vertex_list.
+	  */
+	 if (t->save.counter > 10) {
+	    t->save.initial_counter = 10;
+	    t->save.counter = 10;
+	 }
+#endif
+      }
+
+      for (i = 0 ; i < tnl->save.copied.nr ; i++) {
+	 for (j = 0 ; j < _TNL_ATTRIB_MAX ; j++) {
+	    if (tnl->save.attrsz[j]) {
+	       if (j == attr) {
+		  if (oldsz) {
+		     COPY_CLEAN_4V( dest, oldsz, data );
+		     data += oldsz;
+		     dest += newsz;
+		  }
+		  else {
+		     COPY_SZ_4V( dest, newsz, tnl->save.current[attr] );
+		     dest += newsz;
+		  }
+	       }
+	       else {
+		  GLint sz = tnl->save.attrsz[j];
+		  COPY_SZ_4V( dest, sz, data );
+		  data += sz;
+		  dest += sz;
+	       }
+	    }
+	 }
+      }
+
+      tnl->save.vbptr = dest;
+      tnl->save.counter -= tnl->save.copied.nr;
+   }
+}
+
+
+
+
+/* Helper function for 'CHOOSE' macro.  Do what's necessary when an
+ * entrypoint is called for the first time.
+ */
+static void do_choose( GLuint attr, GLuint sz, 
+		       void (*attr_func)( const GLfloat *),
+		       void (*choose1)( const GLfloat *),
+		       void (*choose2)( const GLfloat *),
+		       void (*choose3)( const GLfloat *),
+		       void (*choose4)( const GLfloat *),
+		       const GLfloat *v )
+{ 
+   GET_CURRENT_CONTEXT( ctx ); 
+   TNLcontext *tnl = TNL_CONTEXT(ctx); 
+   static GLfloat id[4] = { 0, 0, 0, 1 };
+   int i;
+
+   if (tnl->save.attrsz[attr] < sz) {
+      /* New size is larger.  Need to flush existing vertices and get
+       * an enlarged vertex format.
+       */
+      _save_upgrade_vertex( ctx, attr, sz );
+   }
+   else {
+      /* New size is equal or smaller - just need to fill in some
+       * zeros.
+       */
+      for (i = sz ; i <= tnl->save.attrsz[attr] ; i++)
+	 tnl->save.attrptr[attr][i-1] = id[i-1];
+   }
+
+   /* Reset any active pointers for this attribute 
+    */
+   tnl->save.tabfv[attr][0] = choose1;
+   tnl->save.tabfv[attr][1] = choose2;
+   tnl->save.tabfv[attr][2] = choose3;
+   tnl->save.tabfv[attr][3] = choose4;
+
+   /* Update the secondary dispatch table with the new function
+    */
+   tnl->save.tabfv[attr][sz-1] = attr_func;
+
+   (*attr_func)(v);
+}
+
+
+
+/* Only one size for each attribute may be active at once.  Eg. if
+ * Color3f is installed/active, then Color4f may not be, even if the
+ * vertex actually contains 4 color coordinates.  This is because the
+ * 3f version won't otherwise set color[3] to 1.0 -- this is the job
+ * of the chooser function when switching between Color4f and Color3f.
+ */
+#define ATTRFV( ATTR, N )					\
+static void save_choose_##ATTR##_##N( const GLfloat *v );	\
+								\
+static void save_attrib_##ATTR##_##N( const GLfloat *v )	\
+{								\
+   GET_CURRENT_CONTEXT( ctx );					\
+   TNLcontext *tnl = TNL_CONTEXT(ctx);				\
+								\
+   if ((ATTR) == 0) {						\
+      GLuint i;							\
+								\
+      if (N>0) tnl->save.vbptr[0] = v[0];			\
+      if (N>1) tnl->save.vbptr[1] = v[1];			\
+      if (N>2) tnl->save.vbptr[2] = v[2];			\
+      if (N>3) tnl->save.vbptr[3] = v[3];			\
+								\
+      for (i = N; i < tnl->save.vertex_size; i++)		\
+	 tnl->save.vbptr[i] = tnl->save.vertex[i];		\
+								\
+      tnl->save.vbptr += tnl->save.vertex_size;			\
+								\
+      if (--tnl->save.counter == 0)				\
+	 _save_wrap_filled_vertex( ctx );			\
+   }								\
+   else {							\
+      GLfloat *dest = tnl->save.attrptr[ATTR];			\
+      if (N>0) dest[0] = v[0];					\
+      if (N>1) dest[1] = v[1];					\
+      if (N>2) dest[2] = v[2];					\
+      if (N>3) dest[3] = v[3];					\
+   }								\
+}
+
+#define CHOOSE( ATTR, N )					\
+static void save_choose_##ATTR##_##N( const GLfloat *v )	\
+{								\
+   do_choose(ATTR, N,						\
+	     save_attrib_##ATTR##_##N,				\
+	     save_choose_##ATTR##_1,				\
+	     save_choose_##ATTR##_2,				\
+	     save_choose_##ATTR##_3,				\
+	     save_choose_##ATTR##_4,				\
+	     v );						\
+}
+
+#define INIT(ATTR)					\
+static void save_init_##ATTR( TNLcontext *tnl )		\
+{							\
+   tnl->save.tabfv[ATTR][0] = save_choose_##ATTR##_1;	\
+   tnl->save.tabfv[ATTR][1] = save_choose_##ATTR##_2;	\
+   tnl->save.tabfv[ATTR][2] = save_choose_##ATTR##_3;	\
+   tnl->save.tabfv[ATTR][3] = save_choose_##ATTR##_4;	\
+}
+   
+#define ATTRS( ATTRIB )				\
+   ATTRFV( ATTRIB, 1 )				\
+   ATTRFV( ATTRIB, 2 )				\
+   ATTRFV( ATTRIB, 3 )				\
+   ATTRFV( ATTRIB, 4 )				\
+   CHOOSE( ATTRIB, 1 )				\
+   CHOOSE( ATTRIB, 2 )				\
+   CHOOSE( ATTRIB, 3 )				\
+   CHOOSE( ATTRIB, 4 )				\
+   INIT( ATTRIB )				\
+
+
+/* Generate a lot of functions.  These are the actual worker
+ * functions, which are equivalent to those generated via codegen
+ * elsewhere.
+ */
+ATTRS( 0 )
+ATTRS( 1 )
+ATTRS( 2 )
+ATTRS( 3 )
+ATTRS( 4 )
+ATTRS( 5 )
+ATTRS( 6 )
+ATTRS( 7 )
+ATTRS( 8 )
+ATTRS( 9 )
+ATTRS( 10 )
+ATTRS( 11 )
+ATTRS( 12 )
+ATTRS( 13 )
+ATTRS( 14 )
+ATTRS( 15 )
+
+ATTRS( 16 )
+ATTRS( 17 )
+ATTRS( 18 )
+ATTRS( 19 )
+ATTRS( 20 )
+ATTRS( 21 )
+ATTRS( 22 )
+ATTRS( 23 )
+ATTRS( 24 )
+ATTRS( 25 )
+ATTRS( 26 )
+ATTRS( 27 )
+ATTRS( 28 )
+ATTRS( 29 )
+ATTRS( 30 )
+ATTRS( 31 )
+
+
+static void _save_reset_vertex( GLcontext *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   GLuint i;
+
+   /* conventional attributes */
+   save_init_0( tnl );
+   save_init_1( tnl );
+   save_init_2( tnl );
+   save_init_3( tnl );
+   save_init_4( tnl );
+   save_init_5( tnl );
+   save_init_6( tnl );
+   save_init_7( tnl );
+   save_init_8( tnl );
+   save_init_9( tnl );
+   save_init_10( tnl );
+   save_init_11( tnl );
+   save_init_12( tnl );
+   save_init_13( tnl );
+   save_init_14( tnl );
+   save_init_15( tnl );
+
+   /* generic attributes */
+   save_init_16( tnl );
+   save_init_17( tnl );
+   save_init_18( tnl );
+   save_init_19( tnl );
+   save_init_20( tnl );
+   save_init_21( tnl );
+   save_init_22( tnl );
+   save_init_23( tnl );
+   save_init_24( tnl );
+   save_init_25( tnl );
+   save_init_26( tnl );
+   save_init_27( tnl );
+   save_init_28( tnl );
+   save_init_29( tnl );
+   save_init_30( tnl );
+   save_init_31( tnl );
+      
+   for (i = 0 ; i < _TNL_ATTRIB_MAX ; i++)
+      tnl->save.attrsz[i] = 0;
+      
+   tnl->save.vertex_size = 0;
+   tnl->save.have_materials = 0;
+
+   _save_reset_counters( ctx ); 
+}
+
+
+
+/* Cope with aliasing of classic Vertex, Normal, etc. and the fan-out
+ * of glMultTexCoord and glProgramParamterNV by routing all these
+ * through a second level dispatch table.
+ */
+#define DISPATCH_ATTRFV( ATTR, COUNT, P )	\
+do {						\
+   GET_CURRENT_CONTEXT( ctx ); 			\
+   TNLcontext *tnl = TNL_CONTEXT(ctx); 		\
+   tnl->save.tabfv[ATTR][COUNT-1]( P );		\
+} while (0)
+
+#define DISPATCH_ATTR1FV( ATTR, V ) DISPATCH_ATTRFV( ATTR, 1, V )
+#define DISPATCH_ATTR2FV( ATTR, V ) DISPATCH_ATTRFV( ATTR, 2, V )
+#define DISPATCH_ATTR3FV( ATTR, V ) DISPATCH_ATTRFV( ATTR, 3, V )
+#define DISPATCH_ATTR4FV( ATTR, V ) DISPATCH_ATTRFV( ATTR, 4, V )
+
+#define DISPATCH_ATTR1F( ATTR, S ) DISPATCH_ATTRFV( ATTR, 1, &(S) )
+
+#if defined(USE_X86_ASM) && 0 /* will break register calling convention */
+/* Naughty cheat:
+ */
+#define DISPATCH_ATTR2F( ATTR, S,T ) DISPATCH_ATTRFV( ATTR, 2, &(S) )
+#define DISPATCH_ATTR3F( ATTR, S,T,R ) DISPATCH_ATTRFV( ATTR, 3, &(S) )
+#define DISPATCH_ATTR4F( ATTR, S,T,R,Q ) DISPATCH_ATTRFV( ATTR, 4, &(S) )
+#else
+/* Safe:
+ */
+#define DISPATCH_ATTR2F( ATTR, S,T ) 		\
+do { 						\
+   GLfloat v[2]; 				\
+   v[0] = S; v[1] = T;				\
+   DISPATCH_ATTR2FV( ATTR, v );			\
+} while (0)
+#define DISPATCH_ATTR3F( ATTR, S,T,R ) 		\
+do { 						\
+   GLfloat v[3]; 				\
+   v[0] = S; v[1] = T; v[2] = R;		\
+   DISPATCH_ATTR3FV( ATTR, v );			\
+} while (0)
+#define DISPATCH_ATTR4F( ATTR, S,T,R,Q )	\
+do { 						\
+   GLfloat v[4]; 				\
+   v[0] = S; v[1] = T; v[2] = R; v[3] = Q;	\
+   DISPATCH_ATTR4FV( ATTR, v );			\
+} while (0)
+#endif
+
+
+static void enum_error( void )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   _mesa_compile_error( ctx, GL_INVALID_ENUM, "glVertexAttrib" );
+}
+
+static void GLAPIENTRY _save_Vertex2f( GLfloat x, GLfloat y )
+{
+   DISPATCH_ATTR2F( _TNL_ATTRIB_POS, x, y );
+}
+
+static void GLAPIENTRY _save_Vertex2fv( const GLfloat *v )
+{
+   DISPATCH_ATTR2FV( _TNL_ATTRIB_POS, v );
+}
+
+static void GLAPIENTRY _save_Vertex3f( GLfloat x, GLfloat y, GLfloat z )
+{
+   DISPATCH_ATTR3F( _TNL_ATTRIB_POS, x, y, z );
+}
+
+static void GLAPIENTRY _save_Vertex3fv( const GLfloat *v )
+{
+   DISPATCH_ATTR3FV( _TNL_ATTRIB_POS, v );
+}
+
+static void GLAPIENTRY _save_Vertex4f( GLfloat x, GLfloat y, GLfloat z, GLfloat w )
+{
+   DISPATCH_ATTR4F( _TNL_ATTRIB_POS, x, y, z, w );
+}
+
+static void GLAPIENTRY _save_Vertex4fv( const GLfloat *v )
+{
+   DISPATCH_ATTR4FV( _TNL_ATTRIB_POS, v );
+}
+
+static void GLAPIENTRY _save_TexCoord1f( GLfloat x )
+{
+   DISPATCH_ATTR1F( _TNL_ATTRIB_TEX0, x );
+}
+
+static void GLAPIENTRY _save_TexCoord1fv( const GLfloat *v )
+{
+   DISPATCH_ATTR1FV( _TNL_ATTRIB_TEX0, v );
+}
+
+static void GLAPIENTRY _save_TexCoord2f( GLfloat x, GLfloat y )
+{
+   DISPATCH_ATTR2F( _TNL_ATTRIB_TEX0, x, y );
+}
+
+static void GLAPIENTRY _save_TexCoord2fv( const GLfloat *v )
+{
+   DISPATCH_ATTR2FV( _TNL_ATTRIB_TEX0, v );
+}
+
+static void GLAPIENTRY _save_TexCoord3f( GLfloat x, GLfloat y, GLfloat z )
+{
+   DISPATCH_ATTR3F( _TNL_ATTRIB_TEX0, x, y, z );
+}
+
+static void GLAPIENTRY _save_TexCoord3fv( const GLfloat *v )
+{
+   DISPATCH_ATTR3FV( _TNL_ATTRIB_TEX0, v );
+}
+
+static void GLAPIENTRY _save_TexCoord4f( GLfloat x, GLfloat y, GLfloat z, GLfloat w )
+{
+   DISPATCH_ATTR4F( _TNL_ATTRIB_TEX0, x, y, z, w );
+}
+
+static void GLAPIENTRY _save_TexCoord4fv( const GLfloat *v )
+{
+   DISPATCH_ATTR4FV( _TNL_ATTRIB_TEX0, v );
+}
+
+static void GLAPIENTRY _save_Normal3f( GLfloat x, GLfloat y, GLfloat z )
+{
+   DISPATCH_ATTR3F( _TNL_ATTRIB_NORMAL, x, y, z );
+}
+
+static void GLAPIENTRY _save_Normal3fv( const GLfloat *v )
+{
+   DISPATCH_ATTR3FV( _TNL_ATTRIB_NORMAL, v );
+}
+
+static void GLAPIENTRY _save_FogCoordfEXT( GLfloat x )
+{
+   DISPATCH_ATTR1F( _TNL_ATTRIB_FOG, x );
+}
+
+static void GLAPIENTRY _save_FogCoordfvEXT( const GLfloat *v )
+{
+   DISPATCH_ATTR1FV( _TNL_ATTRIB_FOG, v );
+}
+
+static void GLAPIENTRY _save_Color3f( GLfloat x, GLfloat y, GLfloat z )
+{
+   DISPATCH_ATTR3F( _TNL_ATTRIB_COLOR0, x, y, z );
+}
+
+static void GLAPIENTRY _save_Color3fv( const GLfloat *v )
+{
+   DISPATCH_ATTR3FV( _TNL_ATTRIB_COLOR0, v );
+}
+
+static void GLAPIENTRY _save_Color4f( GLfloat x, GLfloat y, GLfloat z, GLfloat w )
+{
+   DISPATCH_ATTR4F( _TNL_ATTRIB_COLOR0, x, y, z, w );
+}
+
+static void GLAPIENTRY _save_Color4fv( const GLfloat *v )
+{
+   DISPATCH_ATTR4FV( _TNL_ATTRIB_COLOR0, v );
+}
+
+static void GLAPIENTRY _save_SecondaryColor3fEXT( GLfloat x, GLfloat y, GLfloat z )
+{
+   DISPATCH_ATTR3F( _TNL_ATTRIB_COLOR1, x, y, z );
+}
+
+static void GLAPIENTRY _save_SecondaryColor3fvEXT( const GLfloat *v )
+{
+   DISPATCH_ATTR3FV( _TNL_ATTRIB_COLOR1, v );
+}
+
+static void GLAPIENTRY _save_MultiTexCoord1f( GLenum target, GLfloat x  )
+{
+   GLuint attr = (target & 0x7) + _TNL_ATTRIB_TEX0;
+   DISPATCH_ATTR1F( attr, x );
+}
+
+static void GLAPIENTRY _save_MultiTexCoord1fv( GLenum target, const GLfloat *v )
+{
+   GLuint attr = (target & 0x7) + _TNL_ATTRIB_TEX0;
+   DISPATCH_ATTR1FV( attr, v );
+}
+
+static void GLAPIENTRY _save_MultiTexCoord2f( GLenum target, GLfloat x, GLfloat y )
+{
+   GLuint attr = (target & 0x7) + _TNL_ATTRIB_TEX0;
+   DISPATCH_ATTR2F( attr, x, y );
+}
+
+static void GLAPIENTRY _save_MultiTexCoord2fv( GLenum target, const GLfloat *v )
+{
+   GLuint attr = (target & 0x7) + _TNL_ATTRIB_TEX0;
+   DISPATCH_ATTR2FV( attr, v );
+}
+
+static void GLAPIENTRY _save_MultiTexCoord3f( GLenum target, GLfloat x, GLfloat y,
+				    GLfloat z)
+{
+   GLuint attr = (target & 0x7) + _TNL_ATTRIB_TEX0;
+   DISPATCH_ATTR3F( attr, x, y, z );
+}
+
+static void GLAPIENTRY _save_MultiTexCoord3fv( GLenum target, const GLfloat *v )
+{
+   GLuint attr = (target & 0x7) + _TNL_ATTRIB_TEX0;
+   DISPATCH_ATTR3FV( attr, v );
+}
+
+static void GLAPIENTRY _save_MultiTexCoord4f( GLenum target, GLfloat x, GLfloat y,
+				    GLfloat z, GLfloat w )
+{
+   GLuint attr = (target & 0x7) + _TNL_ATTRIB_TEX0;
+   DISPATCH_ATTR4F( attr, x, y, z, w );
+}
+
+static void GLAPIENTRY _save_MultiTexCoord4fv( GLenum target, const GLfloat *v )
+{
+   GLuint attr = (target & 0x7) + _TNL_ATTRIB_TEX0;
+   DISPATCH_ATTR4FV( attr, v );
+}
+
+
+
+static void GLAPIENTRY
+_save_VertexAttrib1fNV(GLuint index, GLfloat x)
+{
+   if (index < MAX_VERTEX_PROGRAM_ATTRIBS) {
+      if (index > 0)
+         index += VERT_ATTRIB_GENERIC0;
+      DISPATCH_ATTR1F( index, x );
+   }
+   else
+      enum_error(); 
+}
+
+static void GLAPIENTRY
+_save_VertexAttrib1fvNV(GLuint index, const GLfloat *v)
+{
+   if (index < MAX_VERTEX_PROGRAM_ATTRIBS) {
+      if (index > 0)
+         index += VERT_ATTRIB_GENERIC0;
+      DISPATCH_ATTR1FV( index, v );
+   }
+   else
+      enum_error();
+}
+
+static void GLAPIENTRY
+_save_VertexAttrib2fNV(GLuint index, GLfloat x, GLfloat y)
+{
+   if (index < MAX_VERTEX_PROGRAM_ATTRIBS) {
+      if (index > 0)
+         index += VERT_ATTRIB_GENERIC0;
+      DISPATCH_ATTR2F( index, x, y );
+   }
+   else
+      enum_error();
+}
+
+static void GLAPIENTRY
+_save_VertexAttrib2fvNV(GLuint index, const GLfloat *v)
+{
+   if (index < MAX_VERTEX_PROGRAM_ATTRIBS) {
+      if (index > 0)
+         index += VERT_ATTRIB_GENERIC0;
+      DISPATCH_ATTR2FV( index, v );
+   }
+   else
+      enum_error();
+}
+
+static void GLAPIENTRY
+_save_VertexAttrib3fNV(GLuint index, GLfloat x, GLfloat y, GLfloat z)
+{
+   if (index < MAX_VERTEX_PROGRAM_ATTRIBS) {
+      if (index > 0)
+         index += VERT_ATTRIB_GENERIC0;
+      DISPATCH_ATTR3F( index, x, y, z );
+   }
+   else
+      enum_error();
+}
+
+static void GLAPIENTRY
+_save_VertexAttrib3fvNV(GLuint index, const GLfloat *v)
+{
+   if (index < MAX_VERTEX_PROGRAM_ATTRIBS) {
+      if (index > 0)
+         index += VERT_ATTRIB_GENERIC0;
+      DISPATCH_ATTR3FV( index, v );
+   }
+   else
+      enum_error();
+}
+
+static void GLAPIENTRY
+_save_VertexAttrib4fNV(GLuint index, GLfloat x, GLfloat y,
+                       GLfloat z, GLfloat w)
+{
+   if (index < MAX_VERTEX_PROGRAM_ATTRIBS) {
+      if (index > 0)
+         index += VERT_ATTRIB_GENERIC0;
+      DISPATCH_ATTR4F( index, x, y, z, w );
+   }
+   else
+      enum_error();
+}
+
+static void GLAPIENTRY
+_save_VertexAttrib4fvNV(GLuint index, const GLfloat *v)
+{
+   if (index < MAX_VERTEX_PROGRAM_ATTRIBS) {
+      if (index > 0)
+         index += VERT_ATTRIB_GENERIC0;
+      DISPATCH_ATTR4FV( index, v );
+   }
+   else
+      enum_error();
+}
+
+static void GLAPIENTRY
+_save_VertexAttrib1fARB(GLuint index, GLfloat x)
+{
+   if (index < MAX_VERTEX_ATTRIBS) {
+      if (index > 0)
+         index += VERT_ATTRIB_GENERIC0;
+      DISPATCH_ATTR1F( index, x );
+   }
+   else
+      enum_error(); 
+}
+
+static void GLAPIENTRY
+_save_VertexAttrib1fvARB(GLuint index, const GLfloat *v)
+{
+   if (index < MAX_VERTEX_ATTRIBS) {
+      if (index > 0)
+         index += VERT_ATTRIB_GENERIC0;
+      DISPATCH_ATTR1FV( index, v );
+   }
+   else
+      enum_error();
+}
+
+static void GLAPIENTRY
+_save_VertexAttrib2fARB(GLuint index, GLfloat x, GLfloat y)
+{
+   if (index < MAX_VERTEX_ATTRIBS) {
+      if (index > 0)
+         index += VERT_ATTRIB_GENERIC0;
+      DISPATCH_ATTR2F( index, x, y );
+   }
+   else
+      enum_error();
+}
+
+static void GLAPIENTRY
+_save_VertexAttrib2fvARB(GLuint index, const GLfloat *v)
+{
+   if (index < MAX_VERTEX_ATTRIBS) {
+      if (index > 0)
+         index += VERT_ATTRIB_GENERIC0;
+      DISPATCH_ATTR2FV( index, v );
+   }
+   else
+      enum_error();
+}
+
+static void GLAPIENTRY
+_save_VertexAttrib3fARB(GLuint index, GLfloat x, GLfloat y, GLfloat z)
+{
+   if (index < MAX_VERTEX_ATTRIBS) {
+      if (index > 0)
+         index += VERT_ATTRIB_GENERIC0;
+      DISPATCH_ATTR3F( index, x, y, z );
+   }
+   else
+      enum_error();
+}
+
+static void GLAPIENTRY
+_save_VertexAttrib3fvARB(GLuint index, const GLfloat *v)
+{
+   if (index < MAX_VERTEX_ATTRIBS) {
+      if (index > 0)
+         index += VERT_ATTRIB_GENERIC0;
+      DISPATCH_ATTR3FV( index, v );
+   }
+   else
+      enum_error();
+}
+
+static void GLAPIENTRY
+_save_VertexAttrib4fARB(GLuint index, GLfloat x, GLfloat y,
+                        GLfloat z, GLfloat w)
+{
+   if (index < MAX_VERTEX_ATTRIBS) {
+      if (index > 0)
+         index += VERT_ATTRIB_GENERIC0;
+      DISPATCH_ATTR4F( index, x, y, z, w );
+   }
+   else
+      enum_error();
+}
+
+static void GLAPIENTRY
+_save_VertexAttrib4fvARB(GLuint index, const GLfloat *v)
+{
+   if (index < MAX_VERTEX_ATTRIBS) {
+      if (index > 0)
+         index += VERT_ATTRIB_GENERIC0;
+      DISPATCH_ATTR4FV( index, v );
+   }
+   else
+      enum_error();
+}
+
+
+/* Materials:  
+ * 
+ * These are treated as per-vertex attributes, at indices above where
+ * the NV_vertex_program leaves off.  There are a lot of good things
+ * about treating materials this way.  
+ *
+ * However: I don't want to double the number of generated functions
+ * just to cope with this, so I unroll the 'C' varients of CHOOSE and
+ * ATTRF into this function, and dispense with codegen and
+ * second-level dispatch.
+ *
+ * There is no aliasing of material attributes with other entrypoints.
+ */
+#define MAT_ATTR( A, N, params )			\
+do {							\
+   if (tnl->save.attrsz[A] < N) {			\
+      _save_upgrade_vertex( ctx, A, N );		\
+      tnl->save.have_materials = GL_TRUE;               \
+   }							\
+							\
+   {							\
+      GLfloat *dest = tnl->save.attrptr[A];	      	\
+      if (N>0) dest[0] = params[0];			\
+      if (N>1) dest[1] = params[1];			\
+      if (N>2) dest[2] = params[2];			\
+      if (N>3) dest[3] = params[3];			\
+   }							\
+} while (0)
+
+
+#define MAT( ATTR, N, face, params )			\
+do {							\
+   if (face != GL_BACK)					\
+      MAT_ATTR( ATTR, N, params ); /* front */		\
+   if (face != GL_FRONT)				\
+      MAT_ATTR( ATTR + 1, N, params ); /* back */	\
+} while (0)
+
+
+/* NOTE: Have to remove/deal-with colormaterial crossovers, probably
+ * later on - in the meantime just store everything.  
+ */
+static void GLAPIENTRY _save_Materialfv( GLenum face, GLenum pname, 
+			       const GLfloat *params )
+{
+   GET_CURRENT_CONTEXT( ctx ); 
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+
+   switch (pname) {
+   case GL_EMISSION:
+      MAT( _TNL_ATTRIB_MAT_FRONT_EMISSION, 4, face, params );
+      break;
+   case GL_AMBIENT:
+      MAT( _TNL_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params );
+      break;
+   case GL_DIFFUSE:
+      MAT( _TNL_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params );
+      break;
+   case GL_SPECULAR:
+      MAT( _TNL_ATTRIB_MAT_FRONT_SPECULAR, 4, face, params );
+      break;
+   case GL_SHININESS:
+      MAT( _TNL_ATTRIB_MAT_FRONT_SHININESS, 1, face, params );
+      break;
+   case GL_COLOR_INDEXES:
+      MAT( _TNL_ATTRIB_MAT_FRONT_INDEXES, 3, face, params );
+      break;
+   case GL_AMBIENT_AND_DIFFUSE:
+      MAT( _TNL_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params );
+      MAT( _TNL_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params );
+      break;
+   default:
+      _mesa_compile_error( ctx, GL_INVALID_ENUM, "glMaterialfv" );
+      return;
+   }
+}
+
+
+#define IDX_ATTR( A, IDX )				\
+do {							\
+   GET_CURRENT_CONTEXT( ctx );				\
+   TNLcontext *tnl = TNL_CONTEXT(ctx);			\
+							\
+   if (tnl->save.attrsz[A] < 1) {			\
+      _save_upgrade_vertex( ctx, A, 1 );		\
+   }							\
+							\
+   {							\
+      GLfloat *dest = tnl->save.attrptr[A];		\
+      dest[0] = IDX;				\
+   }							\
+} while (0)
+
+
+static void GLAPIENTRY _save_EdgeFlag( GLboolean b )
+{
+   IDX_ATTR( _TNL_ATTRIB_EDGEFLAG, (GLfloat)b );
+}
+
+
+static void GLAPIENTRY _save_Indexf( GLfloat f )
+{
+   IDX_ATTR( _TNL_ATTRIB_COLOR_INDEX, f );
+}
+
+static void GLAPIENTRY _save_Indexfv( const GLfloat *f )
+{
+   IDX_ATTR( _TNL_ATTRIB_COLOR_INDEX, f[0] );
+}
+
+
+
+
+/* Cope with EvalCoord/CallList called within a begin/end object:
+ *     -- Flush current buffer
+ *     -- Fallback to opcodes for the rest of the begin/end object.
+ */
+#define FALLBACK(ctx) 							\
+do {									\
+   TNLcontext *tnl = TNL_CONTEXT(ctx);					\
+									\
+   if (tnl->save.initial_counter != tnl->save.counter ||		\
+       tnl->save.prim_count) 						\
+      _save_compile_vertex_list( ctx );					\
+									\
+   _save_copy_to_current( ctx );					\
+   _save_reset_vertex( ctx );						\
+   _mesa_install_save_vtxfmt( ctx, &ctx->ListState.ListVtxfmt );	\
+   ctx->Driver.SaveNeedFlush = 0;					\
+} while (0)
+
+static void GLAPIENTRY _save_EvalCoord1f( GLfloat u )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   FALLBACK(ctx);
+   CALL_EvalCoord1f(ctx->Save, ( u ));
+}
+
+static void GLAPIENTRY _save_EvalCoord1fv( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   FALLBACK(ctx);
+   CALL_EvalCoord1fv(ctx->Save, ( v ));
+}
+
+static void GLAPIENTRY _save_EvalCoord2f( GLfloat u, GLfloat v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   FALLBACK(ctx);
+   CALL_EvalCoord2f(ctx->Save, ( u, v ));
+}
+
+static void GLAPIENTRY _save_EvalCoord2fv( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   FALLBACK(ctx);
+   CALL_EvalCoord2fv(ctx->Save, ( v ));
+}
+
+static void GLAPIENTRY _save_EvalPoint1( GLint i )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   FALLBACK(ctx);
+   CALL_EvalPoint1(ctx->Save, ( i ));
+}
+
+static void GLAPIENTRY _save_EvalPoint2( GLint i, GLint j )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   FALLBACK(ctx);
+   CALL_EvalPoint2(ctx->Save, ( i, j ));
+}
+
+static void GLAPIENTRY _save_CallList( GLuint l )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   FALLBACK(ctx);
+   CALL_CallList(ctx->Save, ( l ));
+}
+
+static void GLAPIENTRY _save_CallLists( GLsizei n, GLenum type, const GLvoid *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   FALLBACK(ctx);
+   CALL_CallLists(ctx->Save, ( n, type, v ));
+}
+
+
+
+
+/**
+ * Called via ctx->Driver.NotifySaveBegin(ctx, mode) when we get a
+ * glBegin() call while compiling a display list.
+ * See save_Begin() in dlist.c
+ *
+ * This plugs in our special TNL-related display list functions.
+ * All subsequent glBegin/glVertex/glEnd()s found while compiling a
+ * display list will get routed to the functions in this file.
+ *
+ * Updating of ctx->Driver.CurrentSavePrimitive is already taken care of.
+ */
+static GLboolean _save_NotifyBegin( GLcontext *ctx, GLenum mode )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx); 
+
+   if (1) {
+      GLuint i = tnl->save.prim_count++;
+
+      assert(i < tnl->save.prim_max);
+      tnl->save.prim[i].mode = mode | PRIM_BEGIN;
+      tnl->save.prim[i].start = tnl->save.initial_counter - tnl->save.counter;
+      tnl->save.prim[i].count = 0;   
+
+      _mesa_install_save_vtxfmt( ctx, &tnl->save_vtxfmt );      
+      ctx->Driver.SaveNeedFlush = 1;
+      return GL_TRUE;
+   }
+   else 
+      return GL_FALSE;
+}
+
+
+
+static void GLAPIENTRY _save_End( void )
+{
+   GET_CURRENT_CONTEXT( ctx ); 
+   TNLcontext *tnl = TNL_CONTEXT(ctx); 
+   GLint i = tnl->save.prim_count - 1;
+
+   ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
+   if (ctx->ExecuteFlag)
+      ctx->Driver.CurrentExecPrimitive = PRIM_OUTSIDE_BEGIN_END;
+
+   tnl->save.prim[i].mode |= PRIM_END;
+   tnl->save.prim[i].count = ((tnl->save.initial_counter - tnl->save.counter) - 
+			      tnl->save.prim[i].start);
+
+   if (i == (GLint) tnl->save.prim_max - 1) {
+      _save_compile_vertex_list( ctx );
+      assert(tnl->save.copied.nr == 0);
+   }
+
+   /* Swap out this vertex format while outside begin/end.  Any color,
+    * etc. received between here and the next begin will be compiled
+    * as opcodes.
+    */   
+   _mesa_install_save_vtxfmt( ctx, &ctx->ListState.ListVtxfmt );
+}
+
+
+/* These are all errors as this vtxfmt is only installed inside
+ * begin/end pairs.
+ */
+static void GLAPIENTRY _save_DrawElements(GLenum mode, GLsizei count, GLenum type,
+			       const GLvoid *indices)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   (void) mode; (void) count; (void) type; (void) indices;
+   _mesa_compile_error( ctx, GL_INVALID_OPERATION, "glDrawElements" );
+}
+
+
+static void GLAPIENTRY _save_DrawRangeElements(GLenum mode,
+				    GLuint start, GLuint end,
+				    GLsizei count, GLenum type,
+				    const GLvoid *indices)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   (void) mode; (void) start; (void) end; (void) count; (void) type; (void) indices;
+   _mesa_compile_error( ctx, GL_INVALID_OPERATION, "glDrawRangeElements" );
+}
+
+static void GLAPIENTRY _save_DrawArrays(GLenum mode, GLint start, GLsizei count)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   (void) mode; (void) start; (void) count;
+   _mesa_compile_error( ctx, GL_INVALID_OPERATION, "glDrawArrays" );
+}
+
+static void GLAPIENTRY _save_Rectf( GLfloat x1, GLfloat y1, GLfloat x2, GLfloat y2 )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   (void) x1; (void) y1; (void) x2; (void) y2;
+   _mesa_compile_error( ctx, GL_INVALID_OPERATION, "glRectf" );
+}
+
+static void GLAPIENTRY _save_EvalMesh1( GLenum mode, GLint i1, GLint i2 )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   (void) mode; (void) i1; (void) i2;
+   _mesa_compile_error( ctx, GL_INVALID_OPERATION, "glEvalMesh1" );
+}
+
+static void GLAPIENTRY _save_EvalMesh2( GLenum mode, GLint i1, GLint i2,
+				  GLint j1, GLint j2 )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   (void) mode; (void) i1; (void) i2; (void) j1; (void) j2;
+   _mesa_compile_error( ctx, GL_INVALID_OPERATION, "glEvalMesh2" );
+}
+
+/**
+ * This is only called if someone tries to compile nested glBegin()s 
+ * in their display list.
+ */
+static void GLAPIENTRY _save_Begin( GLenum mode )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   (void) mode;
+   _mesa_compile_error(ctx, GL_INVALID_OPERATION,
+                       "glBegin(called inside glBegin/End)");
+}
+
+
+/* Unlike the functions above, these are to be hooked into the vtxfmt
+ * maintained in ctx->ListState, active when the list is known or
+ * suspected to be outside any begin/end primitive.
+ */
+static void GLAPIENTRY _save_OBE_Rectf( GLfloat x1, GLfloat y1, GLfloat x2, GLfloat y2 )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   _save_NotifyBegin( ctx, GL_QUADS | PRIM_WEAK );
+   CALL_Vertex2f(GET_DISPATCH(), ( x1, y1 ));
+   CALL_Vertex2f(GET_DISPATCH(), ( x2, y1 ));
+   CALL_Vertex2f(GET_DISPATCH(), ( x2, y2 ));
+   CALL_Vertex2f(GET_DISPATCH(), ( x1, y2 ));
+   CALL_End(GET_DISPATCH(), ());
+}
+
+
+static void GLAPIENTRY _save_OBE_DrawArrays(GLenum mode, GLint start, GLsizei count)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   GLint i;
+
+   if (!_mesa_validate_DrawArrays( ctx, mode, start, count ))
+      return;
+
+   _ae_map_vbos( ctx );
+
+   _save_NotifyBegin( ctx, mode | PRIM_WEAK );
+   for (i = 0; i < count; i++)
+       CALL_ArrayElement(GET_DISPATCH(), (start + i));
+   CALL_End(GET_DISPATCH(), ());
+
+   _ae_unmap_vbos( ctx );
+}
+
+
+static void GLAPIENTRY _save_OBE_DrawElements(GLenum mode, GLsizei count, GLenum type,
+				   const GLvoid *indices)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   GLint i;
+
+   if (!_mesa_validate_DrawElements( ctx, mode, count, type, indices ))
+      return;
+
+   _ae_map_vbos( ctx );
+
+   _save_NotifyBegin( ctx, mode | PRIM_WEAK );
+
+   switch (type) {
+   case GL_UNSIGNED_BYTE:
+      for (i = 0 ; i < count ; i++)
+	  CALL_ArrayElement(GET_DISPATCH(), ( ((GLubyte *)indices)[i] ));
+      break;
+   case GL_UNSIGNED_SHORT:
+      for (i = 0 ; i < count ; i++)
+	  CALL_ArrayElement(GET_DISPATCH(), ( ((GLushort *)indices)[i] ));
+      break;
+   case GL_UNSIGNED_INT:
+      for (i = 0 ; i < count ; i++)
+	  CALL_ArrayElement(GET_DISPATCH(), ( ((GLuint *)indices)[i] ));
+      break;
+   default:
+      _mesa_error( ctx, GL_INVALID_ENUM, "glDrawElements(type)" );
+      break;
+   }
+
+   CALL_End(GET_DISPATCH(), ());
+
+   _ae_unmap_vbos( ctx );
+}
+
+static void GLAPIENTRY _save_OBE_DrawRangeElements(GLenum mode,
+					GLuint start, GLuint end,
+					GLsizei count, GLenum type,
+					const GLvoid *indices)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   if (_mesa_validate_DrawRangeElements( ctx, mode,
+					 start, end,
+					 count, type, indices ))
+      _save_OBE_DrawElements( mode, count, type, indices );
+}
+
+
+
+
+
+static void _save_vtxfmt_init( GLcontext *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   GLvertexformat *vfmt = &tnl->save_vtxfmt;
+
+   vfmt->ArrayElement = _ae_loopback_array_elt;	        /* generic helper */
+   vfmt->Begin = _save_Begin;
+   vfmt->Color3f = _save_Color3f;
+   vfmt->Color3fv = _save_Color3fv;
+   vfmt->Color4f = _save_Color4f;
+   vfmt->Color4fv = _save_Color4fv;
+   vfmt->EdgeFlag = _save_EdgeFlag;
+   vfmt->End = _save_End;
+   vfmt->FogCoordfEXT = _save_FogCoordfEXT;
+   vfmt->FogCoordfvEXT = _save_FogCoordfvEXT;
+   vfmt->Indexf = _save_Indexf;
+   vfmt->Indexfv = _save_Indexfv;
+   vfmt->Materialfv = _save_Materialfv;
+   vfmt->MultiTexCoord1fARB = _save_MultiTexCoord1f;
+   vfmt->MultiTexCoord1fvARB = _save_MultiTexCoord1fv;
+   vfmt->MultiTexCoord2fARB = _save_MultiTexCoord2f;
+   vfmt->MultiTexCoord2fvARB = _save_MultiTexCoord2fv;
+   vfmt->MultiTexCoord3fARB = _save_MultiTexCoord3f;
+   vfmt->MultiTexCoord3fvARB = _save_MultiTexCoord3fv;
+   vfmt->MultiTexCoord4fARB = _save_MultiTexCoord4f;
+   vfmt->MultiTexCoord4fvARB = _save_MultiTexCoord4fv;
+   vfmt->Normal3f = _save_Normal3f;
+   vfmt->Normal3fv = _save_Normal3fv;
+   vfmt->SecondaryColor3fEXT = _save_SecondaryColor3fEXT;
+   vfmt->SecondaryColor3fvEXT = _save_SecondaryColor3fvEXT;
+   vfmt->TexCoord1f = _save_TexCoord1f;
+   vfmt->TexCoord1fv = _save_TexCoord1fv;
+   vfmt->TexCoord2f = _save_TexCoord2f;
+   vfmt->TexCoord2fv = _save_TexCoord2fv;
+   vfmt->TexCoord3f = _save_TexCoord3f;
+   vfmt->TexCoord3fv = _save_TexCoord3fv;
+   vfmt->TexCoord4f = _save_TexCoord4f;
+   vfmt->TexCoord4fv = _save_TexCoord4fv;
+   vfmt->Vertex2f = _save_Vertex2f;
+   vfmt->Vertex2fv = _save_Vertex2fv;
+   vfmt->Vertex3f = _save_Vertex3f;
+   vfmt->Vertex3fv = _save_Vertex3fv;
+   vfmt->Vertex4f = _save_Vertex4f;
+   vfmt->Vertex4fv = _save_Vertex4fv;
+   vfmt->VertexAttrib1fNV = _save_VertexAttrib1fNV;
+   vfmt->VertexAttrib1fvNV = _save_VertexAttrib1fvNV;
+   vfmt->VertexAttrib2fNV = _save_VertexAttrib2fNV;
+   vfmt->VertexAttrib2fvNV = _save_VertexAttrib2fvNV;
+   vfmt->VertexAttrib3fNV = _save_VertexAttrib3fNV;
+   vfmt->VertexAttrib3fvNV = _save_VertexAttrib3fvNV;
+   vfmt->VertexAttrib4fNV = _save_VertexAttrib4fNV;
+   vfmt->VertexAttrib4fvNV = _save_VertexAttrib4fvNV;
+   vfmt->VertexAttrib1fARB = _save_VertexAttrib1fARB;
+   vfmt->VertexAttrib1fvARB = _save_VertexAttrib1fvARB;
+   vfmt->VertexAttrib2fARB = _save_VertexAttrib2fARB;
+   vfmt->VertexAttrib2fvARB = _save_VertexAttrib2fvARB;
+   vfmt->VertexAttrib3fARB = _save_VertexAttrib3fARB;
+   vfmt->VertexAttrib3fvARB = _save_VertexAttrib3fvARB;
+   vfmt->VertexAttrib4fARB = _save_VertexAttrib4fARB;
+   vfmt->VertexAttrib4fvARB = _save_VertexAttrib4fvARB;
+   
+   /* This will all require us to fallback to saving the list as opcodes:
+    */ 
+   vfmt->CallList = _save_CallList; /* inside begin/end */
+   vfmt->CallLists = _save_CallLists; /* inside begin/end */
+   vfmt->EvalCoord1f = _save_EvalCoord1f;
+   vfmt->EvalCoord1fv = _save_EvalCoord1fv;
+   vfmt->EvalCoord2f = _save_EvalCoord2f;
+   vfmt->EvalCoord2fv = _save_EvalCoord2fv;
+   vfmt->EvalPoint1 = _save_EvalPoint1;
+   vfmt->EvalPoint2 = _save_EvalPoint2;
+
+   /* These are all errors as we at least know we are in some sort of
+    * begin/end pair:
+    */
+   vfmt->EvalMesh1 = _save_EvalMesh1;	
+   vfmt->EvalMesh2 = _save_EvalMesh2;
+   vfmt->Begin = _save_Begin;
+   vfmt->Rectf = _save_Rectf;
+   vfmt->DrawArrays = _save_DrawArrays;
+   vfmt->DrawElements = _save_DrawElements;
+   vfmt->DrawRangeElements = _save_DrawRangeElements;
+
+}
+
+
+void _tnl_SaveFlushVertices( GLcontext *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+
+   /* Noop when we are actually active:
+    */
+   if (ctx->Driver.CurrentSavePrimitive == PRIM_INSIDE_UNKNOWN_PRIM ||
+       ctx->Driver.CurrentSavePrimitive <= GL_POLYGON)
+      return;
+
+   if (tnl->save.initial_counter != tnl->save.counter ||
+       tnl->save.prim_count) 
+      _save_compile_vertex_list( ctx );
+   
+   _save_copy_to_current( ctx );
+   _save_reset_vertex( ctx );
+   ctx->Driver.SaveNeedFlush = 0;
+}
+
+void _tnl_NewList( GLcontext *ctx, GLuint list, GLenum mode )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+
+   (void) list; (void) mode;
+
+   if (!tnl->save.prim_store)
+      tnl->save.prim_store = alloc_prim_store( ctx );
+
+   if (!tnl->save.vertex_store) {
+      tnl->save.vertex_store = alloc_vertex_store( ctx );
+      tnl->save.vbptr = tnl->save.vertex_store->buffer;
+   }
+   
+   _save_reset_vertex( ctx );
+   ctx->Driver.SaveNeedFlush = 0;
+}
+
+void _tnl_EndList( GLcontext *ctx )
+{
+   (void) ctx;
+   assert(TNL_CONTEXT(ctx)->save.vertex_size == 0);
+}
+ 
+void _tnl_BeginCallList( GLcontext *ctx, struct mesa_display_list *dlist )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   tnl->save.replay_flags |= dlist->flags;
+   tnl->save.replay_flags |= tnl->LoopbackDListCassettes;
+}
+
+void _tnl_EndCallList( GLcontext *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   
+   if (ctx->ListState.CallDepth == 1)
+      tnl->save.replay_flags = 0;
+}
+
+
+static void _tnl_destroy_vertex_list( GLcontext *ctx, void *data )
+{
+   struct tnl_vertex_list *node = (struct tnl_vertex_list *)data;
+   (void) ctx;
+
+   if ( --node->vertex_store->refcount == 0 )
+      FREE( node->vertex_store );
+
+   if ( --node->prim_store->refcount == 0 )
+      FREE( node->prim_store );
+
+   if ( node->normal_lengths )
+      FREE( node->normal_lengths );
+}
+
+
+static void _tnl_print_vertex_list( GLcontext *ctx, void *data )
+{
+   struct tnl_vertex_list *node = (struct tnl_vertex_list *)data;
+   GLuint i;
+   (void) ctx;
+
+   _mesa_debug(NULL, "TNL-VERTEX-LIST, %u vertices %d primitives, %d vertsize\n",
+               node->count,
+	       node->prim_count,
+	       node->vertex_size);
+
+   for (i = 0 ; i < node->prim_count ; i++) {
+      struct tnl_prim *prim = &node->prim[i];
+      _mesa_debug(NULL, "   prim %d: %s %d..%d %s %s\n",
+		  i, 
+		  _mesa_lookup_enum_by_nr(prim->mode & PRIM_MODE_MASK),
+		  prim->start, 
+		  prim->start + prim->count,
+		  (prim->mode & PRIM_BEGIN) ? "BEGIN" : "(wrap)",
+		  (prim->mode & PRIM_END) ? "END" : "(wrap)");
+   }
+}
+
+
+static void _save_current_init( GLcontext *ctx ) 
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   GLint i;
+
+   for (i = 0; i < _TNL_ATTRIB_MAT_FRONT_AMBIENT; i++) {
+      ASSERT(i < VERT_ATTRIB_MAX);
+      tnl->save.currentsz[i] = &ctx->ListState.ActiveAttribSize[i];
+      tnl->save.current[i] = ctx->ListState.CurrentAttrib[i];
+   }
+
+   for (i = _TNL_FIRST_MAT; i <= _TNL_LAST_MAT; i++) {
+      const GLuint j = i - _TNL_FIRST_MAT;
+      ASSERT(j < MAT_ATTRIB_MAX);
+      tnl->save.currentsz[i] = &ctx->ListState.ActiveMaterialSize[j];
+      tnl->save.current[i] = ctx->ListState.CurrentMaterial[j];
+   }
+
+   tnl->save.currentsz[_TNL_ATTRIB_EDGEFLAG] = &ctx->ListState.ActiveEdgeFlag;
+   tnl->save.current[_TNL_ATTRIB_EDGEFLAG] = &tnl->save.CurrentFloatEdgeFlag;
+}
+
+/**
+ * Initialize the display list compiler
+ */
+void _tnl_save_init( GLcontext *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   struct tnl_vertex_arrays *tmp = &tnl->save_inputs;
+   GLuint i;
+
+
+   for (i = 0; i < _TNL_ATTRIB_MAX; i++)
+      _mesa_vector4f_init( &tmp->Attribs[i], 0, NULL);
+
+   tnl->save.opcode_vertex_list =
+      _mesa_alloc_opcode( ctx,
+			  sizeof(struct tnl_vertex_list),
+			  _tnl_playback_vertex_list,
+			  _tnl_destroy_vertex_list,
+			  _tnl_print_vertex_list );
+
+   ctx->Driver.NotifySaveBegin = _save_NotifyBegin;
+
+   _save_vtxfmt_init( ctx );
+   _save_current_init( ctx );
+
+   /* Hook our array functions into the outside-begin-end vtxfmt in 
+    * ctx->ListState.
+    */
+   ctx->ListState.ListVtxfmt.Rectf = _save_OBE_Rectf;
+   ctx->ListState.ListVtxfmt.DrawArrays = _save_OBE_DrawArrays;
+   ctx->ListState.ListVtxfmt.DrawElements = _save_OBE_DrawElements;
+   ctx->ListState.ListVtxfmt.DrawRangeElements = _save_OBE_DrawRangeElements;
+   _mesa_install_save_vtxfmt( ctx, &ctx->ListState.ListVtxfmt );
+}
+
+
+/**
+ * Deallocate the immediate-mode buffer for the given context, if
+ * its reference count goes to zero.
+ */
+void _tnl_save_destroy( GLcontext *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+
+   /* Decrement the refcounts.  References may still be held by
+    * display lists yet to be destroyed, so it may not yet be time to
+    * free these items.
+    */
+   if (tnl->save.prim_store &&
+       --tnl->save.prim_store->refcount == 0 )
+      FREE( tnl->save.prim_store );
+
+   if (tnl->save.vertex_store &&
+       --tnl->save.vertex_store->refcount == 0 )
+      FREE( tnl->save.vertex_store );
+}
diff --git a/src/mesa/tnl/t_vb_arbprogram.c b/src/mesa/tnl/t_vb_arbprogram.c
index 4c8f967fdff..425a8669948 100644
--- a/src/mesa/tnl/t_vb_arbprogram.c
+++ b/src/mesa/tnl/t_vb_arbprogram.c
@@ -1,8 +1,8 @@
 /*
  * Mesa 3-D graphics library
- * Version:  6.5.1
+ * Version:  6.5.3
  *
- * Copyright (C) 1999-2006  Brian Paul   All Rights Reserved.
+ * Copyright (C) 1999-2007  Brian Paul   All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -37,12 +37,15 @@
 #include "arbprogparse.h"
 #include "light.h"
 #include "program.h"
+#include "prog_instruction.h"
+#include "prog_parameter.h"
+#include "prog_statevars.h"
+#include "programopt.h"
 #include "math/m_matrix.h"
 #include "t_context.h"
 #include "t_pipeline.h"
 #include "t_vb_arbprogram.h"
 #include "tnl.h"
-#include "program_instruction.h"
 
 
 #define DISASSEM 0
@@ -292,7 +295,7 @@ static void do_EX2( struct arb_vp_machine *m, union instruction op )
    GLfloat *result = m->File[0][op.alu.dst];
    const GLfloat *arg0 = m->File[op.alu.file0][op.alu.idx0];
 
-   result[0] = (GLfloat)ApproxExp2(arg0[0]);
+   result[0] = ApproxExp2(arg0[0]);
    PUFF(result);
 }
 
@@ -335,6 +338,17 @@ static void do_FRC( struct arb_vp_machine *m, union instruction op )
    result[3] = arg0[3] - FLOORF(arg0[3]);
 }
 
+static void do_INT( struct arb_vp_machine *m, union instruction op ) 
+{
+   GLfloat *result = m->File[0][op.alu.dst];
+   const GLfloat *arg0 = m->File[op.alu.file0][op.alu.idx0];
+
+   result[0] = (GLfloat) (GLint) arg0[0];
+   result[1] = (GLfloat) (GLint) arg0[1];
+   result[2] = (GLfloat) (GLint) arg0[2];
+   result[3] = (GLfloat) (GLint) arg0[3];
+}
+
 /* High precision log base 2:
  */
 static void do_LG2( struct arb_vp_machine *m, union instruction op ) 
@@ -566,7 +580,7 @@ static void print_reg( GLuint file, GLuint reg )
 	 _mesa_printf("TMP%d", reg - REG_TMP0);
       else if (reg >= REG_IN0 && reg <= REG_IN31)
 	 _mesa_printf("IN%d", reg - REG_IN0);
-      else if (reg >= REG_OUT0 && reg <= REG_OUT14)
+      else if (reg >= REG_OUT0 && reg <= REG_OUT23)
 	 _mesa_printf("OUT%d", reg - REG_OUT0);
       else if (reg == REG_ADDR)
 	 _mesa_printf("ADDR");
@@ -662,6 +676,7 @@ _tnl_disassem_vba_insn( union instruction op )
    case OPCODE_EXP:
    case OPCODE_FLR:
    case OPCODE_FRC:
+   case OPCODE_INT:
    case OPCODE_LG2:
    case OPCODE_LIT:
    case OPCODE_LOG:
@@ -711,18 +726,28 @@ _tnl_disassem_vba_insn( union instruction op )
    }
 }
 
+typedef void (*gpu_function)(struct arb_vp_machine *m, union instruction op);
 
-static void (* const opcode_func[MAX_OPCODE+3])(struct arb_vp_machine *, union instruction) = 
+static gpu_function opcode_func[MAX_OPCODE+3] =
 {
+   do_NOP,
    do_ABS,
    do_ADD,
    do_NOP,/*ARA*/
    do_NOP,/*ARL*/
    do_NOP,/*ARL_NV*/
    do_NOP,/*ARR*/
+   do_NOP,/*BGNLOOP*/
+   do_NOP,/*BGNSUB*/
    do_NOP,/*BRA*/
+   do_NOP,/*BRK*/
+   do_NOP,/*BRK0*/
+   do_NOP,/*BRK1*/
    do_NOP,/*CAL*/
    do_NOP,/*CMP*/
+   do_NOP,/*CONT*/
+   do_NOP,/*CONT0*/
+   do_NOP,/*CONT1*/
    do_NOP,/*COS*/
    do_NOP,/*DDX*/
    do_NOP,/*DDY*/
@@ -730,11 +755,17 @@ static void (* const opcode_func[MAX_OPCODE+3])(struct arb_vp_machine *, union i
    do_DP4,
    do_DPH,
    do_DST,
-   do_NOP,
+   do_NOP,/*ELSE*/
+   do_NOP,/*END*/
+   do_NOP,/*ENDIF*/
+   do_NOP,/*ENDLOOP*/
+   do_NOP,/*ENDSUB*/
    do_EX2,
    do_EXP,
    do_FLR,
    do_FRC,
+   do_NOP,/*IF*/
+   do_INT,
    do_NOP,/*KIL*/
    do_NOP,/*KIL_NV*/
    do_LG2,
@@ -746,6 +777,10 @@ static void (* const opcode_func[MAX_OPCODE+3])(struct arb_vp_machine *, union i
    do_MIN,
    do_MOV,
    do_MUL,
+   do_NOP,/*NOISE1*/
+   do_NOP,/*NOISE2*/
+   do_NOP,/*NOISE3*/
+   do_NOP,/*NOISE4*/
    do_NOP,/*PK2H*/
    do_NOP,/*PK2US*/
    do_NOP,/*PK4B*/
@@ -828,6 +863,7 @@ static struct reg cvp_load_reg( struct compilation *cp,
 
    switch (file) {
    case PROGRAM_TEMPORARY:
+      assert(REG_TMP0 + index <= REG_TMP11);
       return cvp_make_reg(FILE_REG, REG_TMP0 + index);
 
    case PROGRAM_INPUT:
@@ -853,6 +889,8 @@ static struct reg cvp_load_reg( struct compilation *cp,
 	 return reg;
 
    case PROGRAM_STATE_VAR:
+   case PROGRAM_CONSTANT:
+   case PROGRAM_UNIFORM:
       reg = cvp_make_reg(FILE_STATE_PARAM, index);
       if (rel) 
 	 return cvp_emit_rel(cp, reg, tmpreg);
@@ -864,7 +902,7 @@ static struct reg cvp_load_reg( struct compilation *cp,
    case PROGRAM_WRITE_ONLY:
    case PROGRAM_ADDRESS:
    default:
-      _mesa_problem(NULL, "Invalid register file %d in cvp_load_reg()");
+      _mesa_problem(NULL, "Invalid register file %d in cvp_load_reg()", file);
       assert(0);
       return tmpreg;		/* can't happen */
    }
@@ -925,7 +963,10 @@ static GLuint cvp_choose_result( struct compilation *cp,
       idx = REG_OUT0 + dst->Index;
       break;
    default:
+#if 0
+      /* IF/ELSE/ENDIF instructions will hit this */
       assert(0);
+#endif
       return REG_RES;		/* can't happen */
    }
 
@@ -1020,6 +1061,13 @@ static void cvp_emit_inst( struct compilation *cp,
       }
       break;
 
+   case OPCODE_NOP:
+      break;
+
+   case OPCODE_BRA:
+      /* XXX implement */
+      break;
+
    default:
       result = cvp_choose_result( cp, &inst->DstReg, &fixup );
       nr_args = _mesa_num_inst_src_regs(inst->Opcode);
@@ -1058,6 +1106,16 @@ static void compile_vertex_program( struct gl_vertex_program *program,
    struct tnl_compiled_program *p = CALLOC_STRUCT(tnl_compiled_program);
    GLint i;
 
+#if 1
+   if (!program->IsNVProgram && program->IsPositionInvariant) {
+      printf("Adding MVP code\n");
+      if (!program->Base.Parameters)
+         program->Base.Parameters = _mesa_new_parameter_list();
+      _mesa_insert_mvp_code(NULL, program);
+      program->IsPositionInvariant = 0;
+   }
+#endif
+
    if (program->TnlData) 
       free_tnl_data( program );
    
@@ -1097,7 +1155,6 @@ static void compile_vertex_program( struct gl_vertex_program *program,
 
 
 
-
 /* ----------------------------------------------------------------------
  * Execution
  */
@@ -1182,6 +1239,7 @@ do_ndc_cliptest(GLcontext *ctx, struct arb_vp_machine *m)
 
    /* Test userclip planes.  This contributes to VB->ClipMask.
     */
+   /** XXX NEW_SLANG _Enabled ??? */
    if (ctx->Transform.ClipPlanesEnabled && (!ctx->VertexProgram._Enabled ||
       ctx->VertexProgram.Current->IsPositionInvariant)) {
       userclip( ctx,
@@ -1209,6 +1267,7 @@ static INLINE void call_func( struct tnl_compiled_program *p,
    p->compiled_func(m);
 }
 
+
 /**
  * Execute the given vertex program.  
  * 
@@ -1221,22 +1280,20 @@ static INLINE void call_func( struct tnl_compiled_program *p,
 static GLboolean
 run_arb_vertex_program(GLcontext *ctx, struct tnl_pipeline_stage *stage)
 {
-   const struct gl_vertex_program *program;
+   const struct gl_vertex_program *program = ctx->VertexProgram._Current;
    struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
    struct arb_vp_machine *m = ARB_VP_MACHINE(stage);
    struct tnl_compiled_program *p;
    GLuint i, j;
    GLbitfield outputs;
 
-   if (ctx->ShaderObjects._VertexShaderPresent)
-      return GL_TRUE;
-
-   program = ctx->VertexProgram._Enabled ? ctx->VertexProgram.Current : NULL;
-   if (!program && ctx->_MaintainTnlProgram) {
-      program = ctx->_TnlProgram;
-   }
-   if (!program || program->IsNVProgram)
+#define FORCE_PROG_EXECUTE_C 1
+#if FORCE_PROG_EXECUTE_C
+   return GL_TRUE;   
+#else
+   if (!program)
       return GL_TRUE;   
+#endif
 
    if (program->Base.Parameters) {
       _mesa_load_state_parameters(ctx, program->Base.Parameters);
@@ -1279,7 +1336,13 @@ run_arb_vertex_program(GLcontext *ctx, struct tnl_pipeline_stage *stage)
 	 case 2: m->File[0][idx][1] = m->input[j].data[1];
 	 case 1: m->File[0][idx][0] = m->input[j].data[0];
 	 }
-
+#if 0
+         printf("  attr %d/%d: %g %g %g %g\n", j, idx-REG_IN0,
+                m->input[j].data[0],
+                m->input[j].data[1],
+                m->input[j].data[2],
+                m->input[j].data[3]);
+#endif
 	 STRIDE_F(m->input[j].data, m->input[j].stride);
       }
 
@@ -1288,9 +1351,9 @@ run_arb_vertex_program(GLcontext *ctx, struct tnl_pipeline_stage *stage)
 	 call_func( p, m );
       }
       else {
-         GLint j;
-	 for (j = 0; j < p->nr_instructions; j++) {
-	    union instruction inst = p->instructions[j];	 
+         GLint pc;
+	 for (pc = 0; pc < p->nr_instructions; pc++) {
+	    union instruction inst = p->instructions[pc];	 
 	    opcode_func[inst.alu.opcode]( m, inst );
 	 }
       }
@@ -1384,6 +1447,14 @@ run_arb_vertex_program(GLcontext *ctx, struct tnl_pipeline_stage *stage)
       }
    }
 
+   for (i = 0; i < ctx->Const.MaxVarying; i++) {
+      if (outputs & (1 << (VERT_RESULT_VAR0 + i))) {
+         /* Note: varying results get put into the generic attributes */
+	 VB->AttribPtr[VERT_ATTRIB_GENERIC0+i]
+            = &m->attribs[VERT_RESULT_VAR0 + i];
+      }
+   }
+
 #if 0
    for (i = 0; i < VB->Count; i++) {
       printf("Out %d: %f %f %f %f %f %f %f %f\n", i,
@@ -1408,17 +1479,13 @@ static void
 validate_vertex_program( GLcontext *ctx, struct tnl_pipeline_stage *stage )
 {
    struct arb_vp_machine *m = ARB_VP_MACHINE(stage);
-   struct gl_vertex_program *program;
-
-   if (ctx->ShaderObjects._VertexShaderPresent)
-      return;
-
-   program = (ctx->VertexProgram._Enabled ? ctx->VertexProgram.Current : 0);
-   if (!program && ctx->_MaintainTnlProgram) {
-      program = ctx->_TnlProgram;
-   }
+   struct gl_vertex_program *program = ctx->VertexProgram._Current;
 
+#if FORCE_OLD
+   if (0 &&program) {
+#else
    if (program) {
+#endif
       if (!program->TnlData)
 	 compile_vertex_program( program, m->try_codegen );
       
@@ -1435,11 +1502,6 @@ validate_vertex_program( GLcontext *ctx, struct tnl_pipeline_stage *stage )
 }
 
 
-
-
-
-
-
 /**
  * Called the first time stage->run is called.  In effect, don't
  * allocate data until the first time the stage is run.
@@ -1453,6 +1515,10 @@ static GLboolean init_vertex_program( GLcontext *ctx,
    const GLuint size = VB->Size;
    GLuint i;
 
+   /* spot checks to be sure the opcode table is correct */
+   assert(opcode_func[OPCODE_SGE] == do_SGE);
+   assert(opcode_func[OPCODE_XPD] == do_XPD);
+
    stage->privatePtr = _mesa_calloc(sizeof(*m));
    m = ARB_VP_MACHINE(stage);
    if (!m)
@@ -1486,7 +1552,7 @@ static GLboolean init_vertex_program( GLcontext *ctx,
    _mesa_vector4f_alloc( &m->ndcCoords, 0, size, 32 );
    m->clipmask = (GLubyte *) ALIGN_MALLOC(sizeof(GLubyte)*size, 32 );
 
-   if (ctx->_MaintainTnlProgram)
+   if (ctx->VertexProgram._MaintainTnlProgram)
       _mesa_allow_light_in_model( ctx, GL_FALSE );
 
    m->fpucntl_rnd_neg = RND_NEG_FPU; /* const value */
@@ -1496,8 +1562,6 @@ static GLboolean init_vertex_program( GLcontext *ctx,
 }
 
 
-
-
 /**
  * Destructor for this pipeline stage.
  */
@@ -1522,6 +1586,7 @@ static void dtr( struct tnl_pipeline_stage *stage )
    }
 }
 
+
 /**
  * Public description of this pipeline stage.
  */
@@ -1543,7 +1608,7 @@ const struct tnl_pipeline_stage _tnl_arb_vertex_program_stage =
 void
 _tnl_program_string(GLcontext *ctx, GLenum target, struct gl_program *program)
 {
-   if (target == GL_VERTEX_PROGRAM_ARB) {
+   if (program->Target == GL_VERTEX_PROGRAM_ARB) {
       /* free any existing tnl data hanging off the program */
       struct gl_vertex_program *vprog = (struct gl_vertex_program *) program;
       if (vprog->TnlData) {
diff --git a/src/mesa/tnl/t_vb_arbprogram.h b/src/mesa/tnl/t_vb_arbprogram.h
index 6de8dca328c..1bec2a411e8 100644
--- a/src/mesa/tnl/t_vb_arbprogram.h
+++ b/src/mesa/tnl/t_vb_arbprogram.h
@@ -56,6 +56,22 @@
 #define REG_TMP0   5
 #define REG_TMP11  16
 #define REG_OUT0   17
+#define REG_OUT23  40
+#define REG_IN0    41
+#define REG_IN31   72
+#define REG_ID     73		/* 0,0,0,1 */
+#define REG_ONES   74		/* 1,1,1,1 */
+#define REG_SWZ    75		/* 1,-1,0,0 */
+#define REG_NEG    76		/* -1,-1,-1,-1 */
+#define REG_LIT    77           /* 1,0,0,1 */
+#define REG_LIT2    78           /* 1,0,0,1 */
+#define REG_SCRATCH 79		/* internal temporary. XXX we can't actually use this because 70 doesn't fit in the 5-bit 'dst' instruction field! */
+#define REG_UNDEF  127		/* special case - never used */
+#define REG_MAX    128
+#define REG_INVALID ~0
+
+
+#if 0
 #define REG_OUT14  31
 #define REG_IN0    32
 #define REG_IN31   63
@@ -69,6 +85,7 @@
 #define REG_UNDEF  127		/* special case - never used */
 #define REG_MAX    128
 #define REG_INVALID ~0
+#endif
 
 /* ARB_vp instructions are broken down into one or more of the
  * following micro-instructions, each representable in a 64 bit packed
@@ -76,16 +93,16 @@
  */
 struct reg {
    GLuint file:2;
-   GLuint idx:7;
+   GLuint idx:8;
 };
 
 
 union instruction {
    struct {
       GLuint opcode:7;
-      GLuint dst:5;
+      GLuint dst:6;
       GLuint file0:2;
-      GLuint idx0:7;
+      GLuint idx0:8;
       GLuint file1:2;
       GLuint idx1:7;
       GLuint pad:2;
@@ -94,18 +111,18 @@ union instruction {
 
    struct {
       GLuint opcode:7;
-      GLuint dst:5;
+      GLuint dst:6;
       GLuint file0:2;
-      GLuint idx0:7;
+      GLuint idx0:8;
       GLuint neg:4;
       GLuint swz:12;		/* xyzw01 */
    } rsw;
 
    struct {
       GLuint opcode:7;
-      GLuint dst:5;
+      GLuint dst:6;
       GLuint file:2;
-      GLuint idx:7;
+      GLuint idx:8;
       GLuint mask:4;
       GLuint pad:7;
       GLuint pad2;
diff --git a/src/mesa/tnl/t_vb_arbprogram_sse.c b/src/mesa/tnl/t_vb_arbprogram_sse.c
index b9126d6d886..d0f66fd6196 100644
--- a/src/mesa/tnl/t_vb_arbprogram_sse.c
+++ b/src/mesa/tnl/t_vb_arbprogram_sse.c
@@ -43,7 +43,7 @@
 #include "mtypes.h"
 #include "arbprogparse.h"
 #include "program.h"
-#include "program_instruction.h"
+#include "prog_instruction.h"
 #include "math/m_matrix.h"
 #include "math/m_translate.h"
 #include "t_context.h"
diff --git a/src/mesa/tnl/t_vb_arbshader.c b/src/mesa/tnl/t_vb_arbshader.c
deleted file mode 100644
index 13aa3ea9105..00000000000
--- a/src/mesa/tnl/t_vb_arbshader.c
+++ /dev/null
@@ -1,301 +0,0 @@
-/*
- * Mesa 3-D graphics library
- * Version:  6.5
- *
- * Copyright (C) 2006  Brian Paul   All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * the rights to use, copy, modify, merge, publish, distribute, sublicense,
- * and/or sell copies of the Software, and to permit persons to whom the
- * Software is furnished to do so, subject to the following conditions:
- *
- * The above copyright notice and this permission notice shall be included
- * in all copies or substantial portions of the Software.
- *
- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
- * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
- * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
- * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
- * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
- * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
- *
- * Authors:
- *    Michal Krol
- */
-
-#include "glheader.h"
-#include "imports.h"
-#include "macros.h"
-#include "shaderobjects.h"
-#include "shaderobjects_3dlabs.h"
-#include "t_pipeline.h"
-#include "slang_utility.h"
-#include "slang_link.h"
-
-#if FEATURE_ARB_vertex_shader
-
-typedef struct
-{
-	GLvector4f outputs[VERT_RESULT_MAX];
-	GLvector4f varyings[MAX_VARYING_VECTORS];
-	GLvector4f ndc_coords;
-	GLubyte *clipmask;
-	GLubyte ormask;
-	GLubyte andmask;
-} arbvs_stage_data;
-
-#define ARBVS_STAGE_DATA(stage) ((arbvs_stage_data *) stage->privatePtr)
-
-static GLboolean construct_arb_vertex_shader (GLcontext *ctx, struct tnl_pipeline_stage *stage)
-{
-	TNLcontext *tnl = TNL_CONTEXT(ctx);
-	struct vertex_buffer *vb = &tnl->vb;
-	arbvs_stage_data *store;
-	GLuint size = vb->Size;
-	GLuint i;
-
-	stage->privatePtr = _mesa_malloc (sizeof (arbvs_stage_data));
-	store = ARBVS_STAGE_DATA(stage);
-	if (store == NULL)
-		return GL_FALSE;
-
-	for (i = 0; i < VERT_RESULT_MAX; i++)
-	{
-		_mesa_vector4f_alloc (&store->outputs[i], 0, size, 32);
-		store->outputs[i].size = 4;
-	}
-	for (i = 0; i < MAX_VARYING_VECTORS; i++)
-	{
-		_mesa_vector4f_alloc (&store->varyings[i], 0, size, 32);
-		store->varyings[i].size = 4;
-	}
-	_mesa_vector4f_alloc (&store->ndc_coords, 0, size, 32);
-	store->clipmask = (GLubyte *) ALIGN_MALLOC (size, 32);
-
-	return GL_TRUE;
-}
-
-static void destruct_arb_vertex_shader (struct tnl_pipeline_stage *stage)
-{
-	arbvs_stage_data *store = ARBVS_STAGE_DATA(stage);
-
-	if (store != NULL)
-	{
-		GLuint i;
-
-		for (i = 0; i < VERT_RESULT_MAX; i++)
-			_mesa_vector4f_free (&store->outputs[i]);
-		for (i = 0; i < MAX_VARYING_VECTORS; i++)
-			_mesa_vector4f_free (&store->varyings[i]);
-		_mesa_vector4f_free (&store->ndc_coords);
-		ALIGN_FREE (store->clipmask);
-
-		_mesa_free (store);
-		stage->privatePtr = NULL;
-	}
-}
-
-static void validate_arb_vertex_shader (GLcontext *ctx, struct tnl_pipeline_stage *stage)
-{
-}
-
-static GLvoid fetch_input_float (struct gl2_program_intf **pro, GLuint index, GLuint attr, GLuint i,
-	struct vertex_buffer *vb)
-{
-	const GLubyte *ptr = (const GLubyte *) vb->AttribPtr[attr]->data;
-	const GLuint stride = vb->AttribPtr[attr]->stride;
-   GLfloat *data = (GLfloat *) (ptr + stride * i);
-
-   (**pro).UpdateFixedAttrib (pro, index, data, 0, sizeof (GLfloat), GL_TRUE);
-}
-
-static GLvoid fetch_input_vec3 (struct gl2_program_intf **pro, GLuint index, GLuint attr, GLuint i,
-	struct vertex_buffer *vb)
-{
-	const GLubyte *ptr = (const GLubyte *) vb->AttribPtr[attr]->data;
-	const GLuint stride = vb->AttribPtr[attr]->stride;
-   GLfloat *data = (GLfloat *) (ptr + stride * i);
-
-   (**pro).UpdateFixedAttrib (pro, index, data, 0, 3 * sizeof (GLfloat), GL_TRUE);
-}
-
-static void fetch_input_vec4 (struct gl2_program_intf **pro, GLuint index, GLuint attr, GLuint i,
-	struct vertex_buffer *vb)
-{
-	const GLubyte *ptr = (const GLubyte *) vb->AttribPtr[attr]->data;
-	const GLuint size = vb->AttribPtr[attr]->size;
-	const GLuint stride = vb->AttribPtr[attr]->stride;
-	const GLfloat *data = (const GLfloat *) (ptr + stride * i);
-	GLfloat vec[4];
-
-	switch (size)
-	{
-	case 2:
-		vec[0] = data[0];
-		vec[1] = data[1];
-		vec[2] = 0.0f;
-		vec[3] = 1.0f;
-		break;
-	case 3:
-		vec[0] = data[0];
-		vec[1] = data[1];
-		vec[2] = data[2];
-		vec[3] = 1.0f;
-		break;
-	case 4:
-		vec[0] = data[0];
-		vec[1] = data[1];
-		vec[2] = data[2];
-		vec[3] = data[3];
-		break;
-	}
-   (**pro).UpdateFixedAttrib (pro, index, vec, 0, 4 * sizeof (GLfloat), GL_TRUE);
-}
-
-static GLvoid
-fetch_gen_attrib (struct gl2_program_intf **pro, GLuint index, GLuint i, struct vertex_buffer *vb)
-{
-   const GLuint attr = _TNL_ATTRIB_GENERIC0 + index;
-   const GLubyte *ptr = (const GLubyte *) (vb->AttribPtr[attr]->data);
-   const GLuint stride = vb->AttribPtr[attr]->stride;
-   const GLfloat *data = (const GLfloat *) (ptr + stride * i);
-
-   (**pro).WriteAttrib (pro, index, data);
-}
-
-static GLvoid fetch_output_float (struct gl2_program_intf **pro, GLuint index, GLuint attr, GLuint i,
-	arbvs_stage_data *store)
-{
-   (**pro).UpdateFixedAttrib (pro, index, &store->outputs[attr].data[i], 0, sizeof (GLfloat),
-                              GL_FALSE);
-}
-
-static void fetch_output_vec4 (struct gl2_program_intf **pro, GLuint index, GLuint attr, GLuint i,
-	GLuint offset, arbvs_stage_data *store)
-{
-   (**pro).UpdateFixedAttrib (pro, index, &store->outputs[attr].data[i], offset,
-                              4 * sizeof (GLfloat), GL_FALSE);
-}
-
-static GLboolean run_arb_vertex_shader (GLcontext *ctx, struct tnl_pipeline_stage *stage)
-{
-	TNLcontext *tnl = TNL_CONTEXT(ctx);
-	struct vertex_buffer *vb = &tnl->vb;
-	arbvs_stage_data *store = ARBVS_STAGE_DATA(stage);
-	struct gl2_program_intf **pro;
-	GLsizei i, j;
-
-	if (!ctx->ShaderObjects._VertexShaderPresent)
-		return GL_TRUE;
-
-	pro = ctx->ShaderObjects.CurrentProgram;
-	(**pro).UpdateFixedUniforms (pro);
-
-	for (i = 0; i < vb->Count; i++)
-	{
-		fetch_input_vec4 (pro, SLANG_VERTEX_FIXED_VERTEX, _TNL_ATTRIB_POS, i, vb);
-		fetch_input_vec3 (pro, SLANG_VERTEX_FIXED_NORMAL, _TNL_ATTRIB_NORMAL, i, vb);
-		fetch_input_vec4 (pro, SLANG_VERTEX_FIXED_COLOR, _TNL_ATTRIB_COLOR0, i, vb);
-		fetch_input_vec4 (pro, SLANG_VERTEX_FIXED_SECONDARYCOLOR, _TNL_ATTRIB_COLOR1, i, vb);
-		fetch_input_float (pro, SLANG_VERTEX_FIXED_FOGCOORD, _TNL_ATTRIB_FOG, i, vb);
-		fetch_input_vec4 (pro, SLANG_VERTEX_FIXED_MULTITEXCOORD0, _TNL_ATTRIB_TEX0, i, vb);
-		fetch_input_vec4 (pro, SLANG_VERTEX_FIXED_MULTITEXCOORD1, _TNL_ATTRIB_TEX1, i, vb);
-		fetch_input_vec4 (pro, SLANG_VERTEX_FIXED_MULTITEXCOORD2, _TNL_ATTRIB_TEX2, i, vb);
-		fetch_input_vec4 (pro, SLANG_VERTEX_FIXED_MULTITEXCOORD3, _TNL_ATTRIB_TEX3, i, vb);
-		fetch_input_vec4 (pro, SLANG_VERTEX_FIXED_MULTITEXCOORD4, _TNL_ATTRIB_TEX4, i, vb);
-		fetch_input_vec4 (pro, SLANG_VERTEX_FIXED_MULTITEXCOORD5, _TNL_ATTRIB_TEX5, i, vb);
-		fetch_input_vec4 (pro, SLANG_VERTEX_FIXED_MULTITEXCOORD6, _TNL_ATTRIB_TEX6, i, vb);
-		fetch_input_vec4 (pro, SLANG_VERTEX_FIXED_MULTITEXCOORD7, _TNL_ATTRIB_TEX7, i, vb);
-      for (j = 0; j < MAX_VERTEX_ATTRIBS; j++)
-         fetch_gen_attrib (pro, j, i, vb);
-
-		_slang_exec_vertex_shader (pro);
-
-		fetch_output_vec4 (pro, SLANG_VERTEX_FIXED_POSITION, VERT_RESULT_HPOS, i, 0, store);
-		fetch_output_vec4 (pro, SLANG_VERTEX_FIXED_FRONTCOLOR, VERT_RESULT_COL0, i, 0, store);
-		fetch_output_vec4 (pro, SLANG_VERTEX_FIXED_FRONTSECONDARYCOLOR, VERT_RESULT_COL1, i, 0, store);
-		fetch_output_float (pro, SLANG_VERTEX_FIXED_FOGFRAGCOORD, VERT_RESULT_FOGC, i, store);
-		for (j = 0; j < 8; j++)
-			fetch_output_vec4 (pro, SLANG_VERTEX_FIXED_TEXCOORD, VERT_RESULT_TEX0 + j, i, j, store);
-		fetch_output_float (pro, SLANG_VERTEX_FIXED_POINTSIZE, VERT_RESULT_PSIZ, i, store);
-		fetch_output_vec4 (pro, SLANG_VERTEX_FIXED_BACKCOLOR, VERT_RESULT_BFC0, i, 0, store);
-		fetch_output_vec4 (pro, SLANG_VERTEX_FIXED_BACKSECONDARYCOLOR, VERT_RESULT_BFC1, i, 0, store);
-		/* XXX: fetch output SLANG_VERTEX_FIXED_CLIPVERTEX */
-
-		for (j = 0; j < MAX_VARYING_VECTORS; j++)
-		{
-			GLuint k;
-
-			for (k = 0; k < VARYINGS_PER_VECTOR; k++)
-			{
-				(**pro).UpdateVarying (pro, j * VARYINGS_PER_VECTOR + k,
-					&store->varyings[j].data[i][k], GL_TRUE);
-			}
-		}
-	}
-
-	vb->ClipPtr = &store->outputs[VERT_RESULT_HPOS];
-	vb->ClipPtr->count = vb->Count;
-
-	vb->ColorPtr[0] = &store->outputs[VERT_RESULT_COL0];
-	vb->AttribPtr[VERT_ATTRIB_COLOR0] = vb->ColorPtr[0];
-	vb->ColorPtr[1] = &store->outputs[VERT_RESULT_BFC0];
-
-	vb->SecondaryColorPtr[0] =
-	vb->AttribPtr[VERT_ATTRIB_COLOR1] = &store->outputs[VERT_RESULT_COL1];
-
-	vb->SecondaryColorPtr[1] = &store->outputs[VERT_RESULT_BFC1];
-
-	for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
-		vb->TexCoordPtr[i] =
-		vb->AttribPtr[VERT_ATTRIB_TEX0 + i] = &store->outputs[VERT_RESULT_TEX0 + i];
-        }
-
-	vb->FogCoordPtr =
-	vb->AttribPtr[VERT_ATTRIB_FOG] = &store->outputs[VERT_RESULT_FOGC];
-
-	vb->AttribPtr[_TNL_ATTRIB_POINTSIZE] = &store->outputs[VERT_RESULT_PSIZ];
-
-	for (i = 0; i < MAX_VARYING_VECTORS; i++) {
-		vb->VaryingPtr[i] = &store->varyings[i];
-		vb->AttribPtr[_TNL_ATTRIB_GENERIC0 + i] = vb->VaryingPtr[i];
-        }
-
-	store->ormask = 0;
-	store->andmask = CLIP_FRUSTUM_BITS;
-
-	if (tnl->NeedNdcCoords)
-	{
-		vb->NdcPtr = _mesa_clip_tab[vb->ClipPtr->size] (vb->ClipPtr, &store->ndc_coords,
-			store->clipmask, &store->ormask, &store->andmask);
-	}
-	else
-	{
-		vb->NdcPtr = NULL;
-		_mesa_clip_np_tab[vb->ClipPtr->size] (vb->ClipPtr, NULL, store->clipmask, &store->ormask,
-			&store->andmask);
-	}
-
-	if (store->andmask)
-		return GL_FALSE;
-
-	vb->ClipAndMask = store->andmask;
-	vb->ClipOrMask = store->ormask;
-	vb->ClipMask = store->clipmask;
-
-	return GL_TRUE;
-}
-
-const struct tnl_pipeline_stage _tnl_arb_vertex_shader_stage = {
-	"ARB_vertex_shader",
-	NULL,
-	construct_arb_vertex_shader,
-	destruct_arb_vertex_shader,
-	validate_arb_vertex_shader,
-	run_arb_vertex_shader
-};
-
-#endif /* FEATURE_ARB_vertex_shader */
-
diff --git a/src/mesa/tnl/t_vb_cull.c b/src/mesa/tnl/t_vb_cull.c
index 8848dac10c1..21a32e5b1d8 100644
--- a/src/mesa/tnl/t_vb_cull.c
+++ b/src/mesa/tnl/t_vb_cull.c
@@ -57,10 +57,7 @@ static GLboolean run_cull_stage( GLcontext *ctx,
    GLuint count = VB->Count;
    GLuint i;
 
-   if (ctx->ShaderObjects._VertexShaderPresent)
-      return GL_TRUE;
-
-   if (ctx->VertexProgram._Enabled ||
+   if (ctx->VertexProgram._Current ||
        !ctx->Transform.CullVertexFlag) 
       return GL_TRUE;
 
diff --git a/src/mesa/tnl/t_vb_fog.c b/src/mesa/tnl/t_vb_fog.c
index 51f28c4059f..5440ff7894d 100644
--- a/src/mesa/tnl/t_vb_fog.c
+++ b/src/mesa/tnl/t_vb_fog.c
@@ -148,10 +148,7 @@ run_fog_stage(GLcontext *ctx, struct tnl_pipeline_stage *stage)
    struct fog_stage_data *store = FOG_STAGE_DATA(stage);
    GLvector4f *input;
 
-   if (ctx->ShaderObjects._VertexShaderPresent)
-      return GL_TRUE;
-
-   if (!ctx->Fog.Enabled || ctx->VertexProgram._Enabled)
+   if (!ctx->Fog.Enabled || ctx->VertexProgram._Current)
       return GL_TRUE;
 
 
diff --git a/src/mesa/tnl/t_vb_light.c b/src/mesa/tnl/t_vb_light.c
index 47c5b400aae..12f2cc7735e 100644
--- a/src/mesa/tnl/t_vb_light.c
+++ b/src/mesa/tnl/t_vb_light.c
@@ -203,10 +203,7 @@ static GLboolean run_lighting( GLcontext *ctx,
    GLvector4f *input = ctx->_NeedEyeCoords ? VB->EyePtr : VB->ObjPtr;
    GLuint idx;
 
-   if (ctx->ShaderObjects._VertexShaderPresent)
-      return GL_TRUE;
-
-   if (!ctx->Light.Enabled || ctx->VertexProgram._Enabled)
+   if (!ctx->Light.Enabled || ctx->VertexProgram._Current)
       return GL_TRUE;
 
    /* Make sure we can talk about position x,y and z:
@@ -264,10 +261,7 @@ static void validate_lighting( GLcontext *ctx,
 {
    light_func *tab;
 
-   if (ctx->ShaderObjects._VertexShaderPresent)
-      return;
-
-   if (!ctx->Light.Enabled || ctx->VertexProgram._Enabled)
+   if (!ctx->Light.Enabled || ctx->VertexProgram._Current)
       return;
 
    if (ctx->Visual.rgbMode) {
diff --git a/src/mesa/tnl/t_vb_normals.c b/src/mesa/tnl/t_vb_normals.c
index 0f91784e809..01fad0cee21 100644
--- a/src/mesa/tnl/t_vb_normals.c
+++ b/src/mesa/tnl/t_vb_normals.c
@@ -95,12 +95,7 @@ validate_normal_stage(GLcontext *ctx, struct tnl_pipeline_stage *stage)
 {
    struct normal_stage_data *store = NORMAL_STAGE_DATA(stage);
 
-   if (ctx->ShaderObjects._VertexShaderPresent) {
-      store->NormalTransform = NULL;
-      return;
-   }
-
-   if (ctx->VertexProgram._Enabled ||
+   if (ctx->VertexProgram._Current ||
        (!ctx->Light.Enabled &&
 	!(ctx->Texture._GenFlags & TEXGEN_NEED_NORMALS))) {
       store->NormalTransform = NULL;
diff --git a/src/mesa/tnl/t_vb_points.c b/src/mesa/tnl/t_vb_points.c
index 2f502d61d1d..9327f8c273c 100644
--- a/src/mesa/tnl/t_vb_points.c
+++ b/src/mesa/tnl/t_vb_points.c
@@ -47,10 +47,7 @@ struct point_stage_data {
 static GLboolean
 run_point_stage(GLcontext *ctx, struct tnl_pipeline_stage *stage)
 {
-   if (ctx->ShaderObjects._VertexShaderPresent)
-      return GL_TRUE;
-
-   if (ctx->Point._Attenuated && !ctx->VertexProgram._Enabled) {
+   if (ctx->Point._Attenuated && !ctx->VertexProgram._Current) {
       struct point_stage_data *store = POINT_STAGE_DATA(stage);
       struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
       const GLfloat (*eye)[4] = (const GLfloat (*)[4]) VB->EyePtr->data;
diff --git a/src/mesa/tnl/t_vb_program.c b/src/mesa/tnl/t_vb_program.c
index f11ac616f16..17eefe70326 100644
--- a/src/mesa/tnl/t_vb_program.c
+++ b/src/mesa/tnl/t_vb_program.c
@@ -1,8 +1,8 @@
 /*
  * Mesa 3-D graphics library
- * Version:  6.5
+ * Version:  6.5.3
  *
- * Copyright (C) 1999-2006  Brian Paul   All Rights Reserved.
+ * Copyright (C) 1999-2007  Brian Paul   All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -31,20 +31,14 @@
 
 
 #include "glheader.h"
-#include "api_noop.h"
-#include "colormac.h"
 #include "context.h"
-#include "dlist.h"
-#include "hash.h"
-#include "light.h"
 #include "macros.h"
 #include "imports.h"
-#include "simple_list.h"
-#include "mtypes.h"
-#include "program_instruction.h"
-#include "nvvertexec.h"
-#include "nvprogram.h"
+#include "prog_instruction.h"
+#include "prog_statevars.h"
+#include "prog_execute.h"
 
+#include "tnl.h"
 #include "t_context.h"
 #include "t_pipeline.h"
 
@@ -55,7 +49,7 @@
  */
 struct vp_stage_data {
    /** The results of running the vertex program go into these arrays. */
-   GLvector4f attribs[15];
+   GLvector4f results[VERT_RESULT_MAX];
 
    GLvector4f ndcCoords;              /**< normalized device coords */
    GLubyte *clipmask;                 /**< clip flags */
@@ -67,6 +61,131 @@ struct vp_stage_data {
 
 
 /**
+ * Initialize virtual machine state prior to executing vertex program.
+ */
+static void
+init_machine(GLcontext *ctx, struct gl_program_machine *machine)
+{
+   /* Input registers get initialized from the current vertex attribs */
+   MEMCPY(machine->VertAttribs, ctx->Current.Attrib,
+          MAX_VERTEX_PROGRAM_ATTRIBS * 4 * sizeof(GLfloat));
+
+   if (ctx->VertexProgram.Current->IsNVProgram) {
+      GLuint i;
+      /* Output/result regs are initialized to [0,0,0,1] */
+      for (i = 0; i < MAX_NV_VERTEX_PROGRAM_OUTPUTS; i++) {
+         ASSIGN_4V(machine->Outputs[i], 0.0F, 0.0F, 0.0F, 1.0F);
+      }
+      /* Temp regs are initialized to [0,0,0,0] */
+      for (i = 0; i < MAX_NV_VERTEX_PROGRAM_TEMPS; i++) {
+         ASSIGN_4V(machine->Temporaries[i], 0.0F, 0.0F, 0.0F, 0.0F);
+      }
+      for (i = 0; i < MAX_VERTEX_PROGRAM_ADDRESS_REGS; i++) {
+         ASSIGN_4V(machine->AddressReg[i], 0, 0, 0, 0);
+      }
+   }
+
+   /* init condition codes */
+   machine->CondCodes[0] = COND_EQ;
+   machine->CondCodes[1] = COND_EQ;
+   machine->CondCodes[2] = COND_EQ;
+   machine->CondCodes[3] = COND_EQ;
+}
+
+
+/**
+ * Copy the 16 elements of a matrix into four consecutive program
+ * registers starting at 'pos'.
+ */
+static void
+load_matrix(GLfloat registers[][4], GLuint pos, const GLfloat mat[16])
+{
+   GLuint i;
+   for (i = 0; i < 4; i++) {
+      registers[pos + i][0] = mat[0 + i];
+      registers[pos + i][1] = mat[4 + i];
+      registers[pos + i][2] = mat[8 + i];
+      registers[pos + i][3] = mat[12 + i];
+   }
+}
+
+
+/**
+ * As above, but transpose the matrix.
+ */
+static void
+load_transpose_matrix(GLfloat registers[][4], GLuint pos,
+                      const GLfloat mat[16])
+{
+   MEMCPY(registers[pos], mat, 16 * sizeof(GLfloat));
+}
+
+
+/**
+ * Load current vertex program's parameter registers with tracked
+ * matrices (if NV program).  This only needs to be done per
+ * glBegin/glEnd, not per-vertex.
+ */
+void
+_mesa_load_tracked_matrices(GLcontext *ctx)
+{
+   GLuint i;
+
+   for (i = 0; i < MAX_NV_VERTEX_PROGRAM_PARAMS / 4; i++) {
+      /* point 'mat' at source matrix */
+      GLmatrix *mat;
+      if (ctx->VertexProgram.TrackMatrix[i] == GL_MODELVIEW) {
+         mat = ctx->ModelviewMatrixStack.Top;
+      }
+      else if (ctx->VertexProgram.TrackMatrix[i] == GL_PROJECTION) {
+         mat = ctx->ProjectionMatrixStack.Top;
+      }
+      else if (ctx->VertexProgram.TrackMatrix[i] == GL_TEXTURE) {
+         mat = ctx->TextureMatrixStack[ctx->Texture.CurrentUnit].Top;
+      }
+      else if (ctx->VertexProgram.TrackMatrix[i] == GL_COLOR) {
+         mat = ctx->ColorMatrixStack.Top;
+      }
+      else if (ctx->VertexProgram.TrackMatrix[i]==GL_MODELVIEW_PROJECTION_NV) {
+         /* XXX verify the combined matrix is up to date */
+         mat = &ctx->_ModelProjectMatrix;
+      }
+      else if (ctx->VertexProgram.TrackMatrix[i] >= GL_MATRIX0_NV &&
+               ctx->VertexProgram.TrackMatrix[i] <= GL_MATRIX7_NV) {
+         GLuint n = ctx->VertexProgram.TrackMatrix[i] - GL_MATRIX0_NV;
+         ASSERT(n < MAX_PROGRAM_MATRICES);
+         mat = ctx->ProgramMatrixStack[n].Top;
+      }
+      else {
+         /* no matrix is tracked, but we leave the register values as-is */
+         assert(ctx->VertexProgram.TrackMatrix[i] == GL_NONE);
+         continue;
+      }
+
+      /* load the matrix values into sequential registers */
+      if (ctx->VertexProgram.TrackMatrixTransform[i] == GL_IDENTITY_NV) {
+         load_matrix(ctx->VertexProgram.Parameters, i*4, mat->m);
+      }
+      else if (ctx->VertexProgram.TrackMatrixTransform[i] == GL_INVERSE_NV) {
+         _math_matrix_analyse(mat); /* update the inverse */
+         ASSERT(!_math_matrix_is_dirty(mat));
+         load_matrix(ctx->VertexProgram.Parameters, i*4, mat->inv);
+      }
+      else if (ctx->VertexProgram.TrackMatrixTransform[i] == GL_TRANSPOSE_NV) {
+         load_transpose_matrix(ctx->VertexProgram.Parameters, i*4, mat->m);
+      }
+      else {
+         assert(ctx->VertexProgram.TrackMatrixTransform[i]
+                == GL_INVERSE_TRANSPOSE_NV);
+         _math_matrix_analyse(mat); /* update the inverse */
+         ASSERT(!_math_matrix_is_dirty(mat));
+         load_transpose_matrix(ctx->VertexProgram.Parameters, i*4, mat->inv);
+      }
+   }
+}
+
+
+/**
  * This function executes vertex programs
  */
 static GLboolean
@@ -75,24 +194,38 @@ run_vp( GLcontext *ctx, struct tnl_pipeline_stage *stage )
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    struct vp_stage_data *store = VP_STAGE_DATA(stage);
    struct vertex_buffer *VB = &tnl->vb;
-   struct gl_vertex_program *program = ctx->VertexProgram.Current;
-   struct vp_machine machine;
-   GLuint i;
-
-   if (ctx->ShaderObjects._VertexShaderPresent)
+   struct gl_vertex_program *program = ctx->VertexProgram._Current;
+   struct gl_program_machine machine;
+   GLuint outputs[VERT_RESULT_MAX], numOutputs;
+   GLuint i, j;
+
+#define FORCE_PROG_EXECUTE_C 1
+#if FORCE_PROG_EXECUTE_C
+   if (!program)
       return GL_TRUE;
-
-   if (!ctx->VertexProgram._Enabled ||
-       !program->IsNVProgram)
+#else
+   if (!program || !program->IsNVProgram)
       return GL_TRUE;
+#endif
 
-   /* load program parameter registers (they're read-only) */
-   _mesa_init_vp_per_primitive_registers(ctx);
+   if (ctx->VertexProgram.Current->IsNVProgram) {
+      _mesa_load_tracked_matrices(ctx);
+   }
+   else {
+      _mesa_load_state_parameters(ctx, program->Base.Parameters);
+   }
+
+   numOutputs = 0;
+   for (i = 0; i < VERT_RESULT_MAX; i++) {
+      if (program->Base.OutputsWritten & (1 << i)) {
+         outputs[numOutputs++] = i;
+      }
+   }
 
    for (i = 0; i < VB->Count; i++) {
       GLuint attr;
 
-      _mesa_init_vp_per_vertex_registers(ctx, &machine);
+      init_machine(ctx, &machine);
 
 #if 0
       printf("Input  %d: %f, %f, %f, %f\n", i,
@@ -119,13 +252,12 @@ run_vp( GLcontext *ctx, struct tnl_pipeline_stage *stage )
 	    const GLuint size = VB->AttribPtr[attr]->size;
 	    const GLuint stride = VB->AttribPtr[attr]->stride;
 	    const GLfloat *data = (GLfloat *) (ptr + stride * i);
-	    COPY_CLEAN_4V(machine.Inputs[attr], size, data);
+	    COPY_CLEAN_4V(machine.VertAttribs[attr], size, data);
 	 }
       }
 
       /* execute the program */
-      ASSERT(program);
-      _mesa_exec_vertex_program(ctx, &machine, program);
+      _mesa_execute_program(ctx, &program->Base, &machine);
 
       /* Fixup fog an point size results if needed */
       if (ctx->Fog.Enabled &&
@@ -139,33 +271,48 @@ run_vp( GLcontext *ctx, struct tnl_pipeline_stage *stage )
       }
 
       /* copy the output registers into the VB->attribs arrays */
-      /* XXX (optimize) could use a conditional and smaller loop limit here */
-      for (attr = 0; attr < 15; attr++) {
-         COPY_4V(store->attribs[attr].data[i], machine.Outputs[attr]);
+      for (j = 0; j < numOutputs; j++) {
+         const GLuint attr = outputs[j];
+         COPY_4V(store->results[attr].data[i], machine.Outputs[attr]);
       }
+#if 0
+      printf("HPOS: %f %f %f %f\n",
+             machine.Outputs[0][0], 
+             machine.Outputs[0][1], 
+             machine.Outputs[0][2], 
+             machine.Outputs[0][3]);
+#endif
    }
 
    /* Setup the VB pointers so that the next pipeline stages get
     * their data from the right place (the program output arrays).
     */
-   VB->ClipPtr = &store->attribs[VERT_RESULT_HPOS];
+   VB->ClipPtr = &store->results[VERT_RESULT_HPOS];
    VB->ClipPtr->size = 4;
    VB->ClipPtr->count = VB->Count;
-   VB->ColorPtr[0] = &store->attribs[VERT_RESULT_COL0];
-   VB->ColorPtr[1] = &store->attribs[VERT_RESULT_BFC0];
-   VB->SecondaryColorPtr[0] = &store->attribs[VERT_RESULT_COL1];
-   VB->SecondaryColorPtr[1] = &store->attribs[VERT_RESULT_BFC1];
-   VB->FogCoordPtr = &store->attribs[VERT_RESULT_FOGC];
+   VB->ColorPtr[0] = &store->results[VERT_RESULT_COL0];
+   VB->ColorPtr[1] = &store->results[VERT_RESULT_BFC0];
+   VB->SecondaryColorPtr[0] = &store->results[VERT_RESULT_COL1];
+   VB->SecondaryColorPtr[1] = &store->results[VERT_RESULT_BFC1];
+   VB->FogCoordPtr = &store->results[VERT_RESULT_FOGC];
 
-   VB->AttribPtr[VERT_ATTRIB_COLOR0] = &store->attribs[VERT_RESULT_COL0];
-   VB->AttribPtr[VERT_ATTRIB_COLOR1] = &store->attribs[VERT_RESULT_COL1];
-   VB->AttribPtr[VERT_ATTRIB_FOG] = &store->attribs[VERT_RESULT_FOGC];
-   VB->AttribPtr[_TNL_ATTRIB_POINTSIZE] = &store->attribs[VERT_RESULT_PSIZ];
+   VB->AttribPtr[VERT_ATTRIB_COLOR0] = &store->results[VERT_RESULT_COL0];
+   VB->AttribPtr[VERT_ATTRIB_COLOR1] = &store->results[VERT_RESULT_COL1];
+   VB->AttribPtr[VERT_ATTRIB_FOG] = &store->results[VERT_RESULT_FOGC];
+   VB->AttribPtr[_TNL_ATTRIB_POINTSIZE] = &store->results[VERT_RESULT_PSIZ];
 
    for (i = 0; i < ctx->Const.MaxTextureCoordUnits; i++) {
       VB->TexCoordPtr[i] = 
       VB->AttribPtr[_TNL_ATTRIB_TEX0 + i]
-         = &store->attribs[VERT_RESULT_TEX0 + i];
+         = &store->results[VERT_RESULT_TEX0 + i];
+   }
+
+   for (i = 0; i < ctx->Const.MaxVarying; i++) {
+      if (program->Base.OutputsWritten & (1 << (VERT_RESULT_VAR0 + i))) {
+         /* Note: varying results get put into the generic attributes */
+	 VB->AttribPtr[VERT_ATTRIB_GENERIC0+i]
+            = &store->results[VERT_RESULT_VAR0 + i];
+      }
    }
 
    /* Cliptest and perspective divide.  Clip functions must clear
@@ -206,8 +353,6 @@ run_vp( GLcontext *ctx, struct tnl_pipeline_stage *stage )
 }
 
 
-
-
 /**
  * Called the first time stage->run is called.  In effect, don't
  * allocate data until the first time the stage is run.
@@ -227,10 +372,9 @@ static GLboolean init_vp( GLcontext *ctx,
       return GL_FALSE;
 
    /* Allocate arrays of vertex output values */
-   /* XXX change '15' to a named constant */
-   for (i = 0; i < 15; i++) {
-      _mesa_vector4f_alloc( &store->attribs[i], 0, size, 32 );
-      store->attribs[i].size = 4;
+   for (i = 0; i < VERT_RESULT_MAX; i++) {
+      _mesa_vector4f_alloc( &store->results[i], 0, size, 32 );
+      store->results[i].size = 4;
    }
 
    /* a few other misc allocations */
@@ -241,9 +385,6 @@ static GLboolean init_vp( GLcontext *ctx,
 }
 
 
-
-
-
 /**
  * Destructor for this pipeline stage.
  */
@@ -256,7 +397,7 @@ static void dtr( struct tnl_pipeline_stage *stage )
 
       /* free the vertex program result arrays */
       for (i = 0; i < VERT_RESULT_MAX; i++)
-         _mesa_vector4f_free( &store->attribs[i] );
+         _mesa_vector4f_free( &store->results[i] );
 
       /* free misc arrays */
       _mesa_vector4f_free( &store->ndcCoords );
@@ -267,6 +408,7 @@ static void dtr( struct tnl_pipeline_stage *stage )
    }
 }
 
+
 /**
  * Public description of this pipeline stage.
  */
diff --git a/src/mesa/tnl/t_vb_texgen.c b/src/mesa/tnl/t_vb_texgen.c
index 5f7b2dccd8f..e98ab743c5e 100644
--- a/src/mesa/tnl/t_vb_texgen.c
+++ b/src/mesa/tnl/t_vb_texgen.c
@@ -488,10 +488,7 @@ static GLboolean run_texgen_stage( GLcontext *ctx,
    struct texgen_stage_data *store = TEXGEN_STAGE_DATA(stage);
    GLuint i;
 
-   if (ctx->ShaderObjects._VertexShaderPresent)
-      return GL_TRUE;
-
-   if (!ctx->Texture._TexGenEnabled || ctx->VertexProgram._Enabled) 
+   if (!ctx->Texture._TexGenEnabled || ctx->VertexProgram._Current) 
       return GL_TRUE;
 
    for (i = 0 ; i < ctx->Const.MaxTextureCoordUnits ; i++) {
@@ -516,10 +513,7 @@ static void validate_texgen_stage( GLcontext *ctx,
    struct texgen_stage_data *store = TEXGEN_STAGE_DATA(stage);
    GLuint i;
 
-   if (ctx->ShaderObjects._VertexShaderPresent)
-      return;
-
-   if (!ctx->Texture._TexGenEnabled || ctx->VertexProgram._Enabled) 
+   if (!ctx->Texture._TexGenEnabled || ctx->VertexProgram._Current) 
       return;
 
    for (i = 0 ; i < ctx->Const.MaxTextureCoordUnits ; i++) {
diff --git a/src/mesa/tnl/t_vb_texmat.c b/src/mesa/tnl/t_vb_texmat.c
index fa9222c0f14..674d36f8cbe 100644
--- a/src/mesa/tnl/t_vb_texmat.c
+++ b/src/mesa/tnl/t_vb_texmat.c
@@ -61,10 +61,7 @@ static GLboolean run_texmat_stage( GLcontext *ctx,
    struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
    GLuint i;
 
-   if (ctx->ShaderObjects._VertexShaderPresent)
-      return GL_TRUE;
-
-   if (!ctx->Texture._TexMatEnabled || ctx->VertexProgram._Enabled) 
+   if (!ctx->Texture._TexMatEnabled || ctx->VertexProgram._Current) 
       return GL_TRUE;
 
    /* ENABLE_TEXMAT implies that the texture matrix is not the
diff --git a/src/mesa/tnl/t_vb_vertex.c b/src/mesa/tnl/t_vb_vertex.c
index 100fca2a226..276305b5e67 100644
--- a/src/mesa/tnl/t_vb_vertex.c
+++ b/src/mesa/tnl/t_vb_vertex.c
@@ -126,10 +126,7 @@ static GLboolean run_vertex_stage( GLcontext *ctx,
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    struct vertex_buffer *VB = &tnl->vb;
 
-   if (ctx->ShaderObjects._VertexShaderPresent)
-      return GL_TRUE;
-
-   if (ctx->VertexProgram._Enabled) 
+   if (ctx->VertexProgram._Current) 
       return GL_TRUE;
 
    if (ctx->_NeedEyeCoords) {
diff --git a/src/mesa/tnl/t_vp_build.c b/src/mesa/tnl/t_vp_build.c
index b1c148a1367..63c99ee6ca0 100644
--- a/src/mesa/tnl/t_vp_build.c
+++ b/src/mesa/tnl/t_vp_build.c
@@ -33,11 +33,14 @@
 #include "glheader.h"
 #include "macros.h"
 #include "enums.h"
+#include "program.h"
+#include "prog_instruction.h"
+#include "prog_parameter.h"
+#include "prog_print.h"
+#include "prog_statevars.h"
 #include "t_context.h" /* NOTE: very light dependency on this */
 #include "t_vp_build.h"
 
-#include "shader/program.h"
-#include "shader/program_instruction.h"
 
 struct state_key {
    unsigned light_global_enabled:1;
@@ -382,11 +385,14 @@ static struct ureg register_const4f( struct tnl_program *p,
 {
    GLfloat values[4];
    GLint idx;
+   GLuint swizzle;
    values[0] = s0;
    values[1] = s1;
    values[2] = s2;
    values[3] = s3;
-   idx = _mesa_add_unnamed_constant( p->program->Base.Parameters, values, 4 );
+   idx = _mesa_add_unnamed_constant( p->program->Base.Parameters, values, 4,
+                                     &swizzle );
+   ASSERT(swizzle == SWIZZLE_NOOP);
    return make_ureg(PROGRAM_STATE_VAR, idx);
 }
 
@@ -408,40 +414,37 @@ static struct ureg get_identity_param( struct tnl_program *p )
    return p->identity;
 }
 
-static struct ureg register_param6( struct tnl_program *p, 
+static struct ureg register_param5(struct tnl_program *p, 
 				   GLint s0,
 				   GLint s1,
 				   GLint s2,
 				   GLint s3,
-				   GLint s4,
-				   GLint s5)
+                                   GLint s4)
 {
-   GLint tokens[6];
+   gl_state_index tokens[STATE_LENGTH];
    GLint idx;
    tokens[0] = s0;
    tokens[1] = s1;
    tokens[2] = s2;
    tokens[3] = s3;
    tokens[4] = s4;
-   tokens[5] = s5;
    idx = _mesa_add_state_reference( p->program->Base.Parameters, tokens );
    return make_ureg(PROGRAM_STATE_VAR, idx);
 }
 
 
-#define register_param1(p,s0)          register_param6(p,s0,0,0,0,0,0)
-#define register_param2(p,s0,s1)       register_param6(p,s0,s1,0,0,0,0)
-#define register_param3(p,s0,s1,s2)    register_param6(p,s0,s1,s2,0,0,0)
-#define register_param4(p,s0,s1,s2,s3) register_param6(p,s0,s1,s2,s3,0,0)
+#define register_param1(p,s0)          register_param5(p,s0,0,0,0,0)
+#define register_param2(p,s0,s1)       register_param5(p,s0,s1,0,0,0)
+#define register_param3(p,s0,s1,s2)    register_param5(p,s0,s1,s2,0,0)
+#define register_param4(p,s0,s1,s2,s3) register_param5(p,s0,s1,s2,s3,0)
 
 
-static void register_matrix_param6( struct tnl_program *p,
-				    GLint s0,
-				    GLint s1,
-				    GLint s2,
-				    GLint s3,
-				    GLint s4,
-				    GLint s5,
+static void register_matrix_param5( struct tnl_program *p,
+				    GLint s0, /* modelview, projection, etc */
+				    GLint s1, /* texture matrix number */
+				    GLint s2, /* first row */
+				    GLint s3, /* last row */
+				    GLint s4, /* inverse, transpose, etc */
 				    struct ureg *matrix )
 {
    GLint i;
@@ -449,8 +452,8 @@ static void register_matrix_param6( struct tnl_program *p,
    /* This is a bit sad as the support is there to pull the whole
     * matrix out in one go:
     */
-   for (i = 0; i <= s4 - s3; i++) 
-      matrix[i] = register_param6( p, s0, s1, s2, i, i, s5 );
+   for (i = 0; i <= s3 - s2; i++) 
+      matrix[i] = register_param5( p, s0, s1, i, i, s4 );
 }
 
 
@@ -630,13 +633,13 @@ static struct ureg get_eye_position( struct tnl_program *p )
       p->eye_position = reserve_temp(p);
 
       if (PREFER_DP4) {
-	 register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 3, 
-				 STATE_MATRIX, modelview );
+	 register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
+                                 0, modelview );
 
 	 emit_matrix_transform_vec4(p, p->eye_position, modelview, pos);
       }
       else {
-	 register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 3, 
+	 register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 3,
 				 STATE_MATRIX_TRANSPOSE, modelview );
 
 	 emit_transpose_matrix_transform_vec4(p, p->eye_position, modelview, pos);
@@ -665,7 +668,7 @@ static struct ureg get_eye_normal( struct tnl_program *p )
       struct ureg normal = register_input(p, VERT_ATTRIB_NORMAL );
       struct ureg mvinv[3];
 
-      register_matrix_param6( p, STATE_MATRIX, STATE_MODELVIEW, 0, 0, 2,
+      register_matrix_param5( p, STATE_MODELVIEW_MATRIX, 0, 0, 2,
 			      STATE_MATRIX_INVTRANS, mvinv );
 
       p->eye_normal = reserve_temp(p);
@@ -700,12 +703,12 @@ static void build_hpos( struct tnl_program *p )
    struct ureg mvp[4];
 
    if (PREFER_DP4) {
-      register_matrix_param6( p, STATE_MATRIX, STATE_MVP, 0, 0, 3, 
-			      STATE_MATRIX, mvp );
+      register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3, 
+			      0, mvp );
       emit_matrix_transform_vec4( p, hpos, mvp, pos );
    }
    else {
-      register_matrix_param6( p, STATE_MATRIX, STATE_MVP, 0, 0, 3, 
+      register_matrix_param5( p, STATE_MVP_MATRIX, 0, 0, 3, 
 			      STATE_MATRIX_TRANSPOSE, mvp );
       emit_transpose_matrix_transform_vec4( p, hpos, mvp, pos );
    }
@@ -850,7 +853,7 @@ static struct ureg calculate_light_attenuation( struct tnl_program *p,
 
 
 /* Need to add some addtional parameters to allow lighting in object
- * space - STATE_SPOT_DIRECTION and STATE_HALF implicitly assume eye
+ * space - STATE_SPOT_DIRECTION and STATE_HALF_VECTOR implicitly assume eye
  * space lighting.
  */
 static void build_lighting( struct tnl_program *p )
@@ -945,7 +948,7 @@ static void build_lighting( struct tnl_program *p )
                 emit_op2(p, OPCODE_SUB, half, 0, VPpli, eye_hat);
                 emit_normalize_vec3(p, half, half);
             } else {
-                half = register_param3(p, STATE_LIGHT, i, STATE_HALF);
+                half = register_param3(p, STATE_LIGHT, i, STATE_HALF_VECTOR);
             }
 	 } 
 	 else {
@@ -1211,7 +1214,7 @@ static void build_texture_transform( struct tnl_program *p )
 
    for (i = 0; i < MAX_TEXTURE_UNITS; i++) {
 
-      if (!(p->state->fragprog_inputs_read & (FRAG_BIT_TEX0<<i)))
+      if (!(p->state->fragprog_inputs_read & FRAG_BIT_TEX(i)))
 	 continue;
 							     
       if (p->state->unit[i].texgen_enabled || 
@@ -1301,13 +1304,13 @@ static void build_texture_transform( struct tnl_program *p )
 			      out_texgen : 
 			      register_input(p, VERT_ATTRIB_TEX0+i));
 	    if (PREFER_DP4) {
-	       register_matrix_param6( p, STATE_MATRIX, STATE_TEXTURE, i, 
-				       0, 3, STATE_MATRIX, texmat );
+	       register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3,
+				       0, texmat );
 	       emit_matrix_transform_vec4( p, out, texmat, in );
 	    }
 	    else {
-	       register_matrix_param6( p, STATE_MATRIX, STATE_TEXTURE, i, 
-				       0, 3, STATE_MATRIX_TRANSPOSE, texmat );
+	       register_matrix_param5( p, STATE_TEXTURE_MATRIX, i, 0, 3,
+				       STATE_MATRIX_TRANSPOSE, texmat );
 	       emit_transpose_matrix_transform_vec4( p, out, texmat, in );
 	    }
 	 }
@@ -1503,7 +1506,7 @@ void _tnl_UpdateFixedFunctionProgram( GLcontext *ctx )
    GLuint hash;
    const struct gl_vertex_program *prev = ctx->VertexProgram._Current;
 
-   if (ctx->VertexProgram._Enabled == GL_FALSE) { 
+   if (!ctx->VertexProgram._Current) {
       /* Grab all the relevent state and put it in a single structure:
        */
       key = make_state_key(ctx);
@@ -1511,45 +1514,42 @@ void _tnl_UpdateFixedFunctionProgram( GLcontext *ctx )
 
       /* Look for an already-prepared program for this state:
        */
-      ctx->_TnlProgram = (struct gl_vertex_program *)
+      ctx->VertexProgram._TnlProgram = (struct gl_vertex_program *)
 	 search_cache( tnl->vp_cache, hash, key, sizeof(*key) );
    
       /* OK, we'll have to build a new one:
        */
-      if (!ctx->_TnlProgram) {
+      if (!ctx->VertexProgram._TnlProgram) {
 	 if (0)
 	    _mesa_printf("Build new TNL program\n");
 	 
-	 ctx->_TnlProgram = (struct gl_vertex_program *)
+	 ctx->VertexProgram._TnlProgram = (struct gl_vertex_program *)
 	    ctx->Driver.NewProgram(ctx, GL_VERTEX_PROGRAM_ARB, 0); 
 
-	 create_new_program( key, ctx->_TnlProgram, 
+	 create_new_program( key, ctx->VertexProgram._TnlProgram, 
 			     ctx->Const.VertexProgram.MaxTemps );
 
 	 if (ctx->Driver.ProgramStringNotify)
 	    ctx->Driver.ProgramStringNotify( ctx, GL_VERTEX_PROGRAM_ARB, 
-					     &ctx->_TnlProgram->Base );
+                                       &ctx->VertexProgram._TnlProgram->Base );
 
-	 cache_item(tnl->vp_cache, hash, key, ctx->_TnlProgram );
+	 cache_item(tnl->vp_cache, hash, key, ctx->VertexProgram._TnlProgram );
       }
       else {
 	 FREE(key);
 	 if (0) 
 	    _mesa_printf("Found existing TNL program for key %x\n", hash);
       }
-      ctx->VertexProgram._Current = ctx->_TnlProgram;
-   }
-   else {
-      ctx->VertexProgram._Current = ctx->VertexProgram.Current;
+      ctx->VertexProgram._Current = ctx->VertexProgram._TnlProgram;
    }
 
    /* Tell the driver about the change.  Could define a new target for
     * this?
     */
-   if (ctx->VertexProgram._Current != prev &&
-       ctx->Driver.BindProgram) 
+   if (ctx->VertexProgram._Current != prev && ctx->Driver.BindProgram) {
       ctx->Driver.BindProgram(ctx, GL_VERTEX_PROGRAM_ARB,
                             (struct gl_program *) ctx->VertexProgram._Current);
+   }
 }
 
 void _tnl_ProgramCacheInit( GLcontext *ctx )
diff --git a/src/mesa/tnl/tnl.h b/src/mesa/tnl/tnl.h
index 17cb30adc76..20bed5546de 100644
--- a/src/mesa/tnl/tnl.h
+++ b/src/mesa/tnl/tnl.h
@@ -82,4 +82,7 @@ _tnl_draw_prims( GLcontext *ctx,
 		 GLuint min_index,
 		 GLuint max_index);
 
+extern void
+_mesa_load_tracked_matrices(GLcontext *ctx);
+
 #endif