svn merge ^/branches/released . --username marha

22 files changed, 7286 insertions, 0 deletions

diff --git a/mesalib/src/mesa/vbo/descrip.mms b/mesalib/src/mesa/vbo/descrip.mms
new file mode 100644
index 000000000..c2a16a04f
--- /dev/null
+++ b/mesalib/src/mesa/vbo/descrip.mms
@@ -0,0 +1,62 @@
+# Makefile for core library for VMS
+# contributed by Jouk Jansen  joukj@hrem.nano.tudelft.nl
+# Last revision : 3 October 2007
+
+.first
+	define gl [---.include.gl]
+	define math [-.math]
+	define vbo [-.vbo]
+	define tnl [-.tnl]
+	define shader [-.shader]
+	define swrast [-.swrast]
+	define swrast_setup [-.swrast_setup]
+	define main [-.main]
+	define glapi [-.glapi]
+
+.include [---]mms-config.
+
+##### MACROS #####
+
+VPATH = RCS
+
+INCDIR = [---.include],[-.main],[-.glapi],[-.shader],[-.shader.slang]
+LIBDIR = [---.lib]
+CFLAGS = /include=($(INCDIR),[])/define=(PTHREADS=1)/name=(as_is,short)/float=ieee/ieee=denorm
+
+SOURCES =vbo_context.c,vbo_exec.c,vbo_exec_api.c,vbo_exec_array.c,\
+	vbo_exec_draw.c,vbo_exec_eval.c,vbo_rebase.c,vbo_save.c,\
+	vbo_save_api.c,vbo_save_draw.c,vbo_save_loopback.c,\
+	vbo_split.c,vbo_split_copy.c,vbo_split_inplace.c
+
+OBJECTS =vbo_context.obj,vbo_exec.obj,vbo_exec_api.obj,vbo_exec_array.obj,\
+	vbo_exec_draw.obj,vbo_exec_eval.obj,vbo_rebase.obj,vbo_save.obj,\
+	vbo_save_api.obj,vbo_save_draw.obj,vbo_save_loopback.obj,\
+	vbo_split.obj,vbo_split_copy.obj,vbo_split_inplace.obj
+
+##### RULES #####
+
+VERSION=Mesa V3.4
+
+##### TARGETS #####
+# Make the library
+$(LIBDIR)$(GL_LIB) : $(OBJECTS)
+  @ library $(LIBDIR)$(GL_LIB) $(OBJECTS)
+
+clean :
+	purge
+	delete *.obj;*
+
+vbo_context.obj : vbo_context.c
+vbo_exec.obj : vbo_exec.c
+vbo_exec_api.obj : vbo_exec_api.c
+vbo_exec_array.obj : vbo_exec_array.c
+vbo_exec_draw.obj : vbo_exec_draw.c
+vbo_exec_eval.obj : vbo_exec_eval.c
+vbo_rebase.obj : vbo_rebase.c
+vbo_save.obj : vbo_save.c
+vbo_save_api.obj : vbo_save_api.c
+vbo_save_draw.obj : vbo_save_draw.c
+vbo_save_loopback.obj : vbo_save_loopback.c
+vbo_split.obj : vbo_split.c
+vbo_split_copy.obj : vbo_split_copy.c
+vbo_split_inplace.obj : vbo_split_inplace.c
diff --git a/mesalib/src/mesa/vbo/vbo.h b/mesalib/src/mesa/vbo/vbo.h
new file mode 100644
index 000000000..5986e9357
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo.h
@@ -0,0 +1,142 @@
+/*
+ * mesa 3-D graphics library
+ * Version:  6.5
+ *
+ * Copyright (C) 1999-2006  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/**
+ * \file vbo_context.h
+ * \brief VBO builder module datatypes and definitions.
+ * \author Keith Whitwell
+ */
+
+
+#ifndef _VBO_H
+#define _VBO_H
+
+#include "main/mtypes.h"
+
+struct _mesa_prim {
+   GLuint mode:8;
+   GLuint indexed:1;
+   GLuint begin:1;
+   GLuint end:1;
+   GLuint weak:1;
+   GLuint pad:20;
+
+   GLuint start;
+   GLuint count;
+};
+
+/* Would like to call this a "vbo_index_buffer", but this would be
+ * confusing as the indices are not neccessarily yet in a non-null
+ * buffer object.
+ */
+struct _mesa_index_buffer {
+   GLuint count;
+   GLenum type;
+   struct gl_buffer_object *obj;
+   const void *ptr;
+};
+
+
+
+GLboolean _vbo_CreateContext( GLcontext *ctx );
+void _vbo_DestroyContext( GLcontext *ctx );
+void _vbo_InvalidateState( GLcontext *ctx, GLuint new_state );
+
+
+typedef void (*vbo_draw_func)( GLcontext *ctx,
+			       const struct gl_client_array **arrays,
+			       const struct _mesa_prim *prims,
+			       GLuint nr_prims,
+			       const struct _mesa_index_buffer *ib,
+			       GLboolean index_bounds_valid,
+			       GLuint min_index,
+			       GLuint max_index );
+
+
+
+
+/* Utility function to cope with various constraints on tnl modules or
+ * hardware.  This can be used to split an incoming set of arrays and
+ * primitives against the following constraints:
+ *    - Maximum number of indices in index buffer.
+ *    - Maximum number of vertices referenced by index buffer.
+ *    - Maximum hardware vertex buffer size.
+ */
+struct split_limits {
+   GLuint max_verts;
+   GLuint max_indices;
+   GLuint max_vb_size;		/* bytes */
+};
+
+
+void vbo_split_prims( GLcontext *ctx,
+		      const struct gl_client_array *arrays[],
+		      const struct _mesa_prim *prim,
+		      GLuint nr_prims,
+		      const struct _mesa_index_buffer *ib,
+		      GLuint min_index,
+		      GLuint max_index,
+		      vbo_draw_func draw,
+		      const struct split_limits *limits );
+
+
+/* Helpers for dealing translating away non-zero min_index.
+ */
+GLboolean vbo_all_varyings_in_vbos( const struct gl_client_array *arrays[] );
+
+void vbo_rebase_prims( GLcontext *ctx,
+		       const struct gl_client_array *arrays[],
+		       const struct _mesa_prim *prim,
+		       GLuint nr_prims,
+		       const struct _mesa_index_buffer *ib,
+		       GLuint min_index,
+		       GLuint max_index,
+		       vbo_draw_func draw );
+void
+vbo_get_minmax_index(GLcontext *ctx, const struct _mesa_prim *prim,
+		     const struct _mesa_index_buffer *ib,
+		     GLuint *min_index, GLuint *max_index);
+
+void vbo_use_buffer_objects(GLcontext *ctx);
+
+
+void vbo_set_draw_func(GLcontext *ctx, vbo_draw_func func);
+
+
+void GLAPIENTRY
+_vbo_Color4f(GLfloat r, GLfloat g, GLfloat b, GLfloat a);
+
+void GLAPIENTRY
+_vbo_Normal3f(GLfloat x, GLfloat y, GLfloat z);
+
+void GLAPIENTRY
+_vbo_MultiTexCoord4f(GLenum target, GLfloat s, GLfloat t, GLfloat r, GLfloat q);
+
+void GLAPIENTRY
+_vbo_Materialfv(GLenum face, GLenum pname, const GLfloat *params);
+
+void GLAPIENTRY
+_vbo_VertexAttrib4f(GLuint index, GLfloat x, GLfloat y, GLfloat z, GLfloat w);
+
+#endif
diff --git a/mesalib/src/mesa/vbo/vbo_attrib.h b/mesalib/src/mesa/vbo/vbo_attrib.h
new file mode 100644
index 000000000..0ae928f2a
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_attrib.h
@@ -0,0 +1,107 @@
+/*
+ Copyright (C) Intel Corp.  2006.  All Rights Reserved.
+ Intel funded Tungsten Graphics (http://www.tungstengraphics.com) to
+ develop this 3D driver.
+ 
+ Permission is hereby granted, free of charge, to any person obtaining
+ a copy of this software and associated documentation files (the
+ "Software"), to deal in the Software without restriction, including
+ without limitation the rights to use, copy, modify, merge, publish,
+ distribute, sublicense, and/or sell copies of the Software, and to
+ permit persons to whom the Software is furnished to do so, subject to
+ the following conditions:
+ 
+ The above copyright notice and this permission notice (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 NONINFRINGEMENT.
+ IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+ LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+ OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ 
+ **********************************************************************/
+ /*
+  * Authors:
+  *   Keith Whitwell <keith@tungstengraphics.com>
+  */
+
+#ifndef VBO_ATTRIB_H
+#define VBO_ATTRIB_H
+
+
+/*
+ * Note: The first attributes match the VERT_ATTRIB_* definitions
+ * in mtypes.h.  However, the tnl module has additional attributes
+ * for materials, color indexes, edge flags, etc.
+ */
+/* Although it's nice to use these as bit indexes in a DWORD flag, we
+ * could manage without if necessary.  Another limit currently is the
+ * number of bits allocated for these numbers in places like vertex
+ * program instruction formats and register layouts.
+ */
+enum {
+	VBO_ATTRIB_POS = 0,
+	VBO_ATTRIB_WEIGHT = 1,
+	VBO_ATTRIB_NORMAL = 2,
+	VBO_ATTRIB_COLOR0 = 3,
+	VBO_ATTRIB_COLOR1 = 4,
+	VBO_ATTRIB_FOG = 5,
+	VBO_ATTRIB_INDEX = 6,        
+	VBO_ATTRIB_EDGEFLAG = 7,     
+	VBO_ATTRIB_TEX0 = 8,
+	VBO_ATTRIB_TEX1 = 9,
+	VBO_ATTRIB_TEX2 = 10,
+	VBO_ATTRIB_TEX3 = 11,
+	VBO_ATTRIB_TEX4 = 12,
+	VBO_ATTRIB_TEX5 = 13,
+	VBO_ATTRIB_TEX6 = 14,
+	VBO_ATTRIB_TEX7 = 15,
+
+	VBO_ATTRIB_GENERIC0 = 16, /* Not used? */
+	VBO_ATTRIB_GENERIC1 = 17,
+	VBO_ATTRIB_GENERIC2 = 18,
+	VBO_ATTRIB_GENERIC3 = 19,
+	VBO_ATTRIB_GENERIC4 = 20,
+	VBO_ATTRIB_GENERIC5 = 21,
+	VBO_ATTRIB_GENERIC6 = 22,
+	VBO_ATTRIB_GENERIC7 = 23,
+	VBO_ATTRIB_GENERIC8 = 24,
+	VBO_ATTRIB_GENERIC9 = 25,
+	VBO_ATTRIB_GENERIC10 = 26,
+	VBO_ATTRIB_GENERIC11 = 27,
+	VBO_ATTRIB_GENERIC12 = 28,
+	VBO_ATTRIB_GENERIC13 = 29,
+	VBO_ATTRIB_GENERIC14 = 30,
+	VBO_ATTRIB_GENERIC15 = 31,
+
+	/* XXX: in the vertex program InputsRead flag, we alias
+	 * materials and generics and use knowledge about the program
+	 * (whether it is a fixed-function emulation) to
+	 * differentiate.  Here we must keep them apart instead.
+	 */
+	VBO_ATTRIB_MAT_FRONT_AMBIENT = 32, 
+	VBO_ATTRIB_MAT_BACK_AMBIENT = 33,
+	VBO_ATTRIB_MAT_FRONT_DIFFUSE = 34,
+	VBO_ATTRIB_MAT_BACK_DIFFUSE = 35,
+	VBO_ATTRIB_MAT_FRONT_SPECULAR = 36,
+	VBO_ATTRIB_MAT_BACK_SPECULAR = 37,
+	VBO_ATTRIB_MAT_FRONT_EMISSION = 38,
+	VBO_ATTRIB_MAT_BACK_EMISSION = 39,
+	VBO_ATTRIB_MAT_FRONT_SHININESS = 40,
+	VBO_ATTRIB_MAT_BACK_SHININESS = 41,
+	VBO_ATTRIB_MAT_FRONT_INDEXES = 42,
+	VBO_ATTRIB_MAT_BACK_INDEXES = 43, 
+
+	VBO_ATTRIB_MAX = 44
+};
+
+#define VBO_ATTRIB_FIRST_MATERIAL VBO_ATTRIB_MAT_FRONT_AMBIENT
+#define VBO_ATTRIB_LAST_MATERIAL VBO_ATTRIB_MAT_BACK_INDEXES
+
+#define VBO_MAX_COPIED_VERTS 3
+
+#endif
diff --git a/mesalib/src/mesa/vbo/vbo_attrib_tmp.h b/mesalib/src/mesa/vbo/vbo_attrib_tmp.h
new file mode 100644
index 000000000..7a889b8e2
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_attrib_tmp.h
@@ -0,0 +1,486 @@
+/**************************************************************************
+
+Copyright 2002 Tungsten Graphics Inc., Cedar Park, Texas.
+
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+#define ATTR1FV( A, V ) ATTR( A, 1, (V)[0], 0, 0, 1 )
+#define ATTR2FV( A, V ) ATTR( A, 2, (V)[0], (V)[1], 0, 1 )
+#define ATTR3FV( A, V ) ATTR( A, 3, (V)[0], (V)[1], (V)[2], 1 )
+#define ATTR4FV( A, V ) ATTR( A, 4, (V)[0], (V)[1], (V)[2], (V)[3] )
+
+#define ATTR1F( A, X )          ATTR( A, 1, X, 0, 0, 1 )
+#define ATTR2F( A, X, Y )       ATTR( A, 2, X, Y, 0, 1 )
+#define ATTR3F( A, X, Y, Z )    ATTR( A, 3, X, Y, Z, 1 )
+#define ATTR4F( A, X, Y, Z, W ) ATTR( A, 4, X, Y, Z, W )
+
+#define MAT_ATTR( A, N, V ) ATTR( A, N, (V)[0], (V)[1], (V)[2], (V)[3] )
+
+static void GLAPIENTRY TAG(Vertex2f)( GLfloat x, GLfloat y )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR2F( VBO_ATTRIB_POS, x, y );
+}
+
+static void GLAPIENTRY TAG(Vertex2fv)( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR2FV( VBO_ATTRIB_POS, v );
+}
+
+static void GLAPIENTRY TAG(Vertex3f)( GLfloat x, GLfloat y, GLfloat z )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR3F( VBO_ATTRIB_POS, x, y, z );
+}
+
+static void GLAPIENTRY TAG(Vertex3fv)( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR3FV( VBO_ATTRIB_POS, v );
+}
+
+static void GLAPIENTRY TAG(Vertex4f)( GLfloat x, GLfloat y, GLfloat z, GLfloat w )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR4F( VBO_ATTRIB_POS, x, y, z, w );
+}
+
+static void GLAPIENTRY TAG(Vertex4fv)( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR4FV( VBO_ATTRIB_POS, v );
+}
+
+static void GLAPIENTRY TAG(TexCoord1f)( GLfloat x )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR1F( VBO_ATTRIB_TEX0, x );
+}
+
+static void GLAPIENTRY TAG(TexCoord1fv)( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR1FV( VBO_ATTRIB_TEX0, v );
+}
+
+static void GLAPIENTRY TAG(TexCoord2f)( GLfloat x, GLfloat y )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR2F( VBO_ATTRIB_TEX0, x, y );
+}
+
+static void GLAPIENTRY TAG(TexCoord2fv)( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR2FV( VBO_ATTRIB_TEX0, v );
+}
+
+static void GLAPIENTRY TAG(TexCoord3f)( GLfloat x, GLfloat y, GLfloat z )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR3F( VBO_ATTRIB_TEX0, x, y, z );
+}
+
+static void GLAPIENTRY TAG(TexCoord3fv)( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR3FV( VBO_ATTRIB_TEX0, v );
+}
+
+static void GLAPIENTRY TAG(TexCoord4f)( GLfloat x, GLfloat y, GLfloat z, GLfloat w )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR4F( VBO_ATTRIB_TEX0, x, y, z, w );
+}
+
+static void GLAPIENTRY TAG(TexCoord4fv)( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR4FV( VBO_ATTRIB_TEX0, v );
+}
+
+static void GLAPIENTRY TAG(Normal3f)( GLfloat x, GLfloat y, GLfloat z )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR3F( VBO_ATTRIB_NORMAL, x, y, z );
+}
+
+static void GLAPIENTRY TAG(Normal3fv)( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR3FV( VBO_ATTRIB_NORMAL, v );
+}
+
+static void GLAPIENTRY TAG(FogCoordfEXT)( GLfloat x )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR1F( VBO_ATTRIB_FOG, x );
+}
+
+static void GLAPIENTRY TAG(FogCoordfvEXT)( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR1FV( VBO_ATTRIB_FOG, v );
+}
+
+static void GLAPIENTRY TAG(Color3f)( GLfloat x, GLfloat y, GLfloat z )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR3F( VBO_ATTRIB_COLOR0, x, y, z );
+}
+
+static void GLAPIENTRY TAG(Color3fv)( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR3FV( VBO_ATTRIB_COLOR0, v );
+}
+
+static void GLAPIENTRY TAG(Color4f)( GLfloat x, GLfloat y, GLfloat z, GLfloat w )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR4F( VBO_ATTRIB_COLOR0, x, y, z, w );
+}
+
+static void GLAPIENTRY TAG(Color4fv)( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR4FV( VBO_ATTRIB_COLOR0, v );
+}
+
+static void GLAPIENTRY TAG(SecondaryColor3fEXT)( GLfloat x, GLfloat y, GLfloat z )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR3F( VBO_ATTRIB_COLOR1, x, y, z );
+}
+
+static void GLAPIENTRY TAG(SecondaryColor3fvEXT)( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR3FV( VBO_ATTRIB_COLOR1, v );
+}
+
+
+static void GLAPIENTRY TAG(EdgeFlag)( GLboolean b )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR1F( VBO_ATTRIB_EDGEFLAG, (GLfloat)b );
+}
+
+static void GLAPIENTRY TAG(Indexf)( GLfloat f )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR1F( VBO_ATTRIB_INDEX, f );
+}
+
+static void GLAPIENTRY TAG(Indexfv)( const GLfloat *f )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   ATTR1FV( VBO_ATTRIB_INDEX, f );
+}
+
+
+static void GLAPIENTRY TAG(MultiTexCoord1f)( GLenum target, GLfloat x  )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   GLuint attr = (target & 0x7) + VBO_ATTRIB_TEX0;
+   ATTR1F( attr, x );
+}
+
+static void GLAPIENTRY TAG(MultiTexCoord1fv)( GLenum target, const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   GLuint attr = (target & 0x7) + VBO_ATTRIB_TEX0;
+   ATTR1FV( attr, v );
+}
+
+static void GLAPIENTRY TAG(MultiTexCoord2f)( GLenum target, GLfloat x, GLfloat y )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   GLuint attr = (target & 0x7) + VBO_ATTRIB_TEX0;
+   ATTR2F( attr, x, y );
+}
+
+static void GLAPIENTRY TAG(MultiTexCoord2fv)( GLenum target, const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   GLuint attr = (target & 0x7) + VBO_ATTRIB_TEX0;
+   ATTR2FV( attr, v );
+}
+
+static void GLAPIENTRY TAG(MultiTexCoord3f)( GLenum target, GLfloat x, GLfloat y,
+				    GLfloat z)
+{
+   GET_CURRENT_CONTEXT( ctx );
+   GLuint attr = (target & 0x7) + VBO_ATTRIB_TEX0;
+   ATTR3F( attr, x, y, z );
+}
+
+static void GLAPIENTRY TAG(MultiTexCoord3fv)( GLenum target, const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   GLuint attr = (target & 0x7) + VBO_ATTRIB_TEX0;
+   ATTR3FV( attr, v );
+}
+
+static void GLAPIENTRY TAG(MultiTexCoord4f)( GLenum target, GLfloat x, GLfloat y,
+				    GLfloat z, GLfloat w )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   GLuint attr = (target & 0x7) + VBO_ATTRIB_TEX0;
+   ATTR4F( attr, x, y, z, w );
+}
+
+static void GLAPIENTRY TAG(MultiTexCoord4fv)( GLenum target, const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   GLuint attr = (target & 0x7) + VBO_ATTRIB_TEX0;
+   ATTR4FV( attr, v );
+}
+
+
+static void GLAPIENTRY TAG(VertexAttrib1fARB)( GLuint index, GLfloat x )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   if (index == 0) 
+      ATTR1F(0, x);
+   else if (index < MAX_VERTEX_GENERIC_ATTRIBS)
+      ATTR1F(VBO_ATTRIB_GENERIC0 + index, x);
+   else
+      ERROR();
+}
+
+static void GLAPIENTRY TAG(VertexAttrib1fvARB)( GLuint index, 
+					       const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   if (index == 0) 
+      ATTR1FV(0, v);
+   else if (index < MAX_VERTEX_GENERIC_ATTRIBS)
+      ATTR1FV(VBO_ATTRIB_GENERIC0 + index, v);
+   else
+      ERROR();
+}
+
+static void GLAPIENTRY TAG(VertexAttrib2fARB)( GLuint index, GLfloat x, 
+					      GLfloat y )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   if (index == 0) 
+      ATTR2F(0, x, y);
+   else if (index < MAX_VERTEX_GENERIC_ATTRIBS)
+      ATTR2F(VBO_ATTRIB_GENERIC0 + index, x, y);
+   else
+      ERROR();
+}
+
+static void GLAPIENTRY TAG(VertexAttrib2fvARB)( GLuint index,
+					       const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   if (index == 0) 
+      ATTR2FV(0, v);
+   else if (index < MAX_VERTEX_GENERIC_ATTRIBS)
+      ATTR2FV(VBO_ATTRIB_GENERIC0 + index, v);
+   else
+      ERROR();
+}
+
+static void GLAPIENTRY TAG(VertexAttrib3fARB)( GLuint index, GLfloat x,
+					      GLfloat y, GLfloat z )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   if (index == 0) 
+      ATTR3F(0, x, y, z);
+   else if (index < MAX_VERTEX_GENERIC_ATTRIBS)
+      ATTR3F(VBO_ATTRIB_GENERIC0 + index, x, y, z);
+   else
+      ERROR();
+}
+
+static void GLAPIENTRY TAG(VertexAttrib3fvARB)( GLuint index,
+					       const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   if (index == 0) 
+      ATTR3FV(0, v);
+   else if (index < MAX_VERTEX_GENERIC_ATTRIBS)
+      ATTR3FV(VBO_ATTRIB_GENERIC0 + index, v);
+   else
+      ERROR();
+}
+
+static void GLAPIENTRY TAG(VertexAttrib4fARB)( GLuint index, GLfloat x,
+					      GLfloat y, GLfloat z,
+					      GLfloat w )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   if (index == 0) 
+      ATTR4F(0, x, y, z, w);
+   else if (index < MAX_VERTEX_GENERIC_ATTRIBS)
+      ATTR4F(VBO_ATTRIB_GENERIC0 + index, x, y, z, w);
+   else
+      ERROR();
+}
+
+static void GLAPIENTRY TAG(VertexAttrib4fvARB)( GLuint index, 
+					       const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   if (index == 0) 
+      ATTR4FV(0, v);
+   else if (index < MAX_VERTEX_GENERIC_ATTRIBS)
+      ATTR4FV(VBO_ATTRIB_GENERIC0 + index, v);
+   else
+      ERROR();
+}
+
+
+/* In addition to supporting NV_vertex_program, these entrypoints are
+ * used by the display list and other code specifically because of
+ * their property of aliasing with other attributes.  (See
+ * vbo_save_loopback.c)
+ */
+static void GLAPIENTRY TAG(VertexAttrib1fNV)( GLuint index, GLfloat x )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   if (index < VBO_ATTRIB_MAX)
+      ATTR1F(index, x);
+}
+
+static void GLAPIENTRY TAG(VertexAttrib1fvNV)( GLuint index, 
+					       const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   if (index < VBO_ATTRIB_MAX) 
+      ATTR1FV(index, v);
+}
+
+static void GLAPIENTRY TAG(VertexAttrib2fNV)( GLuint index, GLfloat x, 
+					      GLfloat y )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   if (index < VBO_ATTRIB_MAX) 
+      ATTR2F(index, x, y);
+}
+
+static void GLAPIENTRY TAG(VertexAttrib2fvNV)( GLuint index,
+					       const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   if (index < VBO_ATTRIB_MAX) 
+      ATTR2FV(index, v);
+}
+
+static void GLAPIENTRY TAG(VertexAttrib3fNV)( GLuint index, GLfloat x,
+					      GLfloat y, GLfloat z )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   if (index < VBO_ATTRIB_MAX) 
+      ATTR3F(index, x, y, z);
+}
+
+static void GLAPIENTRY TAG(VertexAttrib3fvNV)( GLuint index,
+					       const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   if (index < VBO_ATTRIB_MAX) 
+      ATTR3FV(index, v);
+}
+
+static void GLAPIENTRY TAG(VertexAttrib4fNV)( GLuint index, GLfloat x,
+					      GLfloat y, GLfloat z,
+					      GLfloat w )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   if (index < VBO_ATTRIB_MAX) 
+      ATTR4F(index, x, y, z, w);
+}
+
+static void GLAPIENTRY TAG(VertexAttrib4fvNV)( GLuint index, 
+					       const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   if (index < VBO_ATTRIB_MAX) 
+      ATTR4FV(index, v);
+}
+
+
+#define MAT( ATTR, N, face, params )			\
+do {							\
+   if (face != GL_BACK)					\
+      MAT_ATTR( ATTR, N, params ); /* front */		\
+   if (face != GL_FRONT)				\
+      MAT_ATTR( ATTR + 1, N, params ); /* back */	\
+} while (0)
+
+
+/* Colormaterial conflicts are dealt with later.
+ */
+static void GLAPIENTRY TAG(Materialfv)( GLenum face, GLenum pname, 
+			       const GLfloat *params )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   switch (pname) {
+   case GL_EMISSION:
+      MAT( VBO_ATTRIB_MAT_FRONT_EMISSION, 4, face, params );
+      break;
+   case GL_AMBIENT:
+      MAT( VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params );
+      break;
+   case GL_DIFFUSE:
+      MAT( VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params );
+      break;
+   case GL_SPECULAR:
+      MAT( VBO_ATTRIB_MAT_FRONT_SPECULAR, 4, face, params );
+      break;
+   case GL_SHININESS:
+      MAT( VBO_ATTRIB_MAT_FRONT_SHININESS, 1, face, params );
+      break;
+   case GL_COLOR_INDEXES:
+      MAT( VBO_ATTRIB_MAT_FRONT_INDEXES, 3, face, params );
+      break;
+   case GL_AMBIENT_AND_DIFFUSE:
+      MAT( VBO_ATTRIB_MAT_FRONT_AMBIENT, 4, face, params );
+      MAT( VBO_ATTRIB_MAT_FRONT_DIFFUSE, 4, face, params );
+      break;
+   default:
+      ERROR();
+      return;
+   }
+}
+
+
+#undef ATTR1FV
+#undef ATTR2FV
+#undef ATTR3FV
+#undef ATTR4FV
+
+#undef ATTR1F
+#undef ATTR2F
+#undef ATTR3F
+#undef ATTR4F
+
+#undef MAT
+#undef MAT_ATTR
diff --git a/mesalib/src/mesa/vbo/vbo_context.c b/mesalib/src/mesa/vbo/vbo_context.c
new file mode 100644
index 000000000..75c32e0b9
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_context.c
@@ -0,0 +1,273 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  6.3
+ *
+ * Copyright (C) 1999-2005  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "main/imports.h"
+#include "main/mtypes.h"
+#include "main/api_arrayelt.h"
+#include "main/bufferobj.h"
+#include "math/m_eval.h"
+#include "vbo.h"
+#include "vbo_context.h"
+
+#if 0
+/* Reach out and grab this to use as the default:
+ */
+extern void _tnl_draw_prims( GLcontext *ctx,
+			     const struct gl_client_array *arrays[],
+			     const struct _mesa_prim *prims,
+			     GLuint nr_prims,
+			     const struct _mesa_index_buffer *ib,
+			     GLuint min_index,
+			     GLuint max_index );
+#endif
+
+
+
+#define NR_LEGACY_ATTRIBS 16
+#define NR_GENERIC_ATTRIBS 16
+#define NR_MAT_ATTRIBS 12
+
+static GLuint check_size( const GLfloat *attr )
+{
+   if (attr[3] != 1.0) return 4;
+   if (attr[2] != 0.0) return 3;
+   if (attr[1] != 0.0) return 2;
+   return 1;		
+}
+
+static void init_legacy_currval(GLcontext *ctx)
+{
+   struct vbo_context *vbo = vbo_context(ctx);
+   struct gl_client_array *arrays = vbo->legacy_currval;
+   GLuint i;
+
+   memset(arrays, 0, sizeof(*arrays) * NR_LEGACY_ATTRIBS);
+
+   /* Set up a constant (StrideB == 0) array for each current
+    * attribute:
+    */
+   for (i = 0; i < NR_LEGACY_ATTRIBS; i++) {
+      struct gl_client_array *cl = &arrays[i];
+
+      /* Size will have to be determined at runtime:
+       */
+      cl->Size = check_size(ctx->Current.Attrib[i]);
+      cl->Stride = 0;
+      cl->StrideB = 0;
+      cl->Enabled = 1;
+      cl->Type = GL_FLOAT;
+      cl->Format = GL_RGBA;
+      cl->Ptr = (const void *)ctx->Current.Attrib[i];
+      _mesa_reference_buffer_object(ctx, &cl->BufferObj,
+                                    ctx->Shared->NullBufferObj);
+   }
+}
+
+
+static void init_generic_currval(GLcontext *ctx)
+{
+   struct vbo_context *vbo = vbo_context(ctx);
+   struct gl_client_array *arrays = vbo->generic_currval;
+   GLuint i;
+
+   memset(arrays, 0, sizeof(*arrays) * NR_GENERIC_ATTRIBS);
+
+   for (i = 0; i < NR_GENERIC_ATTRIBS; i++) {
+      struct gl_client_array *cl = &arrays[i];
+
+      /* This will have to be determined at runtime:
+       */
+      cl->Size = 1;
+      cl->Type = GL_FLOAT;
+      cl->Format = GL_RGBA;
+      cl->Ptr = (const void *)ctx->Current.Attrib[VERT_ATTRIB_GENERIC0 + i];
+      cl->Stride = 0;
+      cl->StrideB = 0;
+      cl->Enabled = 1;
+      _mesa_reference_buffer_object(ctx, &cl->BufferObj,
+                                    ctx->Shared->NullBufferObj);
+   }
+}
+
+
+static void init_mat_currval(GLcontext *ctx)
+{
+   struct vbo_context *vbo = vbo_context(ctx);
+   struct gl_client_array *arrays = vbo->mat_currval;
+   GLuint i;
+
+   ASSERT(NR_MAT_ATTRIBS == MAT_ATTRIB_MAX);
+
+   memset(arrays, 0, sizeof(*arrays) * NR_MAT_ATTRIBS);
+
+   /* Set up a constant (StrideB == 0) array for each current
+    * attribute:
+    */
+   for (i = 0; i < NR_MAT_ATTRIBS; i++) {
+      struct gl_client_array *cl = &arrays[i];
+
+      /* Size is fixed for the material attributes, for others will
+       * be determined at runtime:
+       */
+      switch (i - VERT_ATTRIB_GENERIC0) {
+      case MAT_ATTRIB_FRONT_SHININESS:
+      case MAT_ATTRIB_BACK_SHININESS:
+	 cl->Size = 1;
+	 break;
+      case MAT_ATTRIB_FRONT_INDEXES:
+      case MAT_ATTRIB_BACK_INDEXES:
+	 cl->Size = 3;
+	 break;
+      default:
+	 cl->Size = 4;
+	 break;
+      }
+
+      cl->Ptr = (const void *)ctx->Light.Material.Attrib[i];
+      cl->Type = GL_FLOAT;
+      cl->Format = GL_RGBA;
+      cl->Stride = 0;
+      cl->StrideB = 0;
+      cl->Enabled = 1;
+      _mesa_reference_buffer_object(ctx, &cl->BufferObj,
+                                    ctx->Shared->NullBufferObj);
+   }
+}
+
+#if 0
+
+static void vbo_exec_current_init( struct vbo_exec_context *exec ) 
+{
+   GLcontext *ctx = exec->ctx;
+   GLint i;
+
+   /* setup the pointers for the typical 16 vertex attributes */
+   for (i = 0; i < VBO_ATTRIB_FIRST_MATERIAL; i++) 
+      exec->vtx.current[i] = ctx->Current.Attrib[i];
+
+   /* setup pointers for the 12 material attributes */
+   for (i = 0; i < MAT_ATTRIB_MAX; i++)
+      exec->vtx.current[VBO_ATTRIB_FIRST_MATERIAL + i] = 
+	 ctx->Light.Material.Attrib[i];
+}
+#endif
+
+GLboolean _vbo_CreateContext( GLcontext *ctx )
+{
+   struct vbo_context *vbo = CALLOC_STRUCT(vbo_context);
+
+   ctx->swtnl_im = (void *)vbo;
+
+   /* Initialize the arrayelt helper
+    */
+   if (!ctx->aelt_context &&
+       !_ae_create_context( ctx )) {
+      return GL_FALSE;
+   }
+
+   /* TODO: remove these pointers.
+    */
+   vbo->legacy_currval = &vbo->currval[VBO_ATTRIB_POS];
+   vbo->generic_currval = &vbo->currval[VBO_ATTRIB_GENERIC0];
+   vbo->mat_currval = &vbo->currval[VBO_ATTRIB_MAT_FRONT_AMBIENT];
+
+   init_legacy_currval( ctx );
+   init_generic_currval( ctx );
+   init_mat_currval( ctx );
+
+   /* Build mappings from VERT_ATTRIB -> VBO_ATTRIB depending on type
+    * of vertex program active.
+    */
+   {
+      GLuint i;
+
+      /* When no vertex program, pull in the material attributes in
+       * the 16..32 generic range.
+       */
+      for (i = 0; i < 16; i++) 
+	 vbo->map_vp_none[i] = i;
+      for (i = 0; i < 12; i++) 
+	 vbo->map_vp_none[16+i] = VBO_ATTRIB_MAT_FRONT_AMBIENT + i;
+      for (i = 0; i < 4; i++)
+	 vbo->map_vp_none[28+i] = i;	
+      
+      for (i = 0; i < Elements(vbo->map_vp_arb); i++)
+	 vbo->map_vp_arb[i] = i;
+   }
+
+
+   /* By default: 
+    */
+#if 0 /* dead - see vbo_set_draw_func() */
+   vbo->draw_prims = _tnl_draw_prims;
+#endif
+
+   /* Hook our functions into exec and compile dispatch tables.  These
+    * will pretty much be permanently installed, which means that the
+    * vtxfmt mechanism can be removed now.
+    */
+   vbo_exec_init( ctx );
+#if FEATURE_dlist
+   vbo_save_init( ctx );
+#endif
+
+   _math_init_eval();
+
+   return GL_TRUE;
+}
+
+void _vbo_InvalidateState( GLcontext *ctx, GLuint new_state )
+{
+   _ae_invalidate_state(ctx, new_state);
+   vbo_exec_invalidate_state(ctx, new_state);
+}
+
+
+void _vbo_DestroyContext( GLcontext *ctx )
+{
+   if (ctx->aelt_context) {
+      _ae_destroy_context( ctx );
+      ctx->aelt_context = NULL;
+   }
+
+   if (vbo_context(ctx)) {
+      vbo_exec_destroy(ctx);
+#if FEATURE_dlist
+      vbo_save_destroy(ctx);
+#endif
+      FREE(vbo_context(ctx));
+      ctx->swtnl_im = NULL;
+   }
+}
+
+
+void vbo_set_draw_func(GLcontext *ctx, vbo_draw_func func)
+{
+   struct vbo_context *vbo = vbo_context(ctx);
+   vbo->draw_prims = func;
+}
+
diff --git a/mesalib/src/mesa/vbo/vbo_context.h b/mesalib/src/mesa/vbo/vbo_context.h
new file mode 100644
index 000000000..8b726dc8a
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_context.h
@@ -0,0 +1,114 @@
+/*
+ * mesa 3-D graphics library
+ * Version:  6.5
+ *
+ * Copyright (C) 1999-2006  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/**
+ * \file vbo_context.h
+ * \brief VBO builder module datatypes and definitions.
+ * \author Keith Whitwell
+ */
+
+
+/**
+ * \mainpage The VBO builder module
+ *
+ * This module hooks into the GL dispatch table and catches all vertex
+ * building and drawing commands, such as glVertex3f, glBegin and
+ * glDrawArrays.  The module stores all incoming vertex data as arrays
+ * in GL vertex buffer objects (VBOs), and translates all drawing
+ * commands into calls to a driver supplied DrawPrimitives() callback.
+ *
+ * The module captures both immediate mode and display list drawing,
+ * and manages the allocation, reference counting and deallocation of
+ * vertex buffer objects itself.
+ * 
+ * The DrawPrimitives() callback can be either implemented by the
+ * driver itself or hooked to the tnl module's _tnl_draw_primitives()
+ * function for hardware without tnl capablilties or during fallbacks.
+ */
+
+
+#ifndef _VBO_CONTEXT_H
+#define _VBO_CONTEXT_H
+
+#include "vbo.h"
+#include "vbo_attrib.h"
+#include "vbo_exec.h"
+#if FEATURE_dlist
+#include "vbo_save.h"
+#endif
+
+
+struct vbo_context {
+   struct gl_client_array currval[VBO_ATTRIB_MAX];
+   
+   /* These point into the above.  TODO: remove. 
+    */
+   struct gl_client_array *legacy_currval;
+   struct gl_client_array *generic_currval;
+   struct gl_client_array *mat_currval;
+
+   GLuint map_vp_none[VERT_ATTRIB_MAX];
+   GLuint map_vp_arb[VERT_ATTRIB_MAX];
+
+   GLfloat *current[VBO_ATTRIB_MAX]; /* points into ctx->Current, ctx->Light.Material */
+   GLfloat CurrentFloatEdgeFlag;
+
+
+   struct vbo_exec_context exec;
+#if FEATURE_dlist
+   struct vbo_save_context save;
+#endif
+
+   /* Callback into the driver.  This must always succeed, the driver
+    * is responsible for initiating any fallback actions required:
+    */
+   vbo_draw_func draw_prims;
+};
+
+
+static INLINE struct vbo_context *vbo_context(GLcontext *ctx) 
+{
+   return (struct vbo_context *)(ctx->swtnl_im);
+}
+
+
+/**
+ * Return VP_x token to indicate whether we're running fixed-function
+ * vertex transformation, an NV vertex program or ARB vertex program/shader.
+ */
+static INLINE enum vp_mode
+get_program_mode( GLcontext *ctx )
+{
+   if (!ctx->VertexProgram._Current)
+      return VP_NONE;
+   else if (ctx->VertexProgram._Current == ctx->VertexProgram._TnlProgram)
+      return VP_NONE;
+   else if (ctx->VertexProgram._Current->IsNVProgram)
+      return VP_NV;
+   else
+      return VP_ARB;
+}
+
+
+#endif
diff --git a/mesalib/src/mesa/vbo/vbo_exec.c b/mesalib/src/mesa/vbo/vbo_exec.c
new file mode 100644
index 000000000..e168a89ea
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_exec.c
@@ -0,0 +1,97 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  6.3
+ *
+ * Copyright (C) 1999-2005  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+
+#include "main/api_arrayelt.h"
+#include "main/glheader.h"
+#include "main/imports.h"
+#include "main/context.h"
+#include "main/macros.h"
+#include "main/mtypes.h"
+#include "main/vtxfmt.h"
+
+#include "vbo_context.h"
+
+void vbo_exec_init( GLcontext *ctx )
+{
+   struct vbo_exec_context *exec = &vbo_context(ctx)->exec;
+
+   exec->ctx = ctx;
+
+   /* Initialize the arrayelt helper
+    */
+   if (!ctx->aelt_context &&
+       !_ae_create_context( ctx )) 
+      return;
+
+   vbo_exec_vtx_init( exec );
+   vbo_exec_array_init( exec );
+
+   /* Hook our functions into exec and compile dispatch tables.
+    */
+   _mesa_install_exec_vtxfmt( ctx, &exec->vtxfmt );
+
+   ctx->Driver.NeedFlush = 0;
+   ctx->Driver.CurrentExecPrimitive = PRIM_OUTSIDE_BEGIN_END;
+   ctx->Driver.BeginVertices = vbo_exec_BeginVertices;
+   ctx->Driver.FlushVertices = vbo_exec_FlushVertices;
+
+   vbo_exec_invalidate_state( ctx, ~0 );
+}
+
+
+void vbo_exec_destroy( GLcontext *ctx )
+{
+   struct vbo_exec_context *exec = &vbo_context(ctx)->exec;
+
+   if (ctx->aelt_context) {
+      _ae_destroy_context( ctx );
+      ctx->aelt_context = NULL;
+   }
+
+   vbo_exec_vtx_destroy( exec );
+   vbo_exec_array_destroy( exec );
+}
+
+/* Really want to install these callbacks to a central facility to be
+ * invoked according to the state flags.  That will have to wait for a
+ * mesa rework:
+ */ 
+void vbo_exec_invalidate_state( GLcontext *ctx, GLuint new_state )
+{
+   struct vbo_exec_context *exec = &vbo_context(ctx)->exec;
+
+   if (new_state & (_NEW_PROGRAM|_NEW_EVAL))
+      exec->eval.recalculate_maps = 1;
+
+   _ae_invalidate_state(ctx, new_state);
+}
+
+
+
+
+
diff --git a/mesalib/src/mesa/vbo/vbo_exec.h b/mesalib/src/mesa/vbo/vbo_exec.h
new file mode 100644
index 000000000..e0f44892c
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_exec.h
@@ -0,0 +1,179 @@
+/**************************************************************************
+
+Copyright 2002 Tungsten Graphics Inc., Cedar Park, Texas.
+
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell <keith@tungstengraphics.com>
+ *
+ */
+
+#ifndef __VBO_EXEC_H__
+#define __VBO_EXEC_H__
+
+#include "main/mtypes.h"
+#include "vbo.h"
+#include "vbo_attrib.h"
+
+
+#define VBO_MAX_PRIM 64
+
+/* Wierd implementation stuff:
+ */
+#define VBO_VERT_BUFFER_SIZE (1024*64)	/* bytes */
+#define VBO_MAX_ATTR_CODEGEN 16 
+#define ERROR_ATTRIB 16
+
+
+/** Current vertex program mode */
+enum vp_mode {
+   VP_NONE,   /**< fixed function */
+   VP_NV,     /**< NV vertex program */
+   VP_ARB     /**< ARB vertex program or GLSL vertex shader */
+};
+
+
+struct vbo_exec_eval1_map {
+   struct gl_1d_map *map;
+   GLuint sz;
+};
+
+struct vbo_exec_eval2_map {
+   struct gl_2d_map *map;
+   GLuint sz;
+};
+
+
+
+struct vbo_exec_copied_vtx {
+   GLfloat buffer[VBO_ATTRIB_MAX * 4 * VBO_MAX_COPIED_VERTS];
+   GLuint nr;
+};
+
+
+typedef void (*vbo_attrfv_func)( const GLfloat * );
+
+
+struct vbo_exec_context
+{
+   GLcontext *ctx;   
+   GLvertexformat vtxfmt;
+
+   struct {
+      struct gl_buffer_object *bufferobj;
+
+      GLuint vertex_size;       /* in dwords */
+
+      struct _mesa_prim prim[VBO_MAX_PRIM];
+      GLuint prim_count;
+
+      GLfloat *buffer_map;
+      GLfloat *buffer_ptr;              /* cursor, points into buffer */
+      GLuint   buffer_used;             /* in bytes */
+      GLfloat vertex[VBO_ATTRIB_MAX*4]; /* current vertex */
+
+      GLuint vert_count;
+      GLuint max_vert;
+      struct vbo_exec_copied_vtx copied;
+
+      GLubyte attrsz[VBO_ATTRIB_MAX];
+      GLubyte active_sz[VBO_ATTRIB_MAX];
+
+      GLfloat *attrptr[VBO_ATTRIB_MAX]; 
+      struct gl_client_array arrays[VBO_ATTRIB_MAX];
+
+      /* According to program mode, the values above plus current
+       * values are squashed down to the 32 attributes passed to the
+       * vertex program below:
+       */
+      enum vp_mode program_mode;
+      GLuint enabled_flags;
+      const struct gl_client_array *inputs[VERT_ATTRIB_MAX];
+   } vtx;
+
+   
+   struct {
+      GLboolean recalculate_maps;
+      struct vbo_exec_eval1_map map1[VERT_ATTRIB_MAX];
+      struct vbo_exec_eval2_map map2[VERT_ATTRIB_MAX];
+   } eval;
+
+   struct {
+      enum vp_mode program_mode;
+      GLuint enabled_flags;
+      GLuint array_obj;
+
+      /* These just mirror the current arrayobj (todo: make arrayobj
+       * look like this and remove the mirror):
+       */
+      const struct gl_client_array *legacy_array[16];
+      const struct gl_client_array *generic_array[16];
+
+      /* Arrays and current values manipulated according to program
+       * mode, etc.  These are the attributes as seen by vertex
+       * programs:
+       */
+      const struct gl_client_array *inputs[VERT_ATTRIB_MAX];
+   } array;
+};
+
+
+
+/* External API:
+ */
+void vbo_exec_init( GLcontext *ctx );
+void vbo_exec_destroy( GLcontext *ctx );
+void vbo_exec_invalidate_state( GLcontext *ctx, GLuint new_state );
+void vbo_exec_FlushVertices_internal( GLcontext *ctx, GLboolean unmap );
+
+void vbo_exec_BeginVertices( GLcontext *ctx );
+void vbo_exec_FlushVertices( GLcontext *ctx, GLuint flags );
+
+
+/* Internal functions:
+ */
+void vbo_exec_array_init( struct vbo_exec_context *exec );
+void vbo_exec_array_destroy( struct vbo_exec_context *exec );
+
+
+void vbo_exec_vtx_init( struct vbo_exec_context *exec );
+void vbo_exec_vtx_destroy( struct vbo_exec_context *exec );
+void vbo_exec_vtx_flush( struct vbo_exec_context *exec, GLboolean unmap );
+void vbo_exec_vtx_map( struct vbo_exec_context *exec );
+void vbo_exec_vtx_wrap( struct vbo_exec_context *exec );
+
+void vbo_exec_eval_update( struct vbo_exec_context *exec );
+
+void vbo_exec_do_EvalCoord2f( struct vbo_exec_context *exec, 
+				     GLfloat u, GLfloat v );
+
+void vbo_exec_do_EvalCoord1f( struct vbo_exec_context *exec,
+				     GLfloat u);
+
+extern GLboolean 
+vbo_validate_shaders(GLcontext *ctx);
+
+#endif
diff --git a/mesalib/src/mesa/vbo/vbo_exec_api.c b/mesalib/src/mesa/vbo/vbo_exec_api.c
new file mode 100644
index 000000000..387d4ee3d
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_exec_api.c
@@ -0,0 +1,838 @@
+/**************************************************************************
+
+Copyright 2002-2008 Tungsten Graphics Inc., Cedar Park, Texas.
+
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "main/glheader.h"
+#include "main/bufferobj.h"
+#include "main/context.h"
+#include "main/macros.h"
+#include "main/vtxfmt.h"
+#if FEATURE_dlist
+#include "main/dlist.h"
+#endif
+#include "main/state.h"
+#include "main/light.h"
+#include "main/api_arrayelt.h"
+#include "main/api_noop.h"
+#include "glapi/dispatch.h"
+
+#include "vbo_context.h"
+
+#ifdef ERROR
+#undef ERROR
+#endif
+
+
+/** ID/name for immediate-mode VBO */
+#define IMM_BUFFER_NAME 0xaabbccdd
+
+
+static void reset_attrfv( struct vbo_exec_context *exec );
+
+
+/* Close off the last primitive, execute the buffer, restart the
+ * primitive.  
+ */
+static void vbo_exec_wrap_buffers( struct vbo_exec_context *exec )
+{
+   if (exec->vtx.prim_count == 0) {
+      exec->vtx.copied.nr = 0;
+      exec->vtx.vert_count = 0;
+      exec->vtx.buffer_ptr = exec->vtx.buffer_map;
+   }
+   else {
+      GLuint last_begin = exec->vtx.prim[exec->vtx.prim_count-1].begin;
+      GLuint last_count;
+
+      if (exec->ctx->Driver.CurrentExecPrimitive != PRIM_OUTSIDE_BEGIN_END) {
+	 GLint i = exec->vtx.prim_count - 1;
+	 assert(i >= 0);
+	 exec->vtx.prim[i].count = (exec->vtx.vert_count - 
+				    exec->vtx.prim[i].start);
+      }
+
+      last_count = exec->vtx.prim[exec->vtx.prim_count-1].count;
+
+      /* Execute the buffer and save copied vertices.
+       */
+      if (exec->vtx.vert_count)
+	 vbo_exec_vtx_flush( exec, GL_FALSE );
+      else {
+	 exec->vtx.prim_count = 0;
+	 exec->vtx.copied.nr = 0;
+      }
+
+      /* Emit a glBegin to start the new list.
+       */
+      assert(exec->vtx.prim_count == 0);
+
+      if (exec->ctx->Driver.CurrentExecPrimitive != PRIM_OUTSIDE_BEGIN_END) {
+	 exec->vtx.prim[0].mode = exec->ctx->Driver.CurrentExecPrimitive;
+	 exec->vtx.prim[0].start = 0;
+	 exec->vtx.prim[0].count = 0;
+	 exec->vtx.prim_count++;
+      
+	 if (exec->vtx.copied.nr == last_count)
+	    exec->vtx.prim[0].begin = last_begin;
+      }
+   }
+}
+
+
+/* Deal with buffer wrapping where provoked by the vertex buffer
+ * filling up, as opposed to upgrade_vertex().
+ */
+void vbo_exec_vtx_wrap( struct vbo_exec_context *exec )
+{
+   GLfloat *data = exec->vtx.copied.buffer;
+   GLuint i;
+
+   /* Run pipeline on current vertices, copy wrapped vertices
+    * to exec->vtx.copied.
+    */
+   vbo_exec_wrap_buffers( exec );
+   
+   /* Copy stored stored vertices to start of new list. 
+    */
+   assert(exec->vtx.max_vert - exec->vtx.vert_count > exec->vtx.copied.nr);
+
+   for (i = 0 ; i < exec->vtx.copied.nr ; i++) {
+      _mesa_memcpy( exec->vtx.buffer_ptr, data, 
+		    exec->vtx.vertex_size * sizeof(GLfloat));
+      exec->vtx.buffer_ptr += exec->vtx.vertex_size;
+      data += exec->vtx.vertex_size;
+      exec->vtx.vert_count++;
+   }
+
+   exec->vtx.copied.nr = 0;
+}
+
+
+/*
+ * Copy the active vertex's values to the ctx->Current fields.
+ */
+static void vbo_exec_copy_to_current( struct vbo_exec_context *exec )
+{
+   GLcontext *ctx = exec->ctx;
+   struct vbo_context *vbo = vbo_context(ctx);
+   GLuint i;
+
+   for (i = VBO_ATTRIB_POS+1 ; i < VBO_ATTRIB_MAX ; i++) {
+      if (exec->vtx.attrsz[i]) {
+         /* Note: the exec->vtx.current[i] pointers point into the
+          * ctx->Current.Attrib and ctx->Light.Material.Attrib arrays.
+          */
+	 GLfloat *current = (GLfloat *)vbo->currval[i].Ptr;
+         GLfloat tmp[4];
+
+         COPY_CLEAN_4V(tmp, 
+                       exec->vtx.attrsz[i], 
+                       exec->vtx.attrptr[i]);
+         
+         if (memcmp(current, tmp, sizeof(tmp)) != 0)
+         { 
+            memcpy(current, tmp, sizeof(tmp));
+	 
+            /* Given that we explicitly state size here, there is no need
+             * for the COPY_CLEAN above, could just copy 16 bytes and be
+             * done.  The only problem is when Mesa accesses ctx->Current
+             * directly.
+             */
+            vbo->currval[i].Size = exec->vtx.attrsz[i];
+
+            /* This triggers rather too much recalculation of Mesa state
+             * that doesn't get used (eg light positions).
+             */
+            if (i >= VBO_ATTRIB_MAT_FRONT_AMBIENT &&
+                i <= VBO_ATTRIB_MAT_BACK_INDEXES)
+               ctx->NewState |= _NEW_LIGHT;
+            
+            ctx->NewState |= _NEW_CURRENT_ATTRIB;
+         }
+      }
+   }
+
+   /* Colormaterial -- this kindof sucks.
+    */
+   if (ctx->Light.ColorMaterialEnabled &&
+       exec->vtx.attrsz[VBO_ATTRIB_COLOR0]) {
+      _mesa_update_color_material(ctx, 
+				  ctx->Current.Attrib[VBO_ATTRIB_COLOR0]);
+   }
+}
+
+
+static void vbo_exec_copy_from_current( struct vbo_exec_context *exec )
+{
+   GLcontext *ctx = exec->ctx;
+   struct vbo_context *vbo = vbo_context(ctx);
+   GLint i;
+
+   for (i = VBO_ATTRIB_POS+1 ; i < VBO_ATTRIB_MAX ; i++) {
+      const GLfloat *current = (GLfloat *)vbo->currval[i].Ptr;
+      switch (exec->vtx.attrsz[i]) {
+      case 4: exec->vtx.attrptr[i][3] = current[3];
+      case 3: exec->vtx.attrptr[i][2] = current[2];
+      case 2: exec->vtx.attrptr[i][1] = current[1];
+      case 1: exec->vtx.attrptr[i][0] = current[0];
+	 break;
+      }
+   }
+}
+
+
+/* Flush existing data, set new attrib size, replay copied vertices.
+ */ 
+static void vbo_exec_wrap_upgrade_vertex( struct vbo_exec_context *exec,
+					  GLuint attr,
+					  GLuint newsz )
+{
+   GLcontext *ctx = exec->ctx;
+   struct vbo_context *vbo = vbo_context(ctx);
+   GLint lastcount = exec->vtx.vert_count;
+   GLfloat *tmp;
+   GLuint oldsz;
+   GLuint i;
+
+   /* Run pipeline on current vertices, copy wrapped vertices
+    * to exec->vtx.copied.
+    */
+   vbo_exec_wrap_buffers( exec );
+
+
+   /* Do a COPY_TO_CURRENT to ensure back-copying works for the case
+    * when the attribute already exists in the vertex and is having
+    * its size increased.  
+    */
+   vbo_exec_copy_to_current( exec );
+
+
+   /* Heuristic: Attempt to isolate attributes received outside
+    * begin/end so that they don't bloat the vertices.
+    */
+   if (ctx->Driver.CurrentExecPrimitive == PRIM_OUTSIDE_BEGIN_END &&
+       exec->vtx.attrsz[attr] == 0 && 
+       lastcount > 8 &&
+       exec->vtx.vertex_size) {
+      reset_attrfv( exec );
+   }
+
+   /* Fix up sizes:
+    */
+   oldsz = exec->vtx.attrsz[attr];
+   exec->vtx.attrsz[attr] = newsz;
+
+   exec->vtx.vertex_size += newsz - oldsz;
+   exec->vtx.max_vert = ((VBO_VERT_BUFFER_SIZE - exec->vtx.buffer_used) / 
+                         (exec->vtx.vertex_size * sizeof(GLfloat)));
+   exec->vtx.vert_count = 0;
+   exec->vtx.buffer_ptr = exec->vtx.buffer_map;
+   
+
+   /* Recalculate all the attrptr[] values
+    */
+   for (i = 0, tmp = exec->vtx.vertex ; i < VBO_ATTRIB_MAX ; i++) {
+      if (exec->vtx.attrsz[i]) {
+	 exec->vtx.attrptr[i] = tmp;
+	 tmp += exec->vtx.attrsz[i];
+      }
+      else 
+	 exec->vtx.attrptr[i] = NULL; /* will not be dereferenced */
+   }
+
+   /* Copy from current to repopulate the vertex with correct values.
+    */
+   vbo_exec_copy_from_current( exec );
+
+   /* Replay stored vertices to translate them
+    * to new format here.
+    *
+    * -- No need to replay - just copy piecewise
+    */
+   if (exec->vtx.copied.nr)
+   {
+      GLfloat *data = exec->vtx.copied.buffer;
+      GLfloat *dest = exec->vtx.buffer_ptr;
+      GLuint j;
+
+      assert(exec->vtx.buffer_ptr == exec->vtx.buffer_map);
+      
+      for (i = 0 ; i < exec->vtx.copied.nr ; i++) {
+	 for (j = 0 ; j < VBO_ATTRIB_MAX ; j++) {
+	    if (exec->vtx.attrsz[j]) {
+	       if (j == attr) {
+		  if (oldsz) {
+		     COPY_CLEAN_4V( dest, oldsz, data );
+		     data += oldsz;
+		     dest += newsz;
+		  } else {
+		     const GLfloat *current = (const GLfloat *)vbo->currval[j].Ptr;
+		     COPY_SZ_4V( dest, newsz, current );
+		     dest += newsz;
+		  }
+	       }
+	       else {
+		  GLuint sz = exec->vtx.attrsz[j];
+		  COPY_SZ_4V( dest, sz, data );
+		  dest += sz;
+		  data += sz;
+	       }
+	    }
+	 }
+      }
+
+      exec->vtx.buffer_ptr = dest;
+      exec->vtx.vert_count += exec->vtx.copied.nr;
+      exec->vtx.copied.nr = 0;
+   }
+}
+
+
+static void vbo_exec_fixup_vertex( GLcontext *ctx,
+				   GLuint attr, GLuint sz )
+{
+   struct vbo_exec_context *exec = &vbo_context(ctx)->exec;
+   int i;
+
+   if (sz > exec->vtx.attrsz[attr]) {
+      /* New size is larger.  Need to flush existing vertices and get
+       * an enlarged vertex format.
+       */
+      vbo_exec_wrap_upgrade_vertex( exec, attr, sz );
+   }
+   else if (sz < exec->vtx.active_sz[attr]) {
+      static const GLfloat id[4] = { 0, 0, 0, 1 };
+
+      /* New size is smaller - just need to fill in some
+       * zeros.  Don't need to flush or wrap.
+       */
+      for (i = sz ; i <= exec->vtx.attrsz[attr] ; i++)
+	 exec->vtx.attrptr[attr][i-1] = id[i-1];
+   }
+
+   exec->vtx.active_sz[attr] = sz;
+
+   /* Does setting NeedFlush belong here?  Necessitates resetting
+    * vtxfmt on each flush (otherwise flags won't get reset
+    * afterwards).
+    */
+   if (attr == 0) 
+      exec->ctx->Driver.NeedFlush |= FLUSH_STORED_VERTICES;
+}
+
+
+
+
+/* 
+ */
+#define ATTR( A, N, V0, V1, V2, V3 )				\
+do {								\
+   struct vbo_exec_context *exec = &vbo_context(ctx)->exec;	\
+								\
+   if (exec->vtx.active_sz[A] != N)				\
+      vbo_exec_fixup_vertex(ctx, A, N);			\
+								\
+   {								\
+      GLfloat *dest = exec->vtx.attrptr[A];			\
+      if (N>0) dest[0] = V0;					\
+      if (N>1) dest[1] = V1;					\
+      if (N>2) dest[2] = V2;					\
+      if (N>3) dest[3] = V3;					\
+   }								\
+								\
+   if ((A) == 0) {						\
+      GLuint i;							\
+								\
+      for (i = 0; i < exec->vtx.vertex_size; i++)		\
+	 exec->vtx.buffer_ptr[i] = exec->vtx.vertex[i];		\
+								\
+      exec->vtx.buffer_ptr += exec->vtx.vertex_size;			\
+      exec->ctx->Driver.NeedFlush |= FLUSH_STORED_VERTICES;	\
+								\
+      if (++exec->vtx.vert_count >= exec->vtx.max_vert)		\
+	 vbo_exec_vtx_wrap( exec );				\
+   }								\
+} while (0)
+
+
+#define ERROR() _mesa_error( ctx, GL_INVALID_ENUM, __FUNCTION__ )
+#define TAG(x) vbo_##x
+
+#include "vbo_attrib_tmp.h"
+
+
+
+
+
+/* Eval
+ */
+static void GLAPIENTRY vbo_exec_EvalCoord1f( GLfloat u )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   struct vbo_exec_context *exec = &vbo_context(ctx)->exec;
+
+   {
+      GLint i;
+      if (exec->eval.recalculate_maps) 
+	 vbo_exec_eval_update( exec );
+
+      for (i = 0; i <= VBO_ATTRIB_TEX7; i++) {
+	 if (exec->eval.map1[i].map) 
+	    if (exec->vtx.active_sz[i] != exec->eval.map1[i].sz)
+	       vbo_exec_fixup_vertex( ctx, i, exec->eval.map1[i].sz );
+      }
+   }
+
+
+   _mesa_memcpy( exec->vtx.copied.buffer, exec->vtx.vertex, 
+                 exec->vtx.vertex_size * sizeof(GLfloat));
+
+   vbo_exec_do_EvalCoord1f( exec, u );
+
+   _mesa_memcpy( exec->vtx.vertex, exec->vtx.copied.buffer,
+                 exec->vtx.vertex_size * sizeof(GLfloat));
+}
+
+static void GLAPIENTRY vbo_exec_EvalCoord2f( GLfloat u, GLfloat v )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   struct vbo_exec_context *exec = &vbo_context(ctx)->exec;
+
+   {
+      GLint i;
+      if (exec->eval.recalculate_maps) 
+	 vbo_exec_eval_update( exec );
+
+      for (i = 0; i <= VBO_ATTRIB_TEX7; i++) {
+	 if (exec->eval.map2[i].map) 
+	    if (exec->vtx.active_sz[i] != exec->eval.map2[i].sz)
+	       vbo_exec_fixup_vertex( ctx, i, exec->eval.map2[i].sz );
+      }
+
+      if (ctx->Eval.AutoNormal) 
+	 if (exec->vtx.active_sz[VBO_ATTRIB_NORMAL] != 3)
+	    vbo_exec_fixup_vertex( ctx, VBO_ATTRIB_NORMAL, 3 );
+   }
+
+   _mesa_memcpy( exec->vtx.copied.buffer, exec->vtx.vertex, 
+                 exec->vtx.vertex_size * sizeof(GLfloat));
+
+   vbo_exec_do_EvalCoord2f( exec, u, v );
+
+   _mesa_memcpy( exec->vtx.vertex, exec->vtx.copied.buffer, 
+                 exec->vtx.vertex_size * sizeof(GLfloat));
+}
+
+static void GLAPIENTRY vbo_exec_EvalCoord1fv( const GLfloat *u )
+{
+   vbo_exec_EvalCoord1f( u[0] );
+}
+
+static void GLAPIENTRY vbo_exec_EvalCoord2fv( const GLfloat *u )
+{
+   vbo_exec_EvalCoord2f( u[0], u[1] );
+}
+
+static void GLAPIENTRY vbo_exec_EvalPoint1( GLint i )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   GLfloat du = ((ctx->Eval.MapGrid1u2 - ctx->Eval.MapGrid1u1) /
+		 (GLfloat) ctx->Eval.MapGrid1un);
+   GLfloat u = i * du + ctx->Eval.MapGrid1u1;
+
+   vbo_exec_EvalCoord1f( u );
+}
+
+
+static void GLAPIENTRY vbo_exec_EvalPoint2( GLint i, GLint j )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   GLfloat du = ((ctx->Eval.MapGrid2u2 - ctx->Eval.MapGrid2u1) / 
+		 (GLfloat) ctx->Eval.MapGrid2un);
+   GLfloat dv = ((ctx->Eval.MapGrid2v2 - ctx->Eval.MapGrid2v1) / 
+		 (GLfloat) ctx->Eval.MapGrid2vn);
+   GLfloat u = i * du + ctx->Eval.MapGrid2u1;
+   GLfloat v = j * dv + ctx->Eval.MapGrid2v1;
+
+   vbo_exec_EvalCoord2f( u, v );
+}
+
+
+/* Build a list of primitives on the fly.  Keep
+ * ctx->Driver.CurrentExecPrimitive uptodate as well.
+ */
+static void GLAPIENTRY vbo_exec_Begin( GLenum mode )
+{
+   GET_CURRENT_CONTEXT( ctx ); 
+
+   if (ctx->Driver.CurrentExecPrimitive == PRIM_OUTSIDE_BEGIN_END) {
+      struct vbo_exec_context *exec = &vbo_context(ctx)->exec;
+      int i;
+
+      if (ctx->NewState) {
+	 _mesa_update_state( ctx );
+
+	 CALL_Begin(ctx->Exec, (mode));
+	 return;
+      }
+
+      if (!_mesa_valid_to_render(ctx, "glBegin")) {
+         return;
+      }
+
+      /* Heuristic: attempt to isolate attributes occuring outside
+       * begin/end pairs.
+       */
+      if (exec->vtx.vertex_size && !exec->vtx.attrsz[0]) 
+	 vbo_exec_FlushVertices_internal( ctx, GL_FALSE );
+
+      i = exec->vtx.prim_count++;
+      exec->vtx.prim[i].mode = mode;
+      exec->vtx.prim[i].begin = 1;
+      exec->vtx.prim[i].end = 0;
+      exec->vtx.prim[i].indexed = 0;
+      exec->vtx.prim[i].weak = 0;
+      exec->vtx.prim[i].pad = 0;
+      exec->vtx.prim[i].start = exec->vtx.vert_count;
+      exec->vtx.prim[i].count = 0;
+
+      ctx->Driver.CurrentExecPrimitive = mode;
+   }
+   else 
+      _mesa_error( ctx, GL_INVALID_OPERATION, "glBegin" );
+      
+}
+
+static void GLAPIENTRY vbo_exec_End( void )
+{
+   GET_CURRENT_CONTEXT( ctx ); 
+
+   if (ctx->Driver.CurrentExecPrimitive != PRIM_OUTSIDE_BEGIN_END) {
+      struct vbo_exec_context *exec = &vbo_context(ctx)->exec;
+      int idx = exec->vtx.vert_count;
+      int i = exec->vtx.prim_count - 1;
+
+      exec->vtx.prim[i].end = 1; 
+      exec->vtx.prim[i].count = idx - exec->vtx.prim[i].start;
+
+      ctx->Driver.CurrentExecPrimitive = PRIM_OUTSIDE_BEGIN_END;
+
+      if (exec->vtx.prim_count == VBO_MAX_PRIM)
+	 vbo_exec_vtx_flush( exec, GL_FALSE );
+   }
+   else 
+      _mesa_error( ctx, GL_INVALID_OPERATION, "glEnd" );
+}
+
+
+static void vbo_exec_vtxfmt_init( struct vbo_exec_context *exec )
+{
+   GLvertexformat *vfmt = &exec->vtxfmt;
+
+   vfmt->ArrayElement = _ae_loopback_array_elt;	        /* generic helper */
+   vfmt->Begin = vbo_exec_Begin;
+#if FEATURE_dlist
+   vfmt->CallList = _mesa_CallList;
+   vfmt->CallLists = _mesa_CallLists;
+#endif
+   vfmt->End = vbo_exec_End;
+   vfmt->EvalCoord1f = vbo_exec_EvalCoord1f;
+   vfmt->EvalCoord1fv = vbo_exec_EvalCoord1fv;
+   vfmt->EvalCoord2f = vbo_exec_EvalCoord2f;
+   vfmt->EvalCoord2fv = vbo_exec_EvalCoord2fv;
+   vfmt->EvalPoint1 = vbo_exec_EvalPoint1;
+   vfmt->EvalPoint2 = vbo_exec_EvalPoint2;
+
+   vfmt->Rectf = _mesa_noop_Rectf;
+   vfmt->EvalMesh1 = _mesa_noop_EvalMesh1;
+   vfmt->EvalMesh2 = _mesa_noop_EvalMesh2;
+
+
+   /* from attrib_tmp.h:
+    */
+   vfmt->Color3f = vbo_Color3f;
+   vfmt->Color3fv = vbo_Color3fv;
+   vfmt->Color4f = vbo_Color4f;
+   vfmt->Color4fv = vbo_Color4fv;
+   vfmt->FogCoordfEXT = vbo_FogCoordfEXT;
+   vfmt->FogCoordfvEXT = vbo_FogCoordfvEXT;
+   vfmt->MultiTexCoord1fARB = vbo_MultiTexCoord1f;
+   vfmt->MultiTexCoord1fvARB = vbo_MultiTexCoord1fv;
+   vfmt->MultiTexCoord2fARB = vbo_MultiTexCoord2f;
+   vfmt->MultiTexCoord2fvARB = vbo_MultiTexCoord2fv;
+   vfmt->MultiTexCoord3fARB = vbo_MultiTexCoord3f;
+   vfmt->MultiTexCoord3fvARB = vbo_MultiTexCoord3fv;
+   vfmt->MultiTexCoord4fARB = vbo_MultiTexCoord4f;
+   vfmt->MultiTexCoord4fvARB = vbo_MultiTexCoord4fv;
+   vfmt->Normal3f = vbo_Normal3f;
+   vfmt->Normal3fv = vbo_Normal3fv;
+   vfmt->SecondaryColor3fEXT = vbo_SecondaryColor3fEXT;
+   vfmt->SecondaryColor3fvEXT = vbo_SecondaryColor3fvEXT;
+   vfmt->TexCoord1f = vbo_TexCoord1f;
+   vfmt->TexCoord1fv = vbo_TexCoord1fv;
+   vfmt->TexCoord2f = vbo_TexCoord2f;
+   vfmt->TexCoord2fv = vbo_TexCoord2fv;
+   vfmt->TexCoord3f = vbo_TexCoord3f;
+   vfmt->TexCoord3fv = vbo_TexCoord3fv;
+   vfmt->TexCoord4f = vbo_TexCoord4f;
+   vfmt->TexCoord4fv = vbo_TexCoord4fv;
+   vfmt->Vertex2f = vbo_Vertex2f;
+   vfmt->Vertex2fv = vbo_Vertex2fv;
+   vfmt->Vertex3f = vbo_Vertex3f;
+   vfmt->Vertex3fv = vbo_Vertex3fv;
+   vfmt->Vertex4f = vbo_Vertex4f;
+   vfmt->Vertex4fv = vbo_Vertex4fv;
+   
+   vfmt->VertexAttrib1fARB = vbo_VertexAttrib1fARB;
+   vfmt->VertexAttrib1fvARB = vbo_VertexAttrib1fvARB;
+   vfmt->VertexAttrib2fARB = vbo_VertexAttrib2fARB;
+   vfmt->VertexAttrib2fvARB = vbo_VertexAttrib2fvARB;
+   vfmt->VertexAttrib3fARB = vbo_VertexAttrib3fARB;
+   vfmt->VertexAttrib3fvARB = vbo_VertexAttrib3fvARB;
+   vfmt->VertexAttrib4fARB = vbo_VertexAttrib4fARB;
+   vfmt->VertexAttrib4fvARB = vbo_VertexAttrib4fvARB;
+
+   vfmt->VertexAttrib1fNV = vbo_VertexAttrib1fNV;
+   vfmt->VertexAttrib1fvNV = vbo_VertexAttrib1fvNV;
+   vfmt->VertexAttrib2fNV = vbo_VertexAttrib2fNV;
+   vfmt->VertexAttrib2fvNV = vbo_VertexAttrib2fvNV;
+   vfmt->VertexAttrib3fNV = vbo_VertexAttrib3fNV;
+   vfmt->VertexAttrib3fvNV = vbo_VertexAttrib3fvNV;
+   vfmt->VertexAttrib4fNV = vbo_VertexAttrib4fNV;
+   vfmt->VertexAttrib4fvNV = vbo_VertexAttrib4fvNV;
+
+   vfmt->Materialfv = vbo_Materialfv;
+
+   vfmt->EdgeFlag = vbo_EdgeFlag;
+   vfmt->Indexf = vbo_Indexf;
+   vfmt->Indexfv = vbo_Indexfv;
+
+}
+
+
+/**
+ * Tell the VBO module to use a real OpenGL vertex buffer object to
+ * store accumulated immediate-mode vertex data.
+ * This replaces the malloced buffer which was created in
+ * vb_exec_vtx_init() below.
+ */
+void vbo_use_buffer_objects(GLcontext *ctx)
+{
+   struct vbo_exec_context *exec = &vbo_context(ctx)->exec;
+   /* Any buffer name but 0 can be used here since this bufferobj won't
+    * go into the bufferobj hashtable.
+    */
+   GLuint bufName = IMM_BUFFER_NAME;
+   GLenum target = GL_ARRAY_BUFFER_ARB;
+   GLenum usage = GL_STREAM_DRAW_ARB;
+   GLsizei size = VBO_VERT_BUFFER_SIZE;
+
+   /* Make sure this func is only used once */
+   assert(exec->vtx.bufferobj == ctx->Shared->NullBufferObj);
+   if (exec->vtx.buffer_map) {
+      _mesa_align_free(exec->vtx.buffer_map);
+      exec->vtx.buffer_map = NULL;
+      exec->vtx.buffer_ptr = NULL;
+   }
+
+   /* Allocate a real buffer object now */
+   exec->vtx.bufferobj = ctx->Driver.NewBufferObject(ctx, bufName, target);
+   ctx->Driver.BufferData(ctx, target, size, NULL, usage, exec->vtx.bufferobj);
+}
+
+
+
+void vbo_exec_vtx_init( struct vbo_exec_context *exec )
+{
+   GLcontext *ctx = exec->ctx;
+   struct vbo_context *vbo = vbo_context(ctx);
+   GLuint i;
+
+   /* Allocate a buffer object.  Will just reuse this object
+    * continuously, unless vbo_use_buffer_objects() is called to enable
+    * use of real VBOs.
+    */
+   _mesa_reference_buffer_object(ctx,
+                                 &exec->vtx.bufferobj,
+                                 ctx->Shared->NullBufferObj);
+
+   ASSERT(!exec->vtx.buffer_map);
+   exec->vtx.buffer_map = (GLfloat *)ALIGN_MALLOC(VBO_VERT_BUFFER_SIZE, 64);
+   exec->vtx.buffer_ptr = exec->vtx.buffer_map;
+
+   vbo_exec_vtxfmt_init( exec );
+
+   /* Hook our functions into the dispatch table.
+    */
+   _mesa_install_exec_vtxfmt( exec->ctx, &exec->vtxfmt );
+
+   for (i = 0 ; i < VBO_ATTRIB_MAX ; i++) {
+      exec->vtx.attrsz[i] = 0;
+      exec->vtx.active_sz[i] = 0;
+      exec->vtx.inputs[i] = &exec->vtx.arrays[i];
+   }
+   
+   {
+      struct gl_client_array *arrays = exec->vtx.arrays;
+      memcpy(arrays,      vbo->legacy_currval,  16 * sizeof(arrays[0]));
+      memcpy(arrays + 16, vbo->generic_currval, 16 * sizeof(arrays[0]));
+   }
+
+   exec->vtx.vertex_size = 0;
+}
+
+
+void vbo_exec_vtx_destroy( struct vbo_exec_context *exec )
+{
+   /* using a real VBO for vertex data */
+   GLcontext *ctx = exec->ctx;
+   unsigned i;
+
+   /* True VBOs should already be unmapped
+    */
+   if (exec->vtx.buffer_map) {
+      ASSERT(exec->vtx.bufferobj->Name == 0 ||
+             exec->vtx.bufferobj->Name == IMM_BUFFER_NAME);
+      if (exec->vtx.bufferobj->Name == 0) {
+         ALIGN_FREE(exec->vtx.buffer_map);
+         exec->vtx.buffer_map = NULL;
+         exec->vtx.buffer_ptr = NULL;
+      }
+   }
+
+   /* Drop any outstanding reference to the vertex buffer
+    */
+   for (i = 0; i < Elements(exec->vtx.arrays); i++) {
+      _mesa_reference_buffer_object(ctx,
+                                    &exec->vtx.arrays[i].BufferObj,
+                                    NULL);
+   }
+
+   /* Free the vertex buffer:
+    */
+   _mesa_reference_buffer_object(ctx, &exec->vtx.bufferobj, NULL);
+}
+
+void vbo_exec_BeginVertices( GLcontext *ctx )
+{
+   struct vbo_exec_context *exec = &vbo_context(ctx)->exec;
+   if (0) _mesa_printf("%s\n", __FUNCTION__);
+   vbo_exec_vtx_map( exec );
+
+   assert((exec->ctx->Driver.NeedFlush & FLUSH_UPDATE_CURRENT) == 0);
+   exec->ctx->Driver.NeedFlush |= FLUSH_UPDATE_CURRENT;
+}
+
+void vbo_exec_FlushVertices_internal( GLcontext *ctx, GLboolean unmap )
+{
+   struct vbo_exec_context *exec = &vbo_context(ctx)->exec;
+
+   if (exec->vtx.vert_count || unmap) {
+      vbo_exec_vtx_flush( exec, unmap );
+   }
+
+   if (exec->vtx.vertex_size) {
+      vbo_exec_copy_to_current( exec );
+      reset_attrfv( exec );
+   }
+}
+
+
+
+void vbo_exec_FlushVertices( GLcontext *ctx, GLuint flags )
+{
+   struct vbo_exec_context *exec = &vbo_context(ctx)->exec;
+
+   if (0) _mesa_printf("%s\n", __FUNCTION__);
+
+   if (exec->ctx->Driver.CurrentExecPrimitive != PRIM_OUTSIDE_BEGIN_END) {
+      if (0) _mesa_printf("%s - inside begin/end\n", __FUNCTION__);
+      return;
+   }
+
+   vbo_exec_FlushVertices_internal( ctx, GL_TRUE );
+
+   /* Need to do this to ensure BeginVertices gets called again:
+    */
+   if (exec->ctx->Driver.NeedFlush & FLUSH_UPDATE_CURRENT) {
+      _mesa_restore_exec_vtxfmt( ctx );
+      exec->ctx->Driver.NeedFlush &= ~FLUSH_UPDATE_CURRENT;
+   }
+
+   exec->ctx->Driver.NeedFlush &= ~flags;
+}
+
+
+static void reset_attrfv( struct vbo_exec_context *exec )
+{   
+   GLuint i;
+
+   for (i = 0 ; i < VBO_ATTRIB_MAX ; i++) {
+      exec->vtx.attrsz[i] = 0;
+      exec->vtx.active_sz[i] = 0;
+   }
+
+   exec->vtx.vertex_size = 0;
+}
+      
+
+void GLAPIENTRY
+_vbo_Color4f(GLfloat r, GLfloat g, GLfloat b, GLfloat a)
+{
+   vbo_Color4f(r, g, b, a);
+}
+
+
+void GLAPIENTRY
+_vbo_Normal3f(GLfloat x, GLfloat y, GLfloat z)
+{
+   vbo_Normal3f(x, y, z);
+}
+
+
+void GLAPIENTRY
+_vbo_MultiTexCoord4f(GLenum target, GLfloat s, GLfloat t, GLfloat r, GLfloat q)
+{
+   vbo_MultiTexCoord4f(target, s, t, r, q);
+}
+
+void GLAPIENTRY
+_vbo_Materialfv(GLenum face, GLenum pname, const GLfloat *params)
+{
+   vbo_Materialfv(face, pname, params);
+}
+
+
+void GLAPIENTRY
+_vbo_VertexAttrib4f(GLuint index, GLfloat x, GLfloat y, GLfloat z, GLfloat w)
+{
+   vbo_VertexAttrib4fARB(index, x, y, z, w);
+}
diff --git a/mesalib/src/mesa/vbo/vbo_exec_array.c b/mesalib/src/mesa/vbo/vbo_exec_array.c
new file mode 100644
index 000000000..39c295763
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_exec_array.c
@@ -0,0 +1,965 @@
+/**************************************************************************
+ * 
+ * Copyright 2003 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * Copyright 2009 VMware, Inc.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "main/glheader.h"
+#include "main/context.h"
+#include "main/state.h"
+#include "main/api_validate.h"
+#include "main/api_noop.h"
+#include "main/varray.h"
+#include "main/bufferobj.h"
+#include "main/macros.h"
+#include "glapi/dispatch.h"
+
+#include "vbo_context.h"
+
+
+/**
+ * Compute min and max elements for glDraw[Range]Elements() calls.
+ */
+void
+vbo_get_minmax_index(GLcontext *ctx,
+		     const struct _mesa_prim *prim,
+		     const struct _mesa_index_buffer *ib,
+		     GLuint *min_index, GLuint *max_index)
+{
+   GLuint i;
+   GLsizei count = prim->count;
+   const void *indices;
+
+   if (_mesa_is_bufferobj(ib->obj)) {
+      const GLvoid *map = ctx->Driver.MapBuffer(ctx,
+                                                GL_ELEMENT_ARRAY_BUFFER_ARB,
+                                                GL_READ_ONLY,
+                                                ib->obj);
+      indices = ADD_POINTERS(map, ib->ptr);
+   } else {
+      indices = ib->ptr;
+   }
+
+   switch (ib->type) {
+   case GL_UNSIGNED_INT: {
+      const GLuint *ui_indices = (const GLuint *)indices;
+      GLuint max_ui = ui_indices[count-1];
+      GLuint min_ui = ui_indices[0];
+      for (i = 0; i < count; i++) {
+	 if (ui_indices[i] > max_ui) max_ui = ui_indices[i];
+	 if (ui_indices[i] < min_ui) min_ui = ui_indices[i];
+      }
+      *min_index = min_ui;
+      *max_index = max_ui;
+      break;
+   }
+   case GL_UNSIGNED_SHORT: {
+      const GLushort *us_indices = (const GLushort *)indices;
+      GLuint max_us = us_indices[count-1];
+      GLuint min_us = us_indices[0];
+      for (i = 0; i < count; i++) {
+	 if (us_indices[i] > max_us) max_us = us_indices[i];
+	 if (us_indices[i] < min_us) min_us = us_indices[i];
+      }
+      *min_index = min_us;
+      *max_index = max_us;
+      break;
+   }
+   case GL_UNSIGNED_BYTE: {
+      const GLubyte *ub_indices = (const GLubyte *)indices;
+      GLuint max_ub = ub_indices[count-1];
+      GLuint min_ub = ub_indices[0];
+      for (i = 0; i < count; i++) {
+	 if (ub_indices[i] > max_ub) max_ub = ub_indices[i];
+	 if (ub_indices[i] < min_ub) min_ub = ub_indices[i];
+      }
+      *min_index = min_ub;
+      *max_index = max_ub;
+      break;
+   }
+   default:
+      assert(0);
+      break;
+   }
+
+   if (_mesa_is_bufferobj(ib->obj)) {
+      ctx->Driver.UnmapBuffer(ctx,
+			      GL_ELEMENT_ARRAY_BUFFER_ARB,
+			      ib->obj);
+   }
+}
+
+
+/**
+ * Check that element 'j' of the array has reasonable data.
+ * Map VBO if needed.
+ */
+static void
+check_array_data(GLcontext *ctx, struct gl_client_array *array,
+                 GLuint attrib, GLuint j)
+{
+   if (array->Enabled) {
+      const void *data = array->Ptr;
+      if (_mesa_is_bufferobj(array->BufferObj)) {
+         if (!array->BufferObj->Pointer) {
+            /* need to map now */
+            array->BufferObj->Pointer = ctx->Driver.MapBuffer(ctx,
+                                                              GL_ARRAY_BUFFER_ARB,
+                                                              GL_READ_ONLY,
+                                                              array->BufferObj);
+         }
+         data = ADD_POINTERS(data, array->BufferObj->Pointer);
+      }
+      switch (array->Type) {
+      case GL_FLOAT:
+         {
+            GLfloat *f = (GLfloat *) ((GLubyte *) data + array->StrideB * j);
+            GLuint k;
+            for (k = 0; k < array->Size; k++) {
+               if (IS_INF_OR_NAN(f[k]) ||
+                   f[k] >= 1.0e20 || f[k] <= -1.0e10) {
+                  _mesa_printf("Bad array data:\n");
+                  _mesa_printf("  Element[%u].%u = %f\n", j, k, f[k]);
+                  _mesa_printf("  Array %u at %p\n", attrib, (void* ) array);
+                  _mesa_printf("  Type 0x%x, Size %d, Stride %d\n",
+                               array->Type, array->Size, array->Stride);
+                  _mesa_printf("  Address/offset %p in Buffer Object %u\n",
+                               array->Ptr, array->BufferObj->Name);
+                  f[k] = 1.0; /* XXX replace the bad value! */
+               }
+               //assert(!IS_INF_OR_NAN(f[k]));
+            }
+         }
+         break;
+      default:
+         ;
+      }
+   }
+}
+
+
+/**
+ * Unmap the buffer object referenced by given array, if mapped.
+ */
+static void
+unmap_array_buffer(GLcontext *ctx, struct gl_client_array *array)
+{
+   if (array->Enabled &&
+       _mesa_is_bufferobj(array->BufferObj) &&
+       _mesa_bufferobj_mapped(array->BufferObj)) {
+      ctx->Driver.UnmapBuffer(ctx,
+                              GL_ARRAY_BUFFER_ARB,
+                              array->BufferObj);
+   }
+}
+
+
+/**
+ * Examine the array's data for NaNs, etc.
+ */
+static void
+check_draw_elements_data(GLcontext *ctx, GLsizei count, GLenum elemType,
+                         const void *elements)
+{
+   struct gl_array_object *arrayObj = ctx->Array.ArrayObj;
+   const void *elemMap;
+   GLint i, k;
+
+   if (_mesa_is_bufferobj(ctx->Array.ElementArrayBufferObj)) {
+      elemMap = ctx->Driver.MapBuffer(ctx,
+                                      GL_ELEMENT_ARRAY_BUFFER_ARB,
+                                      GL_READ_ONLY,
+                                      ctx->Array.ElementArrayBufferObj);
+      elements = ADD_POINTERS(elements, elemMap);
+   }
+
+   for (i = 0; i < count; i++) {
+      GLuint j;
+
+      /* j = element[i] */
+      switch (elemType) {
+      case GL_UNSIGNED_BYTE:
+         j = ((const GLubyte *) elements)[i];
+         break;
+      case GL_UNSIGNED_SHORT:
+         j = ((const GLushort *) elements)[i];
+         break;
+      case GL_UNSIGNED_INT:
+         j = ((const GLuint *) elements)[i];
+         break;
+      default:
+         assert(0);
+      }
+
+      /* check element j of each enabled array */
+      check_array_data(ctx, &arrayObj->Vertex, VERT_ATTRIB_POS, j);
+      check_array_data(ctx, &arrayObj->Normal, VERT_ATTRIB_NORMAL, j);
+      check_array_data(ctx, &arrayObj->Color, VERT_ATTRIB_COLOR0, j);
+      check_array_data(ctx, &arrayObj->SecondaryColor, VERT_ATTRIB_COLOR1, j);
+      for (k = 0; k < Elements(arrayObj->TexCoord); k++) {
+         check_array_data(ctx, &arrayObj->TexCoord[k], VERT_ATTRIB_TEX0 + k, j);
+      }
+      for (k = 0; k < Elements(arrayObj->VertexAttrib); k++) {
+         check_array_data(ctx, &arrayObj->VertexAttrib[k], VERT_ATTRIB_GENERIC0 + k, j);
+      }
+   }
+
+   if (_mesa_is_bufferobj(ctx->Array.ElementArrayBufferObj)) {
+      ctx->Driver.UnmapBuffer(ctx,
+			      GL_ELEMENT_ARRAY_BUFFER_ARB,
+			      ctx->Array.ElementArrayBufferObj);
+   }
+
+   unmap_array_buffer(ctx, &arrayObj->Vertex);
+   unmap_array_buffer(ctx, &arrayObj->Normal);
+   unmap_array_buffer(ctx, &arrayObj->Color);
+   for (k = 0; k < Elements(arrayObj->TexCoord); k++) {
+      unmap_array_buffer(ctx, &arrayObj->TexCoord[k]);
+   }
+   for (k = 0; k < Elements(arrayObj->VertexAttrib); k++) {
+      unmap_array_buffer(ctx, &arrayObj->VertexAttrib[k]);
+   }
+}
+
+
+/**
+ * Check array data, looking for NaNs, etc.
+ */
+static void
+check_draw_arrays_data(GLcontext *ctx, GLint start, GLsizei count)
+{
+   /* TO DO */
+}
+
+
+/**
+ * Print info/data for glDrawArrays().
+ */
+static void
+print_draw_arrays(GLcontext *ctx, struct vbo_exec_context *exec,
+                  GLenum mode, GLint start, GLsizei count)
+{
+   int i;
+
+   _mesa_printf("vbo_exec_DrawArrays(mode 0x%x, start %d, count %d):\n",
+                mode, start, count);
+
+   for (i = 0; i < 32; i++) {
+      GLuint bufName = exec->array.inputs[i]->BufferObj->Name;
+      GLint stride = exec->array.inputs[i]->Stride;
+      _mesa_printf("attr %2d: size %d stride %d  enabled %d  "
+                   "ptr %p  Bufobj %u\n",
+                   i,
+                   exec->array.inputs[i]->Size,
+                   stride,
+                   /*exec->array.inputs[i]->Enabled,*/
+                   exec->array.legacy_array[i]->Enabled,
+                   exec->array.inputs[i]->Ptr,
+                   bufName);
+
+      if (bufName) {
+         struct gl_buffer_object *buf = _mesa_lookup_bufferobj(ctx, bufName);
+         GLubyte *p = ctx->Driver.MapBuffer(ctx, GL_ARRAY_BUFFER_ARB,
+                                            GL_READ_ONLY_ARB, buf);
+         int offset = (int) (GLintptr) exec->array.inputs[i]->Ptr;
+         float *f = (float *) (p + offset);
+         int *k = (int *) f;
+         int i;
+         int n = (count * stride) / 4;
+         if (n > 32)
+            n = 32;
+         _mesa_printf("  Data at offset %d:\n", offset);
+         for (i = 0; i < n; i++) {
+            _mesa_printf("    float[%d] = 0x%08x %f\n", i, k[i], f[i]);
+         }
+         ctx->Driver.UnmapBuffer(ctx, GL_ARRAY_BUFFER_ARB, buf);
+      }
+   }
+}
+
+
+/**
+ * Just translate the arrayobj into a sane layout.
+ */
+static void
+bind_array_obj(GLcontext *ctx)
+{
+   struct vbo_context *vbo = vbo_context(ctx);
+   struct vbo_exec_context *exec = &vbo->exec;
+   struct gl_array_object *arrayObj = ctx->Array.ArrayObj;
+   GLuint i;
+
+   /* TODO: Fix the ArrayObj struct to keep legacy arrays in an array
+    * rather than as individual named arrays.  Then this function can
+    * go away.
+    */
+   exec->array.legacy_array[VERT_ATTRIB_POS] = &arrayObj->Vertex;
+   exec->array.legacy_array[VERT_ATTRIB_WEIGHT] = &arrayObj->Weight;
+   exec->array.legacy_array[VERT_ATTRIB_NORMAL] = &arrayObj->Normal;
+   exec->array.legacy_array[VERT_ATTRIB_COLOR0] = &arrayObj->Color;
+   exec->array.legacy_array[VERT_ATTRIB_COLOR1] = &arrayObj->SecondaryColor;
+   exec->array.legacy_array[VERT_ATTRIB_FOG] = &arrayObj->FogCoord;
+   exec->array.legacy_array[VERT_ATTRIB_COLOR_INDEX] = &arrayObj->Index;
+   if (arrayObj->PointSize.Enabled) {
+      /* this aliases COLOR_INDEX */
+      exec->array.legacy_array[VERT_ATTRIB_POINT_SIZE] = &arrayObj->PointSize;
+   }
+   exec->array.legacy_array[VERT_ATTRIB_EDGEFLAG] = &arrayObj->EdgeFlag;
+
+   for (i = 0; i < Elements(arrayObj->TexCoord); i++)
+      exec->array.legacy_array[VERT_ATTRIB_TEX0 + i] = &arrayObj->TexCoord[i];
+
+   for (i = 0; i < Elements(arrayObj->VertexAttrib); i++) {
+      assert(i < Elements(exec->array.generic_array));
+      exec->array.generic_array[i] = &arrayObj->VertexAttrib[i];
+   }
+   
+   exec->array.array_obj = arrayObj->Name;
+}
+
+
+static void
+recalculate_input_bindings(GLcontext *ctx)
+{
+   struct vbo_context *vbo = vbo_context(ctx);
+   struct vbo_exec_context *exec = &vbo->exec;
+   const struct gl_client_array **inputs = &exec->array.inputs[0];
+   GLbitfield const_inputs = 0x0;
+   GLuint i;
+
+   exec->array.program_mode = get_program_mode(ctx);
+   exec->array.enabled_flags = ctx->Array.ArrayObj->_Enabled;
+
+   switch (exec->array.program_mode) {
+   case VP_NONE:
+      /* When no vertex program is active (or the vertex program is generated
+       * from fixed-function state).  We put the material values into the
+       * generic slots.  This is the only situation where material values
+       * are available as per-vertex attributes.
+       */
+      for (i = 0; i <= VERT_ATTRIB_TEX7; i++) {
+	 if (exec->array.legacy_array[i]->Enabled)
+	    inputs[i] = exec->array.legacy_array[i];
+	 else {
+	    inputs[i] = &vbo->legacy_currval[i];
+            const_inputs |= 1 << i;
+         }
+      }
+
+      for (i = 0; i < MAT_ATTRIB_MAX; i++) {
+	 inputs[VERT_ATTRIB_GENERIC0 + i] = &vbo->mat_currval[i];
+         const_inputs |= 1 << (VERT_ATTRIB_GENERIC0 + i);
+      }
+
+      /* Could use just about anything, just to fill in the empty
+       * slots:
+       */
+      for (i = MAT_ATTRIB_MAX; i < VERT_ATTRIB_MAX - VERT_ATTRIB_GENERIC0; i++) {
+	 inputs[VERT_ATTRIB_GENERIC0 + i] = &vbo->generic_currval[i];
+         const_inputs |= 1 << (VERT_ATTRIB_GENERIC0 + i);
+      }
+      break;
+
+   case VP_NV:
+      /* NV_vertex_program - attribute arrays alias and override
+       * conventional, legacy arrays.  No materials, and the generic
+       * slots are vacant.
+       */
+      for (i = 0; i <= VERT_ATTRIB_TEX7; i++) {
+	 if (exec->array.generic_array[i]->Enabled)
+	    inputs[i] = exec->array.generic_array[i];
+	 else if (exec->array.legacy_array[i]->Enabled)
+	    inputs[i] = exec->array.legacy_array[i];
+	 else {
+	    inputs[i] = &vbo->legacy_currval[i];
+            const_inputs |= 1 << i;
+         }
+      }
+
+      /* Could use just about anything, just to fill in the empty
+       * slots:
+       */
+      for (i = VERT_ATTRIB_GENERIC0; i < VERT_ATTRIB_MAX; i++) {
+	 inputs[i] = &vbo->generic_currval[i - VERT_ATTRIB_GENERIC0];
+         const_inputs |= 1 << i;
+      }
+      break;
+
+   case VP_ARB:
+      /* GL_ARB_vertex_program or GLSL vertex shader - Only the generic[0]
+       * attribute array aliases and overrides the legacy position array.  
+       *
+       * Otherwise, legacy attributes available in the legacy slots,
+       * generic attributes in the generic slots and materials are not
+       * available as per-vertex attributes.
+       */
+      if (exec->array.generic_array[0]->Enabled)
+	 inputs[0] = exec->array.generic_array[0];
+      else if (exec->array.legacy_array[0]->Enabled)
+	 inputs[0] = exec->array.legacy_array[0];
+      else {
+	 inputs[0] = &vbo->legacy_currval[0];
+         const_inputs |= 1 << 0;
+      }
+
+      for (i = 1; i <= VERT_ATTRIB_TEX7; i++) {
+	 if (exec->array.legacy_array[i]->Enabled)
+	    inputs[i] = exec->array.legacy_array[i];
+	 else {
+	    inputs[i] = &vbo->legacy_currval[i];
+            const_inputs |= 1 << i;
+         }
+      }
+
+      for (i = 0; i < MAX_VERTEX_GENERIC_ATTRIBS; i++) {
+	 if (exec->array.generic_array[i]->Enabled)
+	    inputs[VERT_ATTRIB_GENERIC0 + i] = exec->array.generic_array[i];
+	 else {
+	    inputs[VERT_ATTRIB_GENERIC0 + i] = &vbo->generic_currval[i];
+            const_inputs |= 1 << (VERT_ATTRIB_GENERIC0 + i);
+         }
+
+      }
+      break;
+   }
+
+   _mesa_set_varying_vp_inputs( ctx, ~const_inputs );
+}
+
+
+static void
+bind_arrays(GLcontext *ctx)
+{
+#if 0
+   if (ctx->Array.ArrayObj.Name != exec->array.array_obj) {
+      bind_array_obj(ctx);
+      recalculate_input_bindings(ctx);
+   }
+   else if (exec->array.program_mode != get_program_mode(ctx) ||
+	    exec->array.enabled_flags != ctx->Array.ArrayObj->_Enabled) {
+      
+      recalculate_input_bindings(ctx);
+   }
+#else
+   bind_array_obj(ctx);
+   recalculate_input_bindings(ctx);
+#endif
+}
+
+
+
+/***********************************************************************
+ * API functions.
+ */
+
+static void GLAPIENTRY
+vbo_exec_DrawArrays(GLenum mode, GLint start, GLsizei count)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   struct vbo_context *vbo = vbo_context(ctx);
+   struct vbo_exec_context *exec = &vbo->exec;
+   struct _mesa_prim prim[1];
+
+   if (!_mesa_validate_DrawArrays( ctx, mode, start, count ))
+      return;
+
+   FLUSH_CURRENT( ctx, 0 );
+
+   if (ctx->NewState)
+      _mesa_update_state( ctx );
+      
+   if (!_mesa_valid_to_render(ctx, "glDrawArrays")) {
+      return;
+   }
+
+#if 0
+   check_draw_arrays_data(ctx, start, count);
+#else
+   (void) check_draw_arrays_data;
+#endif
+
+   bind_arrays( ctx );
+
+   /* Again... because we may have changed the bitmask of per-vertex varying
+    * attributes.  If we regenerate the fixed-function vertex program now
+    * we may be able to prune down the number of vertex attributes which we
+    * need in the shader.
+    */
+   if (ctx->NewState)
+      _mesa_update_state( ctx );
+
+   prim[0].begin = 1;
+   prim[0].end = 1;
+   prim[0].weak = 0;
+   prim[0].pad = 0;
+   prim[0].mode = mode;
+   prim[0].start = start;
+   prim[0].count = count;
+   prim[0].indexed = 0;
+
+   vbo->draw_prims( ctx, exec->array.inputs, prim, 1, NULL,
+                    GL_TRUE, start, start + count - 1 );
+
+#if 0
+   print_draw_arrays(ctx, exec, mode, start, count);
+#else
+   (void) print_draw_arrays;
+#endif
+}
+
+
+/**
+ * Map GL_ELEMENT_ARRAY_BUFFER and print contents.
+ */
+static void
+dump_element_buffer(GLcontext *ctx, GLenum type)
+{
+   const GLvoid *map = ctx->Driver.MapBuffer(ctx,
+                                             GL_ELEMENT_ARRAY_BUFFER_ARB,
+                                             GL_READ_ONLY,
+                                             ctx->Array.ElementArrayBufferObj);
+   switch (type) {
+   case GL_UNSIGNED_BYTE:
+      {
+         const GLubyte *us = (const GLubyte *) map;
+         GLuint i;
+         for (i = 0; i < ctx->Array.ElementArrayBufferObj->Size; i++) {
+            _mesa_printf("%02x ", us[i]);
+            if (i % 32 == 31)
+               _mesa_printf("\n");
+         }
+         _mesa_printf("\n");
+      }
+      break;
+   case GL_UNSIGNED_SHORT:
+      {
+         const GLushort *us = (const GLushort *) map;
+         GLuint i;
+         for (i = 0; i < ctx->Array.ElementArrayBufferObj->Size / 2; i++) {
+            _mesa_printf("%04x ", us[i]);
+            if (i % 16 == 15)
+               _mesa_printf("\n");
+         }
+         _mesa_printf("\n");
+      }
+      break;
+   case GL_UNSIGNED_INT:
+      {
+         const GLuint *us = (const GLuint *) map;
+         GLuint i;
+         for (i = 0; i < ctx->Array.ElementArrayBufferObj->Size / 4; i++) {
+            _mesa_printf("%08x ", us[i]);
+            if (i % 8 == 7)
+               _mesa_printf("\n");
+         }
+         _mesa_printf("\n");
+      }
+      break;
+   default:
+      ;
+   }
+
+   ctx->Driver.UnmapBuffer(ctx,
+                           GL_ELEMENT_ARRAY_BUFFER_ARB,
+                           ctx->Array.ElementArrayBufferObj);
+}
+
+/* Inner support for both _mesa_DrawElements and _mesa_DrawRangeElements */
+static void
+vbo_validated_drawrangeelements(GLcontext *ctx, GLenum mode,
+				GLboolean index_bounds_valid,
+				GLuint start, GLuint end,
+				GLsizei count, GLenum type,
+				const GLvoid *indices)
+{
+   struct vbo_context *vbo = vbo_context(ctx);
+   struct vbo_exec_context *exec = &vbo->exec;
+   struct _mesa_index_buffer ib;
+   struct _mesa_prim prim[1];
+
+   FLUSH_CURRENT( ctx, 0 );
+
+   if (ctx->NewState)
+      _mesa_update_state( ctx );
+
+   if (!_mesa_valid_to_render(ctx, "glDraw[Range]Elements")) {
+      return;
+   }
+
+   if (ctx->NewState)
+      _mesa_update_state( ctx );
+
+   bind_arrays( ctx );
+
+   ib.count = count;
+   ib.type = type;
+   ib.obj = ctx->Array.ElementArrayBufferObj;
+   ib.ptr = indices;
+
+   prim[0].begin = 1;
+   prim[0].end = 1;
+   prim[0].weak = 0;
+   prim[0].pad = 0;
+   prim[0].mode = mode;
+   prim[0].start = 0;
+   prim[0].count = count;
+   prim[0].indexed = 1;
+
+   /* Need to give special consideration to rendering a range of
+    * indices starting somewhere above zero.  Typically the
+    * application is issuing multiple DrawRangeElements() to draw
+    * successive primitives layed out linearly in the vertex arrays.
+    * Unless the vertex arrays are all in a VBO (or locked as with
+    * CVA), the OpenGL semantics imply that we need to re-read or
+    * re-upload the vertex data on each draw call.  
+    *
+    * In the case of hardware tnl, we want to avoid starting the
+    * upload at zero, as it will mean every draw call uploads an
+    * increasing amount of not-used vertex data.  Worse - in the
+    * software tnl module, all those vertices might be transformed and
+    * lit but never rendered.
+    *
+    * If we just upload or transform the vertices in start..end,
+    * however, the indices will be incorrect.
+    *
+    * At this level, we don't know exactly what the requirements of
+    * the backend are going to be, though it will likely boil down to
+    * either:
+    *
+    * 1) Do nothing, everything is in a VBO and is processed once
+    *       only.
+    *
+    * 2) Adjust the indices and vertex arrays so that start becomes
+    *    zero.
+    *
+    * Rather than doing anything here, I'll provide a helper function
+    * for the latter case elsewhere.
+    */
+
+   vbo->draw_prims( ctx, exec->array.inputs, prim, 1, &ib,
+		    index_bounds_valid, start, end );
+}
+
+static void GLAPIENTRY
+vbo_exec_DrawRangeElements(GLenum mode,
+			   GLuint start, GLuint end,
+			   GLsizei count, GLenum type, const GLvoid *indices)
+{
+   static GLuint warnCount = 0;
+   GET_CURRENT_CONTEXT(ctx);
+
+   if (!_mesa_validate_DrawRangeElements( ctx, mode, start, end, count,
+                                          type, indices ))
+      return;
+
+   if (end >= ctx->Array.ArrayObj->_MaxElement) {
+      /* the max element is out of bounds of one or more enabled arrays */
+      warnCount++;
+
+      if (warnCount < 10) {
+         _mesa_warning(ctx, "glDraw[Range]Elements(start %u, end %u, count %d, "
+                       "type 0x%x, indices=%p)\n"
+                       "\tend is out of bounds (max=%u)  "
+                       "Element Buffer %u (size %d)\n"
+                       "\tThis should probably be fixed in the application.",
+                       start, end, count, type, indices,
+                       ctx->Array.ArrayObj->_MaxElement - 1,
+                       ctx->Array.ElementArrayBufferObj->Name,
+                       ctx->Array.ElementArrayBufferObj->Size);
+      }
+
+      if (0)
+         dump_element_buffer(ctx, type);
+
+      if (0)
+         _mesa_print_arrays(ctx);
+
+#ifdef DEBUG
+      /* 'end' was out of bounds, but now let's check the actual array
+       * indexes to see if any of them are out of bounds.  If so, warn
+       * and skip the draw to avoid potential segfault, etc.
+       */
+      {
+         GLuint max = _mesa_max_buffer_index(ctx, count, type, indices,
+                                             ctx->Array.ElementArrayBufferObj);
+         if (max >= ctx->Array.ArrayObj->_MaxElement) {
+            if (warnCount < 10) {
+               _mesa_warning(ctx, "glDraw[Range]Elements(start %u, end %u, "
+                             "count %d, type 0x%x, indices=%p)\n"
+                             "\tindex=%u is out of bounds (max=%u)  "
+                             "Element Buffer %u (size %d)\n"
+                             "\tSkipping the glDrawRangeElements() call",
+                             start, end, count, type, indices, max,
+                             ctx->Array.ArrayObj->_MaxElement - 1,
+                             ctx->Array.ElementArrayBufferObj->Name,
+                             ctx->Array.ElementArrayBufferObj->Size);
+            }
+            return;
+         }
+         /* XXX we could also find the min index and compare to 'start'
+          * to see if start is correct.  But it's more likely to get the
+          * upper bound wrong.
+          */
+      }
+#endif
+   }
+   else if (0) {
+      _mesa_printf("glDraw[Range]Elements"
+                   "(start %u, end %u, type 0x%x, count %d) ElemBuf %u\n",
+                   start, end, type, count,
+                   ctx->Array.ElementArrayBufferObj->Name);
+   }
+
+#if 0
+   check_draw_elements_data(ctx, count, type, indices);
+#else
+   (void) check_draw_elements_data;
+#endif
+
+   vbo_validated_drawrangeelements(ctx, mode, GL_TRUE, start, end,
+				   count, type, indices);
+}
+
+
+static void GLAPIENTRY
+vbo_exec_DrawElements(GLenum mode, GLsizei count, GLenum type,
+                      const GLvoid *indices)
+{
+   GET_CURRENT_CONTEXT(ctx);
+
+   if (!_mesa_validate_DrawElements( ctx, mode, count, type, indices ))
+      return;
+
+   vbo_validated_drawrangeelements(ctx, mode, GL_FALSE, ~0, ~0,
+				   count, type, indices);
+}
+
+/* Inner support for both _mesa_DrawElements and _mesa_DrawRangeElements */
+static void
+vbo_validated_multidrawelements(GLcontext *ctx, GLenum mode,
+				const GLsizei *count, GLenum type,
+				const GLvoid **indices, GLsizei primcount)
+{
+   struct vbo_context *vbo = vbo_context(ctx);
+   struct vbo_exec_context *exec = &vbo->exec;
+   struct _mesa_index_buffer ib;
+   struct _mesa_prim *prim;
+   unsigned int index_type_size = 0;
+   uintptr_t min_index_ptr, max_index_ptr;
+   GLboolean fallback = GL_FALSE;
+   int i;
+
+   if (primcount == 0)
+      return;
+
+   FLUSH_CURRENT( ctx, 0 );
+
+   if (ctx->NewState)
+      _mesa_update_state( ctx );
+
+   if (!_mesa_valid_to_render(ctx, "glMultiDrawElements")) {
+      return;
+   }
+
+   if (ctx->NewState)
+      _mesa_update_state( ctx );
+
+   prim = _mesa_calloc(primcount * sizeof(*prim));
+   if (prim == NULL) {
+      _mesa_error(ctx, GL_OUT_OF_MEMORY, "glMultiDrawElements");
+      return;
+   }
+
+   /* Decide if we can do this all as one set of primitives sharing the
+    * same index buffer, or if we have to reset the index pointer per primitive.
+    */
+   bind_arrays( ctx );
+
+   switch (type) {
+   case GL_UNSIGNED_INT:
+      index_type_size = 4;
+      break;
+   case GL_UNSIGNED_SHORT:
+      index_type_size = 2;
+      break;
+   case GL_UNSIGNED_BYTE:
+      index_type_size = 1;
+      break;
+   default:
+      assert(0);
+   }
+
+   min_index_ptr = (uintptr_t)indices[0];
+   max_index_ptr = 0;
+   for (i = 0; i < primcount; i++) {
+      min_index_ptr = MIN2(min_index_ptr, (uintptr_t)indices[i]);
+      max_index_ptr = MAX2(max_index_ptr, (uintptr_t)indices[i] +
+			   index_type_size * count[i]);
+   }
+
+   /* Check if we can handle this thing as a bunch of index offsets from the
+    * same index pointer.  If we can't, then we have to fall back to doing
+    * a draw_prims per primitive.
+    */
+   if (index_type_size != 1) {
+      for (i = 0; i < primcount; i++) {
+	 if ((((uintptr_t)indices[i] - min_index_ptr) % index_type_size) != 0) {
+	    fallback = GL_TRUE;
+	    break;
+	 }
+      }
+   }
+
+   /* If the index buffer isn't in a VBO, then treating the application's
+    * subranges of the index buffer as one large index buffer may lead to
+    * us reading unmapped memory.
+    */
+   if (!_mesa_is_bufferobj(ctx->Array.ElementArrayBufferObj))
+      fallback = GL_TRUE;
+
+   if (!fallback) {
+      ib.count = (max_index_ptr - min_index_ptr) / index_type_size;
+      ib.type = type;
+      ib.obj = ctx->Array.ElementArrayBufferObj;
+      ib.ptr = (void *)min_index_ptr;
+
+      for (i = 0; i < primcount; i++) {
+	 prim[i].begin = (i == 0);
+	 prim[i].end = (i == primcount - 1);
+	 prim[i].weak = 0;
+	 prim[i].pad = 0;
+	 prim[i].mode = mode;
+	 prim[i].start = ((uintptr_t)indices[i] - min_index_ptr) / index_type_size;
+	 prim[i].count = count[i];
+	 prim[i].indexed = 1;
+      }
+
+      vbo->draw_prims(ctx, exec->array.inputs, prim, primcount, &ib,
+		      GL_FALSE, ~0, ~0);
+   } else {
+      for (i = 0; i < primcount; i++) {
+	 ib.count = count[i];
+	 ib.type = type;
+	 ib.obj = ctx->Array.ElementArrayBufferObj;
+	 ib.ptr = indices[i];
+
+
+	 prim[0].begin = 1;
+	 prim[0].end = 1;
+	 prim[0].weak = 0;
+	 prim[0].pad = 0;
+	 prim[0].mode = mode;
+	 prim[0].start = 0;
+	 prim[0].count = count[i];
+	 prim[0].indexed = 1;
+      }
+
+      vbo->draw_prims(ctx, exec->array.inputs, prim, 1, &ib,
+		      GL_FALSE, ~0, ~0);
+   }
+   _mesa_free(prim);
+}
+
+static void GLAPIENTRY
+vbo_exec_MultiDrawElements(GLenum mode,
+			   const GLsizei *count, GLenum type,
+			   const GLvoid **indices,
+			   GLsizei primcount)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   GLint i;
+
+   ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx);
+
+   for (i = 0; i < primcount; i++) {
+      if (!_mesa_validate_DrawElements( ctx, mode, count[i], type, indices[i] ))
+	 return;
+   }
+
+   vbo_validated_multidrawelements(ctx, mode, count, type, indices, primcount);
+}
+
+
+
+/***********************************************************************
+ * Initialization
+ */
+
+void
+vbo_exec_array_init( struct vbo_exec_context *exec )
+{
+#if 1
+   exec->vtxfmt.DrawArrays = vbo_exec_DrawArrays;
+   exec->vtxfmt.DrawElements = vbo_exec_DrawElements;
+   exec->vtxfmt.DrawRangeElements = vbo_exec_DrawRangeElements;
+   exec->vtxfmt.MultiDrawElementsEXT = vbo_exec_MultiDrawElements;
+#else
+   exec->vtxfmt.DrawArrays = _mesa_noop_DrawArrays;
+   exec->vtxfmt.DrawElements = _mesa_noop_DrawElements;
+   exec->vtxfmt.DrawRangeElements = _mesa_noop_DrawRangeElements;
+   exec->vtxfmt.MultiDrawElementsEXT = _mesa_noop_MultiDrawElements;
+#endif
+}
+
+
+void
+vbo_exec_array_destroy( struct vbo_exec_context *exec )
+{
+   /* nothing to do */
+}
+
+
+/* This API entrypoint is not ordinarily used */
+void GLAPIENTRY
+_mesa_DrawArrays(GLenum mode, GLint first, GLsizei count)
+{
+   vbo_exec_DrawArrays(mode, first, count);
+}
+
+
+/* This API entrypoint is not ordinarily used */
+void GLAPIENTRY
+_mesa_DrawElements(GLenum mode, GLsizei count, GLenum type,
+                   const GLvoid *indices)
+{
+   vbo_exec_DrawElements(mode, count, type, indices);
+}
+
+
+/* This API entrypoint is not ordinarily used */
+void GLAPIENTRY
+_mesa_DrawRangeElements(GLenum mode, GLuint start, GLuint end, GLsizei count,
+                        GLenum type, const GLvoid *indices)
+{
+   vbo_exec_DrawRangeElements(mode, start, end, count, type, indices);
+}
+
+/* GL_EXT_multi_draw_arrays */
+void GLAPIENTRY
+_mesa_MultiDrawElementsEXT(GLenum mode, const GLsizei *count, GLenum type,
+			   const GLvoid **indices, GLsizei primcount)
+{
+   vbo_exec_MultiDrawElements(mode, count, type, indices, primcount);
+}
diff --git a/mesalib/src/mesa/vbo/vbo_exec_draw.c b/mesalib/src/mesa/vbo/vbo_exec_draw.c
new file mode 100644
index 000000000..0c258c535
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_exec_draw.c
@@ -0,0 +1,413 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.2
+ *
+ * Copyright (C) 1999-2008  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "main/glheader.h"
+#include "main/bufferobj.h"
+#include "main/context.h"
+#include "main/enums.h"
+#include "main/state.h"
+#include "main/macros.h"
+
+#include "vbo_context.h"
+
+
+static void
+vbo_exec_debug_verts( struct vbo_exec_context *exec )
+{
+   GLuint count = exec->vtx.vert_count;
+   GLuint i;
+
+   _mesa_printf("%s: %u vertices %d primitives, %d vertsize\n",
+		__FUNCTION__,
+		count,
+		exec->vtx.prim_count,
+		exec->vtx.vertex_size);
+
+   for (i = 0 ; i < exec->vtx.prim_count ; i++) {
+      struct _mesa_prim *prim = &exec->vtx.prim[i];
+      _mesa_printf("   prim %d: %s%s %d..%d %s %s\n",
+		   i, 
+		   _mesa_lookup_prim_by_nr(prim->mode),
+		   prim->weak ? " (weak)" : "",
+		   prim->start, 
+		   prim->start + prim->count,
+		   prim->begin ? "BEGIN" : "(wrap)",
+		   prim->end ? "END" : "(wrap)");
+   }
+}
+
+
+/*
+ * NOTE: Need to have calculated primitives by this point -- do it on the fly.
+ * NOTE: Old 'parity' issue is gone.
+ */
+static GLuint
+vbo_copy_vertices( struct vbo_exec_context *exec )
+{
+   GLuint nr = exec->vtx.prim[exec->vtx.prim_count-1].count;
+   GLuint ovf, i;
+   GLuint sz = exec->vtx.vertex_size;
+   GLfloat *dst = exec->vtx.copied.buffer;
+   const GLfloat *src = (exec->vtx.buffer_map + 
+                         exec->vtx.prim[exec->vtx.prim_count-1].start * 
+                         exec->vtx.vertex_size);
+
+
+   switch (exec->ctx->Driver.CurrentExecPrimitive) {
+   case GL_POINTS:
+      return 0;
+   case GL_LINES:
+      ovf = nr&1;
+      for (i = 0 ; i < ovf ; i++)
+	 _mesa_memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz * sizeof(GLfloat) );
+      return i;
+   case GL_TRIANGLES:
+      ovf = nr%3;
+      for (i = 0 ; i < ovf ; i++)
+	 _mesa_memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz * sizeof(GLfloat) );
+      return i;
+   case GL_QUADS:
+      ovf = nr&3;
+      for (i = 0 ; i < ovf ; i++)
+	 _mesa_memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz * sizeof(GLfloat) );
+      return i;
+   case GL_LINE_STRIP:
+      if (nr == 0) {
+	 return 0;
+      }
+      else {
+	 _mesa_memcpy( dst, src+(nr-1)*sz, sz * sizeof(GLfloat) );
+	 return 1;
+      }
+   case GL_LINE_LOOP:
+   case GL_TRIANGLE_FAN:
+   case GL_POLYGON:
+      if (nr == 0) {
+	 return 0;
+      }
+      else if (nr == 1) {
+	 _mesa_memcpy( dst, src+0, sz * sizeof(GLfloat) );
+	 return 1;
+      }
+      else {
+	 _mesa_memcpy( dst, src+0, sz * sizeof(GLfloat) );
+	 _mesa_memcpy( dst+sz, src+(nr-1)*sz, sz * sizeof(GLfloat) );
+	 return 2;
+      }
+   case GL_TRIANGLE_STRIP:
+      /* no parity issue, but need to make sure the tri is not drawn twice */
+      if (nr & 1) {
+	 exec->vtx.prim[exec->vtx.prim_count-1].count--;
+      }
+      /* fallthrough */
+   case GL_QUAD_STRIP:
+      switch (nr) {
+      case 0:
+         ovf = 0;
+         break;
+      case 1:
+         ovf = 1;
+         break;
+      default:
+         ovf = 2 + (nr & 1);
+         break;
+      }
+      for (i = 0 ; i < ovf ; i++)
+	 _mesa_memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz * sizeof(GLfloat) );
+      return i;
+   case PRIM_OUTSIDE_BEGIN_END:
+      return 0;
+   default:
+      assert(0);
+      return 0;
+   }
+}
+
+
+
+/* TODO: populate these as the vertex is defined:
+ */
+static void
+vbo_exec_bind_arrays( GLcontext *ctx )
+{
+   struct vbo_context *vbo = vbo_context(ctx);
+   struct vbo_exec_context *exec = &vbo->exec;
+   struct gl_client_array *arrays = exec->vtx.arrays;
+   const GLuint count = exec->vtx.vert_count;
+   const GLubyte *data = (GLubyte *) exec->vtx.buffer_map;
+   const GLuint *map;
+   GLuint attr;
+   GLbitfield varying_inputs = 0x0;
+
+   /* Install the default (ie Current) attributes first, then overlay
+    * all active ones.
+    */
+   switch (get_program_mode(exec->ctx)) {
+   case VP_NONE:
+      for (attr = 0; attr < 16; attr++) {
+         exec->vtx.inputs[attr] = &vbo->legacy_currval[attr];
+      }
+      for (attr = 0; attr < MAT_ATTRIB_MAX; attr++) {
+         exec->vtx.inputs[attr + 16] = &vbo->mat_currval[attr];
+      }
+      map = vbo->map_vp_none;
+      break;
+   case VP_NV:
+   case VP_ARB:
+      /* The aliasing of attributes for NV vertex programs has already
+       * occurred.  NV vertex programs cannot access material values,
+       * nor attributes greater than VERT_ATTRIB_TEX7.  
+       */
+      for (attr = 0; attr < 16; attr++) {
+         exec->vtx.inputs[attr] = &vbo->legacy_currval[attr];
+         exec->vtx.inputs[attr + 16] = &vbo->generic_currval[attr];
+      }
+      map = vbo->map_vp_arb;
+
+      /* check if VERT_ATTRIB_POS is not read but VERT_BIT_GENERIC0 is read.
+       * In that case we effectively need to route the data from
+       * glVertexAttrib(0, val) calls to feed into the GENERIC0 input.
+       */
+      if ((ctx->VertexProgram._Current->Base.InputsRead & VERT_BIT_POS) == 0 &&
+          (ctx->VertexProgram._Current->Base.InputsRead & VERT_BIT_GENERIC0)) {
+         exec->vtx.inputs[16] = exec->vtx.inputs[0];
+         exec->vtx.attrsz[16] = exec->vtx.attrsz[0];
+         exec->vtx.attrptr[16] = exec->vtx.attrptr[0];
+         exec->vtx.attrsz[0] = 0;
+      }
+      break;
+   default:
+      assert(0);
+   }
+
+   /* Make all active attributes (including edgeflag) available as
+    * arrays of floats.
+    */
+   for (attr = 0; attr < VERT_ATTRIB_MAX ; attr++) {
+      const GLuint src = map[attr];
+
+      if (exec->vtx.attrsz[src]) {
+         /* override the default array set above */
+         exec->vtx.inputs[attr] = &arrays[attr];
+
+         if (_mesa_is_bufferobj(exec->vtx.bufferobj)) {
+            /* a real buffer obj: Ptr is an offset, not a pointer*/
+            GLsizeiptr offset;
+            assert(exec->vtx.bufferobj->Pointer);  /* buf should be mapped */
+            offset = (GLbyte *) data -
+	       (GLbyte *) exec->vtx.bufferobj->Pointer +
+	       exec->vtx.bufferobj->Offset;
+            assert(offset >= 0);
+            arrays[attr].Ptr = (void *) offset;
+         }
+         else {
+            /* Ptr into ordinary app memory */
+            arrays[attr].Ptr = (void *) data;
+         }
+	 arrays[attr].Size = exec->vtx.attrsz[src];
+	 arrays[attr].StrideB = exec->vtx.vertex_size * sizeof(GLfloat);
+	 arrays[attr].Stride = exec->vtx.vertex_size * sizeof(GLfloat);
+	 arrays[attr].Type = GL_FLOAT;
+         arrays[attr].Format = GL_RGBA;
+	 arrays[attr].Enabled = 1;
+         _mesa_reference_buffer_object(ctx,
+                                       &arrays[attr].BufferObj,
+                                       exec->vtx.bufferobj);
+	 arrays[attr]._MaxElement = count; /* ??? */
+
+	 data += exec->vtx.attrsz[src] * sizeof(GLfloat);
+         varying_inputs |= 1 << attr;
+      }
+   }
+
+   _mesa_set_varying_vp_inputs( ctx, varying_inputs );
+}
+
+
+static void
+vbo_exec_vtx_unmap( struct vbo_exec_context *exec )
+{
+   GLenum target = GL_ARRAY_BUFFER_ARB;
+
+   if (_mesa_is_bufferobj(exec->vtx.bufferobj)) {
+      GLcontext *ctx = exec->ctx;
+      
+      if (ctx->Driver.FlushMappedBufferRange) {
+         GLintptr offset = exec->vtx.buffer_used - exec->vtx.bufferobj->Offset;
+         GLsizeiptr length = (exec->vtx.buffer_ptr - exec->vtx.buffer_map) * sizeof(float);
+
+         if (length)
+            ctx->Driver.FlushMappedBufferRange(ctx, target,
+                                               offset, length,
+                                               exec->vtx.bufferobj);
+      }
+
+      exec->vtx.buffer_used += (exec->vtx.buffer_ptr -
+                                exec->vtx.buffer_map) * sizeof(float);
+
+      assert(exec->vtx.buffer_used <= VBO_VERT_BUFFER_SIZE);
+      assert(exec->vtx.buffer_ptr != NULL);
+      
+      ctx->Driver.UnmapBuffer(ctx, target, exec->vtx.bufferobj);
+      exec->vtx.buffer_map = NULL;
+      exec->vtx.buffer_ptr = NULL;
+      exec->vtx.max_vert = 0;
+   }
+}
+
+
+void
+vbo_exec_vtx_map( struct vbo_exec_context *exec )
+{
+   GLcontext *ctx = exec->ctx;
+   const GLenum target = GL_ARRAY_BUFFER_ARB;
+   const GLenum access = GL_READ_WRITE_ARB; /* for MapBuffer */
+   const GLenum accessRange = GL_MAP_WRITE_BIT |  /* for MapBufferRange */
+                              GL_MAP_INVALIDATE_RANGE_BIT |
+                              GL_MAP_UNSYNCHRONIZED_BIT |
+                              GL_MAP_FLUSH_EXPLICIT_BIT |
+                              MESA_MAP_NOWAIT_BIT;
+   const GLenum usage = GL_STREAM_DRAW_ARB;
+   
+   if (!_mesa_is_bufferobj(exec->vtx.bufferobj))
+      return;
+
+   if (exec->vtx.buffer_map != NULL) {
+      assert(0);
+      exec->vtx.buffer_map = NULL;
+      exec->vtx.buffer_ptr = NULL;
+   }
+
+   if (VBO_VERT_BUFFER_SIZE > exec->vtx.buffer_used + 1024 &&
+       ctx->Driver.MapBufferRange) {
+      exec->vtx.buffer_map = 
+         (GLfloat *)ctx->Driver.MapBufferRange(ctx, 
+                                               target, 
+                                               exec->vtx.buffer_used,
+                                               (VBO_VERT_BUFFER_SIZE - 
+                                                exec->vtx.buffer_used),
+                                               accessRange,
+                                               exec->vtx.bufferobj);
+      exec->vtx.buffer_ptr = exec->vtx.buffer_map;
+   }
+   
+   if (!exec->vtx.buffer_map) {
+      exec->vtx.buffer_used = 0;
+
+      ctx->Driver.BufferData(ctx, target, 
+                             VBO_VERT_BUFFER_SIZE, 
+                             NULL, usage, exec->vtx.bufferobj);
+
+
+      if (ctx->Driver.MapBufferRange)
+         exec->vtx.buffer_map = 
+            (GLfloat *)ctx->Driver.MapBufferRange(ctx, target,
+                                                  0, VBO_VERT_BUFFER_SIZE,
+                                                  accessRange,
+                                                  exec->vtx.bufferobj);
+      if (!exec->vtx.buffer_map)
+         exec->vtx.buffer_map =
+            (GLfloat *)ctx->Driver.MapBuffer(ctx, target, access, exec->vtx.bufferobj);
+      assert(exec->vtx.buffer_map);
+      exec->vtx.buffer_ptr = exec->vtx.buffer_map;
+   }
+
+   if (0)
+      _mesa_printf("map %d..\n", exec->vtx.buffer_used);
+}
+
+
+
+/**
+ * Execute the buffer and save copied verts.
+ */
+void
+vbo_exec_vtx_flush( struct vbo_exec_context *exec, GLboolean unmap )
+{
+   if (0)
+      vbo_exec_debug_verts( exec );
+
+   if (exec->vtx.prim_count && 
+       exec->vtx.vert_count) {
+
+      exec->vtx.copied.nr = vbo_copy_vertices( exec ); 
+
+      if (exec->vtx.copied.nr != exec->vtx.vert_count) {
+	 GLcontext *ctx = exec->ctx;
+	 
+	 /* Before the update_state() as this may raise _NEW_ARRAY
+          * from _mesa_set_varying_vp_inputs().
+	  */
+	 vbo_exec_bind_arrays( ctx );
+
+         if (ctx->NewState)
+            _mesa_update_state( ctx );
+
+         if (_mesa_is_bufferobj(exec->vtx.bufferobj)) {
+            vbo_exec_vtx_unmap( exec );
+         }
+
+         if (0)
+            _mesa_printf("%s %d %d\n", __FUNCTION__, exec->vtx.prim_count,
+                         exec->vtx.vert_count);
+
+	 vbo_context(ctx)->draw_prims( ctx, 
+				       exec->vtx.inputs, 
+				       exec->vtx.prim, 
+				       exec->vtx.prim_count,
+				       NULL,
+				       GL_TRUE,
+				       0,
+				       exec->vtx.vert_count - 1);
+
+	 /* If using a real VBO, get new storage -- unless asked not to.
+          */
+         if (_mesa_is_bufferobj(exec->vtx.bufferobj) && !unmap) {
+            vbo_exec_vtx_map( exec );
+         }
+      }
+   }
+
+   /* May have to unmap explicitly if we didn't draw:
+    */
+   if (unmap && 
+       _mesa_is_bufferobj(exec->vtx.bufferobj) &&
+       exec->vtx.buffer_map) {
+      vbo_exec_vtx_unmap( exec );
+   }
+
+
+   if (unmap || exec->vtx.vertex_size == 0)
+      exec->vtx.max_vert = 0;
+   else
+      exec->vtx.max_vert = ((VBO_VERT_BUFFER_SIZE - exec->vtx.buffer_used) / 
+                            (exec->vtx.vertex_size * sizeof(GLfloat)));
+
+   exec->vtx.buffer_ptr = exec->vtx.buffer_map;
+   exec->vtx.prim_count = 0;
+   exec->vtx.vert_count = 0;
+}
diff --git a/mesalib/src/mesa/vbo/vbo_exec_eval.c b/mesalib/src/mesa/vbo/vbo_exec_eval.c
new file mode 100644
index 000000000..0c691b3a5
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_exec_eval.c
@@ -0,0 +1,254 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  6.1
+ *
+ * Copyright (C) 1999-2004  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "main/glheader.h"
+#include "main/context.h"
+#include "main/macros.h"
+#include "math/m_eval.h"
+#include "glapi/dispatch.h"
+#include "vbo_exec.h"
+
+
+static void clear_active_eval1( struct vbo_exec_context *exec, GLuint attr ) 
+{
+   exec->eval.map1[attr].map = NULL;
+}
+
+static void clear_active_eval2( struct vbo_exec_context *exec, GLuint attr ) 
+{
+   exec->eval.map2[attr].map = NULL;
+}
+
+static void set_active_eval1( struct vbo_exec_context *exec, GLuint attr, GLuint dim, 
+			      struct gl_1d_map *map )
+{
+   if (!exec->eval.map1[attr].map) {
+      exec->eval.map1[attr].map = map;
+      exec->eval.map1[attr].sz = dim;
+   }
+} 
+
+static void set_active_eval2( struct vbo_exec_context *exec, GLuint attr, GLuint dim, 
+			      struct gl_2d_map *map )
+{
+   if (!exec->eval.map2[attr].map) {
+      exec->eval.map2[attr].map = map;
+      exec->eval.map2[attr].sz = dim;
+   }
+} 
+
+void vbo_exec_eval_update( struct vbo_exec_context *exec )
+{
+   GLcontext *ctx = exec->ctx;
+   GLuint attr;
+
+   /* Vertex program maps have priority over conventional attribs */
+
+   for (attr = 0; attr < VBO_ATTRIB_FIRST_MATERIAL; attr++) {
+      clear_active_eval1( exec, attr );
+      clear_active_eval2( exec, attr );
+   }
+
+   /* _NEW_PROGRAM */
+   if (ctx->VertexProgram._Enabled) {
+      for (attr = 0; attr < VBO_ATTRIB_FIRST_MATERIAL; attr++) {
+	 /* _NEW_EVAL */
+	 if (ctx->Eval.Map1Attrib[attr]) 
+	    set_active_eval1( exec, attr, 4, &ctx->EvalMap.Map1Attrib[attr] );
+
+	 if (ctx->Eval.Map2Attrib[attr]) 
+	    set_active_eval2( exec, attr, 4, &ctx->EvalMap.Map2Attrib[attr] );
+      }
+   }
+
+   if (ctx->Eval.Map1Color4) 
+      set_active_eval1( exec, VBO_ATTRIB_COLOR0, 4, &ctx->EvalMap.Map1Color4 );
+      
+   if (ctx->Eval.Map2Color4) 
+      set_active_eval2( exec, VBO_ATTRIB_COLOR0, 4, &ctx->EvalMap.Map2Color4 );
+
+   if (ctx->Eval.Map1TextureCoord4) 
+      set_active_eval1( exec, VBO_ATTRIB_TEX0, 4, &ctx->EvalMap.Map1Texture4 );
+   else if (ctx->Eval.Map1TextureCoord3) 
+      set_active_eval1( exec, VBO_ATTRIB_TEX0, 3, &ctx->EvalMap.Map1Texture3 );
+   else if (ctx->Eval.Map1TextureCoord2) 
+      set_active_eval1( exec, VBO_ATTRIB_TEX0, 2, &ctx->EvalMap.Map1Texture2 );
+   else if (ctx->Eval.Map1TextureCoord1) 
+      set_active_eval1( exec, VBO_ATTRIB_TEX0, 1, &ctx->EvalMap.Map1Texture1 );
+
+   if (ctx->Eval.Map2TextureCoord4) 
+      set_active_eval2( exec, VBO_ATTRIB_TEX0, 4, &ctx->EvalMap.Map2Texture4 );
+   else if (ctx->Eval.Map2TextureCoord3) 
+      set_active_eval2( exec, VBO_ATTRIB_TEX0, 3, &ctx->EvalMap.Map2Texture3 );
+   else if (ctx->Eval.Map2TextureCoord2) 
+      set_active_eval2( exec, VBO_ATTRIB_TEX0, 2, &ctx->EvalMap.Map2Texture2 );
+   else if (ctx->Eval.Map2TextureCoord1) 
+      set_active_eval2( exec, VBO_ATTRIB_TEX0, 1, &ctx->EvalMap.Map2Texture1 );
+
+   if (ctx->Eval.Map1Normal) 
+      set_active_eval1( exec, VBO_ATTRIB_NORMAL, 3, &ctx->EvalMap.Map1Normal );
+
+   if (ctx->Eval.Map2Normal) 
+      set_active_eval2( exec, VBO_ATTRIB_NORMAL, 3, &ctx->EvalMap.Map2Normal );
+
+   if (ctx->Eval.Map1Vertex4) 
+      set_active_eval1( exec, VBO_ATTRIB_POS, 4, &ctx->EvalMap.Map1Vertex4 );
+   else if (ctx->Eval.Map1Vertex3) 
+      set_active_eval1( exec, VBO_ATTRIB_POS, 3, &ctx->EvalMap.Map1Vertex3 );
+
+   if (ctx->Eval.Map2Vertex4) 
+      set_active_eval2( exec, VBO_ATTRIB_POS, 4, &ctx->EvalMap.Map2Vertex4 );
+   else if (ctx->Eval.Map2Vertex3) 
+      set_active_eval2( exec, VBO_ATTRIB_POS, 3, &ctx->EvalMap.Map2Vertex3 );
+
+   exec->eval.recalculate_maps = 0;
+}
+
+
+
+void vbo_exec_do_EvalCoord1f(struct vbo_exec_context *exec, GLfloat u)
+{
+   GLuint attr;
+
+   for (attr = 1; attr <= VBO_ATTRIB_TEX7; attr++) {
+      struct gl_1d_map *map = exec->eval.map1[attr].map;
+      if (map) {
+	 GLfloat uu = (u - map->u1) * map->du;
+	 GLfloat data[4];
+
+	 ASSIGN_4V(data, 0, 0, 0, 1);
+
+	 _math_horner_bezier_curve(map->Points, data, uu, 
+				   exec->eval.map1[attr].sz, 
+				   map->Order);
+
+	 COPY_SZ_4V( exec->vtx.attrptr[attr],
+		     exec->vtx.attrsz[attr],
+		     data );
+      }
+   }
+
+   /** Vertex -- EvalCoord1f is a noop if this map not enabled:
+    **/
+   if (exec->eval.map1[0].map) {
+      struct gl_1d_map *map = exec->eval.map1[0].map;
+      GLfloat uu = (u - map->u1) * map->du;
+      GLfloat vertex[4];
+
+      ASSIGN_4V(vertex, 0, 0, 0, 1);
+
+      _math_horner_bezier_curve(map->Points, vertex, uu, 
+				exec->eval.map1[0].sz, 
+				map->Order);
+
+      if (exec->eval.map1[0].sz == 4) 
+	 CALL_Vertex4fv(GET_DISPATCH(), ( vertex ));
+      else
+	 CALL_Vertex3fv(GET_DISPATCH(), ( vertex )); 
+   }
+}
+
+
+
+void vbo_exec_do_EvalCoord2f( struct vbo_exec_context *exec, 
+			      GLfloat u, GLfloat v )
+{   
+   GLuint attr;
+
+   for (attr = 1; attr <= VBO_ATTRIB_TEX7; attr++) {
+      struct gl_2d_map *map = exec->eval.map2[attr].map;
+      if (map) {
+	 GLfloat uu = (u - map->u1) * map->du;
+	 GLfloat vv = (v - map->v1) * map->dv;
+	 GLfloat data[4];
+
+	 ASSIGN_4V(data, 0, 0, 0, 1);
+
+	 _math_horner_bezier_surf(map->Points, 
+				  data, 
+				  uu, vv, 
+				  exec->eval.map2[attr].sz, 
+				  map->Uorder, map->Vorder);
+
+	 COPY_SZ_4V( exec->vtx.attrptr[attr],
+		     exec->vtx.attrsz[attr],
+		     data );
+      }
+   }
+
+   /** Vertex -- EvalCoord2f is a noop if this map not enabled:
+    **/
+   if (exec->eval.map2[0].map) {
+      struct gl_2d_map *map = exec->eval.map2[0].map;
+      GLfloat uu = (u - map->u1) * map->du;
+      GLfloat vv = (v - map->v1) * map->dv;
+      GLfloat vertex[4];
+
+      ASSIGN_4V(vertex, 0, 0, 0, 1);
+
+      if (exec->ctx->Eval.AutoNormal) {
+	 GLfloat normal[4];
+         GLfloat du[4], dv[4];
+
+         _math_de_casteljau_surf(map->Points, vertex, du, dv, uu, vv, 
+				 exec->eval.map2[0].sz,
+				 map->Uorder, map->Vorder);
+
+	 if (exec->eval.map2[0].sz == 4) {
+	    du[0] = du[0]*vertex[3] - du[3]*vertex[0];
+	    du[1] = du[1]*vertex[3] - du[3]*vertex[1];
+	    du[2] = du[2]*vertex[3] - du[3]*vertex[2];
+	 
+	    dv[0] = dv[0]*vertex[3] - dv[3]*vertex[0];
+	    dv[1] = dv[1]*vertex[3] - dv[3]*vertex[1];
+	    dv[2] = dv[2]*vertex[3] - dv[3]*vertex[2];
+	 }
+
+
+         CROSS3(normal, du, dv);
+         NORMALIZE_3FV(normal);
+	 normal[3] = 1.0;
+
+	 COPY_SZ_4V( exec->vtx.attrptr[VBO_ATTRIB_NORMAL],
+		     exec->vtx.attrsz[VBO_ATTRIB_NORMAL],
+		     normal );
+
+      }
+      else {
+         _math_horner_bezier_surf(map->Points, vertex, uu, vv, 
+				  exec->eval.map2[0].sz,
+				  map->Uorder, map->Vorder);
+      }
+
+      if (exec->vtx.attrsz[0] == 4) 
+	 CALL_Vertex4fv(GET_DISPATCH(), ( vertex ));
+      else
+	 CALL_Vertex3fv(GET_DISPATCH(), ( vertex )); 
+   }
+}
+
+
diff --git a/mesalib/src/mesa/vbo/vbo_rebase.c b/mesalib/src/mesa/vbo/vbo_rebase.c
new file mode 100644
index 000000000..3bf7ef580
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_rebase.c
@@ -0,0 +1,223 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  6.5
+ *
+ * Copyright (C) 1999-2006  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+/* Helper for drivers which find themselves rendering a range of
+ * indices starting somewhere above zero.  Typically the application
+ * is issuing multiple DrawArrays() or DrawElements() to draw
+ * successive primitives layed out linearly in the vertex arrays.
+ * Unless the vertex arrays are all in a VBO, the OpenGL semantics
+ * imply that we need to re-upload the vertex data on each draw call.
+ * In that case, we want to avoid starting the upload at zero, as it
+ * will mean every draw call uploads an increasing amount of not-used
+ * vertex data.  Worse - in the software tnl module, all those
+ * vertices will be transformed and lit.
+ *
+ * If we just upload the new data, however, the indices will be
+ * incorrect as we tend to upload each set of vertex data to a new
+ * region.  
+ *
+ * This file provides a helper to adjust the arrays, primitives and
+ * indices of a draw call so that it can be re-issued with a min_index
+ * of zero.
+ */
+
+#include "main/glheader.h"
+#include "main/imports.h"
+#include "main/mtypes.h"
+
+#include "vbo.h"
+
+
+#define REBASE(TYPE) 						\
+static void *rebase_##TYPE( const void *ptr,			\
+			  GLuint count, 			\
+			  TYPE min_index )			\
+{								\
+   const TYPE *in = (TYPE *)ptr;				\
+   TYPE *tmp_indices = malloc(count * sizeof(TYPE));	\
+   GLuint i;							\
+								\
+   for (i = 0; i < count; i++)  				\
+      tmp_indices[i] = in[i] - min_index;			\
+								\
+   return (void *)tmp_indices;					\
+}
+
+
+REBASE(GLuint)
+REBASE(GLushort)
+REBASE(GLubyte)
+
+GLboolean vbo_all_varyings_in_vbos( const struct gl_client_array *arrays[] )
+{
+   GLuint i;
+   
+   for (i = 0; i < VERT_ATTRIB_MAX; i++)
+      if (arrays[i]->StrideB &&
+	  arrays[i]->BufferObj->Name == 0)
+	 return GL_FALSE;
+
+   return GL_TRUE;
+}
+
+/* Adjust primitives, indices and vertex definitions so that min_index
+ * becomes zero. There are lots of reasons for wanting to do this, eg:
+ *
+ * Software tnl:
+ *    - any time min_index != 0, otherwise unused vertices lower than
+ *      min_index will be transformed.
+ *
+ * Hardware tnl:
+ *    - if ib != NULL and min_index != 0, otherwise vertices lower than 
+ *      min_index will be uploaded.  Requires adjusting index values.
+ *
+ *    - if ib == NULL and min_index != 0, just for convenience so this doesn't
+ *      have to be handled within the driver.
+ *
+ * Hardware tnl with VBO support:
+ *    - as above, but only when vertices are not (all?) in VBO's.
+ *    - can't save time by trying to upload half a vbo - typically it is
+ *      all or nothing.
+ */
+void vbo_rebase_prims( GLcontext *ctx,
+		       const struct gl_client_array *arrays[],
+		       const struct _mesa_prim *prim,
+		       GLuint nr_prims,
+		       const struct _mesa_index_buffer *ib,
+		       GLuint min_index,
+		       GLuint max_index,
+		       vbo_draw_func draw )
+{
+   struct gl_client_array tmp_arrays[VERT_ATTRIB_MAX];
+   const struct gl_client_array *tmp_array_pointers[VERT_ATTRIB_MAX];
+
+   struct _mesa_index_buffer tmp_ib;
+   struct _mesa_prim *tmp_prims = NULL;
+   void *tmp_indices = NULL;
+   GLuint i;
+
+   assert(min_index != 0);
+
+   if (0)
+      _mesa_printf("%s %d..%d\n", __FUNCTION__, min_index, max_index);
+
+   if (ib) {
+      /* Unfortunately need to adjust each index individually.
+       */
+      GLboolean map_ib = ib->obj->Name && !ib->obj->Pointer;
+      void *ptr;
+
+      if (map_ib) 
+	 ctx->Driver.MapBuffer(ctx, 
+			       GL_ELEMENT_ARRAY_BUFFER,
+			       GL_READ_ONLY_ARB,
+			       ib->obj);
+
+
+      ptr = ADD_POINTERS(ib->obj->Pointer, ib->ptr);
+
+      /* Some users might prefer it if we translated elements to
+       * GLuints here.  Others wouldn't...
+       */
+      switch (ib->type) {
+      case GL_UNSIGNED_INT: 
+	 tmp_indices = rebase_GLuint( ptr, ib->count, min_index );
+	 break;
+      case GL_UNSIGNED_SHORT: 
+	 tmp_indices = rebase_GLushort( ptr, ib->count, min_index );
+	 break;
+      case GL_UNSIGNED_BYTE: 
+	 tmp_indices = rebase_GLubyte( ptr, ib->count, min_index );
+	 break;
+      }      
+
+      if (map_ib) 
+	 ctx->Driver.UnmapBuffer(ctx, 
+				 GL_ELEMENT_ARRAY_BUFFER,
+				 ib->obj);
+
+      tmp_ib.obj = ctx->Shared->NullBufferObj;
+      tmp_ib.ptr = tmp_indices;
+      tmp_ib.count = ib->count;
+      tmp_ib.type = ib->type;
+
+      ib = &tmp_ib;
+   }
+   else {
+      /* Otherwise the primitives need adjustment.
+       */
+      tmp_prims = (struct _mesa_prim *)_mesa_malloc(sizeof(*prim) * nr_prims);
+
+      for (i = 0; i < nr_prims; i++) {
+	 /* If this fails, it could indicate an application error:
+	  */
+	 assert(prim[i].start >= min_index);
+
+	 tmp_prims[i] = prim[i];
+	 tmp_prims[i].start -= min_index;
+      }
+
+      prim = tmp_prims;
+   }
+
+   /* Just need to adjust the pointer values on each incoming array.
+    * This works for VBO and non-vbo rendering and shouldn't pesimize
+    * VBO-based upload schemes.  However this may still not be a fast
+    * path for hardware tnl for VBO based rendering as most machines
+    * will be happier if you just specify a starting vertex value in
+    * each primitive.
+    *
+    * For drivers with hardware tnl, you only want to do this if you
+    * are forced to, eg non-VBO indexed rendering with start != 0.
+    */
+   for (i = 0; i < VERT_ATTRIB_MAX; i++) {
+      tmp_arrays[i] = *arrays[i];
+      tmp_arrays[i].Ptr += min_index * tmp_arrays[i].StrideB;
+      tmp_array_pointers[i] = &tmp_arrays[i];
+   }
+   
+   /* Re-issue the draw call.
+    */
+   draw( ctx, 
+	 tmp_array_pointers, 
+	 prim, 
+	 nr_prims, 
+	 ib, 
+	 GL_TRUE,
+	 0, 
+	 max_index - min_index );
+   
+   if (tmp_indices)
+      _mesa_free(tmp_indices);
+   
+   if (tmp_prims)
+      _mesa_free(tmp_prims);
+}
+
+
+
diff --git a/mesalib/src/mesa/vbo/vbo_save.c b/mesalib/src/mesa/vbo/vbo_save.c
new file mode 100644
index 000000000..9757c3d9f
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_save.c
@@ -0,0 +1,110 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.2
+ *
+ * Copyright (C) 1999-2008  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+
+#include "main/mtypes.h"
+#include "main/bufferobj.h"
+#include "main/dlist.h"
+#include "main/vtxfmt.h"
+#include "main/imports.h"
+
+#include "vbo_context.h"
+
+
+
+static void vbo_save_callback_init( GLcontext *ctx )
+{
+   ctx->Driver.NewList = vbo_save_NewList;
+   ctx->Driver.EndList = vbo_save_EndList;
+   ctx->Driver.SaveFlushVertices = vbo_save_SaveFlushVertices;
+   ctx->Driver.BeginCallList = vbo_save_BeginCallList;
+   ctx->Driver.EndCallList = vbo_save_EndCallList;
+   ctx->Driver.NotifySaveBegin = vbo_save_NotifyBegin;
+}
+
+
+
+void vbo_save_init( GLcontext *ctx )
+{
+   struct vbo_context *vbo = vbo_context(ctx);
+   struct vbo_save_context *save = &vbo->save;
+
+   save->ctx = ctx;
+
+   vbo_save_api_init( save );
+   vbo_save_callback_init(ctx);
+
+   {
+      struct gl_client_array *arrays = save->arrays;
+      memcpy(arrays,      vbo->legacy_currval,  16 * sizeof(arrays[0]));
+      memcpy(arrays + 16, vbo->generic_currval, 16 * sizeof(arrays[0]));
+   }
+
+   ctx->Driver.CurrentSavePrimitive = PRIM_UNKNOWN;
+}
+
+
+void vbo_save_destroy( GLcontext *ctx )
+{
+   struct vbo_context *vbo = vbo_context(ctx);
+   struct vbo_save_context *save = &vbo->save;
+   GLuint i;
+
+   if (save->prim_store) {
+      if ( --save->prim_store->refcount == 0 ) {
+         FREE( save->prim_store );
+         save->prim_store = NULL;
+      }
+      if ( --save->vertex_store->refcount == 0 ) {
+         _mesa_reference_buffer_object(ctx,
+                                       &save->vertex_store->bufferobj, NULL);
+         FREE( save->vertex_store );
+         save->vertex_store = NULL;
+      }
+   }
+
+   for (i = 0; i < VBO_ATTRIB_MAX; i++) {
+      _mesa_reference_buffer_object(ctx, &save->arrays[i].BufferObj, NULL);
+   }
+}
+
+
+
+
+/* Note that this can occur during the playback of a display list:
+ */
+void vbo_save_fallback( GLcontext *ctx, GLboolean fallback )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save;
+
+   if (fallback)
+      save->replay_flags |= VBO_SAVE_FALLBACK;
+   else
+      save->replay_flags &= ~VBO_SAVE_FALLBACK;
+}
+
+
diff --git a/mesalib/src/mesa/vbo/vbo_save.h b/mesalib/src/mesa/vbo/vbo_save.h
new file mode 100644
index 000000000..86bbd24f7
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_save.h
@@ -0,0 +1,184 @@
+/**************************************************************************
+
+Copyright 2002 Tungsten Graphics Inc., Cedar Park, Texas.
+
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell <keith@tungstengraphics.com>
+ *
+ */
+
+#ifndef VBO_SAVE_H
+#define VBO_SAVE_H
+
+#include "main/mtypes.h"
+#include "vbo.h"
+#include "vbo_attrib.h"
+
+
+struct vbo_save_copied_vtx {
+   GLfloat buffer[VBO_ATTRIB_MAX * 4 * VBO_MAX_COPIED_VERTS];
+   GLuint nr;
+};
+
+
+/* For display lists, this structure holds a run of vertices of the
+ * same format, and a strictly well-formed set of begin/end pairs,
+ * starting on the first vertex and ending at the last.  Vertex
+ * copying on buffer breaks is precomputed according to these
+ * primitives, though there are situations where the copying will need
+ * correction at execute-time, perhaps by replaying the list as
+ * immediate mode commands.
+ *
+ * On executing this list, the 'current' values may be updated with
+ * the values of the final vertex, and often no fixup of the start of
+ * the vertex list is required.
+ *
+ * Eval and other commands that don't fit into these vertex lists are
+ * compiled using the fallback opcode mechanism provided by dlist.c.
+ */
+struct vbo_save_vertex_list {
+   GLubyte attrsz[VBO_ATTRIB_MAX];
+   GLuint vertex_size;
+
+   /* Copy of the final vertex from node->vertex_store->bufferobj.
+    * Keep this in regular (non-VBO) memory to avoid repeated
+    * map/unmap of the VBO when updating GL current data.
+    */
+   GLfloat *current_data;
+   GLuint current_size;
+
+   GLuint buffer_offset;
+   GLuint count;
+   GLuint wrap_count;		/* number of copied vertices at start */
+   GLboolean dangling_attr_ref;	/* current attr implicitly referenced 
+				   outside the list */
+
+   struct _mesa_prim *prim;
+   GLuint prim_count;
+
+   struct vbo_save_vertex_store *vertex_store;
+   struct vbo_save_primitive_store *prim_store;
+};
+
+/* These buffers should be a reasonable size to support upload to
+ * hardware.  Current vbo implementation will re-upload on any
+ * changes, so don't make too big or apps which dynamically create
+ * dlists and use only a few times will suffer.
+ *
+ * Consider stategy of uploading regions from the VBO on demand in the
+ * case of dynamic vbos.  Then make the dlist code signal that
+ * likelyhood as it occurs.  No reason we couldn't change usage
+ * internally even though this probably isn't allowed for client VBOs?
+ */
+#define VBO_SAVE_BUFFER_SIZE (8*1024) /* dwords */
+#define VBO_SAVE_PRIM_SIZE   128
+#define VBO_SAVE_PRIM_WEAK 0x40
+
+#define VBO_SAVE_FALLBACK    0x10000000
+
+/* Storage to be shared among several vertex_lists.
+ */
+struct vbo_save_vertex_store {
+   struct gl_buffer_object *bufferobj;
+   GLfloat *buffer;
+   GLuint used;
+   GLuint refcount;
+};
+
+struct vbo_save_primitive_store {
+   struct _mesa_prim buffer[VBO_SAVE_PRIM_SIZE];
+   GLuint used;
+   GLuint refcount;
+};
+
+
+struct vbo_save_context {
+   GLcontext *ctx;
+   GLvertexformat vtxfmt;
+   struct gl_client_array arrays[VBO_ATTRIB_MAX];
+   const struct gl_client_array *inputs[VBO_ATTRIB_MAX];
+
+   GLubyte attrsz[VBO_ATTRIB_MAX];
+   GLubyte active_sz[VBO_ATTRIB_MAX];
+   GLuint vertex_size;
+
+   GLfloat *buffer;
+   GLuint count;
+   GLuint wrap_count;
+   GLuint replay_flags;
+
+   struct _mesa_prim *prim;
+   GLuint prim_count, prim_max;
+
+   struct vbo_save_vertex_store *vertex_store;
+   struct vbo_save_primitive_store *prim_store;
+
+   GLfloat *buffer_ptr;		   /* cursor, points into buffer */
+   GLfloat vertex[VBO_ATTRIB_MAX*4];	   /* current values */
+   GLfloat *attrptr[VBO_ATTRIB_MAX];
+   GLuint vert_count;
+   GLuint max_vert;
+   GLboolean dangling_attr_ref;
+   GLboolean have_materials;
+
+   GLuint opcode_vertex_list;
+
+   struct vbo_save_copied_vtx copied;
+   
+   GLfloat *current[VBO_ATTRIB_MAX]; /* points into ctx->ListState */
+   GLubyte *currentsz[VBO_ATTRIB_MAX];
+};
+
+
+void vbo_save_init( GLcontext *ctx );
+void vbo_save_destroy( GLcontext *ctx );
+void vbo_save_fallback( GLcontext *ctx, GLboolean fallback );
+
+/* save_loopback.c:
+ */
+void vbo_loopback_vertex_list( GLcontext *ctx,
+			       const GLfloat *buffer,
+			       const GLubyte *attrsz,
+			       const struct _mesa_prim *prim,
+			       GLuint prim_count,
+			       GLuint wrap_count,
+			       GLuint vertex_size);
+
+/* Callbacks:
+ */
+void vbo_save_EndList( GLcontext *ctx );
+void vbo_save_NewList( GLcontext *ctx, GLuint list, GLenum mode );
+void vbo_save_EndCallList( GLcontext *ctx );
+void vbo_save_BeginCallList( GLcontext *ctx, struct gl_display_list *list );
+void vbo_save_SaveFlushVertices( GLcontext *ctx );
+GLboolean vbo_save_NotifyBegin( GLcontext *ctx, GLenum mode );
+
+void vbo_save_playback_vertex_list( GLcontext *ctx, void *data );
+
+void vbo_save_api_init( struct vbo_save_context *save );
+
+#endif
diff --git a/mesalib/src/mesa/vbo/vbo_save_api.c b/mesalib/src/mesa/vbo/vbo_save_api.c
new file mode 100644
index 000000000..1771510d8
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_save_api.c
@@ -0,0 +1,1238 @@
+/**************************************************************************
+
+Copyright 2002-2008 Tungsten Graphics Inc., Cedar Park, Texas.
+
+All Rights Reserved.
+
+Permission is hereby granted, free of charge, to any person obtaining a
+copy of this software and associated documentation files (the "Software"),
+to deal in the Software without restriction, including without limitation
+on the rights to use, copy, modify, merge, publish, distribute, sub
+license, and/or sell copies of the Software, and to permit persons to whom
+the Software is furnished to do so, subject to the following conditions:
+
+The above copyright notice and this permission notice (including the next
+paragraph) shall be included in all copies or substantial portions of the
+Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT. IN NO EVENT SHALL
+TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+
+
+/* Display list compiler attempts to store lists of vertices with the
+ * same vertex layout.  Additionally it attempts to minimize the need
+ * for execute-time fixup of these vertex lists, allowing them to be
+ * cached on hardware.
+ *
+ * There are still some circumstances where this can be thwarted, for
+ * example by building a list that consists of one very long primitive
+ * (eg Begin(Triangles), 1000 vertices, End), and calling that list
+ * from inside a different begin/end object (Begin(Lines), CallList,
+ * End).  
+ *
+ * In that case the code will have to replay the list as individual
+ * commands through the Exec dispatch table, or fix up the copied
+ * vertices at execute-time.
+ *
+ * The other case where fixup is required is when a vertex attribute
+ * is introduced in the middle of a primitive.  Eg:
+ *  Begin(Lines)
+ *  TexCoord1f()           Vertex2f()
+ *  TexCoord1f() Color3f() Vertex2f()
+ *  End()
+ *
+ *  If the current value of Color isn't known at compile-time, this
+ *  primitive will require fixup.
+ *
+ *
+ * The list compiler currently doesn't attempt to compile lists
+ * containing EvalCoord or EvalPoint commands.  On encountering one of
+ * these, compilation falls back to opcodes.  
+ *
+ * This could be improved to fallback only when a mix of EvalCoord and
+ * Vertex commands are issued within a single primitive.
+ */
+
+
+#include "main/glheader.h"
+#include "main/bufferobj.h"
+#include "main/context.h"
+#include "main/dlist.h"
+#include "main/enums.h"
+#include "main/macros.h"
+#include "main/api_noop.h"
+#include "main/api_validate.h"
+#include "main/api_arrayelt.h"
+#include "main/vtxfmt.h"
+#include "glapi/dispatch.h"
+
+#include "vbo_context.h"
+
+
+#ifdef ERROR
+#undef ERROR
+#endif
+
+
+/* An interesting VBO number/name to help with debugging */
+#define VBO_BUF_ID  12345
+
+
+/*
+ * NOTE: Old 'parity' issue is gone, but copying can still be
+ * wrong-footed on replay.
+ */
+static GLuint _save_copy_vertices( GLcontext *ctx, 
+				   const struct vbo_save_vertex_list *node,
+				   const GLfloat *src_buffer)
+{
+   struct vbo_save_context *save = &vbo_context( ctx )->save;
+   const struct _mesa_prim *prim = &node->prim[node->prim_count-1];
+   GLuint nr = prim->count;
+   GLuint sz = save->vertex_size;
+   const GLfloat *src = src_buffer + prim->start * sz;
+   GLfloat *dst = save->copied.buffer;
+   GLuint ovf, i;
+
+   if (prim->end)
+      return 0;
+	 
+   switch( prim->mode )
+   {
+   case GL_POINTS:
+      return 0;
+   case GL_LINES:
+      ovf = nr&1;
+      for (i = 0 ; i < ovf ; i++)
+	 _mesa_memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz*sizeof(GLfloat) );
+      return i;
+   case GL_TRIANGLES:
+      ovf = nr%3;
+      for (i = 0 ; i < ovf ; i++)
+	 _mesa_memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz*sizeof(GLfloat) );
+      return i;
+   case GL_QUADS:
+      ovf = nr&3;
+      for (i = 0 ; i < ovf ; i++)
+	 _mesa_memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz*sizeof(GLfloat) );
+      return i;
+   case GL_LINE_STRIP:
+      if (nr == 0) 
+	 return 0;
+      else {
+	 _mesa_memcpy( dst, src+(nr-1)*sz, sz*sizeof(GLfloat) );
+	 return 1;
+      }
+   case GL_LINE_LOOP:
+   case GL_TRIANGLE_FAN:
+   case GL_POLYGON:
+      if (nr == 0) 
+	 return 0;
+      else if (nr == 1) {
+	 _mesa_memcpy( dst, src+0, sz*sizeof(GLfloat) );
+	 return 1;
+      } else {
+	 _mesa_memcpy( dst, src+0, sz*sizeof(GLfloat) );
+	 _mesa_memcpy( dst+sz, src+(nr-1)*sz, sz*sizeof(GLfloat) );
+	 return 2;
+      }
+   case GL_TRIANGLE_STRIP:
+   case GL_QUAD_STRIP:
+      switch (nr) {
+      case 0: ovf = 0; break;
+      case 1: ovf = 1; break;
+      default: ovf = 2 + (nr&1); break;
+      }
+      for (i = 0 ; i < ovf ; i++)
+	 _mesa_memcpy( dst+i*sz, src+(nr-ovf+i)*sz, sz*sizeof(GLfloat) );
+      return i;
+   default:
+      assert(0);
+      return 0;
+   }
+}
+
+
+static struct vbo_save_vertex_store *alloc_vertex_store( GLcontext *ctx )
+{
+   struct vbo_save_vertex_store *vertex_store = CALLOC_STRUCT(vbo_save_vertex_store);
+
+   /* obj->Name needs to be non-zero, but won't ever be examined more
+    * closely than that.  In particular these buffers won't be entered
+    * into the hash and can never be confused with ones visible to the
+    * user.  Perhaps there could be a special number for internal
+    * buffers:
+    */
+   vertex_store->bufferobj = ctx->Driver.NewBufferObject(ctx,
+                                                         VBO_BUF_ID,
+                                                         GL_ARRAY_BUFFER_ARB);
+
+   ctx->Driver.BufferData( ctx, 
+			   GL_ARRAY_BUFFER_ARB, 
+			   VBO_SAVE_BUFFER_SIZE * sizeof(GLfloat),
+			   NULL,
+			   GL_STATIC_DRAW_ARB,
+			   vertex_store->bufferobj);
+
+   vertex_store->buffer = NULL;
+   vertex_store->used = 0;
+   vertex_store->refcount = 1;
+
+   return vertex_store;
+}
+
+static void free_vertex_store( GLcontext *ctx, struct vbo_save_vertex_store *vertex_store )
+{
+   assert(!vertex_store->buffer);
+
+   if (vertex_store->bufferobj) {
+      _mesa_reference_buffer_object(ctx, &vertex_store->bufferobj, NULL);
+   }
+
+   FREE( vertex_store );
+}
+
+static GLfloat *map_vertex_store( GLcontext *ctx, struct vbo_save_vertex_store *vertex_store )
+{
+   assert(vertex_store->bufferobj);
+   assert(!vertex_store->buffer);
+   vertex_store->buffer = (GLfloat *)ctx->Driver.MapBuffer(ctx, 
+							   GL_ARRAY_BUFFER_ARB,	/* not used */
+							   GL_WRITE_ONLY, /* not used */
+							   vertex_store->bufferobj); 
+
+   assert(vertex_store->buffer);
+   return vertex_store->buffer + vertex_store->used;
+}
+
+static void unmap_vertex_store( GLcontext *ctx, struct vbo_save_vertex_store *vertex_store )
+{
+   ctx->Driver.UnmapBuffer( ctx, GL_ARRAY_BUFFER_ARB, vertex_store->bufferobj );
+   vertex_store->buffer = NULL;
+}
+
+
+static struct vbo_save_primitive_store *alloc_prim_store( GLcontext *ctx )
+{
+   struct vbo_save_primitive_store *store = CALLOC_STRUCT(vbo_save_primitive_store);
+   (void) ctx;
+   store->used = 0;
+   store->refcount = 1;
+   return store;
+}
+
+static void _save_reset_counters( GLcontext *ctx )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save;
+
+   save->prim = save->prim_store->buffer + save->prim_store->used;
+   save->buffer = (save->vertex_store->buffer + 
+		   save->vertex_store->used);
+
+   assert(save->buffer == save->buffer_ptr);
+
+   if (save->vertex_size)
+      save->max_vert = ((VBO_SAVE_BUFFER_SIZE - save->vertex_store->used) / 
+			 save->vertex_size);
+   else
+      save->max_vert = 0;
+
+   save->vert_count = 0;
+   save->prim_count = 0;
+   save->prim_max = VBO_SAVE_PRIM_SIZE - save->prim_store->used;
+   save->dangling_attr_ref = 0;
+}
+
+
+/* Insert the active immediate struct onto the display list currently
+ * being built.
+ */
+static void _save_compile_vertex_list( GLcontext *ctx )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save;
+   struct vbo_save_vertex_list *node;
+
+   /* Allocate space for this structure in the display list currently
+    * being compiled.
+    */
+   node = (struct vbo_save_vertex_list *)
+      _mesa_alloc_instruction(ctx, save->opcode_vertex_list, sizeof(*node));
+
+   if (!node)
+      return;
+
+   /* Duplicate our template, increment refcounts to the storage structs:
+    */
+   _mesa_memcpy(node->attrsz, save->attrsz, sizeof(node->attrsz)); 
+   node->vertex_size = save->vertex_size;
+   node->buffer_offset = (save->buffer - save->vertex_store->buffer) * sizeof(GLfloat); 
+   node->count = save->vert_count;
+   node->wrap_count = save->copied.nr;
+   node->dangling_attr_ref = save->dangling_attr_ref;
+   node->prim = save->prim;
+   node->prim_count = save->prim_count;
+   node->vertex_store = save->vertex_store;
+   node->prim_store = save->prim_store;
+
+   node->vertex_store->refcount++;
+   node->prim_store->refcount++;
+
+
+   node->current_size = node->vertex_size - node->attrsz[0];
+   node->current_data = NULL;
+
+   if (node->current_size) {
+      /* If the malloc fails, we just pull the data out of the VBO
+       * later instead.
+       */
+      node->current_data = MALLOC( node->current_size * sizeof(GLfloat) );
+      if (node->current_data) {
+         const char *buffer = (const char *)save->vertex_store->buffer;
+         unsigned attr_offset = node->attrsz[0] * sizeof(GLfloat);
+         unsigned vertex_offset = 0;
+
+         if (node->count)
+            vertex_offset = (node->count-1) * node->vertex_size * sizeof(GLfloat);
+
+         memcpy( node->current_data,
+                 buffer + node->buffer_offset + vertex_offset + attr_offset,
+                 node->current_size * sizeof(GLfloat) );
+      }
+   }
+
+
+
+   assert(node->attrsz[VBO_ATTRIB_POS] != 0 ||
+	  node->count == 0);
+
+   if (save->dangling_attr_ref)
+      ctx->ListState.CurrentList->Flags |= DLIST_DANGLING_REFS;
+
+   save->vertex_store->used += save->vertex_size * node->count;
+   save->prim_store->used += node->prim_count;
+
+
+   /* Copy duplicated vertices 
+    */
+   save->copied.nr = _save_copy_vertices( ctx, node, save->buffer );
+
+
+   /* Deal with GL_COMPILE_AND_EXECUTE:
+    */
+   if (ctx->ExecuteFlag) {
+      struct _glapi_table *dispatch = GET_DISPATCH();
+
+      _glapi_set_dispatch(ctx->Exec);
+
+      vbo_loopback_vertex_list( ctx,
+				(const GLfloat *)((const char *)save->vertex_store->buffer + 
+						  node->buffer_offset),
+				node->attrsz,
+				node->prim,
+				node->prim_count,
+				node->wrap_count,
+				node->vertex_size);
+
+      _glapi_set_dispatch(dispatch);
+   }
+
+
+   /* Decide whether the storage structs are full, or can be used for
+    * the next vertex lists as well.
+    */
+   if (save->vertex_store->used > 
+       VBO_SAVE_BUFFER_SIZE - 16 * (save->vertex_size + 4)) {
+
+      /* Unmap old store:
+       */
+      unmap_vertex_store( ctx, save->vertex_store );
+
+      /* Release old reference:
+       */
+      save->vertex_store->refcount--; 
+      assert(save->vertex_store->refcount != 0);
+      save->vertex_store = NULL;
+
+      /* Allocate and map new store:
+       */
+      save->vertex_store = alloc_vertex_store( ctx );
+      save->buffer_ptr = map_vertex_store( ctx, save->vertex_store );
+   } 
+
+   if (save->prim_store->used > VBO_SAVE_PRIM_SIZE - 6) {
+      save->prim_store->refcount--; 
+      assert(save->prim_store->refcount != 0);
+      save->prim_store = alloc_prim_store( ctx );
+   } 
+
+   /* Reset our structures for the next run of vertices:
+    */
+   _save_reset_counters( ctx );
+}
+
+
+/* TODO -- If no new vertices have been stored, don't bother saving
+ * it.
+ */
+static void _save_wrap_buffers( GLcontext *ctx )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save;
+   GLint i = save->prim_count - 1;
+   GLenum mode;
+   GLboolean weak;
+
+   assert(i < (GLint) save->prim_max);
+   assert(i >= 0);
+
+   /* Close off in-progress primitive.
+    */
+   save->prim[i].count = (save->vert_count - 
+			  save->prim[i].start);
+   mode = save->prim[i].mode;
+   weak = save->prim[i].weak;
+   
+   /* store the copied vertices, and allocate a new list.
+    */
+   _save_compile_vertex_list( ctx );
+
+   /* Restart interrupted primitive
+    */
+   save->prim[0].mode = mode;
+   save->prim[0].weak = weak;
+   save->prim[0].begin = 0;
+   save->prim[0].end = 0;
+   save->prim[0].pad = 0;
+   save->prim[0].start = 0;
+   save->prim[0].count = 0;
+   save->prim_count = 1;
+}
+
+
+
+/* Called only when buffers are wrapped as the result of filling the
+ * vertex_store struct.  
+ */
+static void _save_wrap_filled_vertex( GLcontext *ctx )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save;
+   GLfloat *data = save->copied.buffer;
+   GLuint i;
+
+   /* Emit a glEnd to close off the last vertex list.
+    */
+   _save_wrap_buffers( ctx );
+   
+    /* Copy stored stored vertices to start of new list.
+    */
+   assert(save->max_vert - save->vert_count > save->copied.nr);
+
+   for (i = 0 ; i < save->copied.nr ; i++) {
+      _mesa_memcpy( save->buffer_ptr, data, save->vertex_size * sizeof(GLfloat));
+      data += save->vertex_size;
+      save->buffer_ptr += save->vertex_size;
+      save->vert_count++;
+   }
+}
+
+
+static void _save_copy_to_current( GLcontext *ctx )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save; 
+   GLuint i;
+
+   for (i = VBO_ATTRIB_POS+1 ; i < VBO_ATTRIB_MAX ; i++) {
+      if (save->attrsz[i]) {
+	 save->currentsz[i][0] = save->attrsz[i];
+	 COPY_CLEAN_4V(save->current[i], 
+		       save->attrsz[i], 
+		       save->attrptr[i]);
+      }
+   }
+}
+
+
+static void _save_copy_from_current( GLcontext *ctx )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save; 
+   GLint i;
+
+   for (i = VBO_ATTRIB_POS+1 ; i < VBO_ATTRIB_MAX ; i++) {
+      switch (save->attrsz[i]) {
+      case 4: save->attrptr[i][3] = save->current[i][3];
+      case 3: save->attrptr[i][2] = save->current[i][2];
+      case 2: save->attrptr[i][1] = save->current[i][1];
+      case 1: save->attrptr[i][0] = save->current[i][0];
+      case 0: break;
+      }
+   }
+}
+
+
+
+
+/* Flush existing data, set new attrib size, replay copied vertices.
+ */ 
+static void _save_upgrade_vertex( GLcontext *ctx, 
+				 GLuint attr,
+				 GLuint newsz )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save;
+   GLuint oldsz;
+   GLuint i;
+   GLfloat *tmp;
+
+   /* Store the current run of vertices, and emit a GL_END.  Emit a
+    * BEGIN in the new buffer.
+    */
+   if (save->vert_count) 
+      _save_wrap_buffers( ctx );
+   else
+      assert( save->copied.nr == 0 );
+
+   /* Do a COPY_TO_CURRENT to ensure back-copying works for the case
+    * when the attribute already exists in the vertex and is having
+    * its size increased.  
+    */
+   _save_copy_to_current( ctx );
+
+   /* Fix up sizes:
+    */
+   oldsz = save->attrsz[attr];
+   save->attrsz[attr] = newsz;
+
+   save->vertex_size += newsz - oldsz;
+   save->max_vert = ((VBO_SAVE_BUFFER_SIZE - save->vertex_store->used) / 
+		      save->vertex_size);
+   save->vert_count = 0;
+
+   /* Recalculate all the attrptr[] values:
+    */
+   for (i = 0, tmp = save->vertex ; i < VBO_ATTRIB_MAX ; i++) {
+      if (save->attrsz[i]) {
+	 save->attrptr[i] = tmp;
+	 tmp += save->attrsz[i];
+      }
+      else 
+	 save->attrptr[i] = NULL; /* will not be dereferenced. */
+   }
+
+   /* Copy from current to repopulate the vertex with correct values.
+    */
+   _save_copy_from_current( ctx );
+
+   /* Replay stored vertices to translate them to new format here.
+    *
+    * If there are copied vertices and the new (upgraded) attribute
+    * has not been defined before, this list is somewhat degenerate,
+    * and will need fixup at runtime.
+    */
+   if (save->copied.nr)
+   {
+      GLfloat *data = save->copied.buffer;
+      GLfloat *dest = save->buffer;
+      GLuint j;
+
+      /* Need to note this and fix up at runtime (or loopback):
+       */
+      if (attr != VBO_ATTRIB_POS && save->currentsz[attr][0] == 0) {
+	 assert(oldsz == 0);
+	 save->dangling_attr_ref = GL_TRUE;
+      }
+
+      for (i = 0 ; i < save->copied.nr ; i++) {
+	 for (j = 0 ; j < VBO_ATTRIB_MAX ; j++) {
+	    if (save->attrsz[j]) {
+	       if (j == attr) {
+		  if (oldsz) {
+		     COPY_CLEAN_4V( dest, oldsz, data );
+		     data += oldsz;
+		     dest += newsz;
+		  }
+		  else {
+		     COPY_SZ_4V( dest, newsz, save->current[attr] );
+		     dest += newsz;
+		  }
+	       }
+	       else {
+		  GLint sz = save->attrsz[j];
+		  COPY_SZ_4V( dest, sz, data );
+		  data += sz;
+		  dest += sz;
+	       }
+	    }
+	 }
+      }
+
+      save->buffer_ptr = dest;
+      save->vert_count += save->copied.nr;
+   }
+}
+
+static void save_fixup_vertex( GLcontext *ctx, GLuint attr, GLuint sz )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save; 
+
+   if (sz > save->attrsz[attr]) {
+      /* New size is larger.  Need to flush existing vertices and get
+       * an enlarged vertex format.
+       */
+      _save_upgrade_vertex( ctx, attr, sz );
+   }
+   else if (sz < save->active_sz[attr]) {
+      static GLfloat id[4] = { 0, 0, 0, 1 };
+      GLuint i;
+
+      /* New size is equal or smaller - just need to fill in some
+       * zeros.
+       */
+      for (i = sz ; i <= save->attrsz[attr] ; i++)
+	 save->attrptr[attr][i-1] = id[i-1];
+   }
+
+   save->active_sz[attr] = sz;
+}
+
+static void _save_reset_vertex( GLcontext *ctx )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save;
+   GLuint i;
+
+   for (i = 0 ; i < VBO_ATTRIB_MAX ; i++) {
+      save->attrsz[i] = 0;
+      save->active_sz[i] = 0;
+   }
+      
+   save->vertex_size = 0;
+}
+
+
+
+#define ERROR()   _mesa_compile_error( ctx, GL_INVALID_ENUM, __FUNCTION__ );
+
+
+/* Only one size for each attribute may be active at once.  Eg. if
+ * Color3f is installed/active, then Color4f may not be, even if the
+ * vertex actually contains 4 color coordinates.  This is because the
+ * 3f version won't otherwise set color[3] to 1.0 -- this is the job
+ * of the chooser function when switching between Color4f and Color3f.
+ */
+#define ATTR( A, N, V0, V1, V2, V3 )				\
+do {								\
+   struct vbo_save_context *save = &vbo_context(ctx)->save;	\
+								\
+   if (save->active_sz[A] != N)				\
+      save_fixup_vertex(ctx, A, N);				\
+								\
+   {								\
+      GLfloat *dest = save->attrptr[A];			\
+      if (N>0) dest[0] = V0;					\
+      if (N>1) dest[1] = V1;					\
+      if (N>2) dest[2] = V2;					\
+      if (N>3) dest[3] = V3;					\
+   }								\
+								\
+   if ((A) == 0) {						\
+      GLuint i;							\
+								\
+      for (i = 0; i < save->vertex_size; i++)			\
+	 save->buffer_ptr[i] = save->vertex[i];			\
+								\
+      save->buffer_ptr += save->vertex_size;				\
+								\
+      if (++save->vert_count >= save->max_vert)			\
+	 _save_wrap_filled_vertex( ctx );			\
+   }								\
+} while (0)
+
+#define TAG(x) _save_##x
+
+#include "vbo_attrib_tmp.h"
+
+
+
+
+/* Cope with EvalCoord/CallList called within a begin/end object:
+ *     -- Flush current buffer
+ *     -- Fallback to opcodes for the rest of the begin/end object.
+ */
+static void DO_FALLBACK( GLcontext *ctx )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save;
+
+   if (save->vert_count || save->prim_count) {
+      GLint i = save->prim_count - 1;
+
+      /* Close off in-progress primitive.
+       */
+      save->prim[i].count = (save->vert_count - 
+                             save->prim[i].start);
+
+      /* Need to replay this display list with loopback,
+       * unfortunately, otherwise this primitive won't be handled
+       * properly:
+       */
+      save->dangling_attr_ref = 1;
+      
+      _save_compile_vertex_list( ctx );
+   }
+
+   _save_copy_to_current( ctx );
+   _save_reset_vertex( ctx );
+   _save_reset_counters( ctx );
+   _mesa_install_save_vtxfmt( ctx, &ctx->ListState.ListVtxfmt );
+   ctx->Driver.SaveNeedFlush = 0;
+}
+
+static void GLAPIENTRY _save_EvalCoord1f( GLfloat u )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   DO_FALLBACK(ctx);
+   ctx->Save->EvalCoord1f( u );
+}
+
+static void GLAPIENTRY _save_EvalCoord1fv( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   DO_FALLBACK(ctx);
+   ctx->Save->EvalCoord1fv( v );
+}
+
+static void GLAPIENTRY _save_EvalCoord2f( GLfloat u, GLfloat v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   DO_FALLBACK(ctx);
+   ctx->Save->EvalCoord2f( u, v );
+}
+
+static void GLAPIENTRY _save_EvalCoord2fv( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   DO_FALLBACK(ctx);
+   ctx->Save->EvalCoord2fv( v );
+}
+
+static void GLAPIENTRY _save_EvalPoint1( GLint i )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   DO_FALLBACK(ctx);
+   ctx->Save->EvalPoint1( i );
+}
+
+static void GLAPIENTRY _save_EvalPoint2( GLint i, GLint j )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   DO_FALLBACK(ctx);
+   ctx->Save->EvalPoint2( i, j );
+}
+
+static void GLAPIENTRY _save_CallList( GLuint l )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   DO_FALLBACK(ctx);
+   ctx->Save->CallList( l );
+}
+
+static void GLAPIENTRY _save_CallLists( GLsizei n, GLenum type, const GLvoid *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   DO_FALLBACK(ctx);
+   ctx->Save->CallLists( n, type, v );
+}
+
+
+
+
+/* This begin is hooked into ...  Updating of
+ * ctx->Driver.CurrentSavePrimitive is already taken care of.
+ */
+GLboolean vbo_save_NotifyBegin( GLcontext *ctx, GLenum mode )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save; 
+
+   GLuint i = save->prim_count++;
+
+   assert(i < save->prim_max);
+   save->prim[i].mode = mode & ~VBO_SAVE_PRIM_WEAK;
+   save->prim[i].begin = 1;
+   save->prim[i].end = 0;
+   save->prim[i].weak = (mode & VBO_SAVE_PRIM_WEAK) ? 1 : 0;
+   save->prim[i].pad = 0;
+   save->prim[i].start = save->vert_count;
+   save->prim[i].count = 0;   
+
+   _mesa_install_save_vtxfmt( ctx, &save->vtxfmt );      
+   ctx->Driver.SaveNeedFlush = 1;
+   return GL_TRUE;
+}
+
+
+
+static void GLAPIENTRY _save_End( void )
+{
+   GET_CURRENT_CONTEXT( ctx ); 
+   struct vbo_save_context *save = &vbo_context(ctx)->save; 
+   GLint i = save->prim_count - 1;
+
+   ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
+   save->prim[i].end = 1;
+   save->prim[i].count = (save->vert_count - 
+			  save->prim[i].start);
+
+   if (i == (GLint) save->prim_max - 1) {
+      _save_compile_vertex_list( ctx );
+      assert(save->copied.nr == 0);
+   }
+
+   /* Swap out this vertex format while outside begin/end.  Any color,
+    * etc. received between here and the next begin will be compiled
+    * as opcodes.
+    */   
+   _mesa_install_save_vtxfmt( ctx, &ctx->ListState.ListVtxfmt );
+}
+
+
+/* These are all errors as this vtxfmt is only installed inside
+ * begin/end pairs.
+ */
+static void GLAPIENTRY _save_DrawElements(GLenum mode, GLsizei count, GLenum type,
+			       const GLvoid *indices)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   (void) mode; (void) count; (void) type; (void) indices;
+   _mesa_compile_error( ctx, GL_INVALID_OPERATION, "glDrawElements" );
+}
+
+
+static void GLAPIENTRY _save_DrawRangeElements(GLenum mode,
+				    GLuint start, GLuint end,
+				    GLsizei count, GLenum type,
+				    const GLvoid *indices)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   (void) mode; (void) start; (void) end; (void) count; (void) type; (void) indices;
+   _mesa_compile_error( ctx, GL_INVALID_OPERATION, "glDrawRangeElements" );
+}
+
+static void GLAPIENTRY _save_DrawArrays(GLenum mode, GLint start, GLsizei count)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   (void) mode; (void) start; (void) count;
+   _mesa_compile_error( ctx, GL_INVALID_OPERATION, "glDrawArrays" );
+}
+
+static void GLAPIENTRY _save_Rectf( GLfloat x1, GLfloat y1, GLfloat x2, GLfloat y2 )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   (void) x1; (void) y1; (void) x2; (void) y2;
+   _mesa_compile_error( ctx, GL_INVALID_OPERATION, "glRectf" );
+}
+
+static void GLAPIENTRY _save_EvalMesh1( GLenum mode, GLint i1, GLint i2 )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   (void) mode; (void) i1; (void) i2;
+   _mesa_compile_error( ctx, GL_INVALID_OPERATION, "glEvalMesh1" );
+}
+
+static void GLAPIENTRY _save_EvalMesh2( GLenum mode, GLint i1, GLint i2,
+				  GLint j1, GLint j2 )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   (void) mode; (void) i1; (void) i2; (void) j1; (void) j2;
+   _mesa_compile_error( ctx, GL_INVALID_OPERATION, "glEvalMesh2" );
+}
+
+static void GLAPIENTRY _save_Begin( GLenum mode )
+{
+   GET_CURRENT_CONTEXT( ctx );
+   (void) mode;
+   _mesa_compile_error( ctx, GL_INVALID_OPERATION, "Recursive glBegin" );
+}
+
+
+/* Unlike the functions above, these are to be hooked into the vtxfmt
+ * maintained in ctx->ListState, active when the list is known or
+ * suspected to be outside any begin/end primitive.
+ */
+static void GLAPIENTRY _save_OBE_Rectf( GLfloat x1, GLfloat y1, GLfloat x2, GLfloat y2 )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   vbo_save_NotifyBegin( ctx, GL_QUADS | VBO_SAVE_PRIM_WEAK );
+   CALL_Vertex2f(GET_DISPATCH(), ( x1, y1 ));
+   CALL_Vertex2f(GET_DISPATCH(), ( x2, y1 ));
+   CALL_Vertex2f(GET_DISPATCH(), ( x2, y2 ));
+   CALL_Vertex2f(GET_DISPATCH(), ( x1, y2 ));
+   CALL_End(GET_DISPATCH(), ());
+}
+
+
+static void GLAPIENTRY _save_OBE_DrawArrays(GLenum mode, GLint start, GLsizei count)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   GLint i;
+
+   if (!_mesa_validate_DrawArrays( ctx, mode, start, count ))
+      return;
+
+   _ae_map_vbos( ctx );
+
+   vbo_save_NotifyBegin( ctx, mode | VBO_SAVE_PRIM_WEAK );
+
+   for (i = 0; i < count; i++)
+       CALL_ArrayElement(GET_DISPATCH(), (start + i));
+   CALL_End(GET_DISPATCH(), ());
+
+   _ae_unmap_vbos( ctx );
+}
+
+/* Could do better by copying the arrays and element list intact and
+ * then emitting an indexed prim at runtime.
+ */
+static void GLAPIENTRY _save_OBE_DrawElements(GLenum mode, GLsizei count, GLenum type,
+				   const GLvoid *indices)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   GLint i;
+
+   if (!_mesa_validate_DrawElements( ctx, mode, count, type, indices ))
+      return;
+
+   _ae_map_vbos( ctx );
+
+   if (_mesa_is_bufferobj(ctx->Array.ElementArrayBufferObj))
+      indices = ADD_POINTERS(ctx->Array.ElementArrayBufferObj->Pointer, indices);
+
+   vbo_save_NotifyBegin( ctx, mode | VBO_SAVE_PRIM_WEAK );
+
+   switch (type) {
+   case GL_UNSIGNED_BYTE:
+      for (i = 0 ; i < count ; i++)
+	  CALL_ArrayElement(GET_DISPATCH(), ( ((GLubyte *)indices)[i] ));
+      break;
+   case GL_UNSIGNED_SHORT:
+      for (i = 0 ; i < count ; i++)
+	  CALL_ArrayElement(GET_DISPATCH(), ( ((GLushort *)indices)[i] ));
+      break;
+   case GL_UNSIGNED_INT:
+      for (i = 0 ; i < count ; i++)
+	  CALL_ArrayElement(GET_DISPATCH(), ( ((GLuint *)indices)[i] ));
+      break;
+   default:
+      _mesa_error( ctx, GL_INVALID_ENUM, "glDrawElements(type)" );
+      break;
+   }
+
+   CALL_End(GET_DISPATCH(), ());
+
+   _ae_unmap_vbos( ctx );
+}
+
+static void GLAPIENTRY _save_OBE_DrawRangeElements(GLenum mode,
+					GLuint start, GLuint end,
+					GLsizei count, GLenum type,
+					const GLvoid *indices)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   if (_mesa_validate_DrawRangeElements( ctx, mode,
+					 start, end,
+					 count, type, indices ))
+      _save_OBE_DrawElements( mode, count, type, indices );
+}
+
+
+
+
+
+static void _save_vtxfmt_init( GLcontext *ctx )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save;
+   GLvertexformat *vfmt = &save->vtxfmt;
+
+   vfmt->ArrayElement = _ae_loopback_array_elt;	        /* generic helper */
+   vfmt->Begin = _save_Begin;
+   vfmt->Color3f = _save_Color3f;
+   vfmt->Color3fv = _save_Color3fv;
+   vfmt->Color4f = _save_Color4f;
+   vfmt->Color4fv = _save_Color4fv;
+   vfmt->EdgeFlag = _save_EdgeFlag;
+   vfmt->End = _save_End;
+   vfmt->FogCoordfEXT = _save_FogCoordfEXT;
+   vfmt->FogCoordfvEXT = _save_FogCoordfvEXT;
+   vfmt->Indexf = _save_Indexf;
+   vfmt->Indexfv = _save_Indexfv;
+   vfmt->Materialfv = _save_Materialfv;
+   vfmt->MultiTexCoord1fARB = _save_MultiTexCoord1f;
+   vfmt->MultiTexCoord1fvARB = _save_MultiTexCoord1fv;
+   vfmt->MultiTexCoord2fARB = _save_MultiTexCoord2f;
+   vfmt->MultiTexCoord2fvARB = _save_MultiTexCoord2fv;
+   vfmt->MultiTexCoord3fARB = _save_MultiTexCoord3f;
+   vfmt->MultiTexCoord3fvARB = _save_MultiTexCoord3fv;
+   vfmt->MultiTexCoord4fARB = _save_MultiTexCoord4f;
+   vfmt->MultiTexCoord4fvARB = _save_MultiTexCoord4fv;
+   vfmt->Normal3f = _save_Normal3f;
+   vfmt->Normal3fv = _save_Normal3fv;
+   vfmt->SecondaryColor3fEXT = _save_SecondaryColor3fEXT;
+   vfmt->SecondaryColor3fvEXT = _save_SecondaryColor3fvEXT;
+   vfmt->TexCoord1f = _save_TexCoord1f;
+   vfmt->TexCoord1fv = _save_TexCoord1fv;
+   vfmt->TexCoord2f = _save_TexCoord2f;
+   vfmt->TexCoord2fv = _save_TexCoord2fv;
+   vfmt->TexCoord3f = _save_TexCoord3f;
+   vfmt->TexCoord3fv = _save_TexCoord3fv;
+   vfmt->TexCoord4f = _save_TexCoord4f;
+   vfmt->TexCoord4fv = _save_TexCoord4fv;
+   vfmt->Vertex2f = _save_Vertex2f;
+   vfmt->Vertex2fv = _save_Vertex2fv;
+   vfmt->Vertex3f = _save_Vertex3f;
+   vfmt->Vertex3fv = _save_Vertex3fv;
+   vfmt->Vertex4f = _save_Vertex4f;
+   vfmt->Vertex4fv = _save_Vertex4fv;
+   vfmt->VertexAttrib1fARB = _save_VertexAttrib1fARB;
+   vfmt->VertexAttrib1fvARB = _save_VertexAttrib1fvARB;
+   vfmt->VertexAttrib2fARB = _save_VertexAttrib2fARB;
+   vfmt->VertexAttrib2fvARB = _save_VertexAttrib2fvARB;
+   vfmt->VertexAttrib3fARB = _save_VertexAttrib3fARB;
+   vfmt->VertexAttrib3fvARB = _save_VertexAttrib3fvARB;
+   vfmt->VertexAttrib4fARB = _save_VertexAttrib4fARB;
+   vfmt->VertexAttrib4fvARB = _save_VertexAttrib4fvARB;
+
+   vfmt->VertexAttrib1fNV = _save_VertexAttrib1fNV;
+   vfmt->VertexAttrib1fvNV = _save_VertexAttrib1fvNV;
+   vfmt->VertexAttrib2fNV = _save_VertexAttrib2fNV;
+   vfmt->VertexAttrib2fvNV = _save_VertexAttrib2fvNV;
+   vfmt->VertexAttrib3fNV = _save_VertexAttrib3fNV;
+   vfmt->VertexAttrib3fvNV = _save_VertexAttrib3fvNV;
+   vfmt->VertexAttrib4fNV = _save_VertexAttrib4fNV;
+   vfmt->VertexAttrib4fvNV = _save_VertexAttrib4fvNV;
+   
+   /* This will all require us to fallback to saving the list as opcodes:
+    */ 
+   vfmt->CallList = _save_CallList; /* inside begin/end */
+   vfmt->CallLists = _save_CallLists; /* inside begin/end */
+   vfmt->EvalCoord1f = _save_EvalCoord1f;
+   vfmt->EvalCoord1fv = _save_EvalCoord1fv;
+   vfmt->EvalCoord2f = _save_EvalCoord2f;
+   vfmt->EvalCoord2fv = _save_EvalCoord2fv;
+   vfmt->EvalPoint1 = _save_EvalPoint1;
+   vfmt->EvalPoint2 = _save_EvalPoint2;
+
+   /* These are all errors as we at least know we are in some sort of
+    * begin/end pair:
+    */
+   vfmt->EvalMesh1 = _save_EvalMesh1;	
+   vfmt->EvalMesh2 = _save_EvalMesh2;
+   vfmt->Begin = _save_Begin;
+   vfmt->Rectf = _save_Rectf;
+   vfmt->DrawArrays = _save_DrawArrays;
+   vfmt->DrawElements = _save_DrawElements;
+   vfmt->DrawRangeElements = _save_DrawRangeElements;
+   /* Loops back into vfmt->DrawElements */
+   vfmt->MultiDrawElementsEXT = _mesa_noop_MultiDrawElements;
+
+}
+
+
+void vbo_save_SaveFlushVertices( GLcontext *ctx )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save;
+
+   /* Noop when we are actually active:
+    */
+   if (ctx->Driver.CurrentSavePrimitive == PRIM_INSIDE_UNKNOWN_PRIM ||
+       ctx->Driver.CurrentSavePrimitive <= GL_POLYGON)
+      return;
+
+   if (save->vert_count ||
+       save->prim_count) 
+      _save_compile_vertex_list( ctx );
+   
+   _save_copy_to_current( ctx );
+   _save_reset_vertex( ctx );
+   _save_reset_counters( ctx );  
+   ctx->Driver.SaveNeedFlush = 0;
+}
+
+void vbo_save_NewList( GLcontext *ctx, GLuint list, GLenum mode )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save;
+
+   (void) list; (void) mode;
+
+   if (!save->prim_store)
+      save->prim_store = alloc_prim_store( ctx );
+
+   if (!save->vertex_store) 
+      save->vertex_store = alloc_vertex_store( ctx );
+      
+   save->buffer_ptr = map_vertex_store( ctx, save->vertex_store );
+   
+   _save_reset_vertex( ctx );
+   _save_reset_counters( ctx );  
+   ctx->Driver.SaveNeedFlush = 0;
+}
+
+void vbo_save_EndList( GLcontext *ctx )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save;
+
+   /* EndList called inside a (saved) Begin/End pair?
+    */
+   if (ctx->Driver.CurrentSavePrimitive != PRIM_OUTSIDE_BEGIN_END) {
+
+      if (save->prim_count > 0) {
+         GLint i = save->prim_count - 1;
+         ctx->Driver.CurrentSavePrimitive = PRIM_OUTSIDE_BEGIN_END;
+         save->prim[i].end = 0;
+         save->prim[i].count = (save->vert_count - 
+                                save->prim[i].start);
+      }
+
+      /* Make sure this vertex list gets replayed by the "loopback"
+       * mechanism:
+       */
+      save->dangling_attr_ref = 1;
+      vbo_save_SaveFlushVertices( ctx );
+
+      /* Swap out this vertex format while outside begin/end.  Any color,
+       * etc. received between here and the next begin will be compiled
+       * as opcodes.
+       */   
+      _mesa_install_save_vtxfmt( ctx, &ctx->ListState.ListVtxfmt );
+   }
+
+   unmap_vertex_store( ctx, save->vertex_store );
+
+   assert(save->vertex_size == 0);
+}
+ 
+void vbo_save_BeginCallList( GLcontext *ctx, struct gl_display_list *dlist )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save;
+   save->replay_flags |= dlist->Flags;
+}
+
+void vbo_save_EndCallList( GLcontext *ctx )
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save;
+   
+   if (ctx->ListState.CallDepth == 1) {
+      /* This is correct: want to keep only the VBO_SAVE_FALLBACK
+       * flag, if it is set:
+       */
+      save->replay_flags &= VBO_SAVE_FALLBACK;
+   }
+}
+
+
+static void vbo_destroy_vertex_list( GLcontext *ctx, void *data )
+{
+   struct vbo_save_vertex_list *node = (struct vbo_save_vertex_list *)data;
+   (void) ctx;
+
+   if ( --node->vertex_store->refcount == 0 ) 
+      free_vertex_store( ctx, node->vertex_store );
+
+   if ( --node->prim_store->refcount == 0 )
+      FREE( node->prim_store );
+
+   if (node->current_data) {
+      FREE(node->current_data);
+      node->current_data = NULL;
+   }
+}
+
+
+static void vbo_print_vertex_list( GLcontext *ctx, void *data )
+{
+   struct vbo_save_vertex_list *node = (struct vbo_save_vertex_list *)data;
+   GLuint i;
+   (void) ctx;
+
+   _mesa_printf("VBO-VERTEX-LIST, %u vertices %d primitives, %d vertsize\n",
+		node->count,
+		node->prim_count,
+		node->vertex_size);
+
+   for (i = 0 ; i < node->prim_count ; i++) {
+      struct _mesa_prim *prim = &node->prim[i];
+      _mesa_debug(NULL, "   prim %d: %s%s %d..%d %s %s\n",
+		  i, 
+		  _mesa_lookup_prim_by_nr(prim->mode),
+		  prim->weak ? " (weak)" : "",
+		  prim->start, 
+		  prim->start + prim->count,
+		  (prim->begin) ? "BEGIN" : "(wrap)",
+		  (prim->end) ? "END" : "(wrap)");
+   }
+}
+
+
+static void _save_current_init( GLcontext *ctx ) 
+{
+   struct vbo_save_context *save = &vbo_context(ctx)->save;
+   GLint i;
+
+   for (i = VBO_ATTRIB_POS; i <= VBO_ATTRIB_GENERIC15; i++) {
+      const GLuint j = i - VBO_ATTRIB_POS;
+      ASSERT(j < VERT_ATTRIB_MAX);
+      save->currentsz[i] = &ctx->ListState.ActiveAttribSize[j];
+      save->current[i] = ctx->ListState.CurrentAttrib[j];
+   }
+
+   for (i = VBO_ATTRIB_FIRST_MATERIAL; i <= VBO_ATTRIB_LAST_MATERIAL; i++) {
+      const GLuint j = i - VBO_ATTRIB_FIRST_MATERIAL;
+      ASSERT(j < MAT_ATTRIB_MAX);
+      save->currentsz[i] = &ctx->ListState.ActiveMaterialSize[j];
+      save->current[i] = ctx->ListState.CurrentMaterial[j];
+   }
+}
+
+/**
+ * Initialize the display list compiler
+ */
+void vbo_save_api_init( struct vbo_save_context *save )
+{
+   GLcontext *ctx = save->ctx;
+   GLuint i;
+
+   save->opcode_vertex_list =
+      _mesa_alloc_opcode( ctx,
+			  sizeof(struct vbo_save_vertex_list),
+			  vbo_save_playback_vertex_list,
+			  vbo_destroy_vertex_list,
+			  vbo_print_vertex_list );
+
+   ctx->Driver.NotifySaveBegin = vbo_save_NotifyBegin;
+
+   _save_vtxfmt_init( ctx );
+   _save_current_init( ctx );
+
+   /* These will actually get set again when binding/drawing */
+   for (i = 0; i < VBO_ATTRIB_MAX; i++)
+      save->inputs[i] = &save->arrays[i];
+
+   /* Hook our array functions into the outside-begin-end vtxfmt in 
+    * ctx->ListState.
+    */
+   ctx->ListState.ListVtxfmt.Rectf = _save_OBE_Rectf;
+   ctx->ListState.ListVtxfmt.DrawArrays = _save_OBE_DrawArrays;
+   ctx->ListState.ListVtxfmt.DrawElements = _save_OBE_DrawElements;
+   ctx->ListState.ListVtxfmt.DrawRangeElements = _save_OBE_DrawRangeElements;
+   /* loops back into _save_OBE_DrawElements */
+   ctx->ListState.ListVtxfmt.MultiDrawElementsEXT = _mesa_noop_MultiDrawElements;
+   _mesa_install_save_vtxfmt( ctx, &ctx->ListState.ListVtxfmt );
+}
+
diff --git a/mesalib/src/mesa/vbo/vbo_save_draw.c b/mesalib/src/mesa/vbo/vbo_save_draw.c
new file mode 100644
index 000000000..d834fa1f2
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_save_draw.c
@@ -0,0 +1,290 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  7.2
+ *
+ * Copyright (C) 1999-2008  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/* Author:
+ *    Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#include "main/glheader.h"
+#include "main/bufferobj.h"
+#include "main/context.h"
+#include "main/imports.h"
+#include "main/mtypes.h"
+#include "main/macros.h"
+#include "main/light.h"
+#include "main/state.h"
+
+#include "vbo_context.h"
+
+
+/*
+ * After playback, copy everything but the position from the
+ * last vertex to the saved state
+ */
+static void _playback_copy_to_current( GLcontext *ctx,
+				       const struct vbo_save_vertex_list *node )
+{
+   struct vbo_context *vbo = vbo_context(ctx);
+   GLfloat vertex[VBO_ATTRIB_MAX * 4];
+   GLfloat *data;
+   GLuint i, offset;
+
+   if (node->current_size == 0)
+      return;
+
+   if (node->current_data) {
+      data = node->current_data;
+   }
+   else {
+      data = vertex;
+
+      if (node->count)
+         offset = (node->buffer_offset + 
+                   (node->count-1) * node->vertex_size * sizeof(GLfloat));
+      else
+         offset = node->buffer_offset;
+
+      ctx->Driver.GetBufferSubData( ctx, 0, offset, 
+                                    node->vertex_size * sizeof(GLfloat), 
+                                    data, node->vertex_store->bufferobj );
+
+      data += node->attrsz[0]; /* skip vertex position */
+   }
+
+   for (i = VBO_ATTRIB_POS+1 ; i < VBO_ATTRIB_MAX ; i++) {
+      if (node->attrsz[i]) {
+	 GLfloat *current = (GLfloat *)vbo->currval[i].Ptr;
+         GLfloat tmp[4];
+
+         COPY_CLEAN_4V(tmp, 
+                       node->attrsz[i], 
+                       data);
+         
+         if (memcmp(current, tmp, 4 * sizeof(GLfloat)) != 0)
+         {
+            memcpy(current, tmp, 4 * sizeof(GLfloat));
+
+            vbo->currval[i].Size = node->attrsz[i];
+
+            if (i >= VBO_ATTRIB_FIRST_MATERIAL &&
+                i <= VBO_ATTRIB_LAST_MATERIAL)
+               ctx->NewState |= _NEW_LIGHT;
+
+            ctx->NewState |= _NEW_CURRENT_ATTRIB;
+         }
+
+	 data += node->attrsz[i];
+      }
+   }
+
+   /* Colormaterial -- this kindof sucks.
+    */
+   if (ctx->Light.ColorMaterialEnabled) {
+      _mesa_update_color_material(ctx, ctx->Current.Attrib[VBO_ATTRIB_COLOR0]);
+   }
+
+   /* CurrentExecPrimitive
+    */
+   if (node->prim_count) {
+      const struct _mesa_prim *prim = &node->prim[node->prim_count - 1];
+      if (prim->end)
+	 ctx->Driver.CurrentExecPrimitive = PRIM_OUTSIDE_BEGIN_END;
+      else
+	 ctx->Driver.CurrentExecPrimitive = prim->mode;
+   }
+}
+
+
+
+/* Treat the vertex storage as a VBO, define vertex arrays pointing
+ * into it:
+ */
+static void vbo_bind_vertex_list( GLcontext *ctx,
+                                   const struct vbo_save_vertex_list *node )
+{
+   struct vbo_context *vbo = vbo_context(ctx);
+   struct vbo_save_context *save = &vbo->save;
+   struct gl_client_array *arrays = save->arrays;
+   GLuint buffer_offset = node->buffer_offset;
+   const GLuint *map;
+   GLuint attr;
+   GLubyte node_attrsz[VBO_ATTRIB_MAX];  /* copy of node->attrsz[] */
+   GLbitfield varying_inputs = 0x0;
+
+   memcpy(node_attrsz, node->attrsz, sizeof(node->attrsz));
+
+   /* Install the default (ie Current) attributes first, then overlay
+    * all active ones.
+    */
+   switch (get_program_mode(ctx)) {
+   case VP_NONE:
+      for (attr = 0; attr < 16; attr++) {
+         save->inputs[attr] = &vbo->legacy_currval[attr];
+      }
+      for (attr = 0; attr < MAT_ATTRIB_MAX; attr++) {
+         save->inputs[attr + 16] = &vbo->mat_currval[attr];
+      }
+      map = vbo->map_vp_none;
+      break;
+   case VP_NV:
+   case VP_ARB:
+      /* The aliasing of attributes for NV vertex programs has already
+       * occurred.  NV vertex programs cannot access material values,
+       * nor attributes greater than VERT_ATTRIB_TEX7.  
+       */
+      for (attr = 0; attr < 16; attr++) {
+         save->inputs[attr] = &vbo->legacy_currval[attr];
+         save->inputs[attr + 16] = &vbo->generic_currval[attr];
+      }
+      map = vbo->map_vp_arb;
+
+      /* check if VERT_ATTRIB_POS is not read but VERT_BIT_GENERIC0 is read.
+       * In that case we effectively need to route the data from
+       * glVertexAttrib(0, val) calls to feed into the GENERIC0 input.
+       */
+      if ((ctx->VertexProgram._Current->Base.InputsRead & VERT_BIT_POS) == 0 &&
+          (ctx->VertexProgram._Current->Base.InputsRead & VERT_BIT_GENERIC0)) {
+         save->inputs[16] = save->inputs[0];
+         node_attrsz[16] = node_attrsz[0];
+         node_attrsz[0] = 0;
+      }
+      break;
+   default:
+      assert(0);
+   }
+
+   for (attr = 0; attr < VERT_ATTRIB_MAX; attr++) {
+      GLuint src = map[attr];
+
+      if (node_attrsz[src]) {
+         /* override the default array set above */
+         save->inputs[attr] = &arrays[attr];
+
+	 arrays[attr].Ptr = (const GLubyte *) NULL + buffer_offset;
+	 arrays[attr].Size = node->attrsz[src];
+	 arrays[attr].StrideB = node->vertex_size * sizeof(GLfloat);
+	 arrays[attr].Stride = node->vertex_size * sizeof(GLfloat);
+	 arrays[attr].Type = GL_FLOAT;
+         arrays[attr].Format = GL_RGBA;
+	 arrays[attr].Enabled = 1;
+         _mesa_reference_buffer_object(ctx,
+                                       &arrays[attr].BufferObj,
+                                       node->vertex_store->bufferobj);
+	 arrays[attr]._MaxElement = node->count; /* ??? */
+	 
+	 assert(arrays[attr].BufferObj->Name);
+
+	 buffer_offset += node->attrsz[src] * sizeof(GLfloat);
+         varying_inputs |= 1<<attr;
+      }
+   }
+
+   _mesa_set_varying_vp_inputs( ctx, varying_inputs );
+}
+
+static void vbo_save_loopback_vertex_list( GLcontext *ctx,
+					   const struct vbo_save_vertex_list *list )
+{
+   const char *buffer = ctx->Driver.MapBuffer(ctx, 
+					      GL_ARRAY_BUFFER_ARB, 
+					      GL_READ_ONLY, /* ? */
+					       list->vertex_store->bufferobj);
+
+   vbo_loopback_vertex_list( ctx,
+			     (const GLfloat *)(buffer + list->buffer_offset),
+			     list->attrsz,
+			     list->prim,
+			     list->prim_count,
+			     list->wrap_count,
+			     list->vertex_size);
+
+   ctx->Driver.UnmapBuffer(ctx, GL_ARRAY_BUFFER_ARB, 
+			   list->vertex_store->bufferobj);
+}
+
+
+/**
+ * Execute the buffer and save copied verts.
+ */
+void vbo_save_playback_vertex_list( GLcontext *ctx, void *data )
+{
+   const struct vbo_save_vertex_list *node = (const struct vbo_save_vertex_list *) data;
+   struct vbo_save_context *save = &vbo_context(ctx)->save;
+
+   FLUSH_CURRENT(ctx, 0);
+
+   if (node->prim_count > 0 && node->count > 0) {
+
+      if (ctx->Driver.CurrentExecPrimitive != PRIM_OUTSIDE_BEGIN_END &&
+	  node->prim[0].begin) {
+
+	 /* Degenerate case: list is called inside begin/end pair and
+	  * includes operations such as glBegin or glDrawArrays.
+	  */
+	 if (0)
+	    _mesa_printf("displaylist recursive begin");
+
+	 vbo_save_loopback_vertex_list( ctx, node );
+	 return;
+      }
+      else if (save->replay_flags) {
+	 /* Various degnerate cases: translate into immediate mode
+	  * calls rather than trying to execute in place.
+	  */
+	 vbo_save_loopback_vertex_list( ctx, node );
+	 return;
+      }
+      
+      if (ctx->NewState)
+	 _mesa_update_state( ctx );
+
+      /* XXX also need to check if shader enabled, but invalid */
+      if ((ctx->VertexProgram.Enabled && !ctx->VertexProgram._Enabled) ||
+          (ctx->FragmentProgram.Enabled && !ctx->FragmentProgram._Enabled)) {
+         _mesa_error(ctx, GL_INVALID_OPERATION,
+                     "glBegin (invalid vertex/fragment program)");
+         return;
+      }
+
+      vbo_bind_vertex_list( ctx, node );
+
+      /* Again...
+       */
+      if (ctx->NewState)
+	 _mesa_update_state( ctx );
+
+      vbo_context(ctx)->draw_prims( ctx, 
+				    save->inputs, 
+				    node->prim, 
+				    node->prim_count,
+				    NULL,
+				    GL_TRUE,
+				    0,	/* Node is a VBO, so this is ok */
+				    node->count - 1);
+   }
+
+   /* Copy to current?
+    */
+   _playback_copy_to_current( ctx, node );
+}
diff --git a/mesalib/src/mesa/vbo/vbo_save_loopback.c b/mesalib/src/mesa/vbo/vbo_save_loopback.c
new file mode 100644
index 000000000..b7a74e453
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_save_loopback.c
@@ -0,0 +1,191 @@
+/**************************************************************************
+ * 
+ * Copyright 2005 Tungsten Graphics, Inc., Cedar Park, Texas.
+ * All Rights Reserved.
+ * 
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the
+ * "Software"), to deal in the Software without restriction, including
+ * without limitation the rights to use, copy, modify, merge, publish,
+ * distribute, sub license, and/or sell copies of the Software, and to
+ * permit persons to whom the Software is furnished to do so, subject to
+ * the following conditions:
+ * 
+ * The above copyright notice and this permission notice (including the
+ * next paragraph) shall be included in all copies or substantial portions
+ * of the Software.
+ * 
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+ * MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NON-INFRINGEMENT.
+ * IN NO EVENT SHALL TUNGSTEN GRAPHICS AND/OR ITS SUPPLIERS BE LIABLE FOR
+ * ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT,
+ * TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE
+ * SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ * 
+ **************************************************************************/
+
+#include "main/context.h"
+#include "main/glheader.h"
+#include "main/enums.h"
+#include "main/imports.h"
+#include "main/macros.h"
+#include "main/mtypes.h"
+#include "glapi/dispatch.h"
+#include "glapi/glapi.h"
+
+#include "vbo_context.h"
+
+
+
+typedef void (*attr_func)( GLcontext *ctx, GLint target, const GLfloat * );
+
+
+/* This file makes heavy use of the aliasing of NV vertex attributes
+ * with the legacy attributes, and also with ARB and Material
+ * attributes as currently implemented.
+ */
+static void VertexAttrib1fvNV(GLcontext *ctx, GLint target, const GLfloat *v)
+{
+   CALL_VertexAttrib1fvNV(ctx->Exec, (target, v));
+}
+
+static void VertexAttrib2fvNV(GLcontext *ctx, GLint target, const GLfloat *v)
+{
+   CALL_VertexAttrib2fvNV(ctx->Exec, (target, v));
+}
+
+static void VertexAttrib3fvNV(GLcontext *ctx, GLint target, const GLfloat *v)
+{
+   CALL_VertexAttrib3fvNV(ctx->Exec, (target, v));
+}
+
+static void VertexAttrib4fvNV(GLcontext *ctx, GLint target, const GLfloat *v)
+{
+   CALL_VertexAttrib4fvNV(ctx->Exec, (target, v));
+}
+
+static attr_func vert_attrfunc[4] = {
+   VertexAttrib1fvNV,
+   VertexAttrib2fvNV,
+   VertexAttrib3fvNV,
+   VertexAttrib4fvNV
+};
+
+struct loopback_attr {
+   GLint target;
+   GLint sz;
+   attr_func func;
+};
+
+/* Don't emit ends and begins on wrapped primitives.  Don't replay
+ * wrapped vertices.  If we get here, it's probably because the the
+ * precalculated wrapping is wrong.
+ */
+static void loopback_prim( GLcontext *ctx,
+			   const GLfloat *buffer,
+			   const struct _mesa_prim *prim,
+			   GLuint wrap_count,
+			   GLuint vertex_size,
+			   const struct loopback_attr *la, GLuint nr )
+{
+   GLint start = prim->start;
+   GLint end = start + prim->count;
+   const GLfloat *data;
+   GLint j;
+   GLuint k;
+
+   if (0)
+      _mesa_printf("loopback prim %s(%s,%s) verts %d..%d\n",
+		   _mesa_lookup_prim_by_nr(prim->mode),
+		   prim->begin ? "begin" : "..",
+		   prim->end ? "end" : "..",
+		   start, 
+		   end);
+
+   if (prim->begin) {
+      CALL_Begin(GET_DISPATCH(), ( prim->mode ));
+   }
+   else {
+      assert(start == 0);
+      start += wrap_count;
+   }
+
+   data = buffer + start * vertex_size;
+
+   for (j = start ; j < end ; j++) {
+      const GLfloat *tmp = data + la[0].sz;
+
+      for (k = 1 ; k < nr ; k++) {
+	 la[k].func( ctx, la[k].target, tmp );
+	 tmp += la[k].sz;
+      }
+	 
+      /* Fire the vertex
+       */
+      la[0].func( ctx, VBO_ATTRIB_POS, data );
+      data = tmp;
+   }
+
+   if (prim->end) {
+      CALL_End(GET_DISPATCH(), ());
+   }
+}
+
+/* Primitives generated by DrawArrays/DrawElements/Rectf may be
+ * caught here.  If there is no primitive in progress, execute them
+ * normally, otherwise need to track and discard the generated
+ * primitives.
+ */
+static void loopback_weak_prim( GLcontext *ctx,
+				const struct _mesa_prim *prim )
+{
+   /* Use the prim_weak flag to ensure that if this primitive
+    * wraps, we don't mistake future vertex_lists for part of the
+    * surrounding primitive.
+    *
+    * While this flag is set, we are simply disposing of data
+    * generated by an operation now known to be a noop.
+    */
+   if (prim->begin)
+      ctx->Driver.CurrentExecPrimitive |= VBO_SAVE_PRIM_WEAK;
+   if (prim->end)
+      ctx->Driver.CurrentExecPrimitive &= ~VBO_SAVE_PRIM_WEAK;
+}
+
+
+void vbo_loopback_vertex_list( GLcontext *ctx,
+			       const GLfloat *buffer,
+			       const GLubyte *attrsz,
+			       const struct _mesa_prim *prim,
+			       GLuint prim_count,
+			       GLuint wrap_count,
+			       GLuint vertex_size)
+{
+   struct loopback_attr la[VBO_ATTRIB_MAX];
+   GLuint i, nr = 0;
+
+   /* All Legacy, NV, ARB and Material attributes are routed through
+    * the NV attributes entrypoints:
+    */
+   for (i = 0 ; i < VBO_ATTRIB_MAX ; i++) {
+      if (attrsz[i]) {
+	 la[nr].target = i;
+	 la[nr].sz = attrsz[i];
+	 la[nr].func = vert_attrfunc[attrsz[i]-1];
+	 nr++;
+      }
+   }
+
+   for (i = 0 ; i < prim_count ; i++) {
+      if ((prim[i].mode & VBO_SAVE_PRIM_WEAK) &&
+	  (ctx->Driver.CurrentExecPrimitive != PRIM_OUTSIDE_BEGIN_END))
+      {
+	 loopback_weak_prim( ctx, &prim[i] );
+      }
+      else
+      {
+	 loopback_prim( ctx, buffer, &prim[i], wrap_count, vertex_size, la, nr );
+      }
+   }
+}
diff --git a/mesalib/src/mesa/vbo/vbo_split.c b/mesalib/src/mesa/vbo/vbo_split.c
new file mode 100644
index 000000000..58e879628
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_split.c
@@ -0,0 +1,161 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  6.5
+ *
+ * Copyright (C) 1999-2006  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+/* Deal with hardware and/or swtnl maximums:
+ * - maximum number of vertices in buffer
+ * - maximum number of elements (maybe zero)
+ *
+ * The maximums may vary with opengl state (eg if a larger hardware
+ * vertex is required in this state, the maximum number of vertices
+ * may be smaller than in another state).
+ *
+ * We want buffer splitting to be a convenience function for the code
+ * actually drawing the primitives rather than a system-wide maximum,
+ * otherwise it is hard to avoid pessimism.  
+ *
+ * For instance, if a driver has no hardware limits on vertex buffer
+ * dimensions, it would not ordinarily want to split vbos.  But if
+ * there is an unexpected fallback, eg memory manager fails to upload
+ * textures, it will want to pass the drawing commands onto swtnl,
+ * which does have limitations.  A convenience function allows swtnl
+ * to split the drawing and vbos internally without imposing its
+ * limitations on drivers which want to use it as a fallback path.
+ */
+
+#include "main/glheader.h"
+#include "main/imports.h"
+#include "main/mtypes.h"
+
+#include "vbo_split.h"
+#include "vbo.h"
+
+/* True if a primitive can be split without copying of vertices, false
+ * otherwise.
+ */
+GLboolean split_prim_inplace(GLenum mode, GLuint *first, GLuint *incr)
+{
+   switch (mode) {
+   case GL_POINTS:
+      *first = 1;
+      *incr = 1;
+      return GL_TRUE;
+   case GL_LINES:
+      *first = 2;
+      *incr = 2;
+      return GL_TRUE;
+   case GL_LINE_STRIP:
+      *first = 2;
+      *incr = 1;
+      return GL_TRUE;
+   case GL_TRIANGLES:
+      *first = 3;
+      *incr = 3;
+      return GL_TRUE;
+   case GL_TRIANGLE_STRIP:
+      *first = 3;
+      *incr = 1;
+      return GL_TRUE;
+   case GL_QUADS:
+      *first = 4;
+      *incr = 4;
+      return GL_TRUE;
+   case GL_QUAD_STRIP:
+      *first = 4;
+      *incr = 2;
+      return GL_TRUE;
+   default:
+      *first = 0;
+      *incr = 1;		/* so that count % incr works */
+      return GL_FALSE;
+   }
+}
+
+
+
+void vbo_split_prims( GLcontext *ctx,
+		      const struct gl_client_array *arrays[],
+		      const struct _mesa_prim *prim,
+		      GLuint nr_prims,
+		      const struct _mesa_index_buffer *ib,
+		      GLuint min_index,
+		      GLuint max_index,
+		      vbo_draw_func draw,
+		      const struct split_limits *limits )
+{
+  
+   if (ib) {
+      if (limits->max_indices == 0) {
+	 /* Could traverse the indices, re-emitting vertices in turn.
+	  * But it's hard to see why this case would be needed - for
+	  * software tnl, it is better to convert to non-indexed
+	  * rendering after transformation is complete, as is done in
+	  * the t_dd_rendertmp.h templates.  Are there any devices
+	  * with hardware tnl that cannot do indexed rendering?
+	  *
+	  * For now, this path is disabled.
+	  */
+	 assert(0);
+      }
+      else if (max_index - min_index >= limits->max_verts) {
+	 /* The vertex buffers are too large for hardware (or the
+	  * swtnl module).  Traverse the indices, re-emitting vertices
+	  * in turn.  Use a vertex cache to preserve some of the
+	  * sharing from the original index list.
+	  */
+	 vbo_split_copy(ctx, arrays, prim, nr_prims, ib,
+			draw, limits );
+      }
+      else if (ib->count > limits->max_indices) {
+	 /* The index buffer is too large for hardware.  Try to split
+	  * on whole-primitive boundaries, otherwise try to split the
+	  * individual primitives.
+	  */
+	 vbo_split_inplace(ctx, arrays, prim, nr_prims, ib,
+			   min_index, max_index, draw, limits );
+      }
+      else {
+	 /* Why were we called? */
+	 assert(0);
+      }
+   }
+   else {
+      if (max_index - min_index >= limits->max_verts) {
+	 /* The vertex buffer is too large for hardware (or the swtnl
+	  * module).  Try to split on whole-primitive boundaries,
+	  * otherwise try to split the individual primitives.
+	  */
+	 vbo_split_inplace(ctx, arrays, prim, nr_prims, ib,
+			   min_index, max_index, draw, limits );
+      }
+      else {
+	 /* Why were we called? */
+	 assert(0);
+      }
+   }
+}
+
diff --git a/mesalib/src/mesa/vbo/vbo_split.h b/mesalib/src/mesa/vbo/vbo_split.h
new file mode 100644
index 000000000..05888d048
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_split.h
@@ -0,0 +1,72 @@
+/*
+ * mesa 3-D graphics library
+ * Version:  6.5
+ *
+ * Copyright (C) 1999-2006  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ */
+
+/**
+ * \file vbo_context.h
+ * \brief VBO builder module datatypes and definitions.
+ * \author Keith Whitwell
+ */
+
+
+/**
+ * \mainpage The VBO splitter
+ *
+ * This is the private data used internally to the vbo_split_prims()
+ * helper function.  Nobody outside the vbo_split* files needs to
+ * include or know about this structure.
+ */
+
+
+#ifndef _VBO_SPLIT_H
+#define _VBO_SPLIT_H
+
+#include "vbo.h"
+
+
+/* True if a primitive can be split without copying of vertices, false
+ * otherwise.
+ */
+GLboolean split_prim_inplace(GLenum mode, GLuint *first, GLuint *incr);
+
+void vbo_split_inplace( GLcontext *ctx,
+			const struct gl_client_array *arrays[],
+			const struct _mesa_prim *prim,
+			GLuint nr_prims,
+			const struct _mesa_index_buffer *ib,
+			GLuint min_index,
+			GLuint max_index,
+			vbo_draw_func draw,
+			const struct split_limits *limits );
+
+/* Requires ib != NULL:
+ */
+void vbo_split_copy( GLcontext *ctx,
+		     const struct gl_client_array *arrays[],
+		     const struct _mesa_prim *prim,
+		     GLuint nr_prims,
+		     const struct _mesa_index_buffer *ib,
+		     vbo_draw_func draw,
+		     const struct split_limits *limits );
+
+#endif
diff --git a/mesalib/src/mesa/vbo/vbo_split_copy.c b/mesalib/src/mesa/vbo/vbo_split_copy.c
new file mode 100644
index 000000000..8ec180d55
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_split_copy.c
@@ -0,0 +1,616 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  6.5
+ *
+ * Copyright (C) 1999-2006  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+/* Split indexed primitives with per-vertex copying.
+ */
+
+#include "main/glheader.h"
+#include "main/bufferobj.h"
+#include "main/imports.h"
+#include "main/image.h"
+#include "main/macros.h"
+#include "main/enums.h"
+#include "main/mtypes.h"
+
+#include "vbo_split.h"
+#include "vbo.h"
+
+
+#define ELT_TABLE_SIZE 16
+
+/**
+ * Used for vertex-level splitting of indexed buffers.  Note that
+ * non-indexed primitives may be converted to indexed in some cases
+ * (eg loops, fans) in order to use this splitting path.
+ */
+struct copy_context {
+
+   GLcontext *ctx;
+   const struct gl_client_array **array;
+   const struct _mesa_prim *prim;
+   GLuint nr_prims;
+   const struct _mesa_index_buffer *ib;
+   vbo_draw_func draw;
+
+   const struct split_limits *limits;
+
+   struct {
+      GLuint attr;
+      GLuint size;
+      const struct gl_client_array *array;
+      const GLubyte *src_ptr;
+
+      struct gl_client_array dstarray;
+
+   } varying[VERT_ATTRIB_MAX];
+   GLuint nr_varying;
+
+   const struct gl_client_array *dstarray_ptr[VERT_ATTRIB_MAX];
+   struct _mesa_index_buffer dstib;
+
+   GLuint *translated_elt_buf;
+   const GLuint *srcelt;
+
+   /** A baby hash table to avoid re-emitting (some) duplicate
+    * vertices when splitting indexed primitives.
+    */
+   struct { 
+      GLuint in;
+      GLuint out;
+   } vert_cache[ELT_TABLE_SIZE];
+
+   GLuint vertex_size;
+   GLubyte *dstbuf;
+   GLubyte *dstptr;     /**< dstptr == dstbuf + dstelt_max * vertsize */
+   GLuint dstbuf_size;  /**< in vertices */
+   GLuint dstbuf_nr;    /**< count of emitted vertices, also the largest value
+                         * in dstelt.  Our MaxIndex.
+                         */
+
+   GLuint *dstelt;
+   GLuint dstelt_nr;
+   GLuint dstelt_size;
+
+#define MAX_PRIM 32
+   struct _mesa_prim dstprim[MAX_PRIM];
+   GLuint dstprim_nr;
+
+};
+
+
+static GLuint attr_size( const struct gl_client_array *array )
+{
+   return array->Size * _mesa_sizeof_type(array->Type);
+}
+
+
+/**
+ * Starts returning true slightly before the buffer fills, to ensure
+ * that there is sufficient room for any remaining vertices to finish
+ * off the prim:
+ */
+static GLboolean
+check_flush( struct copy_context *copy )
+{
+   GLenum mode = copy->dstprim[copy->dstprim_nr].mode; 
+
+   if (GL_TRIANGLE_STRIP == mode &&
+       copy->dstelt_nr & 1) { /* see bug9962 */
+       return GL_FALSE;
+   }
+
+   if (copy->dstbuf_nr + 4 > copy->dstbuf_size)
+      return GL_TRUE;
+
+   if (copy->dstelt_nr + 4 > copy->dstelt_size)
+      return GL_TRUE;
+
+   return GL_FALSE;
+}
+
+
+/**
+ * Dump the parameters/info for a vbo->draw() call.
+ */
+static void
+dump_draw_info(GLcontext *ctx,
+               const struct gl_client_array **arrays,
+               const struct _mesa_prim *prims,
+               GLuint nr_prims,
+               const struct _mesa_index_buffer *ib,
+               GLuint min_index,
+               GLuint max_index)
+{
+   GLuint i, j;
+
+   _mesa_printf("VBO Draw:\n");
+   for (i = 0; i < nr_prims; i++) {
+      _mesa_printf("Prim %u of %u\n", i, nr_prims);
+      _mesa_printf("  Prim mode 0x%x\n", prims[i].mode);
+      _mesa_printf("  IB: %p\n", (void*) ib);
+      for (j = 0; j < VERT_ATTRIB_MAX; j++) {
+         _mesa_printf("    array %d at %p:\n", j, (void*) arrays[j]);
+         _mesa_printf("      enabled %d, ptr %p, size %d, type 0x%x, stride %d\n",
+                      arrays[j]->Enabled, arrays[j]->Ptr,
+                      arrays[j]->Size, arrays[j]->Type, arrays[j]->StrideB);
+         if (0) {
+            GLint k = prims[i].start + prims[i].count - 1;
+            GLfloat *last = (GLfloat *) (arrays[j]->Ptr + arrays[j]->Stride * k);
+            _mesa_printf("        last: %f %f %f\n",
+                         last[0], last[1], last[2]);
+         }
+      }
+   }
+}
+
+
+static void
+flush( struct copy_context *copy )
+{
+   GLuint i;
+
+   /* Set some counters: 
+    */
+   copy->dstib.count = copy->dstelt_nr;
+
+#if 0
+   dump_draw_info(copy->ctx,
+                  copy->dstarray_ptr,
+                  copy->dstprim,
+                  copy->dstprim_nr,
+                  &copy->dstib,
+                  0,
+                  copy->dstbuf_nr);
+#else
+   (void) dump_draw_info;
+#endif
+
+   copy->draw( copy->ctx,
+	       copy->dstarray_ptr,
+	       copy->dstprim,
+	       copy->dstprim_nr,
+	       &copy->dstib,
+	       GL_TRUE,
+	       0,
+	       copy->dstbuf_nr );
+
+   /* Reset all pointers: 
+    */
+   copy->dstprim_nr = 0;
+   copy->dstelt_nr = 0;
+   copy->dstbuf_nr = 0;
+   copy->dstptr = copy->dstbuf;
+
+   /* Clear the vertex cache:
+    */
+   for (i = 0; i < ELT_TABLE_SIZE; i++)
+      copy->vert_cache[i].in = ~0;
+}
+
+
+/**
+ * Called at begin of each primitive during replay.
+ */
+static void
+begin( struct copy_context *copy, GLenum mode, GLboolean begin_flag )
+{
+   struct _mesa_prim *prim = &copy->dstprim[copy->dstprim_nr];
+
+/*    _mesa_printf("begin %s (%d)\n", _mesa_lookup_prim_by_nr(mode), begin_flag); */
+		
+   prim->mode = mode;
+   prim->begin = begin_flag;
+}
+
+
+/**
+ * Use a hashtable to attempt to identify recently-emitted vertices
+ * and avoid re-emitting them.
+ */
+static GLuint
+elt(struct copy_context *copy, GLuint elt_idx)
+{
+   GLuint elt = copy->srcelt[elt_idx];
+   GLuint slot = elt & (ELT_TABLE_SIZE-1);
+
+/*    _mesa_printf("elt %d\n", elt); */
+
+   /* Look up the incoming element in the vertex cache.  Re-emit if
+    * necessary.   
+    */
+   if (copy->vert_cache[slot].in != elt) {
+      GLubyte *csr = copy->dstptr;
+      GLuint i;
+
+/*       _mesa_printf("  --> emit to dstelt %d\n", copy->dstbuf_nr); */
+
+      for (i = 0; i < copy->nr_varying; i++) {
+	 const struct gl_client_array *srcarray = copy->varying[i].array;
+	 const GLubyte *srcptr = copy->varying[i].src_ptr + elt * srcarray->StrideB;
+
+	 memcpy(csr, srcptr, copy->varying[i].size);
+	 csr += copy->varying[i].size;
+
+#ifdef NAN_CHECK
+         if (srcarray->Type == GL_FLOAT) {
+            GLuint k;
+            GLfloat *f = (GLfloat *) srcptr;
+            for (k = 0; k < srcarray->Size; k++) {
+               assert(!IS_INF_OR_NAN(f[k]));
+               assert(f[k] <= 1.0e20 && f[k] >= -1.0e20);
+            }
+         }
+#endif
+
+	 if (0) 
+	 {
+	    const GLuint *f = (const GLuint *)srcptr;
+	    GLuint j;
+	    _mesa_printf("  varying %d: ", i);
+	    for(j = 0; j < copy->varying[i].size / 4; j++)
+	       _mesa_printf("%x ", f[j]);
+	    _mesa_printf("\n");
+	 }
+      }
+
+      copy->vert_cache[slot].in = elt;
+      copy->vert_cache[slot].out = copy->dstbuf_nr++;
+      copy->dstptr += copy->vertex_size;
+
+      assert(csr == copy->dstptr);
+      assert(copy->dstptr == (copy->dstbuf +
+                              copy->dstbuf_nr * copy->vertex_size));
+   }
+/*    else */
+/*       _mesa_printf("  --> reuse vertex\n"); */
+   
+/*    _mesa_printf("  --> emit %d\n", copy->vert_cache[slot].out); */
+   copy->dstelt[copy->dstelt_nr++] = copy->vert_cache[slot].out;
+   return check_flush(copy);
+}
+
+
+/**
+ * Called at end of each primitive during replay.
+ */
+static void
+end( struct copy_context *copy, GLboolean end_flag )
+{
+   struct _mesa_prim *prim = &copy->dstprim[copy->dstprim_nr];
+
+/*    _mesa_printf("end (%d)\n", end_flag); */
+
+   prim->end = end_flag;
+   prim->count = copy->dstelt_nr - prim->start;
+
+   if (++copy->dstprim_nr == MAX_PRIM ||
+       check_flush(copy)) 
+      flush(copy);
+}
+
+
+static void
+replay_elts( struct copy_context *copy )
+{
+   GLuint i, j, k;
+   GLboolean split;
+
+   for (i = 0; i < copy->nr_prims; i++) {
+      const struct _mesa_prim *prim = &copy->prim[i];
+      const GLuint start = prim->start;
+      GLuint first, incr;
+
+      switch (prim->mode) {
+	 
+      case GL_LINE_LOOP:
+	 /* Convert to linestrip and emit the final vertex explicitly,
+	  * but only in the resultant strip that requires it.
+	  */
+	 j = 0;
+	 while (j != prim->count) {
+	    begin(copy, GL_LINE_STRIP, prim->begin && j == 0);
+
+	    for (split = GL_FALSE; j != prim->count && !split; j++)
+	       split = elt(copy, start + j);
+
+	    if (j == prim->count) {
+	       /* Done, emit final line.  Split doesn't matter as
+		* it is always raised a bit early so we can emit
+		* the last verts if necessary!
+		*/
+	       if (prim->end) 
+		  (void)elt(copy, start + 0);
+
+	       end(copy, prim->end);
+	    }
+	    else {
+	       /* Wrap
+		*/
+	       assert(split);
+	       end(copy, 0);
+	       j--;
+	    }
+	 }
+	 break;
+
+      case GL_TRIANGLE_FAN:
+      case GL_POLYGON:
+	 j = 2;
+	 while (j != prim->count) {
+	    begin(copy, prim->mode, prim->begin && j == 0);
+
+	    split = elt(copy, start+0); 
+	    assert(!split);
+
+	    split = elt(copy, start+j-1); 
+	    assert(!split);
+
+	    for (; j != prim->count && !split; j++)
+	       split = elt(copy, start+j);
+
+	    end(copy, prim->end && j == prim->count);
+
+	    if (j != prim->count) {
+	       /* Wrapped the primitive, need to repeat some vertices:
+		*/
+	       j -= 1;
+	    }
+	 }
+	 break;
+
+      default:
+	 (void)split_prim_inplace(prim->mode, &first, &incr);
+	 
+	 j = 0;
+	 while (j != prim->count) {
+
+	    begin(copy, prim->mode, prim->begin && j == 0);
+
+	    split = 0;
+	    for (k = 0; k < first; k++, j++)
+	       split |= elt(copy, start+j);
+
+	    assert(!split);
+
+	    for (; j != prim->count && !split; )
+	       for (k = 0; k < incr; k++, j++)
+		  split |= elt(copy, start+j);
+
+	    end(copy, prim->end && j == prim->count);
+
+	    if (j != prim->count) {
+	       /* Wrapped the primitive, need to repeat some vertices:
+		*/
+	       assert(j > first - incr);
+	       j -= (first - incr);
+	    }
+	 }
+	 break;
+      }
+   }
+
+   if (copy->dstprim_nr)
+      flush(copy);
+}
+
+
+static void
+replay_init( struct copy_context *copy )
+{
+   GLcontext *ctx = copy->ctx;
+   GLuint i;
+   GLuint offset;
+   const GLvoid *srcptr;
+
+   /* Make a list of varying attributes and their vbo's.  Also
+    * calculate vertex size.
+    */
+   copy->vertex_size = 0;
+   for (i = 0; i < VERT_ATTRIB_MAX; i++) {
+      struct gl_buffer_object *vbo = copy->array[i]->BufferObj;
+
+      if (copy->array[i]->StrideB == 0) {
+	 copy->dstarray_ptr[i] = copy->array[i];
+      }
+      else {
+	 GLuint j = copy->nr_varying++;
+	 
+	 copy->varying[j].attr = i;
+	 copy->varying[j].array = copy->array[i];
+	 copy->varying[j].size = attr_size(copy->array[i]);
+	 copy->vertex_size += attr_size(copy->array[i]);
+      
+	 if (_mesa_is_bufferobj(vbo) && !_mesa_bufferobj_mapped(vbo)) 
+	    ctx->Driver.MapBuffer(ctx, GL_ARRAY_BUFFER, GL_READ_ONLY, vbo);
+
+	 copy->varying[j].src_ptr = ADD_POINTERS(vbo->Pointer,
+						 copy->array[i]->Ptr);
+
+	 copy->dstarray_ptr[i] = &copy->varying[j].dstarray;
+      }
+   }
+
+   /* There must always be an index buffer.  Currently require the
+    * caller convert non-indexed prims to indexed.  Could alternately
+    * do it internally.
+    */
+   if (_mesa_is_bufferobj(copy->ib->obj) &&
+       !_mesa_bufferobj_mapped(copy->ib->obj)) 
+      ctx->Driver.MapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER, GL_READ_ONLY,
+			    copy->ib->obj);
+
+   srcptr = (const GLubyte *) ADD_POINTERS(copy->ib->obj->Pointer,
+                                           copy->ib->ptr);
+
+   switch (copy->ib->type) {
+   case GL_UNSIGNED_BYTE:
+      copy->translated_elt_buf = _mesa_malloc(sizeof(GLuint) * copy->ib->count);
+      copy->srcelt = copy->translated_elt_buf;
+
+      for (i = 0; i < copy->ib->count; i++)
+	 copy->translated_elt_buf[i] = ((const GLubyte *)srcptr)[i];
+      break;
+
+   case GL_UNSIGNED_SHORT:
+      copy->translated_elt_buf = _mesa_malloc(sizeof(GLuint) * copy->ib->count);
+      copy->srcelt = copy->translated_elt_buf;
+
+      for (i = 0; i < copy->ib->count; i++)
+	 copy->translated_elt_buf[i] = ((const GLushort *)srcptr)[i];
+      break;
+
+   case GL_UNSIGNED_INT:
+      copy->translated_elt_buf = NULL;
+      copy->srcelt = (const GLuint *)srcptr;
+      break;
+   }
+
+   /* Figure out the maximum allowed vertex buffer size:
+    */
+   if (copy->vertex_size * copy->limits->max_verts <= copy->limits->max_vb_size) {
+      copy->dstbuf_size = copy->limits->max_verts;
+   }
+   else {
+      copy->dstbuf_size = copy->limits->max_vb_size / copy->vertex_size;
+   }
+
+   /* Allocate an output vertex buffer:
+    *
+    * XXX:  This should be a VBO!
+    */
+   copy->dstbuf = _mesa_malloc(copy->dstbuf_size * copy->vertex_size);   
+   copy->dstptr = copy->dstbuf;
+
+   /* Setup new vertex arrays to point into the output buffer: 
+    */
+   for (offset = 0, i = 0; i < copy->nr_varying; i++) {
+      const struct gl_client_array *src = copy->varying[i].array;
+      struct gl_client_array *dst = &copy->varying[i].dstarray;
+
+      dst->Size = src->Size;
+      dst->Type = src->Type;
+      dst->Format = GL_RGBA;
+      dst->Stride = copy->vertex_size;
+      dst->StrideB = copy->vertex_size;
+      dst->Ptr = copy->dstbuf + offset;
+      dst->Enabled = GL_TRUE;
+      dst->Normalized = src->Normalized; 
+      dst->BufferObj = ctx->Shared->NullBufferObj;
+      dst->_MaxElement = copy->dstbuf_size; /* may be less! */
+
+      offset += copy->varying[i].size;
+   }
+
+   /* Allocate an output element list:
+    */
+   copy->dstelt_size = MIN2(65536,
+			    copy->ib->count * 2 + 3);
+   copy->dstelt_size = MIN2(copy->dstelt_size,
+			    copy->limits->max_indices);
+   copy->dstelt = _mesa_malloc(sizeof(GLuint) * copy->dstelt_size);
+   copy->dstelt_nr = 0;
+
+   /* Setup the new index buffer to point to the allocated element
+    * list:
+    */
+   copy->dstib.count = 0;	/* duplicates dstelt_nr */
+   copy->dstib.type = GL_UNSIGNED_INT;
+   copy->dstib.obj = ctx->Shared->NullBufferObj;
+   copy->dstib.ptr = copy->dstelt;
+}
+
+
+/**
+ * Free up everything allocated during split/replay.
+ */
+static void
+replay_finish( struct copy_context *copy )
+{
+   GLcontext *ctx = copy->ctx;
+   GLuint i;
+
+   /* Free our vertex and index buffers: 
+    */
+   _mesa_free(copy->translated_elt_buf);
+   _mesa_free(copy->dstbuf);
+   _mesa_free(copy->dstelt);
+
+   /* Unmap VBO's 
+    */
+   for (i = 0; i < copy->nr_varying; i++) {
+      struct gl_buffer_object *vbo = copy->varying[i].array->BufferObj;
+      if (_mesa_is_bufferobj(vbo) && _mesa_bufferobj_mapped(vbo)) 
+	 ctx->Driver.UnmapBuffer(ctx, GL_ARRAY_BUFFER, vbo);
+   }
+
+   /* Unmap index buffer:
+    */
+   if (_mesa_is_bufferobj(copy->ib->obj) &&
+       _mesa_bufferobj_mapped(copy->ib->obj)) {
+      ctx->Driver.UnmapBuffer(ctx, GL_ELEMENT_ARRAY_BUFFER, copy->ib->obj);
+   }
+}
+
+
+/**
+ * Split VBO into smaller pieces, draw the pieces.
+ */
+void vbo_split_copy( GLcontext *ctx,
+		     const struct gl_client_array *arrays[],
+		     const struct _mesa_prim *prim,
+		     GLuint nr_prims,
+		     const struct _mesa_index_buffer *ib,
+		     vbo_draw_func draw,
+		     const struct split_limits *limits )
+{
+   struct copy_context copy;
+   GLuint i;
+
+   memset(&copy, 0, sizeof(copy));
+
+   /* Require indexed primitives:
+    */
+   assert(ib);
+   
+   copy.ctx = ctx;
+   copy.array = arrays;
+   copy.prim = prim;
+   copy.nr_prims = nr_prims;
+   copy.ib = ib;
+   copy.draw = draw;
+   copy.limits = limits;
+
+   /* Clear the vertex cache:
+    */
+   for (i = 0; i < ELT_TABLE_SIZE; i++)
+      copy.vert_cache[i].in = ~0;
+
+   replay_init(&copy);
+   replay_elts(&copy);
+   replay_finish(&copy);
+}
diff --git a/mesalib/src/mesa/vbo/vbo_split_inplace.c b/mesalib/src/mesa/vbo/vbo_split_inplace.c
new file mode 100644
index 000000000..da84eaa6e
--- /dev/null
+++ b/mesalib/src/mesa/vbo/vbo_split_inplace.c
@@ -0,0 +1,271 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  6.5
+ *
+ * Copyright (C) 1999-2006  Brian Paul   All Rights Reserved.
+ *
+ * Permission is hereby granted, free of charge, to any person obtaining a
+ * copy of this software and associated documentation files (the "Software"),
+ * to deal in the Software without restriction, including without limitation
+ * the rights to use, copy, modify, merge, publish, distribute, sublicense,
+ * and/or sell copies of the Software, and to permit persons to whom the
+ * Software is furnished to do so, subject to the following conditions:
+ *
+ * The above copyright notice and this permission notice shall be included
+ * in all copies or substantial portions of the Software.
+ *
+ * THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS
+ * OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+ * FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.  IN NO EVENT SHALL
+ * BRIAN PAUL BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN
+ * AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN
+ * CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+ *
+ * Authors:
+ *    Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+
+#include "main/mtypes.h"
+#include "main/macros.h"
+#include "main/enums.h"
+#include "vbo_split.h"
+
+
+#define MAX_PRIM 32
+
+/* Used for splitting without copying.
+ */
+struct split_context {
+   GLcontext *ctx;
+   const struct gl_client_array **array;
+   const struct _mesa_prim *prim;
+   GLuint nr_prims;
+   const struct _mesa_index_buffer *ib;
+   GLuint min_index;
+   GLuint max_index;
+   vbo_draw_func draw;
+
+   const struct split_limits *limits;
+
+   struct _mesa_prim dstprim[MAX_PRIM];
+   GLuint dstprim_nr;
+};
+
+
+
+
+static void flush_vertex( struct split_context *split )
+{
+   GLuint min_index, max_index;
+   GLuint i;
+
+   if (!split->dstprim_nr) 
+      return;
+
+   min_index = split->dstprim[0].start;
+   max_index = min_index + split->dstprim[0].count - 1;
+
+   for (i = 1; i < split->dstprim_nr; i++) {
+      GLuint tmp_min = split->dstprim[i].start;
+      GLuint tmp_max = tmp_min + split->dstprim[i].count - 1;
+
+      if (tmp_min < min_index)
+	 min_index = tmp_min;
+
+      if (tmp_max > max_index)
+	 max_index = tmp_max;
+   }
+
+   assert(max_index >= min_index);
+
+   split->draw( split->ctx, 
+		split->array, 
+		split->dstprim,
+		split->dstprim_nr,
+		NULL,
+		GL_TRUE,
+		min_index,
+		max_index);
+
+   split->dstprim_nr = 0;
+}
+
+
+static struct _mesa_prim *next_outprim( struct split_context *split )
+{
+   if (split->dstprim_nr == MAX_PRIM-1) {
+      flush_vertex(split);
+   }
+
+   {
+      struct _mesa_prim *prim = &split->dstprim[split->dstprim_nr++];
+      memset(prim, 0, sizeof(*prim));
+      return prim;
+   }
+}
+
+static int align(int value, int alignment)
+{
+   return (value + alignment - 1) & ~(alignment - 1);
+}
+
+
+
+/* Break large primitives into smaller ones.  If not possible, convert
+ * the primitive to indexed and pass to split_elts().
+ */
+static void split_prims( struct split_context *split) 
+{
+   GLuint csr = 0;
+   GLuint i;
+
+   for (i = 0; i < split->nr_prims; i++) {
+      const struct _mesa_prim *prim = &split->prim[i];
+      GLuint first, incr;
+      GLboolean split_inplace = split_prim_inplace(prim->mode, &first, &incr);
+      GLuint count;
+
+      /* Always wrap on an even numbered vertex to avoid problems with
+       * triangle strips.  
+       */
+      GLuint available = align(split->limits->max_verts - csr - 1, 2); 
+      assert(split->limits->max_verts >= csr);
+
+      if (prim->count < first)
+	 continue;
+      
+      count = prim->count - (prim->count - first) % incr; 
+
+
+      if ((available < count && !split_inplace) || 
+	  (available < first && split_inplace)) {
+	 flush_vertex(split);
+	 csr = 0;
+	 available = align(split->limits->max_verts - csr - 1, 2);
+      }
+      
+      if (available >= count) {
+	 struct _mesa_prim *outprim = next_outprim(split);
+	 *outprim = *prim;
+	 csr += prim->count;
+	 available = align(split->limits->max_verts - csr - 1, 2); 
+      } 
+      else if (split_inplace) {
+	 GLuint j, nr;
+
+
+	 for (j = 0 ; j < count ; ) {
+	    GLuint remaining = count - j;
+	    struct _mesa_prim *outprim = next_outprim(split);
+
+	    nr = MIN2( available, remaining );
+	    
+	    nr -= (nr - first) % incr;
+	    
+	    outprim->mode = prim->mode;
+	    outprim->begin = (j == 0 && prim->begin);
+	    outprim->end = (nr == remaining && prim->end);
+	    outprim->start = prim->start + j;
+	    outprim->count = nr;
+	    
+	    if (nr == remaining) {
+	       /* Finished. 
+		*/
+	       j += nr;		
+	       csr += nr;
+	       available = align(split->limits->max_verts - csr - 1, 2); 
+	    }
+	    else {
+	       /* Wrapped the primitive: 
+		*/
+	       j += nr - (first - incr);
+	       flush_vertex(split);
+	       csr = 0;
+	       available = align(split->limits->max_verts - csr - 1, 2); 
+	    }
+	 }
+      }
+      else if (split->ib == NULL) {
+	 /* XXX: could at least send the first max_verts off from the
+	  * inplace buffers.
+	  */
+
+	 /* else convert to indexed primitive and pass to split_elts,
+	  * which will do the necessary copying and turn it back into a
+	  * vertex primitive for rendering...
+	  */
+	 struct _mesa_index_buffer ib;
+	 struct _mesa_prim tmpprim;
+	 GLuint *elts = malloc(count * sizeof(GLuint));
+	 GLuint j;
+	 
+	 for (j = 0; j < count; j++)
+	    elts[j] = prim->start + j;
+
+	 ib.count = count;
+	 ib.type = GL_UNSIGNED_INT;
+	 ib.obj = split->ctx->Shared->NullBufferObj;
+	 ib.ptr = elts;
+	    
+	 tmpprim = *prim;
+	 tmpprim.indexed = 1;
+	 tmpprim.start = 0;
+	 tmpprim.count = count;
+
+	 flush_vertex(split);
+
+	 vbo_split_copy(split->ctx,
+			split->array,
+			&tmpprim, 1, 
+			&ib,
+			split->draw,
+			split->limits);
+	    
+	 free(elts);
+      }
+      else {
+	 flush_vertex(split);
+
+	 vbo_split_copy(split->ctx,
+			split->array,
+			prim, 1, 
+			split->ib,
+			split->draw,
+			split->limits);
+      }
+   }
+
+   flush_vertex(split);
+}
+
+
+void vbo_split_inplace( GLcontext *ctx,
+			const struct gl_client_array *arrays[],
+			const struct _mesa_prim *prim,
+			GLuint nr_prims,
+			const struct _mesa_index_buffer *ib,
+			GLuint min_index,
+			GLuint max_index,
+			vbo_draw_func draw,
+			const struct split_limits *limits )
+{
+   struct split_context split;
+
+   memset(&split, 0, sizeof(split));
+
+   split.ctx = ctx;
+   split.array = arrays;
+   split.prim = prim;
+   split.nr_prims = nr_prims;
+   split.ib = ib;
+   split.min_index = min_index;
+   split.max_index = max_index;
+   split.draw = draw;
+   split.limits = limits;
+
+   split_prims( &split );
+}
+
+