libxtrans libX11 libX11 libXext mesa xserver git update 22 sep 2011

3 files changed, 2405 insertions, 2403 deletions

diff --git a/mesalib/src/mesa/tnl/t_vertex.c b/mesalib/src/mesa/tnl/t_vertex.c
index 0ca324402..6582949a0 100644
--- a/mesalib/src/mesa/tnl/t_vertex.c
+++ b/mesalib/src/mesa/tnl/t_vertex.c
@@ -1,564 +1,564 @@
-/*

- * Copyright 2003 Tungsten Graphics, inc.

- * All Rights Reserved.

- *

- * Permission is hereby granted, free of charge, to any person obtaining a

- * copy of this software and associated documentation files (the "Software"),

- * to deal in the Software without restriction, including without limitation

- * on the rights to use, copy, modify, merge, publish, distribute, sub

- * license, and/or sell copies of the Software, and to permit persons to whom

- * the Software is furnished to do so, subject to the following conditions:

- *

- * The above copyright notice and this permission notice (including the next

- * paragraph) shall be included in all copies or substantial portions of the

- * Software.

- *

- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR

- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,

- * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL

- * TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,

- * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR

- * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE

- * USE OR OTHER DEALINGS IN THE SOFTWARE.

- *

- * Authors:

- *    Keith Whitwell <keithw@tungstengraphics.com>

- */

-

-#include "main/glheader.h"

-#include "main/context.h"

-#include "main/colormac.h"

-

-#include "t_context.h"

-#include "t_vertex.h"

-

-#define DBG 0

-

-/* Build and manage clipspace/ndc/window vertices.

- */

-

-static GLboolean match_fastpath( struct tnl_clipspace *vtx,

-				 const struct tnl_clipspace_fastpath *fp)

-{

-   GLuint j;

-

-   if (vtx->attr_count != fp->attr_count) 

-      return GL_FALSE;

-

-   for (j = 0; j < vtx->attr_count; j++) 

-      if (vtx->attr[j].format != fp->attr[j].format ||

-	  vtx->attr[j].inputsize != fp->attr[j].size ||

-	  vtx->attr[j].vertoffset != fp->attr[j].offset) 

-	 return GL_FALSE;

-      

-   if (fp->match_strides) {

-      if (vtx->vertex_size != fp->vertex_size)

-	 return GL_FALSE;

-

-      for (j = 0; j < vtx->attr_count; j++) 

-	 if (vtx->attr[j].inputstride != fp->attr[j].stride) 

-	    return GL_FALSE;

-   }

-   

-   return GL_TRUE;

-}

-

-static GLboolean search_fastpath_emit( struct tnl_clipspace *vtx )

-{

-   struct tnl_clipspace_fastpath *fp = vtx->fastpath;

-

-   for ( ; fp ; fp = fp->next) {

-      if (match_fastpath(vtx, fp)) {

-         vtx->emit = fp->func;

-	 return GL_TRUE;

-      }

-   }

-

-   return GL_FALSE;

-}

-

-void _tnl_register_fastpath( struct tnl_clipspace *vtx,

-			     GLboolean match_strides )

-{

-   struct tnl_clipspace_fastpath *fastpath = CALLOC_STRUCT(tnl_clipspace_fastpath);

-   GLuint i;

-

-   fastpath->vertex_size = vtx->vertex_size;

-   fastpath->attr_count = vtx->attr_count;

-   fastpath->match_strides = match_strides;

-   fastpath->func = vtx->emit;

-   fastpath->attr = (struct tnl_attr_type *)

-      malloc(vtx->attr_count * sizeof(fastpath->attr[0]));

-

-   for (i = 0; i < vtx->attr_count; i++) {

-      fastpath->attr[i].format = vtx->attr[i].format;

-      fastpath->attr[i].stride = vtx->attr[i].inputstride;

-      fastpath->attr[i].size = vtx->attr[i].inputsize;

-      fastpath->attr[i].offset = vtx->attr[i].vertoffset;

-   }

-

-   fastpath->next = vtx->fastpath;

-   vtx->fastpath = fastpath;

-}

-

-

-

-/***********************************************************************

- * Build codegen functions or return generic ones:

- */

-static void choose_emit_func( struct gl_context *ctx, GLuint count, GLubyte *dest)

-{

-   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-   struct tnl_clipspace_attr *a = vtx->attr;

-   const GLuint attr_count = vtx->attr_count;

-   GLuint j;

-

-   for (j = 0; j < attr_count; j++) {

-      GLvector4f *vptr = VB->AttribPtr[a[j].attrib];

-      a[j].inputstride = vptr->stride;

-      a[j].inputsize = vptr->size;

-      a[j].emit = a[j].insert[vptr->size - 1]; /* not always used */

-   }

-

-   vtx->emit = NULL;

-   

-   /* Does this match an existing (hardwired, codegen or known-bad)

-    * fastpath?

-    */

-   if (search_fastpath_emit(vtx)) {

-      /* Use this result.  If it is null, then it is already known

-       * that the current state will fail for codegen and there is no

-       * point trying again.

-       */

-   }

-   else if (vtx->codegen_emit) {

-      vtx->codegen_emit(ctx);

-   }

-

-   if (!vtx->emit) {

-      _tnl_generate_hardwired_emit(ctx);

-   }

-

-   /* Otherwise use the generic version:

-    */

-   if (!vtx->emit)

-      vtx->emit = _tnl_generic_emit;

-

-   vtx->emit( ctx, count, dest );

-}

-

-

-

-static void choose_interp_func( struct gl_context *ctx,

-				GLfloat t,

-				GLuint edst, GLuint eout, GLuint ein,

-				GLboolean force_boundary )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-

-   if (vtx->need_extras && 

-       (ctx->_TriangleCaps & (DD_TRI_LIGHT_TWOSIDE|DD_TRI_UNFILLED))) {

-      vtx->interp = _tnl_generic_interp_extras;

-   } else {

-      vtx->interp = _tnl_generic_interp;

-   }

-

-   vtx->interp( ctx, t, edst, eout, ein, force_boundary );

-}

-

-

-static void choose_copy_pv_func(  struct gl_context *ctx, GLuint edst, GLuint esrc )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-

-   if (vtx->need_extras && 

-       (ctx->_TriangleCaps & (DD_TRI_LIGHT_TWOSIDE|DD_TRI_UNFILLED))) {

-      vtx->copy_pv = _tnl_generic_copy_pv_extras;

-   } else {

-      vtx->copy_pv = _tnl_generic_copy_pv;

-   }

-

-   vtx->copy_pv( ctx, edst, esrc );

-}

-

-

-/***********************************************************************

- * Public entrypoints, mostly dispatch to the above:

- */

-

-

-/* Interpolate between two vertices to produce a third:

- */

-void _tnl_interp( struct gl_context *ctx,

-		  GLfloat t,

-		  GLuint edst, GLuint eout, GLuint ein,

-		  GLboolean force_boundary )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-   vtx->interp( ctx, t, edst, eout, ein, force_boundary );

-}

-

-/* Copy colors from one vertex to another:

- */

-void _tnl_copy_pv(  struct gl_context *ctx, GLuint edst, GLuint esrc )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-   vtx->copy_pv( ctx, edst, esrc );

-}

-

-

-/* Extract a named attribute from a hardware vertex.  Will have to

- * reverse any viewport transformation, swizzling or other conversions

- * which may have been applied:

- */

-void _tnl_get_attr( struct gl_context *ctx, const void *vin,

-			      GLenum attr, GLfloat *dest )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-   const struct tnl_clipspace_attr *a = vtx->attr;

-   const GLuint attr_count = vtx->attr_count;

-   GLuint j;

-

-   for (j = 0; j < attr_count; j++) {

-      if (a[j].attrib == attr) {

-	 a[j].extract( &a[j], dest, (GLubyte *)vin + a[j].vertoffset );

-	 return;

-      }

-   }

-

-   /* Else return the value from ctx->Current.

-    */

-   if (attr == _TNL_ATTRIB_POINTSIZE) {

-      /* If the hardware vertex doesn't have point size then use size from

-       * struct gl_context.  XXX this will be wrong if drawing attenuated points!

-       */

-      dest[0] = ctx->Point.Size;

-   }

-   else {

-      memcpy( dest, ctx->Current.Attrib[attr], 4*sizeof(GLfloat));

-   }

-}

-

-

-/* Complementary operation to the above.

- */

-void _tnl_set_attr( struct gl_context *ctx, void *vout,

-		    GLenum attr, const GLfloat *src )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-   const struct tnl_clipspace_attr *a = vtx->attr;

-   const GLuint attr_count = vtx->attr_count;

-   GLuint j;

-

-   for (j = 0; j < attr_count; j++) {

-      if (a[j].attrib == attr) {

-	 a[j].insert[4-1]( &a[j], (GLubyte *)vout + a[j].vertoffset, src );

-	 return;

-      }

-   }

-}

-

-

-void *_tnl_get_vertex( struct gl_context *ctx, GLuint nr )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-

-   return vtx->vertex_buf + nr * vtx->vertex_size;

-}

-

-void _tnl_invalidate_vertex_state( struct gl_context *ctx, GLuint new_state )

-{

-   if (new_state & (_DD_NEW_TRI_LIGHT_TWOSIDE|_DD_NEW_TRI_UNFILLED) ) {

-      struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-      vtx->new_inputs = ~0;

-      vtx->interp = choose_interp_func;

-      vtx->copy_pv = choose_copy_pv_func;

-   }

-}

-

-static void invalidate_funcs( struct tnl_clipspace *vtx )

-{

-   vtx->emit = choose_emit_func;

-   vtx->interp = choose_interp_func;

-   vtx->copy_pv = choose_copy_pv_func;

-   vtx->new_inputs = ~0;

-}

-

-GLuint _tnl_install_attrs( struct gl_context *ctx, const struct tnl_attr_map *map,

-			   GLuint nr, const GLfloat *vp, 

-			   GLuint unpacked_size )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-   GLuint offset = 0;

-   GLuint i, j;

-

-   assert(nr < _TNL_ATTRIB_MAX);

-   assert(nr == 0 || map[0].attrib == VERT_ATTRIB_POS);

-

-   vtx->new_inputs = ~0;

-   vtx->need_viewport = GL_FALSE;

-

-   if (vp) {

-      vtx->need_viewport = GL_TRUE;

-   }

-

-   for (j = 0, i = 0; i < nr; i++) {

-      const GLuint format = map[i].format;

-      if (format == EMIT_PAD) {

-	 if (DBG)

-	    printf("%d: pad %d, offset %d\n", i,  

-		   map[i].offset, offset);  

-

-	 offset += map[i].offset;

-

-      }

-      else {

-	 GLuint tmpoffset;

-

-	 if (unpacked_size) 

-	    tmpoffset = map[i].offset;

-	 else

-	    tmpoffset = offset;

-

-	 if (vtx->attr_count != j ||

-	     vtx->attr[j].attrib != map[i].attrib ||

-	     vtx->attr[j].format != format ||

-	     vtx->attr[j].vertoffset != tmpoffset) {

-	    invalidate_funcs(vtx);

-

-	    vtx->attr[j].attrib = map[i].attrib;

-	    vtx->attr[j].format = format;

-	    vtx->attr[j].vp = vp;

-	    vtx->attr[j].insert = _tnl_format_info[format].insert;

-	    vtx->attr[j].extract = _tnl_format_info[format].extract;

-	    vtx->attr[j].vertattrsize = _tnl_format_info[format].attrsize;

-	    vtx->attr[j].vertoffset = tmpoffset;

-	 }

-

-	 

-	 if (DBG)

-	    printf("%d: %s, vp %p, offset %d\n", i,  

-		   _tnl_format_info[format].name, (void *)vp,

-		   vtx->attr[j].vertoffset);   

-

-	 offset += _tnl_format_info[format].attrsize;

-	 j++;

-      }

-   }

-

-   vtx->attr_count = j;

-

-   if (unpacked_size)

-      vtx->vertex_size = unpacked_size;

-   else

-      vtx->vertex_size = offset;

-

-   assert(vtx->vertex_size <= vtx->max_vertex_size);

-   return vtx->vertex_size;

-}

-

-

-

-void _tnl_invalidate_vertices( struct gl_context *ctx, GLuint newinputs )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-   vtx->new_inputs |= newinputs;

-}

-

-

-/* This event has broader use beyond this file - will move elsewhere

- * and probably invoke a driver callback.

- */

-void _tnl_notify_pipeline_output_change( struct gl_context *ctx )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-   invalidate_funcs(vtx);

-}

-

-

-static void adjust_input_ptrs( struct gl_context *ctx, GLint diff)

-{

-   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-   struct tnl_clipspace_attr *a = vtx->attr;

-   const GLuint count = vtx->attr_count;

-   GLuint j;

-

-   diff -= 1;

-   for (j=0; j<count; ++j) {

-           register GLvector4f *vptr = VB->AttribPtr[a->attrib];

-	   (a++)->inputptr += diff*vptr->stride;

-   }

-}

-

-static void update_input_ptrs( struct gl_context *ctx, GLuint start )

-{

-   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-   struct tnl_clipspace_attr *a = vtx->attr;

-   const GLuint count = vtx->attr_count;

-   GLuint j;

-   

-   for (j = 0; j < count; j++) {

-      GLvector4f *vptr = VB->AttribPtr[a[j].attrib];

-

-      if (vtx->emit != choose_emit_func) {

-	 assert(a[j].inputstride == vptr->stride);

-	 assert(a[j].inputsize == vptr->size);

-      }

-

-      a[j].inputptr = ((GLubyte *)vptr->data) + start * vptr->stride;

-   }

-   

-   if (a->vp) {

-      vtx->vp_scale[0] = a->vp[MAT_SX];

-      vtx->vp_scale[1] = a->vp[MAT_SY];

-      vtx->vp_scale[2] = a->vp[MAT_SZ];

-      vtx->vp_scale[3] = 1.0;

-      vtx->vp_xlate[0] = a->vp[MAT_TX];

-      vtx->vp_xlate[1] = a->vp[MAT_TY];

-      vtx->vp_xlate[2] = a->vp[MAT_TZ];

-      vtx->vp_xlate[3] = 0.0;

-   }

-}

-

-

-void _tnl_build_vertices( struct gl_context *ctx,

-			  GLuint start,

-			  GLuint end,

-			  GLuint newinputs )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);  

-   update_input_ptrs( ctx, start );      

-   vtx->emit( ctx, end - start, 

-	      (GLubyte *)(vtx->vertex_buf + 

-			  start * vtx->vertex_size));

-}

-

-/* Emit VB vertices start..end to dest.  Note that VB vertex at

- * postion start will be emitted to dest at position zero.

- */

-void *_tnl_emit_vertices_to_buffer( struct gl_context *ctx,

-				    GLuint start,

-				    GLuint end,

-				    void *dest )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-

-   update_input_ptrs(ctx, start);

-   /* Note: dest should not be adjusted for non-zero 'start' values:

-    */

-   vtx->emit( ctx, end - start, (GLubyte*) dest );	

-   return (void *)((GLubyte *)dest + vtx->vertex_size * (end - start));

-}

-

-/* Emit indexed VB vertices start..end to dest.  Note that VB vertex at

- * postion start will be emitted to dest at position zero.

- */

-

-void *_tnl_emit_indexed_vertices_to_buffer( struct gl_context *ctx,

-					    const GLuint *elts,

-					    GLuint start,

-					    GLuint end,

-					    void *dest )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-   GLuint oldIndex;

-   GLubyte *cdest = dest;

-

-   update_input_ptrs(ctx, oldIndex = elts[start++]);

-   vtx->emit( ctx, 1, cdest );

-   cdest += vtx->vertex_size;

-

-   for (; start < end; ++start) {

-      adjust_input_ptrs(ctx, elts[start] - oldIndex);

-      oldIndex = elts[start];

-      vtx->emit( ctx, 1, cdest);

-      cdest += vtx->vertex_size;

-   }

-

-   return (void *) cdest;

-}

-

-

-void _tnl_init_vertices( struct gl_context *ctx, 

-			GLuint vb_size,

-			GLuint max_vertex_size )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);  

-

-   _tnl_install_attrs( ctx, NULL, 0, NULL, 0 );

-

-   vtx->need_extras = GL_TRUE;

-   if (max_vertex_size > vtx->max_vertex_size) {

-      _tnl_free_vertices( ctx );

-      vtx->max_vertex_size = max_vertex_size;

-      vtx->vertex_buf = (GLubyte *)_mesa_align_calloc(vb_size * max_vertex_size, 32 );

-      invalidate_funcs(vtx);

-   }

-

-   switch(CHAN_TYPE) {

-   case GL_UNSIGNED_BYTE:

-      vtx->chan_scale[0] = 255.0;

-      vtx->chan_scale[1] = 255.0;

-      vtx->chan_scale[2] = 255.0;

-      vtx->chan_scale[3] = 255.0;

-      break;

-   case GL_UNSIGNED_SHORT:

-      vtx->chan_scale[0] = 65535.0;

-      vtx->chan_scale[1] = 65535.0;

-      vtx->chan_scale[2] = 65535.0;

-      vtx->chan_scale[3] = 65535.0;

-      break;

-   default:

-      vtx->chan_scale[0] = 1.0;

-      vtx->chan_scale[1] = 1.0;

-      vtx->chan_scale[2] = 1.0;

-      vtx->chan_scale[3] = 1.0;

-      break;

-   }

-

-   vtx->identity[0] = 0.0;

-   vtx->identity[1] = 0.0;

-   vtx->identity[2] = 0.0;

-   vtx->identity[3] = 1.0;

-

-   vtx->codegen_emit = NULL;

-

-#ifdef USE_SSE_ASM

-   if (!_mesa_getenv("MESA_NO_CODEGEN"))

-      vtx->codegen_emit = _tnl_generate_sse_emit;

-#endif

-}

-

-

-void _tnl_free_vertices( struct gl_context *ctx )

-{

-   TNLcontext *tnl = TNL_CONTEXT(ctx);

-   if (tnl) {

-      struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-      struct tnl_clipspace_fastpath *fp, *tmp;

-

-      if (vtx->vertex_buf) {

-         _mesa_align_free(vtx->vertex_buf);

-         vtx->vertex_buf = NULL;

-      }

-

-      for (fp = vtx->fastpath ; fp ; fp = tmp) {

-         tmp = fp->next;

-         FREE(fp->attr);

-

-         /* KW: At the moment, fp->func is constrained to be allocated by

-          * _mesa_exec_alloc(), as the hardwired fastpaths in

-          * t_vertex_generic.c are handled specially.  It would be nice

-          * to unify them, but this probably won't change until this

-          * module gets another overhaul.

-          */

-         _mesa_exec_free((void *) fp->func);

-         FREE(fp);

-      }

-

-      vtx->fastpath = NULL;

-   }

-}

+/*
+ * Copyright 2003 Tungsten Graphics, inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <keithw@tungstengraphics.com>
+ */
+
+#include "main/glheader.h"
+#include "main/context.h"
+#include "main/colormac.h"
+#include "swrast/s_chan.h"
+#include "t_context.h"
+#include "t_vertex.h"
+
+#define DBG 0
+
+/* Build and manage clipspace/ndc/window vertices.
+ */
+
+static GLboolean match_fastpath( struct tnl_clipspace *vtx,
+				 const struct tnl_clipspace_fastpath *fp)
+{
+   GLuint j;
+
+   if (vtx->attr_count != fp->attr_count) 
+      return GL_FALSE;
+
+   for (j = 0; j < vtx->attr_count; j++) 
+      if (vtx->attr[j].format != fp->attr[j].format ||
+	  vtx->attr[j].inputsize != fp->attr[j].size ||
+	  vtx->attr[j].vertoffset != fp->attr[j].offset) 
+	 return GL_FALSE;
+      
+   if (fp->match_strides) {
+      if (vtx->vertex_size != fp->vertex_size)
+	 return GL_FALSE;
+
+      for (j = 0; j < vtx->attr_count; j++) 
+	 if (vtx->attr[j].inputstride != fp->attr[j].stride) 
+	    return GL_FALSE;
+   }
+   
+   return GL_TRUE;
+}
+
+static GLboolean search_fastpath_emit( struct tnl_clipspace *vtx )
+{
+   struct tnl_clipspace_fastpath *fp = vtx->fastpath;
+
+   for ( ; fp ; fp = fp->next) {
+      if (match_fastpath(vtx, fp)) {
+         vtx->emit = fp->func;
+	 return GL_TRUE;
+      }
+   }
+
+   return GL_FALSE;
+}
+
+void _tnl_register_fastpath( struct tnl_clipspace *vtx,
+			     GLboolean match_strides )
+{
+   struct tnl_clipspace_fastpath *fastpath = CALLOC_STRUCT(tnl_clipspace_fastpath);
+   GLuint i;
+
+   fastpath->vertex_size = vtx->vertex_size;
+   fastpath->attr_count = vtx->attr_count;
+   fastpath->match_strides = match_strides;
+   fastpath->func = vtx->emit;
+   fastpath->attr = (struct tnl_attr_type *)
+      malloc(vtx->attr_count * sizeof(fastpath->attr[0]));
+
+   for (i = 0; i < vtx->attr_count; i++) {
+      fastpath->attr[i].format = vtx->attr[i].format;
+      fastpath->attr[i].stride = vtx->attr[i].inputstride;
+      fastpath->attr[i].size = vtx->attr[i].inputsize;
+      fastpath->attr[i].offset = vtx->attr[i].vertoffset;
+   }
+
+   fastpath->next = vtx->fastpath;
+   vtx->fastpath = fastpath;
+}
+
+
+
+/***********************************************************************
+ * Build codegen functions or return generic ones:
+ */
+static void choose_emit_func( struct gl_context *ctx, GLuint count, GLubyte *dest)
+{
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+   struct tnl_clipspace_attr *a = vtx->attr;
+   const GLuint attr_count = vtx->attr_count;
+   GLuint j;
+
+   for (j = 0; j < attr_count; j++) {
+      GLvector4f *vptr = VB->AttribPtr[a[j].attrib];
+      a[j].inputstride = vptr->stride;
+      a[j].inputsize = vptr->size;
+      a[j].emit = a[j].insert[vptr->size - 1]; /* not always used */
+   }
+
+   vtx->emit = NULL;
+   
+   /* Does this match an existing (hardwired, codegen or known-bad)
+    * fastpath?
+    */
+   if (search_fastpath_emit(vtx)) {
+      /* Use this result.  If it is null, then it is already known
+       * that the current state will fail for codegen and there is no
+       * point trying again.
+       */
+   }
+   else if (vtx->codegen_emit) {
+      vtx->codegen_emit(ctx);
+   }
+
+   if (!vtx->emit) {
+      _tnl_generate_hardwired_emit(ctx);
+   }
+
+   /* Otherwise use the generic version:
+    */
+   if (!vtx->emit)
+      vtx->emit = _tnl_generic_emit;
+
+   vtx->emit( ctx, count, dest );
+}
+
+
+
+static void choose_interp_func( struct gl_context *ctx,
+				GLfloat t,
+				GLuint edst, GLuint eout, GLuint ein,
+				GLboolean force_boundary )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+
+   if (vtx->need_extras && 
+       (ctx->_TriangleCaps & (DD_TRI_LIGHT_TWOSIDE|DD_TRI_UNFILLED))) {
+      vtx->interp = _tnl_generic_interp_extras;
+   } else {
+      vtx->interp = _tnl_generic_interp;
+   }
+
+   vtx->interp( ctx, t, edst, eout, ein, force_boundary );
+}
+
+
+static void choose_copy_pv_func(  struct gl_context *ctx, GLuint edst, GLuint esrc )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+
+   if (vtx->need_extras && 
+       (ctx->_TriangleCaps & (DD_TRI_LIGHT_TWOSIDE|DD_TRI_UNFILLED))) {
+      vtx->copy_pv = _tnl_generic_copy_pv_extras;
+   } else {
+      vtx->copy_pv = _tnl_generic_copy_pv;
+   }
+
+   vtx->copy_pv( ctx, edst, esrc );
+}
+
+
+/***********************************************************************
+ * Public entrypoints, mostly dispatch to the above:
+ */
+
+
+/* Interpolate between two vertices to produce a third:
+ */
+void _tnl_interp( struct gl_context *ctx,
+		  GLfloat t,
+		  GLuint edst, GLuint eout, GLuint ein,
+		  GLboolean force_boundary )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+   vtx->interp( ctx, t, edst, eout, ein, force_boundary );
+}
+
+/* Copy colors from one vertex to another:
+ */
+void _tnl_copy_pv(  struct gl_context *ctx, GLuint edst, GLuint esrc )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+   vtx->copy_pv( ctx, edst, esrc );
+}
+
+
+/* Extract a named attribute from a hardware vertex.  Will have to
+ * reverse any viewport transformation, swizzling or other conversions
+ * which may have been applied:
+ */
+void _tnl_get_attr( struct gl_context *ctx, const void *vin,
+			      GLenum attr, GLfloat *dest )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+   const struct tnl_clipspace_attr *a = vtx->attr;
+   const GLuint attr_count = vtx->attr_count;
+   GLuint j;
+
+   for (j = 0; j < attr_count; j++) {
+      if (a[j].attrib == attr) {
+	 a[j].extract( &a[j], dest, (GLubyte *)vin + a[j].vertoffset );
+	 return;
+      }
+   }
+
+   /* Else return the value from ctx->Current.
+    */
+   if (attr == _TNL_ATTRIB_POINTSIZE) {
+      /* If the hardware vertex doesn't have point size then use size from
+       * struct gl_context.  XXX this will be wrong if drawing attenuated points!
+       */
+      dest[0] = ctx->Point.Size;
+   }
+   else {
+      memcpy( dest, ctx->Current.Attrib[attr], 4*sizeof(GLfloat));
+   }
+}
+
+
+/* Complementary operation to the above.
+ */
+void _tnl_set_attr( struct gl_context *ctx, void *vout,
+		    GLenum attr, const GLfloat *src )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+   const struct tnl_clipspace_attr *a = vtx->attr;
+   const GLuint attr_count = vtx->attr_count;
+   GLuint j;
+
+   for (j = 0; j < attr_count; j++) {
+      if (a[j].attrib == attr) {
+	 a[j].insert[4-1]( &a[j], (GLubyte *)vout + a[j].vertoffset, src );
+	 return;
+      }
+   }
+}
+
+
+void *_tnl_get_vertex( struct gl_context *ctx, GLuint nr )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+
+   return vtx->vertex_buf + nr * vtx->vertex_size;
+}
+
+void _tnl_invalidate_vertex_state( struct gl_context *ctx, GLuint new_state )
+{
+   if (new_state & (_DD_NEW_TRI_LIGHT_TWOSIDE|_DD_NEW_TRI_UNFILLED) ) {
+      struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+      vtx->new_inputs = ~0;
+      vtx->interp = choose_interp_func;
+      vtx->copy_pv = choose_copy_pv_func;
+   }
+}
+
+static void invalidate_funcs( struct tnl_clipspace *vtx )
+{
+   vtx->emit = choose_emit_func;
+   vtx->interp = choose_interp_func;
+   vtx->copy_pv = choose_copy_pv_func;
+   vtx->new_inputs = ~0;
+}
+
+GLuint _tnl_install_attrs( struct gl_context *ctx, const struct tnl_attr_map *map,
+			   GLuint nr, const GLfloat *vp, 
+			   GLuint unpacked_size )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+   GLuint offset = 0;
+   GLuint i, j;
+
+   assert(nr < _TNL_ATTRIB_MAX);
+   assert(nr == 0 || map[0].attrib == VERT_ATTRIB_POS);
+
+   vtx->new_inputs = ~0;
+   vtx->need_viewport = GL_FALSE;
+
+   if (vp) {
+      vtx->need_viewport = GL_TRUE;
+   }
+
+   for (j = 0, i = 0; i < nr; i++) {
+      const GLuint format = map[i].format;
+      if (format == EMIT_PAD) {
+	 if (DBG)
+	    printf("%d: pad %d, offset %d\n", i,  
+		   map[i].offset, offset);  
+
+	 offset += map[i].offset;
+
+      }
+      else {
+	 GLuint tmpoffset;
+
+	 if (unpacked_size) 
+	    tmpoffset = map[i].offset;
+	 else
+	    tmpoffset = offset;
+
+	 if (vtx->attr_count != j ||
+	     vtx->attr[j].attrib != map[i].attrib ||
+	     vtx->attr[j].format != format ||
+	     vtx->attr[j].vertoffset != tmpoffset) {
+	    invalidate_funcs(vtx);
+
+	    vtx->attr[j].attrib = map[i].attrib;
+	    vtx->attr[j].format = format;
+	    vtx->attr[j].vp = vp;
+	    vtx->attr[j].insert = _tnl_format_info[format].insert;
+	    vtx->attr[j].extract = _tnl_format_info[format].extract;
+	    vtx->attr[j].vertattrsize = _tnl_format_info[format].attrsize;
+	    vtx->attr[j].vertoffset = tmpoffset;
+	 }
+
+	 
+	 if (DBG)
+	    printf("%d: %s, vp %p, offset %d\n", i,  
+		   _tnl_format_info[format].name, (void *)vp,
+		   vtx->attr[j].vertoffset);   
+
+	 offset += _tnl_format_info[format].attrsize;
+	 j++;
+      }
+   }
+
+   vtx->attr_count = j;
+
+   if (unpacked_size)
+      vtx->vertex_size = unpacked_size;
+   else
+      vtx->vertex_size = offset;
+
+   assert(vtx->vertex_size <= vtx->max_vertex_size);
+   return vtx->vertex_size;
+}
+
+
+
+void _tnl_invalidate_vertices( struct gl_context *ctx, GLuint newinputs )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+   vtx->new_inputs |= newinputs;
+}
+
+
+/* This event has broader use beyond this file - will move elsewhere
+ * and probably invoke a driver callback.
+ */
+void _tnl_notify_pipeline_output_change( struct gl_context *ctx )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+   invalidate_funcs(vtx);
+}
+
+
+static void adjust_input_ptrs( struct gl_context *ctx, GLint diff)
+{
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+   struct tnl_clipspace_attr *a = vtx->attr;
+   const GLuint count = vtx->attr_count;
+   GLuint j;
+
+   diff -= 1;
+   for (j=0; j<count; ++j) {
+           register GLvector4f *vptr = VB->AttribPtr[a->attrib];
+	   (a++)->inputptr += diff*vptr->stride;
+   }
+}
+
+static void update_input_ptrs( struct gl_context *ctx, GLuint start )
+{
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+   struct tnl_clipspace_attr *a = vtx->attr;
+   const GLuint count = vtx->attr_count;
+   GLuint j;
+   
+   for (j = 0; j < count; j++) {
+      GLvector4f *vptr = VB->AttribPtr[a[j].attrib];
+
+      if (vtx->emit != choose_emit_func) {
+	 assert(a[j].inputstride == vptr->stride);
+	 assert(a[j].inputsize == vptr->size);
+      }
+
+      a[j].inputptr = ((GLubyte *)vptr->data) + start * vptr->stride;
+   }
+   
+   if (a->vp) {
+      vtx->vp_scale[0] = a->vp[MAT_SX];
+      vtx->vp_scale[1] = a->vp[MAT_SY];
+      vtx->vp_scale[2] = a->vp[MAT_SZ];
+      vtx->vp_scale[3] = 1.0;
+      vtx->vp_xlate[0] = a->vp[MAT_TX];
+      vtx->vp_xlate[1] = a->vp[MAT_TY];
+      vtx->vp_xlate[2] = a->vp[MAT_TZ];
+      vtx->vp_xlate[3] = 0.0;
+   }
+}
+
+
+void _tnl_build_vertices( struct gl_context *ctx,
+			  GLuint start,
+			  GLuint end,
+			  GLuint newinputs )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);  
+   update_input_ptrs( ctx, start );      
+   vtx->emit( ctx, end - start, 
+	      (GLubyte *)(vtx->vertex_buf + 
+			  start * vtx->vertex_size));
+}
+
+/* Emit VB vertices start..end to dest.  Note that VB vertex at
+ * postion start will be emitted to dest at position zero.
+ */
+void *_tnl_emit_vertices_to_buffer( struct gl_context *ctx,
+				    GLuint start,
+				    GLuint end,
+				    void *dest )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+
+   update_input_ptrs(ctx, start);
+   /* Note: dest should not be adjusted for non-zero 'start' values:
+    */
+   vtx->emit( ctx, end - start, (GLubyte*) dest );	
+   return (void *)((GLubyte *)dest + vtx->vertex_size * (end - start));
+}
+
+/* Emit indexed VB vertices start..end to dest.  Note that VB vertex at
+ * postion start will be emitted to dest at position zero.
+ */
+
+void *_tnl_emit_indexed_vertices_to_buffer( struct gl_context *ctx,
+					    const GLuint *elts,
+					    GLuint start,
+					    GLuint end,
+					    void *dest )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+   GLuint oldIndex;
+   GLubyte *cdest = dest;
+
+   update_input_ptrs(ctx, oldIndex = elts[start++]);
+   vtx->emit( ctx, 1, cdest );
+   cdest += vtx->vertex_size;
+
+   for (; start < end; ++start) {
+      adjust_input_ptrs(ctx, elts[start] - oldIndex);
+      oldIndex = elts[start];
+      vtx->emit( ctx, 1, cdest);
+      cdest += vtx->vertex_size;
+   }
+
+   return (void *) cdest;
+}
+
+
+void _tnl_init_vertices( struct gl_context *ctx, 
+			GLuint vb_size,
+			GLuint max_vertex_size )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);  
+
+   _tnl_install_attrs( ctx, NULL, 0, NULL, 0 );
+
+   vtx->need_extras = GL_TRUE;
+   if (max_vertex_size > vtx->max_vertex_size) {
+      _tnl_free_vertices( ctx );
+      vtx->max_vertex_size = max_vertex_size;
+      vtx->vertex_buf = (GLubyte *)_mesa_align_calloc(vb_size * max_vertex_size, 32 );
+      invalidate_funcs(vtx);
+   }
+
+   switch(CHAN_TYPE) {
+   case GL_UNSIGNED_BYTE:
+      vtx->chan_scale[0] = 255.0;
+      vtx->chan_scale[1] = 255.0;
+      vtx->chan_scale[2] = 255.0;
+      vtx->chan_scale[3] = 255.0;
+      break;
+   case GL_UNSIGNED_SHORT:
+      vtx->chan_scale[0] = 65535.0;
+      vtx->chan_scale[1] = 65535.0;
+      vtx->chan_scale[2] = 65535.0;
+      vtx->chan_scale[3] = 65535.0;
+      break;
+   default:
+      vtx->chan_scale[0] = 1.0;
+      vtx->chan_scale[1] = 1.0;
+      vtx->chan_scale[2] = 1.0;
+      vtx->chan_scale[3] = 1.0;
+      break;
+   }
+
+   vtx->identity[0] = 0.0;
+   vtx->identity[1] = 0.0;
+   vtx->identity[2] = 0.0;
+   vtx->identity[3] = 1.0;
+
+   vtx->codegen_emit = NULL;
+
+#ifdef USE_SSE_ASM
+   if (!_mesa_getenv("MESA_NO_CODEGEN"))
+      vtx->codegen_emit = _tnl_generate_sse_emit;
+#endif
+}
+
+
+void _tnl_free_vertices( struct gl_context *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   if (tnl) {
+      struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+      struct tnl_clipspace_fastpath *fp, *tmp;
+
+      if (vtx->vertex_buf) {
+         _mesa_align_free(vtx->vertex_buf);
+         vtx->vertex_buf = NULL;
+      }
+
+      for (fp = vtx->fastpath ; fp ; fp = tmp) {
+         tmp = fp->next;
+         FREE(fp->attr);
+
+         /* KW: At the moment, fp->func is constrained to be allocated by
+          * _mesa_exec_alloc(), as the hardwired fastpaths in
+          * t_vertex_generic.c are handled specially.  It would be nice
+          * to unify them, but this probably won't change until this
+          * module gets another overhaul.
+          */
+         _mesa_exec_free((void *) fp->func);
+         FREE(fp);
+      }
+
+      vtx->fastpath = NULL;
+   }
+}
diff --git a/mesalib/src/mesa/tnl/t_vertex_generic.c b/mesalib/src/mesa/tnl/t_vertex_generic.c
index 144b3669d..9dcecdd57 100644
--- a/mesalib/src/mesa/tnl/t_vertex_generic.c
+++ b/mesalib/src/mesa/tnl/t_vertex_generic.c
@@ -1,1155 +1,1156 @@
-

-/*

- * Copyright 2003 Tungsten Graphics, inc.

- * All Rights Reserved.

- *

- * Permission is hereby granted, free of charge, to any person obtaining a

- * copy of this software and associated documentation files (the "Software"),

- * to deal in the Software without restriction, including without limitation

- * on the rights to use, copy, modify, merge, publish, distribute, sub

- * license, and/or sell copies of the Software, and to permit persons to whom

- * the Software is furnished to do so, subject to the following conditions:

- *

- * The above copyright notice and this permission notice (including the next

- * paragraph) shall be included in all copies or substantial portions of the

- * Software.

- *

- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR

- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,

- * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL

- * TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,

- * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR

- * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE

- * USE OR OTHER DEALINGS IN THE SOFTWARE.

- *

- * Authors:

- *    Keith Whitwell <keithw@tungstengraphics.com>

- */

-

-#include "main/glheader.h"

-#include "main/context.h"

-#include "main/colormac.h"

-#include "main/simple_list.h"

-#include "t_context.h"

-#include "t_vertex.h"

-

-

-#if 0

-#define DEBUG_INSERT printf("%s\n", __FUNCTION__)

-#else

-#define DEBUG_INSERT

-#endif

-

-

-/*

- * These functions take the NDC coordinates pointed to by 'in', apply the

- * NDC->Viewport mapping and store the results at 'v'.

- */

-

-static INLINE void insert_4f_viewport_4( const struct tnl_clipspace_attr *a, GLubyte *v,

-                      const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)v;

-   const GLfloat * const vp = a->vp;

-   DEBUG_INSERT;

-   out[0] = vp[0] * in[0] + vp[12];

-   out[1] = vp[5] * in[1] + vp[13];

-   out[2] = vp[10] * in[2] + vp[14];

-   out[3] = in[3];

-}

-

-static INLINE void insert_4f_viewport_3( const struct tnl_clipspace_attr *a, GLubyte *v,

-				const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)v;

-   const GLfloat * const vp = a->vp;

-   DEBUG_INSERT;

-   out[0] = vp[0] * in[0] + vp[12];

-   out[1] = vp[5] * in[1] + vp[13];

-   out[2] = vp[10] * in[2] + vp[14];

-   out[3] = 1;

-}

-

-static INLINE void insert_4f_viewport_2( const struct tnl_clipspace_attr *a, GLubyte *v,

-				const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)v;

-   const GLfloat * const vp = a->vp;

-   DEBUG_INSERT;

-   out[0] = vp[0] * in[0] + vp[12];

-   out[1] = vp[5] * in[1] + vp[13];

-   out[2] = vp[14];

-   out[3] = 1;

-}

-

-static INLINE void insert_4f_viewport_1( const struct tnl_clipspace_attr *a, GLubyte *v,

-				const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)v;

-   const GLfloat * const vp = a->vp;

-   DEBUG_INSERT;

-   out[0] = vp[0] * in[0] + vp[12];

-   out[1] = vp[13];

-   out[2] = vp[14];

-   out[3] = 1;

-}

-

-static INLINE void insert_3f_viewport_3( const struct tnl_clipspace_attr *a, GLubyte *v,

-				const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)v;

-   const GLfloat * const vp = a->vp;

-   DEBUG_INSERT;

-   out[0] = vp[0] * in[0] + vp[12];

-   out[1] = vp[5] * in[1] + vp[13];

-   out[2] = vp[10] * in[2] + vp[14];

-}

-

-static INLINE void insert_3f_viewport_2( const struct tnl_clipspace_attr *a, GLubyte *v,

-				const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)v;

-   const GLfloat * const vp = a->vp;

-   DEBUG_INSERT;

-   out[0] = vp[0] * in[0] + vp[12];

-   out[1] = vp[5] * in[1] + vp[13];

-   out[2] = vp[14];

-}

-

-static INLINE void insert_3f_viewport_1( const struct tnl_clipspace_attr *a, GLubyte *v,

-				const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)v;

-   const GLfloat * const vp = a->vp;

-   DEBUG_INSERT;

-   out[0] = vp[0] * in[0] + vp[12];

-   out[1] = vp[13];

-   out[2] = vp[14];

-}

-

-static INLINE void insert_2f_viewport_2( const struct tnl_clipspace_attr *a, GLubyte *v,

-				const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)v;

-   const GLfloat * const vp = a->vp;

-   DEBUG_INSERT;

-   out[0] = vp[0] * in[0] + vp[12];

-   out[1] = vp[5] * in[1] + vp[13];

-}

-

-static INLINE void insert_2f_viewport_1( const struct tnl_clipspace_attr *a, GLubyte *v,

-				const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)v;

-   const GLfloat * const vp = a->vp;

-   DEBUG_INSERT;

-   out[0] = vp[0] * in[0] + vp[12];

-   out[1] = vp[13];

-}

-

-

-/*

- * These functions do the same as above, except for the viewport mapping.

- */

-

-static INLINE void insert_4f_4( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)(v);

-   (void) a;

-   DEBUG_INSERT;

-   out[0] = in[0];

-   out[1] = in[1];

-   out[2] = in[2];

-   out[3] = in[3];

-}

-

-static INLINE void insert_4f_3( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)(v);

-   (void) a;

-   DEBUG_INSERT;

-   out[0] = in[0];

-   out[1] = in[1];

-   out[2] = in[2];

-   out[3] = 1;

-}

-

-static INLINE void insert_4f_2( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)(v);

-   (void) a;

-   DEBUG_INSERT;

-   out[0] = in[0];

-   out[1] = in[1];

-   out[2] = 0;

-   out[3] = 1;

-}

-

-static INLINE void insert_4f_1( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)(v);

-   (void) a;

-   DEBUG_INSERT;

-   out[0] = in[0];

-   out[1] = 0;

-   out[2] = 0;

-   out[3] = 1;

-}

-

-static INLINE void insert_3f_xyw_4( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)(v);

-   (void) a;

-   DEBUG_INSERT;

-   out[0] = in[0];

-   out[1] = in[1];

-   out[2] = in[3];

-}

-

-static INLINE void insert_3f_xyw_err( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )

-{

-   (void) a; (void) v; (void) in;

-   DEBUG_INSERT;

-   exit(1);

-}

-

-static INLINE void insert_3f_3( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)(v);

-   (void) a;

-   DEBUG_INSERT;

-   out[0] = in[0];

-   out[1] = in[1];

-   out[2] = in[2];

-}

-

-static INLINE void insert_3f_2( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)(v);

-   (void) a;

-   DEBUG_INSERT;

-   out[0] = in[0];

-   out[1] = in[1];

-   out[2] = 0;

-}

-

-static INLINE void insert_3f_1( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)(v);

-   (void) a;

-   DEBUG_INSERT;

-   out[0] = in[0];

-   out[1] = 0;

-   out[2] = 0;

-}

-

-

-static INLINE void insert_2f_2( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)(v);

-   (void) a;

-   DEBUG_INSERT;

-   out[0] = in[0];

-   out[1] = in[1];

-}

-

-static INLINE void insert_2f_1( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)(v);

-   (void) a;

-   DEBUG_INSERT;

-   out[0] = in[0];

-   out[1] = 0;

-}

-

-static INLINE void insert_1f_1( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )

-{

-   GLfloat *out = (GLfloat *)(v);

-   (void) a;

-   DEBUG_INSERT;

-   out[0] = in[0];

-}

-

-static INLINE void insert_null( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a; (void) v; (void) in;

-}

-

-static INLINE void insert_4chan_4f_rgba_4( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				  const GLfloat *in )

-{

-   GLchan *c = (GLchan *)v;

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_CHAN(c[0], in[0]); 

-   UNCLAMPED_FLOAT_TO_CHAN(c[1], in[1]); 

-   UNCLAMPED_FLOAT_TO_CHAN(c[2], in[2]); 

-   UNCLAMPED_FLOAT_TO_CHAN(c[3], in[3]);

-}

-

-static INLINE void insert_4chan_4f_rgba_3( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				  const GLfloat *in )

-{

-   GLchan *c = (GLchan *)v;

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_CHAN(c[0], in[0]); 

-   UNCLAMPED_FLOAT_TO_CHAN(c[1], in[1]); 

-   UNCLAMPED_FLOAT_TO_CHAN(c[2], in[2]); 

-   c[3] = CHAN_MAX;

-}

-

-static INLINE void insert_4chan_4f_rgba_2( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				  const GLfloat *in )

-{

-   GLchan *c = (GLchan *)v;

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_CHAN(c[0], in[0]); 

-   UNCLAMPED_FLOAT_TO_CHAN(c[1], in[1]); 

-   c[2] = 0;

-   c[3] = CHAN_MAX;

-}

-

-static INLINE void insert_4chan_4f_rgba_1( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				  const GLfloat *in )

-{

-   GLchan *c = (GLchan *)v;

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_CHAN(c[0], in[0]); 

-   c[1] = 0;

-   c[2] = 0;

-   c[3] = CHAN_MAX;

-}

-

-static INLINE void insert_4ub_4f_rgba_4( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[2]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[3]);

-}

-

-static INLINE void insert_4ub_4f_rgba_3( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[2]);

-   v[3] = 0xff;

-}

-

-static INLINE void insert_4ub_4f_rgba_2( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);

-   v[2] = 0;

-   v[3] = 0xff;

-}

-

-static INLINE void insert_4ub_4f_rgba_1( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);

-   v[1] = 0;

-   v[2] = 0;

-   v[3] = 0xff;

-}

-

-static INLINE void insert_4ub_4f_bgra_4( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[2]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[3]);

-}

-

-static INLINE void insert_4ub_4f_bgra_3( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[2]);

-   v[3] = 0xff;

-}

-

-static INLINE void insert_4ub_4f_bgra_2( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);

-   v[0] = 0;

-   v[3] = 0xff;

-}

-

-static INLINE void insert_4ub_4f_bgra_1( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);

-   v[1] = 0;

-   v[0] = 0;

-   v[3] = 0xff;

-}

-

-static INLINE void insert_4ub_4f_argb_4( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[0]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[2]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[3]);

-}

-

-static INLINE void insert_4ub_4f_argb_3( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[0]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[2]);

-   v[0] = 0xff;

-}

-

-static INLINE void insert_4ub_4f_argb_2( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[0]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);

-   v[3] = 0x00;

-   v[0] = 0xff;

-}

-

-static INLINE void insert_4ub_4f_argb_1( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[0]);

-   v[2] = 0x00;

-   v[3] = 0x00;

-   v[0] = 0xff;

-}

-

-static INLINE void insert_4ub_4f_abgr_4( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[0]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[2]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[3]);

-}

-

-static INLINE void insert_4ub_4f_abgr_3( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[0]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[2]);

-   v[0] = 0xff;

-}

-

-static INLINE void insert_4ub_4f_abgr_2( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[0]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);

-   v[1] = 0x00;

-   v[0] = 0xff;

-}

-

-static INLINE void insert_4ub_4f_abgr_1( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[0]);

-   v[2] = 0x00;

-   v[1] = 0x00;

-   v[0] = 0xff;

-}

-

-static INLINE void insert_3ub_3f_rgb_3( const struct tnl_clipspace_attr *a, GLubyte *v, 

-			       const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[2]);

-}

-

-static INLINE void insert_3ub_3f_rgb_2( const struct tnl_clipspace_attr *a, GLubyte *v, 

-			       const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);

-   v[2] = 0;

-}

-

-static INLINE void insert_3ub_3f_rgb_1( const struct tnl_clipspace_attr *a, GLubyte *v, 

-			       const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);

-   v[1] = 0;

-   v[2] = 0;

-}

-

-static INLINE void insert_3ub_3f_bgr_3( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				 const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[2]);

-}

-

-static INLINE void insert_3ub_3f_bgr_2( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				 const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);

-   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);

-   v[0] = 0;

-}

-

-static INLINE void insert_3ub_3f_bgr_1( const struct tnl_clipspace_attr *a, GLubyte *v, 

-				 const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);

-   v[1] = 0;

-   v[0] = 0;

-}

-

-

-static INLINE void insert_1ub_1f_1( const struct tnl_clipspace_attr *a, GLubyte *v, 

-			   const GLfloat *in )

-{

-   DEBUG_INSERT;

-   (void) a;

-   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);

-}

-

-

-/***********************************************************************

- * Functions to perform the reverse operations to the above, for

- * swrast translation and clip-interpolation.

- * 

- * Currently always extracts a full 4 floats.

- */

-

-static void extract_4f_viewport( const struct tnl_clipspace_attr *a, GLfloat *out, 

-				 const GLubyte *v )

-{

-   const GLfloat *in = (const GLfloat *)v;

-   const GLfloat * const vp = a->vp;

-   

-   /* Although included for completeness, the position coordinate is

-    * usually handled differently during clipping.

-    */

-   DEBUG_INSERT;

-   out[0] = (in[0] - vp[12]) / vp[0];

-   out[1] = (in[1] - vp[13]) / vp[5];

-   out[2] = (in[2] - vp[14]) / vp[10];

-   out[3] = in[3];

-}

-

-static void extract_3f_viewport( const struct tnl_clipspace_attr *a, GLfloat *out, 

-				 const GLubyte *v )

-{

-   const GLfloat *in = (const GLfloat *)v;

-   const GLfloat * const vp = a->vp;

-   DEBUG_INSERT;

-   out[0] = (in[0] - vp[12]) / vp[0];

-   out[1] = (in[1] - vp[13]) / vp[5];

-   out[2] = (in[2] - vp[14]) / vp[10];

-   out[3] = 1;

-}

-

-

-static void extract_2f_viewport( const struct tnl_clipspace_attr *a, GLfloat *out, 

-				 const GLubyte *v )

-{

-   const GLfloat *in = (const GLfloat *)v;

-   const GLfloat * const vp = a->vp;

-   DEBUG_INSERT;

-   out[0] = (in[0] - vp[12]) / vp[0];

-   out[1] = (in[1] - vp[13]) / vp[5];

-   out[2] = 0;

-   out[3] = 1;

-}

-

-

-static void extract_4f( const struct tnl_clipspace_attr *a, GLfloat *out, const GLubyte *v  )

-{

-   const GLfloat *in = (const GLfloat *)v;

-   (void) a;

-   

-   out[0] = in[0];

-   out[1] = in[1];

-   out[2] = in[2];

-   out[3] = in[3];

-}

-

-static void extract_3f_xyw( const struct tnl_clipspace_attr *a, GLfloat *out, const GLubyte *v )

-{

-   const GLfloat *in = (const GLfloat *)v;

-   (void) a;

-   

-   out[0] = in[0];

-   out[1] = in[1];

-   out[2] = 0;

-   out[3] = in[2];

-}

-

-

-static void extract_3f( const struct tnl_clipspace_attr *a, GLfloat *out, const GLubyte *v )

-{

-   const GLfloat *in = (const GLfloat *)v;

-   (void) a;

-   

-   out[0] = in[0];

-   out[1] = in[1];

-   out[2] = in[2];

-   out[3] = 1;

-}

-

-

-static void extract_2f( const struct tnl_clipspace_attr *a, GLfloat *out, const GLubyte *v )

-{

-   const GLfloat *in = (const GLfloat *)v;

-   (void) a;

-   

-   out[0] = in[0];

-   out[1] = in[1];

-   out[2] = 0;

-   out[3] = 1;

-}

-

-static void extract_1f( const struct tnl_clipspace_attr *a, GLfloat *out, const GLubyte *v )

-{

-   const GLfloat *in = (const GLfloat *)v;

-   (void) a;

-   

-   out[0] = in[0];

-   out[1] = 0;

-   out[2] = 0;

-   out[3] = 1;

-}

-

-static void extract_4chan_4f_rgba( const struct tnl_clipspace_attr *a, GLfloat *out, 

-				 const GLubyte *v )

-{

-   GLchan *c = (GLchan *)v;

-   (void) a;

-

-   out[0] = CHAN_TO_FLOAT(c[0]);

-   out[1] = CHAN_TO_FLOAT(c[1]);

-   out[2] = CHAN_TO_FLOAT(c[2]);

-   out[3] = CHAN_TO_FLOAT(c[3]);

-}

-

-static void extract_4ub_4f_rgba( const struct tnl_clipspace_attr *a, GLfloat *out, 

-				 const GLubyte *v )

-{

-   (void) a;

-   out[0] = UBYTE_TO_FLOAT(v[0]);

-   out[1] = UBYTE_TO_FLOAT(v[1]);

-   out[2] = UBYTE_TO_FLOAT(v[2]);

-   out[3] = UBYTE_TO_FLOAT(v[3]);

-}

-

-static void extract_4ub_4f_bgra( const struct tnl_clipspace_attr *a, GLfloat *out, 

-				 const GLubyte *v )

-{

-   (void) a;

-   out[2] = UBYTE_TO_FLOAT(v[0]);

-   out[1] = UBYTE_TO_FLOAT(v[1]);

-   out[0] = UBYTE_TO_FLOAT(v[2]);

-   out[3] = UBYTE_TO_FLOAT(v[3]);

-}

-

-static void extract_4ub_4f_argb( const struct tnl_clipspace_attr *a, GLfloat *out, 

-				 const GLubyte *v )

-{

-   (void) a;

-   out[3] = UBYTE_TO_FLOAT(v[0]);

-   out[0] = UBYTE_TO_FLOAT(v[1]);

-   out[1] = UBYTE_TO_FLOAT(v[2]);

-   out[2] = UBYTE_TO_FLOAT(v[3]);

-}

-

-static void extract_4ub_4f_abgr( const struct tnl_clipspace_attr *a, GLfloat *out, 

-				 const GLubyte *v )

-{

-   (void) a;

-   out[3] = UBYTE_TO_FLOAT(v[0]);

-   out[2] = UBYTE_TO_FLOAT(v[1]);

-   out[1] = UBYTE_TO_FLOAT(v[2]);

-   out[0] = UBYTE_TO_FLOAT(v[3]);

-}

-

-static void extract_3ub_3f_rgb( const struct tnl_clipspace_attr *a, GLfloat *out, 

-				const GLubyte *v )

-{

-   (void) a;

-   out[0] = UBYTE_TO_FLOAT(v[0]);

-   out[1] = UBYTE_TO_FLOAT(v[1]);

-   out[2] = UBYTE_TO_FLOAT(v[2]);

-   out[3] = 1;

-}

-

-static void extract_3ub_3f_bgr( const struct tnl_clipspace_attr *a, GLfloat *out, 

-				const GLubyte *v )

-{

-   (void) a;

-   out[2] = UBYTE_TO_FLOAT(v[0]);

-   out[1] = UBYTE_TO_FLOAT(v[1]);

-   out[0] = UBYTE_TO_FLOAT(v[2]);

-   out[3] = 1;

-}

-

-static void extract_1ub_1f( const struct tnl_clipspace_attr *a, GLfloat *out, const GLubyte *v )

-{

-   (void) a;

-   out[0] = UBYTE_TO_FLOAT(v[0]);

-   out[1] = 0;

-   out[2] = 0;

-   out[3] = 1;

-}

-

-

-const struct tnl_format_info _tnl_format_info[EMIT_MAX] = 

-{

-   { "1f",

-     extract_1f,

-     { insert_1f_1, insert_1f_1, insert_1f_1, insert_1f_1 },

-     sizeof(GLfloat) },

-

-   { "2f",

-     extract_2f,

-     { insert_2f_1, insert_2f_2, insert_2f_2, insert_2f_2 },

-     2 * sizeof(GLfloat) },

-

-   { "3f",

-     extract_3f,

-     { insert_3f_1, insert_3f_2, insert_3f_3, insert_3f_3 },

-     3 * sizeof(GLfloat) },

-

-   { "4f",

-     extract_4f,

-     { insert_4f_1, insert_4f_2, insert_4f_3, insert_4f_4 },

-     4 * sizeof(GLfloat) },

-

-   { "2f_viewport",

-     extract_2f_viewport,

-     { insert_2f_viewport_1, insert_2f_viewport_2, insert_2f_viewport_2,

-       insert_2f_viewport_2 },

-     2 * sizeof(GLfloat) },

-

-   { "3f_viewport",

-     extract_3f_viewport,

-     { insert_3f_viewport_1, insert_3f_viewport_2, insert_3f_viewport_3,

-       insert_3f_viewport_3 },

-     3 * sizeof(GLfloat) },

-

-   { "4f_viewport",

-     extract_4f_viewport,

-     { insert_4f_viewport_1, insert_4f_viewport_2, insert_4f_viewport_3,

-       insert_4f_viewport_4 }, 

-     4 * sizeof(GLfloat) },

-

-   { "3f_xyw",

-     extract_3f_xyw,

-     { insert_3f_xyw_err, insert_3f_xyw_err, insert_3f_xyw_err, 

-       insert_3f_xyw_4 },

-     3 * sizeof(GLfloat) },

-

-   { "1ub_1f",

-     extract_1ub_1f,

-     { insert_1ub_1f_1, insert_1ub_1f_1, insert_1ub_1f_1, insert_1ub_1f_1 },

-     sizeof(GLubyte) },

-

-   { "3ub_3f_rgb",

-     extract_3ub_3f_rgb,

-     { insert_3ub_3f_rgb_1, insert_3ub_3f_rgb_2, insert_3ub_3f_rgb_3,

-       insert_3ub_3f_rgb_3 },

-     3 * sizeof(GLubyte) },

-

-   { "3ub_3f_bgr",

-     extract_3ub_3f_bgr,

-     { insert_3ub_3f_bgr_1, insert_3ub_3f_bgr_2, insert_3ub_3f_bgr_3,

-       insert_3ub_3f_bgr_3 },

-     3 * sizeof(GLubyte) },

-

-   { "4ub_4f_rgba",

-     extract_4ub_4f_rgba,

-     { insert_4ub_4f_rgba_1, insert_4ub_4f_rgba_2, insert_4ub_4f_rgba_3, 

-       insert_4ub_4f_rgba_4 },

-     4 * sizeof(GLubyte) },

-

-   { "4ub_4f_bgra",

-     extract_4ub_4f_bgra,

-     { insert_4ub_4f_bgra_1, insert_4ub_4f_bgra_2, insert_4ub_4f_bgra_3,

-       insert_4ub_4f_bgra_4 },

-     4 * sizeof(GLubyte) },

-

-   { "4ub_4f_argb",

-     extract_4ub_4f_argb,

-     { insert_4ub_4f_argb_1, insert_4ub_4f_argb_2, insert_4ub_4f_argb_3,

-       insert_4ub_4f_argb_4 },

-     4 * sizeof(GLubyte) },

-

-   { "4ub_4f_abgr",

-     extract_4ub_4f_abgr,

-     { insert_4ub_4f_abgr_1, insert_4ub_4f_abgr_2, insert_4ub_4f_abgr_3,

-       insert_4ub_4f_abgr_4 },

-     4 * sizeof(GLubyte) },

-

-   { "4chan_4f_rgba",

-     extract_4chan_4f_rgba,

-     { insert_4chan_4f_rgba_1, insert_4chan_4f_rgba_2, insert_4chan_4f_rgba_3,

-       insert_4chan_4f_rgba_4 },

-     4 * sizeof(GLchan) },

-

-   { "pad",

-     NULL,

-     { NULL, NULL, NULL, NULL },

-     0 }

-

-};

-

-

-

-    

-/***********************************************************************

- * Hardwired fastpaths for emitting whole vertices or groups of

- * vertices

- */

-#define EMIT5(NR, F0, F1, F2, F3, F4, NAME)				\

-static void NAME( struct gl_context *ctx,					\

-		  GLuint count,						\

-		  GLubyte *v )						\

-{									\

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);			\

-   struct tnl_clipspace_attr *a = vtx->attr;				\

-   GLuint i;								\

-									\

-   for (i = 0 ; i < count ; i++, v += vtx->vertex_size) {		\

-      if (NR > 0) {							\

-	 F0( &a[0], v + a[0].vertoffset, (GLfloat *)a[0].inputptr );	\

-	 a[0].inputptr += a[0].inputstride;				\

-      }									\

-      									\

-      if (NR > 1) {							\

-	 F1( &a[1], v + a[1].vertoffset, (GLfloat *)a[1].inputptr );	\

-	 a[1].inputptr += a[1].inputstride;				\

-      }									\

-      									\

-      if (NR > 2) {							\

-	 F2( &a[2], v + a[2].vertoffset, (GLfloat *)a[2].inputptr );	\

-	 a[2].inputptr += a[2].inputstride;				\

-      }									\

-      									\

-      if (NR > 3) {							\

-	 F3( &a[3], v + a[3].vertoffset, (GLfloat *)a[3].inputptr );	\

-	 a[3].inputptr += a[3].inputstride;				\

-      }									\

-									\

-      if (NR > 4) {							\

-	 F4( &a[4], v + a[4].vertoffset, (GLfloat *)a[4].inputptr );	\

-	 a[4].inputptr += a[4].inputstride;				\

-      }									\

-   }									\

-}

-

-   

-#define EMIT2(F0, F1, NAME) EMIT5(2, F0, F1, insert_null, \

-				  insert_null, insert_null, NAME)

-

-#define EMIT3(F0, F1, F2, NAME) EMIT5(3, F0, F1, F2, insert_null, \

-				      insert_null, NAME)

-   

-#define EMIT4(F0, F1, F2, F3, NAME) EMIT5(4, F0, F1, F2, F3, \

-				          insert_null, NAME)

-   

-

-EMIT2(insert_3f_viewport_3, insert_4ub_4f_rgba_4, emit_viewport3_rgba4)

-EMIT2(insert_3f_viewport_3, insert_4ub_4f_bgra_4, emit_viewport3_bgra4)

-EMIT2(insert_3f_3, insert_4ub_4f_rgba_4, emit_xyz3_rgba4)

-

-EMIT3(insert_4f_viewport_4, insert_4ub_4f_rgba_4, insert_2f_2, emit_viewport4_rgba4_st2)

-EMIT3(insert_4f_viewport_4, insert_4ub_4f_bgra_4, insert_2f_2,  emit_viewport4_bgra4_st2)

-EMIT3(insert_4f_4, insert_4ub_4f_rgba_4, insert_2f_2, emit_xyzw4_rgba4_st2)

-

-EMIT4(insert_4f_viewport_4, insert_4ub_4f_rgba_4, insert_2f_2, insert_2f_2, emit_viewport4_rgba4_st2_st2)

-EMIT4(insert_4f_viewport_4, insert_4ub_4f_bgra_4, insert_2f_2, insert_2f_2,  emit_viewport4_bgra4_st2_st2)

-EMIT4(insert_4f_4, insert_4ub_4f_rgba_4, insert_2f_2, insert_2f_2, emit_xyzw4_rgba4_st2_st2)

-

-

-/* Use the codegen paths to select one of a number of hardwired

- * fastpaths.

- */

-void _tnl_generate_hardwired_emit( struct gl_context *ctx )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-   tnl_emit_func func = NULL;

-

-   /* Does it fit a hardwired fastpath?  Help! this is growing out of

-    * control!

-    */

-   switch (vtx->attr_count) {

-   case 2:

-      if (vtx->attr[0].emit == insert_3f_viewport_3) {

-	 if (vtx->attr[1].emit == insert_4ub_4f_bgra_4) 

-	    func = emit_viewport3_bgra4;

-	 else if (vtx->attr[1].emit == insert_4ub_4f_rgba_4) 

-	    func = emit_viewport3_rgba4;

-      }

-      else if (vtx->attr[0].emit == insert_3f_3 &&

-	       vtx->attr[1].emit == insert_4ub_4f_rgba_4) {

- 	 func = emit_xyz3_rgba4; 

-      }

-      break;

-   case 3:

-      if (vtx->attr[2].emit == insert_2f_2) {

-	 if (vtx->attr[1].emit == insert_4ub_4f_rgba_4) {

-	    if (vtx->attr[0].emit == insert_4f_viewport_4)

-	       func = emit_viewport4_rgba4_st2;

-	    else if (vtx->attr[0].emit == insert_4f_4) 

-	       func = emit_xyzw4_rgba4_st2;

-	 }

-	 else if (vtx->attr[1].emit == insert_4ub_4f_bgra_4 &&

-		  vtx->attr[0].emit == insert_4f_viewport_4)

-	    func = emit_viewport4_bgra4_st2;

-      }

-      break;

-   case 4:

-      if (vtx->attr[2].emit == insert_2f_2 &&

-	  vtx->attr[3].emit == insert_2f_2) {

-	 if (vtx->attr[1].emit == insert_4ub_4f_rgba_4) {

-	    if (vtx->attr[0].emit == insert_4f_viewport_4)

-	       func = emit_viewport4_rgba4_st2_st2;

-	    else if (vtx->attr[0].emit == insert_4f_4) 

-	       func = emit_xyzw4_rgba4_st2_st2;

-	 }

-	 else if (vtx->attr[1].emit == insert_4ub_4f_bgra_4 &&

-		  vtx->attr[0].emit == insert_4f_viewport_4)

-	    func = emit_viewport4_bgra4_st2_st2;

-      }

-      break;

-   }

-

-   vtx->emit = func;

-}

-

-/***********************************************************************

- * Generic (non-codegen) functions for whole vertices or groups of

- * vertices

- */

-

-void _tnl_generic_emit( struct gl_context *ctx,

-			GLuint count,

-			GLubyte *v )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-   struct tnl_clipspace_attr *a = vtx->attr;

-   const GLuint attr_count = vtx->attr_count;

-   const GLuint stride = vtx->vertex_size;

-   GLuint i, j;

-

-   for (i = 0 ; i < count ; i++, v += stride) {

-      for (j = 0; j < attr_count; j++) {

-	 GLfloat *in = (GLfloat *)a[j].inputptr;

-	 a[j].inputptr += a[j].inputstride;

-	 a[j].emit( &a[j], v + a[j].vertoffset, in );

-      }

-   }

-}

-

-

-void _tnl_generic_interp( struct gl_context *ctx,

-			    GLfloat t,

-			    GLuint edst, GLuint eout, GLuint ein,

-			    GLboolean force_boundary )

-{

-   TNLcontext *tnl = TNL_CONTEXT(ctx);

-   struct vertex_buffer *VB = &tnl->vb;

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-   const GLubyte *vin  = vtx->vertex_buf + ein  * vtx->vertex_size;

-   const GLubyte *vout = vtx->vertex_buf + eout * vtx->vertex_size;

-   GLubyte *vdst = vtx->vertex_buf + edst * vtx->vertex_size;

-   const struct tnl_clipspace_attr *a = vtx->attr;

-   const GLuint attr_count = vtx->attr_count;

-   GLuint j;

-   (void) force_boundary;

-

-   if (tnl->NeedNdcCoords) {

-      const GLfloat *dstclip = VB->ClipPtr->data[edst];

-      if (dstclip[3] != 0.0) {

-	 const GLfloat w = 1.0f / dstclip[3];

-	 GLfloat pos[4];

-

-	 pos[0] = dstclip[0] * w;

-	 pos[1] = dstclip[1] * w;

-	 pos[2] = dstclip[2] * w;

-	 pos[3] = w;

-

-	 a[0].insert[4-1]( &a[0], vdst, pos );

-      }

-   }

-   else {

-      a[0].insert[4-1]( &a[0], vdst, VB->ClipPtr->data[edst] );

-   }

-

-

-   for (j = 1; j < attr_count; j++) {

-      GLfloat fin[4], fout[4], fdst[4];

-	 

-      a[j].extract( &a[j], fin, vin + a[j].vertoffset );

-      a[j].extract( &a[j], fout, vout + a[j].vertoffset );

-

-      INTERP_F( t, fdst[3], fout[3], fin[3] );

-      INTERP_F( t, fdst[2], fout[2], fin[2] );

-      INTERP_F( t, fdst[1], fout[1], fin[1] );

-      INTERP_F( t, fdst[0], fout[0], fin[0] );

-

-      a[j].insert[4-1]( &a[j], vdst + a[j].vertoffset, fdst );

-   }

-}

-

-

-/* Extract color attributes from one vertex and insert them into

- * another.  (Shortcircuit extract/insert with memcpy).

- */

-void _tnl_generic_copy_pv( struct gl_context *ctx, GLuint edst, GLuint esrc )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-   GLubyte *vsrc = vtx->vertex_buf + esrc * vtx->vertex_size;

-   GLubyte *vdst = vtx->vertex_buf + edst * vtx->vertex_size;

-   const struct tnl_clipspace_attr *a = vtx->attr;

-   const GLuint attr_count = vtx->attr_count;

-   GLuint j;

-

-   for (j = 0; j < attr_count; j++) {

-      if (a[j].attrib == VERT_ATTRIB_COLOR0 ||

-	  a[j].attrib == VERT_ATTRIB_COLOR1) {

-

-	 memcpy( vdst + a[j].vertoffset,

-                 vsrc + a[j].vertoffset,

-                 a[j].vertattrsize );

-      }

-   }

-}

-

-

-/* Helper functions for hardware which doesn't put back colors and/or

- * edgeflags into vertices.

- */

-void _tnl_generic_interp_extras( struct gl_context *ctx,

-				   GLfloat t,

-				   GLuint dst, GLuint out, GLuint in,

-				   GLboolean force_boundary )

-{

-   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;

-

-   /* If stride is zero, BackfaceColorPtr is constant across the VB, so

-    * there is no point interpolating between two values as they will

-    * be identical.  In all other cases, this value is generated by

-    * t_vb_lighttmp.h and has a stride of 4 dwords.

-    */

-   if (VB->BackfaceColorPtr && VB->BackfaceColorPtr->stride) {

-      assert(VB->BackfaceColorPtr->stride == 4 * sizeof(GLfloat));

-

-      INTERP_4F( t,

-		 VB->BackfaceColorPtr->data[dst],

-		 VB->BackfaceColorPtr->data[out],

-		 VB->BackfaceColorPtr->data[in] );

-   }

-

-   if (VB->BackfaceSecondaryColorPtr) {

-      assert(VB->BackfaceSecondaryColorPtr->stride == 4 * sizeof(GLfloat));

-      

-      INTERP_3F( t,

-		 VB->BackfaceSecondaryColorPtr->data[dst],

-		 VB->BackfaceSecondaryColorPtr->data[out],

-		 VB->BackfaceSecondaryColorPtr->data[in] );

-   }

-   

-   if (VB->BackfaceIndexPtr) {

-      VB->BackfaceIndexPtr->data[dst][0] = LINTERP( t,

-					       VB->BackfaceIndexPtr->data[out][0],

-					       VB->BackfaceIndexPtr->data[in][0] );

-   }

-

-   if (VB->EdgeFlag) {

-      VB->EdgeFlag[dst] = VB->EdgeFlag[out] || force_boundary;

-   }

-

-   _tnl_generic_interp(ctx, t, dst, out, in, force_boundary);

-}

-

-void _tnl_generic_copy_pv_extras( struct gl_context *ctx, 

-				  GLuint dst, GLuint src )

-{

-   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;

-

-   /* See above comment:

-    */

-   if (VB->BackfaceColorPtr && VB->BackfaceColorPtr->stride) {

-      COPY_4FV( VB->BackfaceColorPtr->data[dst],

-		VB->BackfaceColorPtr->data[src] );

-   }

-

-   if (VB->BackfaceSecondaryColorPtr) {

-      COPY_4FV( VB->BackfaceSecondaryColorPtr->data[dst],

-		VB->BackfaceSecondaryColorPtr->data[src] );

-   }

-

-   if (VB->BackfaceIndexPtr) {

-      VB->BackfaceIndexPtr->data[dst][0] = VB->BackfaceIndexPtr->data[src][0];

-   }

-

-   _tnl_generic_copy_pv(ctx, dst, src);

-}

-

-

+
+/*
+ * Copyright 2003 Tungsten Graphics, inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <keithw@tungstengraphics.com>
+ */
+
+#include "main/glheader.h"
+#include "main/context.h"
+#include "main/colormac.h"
+#include "main/simple_list.h"
+#include "swrast/s_chan.h"
+#include "t_context.h"
+#include "t_vertex.h"
+
+
+#if 0
+#define DEBUG_INSERT printf("%s\n", __FUNCTION__)
+#else
+#define DEBUG_INSERT
+#endif
+
+
+/*
+ * These functions take the NDC coordinates pointed to by 'in', apply the
+ * NDC->Viewport mapping and store the results at 'v'.
+ */
+
+static INLINE void insert_4f_viewport_4( const struct tnl_clipspace_attr *a, GLubyte *v,
+                      const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)v;
+   const GLfloat * const vp = a->vp;
+   DEBUG_INSERT;
+   out[0] = vp[0] * in[0] + vp[12];
+   out[1] = vp[5] * in[1] + vp[13];
+   out[2] = vp[10] * in[2] + vp[14];
+   out[3] = in[3];
+}
+
+static INLINE void insert_4f_viewport_3( const struct tnl_clipspace_attr *a, GLubyte *v,
+				const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)v;
+   const GLfloat * const vp = a->vp;
+   DEBUG_INSERT;
+   out[0] = vp[0] * in[0] + vp[12];
+   out[1] = vp[5] * in[1] + vp[13];
+   out[2] = vp[10] * in[2] + vp[14];
+   out[3] = 1;
+}
+
+static INLINE void insert_4f_viewport_2( const struct tnl_clipspace_attr *a, GLubyte *v,
+				const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)v;
+   const GLfloat * const vp = a->vp;
+   DEBUG_INSERT;
+   out[0] = vp[0] * in[0] + vp[12];
+   out[1] = vp[5] * in[1] + vp[13];
+   out[2] = vp[14];
+   out[3] = 1;
+}
+
+static INLINE void insert_4f_viewport_1( const struct tnl_clipspace_attr *a, GLubyte *v,
+				const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)v;
+   const GLfloat * const vp = a->vp;
+   DEBUG_INSERT;
+   out[0] = vp[0] * in[0] + vp[12];
+   out[1] = vp[13];
+   out[2] = vp[14];
+   out[3] = 1;
+}
+
+static INLINE void insert_3f_viewport_3( const struct tnl_clipspace_attr *a, GLubyte *v,
+				const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)v;
+   const GLfloat * const vp = a->vp;
+   DEBUG_INSERT;
+   out[0] = vp[0] * in[0] + vp[12];
+   out[1] = vp[5] * in[1] + vp[13];
+   out[2] = vp[10] * in[2] + vp[14];
+}
+
+static INLINE void insert_3f_viewport_2( const struct tnl_clipspace_attr *a, GLubyte *v,
+				const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)v;
+   const GLfloat * const vp = a->vp;
+   DEBUG_INSERT;
+   out[0] = vp[0] * in[0] + vp[12];
+   out[1] = vp[5] * in[1] + vp[13];
+   out[2] = vp[14];
+}
+
+static INLINE void insert_3f_viewport_1( const struct tnl_clipspace_attr *a, GLubyte *v,
+				const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)v;
+   const GLfloat * const vp = a->vp;
+   DEBUG_INSERT;
+   out[0] = vp[0] * in[0] + vp[12];
+   out[1] = vp[13];
+   out[2] = vp[14];
+}
+
+static INLINE void insert_2f_viewport_2( const struct tnl_clipspace_attr *a, GLubyte *v,
+				const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)v;
+   const GLfloat * const vp = a->vp;
+   DEBUG_INSERT;
+   out[0] = vp[0] * in[0] + vp[12];
+   out[1] = vp[5] * in[1] + vp[13];
+}
+
+static INLINE void insert_2f_viewport_1( const struct tnl_clipspace_attr *a, GLubyte *v,
+				const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)v;
+   const GLfloat * const vp = a->vp;
+   DEBUG_INSERT;
+   out[0] = vp[0] * in[0] + vp[12];
+   out[1] = vp[13];
+}
+
+
+/*
+ * These functions do the same as above, except for the viewport mapping.
+ */
+
+static INLINE void insert_4f_4( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)(v);
+   (void) a;
+   DEBUG_INSERT;
+   out[0] = in[0];
+   out[1] = in[1];
+   out[2] = in[2];
+   out[3] = in[3];
+}
+
+static INLINE void insert_4f_3( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)(v);
+   (void) a;
+   DEBUG_INSERT;
+   out[0] = in[0];
+   out[1] = in[1];
+   out[2] = in[2];
+   out[3] = 1;
+}
+
+static INLINE void insert_4f_2( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)(v);
+   (void) a;
+   DEBUG_INSERT;
+   out[0] = in[0];
+   out[1] = in[1];
+   out[2] = 0;
+   out[3] = 1;
+}
+
+static INLINE void insert_4f_1( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)(v);
+   (void) a;
+   DEBUG_INSERT;
+   out[0] = in[0];
+   out[1] = 0;
+   out[2] = 0;
+   out[3] = 1;
+}
+
+static INLINE void insert_3f_xyw_4( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)(v);
+   (void) a;
+   DEBUG_INSERT;
+   out[0] = in[0];
+   out[1] = in[1];
+   out[2] = in[3];
+}
+
+static INLINE void insert_3f_xyw_err( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )
+{
+   (void) a; (void) v; (void) in;
+   DEBUG_INSERT;
+   exit(1);
+}
+
+static INLINE void insert_3f_3( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)(v);
+   (void) a;
+   DEBUG_INSERT;
+   out[0] = in[0];
+   out[1] = in[1];
+   out[2] = in[2];
+}
+
+static INLINE void insert_3f_2( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)(v);
+   (void) a;
+   DEBUG_INSERT;
+   out[0] = in[0];
+   out[1] = in[1];
+   out[2] = 0;
+}
+
+static INLINE void insert_3f_1( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)(v);
+   (void) a;
+   DEBUG_INSERT;
+   out[0] = in[0];
+   out[1] = 0;
+   out[2] = 0;
+}
+
+
+static INLINE void insert_2f_2( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)(v);
+   (void) a;
+   DEBUG_INSERT;
+   out[0] = in[0];
+   out[1] = in[1];
+}
+
+static INLINE void insert_2f_1( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)(v);
+   (void) a;
+   DEBUG_INSERT;
+   out[0] = in[0];
+   out[1] = 0;
+}
+
+static INLINE void insert_1f_1( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )
+{
+   GLfloat *out = (GLfloat *)(v);
+   (void) a;
+   DEBUG_INSERT;
+   out[0] = in[0];
+}
+
+static INLINE void insert_null( const struct tnl_clipspace_attr *a, GLubyte *v, const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a; (void) v; (void) in;
+}
+
+static INLINE void insert_4chan_4f_rgba_4( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				  const GLfloat *in )
+{
+   GLchan *c = (GLchan *)v;
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_CHAN(c[0], in[0]); 
+   UNCLAMPED_FLOAT_TO_CHAN(c[1], in[1]); 
+   UNCLAMPED_FLOAT_TO_CHAN(c[2], in[2]); 
+   UNCLAMPED_FLOAT_TO_CHAN(c[3], in[3]);
+}
+
+static INLINE void insert_4chan_4f_rgba_3( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				  const GLfloat *in )
+{
+   GLchan *c = (GLchan *)v;
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_CHAN(c[0], in[0]); 
+   UNCLAMPED_FLOAT_TO_CHAN(c[1], in[1]); 
+   UNCLAMPED_FLOAT_TO_CHAN(c[2], in[2]); 
+   c[3] = CHAN_MAX;
+}
+
+static INLINE void insert_4chan_4f_rgba_2( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				  const GLfloat *in )
+{
+   GLchan *c = (GLchan *)v;
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_CHAN(c[0], in[0]); 
+   UNCLAMPED_FLOAT_TO_CHAN(c[1], in[1]); 
+   c[2] = 0;
+   c[3] = CHAN_MAX;
+}
+
+static INLINE void insert_4chan_4f_rgba_1( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				  const GLfloat *in )
+{
+   GLchan *c = (GLchan *)v;
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_CHAN(c[0], in[0]); 
+   c[1] = 0;
+   c[2] = 0;
+   c[3] = CHAN_MAX;
+}
+
+static INLINE void insert_4ub_4f_rgba_4( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[2]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[3]);
+}
+
+static INLINE void insert_4ub_4f_rgba_3( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[2]);
+   v[3] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_rgba_2( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   v[2] = 0;
+   v[3] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_rgba_1( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);
+   v[1] = 0;
+   v[2] = 0;
+   v[3] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_bgra_4( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[2]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[3]);
+}
+
+static INLINE void insert_4ub_4f_bgra_3( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[2]);
+   v[3] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_bgra_2( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   v[0] = 0;
+   v[3] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_bgra_1( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);
+   v[1] = 0;
+   v[0] = 0;
+   v[3] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_argb_4( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[2]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[3]);
+}
+
+static INLINE void insert_4ub_4f_argb_3( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[2]);
+   v[0] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_argb_2( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);
+   v[3] = 0x00;
+   v[0] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_argb_1( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[0]);
+   v[2] = 0x00;
+   v[3] = 0x00;
+   v[0] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_abgr_4( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[2]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[3]);
+}
+
+static INLINE void insert_4ub_4f_abgr_3( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[2]);
+   v[0] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_abgr_2( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[1]);
+   v[1] = 0x00;
+   v[0] = 0xff;
+}
+
+static INLINE void insert_4ub_4f_abgr_1( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[3], in[0]);
+   v[2] = 0x00;
+   v[1] = 0x00;
+   v[0] = 0xff;
+}
+
+static INLINE void insert_3ub_3f_rgb_3( const struct tnl_clipspace_attr *a, GLubyte *v, 
+			       const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[2]);
+}
+
+static INLINE void insert_3ub_3f_rgb_2( const struct tnl_clipspace_attr *a, GLubyte *v, 
+			       const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   v[2] = 0;
+}
+
+static INLINE void insert_3ub_3f_rgb_1( const struct tnl_clipspace_attr *a, GLubyte *v, 
+			       const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);
+   v[1] = 0;
+   v[2] = 0;
+}
+
+static INLINE void insert_3ub_3f_bgr_3( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				 const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[2]);
+}
+
+static INLINE void insert_3ub_3f_bgr_2( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				 const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);
+   UNCLAMPED_FLOAT_TO_UBYTE(v[1], in[1]);
+   v[0] = 0;
+}
+
+static INLINE void insert_3ub_3f_bgr_1( const struct tnl_clipspace_attr *a, GLubyte *v, 
+				 const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[2], in[0]);
+   v[1] = 0;
+   v[0] = 0;
+}
+
+
+static INLINE void insert_1ub_1f_1( const struct tnl_clipspace_attr *a, GLubyte *v, 
+			   const GLfloat *in )
+{
+   DEBUG_INSERT;
+   (void) a;
+   UNCLAMPED_FLOAT_TO_UBYTE(v[0], in[0]);
+}
+
+
+/***********************************************************************
+ * Functions to perform the reverse operations to the above, for
+ * swrast translation and clip-interpolation.
+ * 
+ * Currently always extracts a full 4 floats.
+ */
+
+static void extract_4f_viewport( const struct tnl_clipspace_attr *a, GLfloat *out, 
+				 const GLubyte *v )
+{
+   const GLfloat *in = (const GLfloat *)v;
+   const GLfloat * const vp = a->vp;
+   
+   /* Although included for completeness, the position coordinate is
+    * usually handled differently during clipping.
+    */
+   DEBUG_INSERT;
+   out[0] = (in[0] - vp[12]) / vp[0];
+   out[1] = (in[1] - vp[13]) / vp[5];
+   out[2] = (in[2] - vp[14]) / vp[10];
+   out[3] = in[3];
+}
+
+static void extract_3f_viewport( const struct tnl_clipspace_attr *a, GLfloat *out, 
+				 const GLubyte *v )
+{
+   const GLfloat *in = (const GLfloat *)v;
+   const GLfloat * const vp = a->vp;
+   DEBUG_INSERT;
+   out[0] = (in[0] - vp[12]) / vp[0];
+   out[1] = (in[1] - vp[13]) / vp[5];
+   out[2] = (in[2] - vp[14]) / vp[10];
+   out[3] = 1;
+}
+
+
+static void extract_2f_viewport( const struct tnl_clipspace_attr *a, GLfloat *out, 
+				 const GLubyte *v )
+{
+   const GLfloat *in = (const GLfloat *)v;
+   const GLfloat * const vp = a->vp;
+   DEBUG_INSERT;
+   out[0] = (in[0] - vp[12]) / vp[0];
+   out[1] = (in[1] - vp[13]) / vp[5];
+   out[2] = 0;
+   out[3] = 1;
+}
+
+
+static void extract_4f( const struct tnl_clipspace_attr *a, GLfloat *out, const GLubyte *v  )
+{
+   const GLfloat *in = (const GLfloat *)v;
+   (void) a;
+   
+   out[0] = in[0];
+   out[1] = in[1];
+   out[2] = in[2];
+   out[3] = in[3];
+}
+
+static void extract_3f_xyw( const struct tnl_clipspace_attr *a, GLfloat *out, const GLubyte *v )
+{
+   const GLfloat *in = (const GLfloat *)v;
+   (void) a;
+   
+   out[0] = in[0];
+   out[1] = in[1];
+   out[2] = 0;
+   out[3] = in[2];
+}
+
+
+static void extract_3f( const struct tnl_clipspace_attr *a, GLfloat *out, const GLubyte *v )
+{
+   const GLfloat *in = (const GLfloat *)v;
+   (void) a;
+   
+   out[0] = in[0];
+   out[1] = in[1];
+   out[2] = in[2];
+   out[3] = 1;
+}
+
+
+static void extract_2f( const struct tnl_clipspace_attr *a, GLfloat *out, const GLubyte *v )
+{
+   const GLfloat *in = (const GLfloat *)v;
+   (void) a;
+   
+   out[0] = in[0];
+   out[1] = in[1];
+   out[2] = 0;
+   out[3] = 1;
+}
+
+static void extract_1f( const struct tnl_clipspace_attr *a, GLfloat *out, const GLubyte *v )
+{
+   const GLfloat *in = (const GLfloat *)v;
+   (void) a;
+   
+   out[0] = in[0];
+   out[1] = 0;
+   out[2] = 0;
+   out[3] = 1;
+}
+
+static void extract_4chan_4f_rgba( const struct tnl_clipspace_attr *a, GLfloat *out, 
+				 const GLubyte *v )
+{
+   GLchan *c = (GLchan *)v;
+   (void) a;
+
+   out[0] = CHAN_TO_FLOAT(c[0]);
+   out[1] = CHAN_TO_FLOAT(c[1]);
+   out[2] = CHAN_TO_FLOAT(c[2]);
+   out[3] = CHAN_TO_FLOAT(c[3]);
+}
+
+static void extract_4ub_4f_rgba( const struct tnl_clipspace_attr *a, GLfloat *out, 
+				 const GLubyte *v )
+{
+   (void) a;
+   out[0] = UBYTE_TO_FLOAT(v[0]);
+   out[1] = UBYTE_TO_FLOAT(v[1]);
+   out[2] = UBYTE_TO_FLOAT(v[2]);
+   out[3] = UBYTE_TO_FLOAT(v[3]);
+}
+
+static void extract_4ub_4f_bgra( const struct tnl_clipspace_attr *a, GLfloat *out, 
+				 const GLubyte *v )
+{
+   (void) a;
+   out[2] = UBYTE_TO_FLOAT(v[0]);
+   out[1] = UBYTE_TO_FLOAT(v[1]);
+   out[0] = UBYTE_TO_FLOAT(v[2]);
+   out[3] = UBYTE_TO_FLOAT(v[3]);
+}
+
+static void extract_4ub_4f_argb( const struct tnl_clipspace_attr *a, GLfloat *out, 
+				 const GLubyte *v )
+{
+   (void) a;
+   out[3] = UBYTE_TO_FLOAT(v[0]);
+   out[0] = UBYTE_TO_FLOAT(v[1]);
+   out[1] = UBYTE_TO_FLOAT(v[2]);
+   out[2] = UBYTE_TO_FLOAT(v[3]);
+}
+
+static void extract_4ub_4f_abgr( const struct tnl_clipspace_attr *a, GLfloat *out, 
+				 const GLubyte *v )
+{
+   (void) a;
+   out[3] = UBYTE_TO_FLOAT(v[0]);
+   out[2] = UBYTE_TO_FLOAT(v[1]);
+   out[1] = UBYTE_TO_FLOAT(v[2]);
+   out[0] = UBYTE_TO_FLOAT(v[3]);
+}
+
+static void extract_3ub_3f_rgb( const struct tnl_clipspace_attr *a, GLfloat *out, 
+				const GLubyte *v )
+{
+   (void) a;
+   out[0] = UBYTE_TO_FLOAT(v[0]);
+   out[1] = UBYTE_TO_FLOAT(v[1]);
+   out[2] = UBYTE_TO_FLOAT(v[2]);
+   out[3] = 1;
+}
+
+static void extract_3ub_3f_bgr( const struct tnl_clipspace_attr *a, GLfloat *out, 
+				const GLubyte *v )
+{
+   (void) a;
+   out[2] = UBYTE_TO_FLOAT(v[0]);
+   out[1] = UBYTE_TO_FLOAT(v[1]);
+   out[0] = UBYTE_TO_FLOAT(v[2]);
+   out[3] = 1;
+}
+
+static void extract_1ub_1f( const struct tnl_clipspace_attr *a, GLfloat *out, const GLubyte *v )
+{
+   (void) a;
+   out[0] = UBYTE_TO_FLOAT(v[0]);
+   out[1] = 0;
+   out[2] = 0;
+   out[3] = 1;
+}
+
+
+const struct tnl_format_info _tnl_format_info[EMIT_MAX] = 
+{
+   { "1f",
+     extract_1f,
+     { insert_1f_1, insert_1f_1, insert_1f_1, insert_1f_1 },
+     sizeof(GLfloat) },
+
+   { "2f",
+     extract_2f,
+     { insert_2f_1, insert_2f_2, insert_2f_2, insert_2f_2 },
+     2 * sizeof(GLfloat) },
+
+   { "3f",
+     extract_3f,
+     { insert_3f_1, insert_3f_2, insert_3f_3, insert_3f_3 },
+     3 * sizeof(GLfloat) },
+
+   { "4f",
+     extract_4f,
+     { insert_4f_1, insert_4f_2, insert_4f_3, insert_4f_4 },
+     4 * sizeof(GLfloat) },
+
+   { "2f_viewport",
+     extract_2f_viewport,
+     { insert_2f_viewport_1, insert_2f_viewport_2, insert_2f_viewport_2,
+       insert_2f_viewport_2 },
+     2 * sizeof(GLfloat) },
+
+   { "3f_viewport",
+     extract_3f_viewport,
+     { insert_3f_viewport_1, insert_3f_viewport_2, insert_3f_viewport_3,
+       insert_3f_viewport_3 },
+     3 * sizeof(GLfloat) },
+
+   { "4f_viewport",
+     extract_4f_viewport,
+     { insert_4f_viewport_1, insert_4f_viewport_2, insert_4f_viewport_3,
+       insert_4f_viewport_4 }, 
+     4 * sizeof(GLfloat) },
+
+   { "3f_xyw",
+     extract_3f_xyw,
+     { insert_3f_xyw_err, insert_3f_xyw_err, insert_3f_xyw_err, 
+       insert_3f_xyw_4 },
+     3 * sizeof(GLfloat) },
+
+   { "1ub_1f",
+     extract_1ub_1f,
+     { insert_1ub_1f_1, insert_1ub_1f_1, insert_1ub_1f_1, insert_1ub_1f_1 },
+     sizeof(GLubyte) },
+
+   { "3ub_3f_rgb",
+     extract_3ub_3f_rgb,
+     { insert_3ub_3f_rgb_1, insert_3ub_3f_rgb_2, insert_3ub_3f_rgb_3,
+       insert_3ub_3f_rgb_3 },
+     3 * sizeof(GLubyte) },
+
+   { "3ub_3f_bgr",
+     extract_3ub_3f_bgr,
+     { insert_3ub_3f_bgr_1, insert_3ub_3f_bgr_2, insert_3ub_3f_bgr_3,
+       insert_3ub_3f_bgr_3 },
+     3 * sizeof(GLubyte) },
+
+   { "4ub_4f_rgba",
+     extract_4ub_4f_rgba,
+     { insert_4ub_4f_rgba_1, insert_4ub_4f_rgba_2, insert_4ub_4f_rgba_3, 
+       insert_4ub_4f_rgba_4 },
+     4 * sizeof(GLubyte) },
+
+   { "4ub_4f_bgra",
+     extract_4ub_4f_bgra,
+     { insert_4ub_4f_bgra_1, insert_4ub_4f_bgra_2, insert_4ub_4f_bgra_3,
+       insert_4ub_4f_bgra_4 },
+     4 * sizeof(GLubyte) },
+
+   { "4ub_4f_argb",
+     extract_4ub_4f_argb,
+     { insert_4ub_4f_argb_1, insert_4ub_4f_argb_2, insert_4ub_4f_argb_3,
+       insert_4ub_4f_argb_4 },
+     4 * sizeof(GLubyte) },
+
+   { "4ub_4f_abgr",
+     extract_4ub_4f_abgr,
+     { insert_4ub_4f_abgr_1, insert_4ub_4f_abgr_2, insert_4ub_4f_abgr_3,
+       insert_4ub_4f_abgr_4 },
+     4 * sizeof(GLubyte) },
+
+   { "4chan_4f_rgba",
+     extract_4chan_4f_rgba,
+     { insert_4chan_4f_rgba_1, insert_4chan_4f_rgba_2, insert_4chan_4f_rgba_3,
+       insert_4chan_4f_rgba_4 },
+     4 * sizeof(GLchan) },
+
+   { "pad",
+     NULL,
+     { NULL, NULL, NULL, NULL },
+     0 }
+
+};
+
+
+
+    
+/***********************************************************************
+ * Hardwired fastpaths for emitting whole vertices or groups of
+ * vertices
+ */
+#define EMIT5(NR, F0, F1, F2, F3, F4, NAME)				\
+static void NAME( struct gl_context *ctx,					\
+		  GLuint count,						\
+		  GLubyte *v )						\
+{									\
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);			\
+   struct tnl_clipspace_attr *a = vtx->attr;				\
+   GLuint i;								\
+									\
+   for (i = 0 ; i < count ; i++, v += vtx->vertex_size) {		\
+      if (NR > 0) {							\
+	 F0( &a[0], v + a[0].vertoffset, (GLfloat *)a[0].inputptr );	\
+	 a[0].inputptr += a[0].inputstride;				\
+      }									\
+      									\
+      if (NR > 1) {							\
+	 F1( &a[1], v + a[1].vertoffset, (GLfloat *)a[1].inputptr );	\
+	 a[1].inputptr += a[1].inputstride;				\
+      }									\
+      									\
+      if (NR > 2) {							\
+	 F2( &a[2], v + a[2].vertoffset, (GLfloat *)a[2].inputptr );	\
+	 a[2].inputptr += a[2].inputstride;				\
+      }									\
+      									\
+      if (NR > 3) {							\
+	 F3( &a[3], v + a[3].vertoffset, (GLfloat *)a[3].inputptr );	\
+	 a[3].inputptr += a[3].inputstride;				\
+      }									\
+									\
+      if (NR > 4) {							\
+	 F4( &a[4], v + a[4].vertoffset, (GLfloat *)a[4].inputptr );	\
+	 a[4].inputptr += a[4].inputstride;				\
+      }									\
+   }									\
+}
+
+   
+#define EMIT2(F0, F1, NAME) EMIT5(2, F0, F1, insert_null, \
+				  insert_null, insert_null, NAME)
+
+#define EMIT3(F0, F1, F2, NAME) EMIT5(3, F0, F1, F2, insert_null, \
+				      insert_null, NAME)
+   
+#define EMIT4(F0, F1, F2, F3, NAME) EMIT5(4, F0, F1, F2, F3, \
+				          insert_null, NAME)
+   
+
+EMIT2(insert_3f_viewport_3, insert_4ub_4f_rgba_4, emit_viewport3_rgba4)
+EMIT2(insert_3f_viewport_3, insert_4ub_4f_bgra_4, emit_viewport3_bgra4)
+EMIT2(insert_3f_3, insert_4ub_4f_rgba_4, emit_xyz3_rgba4)
+
+EMIT3(insert_4f_viewport_4, insert_4ub_4f_rgba_4, insert_2f_2, emit_viewport4_rgba4_st2)
+EMIT3(insert_4f_viewport_4, insert_4ub_4f_bgra_4, insert_2f_2,  emit_viewport4_bgra4_st2)
+EMIT3(insert_4f_4, insert_4ub_4f_rgba_4, insert_2f_2, emit_xyzw4_rgba4_st2)
+
+EMIT4(insert_4f_viewport_4, insert_4ub_4f_rgba_4, insert_2f_2, insert_2f_2, emit_viewport4_rgba4_st2_st2)
+EMIT4(insert_4f_viewport_4, insert_4ub_4f_bgra_4, insert_2f_2, insert_2f_2,  emit_viewport4_bgra4_st2_st2)
+EMIT4(insert_4f_4, insert_4ub_4f_rgba_4, insert_2f_2, insert_2f_2, emit_xyzw4_rgba4_st2_st2)
+
+
+/* Use the codegen paths to select one of a number of hardwired
+ * fastpaths.
+ */
+void _tnl_generate_hardwired_emit( struct gl_context *ctx )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+   tnl_emit_func func = NULL;
+
+   /* Does it fit a hardwired fastpath?  Help! this is growing out of
+    * control!
+    */
+   switch (vtx->attr_count) {
+   case 2:
+      if (vtx->attr[0].emit == insert_3f_viewport_3) {
+	 if (vtx->attr[1].emit == insert_4ub_4f_bgra_4) 
+	    func = emit_viewport3_bgra4;
+	 else if (vtx->attr[1].emit == insert_4ub_4f_rgba_4) 
+	    func = emit_viewport3_rgba4;
+      }
+      else if (vtx->attr[0].emit == insert_3f_3 &&
+	       vtx->attr[1].emit == insert_4ub_4f_rgba_4) {
+ 	 func = emit_xyz3_rgba4; 
+      }
+      break;
+   case 3:
+      if (vtx->attr[2].emit == insert_2f_2) {
+	 if (vtx->attr[1].emit == insert_4ub_4f_rgba_4) {
+	    if (vtx->attr[0].emit == insert_4f_viewport_4)
+	       func = emit_viewport4_rgba4_st2;
+	    else if (vtx->attr[0].emit == insert_4f_4) 
+	       func = emit_xyzw4_rgba4_st2;
+	 }
+	 else if (vtx->attr[1].emit == insert_4ub_4f_bgra_4 &&
+		  vtx->attr[0].emit == insert_4f_viewport_4)
+	    func = emit_viewport4_bgra4_st2;
+      }
+      break;
+   case 4:
+      if (vtx->attr[2].emit == insert_2f_2 &&
+	  vtx->attr[3].emit == insert_2f_2) {
+	 if (vtx->attr[1].emit == insert_4ub_4f_rgba_4) {
+	    if (vtx->attr[0].emit == insert_4f_viewport_4)
+	       func = emit_viewport4_rgba4_st2_st2;
+	    else if (vtx->attr[0].emit == insert_4f_4) 
+	       func = emit_xyzw4_rgba4_st2_st2;
+	 }
+	 else if (vtx->attr[1].emit == insert_4ub_4f_bgra_4 &&
+		  vtx->attr[0].emit == insert_4f_viewport_4)
+	    func = emit_viewport4_bgra4_st2_st2;
+      }
+      break;
+   }
+
+   vtx->emit = func;
+}
+
+/***********************************************************************
+ * Generic (non-codegen) functions for whole vertices or groups of
+ * vertices
+ */
+
+void _tnl_generic_emit( struct gl_context *ctx,
+			GLuint count,
+			GLubyte *v )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+   struct tnl_clipspace_attr *a = vtx->attr;
+   const GLuint attr_count = vtx->attr_count;
+   const GLuint stride = vtx->vertex_size;
+   GLuint i, j;
+
+   for (i = 0 ; i < count ; i++, v += stride) {
+      for (j = 0; j < attr_count; j++) {
+	 GLfloat *in = (GLfloat *)a[j].inputptr;
+	 a[j].inputptr += a[j].inputstride;
+	 a[j].emit( &a[j], v + a[j].vertoffset, in );
+      }
+   }
+}
+
+
+void _tnl_generic_interp( struct gl_context *ctx,
+			    GLfloat t,
+			    GLuint edst, GLuint eout, GLuint ein,
+			    GLboolean force_boundary )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   struct vertex_buffer *VB = &tnl->vb;
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+   const GLubyte *vin  = vtx->vertex_buf + ein  * vtx->vertex_size;
+   const GLubyte *vout = vtx->vertex_buf + eout * vtx->vertex_size;
+   GLubyte *vdst = vtx->vertex_buf + edst * vtx->vertex_size;
+   const struct tnl_clipspace_attr *a = vtx->attr;
+   const GLuint attr_count = vtx->attr_count;
+   GLuint j;
+   (void) force_boundary;
+
+   if (tnl->NeedNdcCoords) {
+      const GLfloat *dstclip = VB->ClipPtr->data[edst];
+      if (dstclip[3] != 0.0) {
+	 const GLfloat w = 1.0f / dstclip[3];
+	 GLfloat pos[4];
+
+	 pos[0] = dstclip[0] * w;
+	 pos[1] = dstclip[1] * w;
+	 pos[2] = dstclip[2] * w;
+	 pos[3] = w;
+
+	 a[0].insert[4-1]( &a[0], vdst, pos );
+      }
+   }
+   else {
+      a[0].insert[4-1]( &a[0], vdst, VB->ClipPtr->data[edst] );
+   }
+
+
+   for (j = 1; j < attr_count; j++) {
+      GLfloat fin[4], fout[4], fdst[4];
+	 
+      a[j].extract( &a[j], fin, vin + a[j].vertoffset );
+      a[j].extract( &a[j], fout, vout + a[j].vertoffset );
+
+      INTERP_F( t, fdst[3], fout[3], fin[3] );
+      INTERP_F( t, fdst[2], fout[2], fin[2] );
+      INTERP_F( t, fdst[1], fout[1], fin[1] );
+      INTERP_F( t, fdst[0], fout[0], fin[0] );
+
+      a[j].insert[4-1]( &a[j], vdst + a[j].vertoffset, fdst );
+   }
+}
+
+
+/* Extract color attributes from one vertex and insert them into
+ * another.  (Shortcircuit extract/insert with memcpy).
+ */
+void _tnl_generic_copy_pv( struct gl_context *ctx, GLuint edst, GLuint esrc )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+   GLubyte *vsrc = vtx->vertex_buf + esrc * vtx->vertex_size;
+   GLubyte *vdst = vtx->vertex_buf + edst * vtx->vertex_size;
+   const struct tnl_clipspace_attr *a = vtx->attr;
+   const GLuint attr_count = vtx->attr_count;
+   GLuint j;
+
+   for (j = 0; j < attr_count; j++) {
+      if (a[j].attrib == VERT_ATTRIB_COLOR0 ||
+	  a[j].attrib == VERT_ATTRIB_COLOR1) {
+
+	 memcpy( vdst + a[j].vertoffset,
+                 vsrc + a[j].vertoffset,
+                 a[j].vertattrsize );
+      }
+   }
+}
+
+
+/* Helper functions for hardware which doesn't put back colors and/or
+ * edgeflags into vertices.
+ */
+void _tnl_generic_interp_extras( struct gl_context *ctx,
+				   GLfloat t,
+				   GLuint dst, GLuint out, GLuint in,
+				   GLboolean force_boundary )
+{
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+
+   /* If stride is zero, BackfaceColorPtr is constant across the VB, so
+    * there is no point interpolating between two values as they will
+    * be identical.  In all other cases, this value is generated by
+    * t_vb_lighttmp.h and has a stride of 4 dwords.
+    */
+   if (VB->BackfaceColorPtr && VB->BackfaceColorPtr->stride) {
+      assert(VB->BackfaceColorPtr->stride == 4 * sizeof(GLfloat));
+
+      INTERP_4F( t,
+		 VB->BackfaceColorPtr->data[dst],
+		 VB->BackfaceColorPtr->data[out],
+		 VB->BackfaceColorPtr->data[in] );
+   }
+
+   if (VB->BackfaceSecondaryColorPtr) {
+      assert(VB->BackfaceSecondaryColorPtr->stride == 4 * sizeof(GLfloat));
+      
+      INTERP_3F( t,
+		 VB->BackfaceSecondaryColorPtr->data[dst],
+		 VB->BackfaceSecondaryColorPtr->data[out],
+		 VB->BackfaceSecondaryColorPtr->data[in] );
+   }
+   
+   if (VB->BackfaceIndexPtr) {
+      VB->BackfaceIndexPtr->data[dst][0] = LINTERP( t,
+					       VB->BackfaceIndexPtr->data[out][0],
+					       VB->BackfaceIndexPtr->data[in][0] );
+   }
+
+   if (VB->EdgeFlag) {
+      VB->EdgeFlag[dst] = VB->EdgeFlag[out] || force_boundary;
+   }
+
+   _tnl_generic_interp(ctx, t, dst, out, in, force_boundary);
+}
+
+void _tnl_generic_copy_pv_extras( struct gl_context *ctx, 
+				  GLuint dst, GLuint src )
+{
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+
+   /* See above comment:
+    */
+   if (VB->BackfaceColorPtr && VB->BackfaceColorPtr->stride) {
+      COPY_4FV( VB->BackfaceColorPtr->data[dst],
+		VB->BackfaceColorPtr->data[src] );
+   }
+
+   if (VB->BackfaceSecondaryColorPtr) {
+      COPY_4FV( VB->BackfaceSecondaryColorPtr->data[dst],
+		VB->BackfaceSecondaryColorPtr->data[src] );
+   }
+
+   if (VB->BackfaceIndexPtr) {
+      VB->BackfaceIndexPtr->data[dst][0] = VB->BackfaceIndexPtr->data[src][0];
+   }
+
+   _tnl_generic_copy_pv(ctx, dst, src);
+}
+
+
diff --git a/mesalib/src/mesa/tnl/t_vertex_sse.c b/mesalib/src/mesa/tnl/t_vertex_sse.c
index f164cd77e..e0141c36f 100644
--- a/mesalib/src/mesa/tnl/t_vertex_sse.c
+++ b/mesalib/src/mesa/tnl/t_vertex_sse.c
@@ -1,684 +1,685 @@
-/*

- * Copyright 2003 Tungsten Graphics, inc.

- * All Rights Reserved.

- *

- * Permission is hereby granted, free of charge, to any person obtaining a

- * copy of this software and associated documentation files (the "Software"),

- * to deal in the Software without restriction, including without limitation

- * on the rights to use, copy, modify, merge, publish, distribute, sub

- * license, and/or sell copies of the Software, and to permit persons to whom

- * the Software is furnished to do so, subject to the following conditions:

- *

- * The above copyright notice and this permission notice (including the next

- * paragraph) shall be included in all copies or substantial portions of the

- * Software.

- *

- * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR

- * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,

- * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL

- * TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,

- * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR

- * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE

- * USE OR OTHER DEALINGS IN THE SOFTWARE.

- *

- * Authors:

- *    Keith Whitwell <keithw@tungstengraphics.com>

- */

-

-#include "main/glheader.h"

-#include "main/context.h"

-#include "main/colormac.h"

-#include "main/simple_list.h"

-#include "main/enums.h"

-#include "t_context.h"

-#include "t_vertex.h"

-

-#if defined(USE_SSE_ASM)

-

-#include "x86/rtasm/x86sse.h"

-#include "x86/common_x86_asm.h"

-

-

-/**

- * Number of bytes to allocate for generated SSE functions

- */

-#define MAX_SSE_CODE_SIZE 1024

-

-

-#define X    0

-#define Y    1

-#define Z    2

-#define W    3

-

-

-struct x86_program {

-   struct x86_function func;

-

-   struct gl_context *ctx;

-   GLboolean inputs_safe;

-   GLboolean outputs_safe;

-   GLboolean have_sse2;

-   

-   struct x86_reg identity;

-   struct x86_reg chan0;

-};

-

-

-static struct x86_reg get_identity( struct x86_program *p )

-{

-   return p->identity;

-}

-

-static void emit_load4f_4( struct x86_program *p, 			   

-			   struct x86_reg dest,

-			   struct x86_reg arg0 )

-{

-   sse_movups(&p->func, dest, arg0);

-}

-

-static void emit_load4f_3( struct x86_program *p, 

-			   struct x86_reg dest,

-			   struct x86_reg arg0 )

-{

-   /* Have to jump through some hoops:

-    *

-    * c 0 0 0

-    * c 0 0 1

-    * 0 0 c 1

-    * a b c 1

-    */

-   sse_movss(&p->func, dest, x86_make_disp(arg0, 8));

-   sse_shufps(&p->func, dest, get_identity(p), SHUF(X,Y,Z,W) );

-   sse_shufps(&p->func, dest, dest, SHUF(Y,Z,X,W) );

-   sse_movlps(&p->func, dest, arg0);

-}

-

-static void emit_load4f_2( struct x86_program *p, 

-			   struct x86_reg dest,

-			   struct x86_reg arg0 )

-{

-   /* Initialize from identity, then pull in low two words:

-    */

-   sse_movups(&p->func, dest, get_identity(p));

-   sse_movlps(&p->func, dest, arg0);

-}

-

-static void emit_load4f_1( struct x86_program *p, 

-			   struct x86_reg dest,

-			   struct x86_reg arg0 )

-{

-   /* Pull in low word, then swizzle in identity */

-   sse_movss(&p->func, dest, arg0);

-   sse_shufps(&p->func, dest, get_identity(p), SHUF(X,Y,Z,W) );

-}

-

-

-

-static void emit_load3f_3( struct x86_program *p, 			   

-			   struct x86_reg dest,

-			   struct x86_reg arg0 )

-{

-   /* Over-reads by 1 dword - potential SEGV if input is a vertex

-    * array.

-    */

-   if (p->inputs_safe) {

-      sse_movups(&p->func, dest, arg0);

-   } 

-   else {

-      /* c 0 0 0

-       * c c c c

-       * a b c c 

-       */

-      sse_movss(&p->func, dest, x86_make_disp(arg0, 8));

-      sse_shufps(&p->func, dest, dest, SHUF(X,X,X,X));

-      sse_movlps(&p->func, dest, arg0);

-   }

-}

-

-static void emit_load3f_2( struct x86_program *p, 

-			   struct x86_reg dest,

-			   struct x86_reg arg0 )

-{

-   emit_load4f_2(p, dest, arg0);

-}

-

-static void emit_load3f_1( struct x86_program *p, 

-			   struct x86_reg dest,

-			   struct x86_reg arg0 )

-{

-   /* Loading from memory erases the upper bits. */

-   sse_movss(&p->func, dest, arg0);

-}

-

-static void emit_load2f_2( struct x86_program *p, 

-			   struct x86_reg dest,

-			   struct x86_reg arg0 )

-{

-   sse_movlps(&p->func, dest, arg0);

-}

-

-static void emit_load2f_1( struct x86_program *p, 

-			   struct x86_reg dest,

-			   struct x86_reg arg0 )

-{

-   /* Loading from memory erases the upper bits. */

-   sse_movss(&p->func, dest, arg0);

-}

-

-static void emit_load1f_1( struct x86_program *p, 

-			   struct x86_reg dest,

-			   struct x86_reg arg0 )

-{

-   sse_movss(&p->func, dest, arg0);

-}

-

-static void (*load[4][4])( struct x86_program *p, 

-			   struct x86_reg dest,

-			   struct x86_reg arg0 ) = {

-   { emit_load1f_1, 

-     emit_load1f_1, 

-     emit_load1f_1, 

-     emit_load1f_1 },

-

-   { emit_load2f_1, 

-     emit_load2f_2, 

-     emit_load2f_2, 

-     emit_load2f_2 },

-

-   { emit_load3f_1, 

-     emit_load3f_2, 

-     emit_load3f_3, 

-     emit_load3f_3 },

-

-   { emit_load4f_1, 

-     emit_load4f_2, 

-     emit_load4f_3, 

-     emit_load4f_4 } 

-};

-

-static void emit_load( struct x86_program *p,

-		       struct x86_reg dest,

-		       GLuint sz,

-		       struct x86_reg src,

-		       GLuint src_sz)

-{

-   load[sz-1][src_sz-1](p, dest, src);

-}

-

-static void emit_store4f( struct x86_program *p, 			   

-			  struct x86_reg dest,

-			  struct x86_reg arg0 )

-{

-   sse_movups(&p->func, dest, arg0);

-}

-

-static void emit_store3f( struct x86_program *p, 

-			  struct x86_reg dest,

-			  struct x86_reg arg0 )

-{

-   if (p->outputs_safe) {

-      /* Emit the extra dword anyway.  This may hurt writecombining,

-       * may cause other problems.

-       */

-      sse_movups(&p->func, dest, arg0);

-   }

-   else {

-      /* Alternate strategy - emit two, shuffle, emit one.

-       */

-      sse_movlps(&p->func, dest, arg0);

-      sse_shufps(&p->func, arg0, arg0, SHUF(Z,Z,Z,Z) ); /* NOTE! destructive */

-      sse_movss(&p->func, x86_make_disp(dest,8), arg0);

-   }

-}

-

-static void emit_store2f( struct x86_program *p, 

-			   struct x86_reg dest,

-			   struct x86_reg arg0 )

-{

-   sse_movlps(&p->func, dest, arg0);

-}

-

-static void emit_store1f( struct x86_program *p, 

-			  struct x86_reg dest,

-			  struct x86_reg arg0 )

-{

-   sse_movss(&p->func, dest, arg0);

-}

-

-

-static void (*store[4])( struct x86_program *p, 

-			 struct x86_reg dest,

-			 struct x86_reg arg0 ) = 

-{

-   emit_store1f, 

-   emit_store2f, 

-   emit_store3f, 

-   emit_store4f 

-};

-

-static void emit_store( struct x86_program *p,

-			struct x86_reg dest,

-			GLuint sz,

-			struct x86_reg temp )

-

-{

-   store[sz-1](p, dest, temp);

-}

-

-static void emit_pack_store_4ub( struct x86_program *p,

-				 struct x86_reg dest,

-				 struct x86_reg temp )

-{

-   /* Scale by 255.0

-    */

-   sse_mulps(&p->func, temp, p->chan0);

-

-   if (p->have_sse2) {

-      sse2_cvtps2dq(&p->func, temp, temp);

-      sse2_packssdw(&p->func, temp, temp);

-      sse2_packuswb(&p->func, temp, temp);

-      sse_movss(&p->func, dest, temp);

-   }

-   else {

-      struct x86_reg mmx0 = x86_make_reg(file_MMX, 0);

-      struct x86_reg mmx1 = x86_make_reg(file_MMX, 1);

-      sse_cvtps2pi(&p->func, mmx0, temp);

-      sse_movhlps(&p->func, temp, temp);

-      sse_cvtps2pi(&p->func, mmx1, temp);

-      mmx_packssdw(&p->func, mmx0, mmx1);

-      mmx_packuswb(&p->func, mmx0, mmx0);

-      mmx_movd(&p->func, dest, mmx0);

-   }

-}

-

-static GLint get_offset( const void *a, const void *b )

-{

-   return (const char *)b - (const char *)a;

-}

-

-/* Not much happens here.  Eventually use this function to try and

- * avoid saving/reloading the source pointers each vertex (if some of

- * them can fit in registers).

- */

-static void get_src_ptr( struct x86_program *p,

-			 struct x86_reg srcREG,

-			 struct x86_reg vtxREG,

-			 struct tnl_clipspace_attr *a )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(p->ctx);

-   struct x86_reg ptr_to_src = x86_make_disp(vtxREG, get_offset(vtx, &a->inputptr));

-

-   /* Load current a[j].inputptr

-    */

-   x86_mov(&p->func, srcREG, ptr_to_src);

-}

-

-static void update_src_ptr( struct x86_program *p,

-			 struct x86_reg srcREG,

-			 struct x86_reg vtxREG,

-			 struct tnl_clipspace_attr *a )

-{

-   if (a->inputstride) {

-      struct tnl_clipspace *vtx = GET_VERTEX_STATE(p->ctx);

-      struct x86_reg ptr_to_src = x86_make_disp(vtxREG, get_offset(vtx, &a->inputptr));

-

-      /* add a[j].inputstride (hardcoded value - could just as easily

-       * pull the stride value from memory each time).

-       */

-      x86_lea(&p->func, srcREG, x86_make_disp(srcREG, a->inputstride));

-      

-      /* save new value of a[j].inputptr 

-       */

-      x86_mov(&p->func, ptr_to_src, srcREG);

-   }

-}

-

-

-/* Lots of hardcoding

- *

- * EAX -- pointer to current output vertex

- * ECX -- pointer to current attribute 

- * 

- */

-static GLboolean build_vertex_emit( struct x86_program *p )

-{

-   struct gl_context *ctx = p->ctx;

-   TNLcontext *tnl = TNL_CONTEXT(ctx);

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-   GLuint j = 0;

-

-   struct x86_reg vertexEAX = x86_make_reg(file_REG32, reg_AX);

-   struct x86_reg srcECX = x86_make_reg(file_REG32, reg_CX);

-   struct x86_reg countEBP = x86_make_reg(file_REG32, reg_BP);

-   struct x86_reg vtxESI = x86_make_reg(file_REG32, reg_SI);

-   struct x86_reg temp = x86_make_reg(file_XMM, 0);

-   struct x86_reg vp0 = x86_make_reg(file_XMM, 1);

-   struct x86_reg vp1 = x86_make_reg(file_XMM, 2);

-   struct x86_reg temp2 = x86_make_reg(file_XMM, 3);

-   GLubyte *fixup, *label;

-

-   /* Push a few regs?

-    */

-   x86_push(&p->func, countEBP);

-   x86_push(&p->func, vtxESI);

-

-

-   /* Get vertex count, compare to zero

-    */

-   x86_xor(&p->func, srcECX, srcECX);

-   x86_mov(&p->func, countEBP, x86_fn_arg(&p->func, 2));

-   x86_cmp(&p->func, countEBP, srcECX);

-   fixup = x86_jcc_forward(&p->func, cc_E);

-

-   /* Initialize destination register. 

-    */

-   x86_mov(&p->func, vertexEAX, x86_fn_arg(&p->func, 3));

-

-   /* Dereference ctx to get tnl, then vtx:

-    */

-   x86_mov(&p->func, vtxESI, x86_fn_arg(&p->func, 1));

-   x86_mov(&p->func, vtxESI, x86_make_disp(vtxESI, get_offset(ctx, &ctx->swtnl_context)));

-   vtxESI = x86_make_disp(vtxESI, get_offset(tnl, &tnl->clipspace));

-

-   

-   /* Possibly load vp0, vp1 for viewport calcs:

-    */

-   if (vtx->need_viewport) {

-      sse_movups(&p->func, vp0, x86_make_disp(vtxESI, get_offset(vtx, &vtx->vp_scale[0])));

-      sse_movups(&p->func, vp1, x86_make_disp(vtxESI, get_offset(vtx, &vtx->vp_xlate[0])));

-   }

-

-   /* always load, needed or not:

-    */

-   sse_movups(&p->func, p->chan0, x86_make_disp(vtxESI, get_offset(vtx, &vtx->chan_scale[0])));

-   sse_movups(&p->func, p->identity, x86_make_disp(vtxESI, get_offset(vtx, &vtx->identity[0])));

-

-   /* Note address for loop jump */

-   label = x86_get_label(&p->func);

-

-   /* Emit code for each of the attributes.  Currently routes

-    * everything through SSE registers, even when it might be more

-    * efficient to stick with regular old x86.  No optimization or

-    * other tricks - enough new ground to cover here just getting

-    * things working.

-    */

-   while (j < vtx->attr_count) {

-      struct tnl_clipspace_attr *a = &vtx->attr[j];

-      struct x86_reg dest = x86_make_disp(vertexEAX, a->vertoffset);

-

-      /* Now, load an XMM reg from src, perhaps transform, then save.

-       * Could be shortcircuited in specific cases:

-       */

-      switch (a->format) {

-      case EMIT_1F:

-	 get_src_ptr(p, srcECX, vtxESI, a);

-	 emit_load(p, temp, 1, x86_deref(srcECX), a->inputsize);

-	 emit_store(p, dest, 1, temp);

-	 update_src_ptr(p, srcECX, vtxESI, a);

-	 break;

-      case EMIT_2F:

-	 get_src_ptr(p, srcECX, vtxESI, a);

-	 emit_load(p, temp, 2, x86_deref(srcECX), a->inputsize);

-	 emit_store(p, dest, 2, temp);

-	 update_src_ptr(p, srcECX, vtxESI, a);

-	 break;

-      case EMIT_3F:

-	 /* Potentially the worst case - hardcode 2+1 copying:

-	  */

-	 if (0) {

-	    get_src_ptr(p, srcECX, vtxESI, a);

-	    emit_load(p, temp, 3, x86_deref(srcECX), a->inputsize);

-	    emit_store(p, dest, 3, temp);

-	    update_src_ptr(p, srcECX, vtxESI, a);

-	 }

-	 else {

-	    get_src_ptr(p, srcECX, vtxESI, a);

-	    emit_load(p, temp, 2, x86_deref(srcECX), a->inputsize);

-	    emit_store(p, dest, 2, temp);

-	    if (a->inputsize > 2) {

-	       emit_load(p, temp, 1, x86_make_disp(srcECX, 8), 1);

-	       emit_store(p, x86_make_disp(dest,8), 1, temp);

-	    }

-	    else {

-	       sse_movss(&p->func, x86_make_disp(dest,8), get_identity(p));

-	    }

-	    update_src_ptr(p, srcECX, vtxESI, a);

-	 }

-	 break;

-      case EMIT_4F:

-	 get_src_ptr(p, srcECX, vtxESI, a);

-	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);

-	 emit_store(p, dest, 4, temp);

-	 update_src_ptr(p, srcECX, vtxESI, a);

-	 break;

-      case EMIT_2F_VIEWPORT: 

-	 get_src_ptr(p, srcECX, vtxESI, a);

-	 emit_load(p, temp, 2, x86_deref(srcECX), a->inputsize);

-	 sse_mulps(&p->func, temp, vp0);

-	 sse_addps(&p->func, temp, vp1);

-	 emit_store(p, dest, 2, temp);

-	 update_src_ptr(p, srcECX, vtxESI, a);

-	 break;

-      case EMIT_3F_VIEWPORT: 

-	 get_src_ptr(p, srcECX, vtxESI, a);

-	 emit_load(p, temp, 3, x86_deref(srcECX), a->inputsize);

-	 sse_mulps(&p->func, temp, vp0);

-	 sse_addps(&p->func, temp, vp1);

-	 emit_store(p, dest, 3, temp);

-	 update_src_ptr(p, srcECX, vtxESI, a);

-	 break;

-      case EMIT_4F_VIEWPORT: 

-	 get_src_ptr(p, srcECX, vtxESI, a);

-	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);

-	 sse_mulps(&p->func, temp, vp0);

-	 sse_addps(&p->func, temp, vp1);

-	 emit_store(p, dest, 4, temp);

-	 update_src_ptr(p, srcECX, vtxESI, a);

-	 break;

-      case EMIT_3F_XYW:

-	 get_src_ptr(p, srcECX, vtxESI, a);

-	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);

-	 sse_shufps(&p->func, temp, temp, SHUF(X,Y,W,Z));

-	 emit_store(p, dest, 3, temp);

-	 update_src_ptr(p, srcECX, vtxESI, a);

-	 break;

-

-      case EMIT_1UB_1F:	 

-	 /* Test for PAD3 + 1UB:

-	  */

-	 if (j > 0 &&

-	     a[-1].vertoffset + a[-1].vertattrsize <= a->vertoffset - 3)

-	 {

-	    get_src_ptr(p, srcECX, vtxESI, a);

-	    emit_load(p, temp, 1, x86_deref(srcECX), a->inputsize);

-	    sse_shufps(&p->func, temp, temp, SHUF(X,X,X,X));

-	    emit_pack_store_4ub(p, x86_make_disp(dest, -3), temp); /* overkill! */

-	    update_src_ptr(p, srcECX, vtxESI, a);

-	 }

-	 else {

-	    printf("Can't emit 1ub %x %x %d\n", a->vertoffset, a[-1].vertoffset, a[-1].vertattrsize );

-	    return GL_FALSE;

-	 }

-	 break;

-      case EMIT_3UB_3F_RGB:

-      case EMIT_3UB_3F_BGR:

-	 /* Test for 3UB + PAD1:

-	  */

-	 if (j == vtx->attr_count - 1 ||

-	     a[1].vertoffset >= a->vertoffset + 4) {

-	    get_src_ptr(p, srcECX, vtxESI, a);

-	    emit_load(p, temp, 3, x86_deref(srcECX), a->inputsize);

-	    if (a->format == EMIT_3UB_3F_BGR)

-	       sse_shufps(&p->func, temp, temp, SHUF(Z,Y,X,W));

-	    emit_pack_store_4ub(p, dest, temp);

-	    update_src_ptr(p, srcECX, vtxESI, a);

-	 }

-	 /* Test for 3UB + 1UB:

-	  */

-	 else if (j < vtx->attr_count - 1 &&

-		  a[1].format == EMIT_1UB_1F &&

-		  a[1].vertoffset == a->vertoffset + 3) {

-	    get_src_ptr(p, srcECX, vtxESI, a);

-	    emit_load(p, temp, 3, x86_deref(srcECX), a->inputsize);

-	    update_src_ptr(p, srcECX, vtxESI, a);

-

-	    /* Make room for incoming value:

-	     */

-	    sse_shufps(&p->func, temp, temp, SHUF(W,X,Y,Z));

-

-	    get_src_ptr(p, srcECX, vtxESI, &a[1]);

-	    emit_load(p, temp2, 1, x86_deref(srcECX), a[1].inputsize);

-	    sse_movss(&p->func, temp, temp2);

-	    update_src_ptr(p, srcECX, vtxESI, &a[1]);

-

-	    /* Rearrange and possibly do BGR conversion:

-	     */

-	    if (a->format == EMIT_3UB_3F_BGR)

-	       sse_shufps(&p->func, temp, temp, SHUF(W,Z,Y,X));

-	    else

-	       sse_shufps(&p->func, temp, temp, SHUF(Y,Z,W,X));

-

-	    emit_pack_store_4ub(p, dest, temp);

-	    j++;		/* NOTE: two attrs consumed */

-	 }

-	 else {

-	    printf("Can't emit 3ub\n");

-	    return GL_FALSE;	/* add this later */

-	 }

-	 break;

-

-      case EMIT_4UB_4F_RGBA:

-	 get_src_ptr(p, srcECX, vtxESI, a);

-	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);

-	 emit_pack_store_4ub(p, dest, temp);

-	 update_src_ptr(p, srcECX, vtxESI, a);

-	 break;

-      case EMIT_4UB_4F_BGRA:

-	 get_src_ptr(p, srcECX, vtxESI, a);

-	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);

-	 sse_shufps(&p->func, temp, temp, SHUF(Z,Y,X,W));

-	 emit_pack_store_4ub(p, dest, temp);

-	 update_src_ptr(p, srcECX, vtxESI, a);

-	 break;

-      case EMIT_4UB_4F_ARGB:

-	 get_src_ptr(p, srcECX, vtxESI, a);

-	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);

-	 sse_shufps(&p->func, temp, temp, SHUF(W,X,Y,Z));

-	 emit_pack_store_4ub(p, dest, temp);

-	 update_src_ptr(p, srcECX, vtxESI, a);

-	 break;

-      case EMIT_4UB_4F_ABGR:

-	 get_src_ptr(p, srcECX, vtxESI, a);

-	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);

-	 sse_shufps(&p->func, temp, temp, SHUF(W,Z,Y,X));

-	 emit_pack_store_4ub(p, dest, temp);

-	 update_src_ptr(p, srcECX, vtxESI, a);

-	 break;

-      case EMIT_4CHAN_4F_RGBA:

-	 switch (CHAN_TYPE) {

-	 case GL_UNSIGNED_BYTE:

-	    get_src_ptr(p, srcECX, vtxESI, a);

-	    emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);

-	    emit_pack_store_4ub(p, dest, temp);

-	    update_src_ptr(p, srcECX, vtxESI, a);

-	    break;

-	 case GL_FLOAT:

-	    get_src_ptr(p, srcECX, vtxESI, a);

-	    emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);

-	    emit_store(p, dest, 4, temp);

-	    update_src_ptr(p, srcECX, vtxESI, a);

-	    break;

-	 case GL_UNSIGNED_SHORT:

-	 default:

-	    printf("unknown CHAN_TYPE %s\n", _mesa_lookup_enum_by_nr(CHAN_TYPE));

-	    return GL_FALSE;

-	 }

-	 break;

-      default:

-	 printf("unknown a[%d].format %d\n", j, a->format);

-	 return GL_FALSE;	/* catch any new opcodes */

-      }

-      

-      /* Increment j by at least 1 - may have been incremented above also:

-       */

-      j++;

-   }

-

-   /* Next vertex:

-    */

-   x86_lea(&p->func, vertexEAX, x86_make_disp(vertexEAX, vtx->vertex_size));

-

-   /* decr count, loop if not zero

-    */

-   x86_dec(&p->func, countEBP);

-   x86_test(&p->func, countEBP, countEBP); 

-   x86_jcc(&p->func, cc_NZ, label);

-

-   /* Exit mmx state?

-    */

-   if (p->func.need_emms)

-      mmx_emms(&p->func);

-

-   /* Land forward jump here:

-    */

-   x86_fixup_fwd_jump(&p->func, fixup);

-

-   /* Pop regs and return

-    */

-   x86_pop(&p->func, x86_get_base_reg(vtxESI));

-   x86_pop(&p->func, countEBP);

-   x86_ret(&p->func);

-

-   assert(!vtx->emit);

-   vtx->emit = (tnl_emit_func)x86_get_func(&p->func);

-

-   assert( (char *) p->func.csr - (char *) p->func.store <= MAX_SSE_CODE_SIZE );

-   return GL_TRUE;

-}

-

-

-

-void _tnl_generate_sse_emit( struct gl_context *ctx )

-{

-   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);

-   struct x86_program p;   

-

-   if (!cpu_has_xmm) {

-      vtx->codegen_emit = NULL;

-      return;

-   }

-

-   memset(&p, 0, sizeof(p));

-

-   p.ctx = ctx;

-   p.inputs_safe = 0;		/* for now */

-   p.outputs_safe = 0;		/* for now */

-   p.have_sse2 = cpu_has_xmm2;

-   p.identity = x86_make_reg(file_XMM, 6);

-   p.chan0 = x86_make_reg(file_XMM, 7);

-

-   if (!x86_init_func_size(&p.func, MAX_SSE_CODE_SIZE)) {

-      vtx->emit = NULL;

-      return;

-   }

-

-   if (build_vertex_emit(&p)) {

-      _tnl_register_fastpath( vtx, GL_TRUE );

-   }

-   else {

-      /* Note the failure so that we don't keep trying to codegen an

-       * impossible state:

-       */

-      _tnl_register_fastpath( vtx, GL_FALSE );

-      x86_release_func(&p.func);

-   }

-}

-

-#else

-

-void _tnl_generate_sse_emit( struct gl_context *ctx )

-{

-   /* Dummy version for when USE_SSE_ASM not defined */

-}

-

-#endif

+/*
+ * Copyright 2003 Tungsten Graphics, inc.
+ * All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * on the rights to use, copy, modify, merge, publish, distribute, sub
+ * license, and/or sell copies of the Software, and to permit persons to whom
+ * the Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice (including the next
+ * paragraph) shall be included in all copies or substantial portions of the
+ * Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+ * IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.  IN NO EVENT SHALL
+ * TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+ * DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+ * OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+ * USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <keithw@tungstengraphics.com>
+ */
+
+#include "main/glheader.h"
+#include "main/context.h"
+#include "main/colormac.h"
+#include "main/simple_list.h"
+#include "main/enums.h"
+#include "swrast/s_chan.h"
+#include "t_context.h"
+#include "t_vertex.h"
+
+#if defined(USE_SSE_ASM)
+
+#include "x86/rtasm/x86sse.h"
+#include "x86/common_x86_asm.h"
+
+
+/**
+ * Number of bytes to allocate for generated SSE functions
+ */
+#define MAX_SSE_CODE_SIZE 1024
+
+
+#define X    0
+#define Y    1
+#define Z    2
+#define W    3
+
+
+struct x86_program {
+   struct x86_function func;
+
+   struct gl_context *ctx;
+   GLboolean inputs_safe;
+   GLboolean outputs_safe;
+   GLboolean have_sse2;
+   
+   struct x86_reg identity;
+   struct x86_reg chan0;
+};
+
+
+static struct x86_reg get_identity( struct x86_program *p )
+{
+   return p->identity;
+}
+
+static void emit_load4f_4( struct x86_program *p, 			   
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   sse_movups(&p->func, dest, arg0);
+}
+
+static void emit_load4f_3( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   /* Have to jump through some hoops:
+    *
+    * c 0 0 0
+    * c 0 0 1
+    * 0 0 c 1
+    * a b c 1
+    */
+   sse_movss(&p->func, dest, x86_make_disp(arg0, 8));
+   sse_shufps(&p->func, dest, get_identity(p), SHUF(X,Y,Z,W) );
+   sse_shufps(&p->func, dest, dest, SHUF(Y,Z,X,W) );
+   sse_movlps(&p->func, dest, arg0);
+}
+
+static void emit_load4f_2( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   /* Initialize from identity, then pull in low two words:
+    */
+   sse_movups(&p->func, dest, get_identity(p));
+   sse_movlps(&p->func, dest, arg0);
+}
+
+static void emit_load4f_1( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   /* Pull in low word, then swizzle in identity */
+   sse_movss(&p->func, dest, arg0);
+   sse_shufps(&p->func, dest, get_identity(p), SHUF(X,Y,Z,W) );
+}
+
+
+
+static void emit_load3f_3( struct x86_program *p, 			   
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   /* Over-reads by 1 dword - potential SEGV if input is a vertex
+    * array.
+    */
+   if (p->inputs_safe) {
+      sse_movups(&p->func, dest, arg0);
+   } 
+   else {
+      /* c 0 0 0
+       * c c c c
+       * a b c c 
+       */
+      sse_movss(&p->func, dest, x86_make_disp(arg0, 8));
+      sse_shufps(&p->func, dest, dest, SHUF(X,X,X,X));
+      sse_movlps(&p->func, dest, arg0);
+   }
+}
+
+static void emit_load3f_2( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   emit_load4f_2(p, dest, arg0);
+}
+
+static void emit_load3f_1( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   /* Loading from memory erases the upper bits. */
+   sse_movss(&p->func, dest, arg0);
+}
+
+static void emit_load2f_2( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   sse_movlps(&p->func, dest, arg0);
+}
+
+static void emit_load2f_1( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   /* Loading from memory erases the upper bits. */
+   sse_movss(&p->func, dest, arg0);
+}
+
+static void emit_load1f_1( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   sse_movss(&p->func, dest, arg0);
+}
+
+static void (*load[4][4])( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 ) = {
+   { emit_load1f_1, 
+     emit_load1f_1, 
+     emit_load1f_1, 
+     emit_load1f_1 },
+
+   { emit_load2f_1, 
+     emit_load2f_2, 
+     emit_load2f_2, 
+     emit_load2f_2 },
+
+   { emit_load3f_1, 
+     emit_load3f_2, 
+     emit_load3f_3, 
+     emit_load3f_3 },
+
+   { emit_load4f_1, 
+     emit_load4f_2, 
+     emit_load4f_3, 
+     emit_load4f_4 } 
+};
+
+static void emit_load( struct x86_program *p,
+		       struct x86_reg dest,
+		       GLuint sz,
+		       struct x86_reg src,
+		       GLuint src_sz)
+{
+   load[sz-1][src_sz-1](p, dest, src);
+}
+
+static void emit_store4f( struct x86_program *p, 			   
+			  struct x86_reg dest,
+			  struct x86_reg arg0 )
+{
+   sse_movups(&p->func, dest, arg0);
+}
+
+static void emit_store3f( struct x86_program *p, 
+			  struct x86_reg dest,
+			  struct x86_reg arg0 )
+{
+   if (p->outputs_safe) {
+      /* Emit the extra dword anyway.  This may hurt writecombining,
+       * may cause other problems.
+       */
+      sse_movups(&p->func, dest, arg0);
+   }
+   else {
+      /* Alternate strategy - emit two, shuffle, emit one.
+       */
+      sse_movlps(&p->func, dest, arg0);
+      sse_shufps(&p->func, arg0, arg0, SHUF(Z,Z,Z,Z) ); /* NOTE! destructive */
+      sse_movss(&p->func, x86_make_disp(dest,8), arg0);
+   }
+}
+
+static void emit_store2f( struct x86_program *p, 
+			   struct x86_reg dest,
+			   struct x86_reg arg0 )
+{
+   sse_movlps(&p->func, dest, arg0);
+}
+
+static void emit_store1f( struct x86_program *p, 
+			  struct x86_reg dest,
+			  struct x86_reg arg0 )
+{
+   sse_movss(&p->func, dest, arg0);
+}
+
+
+static void (*store[4])( struct x86_program *p, 
+			 struct x86_reg dest,
+			 struct x86_reg arg0 ) = 
+{
+   emit_store1f, 
+   emit_store2f, 
+   emit_store3f, 
+   emit_store4f 
+};
+
+static void emit_store( struct x86_program *p,
+			struct x86_reg dest,
+			GLuint sz,
+			struct x86_reg temp )
+
+{
+   store[sz-1](p, dest, temp);
+}
+
+static void emit_pack_store_4ub( struct x86_program *p,
+				 struct x86_reg dest,
+				 struct x86_reg temp )
+{
+   /* Scale by 255.0
+    */
+   sse_mulps(&p->func, temp, p->chan0);
+
+   if (p->have_sse2) {
+      sse2_cvtps2dq(&p->func, temp, temp);
+      sse2_packssdw(&p->func, temp, temp);
+      sse2_packuswb(&p->func, temp, temp);
+      sse_movss(&p->func, dest, temp);
+   }
+   else {
+      struct x86_reg mmx0 = x86_make_reg(file_MMX, 0);
+      struct x86_reg mmx1 = x86_make_reg(file_MMX, 1);
+      sse_cvtps2pi(&p->func, mmx0, temp);
+      sse_movhlps(&p->func, temp, temp);
+      sse_cvtps2pi(&p->func, mmx1, temp);
+      mmx_packssdw(&p->func, mmx0, mmx1);
+      mmx_packuswb(&p->func, mmx0, mmx0);
+      mmx_movd(&p->func, dest, mmx0);
+   }
+}
+
+static GLint get_offset( const void *a, const void *b )
+{
+   return (const char *)b - (const char *)a;
+}
+
+/* Not much happens here.  Eventually use this function to try and
+ * avoid saving/reloading the source pointers each vertex (if some of
+ * them can fit in registers).
+ */
+static void get_src_ptr( struct x86_program *p,
+			 struct x86_reg srcREG,
+			 struct x86_reg vtxREG,
+			 struct tnl_clipspace_attr *a )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(p->ctx);
+   struct x86_reg ptr_to_src = x86_make_disp(vtxREG, get_offset(vtx, &a->inputptr));
+
+   /* Load current a[j].inputptr
+    */
+   x86_mov(&p->func, srcREG, ptr_to_src);
+}
+
+static void update_src_ptr( struct x86_program *p,
+			 struct x86_reg srcREG,
+			 struct x86_reg vtxREG,
+			 struct tnl_clipspace_attr *a )
+{
+   if (a->inputstride) {
+      struct tnl_clipspace *vtx = GET_VERTEX_STATE(p->ctx);
+      struct x86_reg ptr_to_src = x86_make_disp(vtxREG, get_offset(vtx, &a->inputptr));
+
+      /* add a[j].inputstride (hardcoded value - could just as easily
+       * pull the stride value from memory each time).
+       */
+      x86_lea(&p->func, srcREG, x86_make_disp(srcREG, a->inputstride));
+      
+      /* save new value of a[j].inputptr 
+       */
+      x86_mov(&p->func, ptr_to_src, srcREG);
+   }
+}
+
+
+/* Lots of hardcoding
+ *
+ * EAX -- pointer to current output vertex
+ * ECX -- pointer to current attribute 
+ * 
+ */
+static GLboolean build_vertex_emit( struct x86_program *p )
+{
+   struct gl_context *ctx = p->ctx;
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+   GLuint j = 0;
+
+   struct x86_reg vertexEAX = x86_make_reg(file_REG32, reg_AX);
+   struct x86_reg srcECX = x86_make_reg(file_REG32, reg_CX);
+   struct x86_reg countEBP = x86_make_reg(file_REG32, reg_BP);
+   struct x86_reg vtxESI = x86_make_reg(file_REG32, reg_SI);
+   struct x86_reg temp = x86_make_reg(file_XMM, 0);
+   struct x86_reg vp0 = x86_make_reg(file_XMM, 1);
+   struct x86_reg vp1 = x86_make_reg(file_XMM, 2);
+   struct x86_reg temp2 = x86_make_reg(file_XMM, 3);
+   GLubyte *fixup, *label;
+
+   /* Push a few regs?
+    */
+   x86_push(&p->func, countEBP);
+   x86_push(&p->func, vtxESI);
+
+
+   /* Get vertex count, compare to zero
+    */
+   x86_xor(&p->func, srcECX, srcECX);
+   x86_mov(&p->func, countEBP, x86_fn_arg(&p->func, 2));
+   x86_cmp(&p->func, countEBP, srcECX);
+   fixup = x86_jcc_forward(&p->func, cc_E);
+
+   /* Initialize destination register. 
+    */
+   x86_mov(&p->func, vertexEAX, x86_fn_arg(&p->func, 3));
+
+   /* Dereference ctx to get tnl, then vtx:
+    */
+   x86_mov(&p->func, vtxESI, x86_fn_arg(&p->func, 1));
+   x86_mov(&p->func, vtxESI, x86_make_disp(vtxESI, get_offset(ctx, &ctx->swtnl_context)));
+   vtxESI = x86_make_disp(vtxESI, get_offset(tnl, &tnl->clipspace));
+
+   
+   /* Possibly load vp0, vp1 for viewport calcs:
+    */
+   if (vtx->need_viewport) {
+      sse_movups(&p->func, vp0, x86_make_disp(vtxESI, get_offset(vtx, &vtx->vp_scale[0])));
+      sse_movups(&p->func, vp1, x86_make_disp(vtxESI, get_offset(vtx, &vtx->vp_xlate[0])));
+   }
+
+   /* always load, needed or not:
+    */
+   sse_movups(&p->func, p->chan0, x86_make_disp(vtxESI, get_offset(vtx, &vtx->chan_scale[0])));
+   sse_movups(&p->func, p->identity, x86_make_disp(vtxESI, get_offset(vtx, &vtx->identity[0])));
+
+   /* Note address for loop jump */
+   label = x86_get_label(&p->func);
+
+   /* Emit code for each of the attributes.  Currently routes
+    * everything through SSE registers, even when it might be more
+    * efficient to stick with regular old x86.  No optimization or
+    * other tricks - enough new ground to cover here just getting
+    * things working.
+    */
+   while (j < vtx->attr_count) {
+      struct tnl_clipspace_attr *a = &vtx->attr[j];
+      struct x86_reg dest = x86_make_disp(vertexEAX, a->vertoffset);
+
+      /* Now, load an XMM reg from src, perhaps transform, then save.
+       * Could be shortcircuited in specific cases:
+       */
+      switch (a->format) {
+      case EMIT_1F:
+	 get_src_ptr(p, srcECX, vtxESI, a);
+	 emit_load(p, temp, 1, x86_deref(srcECX), a->inputsize);
+	 emit_store(p, dest, 1, temp);
+	 update_src_ptr(p, srcECX, vtxESI, a);
+	 break;
+      case EMIT_2F:
+	 get_src_ptr(p, srcECX, vtxESI, a);
+	 emit_load(p, temp, 2, x86_deref(srcECX), a->inputsize);
+	 emit_store(p, dest, 2, temp);
+	 update_src_ptr(p, srcECX, vtxESI, a);
+	 break;
+      case EMIT_3F:
+	 /* Potentially the worst case - hardcode 2+1 copying:
+	  */
+	 if (0) {
+	    get_src_ptr(p, srcECX, vtxESI, a);
+	    emit_load(p, temp, 3, x86_deref(srcECX), a->inputsize);
+	    emit_store(p, dest, 3, temp);
+	    update_src_ptr(p, srcECX, vtxESI, a);
+	 }
+	 else {
+	    get_src_ptr(p, srcECX, vtxESI, a);
+	    emit_load(p, temp, 2, x86_deref(srcECX), a->inputsize);
+	    emit_store(p, dest, 2, temp);
+	    if (a->inputsize > 2) {
+	       emit_load(p, temp, 1, x86_make_disp(srcECX, 8), 1);
+	       emit_store(p, x86_make_disp(dest,8), 1, temp);
+	    }
+	    else {
+	       sse_movss(&p->func, x86_make_disp(dest,8), get_identity(p));
+	    }
+	    update_src_ptr(p, srcECX, vtxESI, a);
+	 }
+	 break;
+      case EMIT_4F:
+	 get_src_ptr(p, srcECX, vtxESI, a);
+	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);
+	 emit_store(p, dest, 4, temp);
+	 update_src_ptr(p, srcECX, vtxESI, a);
+	 break;
+      case EMIT_2F_VIEWPORT: 
+	 get_src_ptr(p, srcECX, vtxESI, a);
+	 emit_load(p, temp, 2, x86_deref(srcECX), a->inputsize);
+	 sse_mulps(&p->func, temp, vp0);
+	 sse_addps(&p->func, temp, vp1);
+	 emit_store(p, dest, 2, temp);
+	 update_src_ptr(p, srcECX, vtxESI, a);
+	 break;
+      case EMIT_3F_VIEWPORT: 
+	 get_src_ptr(p, srcECX, vtxESI, a);
+	 emit_load(p, temp, 3, x86_deref(srcECX), a->inputsize);
+	 sse_mulps(&p->func, temp, vp0);
+	 sse_addps(&p->func, temp, vp1);
+	 emit_store(p, dest, 3, temp);
+	 update_src_ptr(p, srcECX, vtxESI, a);
+	 break;
+      case EMIT_4F_VIEWPORT: 
+	 get_src_ptr(p, srcECX, vtxESI, a);
+	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);
+	 sse_mulps(&p->func, temp, vp0);
+	 sse_addps(&p->func, temp, vp1);
+	 emit_store(p, dest, 4, temp);
+	 update_src_ptr(p, srcECX, vtxESI, a);
+	 break;
+      case EMIT_3F_XYW:
+	 get_src_ptr(p, srcECX, vtxESI, a);
+	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);
+	 sse_shufps(&p->func, temp, temp, SHUF(X,Y,W,Z));
+	 emit_store(p, dest, 3, temp);
+	 update_src_ptr(p, srcECX, vtxESI, a);
+	 break;
+
+      case EMIT_1UB_1F:	 
+	 /* Test for PAD3 + 1UB:
+	  */
+	 if (j > 0 &&
+	     a[-1].vertoffset + a[-1].vertattrsize <= a->vertoffset - 3)
+	 {
+	    get_src_ptr(p, srcECX, vtxESI, a);
+	    emit_load(p, temp, 1, x86_deref(srcECX), a->inputsize);
+	    sse_shufps(&p->func, temp, temp, SHUF(X,X,X,X));
+	    emit_pack_store_4ub(p, x86_make_disp(dest, -3), temp); /* overkill! */
+	    update_src_ptr(p, srcECX, vtxESI, a);
+	 }
+	 else {
+	    printf("Can't emit 1ub %x %x %d\n", a->vertoffset, a[-1].vertoffset, a[-1].vertattrsize );
+	    return GL_FALSE;
+	 }
+	 break;
+      case EMIT_3UB_3F_RGB:
+      case EMIT_3UB_3F_BGR:
+	 /* Test for 3UB + PAD1:
+	  */
+	 if (j == vtx->attr_count - 1 ||
+	     a[1].vertoffset >= a->vertoffset + 4) {
+	    get_src_ptr(p, srcECX, vtxESI, a);
+	    emit_load(p, temp, 3, x86_deref(srcECX), a->inputsize);
+	    if (a->format == EMIT_3UB_3F_BGR)
+	       sse_shufps(&p->func, temp, temp, SHUF(Z,Y,X,W));
+	    emit_pack_store_4ub(p, dest, temp);
+	    update_src_ptr(p, srcECX, vtxESI, a);
+	 }
+	 /* Test for 3UB + 1UB:
+	  */
+	 else if (j < vtx->attr_count - 1 &&
+		  a[1].format == EMIT_1UB_1F &&
+		  a[1].vertoffset == a->vertoffset + 3) {
+	    get_src_ptr(p, srcECX, vtxESI, a);
+	    emit_load(p, temp, 3, x86_deref(srcECX), a->inputsize);
+	    update_src_ptr(p, srcECX, vtxESI, a);
+
+	    /* Make room for incoming value:
+	     */
+	    sse_shufps(&p->func, temp, temp, SHUF(W,X,Y,Z));
+
+	    get_src_ptr(p, srcECX, vtxESI, &a[1]);
+	    emit_load(p, temp2, 1, x86_deref(srcECX), a[1].inputsize);
+	    sse_movss(&p->func, temp, temp2);
+	    update_src_ptr(p, srcECX, vtxESI, &a[1]);
+
+	    /* Rearrange and possibly do BGR conversion:
+	     */
+	    if (a->format == EMIT_3UB_3F_BGR)
+	       sse_shufps(&p->func, temp, temp, SHUF(W,Z,Y,X));
+	    else
+	       sse_shufps(&p->func, temp, temp, SHUF(Y,Z,W,X));
+
+	    emit_pack_store_4ub(p, dest, temp);
+	    j++;		/* NOTE: two attrs consumed */
+	 }
+	 else {
+	    printf("Can't emit 3ub\n");
+	    return GL_FALSE;	/* add this later */
+	 }
+	 break;
+
+      case EMIT_4UB_4F_RGBA:
+	 get_src_ptr(p, srcECX, vtxESI, a);
+	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);
+	 emit_pack_store_4ub(p, dest, temp);
+	 update_src_ptr(p, srcECX, vtxESI, a);
+	 break;
+      case EMIT_4UB_4F_BGRA:
+	 get_src_ptr(p, srcECX, vtxESI, a);
+	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);
+	 sse_shufps(&p->func, temp, temp, SHUF(Z,Y,X,W));
+	 emit_pack_store_4ub(p, dest, temp);
+	 update_src_ptr(p, srcECX, vtxESI, a);
+	 break;
+      case EMIT_4UB_4F_ARGB:
+	 get_src_ptr(p, srcECX, vtxESI, a);
+	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);
+	 sse_shufps(&p->func, temp, temp, SHUF(W,X,Y,Z));
+	 emit_pack_store_4ub(p, dest, temp);
+	 update_src_ptr(p, srcECX, vtxESI, a);
+	 break;
+      case EMIT_4UB_4F_ABGR:
+	 get_src_ptr(p, srcECX, vtxESI, a);
+	 emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);
+	 sse_shufps(&p->func, temp, temp, SHUF(W,Z,Y,X));
+	 emit_pack_store_4ub(p, dest, temp);
+	 update_src_ptr(p, srcECX, vtxESI, a);
+	 break;
+      case EMIT_4CHAN_4F_RGBA:
+	 switch (CHAN_TYPE) {
+	 case GL_UNSIGNED_BYTE:
+	    get_src_ptr(p, srcECX, vtxESI, a);
+	    emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);
+	    emit_pack_store_4ub(p, dest, temp);
+	    update_src_ptr(p, srcECX, vtxESI, a);
+	    break;
+	 case GL_FLOAT:
+	    get_src_ptr(p, srcECX, vtxESI, a);
+	    emit_load(p, temp, 4, x86_deref(srcECX), a->inputsize);
+	    emit_store(p, dest, 4, temp);
+	    update_src_ptr(p, srcECX, vtxESI, a);
+	    break;
+	 case GL_UNSIGNED_SHORT:
+	 default:
+	    printf("unknown CHAN_TYPE %s\n", _mesa_lookup_enum_by_nr(CHAN_TYPE));
+	    return GL_FALSE;
+	 }
+	 break;
+      default:
+	 printf("unknown a[%d].format %d\n", j, a->format);
+	 return GL_FALSE;	/* catch any new opcodes */
+      }
+      
+      /* Increment j by at least 1 - may have been incremented above also:
+       */
+      j++;
+   }
+
+   /* Next vertex:
+    */
+   x86_lea(&p->func, vertexEAX, x86_make_disp(vertexEAX, vtx->vertex_size));
+
+   /* decr count, loop if not zero
+    */
+   x86_dec(&p->func, countEBP);
+   x86_test(&p->func, countEBP, countEBP); 
+   x86_jcc(&p->func, cc_NZ, label);
+
+   /* Exit mmx state?
+    */
+   if (p->func.need_emms)
+      mmx_emms(&p->func);
+
+   /* Land forward jump here:
+    */
+   x86_fixup_fwd_jump(&p->func, fixup);
+
+   /* Pop regs and return
+    */
+   x86_pop(&p->func, x86_get_base_reg(vtxESI));
+   x86_pop(&p->func, countEBP);
+   x86_ret(&p->func);
+
+   assert(!vtx->emit);
+   vtx->emit = (tnl_emit_func)x86_get_func(&p->func);
+
+   assert( (char *) p->func.csr - (char *) p->func.store <= MAX_SSE_CODE_SIZE );
+   return GL_TRUE;
+}
+
+
+
+void _tnl_generate_sse_emit( struct gl_context *ctx )
+{
+   struct tnl_clipspace *vtx = GET_VERTEX_STATE(ctx);
+   struct x86_program p;   
+
+   if (!cpu_has_xmm) {
+      vtx->codegen_emit = NULL;
+      return;
+   }
+
+   memset(&p, 0, sizeof(p));
+
+   p.ctx = ctx;
+   p.inputs_safe = 0;		/* for now */
+   p.outputs_safe = 0;		/* for now */
+   p.have_sse2 = cpu_has_xmm2;
+   p.identity = x86_make_reg(file_XMM, 6);
+   p.chan0 = x86_make_reg(file_XMM, 7);
+
+   if (!x86_init_func_size(&p.func, MAX_SSE_CODE_SIZE)) {
+      vtx->emit = NULL;
+      return;
+   }
+
+   if (build_vertex_emit(&p)) {
+      _tnl_register_fastpath( vtx, GL_TRUE );
+   }
+   else {
+      /* Note the failure so that we don't keep trying to codegen an
+       * impossible state:
+       */
+      _tnl_register_fastpath( vtx, GL_FALSE );
+      x86_release_func(&p.func);
+   }
+}
+
+#else
+
+void _tnl_generate_sse_emit( struct gl_context *ctx )
+{
+   /* Dummy version for when USE_SSE_ASM not defined */
+}
+
+#endif