patch to import Jon Smirl's work from Bitkeeper

126 files changed, 42756 insertions, 2491 deletions

diff --git a/src/mesa/Makefile.X11 b/src/mesa/Makefile.X11
index c672cce47a5..d014936d366 100644
--- a/src/mesa/Makefile.X11
+++ b/src/mesa/Makefile.X11
@@ -296,7 +296,7 @@ default:
 	@echo "Specify a target configuration"
 
 
-targets: $(LIBDIR)/$(GL_LIB) $(LIBDIR)/$(OSMESA_LIB)
+targets: $(LIBDIR)/$(GL_LIB) $(LIBDIR)/$(OSMESA_LIB) $(LIBMESA)
 
 
 # Make the GL library
@@ -321,7 +321,9 @@ libmesa: $(LIBDIR)/$(MESA_LIB)
 
 $(LIBDIR)/$(MESA_LIB):	$(CORE_OBJECTS)
 
-
+mesa.a:	$(CORE_OBJECTS)
+	rm -f $@ && ar rcv $@ $(CORE_OBJECTS) && ranlib $@
+  
 # Run 'make -f Makefile.X11 dep' to update the dependencies if you change
 # what's included by any source file.
 dep: $(CORE_SOURCES) $(DRIVER_SOURCES) $(OSMESA_DRIVER_SOURCES) $(ASM_SOURCES)
@@ -336,9 +338,9 @@ tags:
 
 # Remove .o and backup files
 clean:
+	-rm *.a
 	-rm -f */*.o */*~ */*.o */*~
 	-rm -f drivers/*/*.o
-	-rm -f drivers/dri/*/*.o
 
 
 include $(TOP)/Make-config
diff --git a/src/mesa/drivers/dri/mga/Doxyfile b/src/mesa/drivers/dri/mga/Doxyfile
new file mode 100644
index 00000000000..0d0c134a728
--- /dev/null
+++ b/src/mesa/drivers/dri/mga/Doxyfile
@@ -0,0 +1,234 @@
+# Doxyfile 1.3.3-Gideon
+
+#---------------------------------------------------------------------------
+# General configuration options
+#---------------------------------------------------------------------------
+PROJECT_NAME           = mga
+PROJECT_NUMBER         = $VERSION$
+OUTPUT_DIRECTORY       = 
+OUTPUT_LANGUAGE        = English
+USE_WINDOWS_ENCODING   = NO
+EXTRACT_ALL            = NO
+EXTRACT_PRIVATE        = NO
+EXTRACT_STATIC         = NO
+EXTRACT_LOCAL_CLASSES  = YES
+HIDE_UNDOC_MEMBERS     = NO
+HIDE_UNDOC_CLASSES     = NO
+HIDE_FRIEND_COMPOUNDS  = NO
+HIDE_IN_BODY_DOCS      = NO
+BRIEF_MEMBER_DESC      = YES
+REPEAT_BRIEF           = YES
+ALWAYS_DETAILED_SEC    = NO
+INLINE_INHERITED_MEMB  = NO
+FULL_PATH_NAMES        = NO
+STRIP_FROM_PATH        = 
+INTERNAL_DOCS          = NO
+CASE_SENSE_NAMES       = YES
+SHORT_NAMES            = NO
+HIDE_SCOPE_NAMES       = NO
+SHOW_INCLUDE_FILES     = YES
+JAVADOC_AUTOBRIEF      = NO
+MULTILINE_CPP_IS_BRIEF = NO
+DETAILS_AT_TOP         = NO
+INHERIT_DOCS           = YES
+INLINE_INFO            = YES
+SORT_MEMBER_DOCS       = YES
+DISTRIBUTE_GROUP_DOC   = NO
+TAB_SIZE               = 8
+GENERATE_TODOLIST      = YES
+GENERATE_TESTLIST      = YES
+GENERATE_BUGLIST       = YES
+GENERATE_DEPRECATEDLIST= YES
+ALIASES                = 
+ENABLED_SECTIONS       = 
+MAX_INITIALIZER_LINES  = 30
+OPTIMIZE_OUTPUT_FOR_C  = NO
+OPTIMIZE_OUTPUT_JAVA   = NO
+SHOW_USED_FILES        = YES
+SUBGROUPING            = YES
+#---------------------------------------------------------------------------
+# configuration options related to warning and progress messages
+#---------------------------------------------------------------------------
+QUIET                  = NO
+WARNINGS               = YES
+WARN_IF_UNDOCUMENTED   = YES
+WARN_IF_DOC_ERROR      = YES
+WARN_FORMAT            = "$file:$line: $text"
+WARN_LOGFILE           = 
+#---------------------------------------------------------------------------
+# configuration options related to the input files
+#---------------------------------------------------------------------------
+INPUT                  = /home/newtree/temp/src/mesa/drivers/dri/mga
+FILE_PATTERNS          = *.c \
+                         *.cc \
+                         *.cxx \
+                         *.cpp \
+                         *.c++ \
+                         *.java \
+                         *.ii \
+                         *.ixx \
+                         *.ipp \
+                         *.i++ \
+                         *.inl \
+                         *.h \
+                         *.hh \
+                         *.hxx \
+                         *.hpp \
+                         *.h++ \
+                         *.idl \
+                         *.odl \
+                         *.cs \
+                         *.C \
+                         *.H \
+                         *.tlh \
+                         *.diff \
+                         *.patch \
+                         *.moc \
+                         *.xpm \
+                         *.dox
+RECURSIVE              = yes
+EXCLUDE                = 
+EXCLUDE_SYMLINKS       = NO
+EXCLUDE_PATTERNS       = 
+EXAMPLE_PATH           = 
+EXAMPLE_PATTERNS       = *
+EXAMPLE_RECURSIVE      = NO
+IMAGE_PATH             = 
+INPUT_FILTER           = 
+FILTER_SOURCE_FILES    = NO
+#---------------------------------------------------------------------------
+# configuration options related to source browsing
+#---------------------------------------------------------------------------
+SOURCE_BROWSER         = NO
+INLINE_SOURCES         = NO
+STRIP_CODE_COMMENTS    = YES
+REFERENCED_BY_RELATION = YES
+REFERENCES_RELATION    = YES
+VERBATIM_HEADERS       = YES
+#---------------------------------------------------------------------------
+# configuration options related to the alphabetical class index
+#---------------------------------------------------------------------------
+ALPHABETICAL_INDEX     = NO
+COLS_IN_ALPHA_INDEX    = 5
+IGNORE_PREFIX          = 
+#---------------------------------------------------------------------------
+# configuration options related to the HTML output
+#---------------------------------------------------------------------------
+GENERATE_HTML          = YES
+HTML_OUTPUT            = html
+HTML_FILE_EXTENSION    = .html
+HTML_HEADER            = 
+HTML_FOOTER            = 
+HTML_STYLESHEET        = 
+HTML_ALIGN_MEMBERS     = YES
+GENERATE_HTMLHELP      = NO
+CHM_FILE               = 
+HHC_LOCATION           = 
+GENERATE_CHI           = NO
+BINARY_TOC             = NO
+TOC_EXPAND             = NO
+DISABLE_INDEX          = NO
+ENUM_VALUES_PER_LINE   = 4
+GENERATE_TREEVIEW      = NO
+TREEVIEW_WIDTH         = 250
+#---------------------------------------------------------------------------
+# configuration options related to the LaTeX output
+#---------------------------------------------------------------------------
+GENERATE_LATEX         = YES
+LATEX_OUTPUT           = latex
+LATEX_CMD_NAME         = latex
+MAKEINDEX_CMD_NAME     = makeindex
+COMPACT_LATEX          = NO
+PAPER_TYPE             = a4wide
+EXTRA_PACKAGES         = 
+LATEX_HEADER           = 
+PDF_HYPERLINKS         = NO
+USE_PDFLATEX           = NO
+LATEX_BATCHMODE        = NO
+LATEX_HIDE_INDICES     = NO
+#---------------------------------------------------------------------------
+# configuration options related to the RTF output
+#---------------------------------------------------------------------------
+GENERATE_RTF           = NO
+RTF_OUTPUT             = rtf
+COMPACT_RTF            = NO
+RTF_HYPERLINKS         = NO
+RTF_STYLESHEET_FILE    = 
+RTF_EXTENSIONS_FILE    = 
+#---------------------------------------------------------------------------
+# configuration options related to the man page output
+#---------------------------------------------------------------------------
+GENERATE_MAN           = NO
+MAN_OUTPUT             = man
+MAN_EXTENSION          = .3
+MAN_LINKS              = NO
+#---------------------------------------------------------------------------
+# configuration options related to the XML output
+#---------------------------------------------------------------------------
+GENERATE_XML           = yes
+XML_OUTPUT             = xml
+XML_SCHEMA             = 
+XML_DTD                = 
+#---------------------------------------------------------------------------
+# configuration options for the AutoGen Definitions output
+#---------------------------------------------------------------------------
+GENERATE_AUTOGEN_DEF   = NO
+#---------------------------------------------------------------------------
+# configuration options related to the Perl module output
+#---------------------------------------------------------------------------
+GENERATE_PERLMOD       = NO
+PERLMOD_LATEX          = NO
+PERLMOD_PRETTY         = YES
+PERLMOD_MAKEVAR_PREFIX = 
+#---------------------------------------------------------------------------
+# Configuration options related to the preprocessor   
+#---------------------------------------------------------------------------
+ENABLE_PREPROCESSING   = YES
+MACRO_EXPANSION        = NO
+EXPAND_ONLY_PREDEF     = NO
+SEARCH_INCLUDES        = YES
+INCLUDE_PATH           = 
+INCLUDE_FILE_PATTERNS  = 
+PREDEFINED             = 
+EXPAND_AS_DEFINED      = 
+SKIP_FUNCTION_MACROS   = YES
+#---------------------------------------------------------------------------
+# Configuration::addtions related to external references   
+#---------------------------------------------------------------------------
+TAGFILES               = 
+GENERATE_TAGFILE       = 
+ALLEXTERNALS           = NO
+EXTERNAL_GROUPS        = YES
+PERL_PATH              = /usr/bin/perl
+#---------------------------------------------------------------------------
+# Configuration options related to the dot tool   
+#---------------------------------------------------------------------------
+CLASS_DIAGRAMS         = YES
+HIDE_UNDOC_RELATIONS   = YES
+HAVE_DOT               = NO
+CLASS_GRAPH            = YES
+COLLABORATION_GRAPH    = YES
+UML_LOOK               = NO
+TEMPLATE_RELATIONS     = NO
+INCLUDE_GRAPH          = YES
+INCLUDED_BY_GRAPH      = YES
+CALL_GRAPH             = NO
+GRAPHICAL_HIERARCHY    = YES
+DOT_IMAGE_FORMAT       = png
+DOT_PATH               = 
+DOTFILE_DIRS           = 
+MAX_DOT_GRAPH_WIDTH    = 1024
+MAX_DOT_GRAPH_HEIGHT   = 1024
+MAX_DOT_GRAPH_DEPTH    = 1000
+GENERATE_LEGEND        = YES
+DOT_CLEANUP            = YES
+#---------------------------------------------------------------------------
+# Configuration::addtions related to the search engine   
+#---------------------------------------------------------------------------
+SEARCHENGINE           = NO
+CGI_NAME               = search.cgi
+CGI_URL                = 
+DOC_URL                = 
+DOC_ABSPATH            = 
+BIN_ABSPATH            = /usr/local/bin/
+EXT_DOC_PATHS          = 
diff --git a/src/mesa/drivers/dri/mga/Makefile.X11 b/src/mesa/drivers/dri/mga/Makefile.X11
index c69d6e5989e..0dce6e7b1a7 100644
--- a/src/mesa/drivers/dri/mga/Makefile.X11
+++ b/src/mesa/drivers/dri/mga/Makefile.X11
@@ -1,4 +1,4 @@
-# $Id: Makefile.X11,v 1.1 2003/08/06 18:01:13 keithw Exp $
+# $Id: Makefile.X11,v 1.2 2003/08/22 20:11:44 brianp Exp $
 
 # Mesa 3-D graphics library
 # Version:  5.0
@@ -6,8 +6,10 @@
 
 TOP = ../../../../..
 
+default: linux-solo
+
 SHARED_INCLUDES = $(INCLUDE_DIRS) -I. -I../common -Iserver
-MINIGLX_INCLUDES = -I$(TOP)/src/miniglx 
+MINIGLX_INCLUDES = -I$(TOP)/src/glx/mini
 
 DEFINES += \
 	-D_HAVE_SWRAST=1 \
@@ -16,24 +18,27 @@ DEFINES += \
 	-D_HAVE_CODEGEN=1 \
 	-D_HAVE_LIGHTING=1 \
 	-D_HAVE_TEXGEN=1 \
-	-D_HAVE_USERCLIP=1
+	-D_HAVE_USERCLIP=1 \
+	-DGLX_DIRECT_RENDERING
 
 MINIGLX_SOURCES = server/mga_dri.c 
 
-DRIVER_SOURCES = mgabuffers.c \
-		 mgadd.c \
+DRIVER_SOURCES = mgadd.c \
 		 mgaioctl.c \
 		 mgarender.c \
 		 mgastate.c \
 		 mgatris.c \
-		 ../common/mm.c 
+		 ../common/mm.c \
+		 ../common/utils.c \
+		 ../common/texmem.c \
+		 ../common/vblank.c
 
 FULL_DRIVER_SOURCES = 	\
 		 mgapixel.c \
 		 mgaspan.c \
 		 mgatex.c \
-		 mgatexcnv.c \
 		 mgatexmem.c \
+		 mga_texstate.c \
 		 mgavb.c \
 		 mga_xmesa.c
 
@@ -54,7 +59,7 @@ WINOBJ=$(MESABUILDDIR)/dri/dri.a
 WINLIB=
 else
 WINOBJ=
-WINLIB=-L$(MESA)/src/miniglx
+WINLIB=-L$(MESA)/src/glx/mini
 endif
 
 ASM_SOURCES = 
@@ -87,7 +92,7 @@ INCLUDE_DIRS = \
 
 ##### TARGETS #####
 
-targets: mga_dri.so
+targets: depend mga_dri.so
 
 mga_dri.so:  $(OBJECTS) $(MESA_MODULES) $(WINOBJ) Makefile.X11
 	rm -f $@ && gcc -o $@ -shared $(OBJECTS) $(MESA_MODULES) $(WINOBJ) $(WINLIB) -lc -lm
@@ -96,7 +101,7 @@ mga_dri.so:  $(OBJECTS) $(MESA_MODULES) $(WINOBJ) Makefile.X11
 
 # Run 'make -f Makefile.X11 dep' to update the dependencies if you change
 # what's included by any source file.
-dep: $(C_SOURCES) $(ASM_SOURCES)
+depend: $(C_SOURCES) $(ASM_SOURCES)
 	makedepend -fdepend -Y $(SHARED_INCLUDES) \
 		$(C_SOURCES) $(ASM_SOURCES)
 
@@ -108,7 +113,7 @@ tags:
 
 # Remove .o and backup files
 clean:
-	-rm -f *.o *~ *.o *~ *.so
+	-rm -f *.o *~ *.o *~ *.so server/*.o
 
 
 include $(TOP)/Make-config
diff --git a/src/mesa/drivers/dri/mga/mga_texstate.c b/src/mesa/drivers/dri/mga/mga_texstate.c
new file mode 100644
index 00000000000..fc1406cab9e
--- /dev/null
+++ b/src/mesa/drivers/dri/mga/mga_texstate.c
@@ -0,0 +1,747 @@
+/*
+ * Copyright 2000-2001 VA Linux Systems, Inc.
+ * (c) Copyright IBM Corporation 2002
+ * 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
+ * VA LINUX SYSTEMS, IBM 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:
+ *    Ian Romanick <[email protected]>
+ *    Keith Whitwell <[email protected]>
+ */
+/* $XFree86:$ */
+
+#include "mm.h"
+#include "mgacontext.h"
+#include "mgatex.h"
+#include "mgaregs.h"
+#include "mgatris.h"
+#include "mgaioctl.h"
+
+#include "context.h"
+#include "enums.h"
+#include "macros.h"
+#include "imports.h"
+
+#include "simple_list.h"
+#include "texformat.h"
+
+#define MGA_USE_TABLE_FOR_FORMAT
+#ifdef MGA_USE_TABLE_FOR_FORMAT
+#define TMC_nr_tformat (MESA_FORMAT_YCBCR_REV + 1)
+static const unsigned TMC_tformat[ TMC_nr_tformat ] =
+{
+    [MESA_FORMAT_ARGB8888] = TMC_tformat_tw32 | TMC_takey_1 | TMC_tamask_0,
+    [MESA_FORMAT_RGB565]   = TMC_tformat_tw16 | TMC_takey_1 | TMC_tamask_0,
+    [MESA_FORMAT_ARGB4444] = TMC_tformat_tw12 | TMC_takey_1 | TMC_tamask_0,
+    [MESA_FORMAT_ARGB1555] = TMC_tformat_tw15 | TMC_takey_1 | TMC_tamask_0,
+    [MESA_FORMAT_CI8]      = TMC_tformat_tw8  | TMC_takey_1 | TMC_tamask_0,
+    [MESA_FORMAT_YCBCR]     = TMC_tformat_tw422uyvy | TMC_takey_1 | TMC_tamask_0,
+    [MESA_FORMAT_YCBCR_REV] = TMC_tformat_tw422 | TMC_takey_1 | TMC_tamask_0,
+};
+#endif
+
+static void
+mgaSetTexImages( mgaContextPtr mmesa,
+		 const struct gl_texture_object * tObj )
+{
+    mgaTextureObjectPtr t = (mgaTextureObjectPtr) tObj->DriverData;
+    struct gl_texture_image *baseImage = tObj->Image[ tObj->BaseLevel ];
+    GLint totalSize;
+    GLint width, height;
+    GLint i;
+    GLint firstLevel, lastLevel, numLevels;
+    GLint log2Width, log2Height;
+    GLuint txformat = 0;
+    GLint ofs;
+
+    /* Set the hardware texture format
+     */
+#ifndef MGA_USE_TABLE_FOR_FORMAT
+    switch (baseImage->TexFormat->MesaFormat) {
+
+	case MESA_FORMAT_ARGB8888: txformat = TMC_tformat_tw32;	break;
+	case MESA_FORMAT_RGB565:   txformat = TMC_tformat_tw16; break;
+	case MESA_FORMAT_ARGB4444: txformat = TMC_tformat_tw12;	break;
+	case MESA_FORMAT_ARGB1555: txformat = TMC_tformat_tw15; break;
+	case MESA_FORMAT_CI8:      txformat = TMC_tformat_tw8;  break;
+        case MESA_FORMAT_YCBCR:    txformat  = TMC_tformat_tw422uyvy; break;
+        case MESA_FORMAT_YCBCR_REV: txformat = TMC_tformat_tw422; break;
+
+	default:
+	_mesa_problem(NULL, "unexpected texture format in %s", __FUNCTION__);
+	return;
+    }
+#else
+    if ( (baseImage->TexFormat->MesaFormat >= TMC_nr_tformat)
+	 || (TMC_tformat[ baseImage->TexFormat->MesaFormat ] == 0) )
+    {
+	_mesa_problem(NULL, "unexpected texture format in %s", __FUNCTION__);
+	return;
+    }
+
+    txformat = TMC_tformat[ baseImage->TexFormat->MesaFormat ];
+
+#endif /* MGA_USE_TABLE_FOR_FORMAT */
+
+   if (tObj->MinFilter == GL_NEAREST || tObj->MinFilter == GL_LINEAR) {
+      /* GL_NEAREST and GL_LINEAR only care about GL_TEXTURE_BASE_LEVEL.
+       */
+
+      firstLevel = lastLevel = tObj->BaseLevel;
+   } else {
+      /* Compute which mipmap levels we really want to send to the hardware.
+       * This depends on the base image size, GL_TEXTURE_MIN_LOD,
+       * GL_TEXTURE_MAX_LOD, GL_TEXTURE_BASE_LEVEL, and GL_TEXTURE_MAX_LEVEL.
+       * Yes, this looks overly complicated, but it's all needed.
+       */
+
+      firstLevel = tObj->BaseLevel + (GLint)(tObj->MinLod + 0.5);
+      firstLevel = MAX2(firstLevel, tObj->BaseLevel);
+      lastLevel = tObj->BaseLevel + (GLint)(tObj->MaxLod + 0.5);
+      lastLevel = MAX2(lastLevel, tObj->BaseLevel);
+      lastLevel = MIN2(lastLevel, tObj->BaseLevel + baseImage->MaxLog2);
+      lastLevel = MIN2(lastLevel, tObj->MaxLevel);
+      lastLevel = MAX2(firstLevel, lastLevel); /* need at least one level */
+   }
+
+   log2Width = tObj->Image[firstLevel]->WidthLog2;
+   log2Height = tObj->Image[firstLevel]->HeightLog2;
+   width = tObj->Image[firstLevel]->Width;
+   height = tObj->Image[firstLevel]->Height;
+
+   numLevels = MIN2( lastLevel - firstLevel + 1,
+                     MGA_IS_G200(mmesa) ? G200_TEX_MAXLEVELS : G400_TEX_MAXLEVELS);
+
+
+   totalSize = 0;
+   for ( i = 0 ; i < numLevels ; i++ ) {
+      const struct gl_texture_image * const texImage = tObj->Image[i+firstLevel];
+
+      if ( (texImage == NULL)
+	   || ((i != 0)
+	       && ((texImage->Width < 8) || (texImage->Height < 8))) ) {
+	 break;
+      }
+
+      t->offsets[i] = totalSize;
+      t->base.dirty_images[0] |= (1<<i);
+
+      totalSize += ((MAX2( texImage->Width, 8 ) *
+                     MAX2( texImage->Height, 8 ) *
+                     baseImage->TexFormat->TexelBytes) + 31) & ~31;
+   }
+
+   numLevels = i;
+   lastLevel = firstLevel + numLevels - 1;
+
+   /* save these values */
+   t->base.firstLevel = firstLevel;
+   t->base.lastLevel = lastLevel;
+
+   t->base.totalSize = totalSize;
+
+   /* setup hardware register values */
+   t->setup.texctl &= (TMC_tformat_MASK & TMC_tpitch_MASK 
+		       & TMC_tpitchext_MASK);
+   t->setup.texctl |= txformat;
+
+
+   /* Set the texture width.  In order to support non-power of 2 textures and
+    * textures larger than 1024 texels wide, "linear" pitch must be used.  For
+    * the linear pitch, if the width is 2048, a value of zero is used.
+    */
+
+   t->setup.texctl |= TMC_tpitchlin_enable;
+   t->setup.texctl |= (width & (2048 - 1)) << TMC_tpitchext_SHIFT;
+
+
+   /* G400 specifies the number of mip levels in a strange way.  Since there
+    * are up to 12 levels, it requires 4 bits.  Three of the bits are at the
+    * high end of TEXFILTER.  The other bit is in the middle.  Weird.
+    */
+
+   t->setup.texfilter &= TF_mapnb_MASK & TF_mapnbhigh_MASK & TF_reserved_MASK;
+   t->setup.texfilter |= (((numLevels-1) & 0x07) << (TF_mapnb_SHIFT));
+   t->setup.texfilter |= (((numLevels-1) & 0x08) << (TF_mapnbhigh_SHIFT - 3));
+
+   /* warp texture registers */
+   ofs = MGA_IS_G200(mmesa) ? 28 : 11;
+
+   t->setup.texwidth = (MGA_FIELD(TW_twmask, width - 1) |
+			MGA_FIELD(TW_rfw, (10 - log2Width - 8) & 63 ) |
+			MGA_FIELD(TW_tw, (log2Width + ofs ) | 0x40 ));
+
+   t->setup.texheight = (MGA_FIELD(TH_thmask, height - 1) |
+			 MGA_FIELD(TH_rfh, (10 - log2Height - 8) & 63 ) |
+			 MGA_FIELD(TH_th, (log2Height + ofs ) | 0x40 ));
+
+   mgaUploadTexImages( mmesa, t );
+}
+
+
+/* ================================================================
+ * Texture unit state management
+ */
+
+static void mgaUpdateTextureEnvG200( GLcontext *ctx, GLuint unit )
+{
+   struct gl_texture_object *tObj = ctx->Texture.Unit[0]._Current;
+   mgaTextureObjectPtr t;
+
+   if (!tObj || !tObj->DriverData)
+      return;
+
+   t = (mgaTextureObjectPtr)tObj->DriverData;
+
+   t->setup.texctl2 &= ~TMC_decalblend_enable;
+
+   switch (ctx->Texture.Unit[0].EnvMode) {
+   case GL_REPLACE:
+      t->setup.texctl &= ~TMC_tmodulate_enable;
+      break;
+   case GL_MODULATE:
+      t->setup.texctl |= TMC_tmodulate_enable;
+      break;
+   case GL_DECAL:
+      t->setup.texctl &= ~TMC_tmodulate_enable;
+      t->setup.texctl2 |= TMC_decalblend_enable;
+      break;
+   case GL_BLEND:
+      FALLBACK( ctx, MGA_FALLBACK_TEXTURE, GL_TRUE );
+      break;
+   default:
+      break;
+   }
+}
+
+
+#define MGA_DISABLE		0
+#define MGA_REPLACE		1
+#define MGA_MODULATE		2
+#define MGA_DECAL		3
+#define MGA_BLEND		4
+#define MGA_ADD			5
+#define MGA_MAX_COMBFUNC	6
+
+static const GLuint g400_color_combine[][MGA_MAX_COMBFUNC] =
+{
+   /* Unit 0:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (0),
+
+      /* GL_REPLACE
+       */
+      (TD0_color_sel_arg1 |
+       TD0_alpha_arg2_diffuse |
+       TD0_alpha_sel_arg2 ),
+      
+      /* GL_MODULATE
+       */
+      (TD0_color_arg2_diffuse |
+       TD0_color_sel_mul |
+       TD0_alpha_arg2_diffuse |
+       TD0_alpha_sel_mul),
+      
+      /* GL_DECAL
+       */
+      (TD0_color_sel_arg1 |
+       TD0_alpha_arg2_diffuse |
+       TD0_alpha_sel_arg2),
+      
+      /* GL_BLEND
+       */
+      (0),
+      
+      /* GL_ADD
+       */
+      (TD0_color_arg2_diffuse |
+       TD0_color_add_add |
+       TD0_color_sel_add |
+       TD0_alpha_arg2_diffuse |
+       TD0_alpha_sel_mul),
+   },
+   
+   /* Unit 1:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (0),
+       
+      /* GL_REPLACE
+       */
+      (TD0_color_sel_arg1 |
+       TD0_alpha_arg2_diffuse |
+       TD0_alpha_sel_arg2 ),
+      
+      /* GL_MODULATE
+       */
+      (TD0_color_arg2_prevstage |
+       TD0_color_alpha_prevstage |
+       TD0_color_sel_mul |
+       TD0_alpha_arg2_prevstage |
+       TD0_alpha_sel_mul),
+
+      /* GL_DECAL
+       */
+      (TD0_color_sel_arg1 |
+       TD0_alpha_arg2_prevstage |
+       TD0_alpha_sel_arg2 ),
+      
+      /* GL_BLEND
+       */
+      (0),
+      
+      /* GL_ADD
+       */
+      (TD0_color_arg2_prevstage |
+       TD0_color_alpha_prevstage |
+       TD0_color_add_add |
+       TD0_color_sel_add |
+       TD0_alpha_arg2_prevstage |
+       TD0_alpha_sel_mul),
+   },
+};
+
+static const GLuint g400_alpha_combine[][MGA_MAX_COMBFUNC] =
+{
+   /* Unit 0:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (0),
+
+      /* GL_REPLACE
+       */
+      (TD0_color_sel_arg2 |
+       TD0_color_arg2_diffuse |
+       TD0_alpha_sel_arg1 ),
+      
+      /* GL_MODULATE
+       * FIXME: Is this correct?
+       */
+      (TD0_color_arg2_diffuse |
+       TD0_color_sel_mul |
+       TD0_alpha_arg2_diffuse |
+       TD0_alpha_sel_mul),
+
+      /* GL_DECAL
+       */
+      (TD0_color_arg2_diffuse |
+       TD0_color_sel_arg2 |
+       TD0_alpha_arg2_diffuse |
+       TD0_alpha_sel_arg2),
+
+      /* GL_BLEND
+       */
+      (TD0_color_arg2_diffuse |
+       TD0_color_sel_mul |
+       TD0_alpha_arg2_diffuse |
+       TD0_alpha_sel_mul),
+
+      /* GL_ADD
+       */
+      (TD0_color_arg2_diffuse |
+       TD0_color_sel_mul |
+       TD0_alpha_arg2_diffuse |
+       TD0_alpha_sel_mul),
+   },
+
+   /* Unit 1:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (0),
+
+      /* GL_REPLACE
+       */
+      (TD0_color_sel_arg2 |
+       TD0_color_arg2_diffuse |
+       TD0_alpha_sel_arg1 ),
+      
+      /* GL_MODULATE
+       * FIXME: Is this correct?
+       */
+      (TD0_color_arg2_prevstage |
+       TD0_color_alpha_prevstage |
+       TD0_color_sel_mul |
+       TD0_alpha_arg2_prevstage |
+       TD0_alpha_sel_mul),
+
+      /* GL_DECAL
+       */
+      (TD0_color_arg2_prevstage |
+       TD0_color_sel_arg2 |
+       TD0_alpha_arg2_prevstage |
+       TD0_alpha_sel_arg2),
+
+      /* GL_BLEND
+       */
+      (TD0_color_arg2_diffuse |
+       TD0_color_sel_mul |
+       TD0_alpha_arg2_diffuse |
+       TD0_alpha_sel_mul),
+
+      /* GL_ADD
+       */
+      (TD0_color_arg2_prevstage |
+       TD0_color_sel_mul |
+       TD0_alpha_arg2_prevstage |
+       TD0_alpha_sel_mul),
+   },
+};
+
+static void mgaUpdateTextureEnvG400( GLcontext *ctx, GLuint unit )
+{
+   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
+   const int source = mmesa->tmu_source[unit];
+   const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[source];
+   const struct gl_texture_object *tObj = texUnit->_Current;
+   GLuint *reg = ((GLuint *)&mmesa->setup.tdualstage0 + unit);
+   GLenum format;
+
+   if ( tObj != ctx->Texture.Unit[source].Current2D || !tObj ) 
+      return;
+
+   format = tObj->Image[tObj->BaseLevel]->Format;
+
+   switch (ctx->Texture.Unit[source].EnvMode) {
+   case GL_REPLACE:
+      if (format == GL_RGB || format == GL_LUMINANCE) {
+	 *reg = g400_color_combine[unit][MGA_REPLACE];
+      }
+      else if (format == GL_ALPHA) {
+         *reg = g400_alpha_combine[unit][MGA_REPLACE];
+      }
+      else {
+         *reg = (TD0_color_sel_arg1 |
+                 TD0_alpha_sel_arg1 );
+      }
+      break;
+
+   case GL_MODULATE:
+      *reg = g400_color_combine[unit][MGA_MODULATE];
+      break;
+   case GL_DECAL:
+      if (format == GL_RGB) {
+	 *reg = g400_color_combine[unit][MGA_DECAL];
+      }
+      else if ( format == GL_RGBA ) {
+#if 0
+         if (unit == 0) {
+            /* this doesn't work */
+            *reg = (TD0_color_arg2_diffuse |
+                    TD0_color_alpha_currtex |
+                    TD0_color_alpha2inv_enable |
+                    TD0_color_arg2mul_alpha2 |
+                    TD0_color_arg1mul_alpha1 |
+                    TD0_color_blend_enable |
+                    TD0_color_arg1add_mulout |
+                    TD0_color_arg2add_mulout |
+                    TD0_color_add_add |
+                    TD0_color_sel_mul |
+                    TD0_alpha_arg2_diffuse |
+                    TD0_alpha_sel_arg2 );
+         }
+         else {
+            *reg = (TD0_color_arg2_prevstage |
+                    TD0_color_alpha_currtex |
+                    TD0_color_alpha2inv_enable |
+                    TD0_color_arg2mul_alpha2 |
+                    TD0_color_arg1mul_alpha1 |
+                    TD0_color_add_add |
+                    TD0_color_sel_add |
+                    TD0_alpha_arg2_prevstage |
+                    TD0_alpha_sel_arg2 );
+         }
+#else
+         /* s/w fallback, pretty sure we can't do in h/w */
+	 FALLBACK( ctx, MGA_FALLBACK_TEXTURE, GL_TRUE );
+	 if ( MGA_DEBUG & DEBUG_VERBOSE_FALLBACK )
+	    fprintf( stderr, "FALLBACK: GL_DECAL RGBA texture, unit=%d\n",
+		     unit );
+#endif
+      }
+      else {
+	*reg = g400_alpha_combine[unit][MGA_DECAL];
+      }
+      break;
+
+   case GL_ADD:
+     if (format == GL_INTENSITY) {
+	if (unit == 0) {
+	   *reg = ( TD0_color_arg2_diffuse |
+		   TD0_color_add_add |
+		   TD0_color_sel_add |
+		   TD0_alpha_arg2_diffuse |
+		   TD0_alpha_add_enable |
+		   TD0_alpha_sel_add);
+	}
+	else {
+	   *reg = ( TD0_color_arg2_prevstage |
+		   TD0_color_add_add |
+		   TD0_color_sel_add |
+		   TD0_alpha_arg2_prevstage |
+		   TD0_alpha_add_enable |
+		   TD0_alpha_sel_add);
+	}
+     }      
+     else if (format == GL_ALPHA) {
+	*reg = g400_alpha_combine[unit][MGA_ADD];
+     }
+     else {
+	*reg = g400_color_combine[unit][MGA_ADD];
+     }
+     break;
+
+   case GL_BLEND:
+      if (format == GL_ALPHA) {
+	 *reg = g400_alpha_combine[unit][MGA_BLEND];
+      }
+      else {
+	 FALLBACK( ctx, MGA_FALLBACK_TEXTURE, GL_TRUE );
+	 if ( MGA_DEBUG & DEBUG_VERBOSE_FALLBACK )
+	    fprintf( stderr, "FALLBACK: GL_BLEND envcolor=0x%08x\n",
+		     mmesa->envcolor );
+
+         /* Do singletexture GL_BLEND with 'all ones' env-color
+          * by using both texture units.  Multitexture gl_blend
+          * is a fallback.
+          */
+         if (unit == 0) {
+            /* Part 1: R1 = Rf ( 1 - Rt )
+             *         A1 = Af At
+             */
+            *reg = ( TD0_color_arg2_diffuse |
+                     TD0_color_arg1_inv_enable |
+                     TD0_color_sel_mul |
+                     TD0_alpha_arg2_diffuse |
+                     TD0_alpha_sel_arg1);
+         } else {
+            /* Part 2: R2 = R1 + Rt
+             *         A2 = A1
+             */
+            *reg = ( TD0_color_arg2_prevstage |
+                     TD0_color_add_add |
+                     TD0_color_sel_add |
+                     TD0_alpha_arg2_prevstage |
+                     TD0_alpha_sel_arg2);
+         }
+      }
+      break;
+   default:
+      break;
+   }
+}
+
+
+static void disable_tex( GLcontext *ctx, int unit )
+{
+   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
+
+   /* Texture unit disabled */
+
+   if ( mmesa->CurrentTexObj[unit] != NULL ) {
+      /* The old texture is no longer bound to this texture unit.
+       * Mark it as such.
+       */
+
+      mmesa->CurrentTexObj[unit]->base.bound &= ~(1UL << unit);
+      mmesa->CurrentTexObj[unit] = NULL;
+   }
+
+   if ( unit != 0 ) {
+      mmesa->setup.tdualstage1 = mmesa->setup.tdualstage0;
+   }
+
+   if ( ctx->Texture._EnabledUnits == 0 ) {
+      mmesa->setup.dwgctl &= DC_opcod_MASK;
+      mmesa->setup.dwgctl |= DC_opcod_trap;
+      mmesa->hw.alpha_sel = AC_alphasel_diffused;
+   }
+
+   mmesa->dirty |= MGA_UPLOAD_CONTEXT | (MGA_UPLOAD_TEX0 << unit);
+}
+
+static GLboolean enable_tex_2d( GLcontext *ctx, int unit )
+{
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+   const int source = mmesa->tmu_source[unit];
+   const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[source];
+   const struct gl_texture_object *tObj = texUnit->_Current;
+   mgaTextureObjectPtr t = (mgaTextureObjectPtr) tObj->DriverData;
+
+   /* Upload teximages (not pipelined)
+    */
+   if (t->base.dirty_images[0]) {
+      FLUSH_BATCH( mmesa );
+      mgaSetTexImages( mmesa, tObj );
+      if ( t->base.memBlock == NULL ) {
+	 return GL_FALSE;
+      }
+   }
+
+   return GL_TRUE;
+}
+
+static GLboolean update_tex_common( GLcontext *ctx, int unit )
+{
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+   const int source = mmesa->tmu_source[unit];
+   const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[source];
+   struct gl_texture_object	*tObj = texUnit->_Current;
+   mgaTextureObjectPtr t = (mgaTextureObjectPtr) tObj->DriverData;
+
+   /* Fallback if there's a texture border */
+   if ( tObj->Image[tObj->BaseLevel]->Border > 0 ) {
+      return GL_FALSE;
+   }
+
+
+   /* Update state if this is a different texture object to last
+    * time.
+    */
+   if ( mmesa->CurrentTexObj[unit] != t ) {
+      if ( mmesa->CurrentTexObj[unit] != NULL ) {
+	 /* The old texture is no longer bound to this texture unit.
+	  * Mark it as such.
+	  */
+
+	 mmesa->CurrentTexObj[unit]->base.bound &= ~(1UL << unit);
+      }
+
+      mmesa->CurrentTexObj[unit] = t;
+      t->base.bound |= (1UL << unit);
+
+      driUpdateTextureLRU( (driTextureObject *) t ); /* done too often */
+   }
+
+   /* register setup */
+   if ( unit == 1 ) {
+      mmesa->setup.tdualstage1 = mmesa->setup.tdualstage0;
+   }
+
+   t->setup.texctl2 &= TMC_dualtex_MASK;
+   if (ctx->Texture._EnabledUnits == 0x03) {
+      t->setup.texctl2 |= TMC_dualtex_enable;
+   }
+
+   /* FIXME: The Radeon has some cached state so that it can avoid calling
+    * FIXME: UpdateTextureEnv in some cases.  Is that possible here?
+    */
+   if (MGA_IS_G400(mmesa)) {
+      /* G400: Regardless of texture env mode, we use the alpha from the
+       * texture unit (AC_alphasel_fromtex) since it will have already
+       * been modulated by the incoming fragment color, if needed.
+       * We don't want (AC_alphasel_modulate) since that'll effectively
+       * do the modulation twice.
+       */
+      mmesa->hw.alpha_sel = AC_alphasel_fromtex;
+
+      mgaUpdateTextureEnvG400( ctx, unit );
+   } else {
+      mmesa->hw.alpha_sel = 0;
+      switch (ctx->Texture.Unit[0].EnvMode) {
+      case GL_DECAL:
+	 mmesa->hw.alpha_sel |= AC_alphasel_diffused;
+      case GL_REPLACE:
+	 mmesa->hw.alpha_sel |= AC_alphasel_fromtex;
+	 break;
+      case GL_BLEND:
+      case GL_MODULATE:
+	 mmesa->hw.alpha_sel |= AC_alphasel_modulated;
+	 break;
+      default:
+	 break;
+      }
+
+      mgaUpdateTextureEnvG200( ctx, unit );
+   }
+
+   mmesa->setup.dwgctl &= DC_opcod_MASK;
+   mmesa->setup.dwgctl |= DC_opcod_texture_trap;
+   mmesa->dirty |= MGA_UPLOAD_CONTEXT | (MGA_UPLOAD_TEX0 << unit);
+
+   FALLBACK( ctx, MGA_FALLBACK_BORDER_MODE, t->border_fallback );
+   return !t->border_fallback;
+}
+
+
+static GLboolean updateTextureUnit( GLcontext *ctx, int unit )
+{
+   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
+   const int source = mmesa->tmu_source[unit];
+   const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[source];
+
+
+   if ( texUnit->_ReallyEnabled == TEXTURE_2D_BIT) {
+      return(enable_tex_2d( ctx, unit ) &&
+	     update_tex_common( ctx, unit ));
+   }
+   else if ( texUnit->_ReallyEnabled ) {
+      return GL_FALSE;
+   }
+   else {
+      disable_tex( ctx, unit );
+      return GL_TRUE;
+   }
+}
+
+/* The G400 is now programmed quite differently wrt texture environment.
+ */
+void mgaUpdateTextureState( GLcontext *ctx )
+{
+   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
+   GLboolean ok;
+   unsigned  i;
+
+
+   /* This works around a quirk with the MGA hardware.  If only OpenGL 
+    * TEXTURE1 is enabled, then the hardware TEXTURE0 must be used.  The
+    * hardware TEXTURE1 can ONLY be used when hardware TEXTURE0 is also used.
+    */
+
+   mmesa->tmu_source[0] = 0;
+   mmesa->tmu_source[1] = 1;
+
+   if ((ctx->Texture._EnabledUnits & 0x03) == 0x02) {
+      /* only texture 1 enabled */
+      mmesa->tmu_source[0] = 1;
+      mmesa->tmu_source[1] = 0;
+   }
+
+   for ( i = 0, ok = GL_TRUE 
+	 ; (i < ctx->Const.MaxTextureUnits) && ok
+	 ; i++ ) {
+      ok = updateTextureUnit( ctx, i );
+   }
+
+   FALLBACK( ctx, MGA_FALLBACK_TEXTURE, !ok );
+   
+   /* FIXME: I believe that ChooseVertexState should be called here instead of
+    * FIXME: in mgaDDValidateState.
+    */
+}
diff --git a/src/mesa/drivers/dri/mga/mga_xmesa.c b/src/mesa/drivers/dri/mga/mga_xmesa.c
index de1bcc2a0f7..16dc349cace 100644
--- a/src/mesa/drivers/dri/mga/mga_xmesa.c
+++ b/src/mesa/drivers/dri/mga/mga_xmesa.c
@@ -1,4 +1,4 @@
-/* $XFree86: xc/lib/GL/mesa/src/drv/mga/mga_xmesa.c,v 1.19 2003/03/26 20:43:49 tsi Exp $ */
+/* $XFree86: xc/lib/GL/mesa/src/drv/mga/mga_xmesa.c,v 1.18 2002/12/16 16:18:52 dawes Exp $ */
 /*
  * Copyright 2000-2001 VA Linux Systems, Inc.
  * All Rights Reserved.
@@ -26,15 +26,14 @@
  *    Keith Whitwell <[email protected]>
  */
 
-#include <stdio.h>
+#ifdef GLX_DIRECT_RENDERING
 
 #include "mga_common.h"
 #include "mga_xmesa.h"
 #include "context.h"
 #include "matrix.h"
-/*#include "mmath.h"*/
 #include "simple_list.h"
-/*#include "mem.h"*/
+#include "imports.h"
 
 #include "swrast/swrast.h"
 #include "swrast_setup/swrast_setup.h"
@@ -50,26 +49,23 @@
 #include "mgaioctl.h"
 #include "mgatris.h"
 #include "mgavb.h"
-#include "mgabuffers.h"
 #include "mgapixel.h"
-
 #include "mga_xmesa.h"
-
 #include "mga_dri.h"
 
 
+#include "utils.h"
+#include "vblank.h"
+#include "dri_util.h"
+#ifndef _SOLO
+#include "glxextensions.h"
+#endif
+
 #ifndef MGA_DEBUG
-int MGA_DEBUG = (0
-/*		 | DEBUG_ALWAYS_SYNC */
-/*		 | DEBUG_VERBOSE_MSG */
-/*		 | DEBUG_VERBOSE_LRU */
-/*		 | DEBUG_VERBOSE_DRI */
-/*		 | DEBUG_VERBOSE_IOCTL */
-/*		 | DEBUG_VERBOSE_2D */
-/*		 | DEBUG_VERBOSE_FALLBACK */
-   );
+int MGA_DEBUG = 0;
 #endif
 
+static int getSwapInfo( __DRIdrawablePrivate *dPriv, __DRIswapInfo * sInfo );
 
 static GLboolean
 mgaInitDriver(__DRIscreenPrivate *sPriv)
@@ -77,16 +73,8 @@ mgaInitDriver(__DRIscreenPrivate *sPriv)
    mgaScreenPrivate *mgaScreen;
    MGADRIPtr         serverInfo = (MGADRIPtr)sPriv->pDevPriv;
 
-   if (MGA_DEBUG&DEBUG_VERBOSE_DRI)
-      fprintf(stderr, "mgaInitDriver\n");
-
-   /* Check that the DRM driver version is compatible */
-   if (sPriv->drmMajor != 3 ||
-       sPriv->drmMinor < 0) {
-      __driUtilMessage("MGA DRI driver expected DRM driver version 3.0.x but got version %d.%d.%d", sPriv->drmMajor, sPriv->drmMinor, sPriv->drmPatch);
+   if ( ! driCheckDriDdxDrmVersions( sPriv, "MGA", 4, 0, 1, 0, 3, 0 ) )
       return GL_FALSE;
-   }
-
 
    /* Allocate the private area */
    mgaScreen = (mgaScreenPrivate *)MALLOC(sizeof(mgaScreenPrivate));
@@ -103,21 +91,37 @@ mgaInitDriver(__DRIscreenPrivate *sPriv)
       drmMGAGetParam gp;
 
       gp.param = MGA_PARAM_IRQ_NR;
-      gp.value = (int *) &mgaScreen->irq;
+      gp.value = &mgaScreen->irq;
 
       ret = drmCommandWriteRead( sPriv->fd, DRM_MGA_GETPARAM,
 				    &gp, sizeof(gp));
       if (ret) {
 	    fprintf(stderr, "drmMgaGetParam (MGA_PARAM_IRQ_NR): %d\n", ret);
-	    FREE(mgaScreen);
+	    free(mgaScreen);
 	    sPriv->private = NULL;
 	    return GL_FALSE;
       }
    }
    
+   mgaScreen->linecomp_sane = (sPriv->ddxMajor > 1) || (sPriv->ddxMinor > 1)
+       || ((sPriv->ddxMinor == 1) && (sPriv->ddxPatch > 0));
+#ifndef _SOLO   
+   if ( ! mgaScreen->linecomp_sane ) {
+      PFNGLXDISABLEEXTENSIONPROC glx_disable_extension;
+
+      glx_disable_extension = (PFNGLXDISABLEEXTENSIONPROC)
+	  glXGetProcAddress( "__glXDisableExtension" );
+
+      if ( glx_disable_extension != NULL ) {
+	 (*glx_disable_extension)( "GLX_SGI_swap_control" );
+	 (*glx_disable_extension)( "GLX_SGI_video_sync" );
+	 (*glx_disable_extension)( "GLX_MESA_swap_control" );
+      }
+   }
+#endif
    if (serverInfo->chipset != MGA_CARD_TYPE_G200 &&
        serverInfo->chipset != MGA_CARD_TYPE_G400) {
-      FREE(mgaScreen);
+      free(mgaScreen);
       sPriv->private = NULL;
       __driUtilMessage("Unrecognized chipset");
       return GL_FALSE;
@@ -164,7 +168,7 @@ mgaInitDriver(__DRIscreenPrivate *sPriv)
 	      mgaScreen->agp.size,
 	      (drmAddress *)&mgaScreen->agp.map) != 0)
    {
-      Xfree(mgaScreen);
+      free(mgaScreen);
       sPriv->private = NULL;
       __driUtilMessage("Couldn't map agp region");
       return GL_FALSE;
@@ -210,7 +214,7 @@ mgaInitDriver(__DRIscreenPrivate *sPriv)
    mgaScreen->bufs = drmMapBufs(sPriv->fd);
    if (!mgaScreen->bufs) {
       /*drmUnmap(mgaScreen->agp_tex.map, mgaScreen->agp_tex.size);*/
-      FREE(mgaScreen);
+      free(mgaScreen);
       sPriv->private = NULL;
       __driUtilMessage("Couldn't map dma buffers");
       return GL_FALSE;
@@ -254,12 +258,63 @@ static const struct gl_pipeline_stage *mga_pipeline[] = {
 };
 
 
+static const char * const g400_extensions[] =
+{
+   "GL_ARB_multitexture",
+   "GL_ARB_texture_env_add",
+   "GL_EXT_texture_env_add",
+   "GL_EXT_texture_edge_clamp",
+   "GL_SGIS_texture_edge_clamp",
+#if defined (MESA_packed_depth_stencil)
+   "GL_MESA_packed_depth_stencil",
+#endif
+   NULL
+};
+
+static const char * const card_extensions[] =
+{
+   "GL_ARB_multisample",
+   "GL_ARB_texture_compression",
+   "GL_EXT_fog_coord",
+   /* paletted_textures currently doesn't work, but we could fix them later */
+#if 0
+   "GL_EXT_shared_texture_palette",
+   "GL_EXT_paletted_texture",
+#endif
+   "GL_EXT_secondary_color",
+   "GL_EXT_stencil_wrap",
+   "GL_MESA_ycbcr_texture",
+   "GL_SGIS_generate_mipmap",
+   "GL_SGIS_texture_lod",
+   NULL
+};
+
+static const struct dri_debug_control debug_control[] =
+{
+    { "fall",  DEBUG_VERBOSE_FALLBACK },
+    { "tex",   DEBUG_VERBOSE_TEXTURE },
+    { "ioctl", DEBUG_VERBOSE_IOCTL },
+    { "verb",  DEBUG_VERBOSE_MSG },
+    { "dri",   DEBUG_VERBOSE_DRI },
+    { NULL,    0 }
+};
+
+
+static int
+get_ust_nop( uint64_t * ust )
+{
+   *ust = 1;
+   return 0;
+}
+
+
 static GLboolean
 mgaCreateContext( const __GLcontextModes *mesaVis,
                   __DRIcontextPrivate *driContextPriv,
                   void *sharedContextPrivate )
 {
    int i;
+   unsigned   maxlevels;
    GLcontext *ctx, *shareCtx;
    mgaContextPtr mmesa;
    __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv;
@@ -281,7 +336,7 @@ mgaCreateContext( const __GLcontextModes *mesaVis,
       shareCtx = ((mgaContextPtr) sharedContextPrivate)->glCtx;
    else 
       shareCtx = NULL;
-   mmesa->glCtx = _mesa_create_context(mesaVis, shareCtx, mmesa, GL_TRUE);
+   mmesa->glCtx = _mesa_create_context(mesaVis, shareCtx, (void *) mmesa, GL_TRUE);
    if (!mmesa->glCtx) {
       FREE(mmesa);
       return GL_FALSE;
@@ -298,13 +353,20 @@ mgaCreateContext( const __GLcontextModes *mesaVis,
    mmesa->sarea = (void *)saPriv;
    mmesa->glBuffer = NULL;
 
-   make_empty_list(&mmesa->SwappedOut);
-
-   mmesa->lastTexHeap = mgaScreen->texVirtual[MGA_AGP_HEAP] ? 2 : 1;
-
-   for (i = 0 ; i < mmesa->lastTexHeap ; i++) {
-      mmesa->texHeap[i] = mmInit( 0, mgaScreen->textureSize[i]);
-      make_empty_list(&mmesa->TexObjList[i]);
+   (void) memset( mmesa->texture_heaps, 0, sizeof( mmesa->texture_heaps ) );
+   make_empty_list( & mmesa->swapped );
+
+   mmesa->nr_heaps = mgaScreen->texVirtual[MGA_AGP_HEAP] ? 2 : 1;
+   for ( i = 0 ; i < mmesa->nr_heaps ; i++ ) {
+      mmesa->texture_heaps[i] = driCreateTextureHeap( i, mmesa,
+	    mgaScreen->textureSize[i],
+	    6,
+	    MGA_NR_TEX_REGIONS,
+	    mmesa->sarea->texList[i],
+	    & mmesa->sarea->texAge[i],
+	    & mmesa->swapped,
+	    sizeof( mgaTextureObject_t ),
+	    (destroy_texture_object_t *) mgaDestroyTexObj );
    }
 
    /* Set the maximum texture size small enough that we can guarentee
@@ -312,22 +374,26 @@ mgaCreateContext( const __GLcontextModes *mesaVis,
     * on the card at once.
     */
    ctx = mmesa->glCtx;
-   { 
-      int nr = 2;
-
-      if (mgaScreen->chipset == MGA_CARD_TYPE_G200)
-	 nr = 1;
-
-      if (mgaScreen->textureSize[0] < nr*1024*1024) {
-	 ctx->Const.MaxTextureLevels = 9;
-      } else if (mgaScreen->textureSize[0] < nr*4*1024*1024) {
-	 ctx->Const.MaxTextureLevels = 10;
-      } else {
-	 ctx->Const.MaxTextureLevels = 11;
-      }
+   if ( mgaScreen->chipset == MGA_CARD_TYPE_G200 ) {
+      ctx->Const.MaxTextureUnits = 1;
+      maxlevels = G200_TEX_MAXLEVELS;
 
-      ctx->Const.MaxTextureUnits = nr;
    }
+   else {
+      ctx->Const.MaxTextureUnits = 2;
+      maxlevels = G400_TEX_MAXLEVELS;
+   }
+
+   driCalculateMaxTextureLevels( mmesa->texture_heaps,
+				 mmesa->nr_heaps,
+				 & ctx->Const,
+				 4,
+				 11, /* max 2D texture size is 1024x1024 */
+				 0,  /* 3D textures unsupported. */
+				 0,  /* cube textures unsupported. */
+				 0,  /* texture rectangles unsupported. */
+				 maxlevels,
+				 GL_FALSE );
 
    ctx->Const.MinLineWidth = 1.0;
    ctx->Const.MinLineWidthAA = 1.0;
@@ -335,6 +401,7 @@ mgaCreateContext( const __GLcontextModes *mesaVis,
    ctx->Const.MaxLineWidthAA = 10.0;
    ctx->Const.LineWidthGranularity = 1.0;
 
+   mmesa->default32BitTextures = (mesaVis->rgbBits >= 24);
    mmesa->hw_stencil = mesaVis->stencilBits && mesaVis->depthBits == 24;
 
    switch (mesaVis->depthBits) {
@@ -361,7 +428,6 @@ mgaCreateContext( const __GLcontextModes *mesaVis,
 
    mmesa->haveHwStipple = GL_FALSE;
    mmesa->RenderIndex = -1;		/* impossible value */
-   mmesa->new_state = ~0;
    mmesa->dirty = ~0;
    mmesa->vertex_format = 0;   
    mmesa->CurrentTexObj[0] = 0;
@@ -395,7 +461,11 @@ mgaCreateContext( const __GLcontextModes *mesaVis,
    ctx->DriverCtx = (void *) mmesa;
    mmesa->glCtx = ctx;
 
-   mgaDDExtensionsInit( ctx );
+   driInitExtensions( ctx, card_extensions, GL_FALSE );
+
+   if (MGA_IS_G400(MGA_CONTEXT(ctx))) {
+      driInitExtensions( ctx, g400_extensions, GL_FALSE );
+   }
 
    mgaDDInitStateFuncs( ctx );
    mgaDDInitTextureFuncs( ctx );
@@ -410,6 +480,24 @@ mgaCreateContext( const __GLcontextModes *mesaVis,
 
    driContextPriv->driverPrivate = (void *) mmesa;
 
+#if DO_DEBUG
+   MGA_DEBUG = driParseDebugString( getenv( "MGA_DEBUG" ),
+				    debug_control );
+#endif
+
+   mmesa->vblank_flags = ((mmesa->mgaScreen->irq == 0) 
+			  || !mmesa->mgaScreen->linecomp_sane)
+       ? VBLANK_FLAG_NO_IRQ : driGetDefaultVBlankFlags();
+#ifndef _SOLO
+   mmesa->get_ust = (PFNGLXGETUSTPROC) glXGetProcAddress( "__glXGetUST" );
+   if ( mmesa->get_ust == NULL ) 
+#endif
+   {
+      mmesa->get_ust = get_ust_nop;
+   }
+
+   (*mmesa->get_ust)( & mmesa->swap_ust );
+
    return GL_TRUE;
 }
 
@@ -419,10 +507,15 @@ mgaDestroyContext(__DRIcontextPrivate *driContextPriv)
    mgaContextPtr mmesa = (mgaContextPtr) driContextPriv->driverPrivate;
 
    if (MGA_DEBUG&DEBUG_VERBOSE_DRI)
-      fprintf(stderr, "mgaDestroyContext\n");
+      fprintf( stderr, "[%s:%d] mgaDestroyContext start\n",
+	       __FILE__, __LINE__ );
 
    assert(mmesa); /* should never be null */
    if (mmesa) {
+      GLboolean   release_texture_heaps;
+
+
+      release_texture_heaps = (mmesa->glCtx->Shared->RefCount == 1);
       _swsetup_DestroyContext( mmesa->glCtx );
       _tnl_DestroyContext( mmesa->glCtx );
       _ac_DestroyContext( mmesa->glCtx );
@@ -433,9 +526,27 @@ mgaDestroyContext(__DRIcontextPrivate *driContextPriv)
       /* free the Mesa context */
       mmesa->glCtx->DriverCtx = NULL;
       _mesa_destroy_context(mmesa->glCtx);
-      /* free the mga context */
+       
+      if ( release_texture_heaps ) {
+         /* This share group is about to go away, free our private
+          * texture object data.
+          */
+         int i;
+
+	 assert( is_empty_list( & mmesa->swapped ) );
+
+         for ( i = 0 ; i < mmesa->nr_heaps ; i++ ) {
+	    driDestroyTextureHeap( mmesa->texture_heaps[ i ] );
+	    mmesa->texture_heaps[ i ] = NULL;
+         }
+      }
+
       FREE(mmesa);
    }
+
+   if (MGA_DEBUG&DEBUG_VERBOSE_DRI)
+      fprintf( stderr, "[%s:%d] mgaDestroyContext done\n",
+	       __FILE__, __LINE__ );
 }
 
 
@@ -482,13 +593,7 @@ mgaUnbindContext(__DRIcontextPrivate *driContextPriv)
 }
 
 static GLboolean
-mgaOpenFullScreen(__DRIcontextPrivate *driContextPriv)
-{
-    return GL_TRUE;
-}
-
-static GLboolean
-mgaCloseFullScreen(__DRIcontextPrivate *driContextPriv)
+mgaOpenCloseFullScreen(__DRIcontextPrivate *driContextPriv)
 {
     return GL_TRUE;
 }
@@ -505,8 +610,6 @@ mgaMakeCurrent(__DRIcontextPrivate *driContextPriv,
                __DRIdrawablePrivate *driDrawPriv,
                __DRIdrawablePrivate *driReadPriv)
 {
-   fprintf(stderr, "%s\n", __FUNCTION__);
-
    if (driContextPriv) {
       mgaContextPtr mmesa = (mgaContextPtr) driContextPriv->driverPrivate;
 
@@ -532,6 +635,7 @@ mgaMakeCurrent(__DRIcontextPrivate *driContextPriv,
    return GL_TRUE;
 }
 
+
 void mgaGetLock( mgaContextPtr mmesa, GLuint flags )
 {
    __DRIdrawablePrivate *dPriv = mmesa->driDrawable;
@@ -539,29 +643,10 @@ void mgaGetLock( mgaContextPtr mmesa, GLuint flags )
    int me = mmesa->hHWContext;
    int i;
 
-   fprintf(stderr, "%s\n", __FUNCTION__);
-
    drmGetLock(mmesa->driFd, mmesa->hHWContext, flags);
-   
-   fprintf(stderr, 
-	   "mmesa->lastStamp %d dpriv->lastStamp %d *(dpriv->pStamp) %d\n",
-	   mmesa->lastStamp, 
-	   dPriv->lastStamp,
-	   *(dPriv->pStamp));
-   
-   /* The window might have moved, so we might need to get new clip
-    * rects.
-    *
-    * NOTE: This releases and regrabs the hw lock to allow the X server
-    * to respond to the DRI protocol request for new drawable info.
-    * Since the hardware state depends on having the latest drawable
-    * clip rects, all state checking must be done _after_ this call.
-    */
-   DRI_VALIDATE_DRAWABLE_INFO( sPriv, dPriv );
 
-   if ( mmesa->lastStamp == 0 ||
-	mmesa->lastStamp != dPriv->lastStamp ) {
-      mmesa->lastStamp = dPriv->lastStamp;
+   if (*(dPriv->pStamp) != mmesa->lastStamp) {
+      mmesa->lastStamp = *(dPriv->pStamp);
       mmesa->SetupNewInputs |= VERT_BIT_CLIP;
       mmesa->dirty_cliprects = (MGA_FRONT|MGA_BACK);
       mgaUpdateRects( mmesa, (MGA_FRONT|MGA_BACK) );
@@ -569,7 +654,7 @@ void mgaGetLock( mgaContextPtr mmesa, GLuint flags )
 
    mmesa->dirty |= MGA_UPLOAD_CONTEXT | MGA_UPLOAD_CLIPRECTS;
 
-   mmesa->sarea->dirty |= MGA_UPLOAD_CONTEXT;
+    mmesa->sarea->dirty |= MGA_UPLOAD_CONTEXT;
 
    if (sarea->ctxOwner != me) {
       mmesa->dirty |= (MGA_UPLOAD_CONTEXT | MGA_UPLOAD_TEX0 |
@@ -577,36 +662,48 @@ void mgaGetLock( mgaContextPtr mmesa, GLuint flags )
       sarea->ctxOwner=me;
    }
 
-   for (i = 0 ; i < mmesa->lastTexHeap ; i++)
-      if (sarea->texAge[i] != mmesa->texAge[i])
-	 mgaAgeTextures( mmesa, i );
+   for ( i = 0 ; i < mmesa->nr_heaps ; i++ ) {
+      DRI_AGE_TEXTURES( mmesa->texture_heaps[ i ] );
+   }
 
    sarea->last_quiescent = -1;	/* just kill it for now */
 }
 
 
-
 static const struct __DriverAPIRec mgaAPI = {
-   mgaInitDriver,
-   mgaDestroyScreen,
-   mgaCreateContext,
-   mgaDestroyContext,
-   mgaCreateBuffer,
-   mgaDestroyBuffer,
-   mgaSwapBuffers,
-   mgaMakeCurrent,
-   mgaUnbindContext,
-   mgaOpenFullScreen,
-   mgaCloseFullScreen
+   .InitDriver      = mgaInitDriver,
+   .DestroyScreen   = mgaDestroyScreen,
+   .CreateContext   = mgaCreateContext,
+   .DestroyContext  = mgaDestroyContext,
+   .CreateBuffer    = mgaCreateBuffer,
+   .DestroyBuffer   = mgaDestroyBuffer,
+   .SwapBuffers     = mgaSwapBuffers,
+   .MakeCurrent     = mgaMakeCurrent,
+   .UnbindContext   = mgaUnbindContext,
+   .OpenFullScreen  = mgaOpenCloseFullScreen,
+   .CloseFullScreen = mgaOpenCloseFullScreen,
+   .GetSwapInfo     = getSwapInfo,
+   .GetMSC          = driGetMSC32,
+   .WaitForMSC      = driWaitForMSC32,
+   .WaitForSBC      = NULL,
+   .SwapBuffersMSC  = NULL
 };
 
 
-
 /*
  * This is the bootstrap function for the driver.
  * The __driCreateScreen name is the symbol that libGL.so fetches.
  * Return:  pointer to a __DRIscreenPrivate.
  */
+#ifndef _SOLO
+void *__driCreateScreen(Display *dpy, int scrn, __DRIscreen *psc,
+                        int numConfigs, __GLXvisualConfig *config)
+{
+   __DRIscreenPrivate *psp;
+   psp = __driUtilCreateScreen(dpy, scrn, psc, numConfigs, config, &mgaAPI);
+   return (void *) psp;
+}
+#else
 void *__driCreateScreen(struct DRIDriverRec *driver,
                         struct DRIDriverContextRec *driverContext)
 {
@@ -614,3 +711,56 @@ void *__driCreateScreen(struct DRIDriverRec *driver,
    psp = __driUtilCreateScreen(driver, driverContext, &mgaAPI);
    return (void *) psp;
 }
+#endif
+
+
+#ifndef _SOLO
+/* This function is called by libGL.so as soon as libGL.so is loaded.
+ * This is where we'd register new extension functions with the dispatcher.
+ */
+void
+__driRegisterExtensions( void )
+{
+   PFNGLXENABLEEXTENSIONPROC glx_enable_extension;
+
+
+   if ( driCompareGLXAPIVersion( 20030317 ) >= 0 ) {
+      glx_enable_extension = (PFNGLXENABLEEXTENSIONPROC)
+	  glXGetProcAddress( "__glXEnableExtension" );
+
+      if ( glx_enable_extension != NULL ) {
+	 (*glx_enable_extension)( "GLX_SGI_swap_control", GL_FALSE );
+	 (*glx_enable_extension)( "GLX_SGI_video_sync", GL_FALSE );
+	 (*glx_enable_extension)( "GLX_MESA_swap_control", GL_FALSE );
+	 (*glx_enable_extension)( "GLX_MESA_swap_frame_usage", GL_FALSE );
+      }
+   }
+}
+#endif
+
+/**
+ * Get information about previous buffer swaps.
+ */
+static int
+getSwapInfo( __DRIdrawablePrivate *dPriv, __DRIswapInfo * sInfo )
+{
+   mgaContextPtr  mmesa;
+
+   if ( (dPriv == NULL) || (dPriv->driContextPriv == NULL)
+	|| (dPriv->driContextPriv->driverPrivate == NULL)
+	|| (sInfo == NULL) ) {
+      return -1;
+   }
+
+   mmesa = (mgaContextPtr) dPriv->driContextPriv->driverPrivate;
+   sInfo->swap_count = mmesa->swap_count;
+   sInfo->swap_ust = mmesa->swap_ust;
+   sInfo->swap_missed_count = mmesa->swap_missed_count;
+
+   sInfo->swap_missed_usage = (sInfo->swap_missed_count != 0)
+       ? driCalculateSwapUsage( dPriv, 0, mmesa->swap_missed_ust )
+       : 0.0;
+
+   return 0;
+}
+#endif
diff --git a/src/mesa/drivers/dri/mga/mga_xmesa.h b/src/mesa/drivers/dri/mga/mga_xmesa.h
index eda996f5a44..de349da0895 100644
--- a/src/mesa/drivers/dri/mga/mga_xmesa.h
+++ b/src/mesa/drivers/dri/mga/mga_xmesa.h
@@ -29,8 +29,9 @@
 #ifndef _MGA_INIT_H_
 #define _MGA_INIT_H_
 
+#ifdef GLX_DIRECT_RENDERING
+
 #include <sys/time.h>
-#include <linux/types.h>
 #include "dri_util.h"
 #include "mtypes.h"
 #include "mgaregs.h"
@@ -46,6 +47,9 @@ typedef struct mga_screen_private_s {
    int cpp;			/* for front and back buffers */
    GLint agpMode;
    unsigned int irq;		/* IRQ number (0 means none) */
+   GLboolean  linecomp_sane;    /* GL_TRUE if line comp. programmed correctly
+				 * by the DDX driver.
+				 */
 
    unsigned int mAccess;
 
@@ -136,8 +140,9 @@ do {						\
 #define MGA_BASE( reg )		((unsigned long)(mmesa->mgaScreen->mmio.map))
 #define MGA_ADDR( reg )		(MGA_BASE(reg) + reg)
 
-#define MGA_DEREF( reg )	*(volatile __u32 *)MGA_ADDR( reg )
+#define MGA_DEREF( reg )	*(volatile CARD32 *)MGA_ADDR( reg )
 #define MGA_READ( reg )		MGA_DEREF( reg )
 #define MGA_WRITE( reg, val )	do { MGA_DEREF( reg ) = val; } while (0)
 
 #endif
+#endif
diff --git a/src/mesa/drivers/dri/mga/mgacontext.h b/src/mesa/drivers/dri/mga/mgacontext.h
index 3065ea9fd1f..7188f6d9552 100644
--- a/src/mesa/drivers/dri/mga/mgacontext.h
+++ b/src/mesa/drivers/dri/mga/mgacontext.h
@@ -29,14 +29,15 @@
 #ifndef MGALIB_INC
 #define MGALIB_INC
 
-/*#include <X11/Xlibint.h>*/
+#include <inttypes.h>
 #include "dri_util.h"
 #include "mtypes.h"
+#include "colormac.h"
 #include "xf86drm.h"
 #include "mm.h"
-/*#include "mem.h"*/
 #include "mga_sarea.h"
-
+#include "texmem.h"
+#include "macros.h"
 
 #define MGA_SET_FIELD(reg,mask,val)  reg = ((reg) & (mask)) | ((val) & ~(mask))
 #define MGA_FIELD(field,val) (((val) << (field ## _SHIFT)) & ~(field ## _MASK))
@@ -59,19 +60,9 @@
 #define MGA_FALLBACK_RENDERMODE     0x10
 #define MGA_FALLBACK_STENCIL        0x20
 #define MGA_FALLBACK_DEPTH          0x40
+#define MGA_FALLBACK_BORDER_MODE    0x80
 
 
-/* For mgaCtx->new_state.
- */
-#define MGA_NEW_DEPTH   0x1
-#define MGA_NEW_ALPHA   0x2
-#define MGA_NEW_CLIP    0x8
-#define MGA_NEW_TEXTURE 0x20
-#define MGA_NEW_CULL    0x40
-#define MGA_NEW_WARP    0x80
-#define MGA_NEW_STENCIL 0x100
-#define MGA_NEW_CONTEXT 0x200
-
 /* Use the templated vertex formats:
  */
 #define TAG(x) mga##x
@@ -96,43 +87,103 @@ typedef void (*mga_point_func)( mgaContextPtr, mgaVertex * );
 struct mga_texture_object_s;
 struct mga_screen_private_s;
 
-#define MGA_TEX_MAXLEVELS 5
+#define G200_TEX_MAXLEVELS 5
+#define G400_TEX_MAXLEVELS 11
 
 typedef struct mga_texture_object_s
 {
-   struct mga_texture_object_s *next;
-   struct mga_texture_object_s *prev;
-   struct gl_texture_object *tObj;
-   struct mga_context_t *ctx;
-   PMemBlock	MemBlock;
-   GLuint		offsets[MGA_TEX_MAXLEVELS];
-   int             lastLevel;
-   GLuint         dirty_images;
-   GLuint		totalSize;
-   int		texelBytes;
-   GLuint 	age;
-   int             bound;
-   int             heap;	/* agp or card */
+   driTextureObject   base;
+
+   /* The G200 only has the ability to use 5 mipmap levels (including the
+    * base level).  The G400 does not have this restriction, but it still
+    * only has 5 offset pointers in the hardware.  The trick on the G400 is
+    * upto the first 4 offset pointers point to mipmap levels.  The last
+    * offset pointer tells how large the preceeding mipmap is.  This value is
+    * then used to determine where the remaining mipmaps are.
+    * 
+    * For example, if the first offsets[0] through offsets[2] are used as
+    * pointers, then offset[3] will be the size of the mipmap pointed to by
+    * offsets[2].  So mipmap level 3 will be at (offsets[2]+offsets[3]).  For
+    * each successive mipmap level, offsets[3] is divided by 4 and added to
+    * the previous address.  So mipmap level 4 will be at 
+    * (offsets[2]+offsets[3]+(offsets[3] / 4)).
+    * 
+    * The last pointer is selected by setting TO_texorgoffsetsel in its
+    * pointer.  In the previous example, offset[2] would have
+    * TO_texorgoffsetsel or'ed in before writing it to the hardware.
+    * 
+    * In the current driver all of the mipmaps are packed together linearly
+    * with mipmap level 0.  Therefore offsets[0] points to the base of the
+    * texture (and has TO_texorgoffsetsel or'ed in), and offsets[1] is the
+    * size of the base texture.
+    *
+    * There is a possible optimization available here.  At times the driver
+    * may not be able to allocate a single block of memory for the complete
+    * texture without ejecting some other textures from memory.  It may be
+    * possible to put some of the lower mipmap levels (i.e., the larger
+    * mipmaps) in memory separate from the higher levels.
+    *
+    * The implementation should be fairly obvious, but getting "right" would
+    * likely be non-trivial.  A first allocation for the entire texture would
+    * be attempted with a flag that says "don't eject other textures."  If
+    * that failed, an additional allocation would be attmpted for just the
+    * base map.  The process would repeat with the block of lower maps.  The
+    * tricky parts would be in detecting when some of the levels had been
+    * ejected from texture memory by other textures and preventing the
+    * 4th allocation (for all the smallest mipmap levels) from kicking out
+    * any of the first three.
+    * 
+    * This array holds G400_TEX_MAXLEVELS pointers to remove an if-statement
+    * in a loop in mgaSetTexImages.  Values past G200_TEX_MAXLEVELS are not
+    * used.
+    */
+   GLuint             offsets[G400_TEX_MAXLEVELS];
+
+   int                texelBytes;
+   GLuint             age;
 
    mga_texture_regs_t setup;
+
+   /* If one texture dimension wraps with GL_CLAMP and the other with
+    * GL_CLAMP_TO_EDGE, we have to fallback to software.  We would also have
+    * to fallback for GL_CLAMP_TO_BORDER.
+    */
+   GLboolean          border_fallback;
 } mgaTextureObject_t;
 
+struct mga_hw_state {
+   GLuint   specen;
+   GLuint   cull;
+   GLuint   cull_dualtex;
+   GLuint   stencil;
+   GLuint   stencilctl;
+   GLuint   stencil_enable;
+   GLuint   zmode;
+   GLuint   rop;
+   GLuint   alpha_func;
+   GLuint   alpha_func_enable;
+   GLuint   blend_func;
+   GLuint   blend_func_enable;
+   GLuint   alpha_sel;
+};
+
 struct mga_context_t {
 
    GLcontext *glCtx;
-   unsigned int lastStamp;	/* fullscreen breaks dpriv->laststamp,
-				 * need to shadow it here. */
+   unsigned int lastStamp;		/* fullscreen breaks dpriv->laststamp,
+					 * need to shadow it here. */
+
+   /* Hardware state management
+    */
+   struct mga_hw_state hw;
 
    /* Bookkeeping for texturing
     */
-   int lastTexHeap;
-   struct mga_texture_object_s TexObjList[MGA_NR_TEX_HEAPS];
-   struct mga_texture_object_s SwappedOut;
+   unsigned           nr_heaps;
+   driTexHeap       * texture_heaps[ MGA_NR_TEX_HEAPS ];
+   driTextureObject   swapped;
+
    struct mga_texture_object_s *CurrentTexObj[2];
-   memHeap_t *texHeap[MGA_NR_TEX_HEAPS];
-   int c_texupload;
-   int c_texusage;
-   int tex_thrash;
 
 
    /* Map GL texture units onto hardware.
@@ -166,7 +217,7 @@ struct mga_context_t {
    GLenum raster_primitive;
    GLenum render_primitive;
 
-   char *verts;
+   GLubyte *verts;
    GLint vertex_stride_shift;
    GLuint vertex_format;		
    GLuint vertex_size;
@@ -180,8 +231,7 @@ struct mga_context_t {
 
    /* Manage driver and hardware state
     */
-   GLuint        new_gl_state; 
-   GLuint        new_state; 
+   GLuint        NewGLState; 
    GLuint        dirty;
 
    mga_context_regs_t setup;
@@ -205,13 +255,21 @@ struct mga_context_t {
    /* VBI
     */
    GLuint vbl_seq;
+   GLuint vblank_flags;
+
+   uint64_t swap_ust;
+   uint64_t swap_missed_ust;
+
+   GLuint swap_count;
+   GLuint swap_missed_count;
+
+   PFNGLXGETUSTPROC get_ust;
 
    /* Drawable, cliprect and scissor information
     */
    int dirty_cliprects;		/* which sets of cliprects are uptodate? */
    int draw_buffer;		/* which buffer are we rendering to */
    unsigned int drawOffset;		/* draw buffer address in  space */
-   int read_buffer;
    int readOffset;
    int drawX, drawY;		/* origin of drawable in draw buffer */
    int lastX, lastY;		/* detect DSTORG bug */
@@ -245,51 +303,38 @@ struct mga_context_t {
 
 #define MGA_CONTEXT(ctx) ((mgaContextPtr)(ctx->DriverCtx))
 
-#define MGAPACKCOLOR555(r,g,b,a) \
-  ((((r) & 0xf8) << 7) | (((g) & 0xf8) << 2) | (((b) & 0xf8) >> 3) | \
-    ((a) ? 0x8000 : 0))
 
-#define MGAPACKCOLOR565(r,g,b) \
-  ((((r) & 0xf8) << 8) | (((g) & 0xfc) << 3) | (((b) & 0xf8) >> 3))
 
-#define MGAPACKCOLOR88(l, a) \
-  (((l) << 8) | (a))
-
-#define MGAPACKCOLOR888(r,g,b) \
-  (((r) << 16) | ((g) << 8) | (b))
-
-#define MGAPACKCOLOR8888(r,g,b,a) \
-  (((a) << 24) | ((r) << 16) | ((g) << 8) | (b))
-
-#define MGAPACKCOLOR4444(r,g,b,a) \
-  ((((a) & 0xf0) << 8) | (((r) & 0xf0) << 4) | ((g) & 0xf0) | ((b) >> 4))
 
+/* ================================================================
+ * Debugging:
+ */
+#define DO_DEBUG		1
 
-#define MGA_DEBUG 0
-#ifndef MGA_DEBUG
+#if DO_DEBUG
 extern int MGA_DEBUG;
+#else
+#define MGA_DEBUG		0
 #endif
 
-#define DEBUG_ALWAYS_SYNC	0x1
-#define DEBUG_VERBOSE_MSG	0x2
-#define DEBUG_VERBOSE_LRU	0x4
-#define DEBUG_VERBOSE_DRI	0x8
-#define DEBUG_VERBOSE_IOCTL	0x10
-#define DEBUG_VERBOSE_2D	0x20
-#define DEBUG_VERBOSE_FALLBACK	0x40
+#define DEBUG_VERBOSE_MSG	0x01
+#define DEBUG_VERBOSE_DRI	0x02
+#define DEBUG_VERBOSE_IOCTL	0x04
+#define DEBUG_VERBOSE_TEXTURE   0x08
+#define DEBUG_VERBOSE_FALLBACK	0x10
 
 static __inline__ GLuint mgaPackColor(GLuint cpp,
 				      GLubyte r, GLubyte g,
 				      GLubyte b, GLubyte a)
 {
-  switch (cpp) {
-  case 2:
-    return MGAPACKCOLOR565(r,g,b);
-  case 4:
-    return MGAPACKCOLOR8888(r,g,b,a);
-  default:
-    return 0;
-  }
+   switch (cpp) {
+   case 2:
+      return PACK_COLOR_565( r, g, b );
+   case 4:
+      return PACK_COLOR_8888( a, r, g, b );
+   default:
+      return 0;
+   }
 }
 
 
diff --git a/src/mesa/drivers/dri/mga/mgadd.c b/src/mesa/drivers/dri/mga/mgadd.c
index 6ce50e6726e..d2f11d8d457 100644
--- a/src/mesa/drivers/dri/mga/mgadd.c
+++ b/src/mesa/drivers/dri/mga/mgadd.c
@@ -29,10 +29,6 @@
 
 #include "mtypes.h"
 
-
-#include <stdio.h>
-#include <stdlib.h>
-
 #include "mm.h"
 #include "mgacontext.h"
 #include "mgadd.h"
@@ -42,12 +38,9 @@
 #include "mgatris.h"
 #include "mgavb.h"
 #include "mga_xmesa.h"
-#include "extensions.h"
-#if defined(USE_X86_ASM)
-#include "X86/common_x86_asm.h"
-#endif
+#include "utils.h"
 
-#define MGA_DATE	"20020221"
+#define DRIVER_DATE	"20030328"
 
 
 /***************************************
@@ -59,52 +52,19 @@ static const GLubyte *mgaDDGetString( GLcontext *ctx, GLenum name )
 {
    mgaContextPtr mmesa = MGA_CONTEXT( ctx );
    static char buffer[128];
+   unsigned   offset;
 
    switch ( name ) {
    case GL_VENDOR:
       return (GLubyte *) "VA Linux Systems Inc.";
 
    case GL_RENDERER:
-      sprintf( buffer, "Mesa DRI %s " MGA_DATE,
-	       MGA_IS_G400(mmesa) ? "G400" :
-	       MGA_IS_G200(mmesa) ? "G200" : "MGA" );
-
-      /* Append any AGP-specific information.
-       */
-      switch ( mmesa->mgaScreen->agpMode ) {
-      case 1:
-	 strncat( buffer, " AGP 1x", 7 );
-	 break;
-      case 2:
-	 strncat( buffer, " AGP 2x", 7 );
-	 break;
-      case 4:
-	 strncat( buffer, " AGP 4x", 7 );
-	 break;
-      }
-
-      /* Append any CPU-specific information.
-       */
-#ifdef USE_X86_ASM
-      if ( _mesa_x86_cpu_features ) {
-	 strncat( buffer, " x86", 4 );
-      }
-#endif
-#ifdef USE_MMX_ASM
-      if ( cpu_has_mmx ) {
-	 strncat( buffer, "/MMX", 4 );
-      }
-#endif
-#ifdef USE_3DNOW_ASM
-      if ( cpu_has_3dnow ) {
-	 strncat( buffer, "/3DNow!", 7 );
-      }
-#endif
-#ifdef USE_SSE_ASM
-      if ( cpu_has_xmm ) {
-	 strncat( buffer, "/SSE", 4 );
-      }
-#endif
+      offset = driGetRendererString( buffer, 
+				     MGA_IS_G400(mmesa) ? "G400" :
+				     MGA_IS_G200(mmesa) ? "G200" : "MGA",
+				     DRIVER_DATE,
+				     mmesa->mgaScreen->agpMode );
+
       return (GLubyte *)buffer;
 
    default:
@@ -129,40 +89,6 @@ static void mgaBufferSize(GLframebuffer *buffer, GLuint *width, GLuint *height)
    UNLOCK_HARDWARE( mmesa );
 }
 
-void mgaDDExtensionsInit( GLcontext *ctx )
-{
-   /* paletted_textures currently doesn't work, but we could fix them later */
-   /*
-   _mesa_enable_extension( ctx, "GL_EXT_shared_texture_palette" );
-   _mesa_enable_extension( ctx, "GL_EXT_paletted_texture" );
-   */
-
-   _mesa_enable_extension( ctx, "GL_ARB_texture_compression" );
-   _mesa_enable_extension( ctx, "GL_ARB_multisample" );
-
-   _mesa_enable_extension( ctx, "GL_SGIS_generate_mipmap" );
-
-   /* Turn on multitexture and texenv_add for the G400.
-    */
-   if (MGA_IS_G400(MGA_CONTEXT(ctx))) {
-      _mesa_enable_extension( ctx, "GL_ARB_multitexture" );
-      _mesa_enable_extension( ctx, "GL_ARB_texture_env_add" );
-
-      _mesa_enable_extension( ctx, "GL_EXT_texture_env_add" );
-
-#if defined (MESA_packed_depth_stencil)
-      _mesa_enable_extension( ctx, "GL_MESA_packed_depth_stencil" );
-#endif
-
-#if defined (MESA_experimetal_agp_allocator)
-      if (!getenv("MGA_DISABLE_AGP_ALLOCATOR"))  
-	 _mesa_enable_extension( ctx, "GL_MESA_experimental_agp_allocator" );
-#endif
-   }
-}
-
-
-
 void mgaDDInitDriverFuncs( GLcontext *ctx )
 {
    ctx->Driver.GetBufferSize = mgaBufferSize;
diff --git a/src/mesa/drivers/dri/mga/mgadd.h b/src/mesa/drivers/dri/mga/mgadd.h
index 6072e6cfc9f..919fd742d61 100644
--- a/src/mesa/drivers/dri/mga/mgadd.h
+++ b/src/mesa/drivers/dri/mga/mgadd.h
@@ -32,6 +32,5 @@
 #include "context.h"
 
 void mgaDDInitDriverFuncs( GLcontext *ctx );
-void mgaDDExtensionsInit( GLcontext *ctx );
 
 #endif
diff --git a/src/mesa/drivers/dri/mga/mgaioctl.c b/src/mesa/drivers/dri/mga/mgaioctl.c
index f88c8780607..6b50af202a8 100644
--- a/src/mesa/drivers/dri/mga/mgaioctl.c
+++ b/src/mesa/drivers/dri/mga/mgaioctl.c
@@ -27,7 +27,7 @@
  */
 /* $XFree86: xc/lib/GL/mesa/src/drv/mga/mgaioctl.c,v 1.16 2002/12/16 16:18:52 dawes Exp $ */
 
-#include <stdio.h>
+#include <sched.h>
 #include <errno.h>
 
 #include "mtypes.h"
@@ -43,11 +43,10 @@
 #include "mgavb.h"
 #include "mgaioctl.h"
 #include "mgatris.h"
-#include "mgabuffers.h"
+#include "mga_common.h"
 
+#include "vblank.h"
 
-#include "xf86drm.h"
-#include "mga_common.h"
 
 static void mga_iload_dma_ioctl(mgaContextPtr mmesa,
 				unsigned long dest,
@@ -61,6 +60,13 @@ static void mga_iload_dma_ioctl(mgaContextPtr mmesa,
       fprintf(stderr, "DRM_IOCTL_MGA_ILOAD idx %d dst %x length %d\n",
 	      buf->idx, (int) dest, length);
 
+   if ( (length & MGA_ILOAD_MASK) != 0 ) {
+      UNLOCK_HARDWARE( mmesa );
+      fprintf( stderr, "%s: Invalid ILOAD datasize (%d), must be "
+	       "multiple of %u.\n", __FUNCTION__, length, MGA_ILOAD_ALIGN );
+      exit( 1 );
+   }
+
    iload.idx = buf->idx;
    iload.dstorg = dest;
    iload.length = length;
@@ -278,65 +284,11 @@ mgaDDClear( GLcontext *ctx, GLbitfield mask, GLboolean all,
 }
 
 
-int nrswaps;
-
-
-void mgaWaitForVBlank( mgaContextPtr mmesa )
-{
-#if 0
-    drmVBlank vbl;
-    int ret;
-
-    if ( !mmesa->mgaScreen->irq )
-	return;
-
-    if ( getenv("LIBGL_SYNC_REFRESH") ) {
-	/* Wait for until the next vertical blank */
-	vbl.request.type = DRM_VBLANK_RELATIVE;
-	vbl.request.sequence = 1;
-    } else if ( getenv("LIBGL_THROTTLE_REFRESH") ) {
-	/* Wait for at least one vertical blank since the last call */
-	vbl.request.type = DRM_VBLANK_ABSOLUTE;
-	vbl.request.sequence = mmesa->vbl_seq + 1;
-    } else {
-	return;
-    }
-
-    if ((ret = drmWaitVBlank( mmesa->driFd, &vbl ))) {
-	fprintf(stderr, "%s: drmWaitVBlank returned %d, IRQs don't seem to be"
-		" working correctly.\nTry running with LIBGL_THROTTLE_REFRESH"
-		" and LIBL_SYNC_REFRESH unset.\n", __FUNCTION__, ret);
-	exit(1);
-    }
-
-    mmesa->vbl_seq = vbl.reply.sequence;
-#endif
-}
-
-
-/*
- * Copy the back buffer to the front buffer.
- */
-void mgaSwapBuffers(__DRIdrawablePrivate *dPriv)
+static void mgaWaitForFrameCompletion( mgaContextPtr mmesa )
 {
-   mgaContextPtr mmesa;
-   XF86DRIClipRectPtr pbox;
-   GLint nbox;
-   GLint ret, wait = 0;
-   GLint i;
+   unsigned wait = 0;
    GLuint last_frame, last_wrap;
 
-   assert(dPriv);
-   assert(dPriv->driContextPriv);
-   assert(dPriv->driContextPriv->driverPrivate);
-
-   mmesa = (mgaContextPtr) dPriv->driContextPriv->driverPrivate;
-
-   FLUSH_BATCH( mmesa );
-
-   mgaWaitForVBlank( mmesa );
-
-   LOCK_HARDWARE( mmesa );
 
    last_frame = mmesa->sarea->last_frame.head;
    last_wrap = mmesa->sarea->last_frame.wrap;
@@ -360,12 +312,46 @@ void mgaSwapBuffers(__DRIdrawablePrivate *dPriv)
       }
       UPDATE_LOCK( mmesa, DRM_LOCK_FLUSH );
 
-      for ( i = 0 ; i < 1024 ; i++ ) {
-	 /* Don't just hammer the register... */
-      }
+      UNLOCK_HARDWARE( mmesa );
+      DO_USLEEP( 1 );
+      LOCK_HARDWARE( mmesa );
    }
    if ( wait )
       fprintf( stderr, "\n" );
+}
+
+
+/*
+ * Copy the back buffer to the front buffer.
+ */
+void mgaSwapBuffers( __DRIdrawablePrivate *dPriv )
+{
+   mgaContextPtr mmesa;
+   XF86DRIClipRectPtr pbox;
+   GLint nbox;
+   GLint ret;
+   GLint i;
+   GLboolean   missed_target;
+
+
+   assert(dPriv);
+   assert(dPriv->driContextPriv);
+   assert(dPriv->driContextPriv->driverPrivate);
+
+   mmesa = (mgaContextPtr) dPriv->driContextPriv->driverPrivate;
+
+   FLUSH_BATCH( mmesa );
+
+   LOCK_HARDWARE( mmesa );
+   mgaWaitForFrameCompletion( mmesa );
+   UNLOCK_HARDWARE( mmesa );
+   driWaitForVBlank( dPriv, & mmesa->vbl_seq, mmesa->vblank_flags,
+		     & missed_target );
+   if ( missed_target ) {
+      mmesa->swap_missed_count++;
+      (void) (*mmesa->get_ust)( & mmesa->swap_missed_ust );
+   }
+   LOCK_HARDWARE( mmesa );
 
    /* Use the frontbuffer cliprects
     */
@@ -399,6 +385,8 @@ void mgaSwapBuffers(__DRIdrawablePrivate *dPriv)
    UNLOCK_HARDWARE( mmesa );
 
    mmesa->dirty |= MGA_UPLOAD_CLIPRECTS;
+   mmesa->swap_count++;
+   (void) (*mmesa->get_ust)( & mmesa->swap_ust );
 }
 
 
@@ -442,18 +430,17 @@ void mgaWaitAge( mgaContextPtr mmesa, int age  )
 }
 
 
-static int intersect_rect( XF86DRIClipRectPtr out,
-			   XF86DRIClipRectPtr a,
-			   XF86DRIClipRectPtr b )
+static GLboolean intersect_rect( XF86DRIClipRectPtr out,
+				 const XF86DRIClipRectPtr a,
+				 const XF86DRIClipRectPtr b )
 {
    *out = *a;
    if (b->x1 > out->x1) out->x1 = b->x1;
    if (b->y1 > out->y1) out->y1 = b->y1;
    if (b->x2 < out->x2) out->x2 = b->x2;
    if (b->y2 < out->y2) out->y2 = b->y2;
-   if (out->x1 > out->x2) return 0;
-   if (out->y1 > out->y2) return 0;
-   return 1;
+
+   return ((out->x1 < out->x2) && (out->y1 < out->y2));
 }
 
 
@@ -559,7 +546,7 @@ void mgaFlushVerticesLocked( mgaContextPtr mmesa )
          vertex.idx = buffer->idx;
          vertex.used = buffer->used;
          vertex.discard = discard;
-         drmCommandWrite( mmesa->driFd, DRM_MGA_VERTEX, 
+         drmCommandWrite( mmesa->driFd, DRM_MGA_VERTEX,
                           &vertex, sizeof(drmMGAVertex) );
 
 	 age_mmesa(mmesa, mmesa->sarea->last_enqueue);
diff --git a/src/mesa/drivers/dri/mga/mgaioctl.h b/src/mesa/drivers/dri/mga/mgaioctl.h
index b7cf44d6ff9..2d959d8246d 100644
--- a/src/mesa/drivers/dri/mga/mgaioctl.h
+++ b/src/mesa/drivers/dri/mga/mgaioctl.h
@@ -34,6 +34,7 @@
 #include "mga_xmesa.h"
 
 void mgaSwapBuffers( __DRIdrawablePrivate *dPriv );
+void mgaWaitForVBlank( mgaContextPtr mmesa );
 
 GLuint *mgaAllocVertexDwords( mgaContextPtr mmesa, int dwords );
 
@@ -41,7 +42,6 @@ GLuint *mgaAllocVertexDwords( mgaContextPtr mmesa, int dwords );
 void mgaGetILoadBufferLocked( mgaContextPtr mmesa );
 drmBufPtr mgaGetBufferLocked( mgaContextPtr mmesa );
 
-void mgaWaitForVBlank( mgaContextPtr mmesa );
 
 void mgaFireILoadLocked( mgaContextPtr mmesa,
 			 GLuint offset, GLuint length );
@@ -104,8 +104,9 @@ do {									\
    if ( ret < 0 ) {							\
       drmCommandNone( mmesa->driFd, DRM_MGA_RESET );			\
       UNLOCK_HARDWARE( mmesa );						\
-      fprintf( stderr, "%s: flush ret=%d\n", __FUNCTION__, ret );	\
-      /*fprintf( stderr, "drmMGAFlushDMA: return = %d\n", ret );*/	\
+      fprintf( stderr, "%s: flush return = %s (%d), flags = 0x%08x\n",	\
+	       __FUNCTION__, strerror( -ret ), -ret,			\
+	       (unsigned)(flags) );					\
       exit( 1 );							\
    }									\
 } while (0)
diff --git a/src/mesa/drivers/dri/mga/mgapixel.c b/src/mesa/drivers/dri/mga/mgapixel.c
index 0bc4b3fac5f..07e0c3a7563 100644
--- a/src/mesa/drivers/dri/mga/mgapixel.c
+++ b/src/mesa/drivers/dri/mga/mgapixel.c
@@ -35,12 +35,12 @@
 #include "mgacontext.h"
 #include "mgaioctl.h"
 #include "mgapixel.h"
-#include "mgabuffers.h"
+#include "mgastate.h"
 
-#include "xf86drm.h"
 #include "mga_common.h"
 
 #include "swrast/swrast.h"
+#include "imports.h"
 
 #define IS_AGP_MEM( mmesa, p )						  \
    ((unsigned long)mmesa->mgaScreen->buffers.map <= ((unsigned long)p) && \
@@ -134,7 +134,7 @@ check_color_per_fragment_ops( const GLcontext *ctx )
 		    !ctx->Color.ColorMask[2] ||
 		    !ctx->Color.ColorMask[3] ||
 		    ctx->Color.ColorLogicOpEnabled ||
-		    ctx->Texture.Unit[0]._ReallyEnabled ||
+		    ctx->Texture._EnabledUnits ||
 		    ctx->Depth.OcclusionTest
            ) &&
 	   ctx->Current.RasterPosValid &&
@@ -596,7 +596,7 @@ mgaTryDrawPixels( GLcontext *ctx,
 	    }
 	 }
 #else
-	 memcpy( address, pixels, rows*bufferpitch );
+	 MEMCPY( address, pixels, rows*bufferpitch );
 #endif
 
 	 do_draw_pix( ctx, x, y, width, rows,
@@ -639,42 +639,8 @@ mgaDDDrawPixels( GLcontext *ctx,
  * the same block of agp space which isn't used for anything else at
  * present.
  */
-#if defined(MESA_hacked_agp_allocator)
-static void mgaDDFreeAgpMemory( GLcontext *ctx, void *ptr )
-{
-   (void) ptr;
-}
-
-static void *mgaDDAllocateAgpMemory( GLcontext *ctx, GLsizei size )
-{
-   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
-
-   if (size < mmesa->mgaScreen->textureSize[MGA_AGP_HEAP])
-      return mmesa->mgaScreen->texVirtual[MGA_AGP_HEAP];
-   else
-      return 0;
-}
-
-static GLint mgaDDGetAgpOffset( GLcontext *ctx, const void *ptr )
-{
-   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
-
-   if (!IS_AGP_MEM(mmesa, ptr))
-      return -1;
-
-   return AGP_OFFSET(mmesa, ptr);
-}
-#endif
-
-
 void mgaDDInitPixelFuncs( GLcontext *ctx )
 {
-#if defined (MESA_experimetal_agp_allocator)
-   ctx->Driver.AllocateAgpMemory = mgaDDAllocateAgpMemory;
-   ctx->Driver.GetAgpOffset = mgaDDGetAgpOffset;
-   ctx->Driver.FreeAgpMemory = mgaDDFreeAgpMemory;
-#endif
-
    /* Pixel path fallbacks.
     */
    ctx->Driver.Accum = _swrast_Accum;
diff --git a/src/mesa/drivers/dri/mga/mgaregs.h b/src/mesa/drivers/dri/mga/mgaregs.h
index f07dc2de0be..e1291ca01bd 100644
--- a/src/mesa/drivers/dri/mga/mgaregs.h
+++ b/src/mesa/drivers/dri/mga/mgaregs.h
@@ -1052,6 +1052,7 @@
 #    define TMC_tformat_tw8a 		0x7 		/* val 7, shift 0 */
 #    define TMC_tformat_tw8al 		0x8 		/* val 8, shift 0 */
 #    define TMC_tformat_tw422 		0xa 		/* val 10, shift 0 */
+#    define TMC_tformat_tw422uyvy	0xb 		/* val 11, shift 0 */
 #    define TMC_tpitchlin_MASK 		0xfffffeff 	/* bit 8 */
 #    define TMC_tpitchlin_disable 	0x0 		
 #    define TMC_tpitchlin_enable 	0x100 		
@@ -1137,6 +1138,13 @@
 #    define TF_magfilter_nrst 		0x0 		/* val 0, shift 4 */
 #    define TF_magfilter_bilin 		0x20 		/* val 2, shift 4 */
 #    define TF_magfilter_cnst 		0x30 		/* val 3, shift 4 */
+#    define TF_uvoffset_SHIFT		17
+#    define TF_uvoffset_OGL		(0U << TF_uvoffset_SHIFT)
+#    define TF_uvoffset_D3D		(1U << TF_uvoffset_SHIFT)
+#    define TF_uvoffset_MASK		(~(1U << TF_uvoffset_SHIFT))
+#    define TF_reserved_MASK		(~0x1ff00)	/* bits 8-16 */
+#    define TF_mapnbhigh_SHIFT 		18
+#    define TF_mapnbhigh_MASK 		(~(1U << TF_mapnbhigh_SHIFT))
 #    define TF_avgstride_MASK 		0xfff7ffff 	/* bit 19 */
 #    define TF_avgstride_disable 	0x0 		
 #    define TF_avgstride_enable 	0x80000 	
@@ -1377,5 +1385,11 @@
 
 /**************** (END) AUTOMATICLY GENERATED REGISTER FILE ******************/
 
+/* Copied from mga_drv.h kernel file.
+ */
+
+#define MGA_ILOAD_ALIGN		64
+#define MGA_ILOAD_MASK		(MGA_ILOAD_ALIGN - 1)
+
 #endif 	/* _MGAREGS_H_ */
 
diff --git a/src/mesa/drivers/dri/mga/mgarender.c b/src/mesa/drivers/dri/mga/mgarender.c
index 53614c223fc..516a99099fd 100644
--- a/src/mesa/drivers/dri/mga/mgarender.c
+++ b/src/mesa/drivers/dri/mga/mgarender.c
@@ -42,9 +42,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "glheader.h"
 #include "context.h"
 #include "macros.h"
-/*//#include "mem.h"*/
+#include "imports.h"
 #include "mtypes.h"
-/*#include "mmath.h" */
 
 #include "tnl/t_context.h"
 
@@ -165,7 +164,7 @@ static GLboolean mga_run_render( GLcontext *ctx,
 
 static void mga_check_render( GLcontext *ctx, struct gl_pipeline_stage *stage )
 {
-   GLuint inputs = VERT_BIT_POS | VERT_BIT_CLIP | VERT_BIT_COLOR0;
+   GLuint inputs = VERT_BIT_CLIP | VERT_BIT_COLOR0;
 
    if (ctx->RenderMode == GL_RENDER) {
       if (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR) 
diff --git a/src/mesa/drivers/dri/mga/mgaspan.c b/src/mesa/drivers/dri/mga/mgaspan.c
index 8dc971cfc57..684239865f7 100644
--- a/src/mesa/drivers/dri/mga/mgaspan.c
+++ b/src/mesa/drivers/dri/mga/mgaspan.c
@@ -116,7 +116,7 @@
 
 #undef INIT_MONO_PIXEL
 #define INIT_MONO_PIXEL(p, color) \
-  p = MGAPACKCOLOR565( color[0], color[1], color[2] )
+  p = PACK_COLOR_565( color[0], color[1], color[2] )
 
 
 #define WRITE_RGBA( _x, _y, r, g, b, a )				\
@@ -148,7 +148,7 @@ do {								\
 
 #undef INIT_MONO_PIXEL
 #define INIT_MONO_PIXEL(p, color) \
-  p = MGAPACKCOLOR8888( color[0], color[1], color[2], color[3] )
+  p = PACK_COLOR_8888( color[3], color[0], color[1], color[2] )
 
 
 #define WRITE_RGBA(_x, _y, r, g, b, a)			\
@@ -233,11 +233,38 @@ do {								\
 
 
 
+/*
+ * This function is called to specify which buffer to read and write
+ * for software rasterization (swrast) fallbacks.  This doesn't necessarily
+ * correspond to glDrawBuffer() or glReadBuffer() calls.
+ */
+static void mgaDDSetBuffer(GLcontext *ctx, GLframebuffer *buffer,
+                           GLuint bufferBit)
+{
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+
+   if (bufferBit == FRONT_LEFT_BIT) 
+   {
+      mmesa->drawOffset = mmesa->mgaScreen->frontOffset;
+      mmesa->readOffset = mmesa->mgaScreen->frontOffset;
+   } 
+   else if (bufferBit == BACK_LEFT_BIT)
+   {
+      mmesa->drawOffset = mmesa->mgaScreen->backOffset;
+      mmesa->readOffset = mmesa->mgaScreen->backOffset;
+   }
+   else {
+      assert(0);
+   }
+}
+
 void mgaDDInitSpanFuncs( GLcontext *ctx )
 {
    mgaContextPtr mmesa = MGA_CONTEXT(ctx);
    struct swrast_device_driver *swdd = _swrast_GetDeviceDriverReference(ctx);
 
+   swdd->SetBuffer = mgaDDSetBuffer;
+
    switch (mmesa->mgaScreen->cpp) {
    case 2:
       swdd->WriteRGBASpan = mgaWriteRGBASpan_565;
diff --git a/src/mesa/drivers/dri/mga/mgastate.c b/src/mesa/drivers/dri/mga/mgastate.c
index 02b98fdb766..0d304966d7e 100644
--- a/src/mesa/drivers/dri/mga/mgastate.c
+++ b/src/mesa/drivers/dri/mga/mgastate.c
@@ -26,9 +26,9 @@
  */
 /* $XFree86: xc/lib/GL/mesa/src/drv/mga/mgastate.c,v 1.13 2002/10/30 12:51:36 alanh Exp $ */
 
-#include <stdio.h>
 
 #include "mtypes.h"
+#include "colormac.h"
 #include "dd.h"
 
 #include "mm.h"
@@ -40,19 +40,22 @@
 #include "mgatris.h"
 #include "mgaioctl.h"
 #include "mgaregs.h"
-#include "mgabuffers.h"
 
 #include "swrast/swrast.h"
 #include "array_cache/acache.h"
 #include "tnl/tnl.h"
+#include "tnl/t_context.h"
+#include "tnl/t_pipeline.h"
 #include "swrast_setup/swrast_setup.h"
 
 
+static void updateSpecularLighting( GLcontext *ctx );
+
 
 /* Some outstanding problems with accelerating logic ops...
  */
 #if defined(ACCEL_ROP)
-static GLuint mgarop_NoBLK[16] = {
+static const GLuint mgarop_NoBLK[16] = {
    DC_atype_rpl  | 0x00000000, DC_atype_rstr | 0x00080000,
    DC_atype_rstr | 0x00040000, DC_atype_rpl  | 0x000c0000,
    DC_atype_rstr | 0x00020000, DC_atype_rstr | 0x000a0000,
@@ -65,221 +68,57 @@ static GLuint mgarop_NoBLK[16] = {
 #endif
 
 
-static void mgaUpdateStencil(const GLcontext *ctx)
-{
-   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
-   GLuint stencil = 0, stencilctl = 0;
-
-   if (ctx->Stencil.Enabled)
-   {
-      stencil = ctx->Stencil.Ref[0] |
-	 ( ctx->Stencil.ValueMask[0] << 8 ) |
-	 ( ctx->Stencil.WriteMask[0] << 16 );
-
-      switch (ctx->Stencil.Function[0])
-      {
-      case GL_NEVER:
-	 MGA_SET_FIELD(stencilctl, SC_smode_MASK, SC_smode_snever);
-	 break;
-      case GL_LESS:
-	 MGA_SET_FIELD(stencilctl, SC_smode_MASK, SC_smode_slt);
-	 break;
-      case GL_LEQUAL:
-	 MGA_SET_FIELD(stencilctl, SC_smode_MASK, SC_smode_slte);
-	 break;
-      case GL_GREATER:
-	 MGA_SET_FIELD(stencilctl, SC_smode_MASK, SC_smode_sgt);
-	 break;
-      case GL_GEQUAL:
-	 MGA_SET_FIELD(stencilctl, SC_smode_MASK, SC_smode_sgte);
-	 break;
-      case GL_NOTEQUAL:
-	 MGA_SET_FIELD(stencilctl, SC_smode_MASK, SC_smode_sne);
-	 break;
-      case GL_EQUAL:
-	 MGA_SET_FIELD(stencilctl, SC_smode_MASK, SC_smode_se);
-	 break;
-      case GL_ALWAYS:
-	 MGA_SET_FIELD(stencilctl, SC_smode_MASK, SC_smode_salways);
-      default:
-	 break;
-      }
-
-      switch (ctx->Stencil.FailFunc[0])
-      {
-      case GL_KEEP:
-	 MGA_SET_FIELD(stencilctl, SC_sfailop_MASK, SC_sfailop_keep);
-	 break;
-      case GL_ZERO:
-	 MGA_SET_FIELD(stencilctl, SC_sfailop_MASK, SC_sfailop_zero);
-	 break;
-      case GL_REPLACE:
-	 MGA_SET_FIELD(stencilctl, SC_sfailop_MASK, SC_sfailop_replace);
-	 break;
-      case GL_INCR:
-	 MGA_SET_FIELD(stencilctl, SC_sfailop_MASK, SC_sfailop_incrsat);
-	 break;
-      case GL_DECR:
-	 MGA_SET_FIELD(stencilctl, SC_sfailop_MASK, SC_sfailop_decrsat);
-	 break;
-      case GL_INVERT:
-	 MGA_SET_FIELD(stencilctl, SC_sfailop_MASK, SC_sfailop_invert);
-	 break;
-      default:
-	 break;
-      }
-
-      switch (ctx->Stencil.ZFailFunc[0])
-      {
-      case GL_KEEP:
-	 MGA_SET_FIELD(stencilctl, SC_szfailop_MASK, SC_szfailop_keep);
-	 break;
-      case GL_ZERO:
-	 MGA_SET_FIELD(stencilctl, SC_szfailop_MASK, SC_szfailop_zero);
-	 break;
-      case GL_REPLACE:
-	 MGA_SET_FIELD(stencilctl, SC_szfailop_MASK, SC_szfailop_replace);
-	 break;
-      case GL_INCR:
-	 MGA_SET_FIELD(stencilctl, SC_szfailop_MASK, SC_szfailop_incrsat);
-	 break;
-      case GL_DECR:
-	 MGA_SET_FIELD(stencilctl, SC_szfailop_MASK, SC_szfailop_decrsat);
-	 break;
-      case GL_INVERT:
-	 MGA_SET_FIELD(stencilctl, SC_szfailop_MASK, SC_szfailop_invert);
-	 break;
-      default:
-	 break;
-      }
-
-      switch (ctx->Stencil.ZPassFunc[0])
-      {
-      case GL_KEEP:
-	 MGA_SET_FIELD(stencilctl, SC_szpassop_MASK, SC_szpassop_keep);
-	 break;
-      case GL_ZERO:
-	 MGA_SET_FIELD(stencilctl, SC_szpassop_MASK, SC_szpassop_zero);
-	 break;
-      case GL_REPLACE:
-	 MGA_SET_FIELD(stencilctl, SC_szpassop_MASK, SC_szpassop_replace);
-	 break;
-      case GL_INCR:
-	 MGA_SET_FIELD(stencilctl, SC_szpassop_MASK, SC_szpassop_incrsat);
-	 break;
-      case GL_DECR:
-	 MGA_SET_FIELD(stencilctl, SC_szpassop_MASK, SC_szpassop_decrsat);
-	 break;
-      case GL_INVERT:
-	 MGA_SET_FIELD(stencilctl, SC_szpassop_MASK, SC_szpassop_invert);
-	 break;
-      default:
-	 break;
-      }
-   }
-
-   mmesa->setup.stencil = stencil;
-   mmesa->setup.stencilctl = stencilctl;
-   mmesa->dirty |= MGA_UPLOAD_CONTEXT;
-}
-
-static void mgaDDStencilFunc(GLcontext *ctx, GLenum func, GLint ref,
-			     GLuint mask)
-{
-   FLUSH_BATCH( MGA_CONTEXT(ctx) );
-   MGA_CONTEXT(ctx)->new_state |= MGA_NEW_STENCIL;
-}
-
-static void mgaDDStencilMask(GLcontext *ctx, GLuint mask)
-{
-   FLUSH_BATCH( MGA_CONTEXT(ctx) );
-   MGA_CONTEXT(ctx)->new_state |= MGA_NEW_STENCIL;
-}
-
-static void mgaDDStencilOp(GLcontext *ctx, GLenum fail, GLenum zfail,
-			   GLenum zpass)
-{
-   FLUSH_BATCH( MGA_CONTEXT(ctx) );
-   MGA_CONTEXT(ctx)->new_state |= MGA_NEW_STENCIL;
-}
+/* =============================================================
+ * Alpha blending
+ */
 
-static void mgaDDClearDepth(GLcontext *ctx, GLclampd d)
+static void mgaDDAlphaFunc(GLcontext *ctx, GLenum func, GLfloat ref)
 {
    mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+   GLubyte refByte;
+   GLuint  a;
 
-   /* KW: should the ~ be there? */
-   switch (mmesa->setup.maccess & ~MA_zwidth_MASK) {
-   case MA_zwidth_16: mmesa->ClearDepth = d * 0x0000ffff; break;
-   case MA_zwidth_24: mmesa->ClearDepth = d * 0xffffff00; break;
-   case MA_zwidth_32: mmesa->ClearDepth = d * 0xffffffff; break;
-   default: return;
-   }
-}
-
-static void mgaUpdateZMode(const GLcontext *ctx)
-{
-   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-   int zmode = 0;
-
-   if (ctx->Depth.Test) {
-      switch(ctx->Depth.Func)  {
-      case GL_NEVER:
-         /* can't do this in h/w, we'll use a s/w fallback */
-	 zmode = DC_zmode_nozcmp;
-         break;
-      case GL_ALWAYS:
-	 zmode = DC_zmode_nozcmp; break;
-      case GL_LESS:
-	 zmode = DC_zmode_zlt; break;
-      case GL_LEQUAL:
-	 zmode = DC_zmode_zlte; break;
-      case GL_EQUAL:
-	 zmode = DC_zmode_ze; break;
-      case GL_GREATER:
-	 zmode = DC_zmode_zgt; break;
-      case GL_GEQUAL:
-	 zmode = DC_zmode_zgte; break;
-      case GL_NOTEQUAL:
-	 zmode = DC_zmode_zne; break;
-      default:
-	 break;
-      }
+   CLAMPED_FLOAT_TO_UBYTE(refByte, ref);
 
-      if (ctx->Depth.Mask)
-         zmode |= DC_atype_zi;
-      else
-         zmode |= DC_atype_i;
-   }
-   else {
-      zmode |= DC_zmode_nozcmp;
-      zmode |= DC_atype_i;  /* don't write to zbuffer */
+   switch ( func ) {
+   case GL_NEVER:
+      a = AC_atmode_alt;
+      refByte = 0;
+      break;
+   case GL_LESS:
+      a = AC_atmode_alt;
+      break;
+   case GL_GEQUAL:
+      a = AC_atmode_agte;
+      break;
+   case GL_LEQUAL:
+      a = AC_atmode_alte;
+      break;
+   case GL_GREATER:
+      a = AC_atmode_agt;
+      break;
+   case GL_NOTEQUAL:
+      a = AC_atmode_ane;
+      break;
+   case GL_EQUAL:
+      a = AC_atmode_ae;
+      break;
+   case GL_ALWAYS:
+      a = AC_atmode_noacmp;
+      break;
+   default:
+      a = 0;
+      break;
    }
 
-#if defined(ACCEL_ROP)
-   mmesa->setup.dwgctl &= DC_bop_MASK;
-   if (ctx->Color.ColorLogicOpEnabled)
-      zmode |= mgarop_NoBLK[(ctx->Color.LogicOp)&0xf];
-   else
-      zmode |= mgarop_NoBLK[GL_COPY & 0xf];
-#endif
-
-   mmesa->setup.dwgctl &= DC_zmode_MASK & DC_atype_MASK;
-   mmesa->setup.dwgctl |= zmode;
+   FLUSH_BATCH( mmesa );
+   mmesa->hw.alpha_func = a | MGA_FIELD( AC_atref, refByte );
    mmesa->dirty |= MGA_UPLOAD_CONTEXT;
 }
 
-
-static void mgaDDAlphaFunc(GLcontext *ctx, GLenum func, GLfloat ref)
-{
-   FLUSH_BATCH( MGA_CONTEXT(ctx) );
-   MGA_CONTEXT(ctx)->new_state |= MGA_NEW_ALPHA;
-}
-
-
 static void mgaDDBlendEquation(GLcontext *ctx, GLenum mode)
 {
    FLUSH_BATCH( MGA_CONTEXT(ctx) );
-   MGA_CONTEXT(ctx)->new_state |= MGA_NEW_ALPHA;
 
    /* BlendEquation sets ColorLogicOpEnabled in an unexpected 
     * manner.  
@@ -291,247 +130,169 @@ static void mgaDDBlendEquation(GLcontext *ctx, GLenum mode)
 
 static void mgaDDBlendFunc(GLcontext *ctx, GLenum sfactor, GLenum dfactor)
 {
-   FLUSH_BATCH( MGA_CONTEXT(ctx) );
-   MGA_CONTEXT(ctx)->new_state |= MGA_NEW_ALPHA;
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+   mgaScreenPrivate *mgaScreen = mmesa->mgaScreen;
+   GLuint   src;
+   GLuint   dst;
+
+   switch (ctx->Color.BlendSrcRGB) {
+   case GL_ZERO:
+      src = AC_src_zero; break;
+   case GL_SRC_ALPHA:
+      src = AC_src_src_alpha; break;
+   case GL_ONE:
+   default:		/* never happens */
+      src = AC_src_one; break;
+   case GL_DST_COLOR:
+      src = AC_src_dst_color; break;
+   case GL_ONE_MINUS_DST_COLOR:
+      src = AC_src_om_dst_color; break;
+   case GL_ONE_MINUS_SRC_ALPHA:
+      src = AC_src_om_src_alpha; break;
+   case GL_DST_ALPHA:
+      src = (mgaScreen->cpp == 4) 
+	  ? AC_src_dst_alpha : AC_src_one;
+      break;
+   case GL_ONE_MINUS_DST_ALPHA:
+      src = (mgaScreen->cpp == 4)
+	  ? AC_src_om_dst_alpha : AC_src_zero;
+      break;
+   case GL_SRC_ALPHA_SATURATE:
+      src = (ctx->Visual.alphaBits > 0)
+	  ? AC_src_src_alpha_sat : AC_src_zero;
+      break;
+   }
+
+   switch (ctx->Color.BlendDstRGB) {
+   case GL_SRC_ALPHA:
+      dst = AC_dst_src_alpha; break;
+   case GL_ONE_MINUS_SRC_ALPHA:
+      dst = AC_dst_om_src_alpha; break;
+   default:		/* never happens */
+   case GL_ZERO:
+      dst = AC_dst_zero; break;
+   case GL_ONE:
+      dst = AC_dst_one; break;
+   case GL_SRC_COLOR:
+      dst = AC_dst_src_color; break;
+   case GL_ONE_MINUS_SRC_COLOR:
+      dst = AC_dst_om_src_color; break;
+   case GL_DST_ALPHA:
+      dst = (mgaScreen->cpp == 4)
+	  ? AC_dst_dst_alpha : AC_dst_one;
+      break;
+   case GL_ONE_MINUS_DST_ALPHA:
+      dst = (mgaScreen->cpp == 4)
+	  ? AC_dst_om_dst_alpha : AC_dst_zero;
+      break;
+   }
+
+   FLUSH_BATCH( mmesa );
+   mmesa->hw.blend_func = (src | dst);
+   mmesa->dirty |= MGA_UPLOAD_CONTEXT;
 }
 
 static void mgaDDBlendFuncSeparate( GLcontext *ctx, GLenum sfactorRGB,
 				    GLenum dfactorRGB, GLenum sfactorA,
 				    GLenum dfactorA )
 {
-   FLUSH_BATCH( MGA_CONTEXT(ctx) );
-   MGA_CONTEXT(ctx)->new_state |= MGA_NEW_ALPHA;
+   mgaDDBlendFunc( ctx, sfactorRGB, dfactorRGB );
 }
 
 
+/* =============================================================
+ * Depth testing
+ */
 
-static void mgaDDLightModelfv(GLcontext *ctx, GLenum pname,
-			      const GLfloat *param)
+static void mgaDDDepthFunc(GLcontext *ctx, GLenum func)
 {
-   if (pname == GL_LIGHT_MODEL_COLOR_CONTROL) {
-      FLUSH_BATCH( MGA_CONTEXT(ctx) );
-      MGA_CONTEXT(ctx)->new_state |= MGA_NEW_TEXTURE;
+   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
+   int zmode;
+
+   switch (func) {
+   case GL_NEVER:
+      /* can't do this in h/w, we'll use a s/w fallback */
+      FALLBACK (ctx, MGA_FALLBACK_DEPTH, ctx->Depth.Test);
+
+      /* FALLTHROUGH */
+   case GL_ALWAYS:
+      zmode = DC_zmode_nozcmp; break;
+   case GL_LESS:
+      zmode = DC_zmode_zlt; break;
+   case GL_LEQUAL:
+      zmode = DC_zmode_zlte; break;
+   case GL_EQUAL:
+      zmode = DC_zmode_ze; break;
+   case GL_GREATER:
+      zmode = DC_zmode_zgt; break;
+   case GL_GEQUAL:
+      zmode = DC_zmode_zgte; break;
+   case GL_NOTEQUAL:
+      zmode = DC_zmode_zne; break;
+   default:
+      zmode = 0; break;
    }
-}
-
 
-static void mgaDDShadeModel(GLcontext *ctx, GLenum mode)
-{
-   FLUSH_BATCH( MGA_CONTEXT(ctx) );
-   MGA_CONTEXT(ctx)->new_state |= MGA_NEW_TEXTURE;
-}
-
-
-static void mgaDDDepthFunc(GLcontext *ctx, GLenum func)
-{
-   FLUSH_BATCH( MGA_CONTEXT(ctx) );
-   MGA_CONTEXT(ctx)->new_state |= MGA_NEW_DEPTH;
+   FLUSH_BATCH( mmesa );
+   mmesa->hw.zmode &= DC_zmode_MASK;
+   mmesa->hw.zmode |= zmode;
+   mmesa->dirty |= MGA_UPLOAD_CONTEXT;
 }
 
 static void mgaDDDepthMask(GLcontext *ctx, GLboolean flag)
 {
-   FLUSH_BATCH( MGA_CONTEXT(ctx) );
-   MGA_CONTEXT(ctx)->new_state |= MGA_NEW_DEPTH;
-}
+   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
 
-#if defined(ACCEL_ROP)
-static void mgaDDLogicOp( GLcontext *ctx, GLenum opcode )
-{
-   FLUSH_BATCH( MGA_CONTEXT(ctx) );
-   MGA_CONTEXT(ctx)->new_state |= MGA_NEW_DEPTH;
-}
-#else
-static void mgaDDLogicOp( GLcontext *ctx, GLenum opcode )
-{
-   FLUSH_BATCH( MGA_CONTEXT(ctx) );
-   FALLBACK( ctx, MGA_FALLBACK_LOGICOP, 
-	     (ctx->Color.ColorLogicOpEnabled && opcode != GL_COPY) );
-}
-#endif
 
+   FLUSH_BATCH( mmesa );
+   mmesa->hw.zmode &= DC_atype_MASK;
+   mmesa->hw.zmode |= (flag) ? DC_atype_zi : DC_atype_i;
+   mmesa->dirty |= MGA_UPLOAD_CONTEXT;
+}
 
 
-static void mgaDDFogfv(GLcontext *ctx, GLenum pname, const GLfloat *param)
+static void mgaDDClearDepth(GLcontext *ctx, GLclampd d)
 {
    mgaContextPtr mmesa = MGA_CONTEXT(ctx);
 
-   if (pname == GL_FOG_COLOR) {
-      GLuint color = MGAPACKCOLOR888((GLubyte)(ctx->Fog.Color[0]*255.0F), 
-				     (GLubyte)(ctx->Fog.Color[1]*255.0F), 
-				     (GLubyte)(ctx->Fog.Color[2]*255.0F));
-
-      MGA_STATECHANGE(mmesa, MGA_UPLOAD_CONTEXT);   
-      mmesa->setup.fogcolor = color;
+   /* Select the Z depth.  The ~ is used because the _MASK values in the
+    * MGA driver are used to mask OFF the selected bits.  In this case,
+    * we want to mask off everything except the MA_zwidth bits.
+    */
+   switch (mmesa->setup.maccess & ~MA_zwidth_MASK) {
+   case MA_zwidth_16: mmesa->ClearDepth = d * 0x0000ffff; break;
+   case MA_zwidth_24: mmesa->ClearDepth = d * 0xffffff00; break;
+   case MA_zwidth_32: mmesa->ClearDepth = d * 0xffffffff; break;
+   default: return;
    }
 }
 
 
-
-
 /* =============================================================
- * Alpha blending
+ * Fog
  */
 
 
-static void mgaUpdateAlphaMode(GLcontext *ctx)
+static void mgaDDFogfv(GLcontext *ctx, GLenum pname, const GLfloat *param)
 {
-   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-   mgaScreenPrivate *mgaScreen = mmesa->mgaScreen;
-   int a = 0;
-
-   /* determine source of alpha for blending and testing */
-   if ( !ctx->Texture.Unit[0]._ReallyEnabled ) {
-      a |= AC_alphasel_diffused;
-   }
-   else {
-      /* G400: Regardless of texture env mode, we use the alpha from the
-       * texture unit (AC_alphasel_fromtex) since it will have already
-       * been modulated by the incoming fragment color, if needed.
-       * We don't want (AC_alphasel_modulate) since that'll effectively
-       * do the modulation twice.
-       */
-      if (MGA_IS_G400(mmesa)) {
-         a |= AC_alphasel_fromtex;
-      }
-      else {
-         /* G200 */
-         switch (ctx->Texture.Unit[0].EnvMode) {
-         case GL_DECAL:
-            a |= AC_alphasel_diffused;
-         case GL_REPLACE:
-            a |= AC_alphasel_fromtex;
-            break;
-         case GL_BLEND:
-         case GL_MODULATE:
-            a |= AC_alphasel_modulated;
-            break;
-         default:
-            break;
-         }
-      }
-   }
-
-
-   /* alpha test control.
-    */
-   if (ctx->Color.AlphaEnabled) {
-      GLubyte ref = ctx->Color.AlphaRef;
-      switch (ctx->Color.AlphaFunc) {
-      case GL_NEVER:
-	 a |= AC_atmode_alt;
-	 ref = 0;
-	 break;
-      case GL_LESS:
-	 a |= AC_atmode_alt;
-	 break;
-      case GL_GEQUAL:
-	 a |= AC_atmode_agte;
-	 break;
-      case GL_LEQUAL:
-	 a |= AC_atmode_alte;
-	 break;
-      case GL_GREATER:
-	 a |= AC_atmode_agt;
-	 break;
-      case GL_NOTEQUAL:
-	 a |= AC_atmode_ane;
-	 break;
-      case GL_EQUAL:
-	 a |= AC_atmode_ae;
-	 break;
-      case GL_ALWAYS:
-	 a |= AC_atmode_noacmp;
-	 break;
-      default:
-	 break;
-      }
-      a |= MGA_FIELD(AC_atref,ref);
-   }
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
 
-   /* blending control */
-   if (ctx->Color.BlendEnabled) {
-      switch (ctx->Color.BlendSrcRGB) {
-      case GL_ZERO:
-	 a |= AC_src_zero; break;
-      case GL_SRC_ALPHA:
-	 a |= AC_src_src_alpha; break;
-      case GL_ONE:
-	 a |= AC_src_one; break;
-      case GL_DST_COLOR:
-	 a |= AC_src_dst_color; break;
-      case GL_ONE_MINUS_DST_COLOR:
-	 a |= AC_src_om_dst_color; break;
-      case GL_ONE_MINUS_SRC_ALPHA:
-	 a |= AC_src_om_src_alpha; break;
-      case GL_DST_ALPHA:
-	 if (mgaScreen->cpp == 4)
-	    a |= AC_src_dst_alpha;
-	 else
-	    a |= AC_src_one;
-	 break;
-      case GL_ONE_MINUS_DST_ALPHA:
-	 if (mgaScreen->cpp == 4)
-	    a |= AC_src_om_dst_alpha;
-	 else
-	    a |= AC_src_zero;
-	 break;
-      case GL_SRC_ALPHA_SATURATE:
-         if (ctx->Visual.alphaBits > 0)
-            a |= AC_src_src_alpha_sat;
-         else
-            a |= AC_src_zero;
-	 break;
-      default:		/* never happens */
-	 break;
-      }
+   if (pname == GL_FOG_COLOR) {
+      GLuint color = PACK_COLOR_888((GLubyte)(ctx->Fog.Color[0]*255.0F), 
+				    (GLubyte)(ctx->Fog.Color[1]*255.0F), 
+				    (GLubyte)(ctx->Fog.Color[2]*255.0F));
 
-      switch (ctx->Color.BlendDstRGB) {
-      case GL_SRC_ALPHA:
-	 a |= AC_dst_src_alpha; break;
-      case GL_ONE_MINUS_SRC_ALPHA:
-	 a |= AC_dst_om_src_alpha; break;
-      case GL_ZERO:
-	 a |= AC_dst_zero; break;
-      case GL_ONE:
-	 a |= AC_dst_one; break;
-      case GL_SRC_COLOR:
-	 a |= AC_dst_src_color; break;
-      case GL_ONE_MINUS_SRC_COLOR:
-	 a |= AC_dst_om_src_color; break;
-      case GL_DST_ALPHA:
-	 if (mgaScreen->cpp == 4)
-	    a |= AC_dst_dst_alpha;
-	 else
-	    a |= AC_dst_one;
-	 break;
-      case GL_ONE_MINUS_DST_ALPHA:
-	 if (mgaScreen->cpp == 4)
-	    a |= AC_dst_om_dst_alpha;
-	 else
-	    a |= AC_dst_zero;
-	 break;
-      default:		/* never happens */
-	 break;
-      }
-   } else {
-      a |= AC_src_one|AC_dst_zero;
+      MGA_STATECHANGE(mmesa, MGA_UPLOAD_CONTEXT);   
+      mmesa->setup.fogcolor = color;
    }
-
-   mmesa->setup.alphactrl = (AC_amode_alpha_channel |
-			     AC_astipple_disable |
-			     AC_aten_disable |
-			     AC_atmode_noacmp |
-			     a);
-
-   mmesa->dirty |= MGA_UPLOAD_CONTEXT;
 }
 
 
-
 /* =============================================================
- * Hardware clipping
+ * Scissoring
  */
 
+
 void mgaUpdateClipping(const GLcontext *ctx)
 {
    mgaContextPtr mmesa = MGA_CONTEXT(ctx);
@@ -554,13 +315,6 @@ void mgaUpdateClipping(const GLcontext *ctx)
       mmesa->scissor_rect.x2 = x2;
       mmesa->scissor_rect.y2 = y2;
 
-      if (MGA_DEBUG&DEBUG_VERBOSE_2D)
-	 fprintf(stderr, "SET SCISSOR %d,%d-%d,%d\n",
-		 mmesa->scissor_rect.x1,
-		 mmesa->scissor_rect.y1,
-		 mmesa->scissor_rect.x2,
-		 mmesa->scissor_rect.y2);
-
       mmesa->dirty |= MGA_UPLOAD_CLIPRECTS;
    }
 }
@@ -569,19 +323,10 @@ void mgaUpdateClipping(const GLcontext *ctx)
 static void mgaDDScissor( GLcontext *ctx, GLint x, GLint y,
 			  GLsizei w, GLsizei h )
 {
-   FLUSH_BATCH( MGA_CONTEXT(ctx) );
-   MGA_CONTEXT(ctx)->new_state |= MGA_NEW_CLIP;
-}
-
-
-static void mgaDDClearColor(GLcontext *ctx, 
-			    const GLfloat color[4] )
-{
-   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
-
-   mmesa->ClearColor = mgaPackColor( mmesa->mgaScreen->cpp,
-				     color[0], color[1], 
-				     color[2], color[3]);
+   if ( ctx->Scissor.Enabled ) {
+      FLUSH_BATCH( MGA_CONTEXT(ctx) );	/* don't pipeline cliprect changes */
+      mgaUpdateClipping( ctx );
+   }
 }
 
 
@@ -589,46 +334,42 @@ static void mgaDDClearColor(GLcontext *ctx,
  * Culling
  */
 
+
 #define _CULL_DISABLE 0
 #define _CULL_NEGATIVE ((1<<11)|(1<<5)|(1<<16))
 #define _CULL_POSITIVE (1<<11)
 
-
-void mgaUpdateCull( GLcontext *ctx )
+static void mgaDDCullFaceFrontFace(GLcontext *ctx, GLenum unused)
 {
    mgaContextPtr mmesa = MGA_CONTEXT(ctx);
-   GLuint mode = _CULL_DISABLE;
 
+
+   FLUSH_BATCH( mmesa );
    if (ctx->Polygon.CullFlag && 
-       mmesa->raster_primitive == GL_TRIANGLES &&       
        ctx->Polygon.CullFaceMode != GL_FRONT_AND_BACK) 
    {
-      mode = _CULL_NEGATIVE;
+      mmesa->hw.cull = _CULL_NEGATIVE;
+
       if (ctx->Polygon.CullFaceMode == GL_FRONT)
-	 mode ^= (_CULL_POSITIVE ^ _CULL_NEGATIVE);
+	 mmesa->hw.cull ^= (_CULL_POSITIVE ^ _CULL_NEGATIVE);
+
       if (ctx->Polygon.FrontFace != GL_CCW)
-	 mode ^= (_CULL_POSITIVE ^ _CULL_NEGATIVE);
-      if ((ctx->Texture.Unit[0]._ReallyEnabled & TEXTURE_2D_BIT) &&
-          (ctx->Texture.Unit[1]._ReallyEnabled & TEXTURE_2D_BIT))
-	 mode ^= (_CULL_POSITIVE ^ _CULL_NEGATIVE); /* warp bug? */
+	 mmesa->hw.cull ^= (_CULL_POSITIVE ^ _CULL_NEGATIVE);
+
+      mmesa->hw.cull_dualtex = mmesa->hw.cull ^
+	  (_CULL_POSITIVE ^ _CULL_NEGATIVE); /* warp bug? */
+   }
+   else {
+      mmesa->hw.cull = _CULL_DISABLE;
+      mmesa->hw.cull_dualtex = _CULL_DISABLE;
    }
 
-   mmesa->setup.wflag = mode;
    mmesa->dirty |= MGA_UPLOAD_CONTEXT;
 }
 
 
-static void mgaDDCullFaceFrontFace(GLcontext *ctx, GLenum mode)
-{
-   FLUSH_BATCH( MGA_CONTEXT(ctx) );
-   MGA_CONTEXT(ctx)->new_state |= MGA_NEW_CULL;
-}
-
-
-
-
 /* =============================================================
- * Color masks
+ * Masks
  */
 
 static void mgaDDColorMask(GLcontext *ctx, 
@@ -654,15 +395,11 @@ static void mgaDDColorMask(GLcontext *ctx,
    }
 }
 
+
 /* =============================================================
- * Polygon stipple
- *
- * The mga supports a subset of possible 4x4 stipples natively, GL
- * wants 32x32.  Fortunately stipple is usually a repeating pattern.
- *
- * Note: the fully opaque pattern (0xffff) has been disabled in order
- * to work around a conformance issue.
+ * Polygon state
  */
+
 static int mgaStipples[16] = {
    0xffff1,  /* See above note */
    0xa5a5,
@@ -682,6 +419,17 @@ static int mgaStipples[16] = {
    0x0000
 };
 
+/**
+ * The MGA supports a subset of possible 4x4 stipples natively, GL
+ * wants 32x32.  Fortunately stipple is usually a repeating pattern.
+ *
+ * \param ctx GL rendering context to be affected
+ * \param mask Pointer to the 32x32 stipple mask
+ *
+ * \note the fully opaque pattern (0xffff) has been disabled in order
+ * to work around a conformance issue.
+ */
+
 static void mgaDDPolygonStipple( GLcontext *ctx, const GLubyte *mask )
 {
    mgaContextPtr mmesa = MGA_CONTEXT(ctx);
@@ -730,93 +478,216 @@ static void mgaDDPolygonStipple( GLcontext *ctx, const GLubyte *mask )
    }
 }
 
+
 /* =============================================================
+ * Rendering attributes
+ *
+ * We really don't want to recalculate all this every time we bind a
+ * texture.  These things shouldn't change all that often, so it makes
+ * sense to break them out of the core texture state update routines.
  */
 
-static void mgaDDPrintDirty( const char *msg, GLuint state )
+static void updateSpecularLighting( GLcontext *ctx )
 {
-   fprintf(stderr, "%s (0x%x): %s%s%s%s%s%s%s\n",
-	   msg,
-	   (unsigned int) state,
-	   (state & MGA_WAIT_AGE) ? "wait-age, " : "",
-	   (state & MGA_UPLOAD_TEX0IMAGE)  ? "upload-tex0-img, " : "",
-	   (state & MGA_UPLOAD_TEX1IMAGE)  ? "upload-tex1-img, " : "",
-	   (state & MGA_UPLOAD_CONTEXT)        ? "upload-ctx, " : "",
-	   (state & MGA_UPLOAD_TEX0)       ? "upload-tex0, " : "",
-	   (state & MGA_UPLOAD_TEX1)       ? "upload-tex1, " : "",
-	   (state & MGA_UPLOAD_PIPE)       ? "upload-pipe, " : ""
-      );
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+   unsigned int specen;
+
+   specen = (ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR &&
+	     ctx->Light.Enabled) ? TMC_specen_enable : 0;
+
+   if ( specen != mmesa->hw.specen ) {
+      mmesa->hw.specen = specen;
+      mmesa->dirty |= MGA_UPLOAD_TEX0 | MGA_UPLOAD_TEX1;
+      
+      mgaChooseVertexState( ctx );
+   }
 }
 
-/* Push the state into the sarea and/or texture memory.
+
+/* =============================================================
+ * Materials
  */
-void mgaEmitHwStateLocked( mgaContextPtr mmesa )
-{
-   MGASAREAPrivPtr sarea = mmesa->sarea;
 
-   if (MGA_DEBUG & DEBUG_VERBOSE_MSG)
-      mgaDDPrintDirty( "mgaEmitHwStateLocked", mmesa->dirty );
 
-   if ((mmesa->dirty & MGA_UPLOAD_TEX0IMAGE) && mmesa->CurrentTexObj[0])
-      mgaUploadTexImages(mmesa, mmesa->CurrentTexObj[0]);
+static void mgaDDLightModelfv(GLcontext *ctx, GLenum pname,
+			      const GLfloat *param)
+{
+   if (pname == GL_LIGHT_MODEL_COLOR_CONTROL) {
+      FLUSH_BATCH( MGA_CONTEXT(ctx) );
+      updateSpecularLighting( ctx );
+   }
+}
 
-   if ((mmesa->dirty & MGA_UPLOAD_TEX1IMAGE) && mmesa->CurrentTexObj[1])
-      mgaUploadTexImages(mmesa, mmesa->CurrentTexObj[1]);
 
-   if (mmesa->dirty & MGA_UPLOAD_CONTEXT) {
-      memcpy( &sarea->ContextState, &mmesa->setup, sizeof(mmesa->setup));
-   }
+static void mgaDDShadeModel(GLcontext *ctx, GLenum mode)
+{
+   /* FIXME: This used to FLUSH_BATCH and set MGA_NEW_TEXTURE in new_state,
+    * FIXME: so I'm not sure what to do here now.
+    */
+}
 
-   if ((mmesa->dirty & MGA_UPLOAD_TEX0) && mmesa->CurrentTexObj[0]) {
-      memcpy(&sarea->TexState[0],
-	     &mmesa->CurrentTexObj[0]->setup,
-	     sizeof(sarea->TexState[0]));
-   }
 
-   if ((mmesa->dirty & MGA_UPLOAD_TEX1) && mmesa->CurrentTexObj[1]) {
-      memcpy(&sarea->TexState[1],
-	     &mmesa->CurrentTexObj[1]->setup,
-	     sizeof(sarea->TexState[1]));
-   }
+/* =============================================================
+ * Stencil
+ */
 
-   if (sarea->TexState[0].texctl2 !=
-       sarea->TexState[1].texctl2) {
-      memcpy(&sarea->TexState[1],
-	     &sarea->TexState[0],
-	     sizeof(sarea->TexState[0]));
-      mmesa->dirty |= MGA_UPLOAD_TEX1|MGA_UPLOAD_TEX0;
-   }
 
-   if (mmesa->dirty & MGA_UPLOAD_PIPE) {
-/*        mmesa->sarea->wacceptseq = mmesa->hw_primitive; */
-      mmesa->sarea->WarpPipe = mmesa->vertex_format;
-      mmesa->sarea->vertsize = mmesa->vertex_size;
+static void mgaDDStencilFunc(GLcontext *ctx, GLenum func, GLint ref,
+			     GLuint mask)
+{
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+   GLuint  stencil;
+   GLuint  stencilctl;
+
+   stencil = (ref << S_sref_SHIFT) | (mask << S_smsk_SHIFT);
+   switch (func)
+   {
+   case GL_NEVER:
+      stencilctl = SC_smode_snever;
+      break;
+   case GL_LESS:
+      stencilctl = SC_smode_slt;
+      break;
+   case GL_LEQUAL:
+      stencilctl = SC_smode_slte;
+      break;
+   case GL_GREATER:
+      stencilctl = SC_smode_sgt;
+      break;
+   case GL_GEQUAL:
+      stencilctl = SC_smode_sgte;
+      break;
+   case GL_NOTEQUAL:
+      stencilctl = SC_smode_sne;
+      break;
+   case GL_EQUAL:
+      stencilctl = SC_smode_se;
+      break;
+   case GL_ALWAYS:
+   default:
+      stencilctl = SC_smode_salways;
+      break;
    }
 
-   mmesa->sarea->dirty |= mmesa->dirty;
+   FLUSH_BATCH( mmesa );
+   mmesa->hw.stencil &= (S_sref_MASK & S_smsk_MASK);
+   mmesa->hw.stencil |= stencil;
+   mmesa->hw.stencilctl &= SC_smode_MASK;
+   mmesa->hw.stencilctl |= stencilctl;
+   mmesa->dirty |= MGA_UPLOAD_CONTEXT;
+}
 
-   mmesa->dirty &= (MGA_UPLOAD_CLIPRECTS|MGA_WAIT_AGE);
+static void mgaDDStencilMask(GLcontext *ctx, GLuint mask)
+{
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
 
-   /* This is a bit of a hack but seems to be the best place to ensure
-    * that separate specular is disabled when not needed.
-    */
-   if (mmesa->glCtx->Texture.Unit[0]._ReallyEnabled == 0 ||
-       !mmesa->glCtx->Light.Enabled ||
-       mmesa->glCtx->Light.Model.ColorControl == GL_SINGLE_COLOR) {
-      sarea->TexState[0].texctl2 &= ~TMC_specen_enable;
-      sarea->TexState[1].texctl2 &= ~TMC_specen_enable;
-   }
+   FLUSH_BATCH( mmesa );
+   mmesa->hw.stencil &= S_swtmsk_MASK;
+   mmesa->hw.stencil |= (mask << S_swtmsk_SHIFT);
+   mmesa->dirty |= MGA_UPLOAD_CONTEXT;
 }
 
-/* Fallback to swrast for select and feedback.
- */
-static void mgaRenderMode( GLcontext *ctx, GLenum mode )
+static void mgaDDStencilOp(GLcontext *ctx, GLenum fail, GLenum zfail,
+			   GLenum zpass)
 {
-   FALLBACK( ctx, MGA_FALLBACK_RENDERMODE, (mode != GL_RENDER) );
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+   GLuint  stencilctl;
+
+   stencilctl = 0;
+   switch (ctx->Stencil.FailFunc[0])
+   {
+   case GL_KEEP:
+      stencilctl |= SC_sfailop_keep;
+      break;
+   case GL_ZERO:
+      stencilctl |= SC_sfailop_zero;
+      break;
+   case GL_REPLACE:
+      stencilctl |= SC_sfailop_replace;
+      break;
+   case GL_INCR:
+      stencilctl |= SC_sfailop_incrsat;
+      break;
+   case GL_DECR:
+      stencilctl |= SC_sfailop_decrsat;
+      break;
+   case GL_INCR_WRAP:
+      stencilctl |= SC_sfailop_incr;
+      break;
+   case GL_DECR_WRAP:
+      stencilctl |= SC_sfailop_decr;
+      break;
+   case GL_INVERT:
+      stencilctl |= SC_sfailop_invert;
+      break;
+   default:
+      break;
+   }
+
+   switch (ctx->Stencil.ZFailFunc[0])
+   {
+   case GL_KEEP:
+      stencilctl |= SC_szfailop_keep;
+      break;
+   case GL_ZERO:
+      stencilctl |= SC_szfailop_zero;
+      break;
+   case GL_REPLACE:
+      stencilctl |= SC_szfailop_replace;
+      break;
+   case GL_INCR:
+      stencilctl |= SC_szfailop_incrsat;
+      break;
+   case GL_DECR:
+      stencilctl |= SC_szfailop_decrsat;
+      break;
+   case GL_INCR_WRAP:
+      stencilctl |= SC_szfailop_incr;
+      break;
+   case GL_DECR_WRAP:
+      stencilctl |= SC_szfailop_decr;
+      break;
+   case GL_INVERT:
+      stencilctl |= SC_szfailop_invert;
+      break;
+   default:
+      break;
+   }
+
+   switch (ctx->Stencil.ZPassFunc[0])
+   {
+   case GL_KEEP:
+      stencilctl |= SC_szpassop_keep;
+      break;
+   case GL_ZERO:
+      stencilctl |= SC_szpassop_zero;
+      break;
+   case GL_REPLACE:
+      stencilctl |= SC_szpassop_replace;
+      break;
+   case GL_INCR:
+      stencilctl |= SC_szpassop_incrsat;
+      break;
+   case GL_DECR:
+      stencilctl |= SC_szpassop_decrsat;
+      break;
+   case GL_INVERT:
+      stencilctl |= SC_szpassop_invert;
+      break;
+   default:
+      break;
+   }
+
+   FLUSH_BATCH( mmesa );
+   mmesa->hw.stencilctl &= (SC_sfailop_MASK & SC_szfailop_MASK 
+			    & SC_szpassop_MASK);
+   mmesa->hw.stencilctl |= stencilctl;
+   mmesa->dirty |= MGA_UPLOAD_CONTEXT;
 }
 
 
 /* =============================================================
+ * Window position and viewport transformation
  */
 
 void mgaCalcViewport( GLcontext *ctx )
@@ -850,8 +721,167 @@ static void mgaDepthRange( GLcontext *ctx,
    mgaCalcViewport( ctx );
 }
 
+
 /* =============================================================
+ * Miscellaneous
+ */
+
+static void mgaDDClearColor(GLcontext *ctx, 
+			    const GLfloat color[4] )
+{
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+   GLubyte c[4];
+   CLAMPED_FLOAT_TO_UBYTE(c[0], color[0]);
+   CLAMPED_FLOAT_TO_UBYTE(c[1], color[1]);
+   CLAMPED_FLOAT_TO_UBYTE(c[2], color[2]);
+   CLAMPED_FLOAT_TO_UBYTE(c[3], color[3]);
+
+   mmesa->ClearColor = mgaPackColor( mmesa->mgaScreen->cpp,
+				     c[0], c[1], c[2], c[3]);
+}
+
+
+/* Fallback to swrast for select and feedback.
  */
+static void mgaRenderMode( GLcontext *ctx, GLenum mode )
+{
+   FALLBACK( ctx, MGA_FALLBACK_RENDERMODE, (mode != GL_RENDER) );
+}
+
+
+static void mgaDDLogicOp( GLcontext *ctx, GLenum opcode )
+{
+   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
+
+   FLUSH_BATCH( mmesa );
+#if defined(ACCEL_ROP)
+   mmesa->hw.rop = mgarop_NoBLK[ opcode & 0x0f ];
+   mmesa->dirty |= MGA_UPLOAD_CONTEXT;
+#else
+   FALLBACK( ctx, MGA_FALLBACK_LOGICOP,
+	     (ctx->Color.ColorLogicOpEnabled && opcode != GL_COPY) );
+#endif
+}
+
+
+static void mgaXMesaSetFrontClipRects( mgaContextPtr mmesa )
+{
+   __DRIdrawablePrivate *driDrawable = mmesa->driDrawable;
+
+   if (driDrawable->numClipRects == 0) {
+       static XF86DRIClipRectRec zeroareacliprect = {0,0,0,0};
+       mmesa->numClipRects = 1;
+       mmesa->pClipRects = &zeroareacliprect;
+   } else {
+       mmesa->numClipRects = driDrawable->numClipRects;
+       mmesa->pClipRects = driDrawable->pClipRects;
+   }
+   mmesa->drawX = driDrawable->x;
+   mmesa->drawY = driDrawable->y;
+
+   mmesa->setup.dstorg = mmesa->drawOffset;
+   mmesa->dirty |= MGA_UPLOAD_CONTEXT | MGA_UPLOAD_CLIPRECTS;
+}
+
+
+static void mgaXMesaSetBackClipRects( mgaContextPtr mmesa )
+{
+   __DRIdrawablePrivate *driDrawable = mmesa->driDrawable;
+
+   if (driDrawable->numBackClipRects == 0)
+   {
+      if (driDrawable->numClipRects == 0) {
+	  static XF86DRIClipRectRec zeroareacliprect = {0,0,0,0};
+	  mmesa->numClipRects = 1;
+	  mmesa->pClipRects = &zeroareacliprect;
+      } else {
+	  mmesa->numClipRects = driDrawable->numClipRects;
+	  mmesa->pClipRects = driDrawable->pClipRects;
+      }
+      mmesa->drawX = driDrawable->x;
+      mmesa->drawY = driDrawable->y;
+   } else {
+      mmesa->numClipRects = driDrawable->numBackClipRects;
+      mmesa->pClipRects = driDrawable->pBackClipRects;
+      mmesa->drawX = driDrawable->backX;
+      mmesa->drawY = driDrawable->backY;
+   }
+
+   mmesa->setup.dstorg = mmesa->drawOffset;
+   mmesa->dirty |= MGA_UPLOAD_CONTEXT | MGA_UPLOAD_CLIPRECTS;
+}
+
+
+void mgaUpdateRects( mgaContextPtr mmesa, GLuint buffers )
+{
+   __DRIdrawablePrivate *driDrawable = mmesa->driDrawable;
+   MGASAREAPrivPtr sarea = mmesa->sarea;
+
+
+   DRI_VALIDATE_DRAWABLE_INFO(mmesa->driScreen, driDrawable); 
+   mmesa->dirty_cliprects = 0;	
+
+   if (mmesa->draw_buffer == MGA_FRONT)
+      mgaXMesaSetFrontClipRects( mmesa );
+   else
+      mgaXMesaSetBackClipRects( mmesa );
+
+   sarea->req_draw_buffer = mmesa->draw_buffer;
+
+   mgaUpdateClipping( mmesa->glCtx );
+   mgaCalcViewport( mmesa->glCtx );
+
+   mmesa->dirty |= MGA_UPLOAD_CLIPRECTS;
+}
+
+
+static void mgaDDDrawBuffer(GLcontext *ctx, GLenum mode )
+{
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+
+   FLUSH_BATCH( mmesa );
+
+   /*
+    * _DrawDestMask is easier to cope with than <mode>.
+    */
+   switch ( ctx->Color._DrawDestMask ) {
+   case FRONT_LEFT_BIT:
+      mmesa->setup.dstorg = mmesa->mgaScreen->frontOffset;
+      mmesa->dirty |= MGA_UPLOAD_CONTEXT;
+      mmesa->draw_buffer = MGA_FRONT;
+      mgaXMesaSetFrontClipRects( mmesa );
+      FALLBACK( ctx, MGA_FALLBACK_DRAW_BUFFER, GL_FALSE );
+      break;
+   case BACK_LEFT_BIT:
+      mmesa->setup.dstorg = mmesa->mgaScreen->backOffset;
+      mmesa->draw_buffer = MGA_BACK;
+      mmesa->dirty |= MGA_UPLOAD_CONTEXT;
+      mgaXMesaSetBackClipRects( mmesa );
+      FALLBACK( ctx, MGA_FALLBACK_DRAW_BUFFER, GL_FALSE );
+      break;
+   default:
+      /* GL_NONE or GL_FRONT_AND_BACK or stereo left&right, etc */
+      FALLBACK( ctx, MGA_FALLBACK_DRAW_BUFFER, GL_TRUE );
+      return;
+   }
+
+   /* We want to update the s/w rast state too so that r200SetBuffer()
+    * gets called.
+    */
+   _swrast_DrawBuffer(ctx, mode);
+}
+
+
+static void mgaDDReadBuffer(GLcontext *ctx, GLenum mode )
+{
+   /* nothing, until we implement h/w glRead/CopyPixels or CopyTexImage */
+}
+
+
+/* =============================================================
+ * State enable/disable
+ */
+
 
 static void mgaDDEnable(GLcontext *ctx, GLenum cap, GLboolean state)
 {
@@ -860,11 +890,11 @@ static void mgaDDEnable(GLcontext *ctx, GLenum cap, GLboolean state)
    switch(cap) {
    case GL_ALPHA_TEST:
       FLUSH_BATCH( mmesa );
-      mmesa->new_state |= MGA_NEW_ALPHA;
+      mmesa->hw.alpha_func_enable = (state) ? ~0 : 0;
       break;
    case GL_BLEND:
       FLUSH_BATCH( mmesa );
-      mmesa->new_state |= MGA_NEW_ALPHA;
+      mmesa->hw.blend_func_enable = (state) ? ~0 : 0;
 
       /* For some reason enable(GL_BLEND) affects ColorLogicOpEnabled.
        */
@@ -874,31 +904,31 @@ static void mgaDDEnable(GLcontext *ctx, GLenum cap, GLboolean state)
       break;
    case GL_DEPTH_TEST:
       FLUSH_BATCH( mmesa );
-      mmesa->new_state |= MGA_NEW_DEPTH;
       FALLBACK (ctx, MGA_FALLBACK_DEPTH,
 		ctx->Depth.Func == GL_NEVER && ctx->Depth.Test);
       break;
+
    case GL_SCISSOR_TEST:
       FLUSH_BATCH( mmesa );
       mmesa->scissor = state;
-      mmesa->new_state |= MGA_NEW_CLIP;
+      mgaUpdateClipping( ctx );
       break;
+
    case GL_FOG:
       MGA_STATECHANGE( mmesa, MGA_UPLOAD_CONTEXT );
       if (ctx->Fog.Enabled) 
 	 mmesa->setup.maccess |= MA_fogen_enable;
       else
 	 mmesa->setup.maccess &= ~MA_fogen_enable;
+      
+      mgaChooseVertexState( ctx );
       break;
    case GL_CULL_FACE:
-      FLUSH_BATCH( mmesa );
-      mmesa->new_state |= MGA_NEW_CULL;
+      mgaDDCullFaceFrontFace( ctx, 0 );
       break;
    case GL_TEXTURE_1D:
    case GL_TEXTURE_2D:
    case GL_TEXTURE_3D:
-      FLUSH_BATCH( mmesa );
-      mmesa->new_state |= (MGA_NEW_TEXTURE|MGA_NEW_ALPHA);
       break;
    case GL_POLYGON_STIPPLE:
       if (mmesa->haveHwStipple && mmesa->raster_primitive == GL_TRIANGLES) {
@@ -914,16 +944,16 @@ static void mgaDDEnable(GLcontext *ctx, GLenum cap, GLboolean state)
 #if !defined(ACCEL_ROP)
       FALLBACK( ctx, MGA_FALLBACK_LOGICOP, 
 		(state && ctx->Color.LogicOp != GL_COPY));
-#else
-      mmesa->new_state |= MGA_NEW_DEPTH;
 #endif
       break;
    case GL_STENCIL_TEST:
       FLUSH_BATCH( mmesa );
-      if (mmesa->hw_stencil)
-	 mmesa->new_state |= MGA_NEW_STENCIL;
-      else
+      if (mmesa->hw_stencil) {
+	 mmesa->hw.stencil_enable = ( state ) ? ~0 : 0;
+      }
+      else {
 	 FALLBACK( ctx, MGA_FALLBACK_STENCIL, state );
+      }
    default:
       break;
    }
@@ -933,56 +963,150 @@ static void mgaDDEnable(GLcontext *ctx, GLenum cap, GLboolean state)
 /* =============================================================
  */
 
-
-
-/* =============================================================
- */
-
-static void mgaDDPrintState( const char *msg, GLuint state )
+static void mgaDDPrintDirty( const char *msg, GLuint state )
 {
-   fprintf(stderr, "%s (0x%x): %s%s%s%s%s%s\n",
+   fprintf(stderr, "%s (0x%03x): %s%s%s%s%s%s%s\n",
 	   msg,
-	   state,
-	   (state & MGA_NEW_DEPTH)   ? "depth, " : "",
-	   (state & MGA_NEW_ALPHA)   ? "alpha, " : "",
-	   (state & MGA_NEW_CLIP)    ? "clip, " : "",
-	   (state & MGA_NEW_CULL)    ? "cull, " : "",
-	   (state & MGA_NEW_TEXTURE) ? "texture, " : "",
-	   (state & MGA_NEW_CONTEXT) ? "context, " : "");
+	   (unsigned int) state,
+	   (state & MGA_WAIT_AGE)          ? "wait-age " : "",
+	   (state & MGA_UPLOAD_TEX0IMAGE)  ? "upload-tex0-img " : "",
+	   (state & MGA_UPLOAD_TEX1IMAGE)  ? "upload-tex1-img " : "",
+	   (state & MGA_UPLOAD_CONTEXT)    ? "upload-ctx " : "",
+	   (state & MGA_UPLOAD_TEX0)       ? "upload-tex0 " : "",
+	   (state & MGA_UPLOAD_TEX1)       ? "upload-tex1 " : "",
+	   (state & MGA_UPLOAD_PIPE)       ? "upload-pipe " : ""
+      );
 }
 
-void mgaDDUpdateHwState( GLcontext *ctx )
+/* Push the state into the sarea and/or texture memory.
+ */
+void mgaEmitHwStateLocked( mgaContextPtr mmesa )
 {
-   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-   int new_state = mmesa->new_state;
+   MGASAREAPrivPtr sarea = mmesa->sarea;
+   GLcontext * ctx = mmesa->glCtx;
 
-   if (new_state)
-   {
-      FLUSH_BATCH( mmesa );
+   if (MGA_DEBUG & DEBUG_VERBOSE_MSG)
+      mgaDDPrintDirty( __FUNCTION__, mmesa->dirty );
 
-      mmesa->new_state = 0;
+   if (mmesa->dirty & MGA_UPLOAD_CONTEXT) {
+      mmesa->setup.wflag = _CULL_DISABLE;
+      if (mmesa->raster_primitive == GL_TRIANGLES) {
+	 if ((ctx->Texture.Unit[0]._ReallyEnabled == TEXTURE_2D_BIT &&
+	      ctx->Texture.Unit[1]._ReallyEnabled == TEXTURE_2D_BIT)) {
+	    mmesa->setup.wflag = mmesa->hw.cull_dualtex;
+	 }
+	 else {
+	    mmesa->setup.wflag = mmesa->hw.cull;
+	 }
+      }
 
-      if (MESA_VERBOSE&VERBOSE_DRIVER)
-	 mgaDDPrintState("UpdateHwState", new_state);
+      mmesa->setup.stencil = mmesa->hw.stencil 
+	  & mmesa->hw.stencil_enable;
+      mmesa->setup.stencilctl = mmesa->hw.stencilctl
+	  & mmesa->hw.stencil_enable;
 
-      if (new_state & MGA_NEW_DEPTH)
-	 mgaUpdateZMode(ctx);
+      /* If depth testing is not enabled, then use the no Z-compare / no
+       * Z-write mode.  Otherwise, use whatever is set in hw.zmode.
+       */
+      mmesa->setup.dwgctl &= (DC_zmode_MASK & DC_atype_MASK);
+      mmesa->setup.dwgctl |= (ctx->Depth.Test)
+	  ? mmesa->hw.zmode : (DC_zmode_nozcmp | DC_atype_i);
 
-      if (new_state & MGA_NEW_ALPHA)
-	 mgaUpdateAlphaMode(ctx);
+#if defined(ACCEL_ROP)
+      mmesa->setup.dwgctl &= DC_bop_MASK;
+      mmesa->setup.dwgctl |= (ctx->Color.ColorLogicOpEnabled)
+	  ? mmesa->hw.rop : mgarop_NoBLK[ GL_COPY & 0x0f ];
+#endif
 
-      if (new_state & MGA_NEW_CLIP)
-	 mgaUpdateClipping(ctx);
+      mmesa->setup.alphactrl &= AC_src_MASK & AC_dst_MASK & AC_atmode_MASK
+	  & AC_atref_MASK & AC_alphasel_MASK;
+      mmesa->setup.alphactrl |= 
+	  ((mmesa->hw.alpha_func & mmesa->hw.alpha_func_enable)
+	   | ((mmesa->hw.blend_func & mmesa->hw.blend_func_enable)
+	      | ((AC_src_one | AC_dst_zero) & ~mmesa->hw.blend_func_enable))
+	   | mmesa->hw.alpha_sel
+	   | (AC_amode_alpha_channel
+	      | AC_astipple_disable 
+	      | AC_aten_disable 
+	      | AC_atmode_noacmp));
 
-      if (new_state & MGA_NEW_STENCIL)
-	 mgaUpdateStencil(ctx);
+      memcpy( &sarea->ContextState, &mmesa->setup, sizeof(mmesa->setup));
+   }
 
-      if (new_state & (MGA_NEW_WARP|MGA_NEW_CULL))
-	 mgaUpdateCull(ctx);
+   if ((mmesa->dirty & MGA_UPLOAD_TEX0) && mmesa->CurrentTexObj[0]) {
+      mmesa->CurrentTexObj[0]->setup.texctl2 &= ~TMC_specen_enable;
+      mmesa->CurrentTexObj[0]->setup.texctl2 |= mmesa->hw.specen;
 
-      if (new_state & (MGA_NEW_WARP|MGA_NEW_TEXTURE))
-	 mgaUpdateTextureState(ctx);
+      memcpy(&sarea->TexState[0],
+	     &mmesa->CurrentTexObj[0]->setup,
+	     sizeof(sarea->TexState[0]));
    }
+
+   if ((mmesa->dirty & MGA_UPLOAD_TEX1) && mmesa->CurrentTexObj[1]) {
+      mmesa->CurrentTexObj[1]->setup.texctl2 &= ~TMC_specen_enable;
+      mmesa->CurrentTexObj[1]->setup.texctl2 |= mmesa->hw.specen;
+
+      memcpy(&sarea->TexState[1],
+	     &mmesa->CurrentTexObj[1]->setup,
+	     sizeof(sarea->TexState[1]));
+   }
+
+   if ( (sarea->TexState[0].texctl2 & TMC_borderen_MASK) !=
+	(sarea->TexState[1].texctl2 & TMC_borderen_MASK) ) {
+      const int borderen = sarea->TexState[1].texctl2 & ~TMC_borderen_MASK;
+
+      memcpy( &sarea->TexState[1], &sarea->TexState[0],
+	      sizeof(sarea->TexState[0]) );
+      sarea->TexState[1].texctl2 |= borderen;
+      mmesa->dirty |= MGA_UPLOAD_TEX1|MGA_UPLOAD_TEX0;
+   }
+
+   if (mmesa->dirty & MGA_UPLOAD_PIPE) {
+/*        mmesa->sarea->wacceptseq = mmesa->hw_primitive; */
+      mmesa->sarea->WarpPipe = mmesa->vertex_format;
+      mmesa->sarea->vertsize = mmesa->vertex_size;
+   }
+
+   mmesa->sarea->dirty |= mmesa->dirty;
+   mmesa->dirty &= MGA_UPLOAD_CLIPRECTS;
+
+   /* This is a bit of a hack but seems to be the best place to ensure
+    * that separate specular is disabled when not needed.
+    */
+   if (ctx->Texture._EnabledUnits == 0 ||
+       !ctx->Light.Enabled ||
+       ctx->Light.Model.ColorControl == GL_SINGLE_COLOR) {
+      sarea->TexState[0].texctl2 &= ~TMC_specen_enable;
+      sarea->TexState[1].texctl2 &= ~TMC_specen_enable;
+   }
+}
+
+
+/* =============================================================
+ */
+
+
+static void mgaDDValidateState( GLcontext *ctx )
+{
+   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
+   int new_state = mmesa->NewGLState;
+
+
+   FLUSH_BATCH( mmesa );
+
+   if (mmesa->NewGLState & _MGA_NEW_RASTERSETUP) {
+      mgaChooseVertexState( ctx );
+   }
+
+   if (mmesa->NewGLState & _MGA_NEW_RENDERSTATE) {
+      mgaChooseRenderState( ctx );
+   }
+
+   if (new_state & _NEW_TEXTURE) {
+      mgaUpdateTextureState(ctx);
+   }
+
+   mmesa->NewGLState = 0;
 }
 
 
@@ -992,25 +1116,40 @@ static void mgaDDInvalidateState( GLcontext *ctx, GLuint new_state )
    _swsetup_InvalidateState( ctx, new_state );
    _ac_InvalidateState( ctx, new_state );
    _tnl_InvalidateState( ctx, new_state );
-   MGA_CONTEXT(ctx)->new_gl_state |= new_state;
+   MGA_CONTEXT(ctx)->NewGLState |= new_state;
 }
 
 
+static void mgaRunPipeline( GLcontext *ctx )
+{
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+
+   if (mmesa->NewGLState) {
+      mgaDDValidateState( ctx );
+   }
+
+   if (mmesa->dirty) {
+       mgaEmitHwStateLocked( mmesa );
+   }
+
+   _tnl_run_pipeline( ctx );
+}
+
 
 void mgaInitState( mgaContextPtr mmesa )
 {
    mgaScreenPrivate *mgaScreen = mmesa->mgaScreen;
    GLcontext *ctx = mmesa->glCtx;
 
-   if (ctx->Color._DrawDestMask == BACK_LEFT_BIT) {
+   if (ctx->Visual.doubleBufferMode) {
+      /* use back buffer by default */
       mmesa->draw_buffer = MGA_BACK;
-      mmesa->read_buffer = MGA_BACK;
       mmesa->drawOffset = mmesa->mgaScreen->backOffset;
       mmesa->readOffset = mmesa->mgaScreen->backOffset;
       mmesa->setup.dstorg = mgaScreen->backOffset;
    } else {
+      /* use front buffer by default */
       mmesa->draw_buffer = MGA_FRONT;
-      mmesa->read_buffer = MGA_FRONT;
       mmesa->drawOffset = mmesa->mgaScreen->frontOffset;
       mmesa->readOffset = mmesa->mgaScreen->frontOffset;
       mmesa->setup.dstorg = mgaScreen->frontOffset;
@@ -1043,14 +1182,21 @@ void mgaInitState( mgaContextPtr mmesa )
       mmesa->setup.maccess |= MA_zwidth_24;
       break;
    case 32:
-      mmesa->setup.maccess |= MA_pwidth_32;
+      mmesa->setup.maccess |= MA_zwidth_32;
       break;
    }
 
+   mmesa->hw.zmode = DC_zmode_zlt | DC_atype_zi;
+   mmesa->hw.stencil = (0x0ff << S_smsk_SHIFT) | (0x0ff << S_swtmsk_SHIFT);
+   mmesa->hw.stencilctl = SC_smode_salways | SC_sfailop_keep 
+       | SC_szfailop_keep | SC_szpassop_keep;
+   mmesa->hw.stencil_enable = 0;
+   mmesa->hw.cull = _CULL_NEGATIVE;
+   mmesa->hw.cull_dualtex = _CULL_POSITIVE;
+   mmesa->hw.specen = 0;
+
    mmesa->setup.dwgctl = (DC_opcod_trap |
-			  DC_atype_i |
 			  DC_linear_xy |
-			  DC_zmode_nozcmp |
 			  DC_solid_disable |
 			  DC_arzero_disable |
 			  DC_sgnzero_disable |
@@ -1072,16 +1218,15 @@ void mgaInitState( mgaContextPtr mmesa )
 			      AC_atmode_noacmp |
 			      AC_alphasel_fromtex );
 
-   mmesa->setup.fogcolor =
-      MGAPACKCOLOR888((GLubyte)(ctx->Fog.Color[0]*255.0F),
-		      (GLubyte)(ctx->Fog.Color[1]*255.0F),
-		      (GLubyte)(ctx->Fog.Color[2]*255.0F));
+   mmesa->setup.fogcolor = PACK_COLOR_888((GLubyte)(ctx->Fog.Color[0]*255.0F),
+					  (GLubyte)(ctx->Fog.Color[1]*255.0F),
+					  (GLubyte)(ctx->Fog.Color[2]*255.0F));
 
    mmesa->setup.wflag = 0;
    mmesa->setup.tdualstage0 = 0;
    mmesa->setup.tdualstage1 = 0;
    mmesa->setup.fcol = 0;
-   mmesa->new_state = ~0;
+   mmesa->dirty |= MGA_UPLOAD_CONTEXT;
 }
 
 
@@ -1103,8 +1248,8 @@ void mgaDDInitStateFuncs( GLcontext *ctx )
    ctx->Driver.FrontFace = mgaDDCullFaceFrontFace;
    ctx->Driver.ColorMask = mgaDDColorMask;
 
-   ctx->Driver.DrawBuffer = mgaDDSetDrawBuffer;
-   ctx->Driver.ReadBuffer = mgaDDSetReadBuffer;
+   ctx->Driver.DrawBuffer = mgaDDDrawBuffer;
+   ctx->Driver.ReadBuffer = mgaDDReadBuffer;
    ctx->Driver.ClearColor = mgaDDClearColor;
    ctx->Driver.ClearDepth = mgaDDClearDepth;
    ctx->Driver.LogicOpcode = mgaDDLogicOp;
@@ -1128,4 +1273,6 @@ void mgaDDInitStateFuncs( GLcontext *ctx )
    ctx->Driver.CopyColorSubTable = _swrast_CopyColorSubTable;
    ctx->Driver.CopyConvolutionFilter1D = _swrast_CopyConvolutionFilter1D;
    ctx->Driver.CopyConvolutionFilter2D = _swrast_CopyConvolutionFilter2D;
+
+   TNL_CONTEXT(ctx)->Driver.RunPipeline = mgaRunPipeline;
 }
diff --git a/src/mesa/drivers/dri/mga/mgastate.h b/src/mesa/drivers/dri/mga/mgastate.h
index a9f1039d760..afbe0aaf904 100644
--- a/src/mesa/drivers/dri/mga/mgastate.h
+++ b/src/mesa/drivers/dri/mga/mgastate.h
@@ -29,14 +29,11 @@
 #ifndef _MGA_STATE_H
 #define _MGA_STATE_H
 
-
 extern void mgaInitState( mgaContextPtr mmesa );
 extern void mgaDDInitStateFuncs(GLcontext *ctx);
-extern void mgaDDUpdateHwState( GLcontext *ctx );
 extern void mgaUpdateClipping(const GLcontext *ctx);
 extern void mgaUpdateCull( GLcontext *ctx );
 extern void mgaCalcViewport( GLcontext *ctx );
-
-
+extern void mgaUpdateRects( mgaContextPtr mmesa, GLuint buffers );
 
 #endif
diff --git a/src/mesa/drivers/dri/mga/mgatex.c b/src/mesa/drivers/dri/mga/mgatex.c
index b4ee787e0b8..82eccc81498 100644
--- a/src/mesa/drivers/dri/mga/mgatex.c
+++ b/src/mesa/drivers/dri/mga/mgatex.c
@@ -26,10 +26,7 @@
  */
 /* $XFree86: xc/lib/GL/mesa/src/drv/mga/mgatex.c,v 1.14 2002/10/30 12:51:36 alanh Exp $ */
 
-#include <stdlib.h>
-#include <stdio.h>
-#include <GL/gl.h>
-
+#include "glheader.h"
 #include "mm.h"
 #include "mgacontext.h"
 #include "mgatex.h"
@@ -37,9 +34,11 @@
 #include "mgatris.h"
 #include "mgaioctl.h"
 
+#include "colormac.h"
+#include "context.h"
 #include "enums.h"
 #include "simple_list.h"
-/*#include "mem.h"*/
+#include "imports.h"
 #include "macros.h"
 #include "texformat.h"
 #include "texstore.h"
@@ -47,67 +46,72 @@
 
 #include "swrast/swrast.h"
 
-#define TEX_0 1
-#define TEX_1 2
-
-/*
- * mgaDestroyTexObj
- * Free all memory associated with a texture and NULL any pointers
- * to it.
+/**
+ * Set the texture wrap modes.
+ * Currently \c GL_REPEAT, \c GL_CLAMP and \c GL_CLAMP_TO_EDGE are supported.
+ * 
+ * \param t Texture object whose wrap modes are to be set
+ * \param swrap Wrap mode for the \a s texture coordinate
+ * \param twrap Wrap mode for the \a t texture coordinate
  */
-void
-mgaDestroyTexObj( mgaContextPtr mmesa, mgaTextureObjectPtr t )
+
+static void 
+mgaSetTexWrapping( mgaTextureObjectPtr t, GLenum swrap, GLenum twrap )
 {
-   if ( !t ) return;
+   GLboolean  is_clamp = GL_FALSE;
+   GLboolean  is_clamp_to_edge = GL_FALSE;
 
-   /* free the texture memory */
-   if (t->MemBlock) {
-      mmFreeMem( t->MemBlock );
-      t->MemBlock = 0;
+   t->setup.texctl &= (TMC_clampu_MASK & TMC_clampv_MASK);
+   t->setup.texctl2 &= (TMC_borderen_MASK);
 
-      if (mmesa && t->age > mmesa->dirtyAge)
-	 mmesa->dirtyAge = t->age;
+   switch( swrap ) {
+   case GL_REPEAT:
+      break;
+   case GL_CLAMP:
+      t->setup.texctl |= TMC_clampu_enable;
+      is_clamp = GL_TRUE;
+      break;
+   case GL_CLAMP_TO_EDGE:
+      t->setup.texctl |= TMC_clampu_enable;
+      is_clamp_to_edge = GL_TRUE;
+      break;
+   default:
+      _mesa_problem(NULL, "bad S wrap mode in %s", __FUNCTION__);
    }
 
-   /* free mesa's link */
-   if (t->tObj)
-      t->tObj->DriverData = NULL;
-
-   /* see if it was the driver's current object */
-   if (mmesa) {
-      if (t->bound & TEX_0) mmesa->CurrentTexObj[0] = 0;
-      if (t->bound & TEX_1) mmesa->CurrentTexObj[1] = 0;
+   switch( twrap ) {
+   case GL_REPEAT:
+      break;
+   case GL_CLAMP:
+      t->setup.texctl |= TMC_clampv_enable;
+      is_clamp = GL_TRUE;
+      break;
+   case GL_CLAMP_TO_EDGE:
+      t->setup.texctl |= TMC_clampv_enable;
+      is_clamp_to_edge = GL_TRUE;
+      break;
+   default:
+      _mesa_problem(NULL, "bad T wrap mode in %s", __FUNCTION__);
    }
-	
-   remove_from_list(t);
-   free( t );
-}
 
+   if ( is_clamp ) {
+      t->setup.texctl2 |= TMC_borderen_enable;
+   }
 
-/*
- * mgaSetTexWrappings
- */
-static void mgaSetTexWrapping( mgaTextureObjectPtr t,
-			       GLenum sWrap,
-			       GLenum tWrap )
-{
-   GLuint val = 0;
-
-   if (sWrap != GL_REPEAT)
-      val |= TMC_clampu_enable;
-
-   if (tWrap != GL_REPEAT)
-      val |= TMC_clampv_enable;
-
-   t->setup.texctl &= ~(TMC_clampu_enable|TMC_clampv_enable);
-   t->setup.texctl |= val;
+   t->border_fallback = (is_clamp && is_clamp_to_edge);
 }
 
 
-/*
- * mgaSetTexFilter
+/**
+ * Set the texture magnification and minification modes.
+ * 
+ * \param t Texture whose filter modes are to be set
+ * \param minf Texture minification mode
+ * \param magf Texture magnification mode
  */
-static void mgaSetTexFilter(mgaTextureObjectPtr t, GLenum minf, GLenum magf)
+
+static void
+mgaSetTexFilter( mgaTextureObjectPtr t, GLenum minf, GLenum magf )
 {
    GLuint val = 0;
 
@@ -136,123 +140,79 @@ static void mgaSetTexFilter(mgaTextureObjectPtr t, GLenum minf, GLenum magf)
    }
 
 
-   t->setup.texfilter &= (TF_minfilter_MASK |
-			  TF_magfilter_MASK |
+   /* Mask off the bits for the fields we are setting.  Remember, the MGA mask
+    * defines have 0s for the bits in the named fields.  This is the opposite
+    * of most of the other drivers.
+    */
+
+   t->setup.texfilter &= (TF_minfilter_MASK &
+			  TF_magfilter_MASK &
 			  TF_fthres_MASK);
    t->setup.texfilter |= val;
 }
 
-/*
- * mgaSetTexBorderColor
- */
 static void mgaSetTexBorderColor(mgaTextureObjectPtr t, GLubyte color[4])
 {
-   t->setup.texbordercol = MGAPACKCOLOR8888(color[0],color[1],
-					    color[2],color[3]);
+   t->setup.texbordercol = PACK_COLOR_8888(color[3], color[0], 
+					   color[1], color[2] );
 }
 
 
-static GLint mgaChooseTexFormat( mgaContextPtr mmesa,
-				 struct gl_texture_image *texImage,
-				 GLenum format, GLenum type )
+static const struct gl_texture_format *
+mgaChooseTextureFormat( GLcontext *ctx, GLint internalFormat,
+		        GLenum format, GLenum type )
 {
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
    const GLboolean do32bpt = mmesa->default32BitTextures;
-   const struct gl_texture_format *texFormat;
-   GLint ret;
-
-   if ( 0 )
-      fprintf( stderr, "internal=%s format=%s type=%s\n",
-	       texImage->IntFormat == 3 ? "GL_RGB (3)" :
-	       texImage->IntFormat == 4 ? "GL_RGBA (4)" :
-	       _mesa_lookup_enum_by_nr( texImage->IntFormat ),
-	       _mesa_lookup_enum_by_nr( format ),
-	       _mesa_lookup_enum_by_nr( type ) );
-
-#define SET_FORMAT( r, gl )						\
-   do {									\
-      ret = (r);							\
-      texFormat = &(gl);						\
-   } while (0)
-
-#define SET_FORMAT_32BPT( r32, gl32, r16, gl16 )			\
-   do {									\
-      if ( do32bpt ) {							\
-	 ret = (r32);							\
-	 texFormat = &(gl32);						\
-      } else {								\
-	 ret = (r16);							\
-	 texFormat = &(gl16);						\
-      }									\
-   } while (0)
-
-   switch ( texImage->IntFormat ) {
-      /* GH: Bias towards GL_RGB, GL_RGBA texture formats.  This has
-       * got to be better than sticking them way down the end of this
-       * huge list.
-       */
+
+   switch ( internalFormat ) {
    case 4:
    case GL_RGBA:
    case GL_COMPRESSED_RGBA:
       if ( format == GL_BGRA ) {
 	 if ( type == GL_UNSIGNED_INT_8_8_8_8_REV ) {
-	    SET_FORMAT( TMC_tformat_tw32, _mesa_texformat_argb8888 );
-	    break;
-	 } else if ( type == GL_UNSIGNED_SHORT_4_4_4_4_REV ) {
-	    SET_FORMAT( TMC_tformat_tw12, _mesa_texformat_argb4444 );
-	    break;
-	 } else if ( type == GL_UNSIGNED_SHORT_1_5_5_5_REV ) {
-	    SET_FORMAT( TMC_tformat_tw15, _mesa_texformat_argb1555 );
-	    break;
+	    return &_mesa_texformat_argb8888;
+	 }
+         else if ( type == GL_UNSIGNED_SHORT_4_4_4_4_REV ) {
+            return &_mesa_texformat_argb4444;
+	 }
+         else if ( type == GL_UNSIGNED_SHORT_1_5_5_5_REV ) {
+	    return &_mesa_texformat_argb1555;
 	 }
       }
-      SET_FORMAT_32BPT( TMC_tformat_tw32, _mesa_texformat_argb8888,
-			TMC_tformat_tw12, _mesa_texformat_argb4444 );
-      break;
+      return do32bpt ? &_mesa_texformat_argb8888 : &_mesa_texformat_argb4444;
 
    case 3:
    case GL_RGB:
    case GL_COMPRESSED_RGB:
       if ( format == GL_RGB && type == GL_UNSIGNED_SHORT_5_6_5 ) {
-	 SET_FORMAT( TMC_tformat_tw16, _mesa_texformat_rgb565 );
-	 break;
+	 return &_mesa_texformat_rgb565;
       }
-      SET_FORMAT_32BPT( TMC_tformat_tw32, _mesa_texformat_argb8888,
-			TMC_tformat_tw16, _mesa_texformat_rgb565 );
-      break;
+      return do32bpt ? &_mesa_texformat_argb8888 : &_mesa_texformat_rgb565;
 
-      /* GH: Okay, keep checking as normal.  Still test for GL_RGB,
-       * GL_RGBA formats first.
-       */
    case GL_RGBA8:
    case GL_RGB10_A2:
    case GL_RGBA12:
    case GL_RGBA16:
-      SET_FORMAT_32BPT( TMC_tformat_tw32, _mesa_texformat_argb8888,
-			TMC_tformat_tw12, _mesa_texformat_argb4444 );
-      break;
+      return do32bpt ? &_mesa_texformat_argb8888 : &_mesa_texformat_argb4444;
 
    case GL_RGBA4:
    case GL_RGBA2:
-      SET_FORMAT( TMC_tformat_tw12, _mesa_texformat_argb4444 );
-      break;
+      return &_mesa_texformat_argb4444;
 
    case GL_RGB5_A1:
-      SET_FORMAT( TMC_tformat_tw15, _mesa_texformat_argb1555 );
-      break;
+      return &_mesa_texformat_argb1555;
 
    case GL_RGB8:
    case GL_RGB10:
    case GL_RGB12:
    case GL_RGB16:
-      SET_FORMAT_32BPT( TMC_tformat_tw32, _mesa_texformat_argb8888,
-			TMC_tformat_tw16, _mesa_texformat_rgb565 );
-      break;
+      return do32bpt ? &_mesa_texformat_argb8888 : &_mesa_texformat_rgb565;
 
    case GL_RGB5:
    case GL_RGB4:
    case GL_R3_G3_B2:
-      SET_FORMAT( TMC_tformat_tw16, _mesa_texformat_rgb565 );
-      break;
+      return &_mesa_texformat_rgb565;
 
    case GL_ALPHA:
    case GL_ALPHA4:
@@ -261,8 +221,7 @@ static GLint mgaChooseTexFormat( mgaContextPtr mmesa,
    case GL_ALPHA16:
    case GL_COMPRESSED_ALPHA:
       /* FIXME: This will report incorrect component sizes... */
-      SET_FORMAT( TMC_tformat_tw12, _mesa_texformat_argb4444 );
-      break;
+      return &_mesa_texformat_argb4444;
 
    case 1:
    case GL_LUMINANCE:
@@ -272,8 +231,7 @@ static GLint mgaChooseTexFormat( mgaContextPtr mmesa,
    case GL_LUMINANCE16:
    case GL_COMPRESSED_LUMINANCE:
       /* FIXME: This will report incorrect component sizes... */
-      SET_FORMAT( TMC_tformat_tw16, _mesa_texformat_rgb565 );
-      break;
+      return &_mesa_texformat_rgb565;
 
    case 2:
    case GL_LUMINANCE_ALPHA:
@@ -285,8 +243,7 @@ static GLint mgaChooseTexFormat( mgaContextPtr mmesa,
    case GL_LUMINANCE16_ALPHA16:
    case GL_COMPRESSED_LUMINANCE_ALPHA:
       /* FIXME: This will report incorrect component sizes... */
-      SET_FORMAT( TMC_tformat_tw12, _mesa_texformat_argb4444 );
-      break;
+      return &_mesa_texformat_argb4444;
 
    case GL_INTENSITY:
    case GL_INTENSITY4:
@@ -295,8 +252,15 @@ static GLint mgaChooseTexFormat( mgaContextPtr mmesa,
    case GL_INTENSITY16:
    case GL_COMPRESSED_INTENSITY:
       /* FIXME: This will report incorrect component sizes... */
-      SET_FORMAT( TMC_tformat_tw12, _mesa_texformat_argb4444 );
-      break;
+      return &_mesa_texformat_argb4444;
+
+   case GL_YCBCR_MESA:
+      if (MGA_IS_G400(mmesa) &&
+          (type == GL_UNSIGNED_SHORT_8_8_APPLE ||
+           type == GL_UNSIGNED_BYTE))
+         return &_mesa_texformat_ycbcr;
+      else
+         return &_mesa_texformat_ycbcr_rev;
 
    case GL_COLOR_INDEX:
    case GL_COLOR_INDEX1_EXT:
@@ -305,502 +269,78 @@ static GLint mgaChooseTexFormat( mgaContextPtr mmesa,
    case GL_COLOR_INDEX8_EXT:
    case GL_COLOR_INDEX12_EXT:
    case GL_COLOR_INDEX16_EXT:
-      SET_FORMAT( TMC_tformat_tw8, _mesa_texformat_ci8 );
-      break;
+      return &_mesa_texformat_ci8;
 
    default:
-      fprintf( stderr, "bad texture format in mgaChooseTexFormat() %d",
-	       texImage->IntFormat );
-      return -1;
+      _mesa_problem( ctx, "unexpected texture format in %s", __FUNCTION__ );
+      return NULL;
    }
 
-   texImage->TexFormat = texFormat;
-
-   return ret;
+   return NULL; /* never get here */
 }
 
 
-/*
- * mgaCreateTexObj
+
+
+/**
  * Allocate space for and load the mesa images into the texture memory block.
  * This will happen before drawing with a new texture, or drawing with a
  * texture after it was swapped out or teximaged again.
  */
-static void mgaCreateTexObj(mgaContextPtr mmesa,
-			    struct gl_texture_object *tObj)
-{
-   const GLint baseLevel = tObj->BaseLevel;
-   struct gl_texture_image *image = tObj->Image[baseLevel];
-   mgaTextureObjectPtr t;
-   int i, ofs;
-   int LastLevel;
-   int s, s2;
-   int tformat;
-
-   if (!image) return;
-
-   tObj->DriverData = t = calloc( 1, sizeof( *t ) );
-   if (!t) {
-      fprintf(stderr, "mgaCreateTexObj: Failed to malloc mgaTextureObject\n" );
-      return;
-   }
-
-   /* FIXME: Use the real DD interface...
-    */
-   tformat = mgaChooseTexFormat( mmesa, image, image->Format,
-				 GL_UNSIGNED_BYTE );
-   t->texelBytes = image->TexFormat->TexelBytes;
-
-   /* We are going to upload all levels that are present, even if
-    * later levels wouldn't be used by the current filtering mode.  This
-    * allows the filtering mode to change without forcing another upload
-    * of the images.
-    */
-   LastLevel = MGA_TEX_MAXLEVELS-1;
-
-   ofs = 0;
-   for ( i = 0 ; i <= LastLevel ; i++ ) {
-      if ( !tObj->Image[i] ) {
-	 LastLevel = i - 1;
-	 break;
-      }
-
-      t->offsets[i] = ofs;
-      t->dirty_images |= (1<<i);
-
-      ofs += ((MAX2( tObj->Image[i]->Width, 8 ) *
-	       MAX2( tObj->Image[i]->Height, 8 ) *
-	       t->texelBytes) + 31) & ~31;
-   }
-   t->totalSize = ofs;
-   t->lastLevel = LastLevel;
-   t->tObj = tObj;
-   t->ctx = mmesa;
-   t->age = 0;
-   t->bound = 0;
-   t->MemBlock = 0;
-
-   insert_at_tail(&(mmesa->SwappedOut), t);
-
-
-   /* setup hardware register values */
-   t->setup.texctl = TMC_takey_1 | TMC_tamask_0 | tformat;
-
-   if (image->WidthLog2 >= 3)
-      t->setup.texctl |= ((image->WidthLog2 - 3) << TMC_tpitch_SHIFT);
-   else
-      t->setup.texctl |= (TMC_tpitchlin_enable |
-			  (image->Width << TMC_tpitchext_SHIFT));
-
-
-   t->setup.texctl2 = TMC_ckstransdis_enable;
-
-   if ( mmesa->glCtx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR )
-      t->setup.texctl2 |= TMC_specen_enable;
-
-
-   t->setup.texfilter = (TF_minfilter_nrst |
-			 TF_magfilter_nrst |
-			 TF_filteralpha_enable |
-			 (0x10 << TF_fthres_SHIFT) |
-			 (LastLevel << TF_mapnb_SHIFT));
-
-   /* warp texture registers */
-   ofs = MGA_IS_G200(mmesa) ? 28 : 11;
-   s = image->Width;
-   s2 = image->WidthLog2;
-   t->setup.texwidth = (MGA_FIELD(TW_twmask, s - 1) |
-			MGA_FIELD(TW_rfw, (10 - s2 - 8) & 63 ) |
-			MGA_FIELD(TW_tw, (s2 + ofs ) | 0x40 ));
-
 
-   s = image->Height;
-   s2 = image->HeightLog2;
-   t->setup.texheight = (MGA_FIELD(TH_thmask, s - 1) |
-			 MGA_FIELD(TH_rfh, (10 - s2 - 8) & 63 ) |
-			 MGA_FIELD(TH_th, (s2 + ofs ) | 0x40 ));
-
-
-   /* set all the register values for filtering, border, etc */
-   mgaSetTexWrapping( t, tObj->WrapS, tObj->WrapT );
-   mgaSetTexFilter( t, tObj->MinFilter, tObj->MagFilter );
-   mgaSetTexBorderColor( t, tObj->_BorderChan );
-}
-
-
-
-
-static void mgaUpdateTextureEnvG200( GLcontext *ctx )
+static mgaTextureObjectPtr
+mgaAllocTexObj( struct gl_texture_object *tObj )
 {
-   struct gl_texture_object *tObj = ctx->Texture.Unit[0]._Current;
    mgaTextureObjectPtr t;
 
-   if (!tObj || !tObj->DriverData)
-      return;
-
-   t = (mgaTextureObjectPtr)tObj->DriverData;
-
-   t->setup.texctl2 &= ~TMC_decalblend_enable;
-
-   switch (ctx->Texture.Unit[0].EnvMode) {
-   case GL_REPLACE:
-      t->setup.texctl &= ~TMC_tmodulate_enable;
-      break;
-   case GL_MODULATE:
-      t->setup.texctl |= TMC_tmodulate_enable;
-      break;
-   case GL_DECAL:
-      t->setup.texctl &= ~TMC_tmodulate_enable;
-      t->setup.texctl2 |= TMC_decalblend_enable;
-      break;
-   case GL_BLEND:
-      FALLBACK( ctx, MGA_FALLBACK_TEXTURE, GL_TRUE );
-      break;
-   default:
-      break;
-   }
-}
-
-static void mgaUpdateTextureEnvG400( GLcontext *ctx, int unit )
-{
-   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-   GLuint *reg = ((GLuint *)&mmesa->setup.tdualstage0 + unit);
-   GLuint source = mmesa->tmu_source[unit];
-   struct gl_texture_object *tObj = ctx->Texture.Unit[source]._Current;
-   GLenum format;
-
-   if ( tObj != ctx->Texture.Unit[source].Current2D || !tObj ) 
-      return;
-
-   format = tObj->Image[tObj->BaseLevel]->Format;
-
-   switch (ctx->Texture.Unit[source].EnvMode) {
-   case GL_REPLACE:
-      if (format == GL_RGB || format == GL_LUMINANCE) {
-	 *reg = (TD0_color_sel_arg1 |
-                 TD0_alpha_arg2_diffuse |
-		 TD0_alpha_sel_arg2 );
-      }
-      else if (format == GL_ALPHA) {
-         *reg = (TD0_color_sel_arg2 |
-                 TD0_color_arg2_diffuse |
-                 TD0_alpha_sel_arg1 );
-      }
-      else {
-         *reg = (TD0_color_sel_arg1 |
-                 TD0_alpha_sel_arg1 );
-      }
-      break;
-
-   case GL_MODULATE:
-      if (unit == 0) {
-	 *reg = ( TD0_color_arg2_diffuse |
-		  TD0_color_sel_mul |
-		  TD0_alpha_arg2_diffuse |
-		  TD0_alpha_sel_mul);
-      }
-      else {
-	 *reg = ( TD0_color_arg2_prevstage |
-		  TD0_color_alpha_prevstage |
-		  TD0_color_sel_mul |
-		  TD0_alpha_arg2_prevstage |
-		  TD0_alpha_sel_mul);
-      }
-      break;
-   case GL_DECAL:
-      if (format == GL_RGB) {
-         if (unit == 0) {
-            *reg = (TD0_color_sel_arg1 |
-                    TD0_alpha_arg2_diffuse |
-                    TD0_alpha_sel_arg2 );
-         }
-         else {
-            *reg = (TD0_color_sel_arg1 |
-                    TD0_alpha_arg2_prevstage |
-                    TD0_alpha_sel_arg2 );
-         }
-      }
-      else if ( format == GL_RGBA ) {
-#if 0
-         if (unit == 0) {
-            /* this doesn't work */
-            *reg = (TD0_color_arg2_diffuse |
-                    TD0_color_alpha_currtex |
-                    TD0_color_alpha2inv_enable |
-                    TD0_color_arg2mul_alpha2 |
-                    TD0_color_arg1mul_alpha1 |
-                    TD0_color_blend_enable |
-                    TD0_color_arg1add_mulout |
-                    TD0_color_arg2add_mulout |
-                    TD0_color_add_add |
-                    TD0_color_sel_mul |
-                    TD0_alpha_arg2_diffuse |
-                    TD0_alpha_sel_arg2 );
-         }
-         else {
-            *reg = (TD0_color_arg2_prevstage |
-                    TD0_color_alpha_currtex |
-                    TD0_color_alpha2inv_enable |
-                    TD0_color_arg2mul_alpha2 |
-                    TD0_color_arg1mul_alpha1 |
-                    TD0_color_add_add |
-                    TD0_color_sel_add |
-                    TD0_alpha_arg2_prevstage |
-                    TD0_alpha_sel_arg2 );
-         }
-#else
-         /* s/w fallback, pretty sure we can't do in h/w */
-	 FALLBACK( ctx, MGA_FALLBACK_TEXTURE, GL_TRUE );
-	 if ( MGA_DEBUG & DEBUG_VERBOSE_FALLBACK )
-	    fprintf( stderr, "FALLBACK: GL_DECAL RGBA texture, unit=%d\n",
-		     unit );
-#endif
-      }
-      else {
-	 if (unit == 0) {
-	    *reg = ( TD0_color_arg2_diffuse |
-		     TD0_color_sel_arg2 |
-		     TD0_alpha_arg2_diffuse |
-		     TD0_alpha_sel_arg2);
-	 }
-	 else {
-	    *reg = ( TD0_color_arg2_prevstage |
-		     TD0_color_sel_arg2 |
-		     TD0_alpha_arg2_prevstage |
-		     TD0_alpha_sel_arg2);
-	 }
-      }
-      break;
-
-   case GL_ADD:
-      if (unit == 0) {
-         if (format == GL_INTENSITY)
-            *reg = ( TD0_color_arg2_diffuse |
-                     TD0_color_add_add |
-                     TD0_color_sel_add |
-                     TD0_alpha_arg2_diffuse |
-                     TD0_alpha_add_enable |
-                     TD0_alpha_sel_add);
-         else if (format == GL_ALPHA)
-            *reg = ( TD0_color_arg2_diffuse |
-                     TD0_color_sel_mul |
-                     TD0_alpha_arg2_diffuse |
-                     TD0_alpha_sel_mul);
-         else
-            *reg = ( TD0_color_arg2_diffuse |
-                     TD0_color_add_add |
-                     TD0_color_sel_add |
-                     TD0_alpha_arg2_diffuse |
-                     TD0_alpha_sel_mul);
-      }
-      else {
-         if (format == GL_INTENSITY) {
-            *reg = ( TD0_color_arg2_prevstage |
-                     TD0_color_add_add |
-                     TD0_color_sel_add |
-                     TD0_alpha_arg2_prevstage |
-                     TD0_alpha_add_enable |
-                     TD0_alpha_sel_add);
-         }
-         else if (format == GL_ALPHA) {
-            *reg = ( TD0_color_arg2_prevstage |
-                     TD0_color_sel_mul |
-                     TD0_alpha_arg2_prevstage |
-                     TD0_alpha_sel_mul);
-         }
-         else {
-            *reg = ( TD0_color_arg2_prevstage |
-                     TD0_color_alpha_prevstage |
-                     TD0_color_add_add |
-                     TD0_color_sel_add |
-                     TD0_alpha_arg2_prevstage |
-                     TD0_alpha_sel_mul);
-         }
-      }
-      break;
-
-   case GL_BLEND:
-      if (format == GL_ALPHA) {
-	 *reg = ( TD0_color_arg2_diffuse |
-		  TD0_color_sel_mul |
-		  TD0_alpha_arg2_diffuse |
-		  TD0_alpha_sel_mul);
-      }
-      else {
-	 FALLBACK( ctx, MGA_FALLBACK_TEXTURE, GL_TRUE );
-	 if ( MGA_DEBUG & DEBUG_VERBOSE_FALLBACK )
-	    fprintf( stderr, "FALLBACK: GL_BLEND envcolor=0x%08x\n",
-		     mmesa->envcolor );
-
-         /* Do singletexture GL_BLEND with 'all ones' env-color
-          * by using both texture units.  Multitexture gl_blend
-          * is a fallback.
-          */
-         if (unit == 0) {
-            /* Part 1: R1 = Rf ( 1 - Rt )
-             *         A1 = Af At
-             */
-            *reg = ( TD0_color_arg2_diffuse |
-                     TD0_color_arg1_inv_enable |
-                     TD0_color_sel_mul |
-                     TD0_alpha_arg2_diffuse |
-                     TD0_alpha_sel_arg1);
-         } else {
-            /* Part 2: R2 = R1 + Rt
-             *         A2 = A1
-             */
-            *reg = ( TD0_color_arg2_prevstage |
-                     TD0_color_add_add |
-                     TD0_color_sel_add |
-                     TD0_alpha_arg2_prevstage |
-                     TD0_alpha_sel_arg2);
-         }
-      }
-      break;
-   default:
-      break;
-   }
-}
-
-
-
-static void mgaUpdateTextureObject( GLcontext *ctx, int hw_unit )
-{
-   mgaTextureObjectPtr t;
-   struct gl_texture_object	*tObj;
-   GLuint enabled;
-   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-   GLuint gl_unit = mmesa->tmu_source[hw_unit];
-
-
-   enabled = ctx->Texture.Unit[gl_unit]._ReallyEnabled;
-   tObj = ctx->Texture.Unit[gl_unit]._Current;
-
-   if (enabled != TEXTURE_2D_BIT) {
-      if (enabled)
-	 FALLBACK( ctx, MGA_FALLBACK_TEXTURE, GL_TRUE );
-      return;
-   }
-
-   if (tObj->Image[tObj->BaseLevel]->Border > 0) {
-      FALLBACK( ctx, MGA_FALLBACK_TEXTURE, GL_TRUE );
-      if ( MGA_DEBUG & DEBUG_VERBOSE_FALLBACK )
-	 fprintf( stderr, "FALLBACK: texture border\n" );
-      return;
-   }
-
-   if ( !tObj->DriverData ) {
-      mgaCreateTexObj( mmesa, tObj );
-      if ( !tObj->DriverData ) {
-	 FALLBACK( ctx, MGA_FALLBACK_TEXTURE, GL_TRUE );
-	 return;		
-      }
-   }
-
-   t = (mgaTextureObjectPtr)tObj->DriverData;
-
-   if (t->dirty_images)
-      mmesa->dirty |= (MGA_UPLOAD_TEX0IMAGE << hw_unit);
-
-   mmesa->CurrentTexObj[hw_unit] = t;
-   t->bound |= hw_unit+1;
-
-/*     if (t->MemBlock) */
-/*        mgaUpdateTexLRU( mmesa, t ); */
-
-   t->setup.texctl2 &= ~TMC_dualtex_enable;
-   if ((ctx->Texture.Unit[0]._ReallyEnabled == TEXTURE_2D_BIT) &&
-       (ctx->Texture.Unit[1]._ReallyEnabled == TEXTURE_2D_BIT)) 
-      t->setup.texctl2 |= TMC_dualtex_enable;
-
-   t->setup.texctl2 &= ~TMC_specen_enable;
-   if (ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR)
-      t->setup.texctl2 |= TMC_specen_enable;
-}
-
-
 
+   t = CALLOC( sizeof( *t ) );
+   tObj->DriverData = t;
+   if ( t != NULL ) {
+      /* Initialize non-image-dependent parts of the state:
+       */
+      t->base.tObj = tObj;
 
+      t->setup.texctl = TMC_takey_1 | TMC_tamask_0;
+      t->setup.texctl2 = TMC_ckstransdis_enable;
+      t->setup.texfilter = (TF_minfilter_nrst 
+			    | TF_magfilter_nrst
+			    | TF_filteralpha_enable
+			    | TF_uvoffset_OGL);
 
+      t->border_fallback = GL_FALSE;
 
-/* The G400 is now programmed quite differently wrt texture environment.
- */
-void mgaUpdateTextureState( GLcontext *ctx )
-{
-   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-   FALLBACK( ctx, MGA_FALLBACK_TEXTURE, GL_FALSE );
+      make_empty_list( & t->base );
 
-   if (mmesa->CurrentTexObj[0]) {
-      mmesa->CurrentTexObj[0]->bound = 0;
-      mmesa->CurrentTexObj[0] = 0;
-   }
-
-   if (mmesa->CurrentTexObj[1]) {
-      mmesa->CurrentTexObj[1]->bound = 0;
-      mmesa->CurrentTexObj[1] = 0;
-   }
-
-   if (ctx->Texture.Unit[1]._ReallyEnabled == TEXTURE_2D_BIT) {
-      mmesa->tmu_source[0] = 1;
-   } else {
-      mmesa->tmu_source[0] = 0;
-   }
-
-   if (MGA_IS_G400(mmesa)) {
-      mgaUpdateTextureObject( ctx, 0 );
-      mgaUpdateTextureEnvG400( ctx, 0 );
-
-      mmesa->setup.tdualstage1 = mmesa->setup.tdualstage0;
-      
-      if ((ctx->Texture.Unit[0]._ReallyEnabled == TEXTURE_2D_BIT) &&
-          (ctx->Texture.Unit[1]._ReallyEnabled == TEXTURE_2D_BIT)) {
-	 mgaUpdateTextureObject( ctx, 1 );	
-	 mgaUpdateTextureEnvG400( ctx, 1 );
-	 mmesa->dirty |= MGA_UPLOAD_TEX1;
-      }
-   } else {
-      mgaUpdateTextureObject( ctx, 0 );
-      mgaUpdateTextureEnvG200( ctx );
+      mgaSetTexWrapping( t, tObj->WrapS, tObj->WrapT );
+      mgaSetTexFilter( t, tObj->MinFilter, tObj->MagFilter );
+      mgaSetTexBorderColor( t, tObj->_BorderChan );
    }
 
-   mmesa->dirty |= MGA_UPLOAD_CONTEXT | MGA_UPLOAD_TEX0;
-
-   mmesa->setup.dwgctl &= DC_opcod_MASK;
-   mmesa->setup.dwgctl |= (ctx->Texture.Unit[0]._ReallyEnabled
-			   ? DC_opcod_texture_trap
-			   : DC_opcod_trap);
+   return( t );
 }
 
 
-
-
 static void mgaDDTexEnv( GLcontext *ctx, GLenum target,
 			 GLenum pname, const GLfloat *param )
 {
+   GLuint unit = ctx->Texture.CurrentUnit;
+   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
    mgaContextPtr mmesa = MGA_CONTEXT(ctx);
 
 
-   if (pname == GL_TEXTURE_ENV_MODE) {
-      /* force the texture state to be updated */
-      FLUSH_BATCH( MGA_CONTEXT(ctx) );
-      MGA_CONTEXT(ctx)->new_state |= (MGA_NEW_TEXTURE |
-				      MGA_NEW_ALPHA);
-   }
-   else if (pname == GL_TEXTURE_ENV_COLOR)
-   {
-      struct gl_texture_unit *texUnit =
-	 &ctx->Texture.Unit[ctx->Texture.CurrentUnit];
-      GLfloat *fc = texUnit->EnvColor;
+   switch( pname ) {
+   case GL_TEXTURE_ENV_COLOR: {
       GLubyte c[4];
-      GLuint col;
+      GLuint envColor;
 
-      COPY_4V(c, fc);
-      col = mgaPackColor( mmesa->mgaScreen->cpp, c[0], c[1], c[2], c[3] );
-      mmesa->envcolor = (c[3]<<24) | (c[0]<<16) | (c[1]<<8) | (c[2]);
+      UNCLAMPED_FLOAT_TO_RGBA_CHAN( c, texUnit->EnvColor );
+      envColor = mgaPackColor( mmesa->mgaScreen->cpp, c[0], c[1], c[2], c[3] );
+      mmesa->envcolor = PACK_COLOR_8888( c[3], c[0], c[1], c[2] );
 
-      if (mmesa->setup.fcol != col) {
+      if (mmesa->setup.fcol != envColor) {
 	 FLUSH_BATCH(mmesa);
-	 mmesa->setup.fcol = col;
+	 mmesa->setup.fcol = envColor;
 	 mmesa->dirty |= MGA_UPLOAD_CONTEXT;
 
 	 mmesa->blend_flags &= ~MGA_BLEND_ENV_COLOR;
@@ -814,6 +354,8 @@ static void mgaDDTexEnv( GLcontext *ctx, GLenum target,
 	 if (mmesa->envcolor != 0x0 && mmesa->envcolor != 0xffffffff)
 	    mmesa->blend_flags |= MGA_BLEND_ENV_COLOR;
       }
+      break;
+   }
    }
 }
 
@@ -826,14 +368,26 @@ static void mgaTexImage2D( GLcontext *ctx, GLenum target, GLint level,
 			    struct gl_texture_object *texObj,
 			    struct gl_texture_image *texImage )
 {
-   mgaTextureObjectPtr t = (mgaTextureObjectPtr) texObj->DriverData;
-   if (t) {
-      mgaDestroyTexObj( MGA_CONTEXT(ctx), t );
-      texObj->DriverData = 0;
+   driTextureObject * t = (driTextureObject *) texObj->DriverData;
+
+
+   if ( t != NULL ) {
+      driSwapOutTextureObject( t );
+   } 
+   else {
+      t = (driTextureObject *) mgaAllocTexObj( texObj );
+      if ( t == NULL ) {
+	 _mesa_error( ctx, GL_OUT_OF_MEMORY, "glTexImage2D" );
+	 return;
+      }
    }
+
    _mesa_store_teximage2d( ctx, target, level, internalFormat,
 			   width, height, border, format, type,
 			   pixels, packing, texObj, texImage );
+   level -= t->firstLevel;
+   if (level >= 0)
+      t->dirty_images[0] |= (1UL << level);
 }
 
 static void mgaTexSubImage2D( GLcontext *ctx, 
@@ -847,44 +401,59 @@ static void mgaTexSubImage2D( GLcontext *ctx,
 			       struct gl_texture_object *texObj,
 			       struct gl_texture_image *texImage )
 {
-   mgaTextureObjectPtr t = (mgaTextureObjectPtr) texObj->DriverData;
-   if (t) {
-      mgaDestroyTexObj( MGA_CONTEXT(ctx), t );
-      texObj->DriverData = 0;
+   driTextureObject * t = (driTextureObject *) texObj->DriverData;
+
+
+   assert( t != NULL ); /* this _should_ be true */
+   if ( t != NULL ) {
+      driSwapOutTextureObject( t );
+   } 
+   else {
+      t = (driTextureObject *) mgaAllocTexObj( texObj );
+      if ( t == NULL ) {
+	 _mesa_error( ctx, GL_OUT_OF_MEMORY, "glTexImage2D" );
+	 return;
+      }
    }
+
    _mesa_store_texsubimage2d(ctx, target, level, xoffset, yoffset, width, 
 			     height, format, type, pixels, packing, texObj,
 			     texImage);
-
+   level -= t->firstLevel;
+   if (level >= 0)
+      t->dirty_images[0] |= (1UL << level);
 }
 
 
-
-
-/*
- * mgaTexParameter
- * This just changes variables and flags for a state update, which
- * will happen at the next mgaUpdateTextureState
+/**
+ * Changes variables and flags for a state update, which will happen at the
+ * next UpdateTextureState
  */
+
 static void
 mgaDDTexParameter( GLcontext *ctx, GLenum target,
 		   struct gl_texture_object *tObj,
 		   GLenum pname, const GLfloat *params )
 {
-   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
+   mgaContextPtr       mmesa = MGA_CONTEXT( ctx );
    mgaTextureObjectPtr t;
 
    t = (mgaTextureObjectPtr) tObj->DriverData;
 
-   /* if we don't have a hardware texture, it will be automatically
-      created with current state before it is used, so we don't have
-      to do anything now */
-   if ( !t || !t->bound || target != GL_TEXTURE_2D ) {
+   /* If we don't have a hardware texture, it will be automatically
+    * created with current state before it is used, so we don't have
+    * to do anything now 
+    */
+
+   if ( (t == NULL)
+        || (target != GL_TEXTURE_2D) ) {
       return;
    }
 
    switch (pname) {
    case GL_TEXTURE_MIN_FILTER:
+      driSwapOutTextureObject( (driTextureObject *) t );
+      /* FALLTHROUGH */
    case GL_TEXTURE_MAG_FILTER:
       FLUSH_BATCH(mmesa);
       mgaSetTexFilter( t, tObj->MinFilter, tObj->MagFilter );
@@ -898,14 +467,24 @@ mgaDDTexParameter( GLcontext *ctx, GLenum target,
 
    case GL_TEXTURE_BORDER_COLOR:
       FLUSH_BATCH(mmesa);
-      mgaSetTexBorderColor(t,tObj->_BorderChan);
+      mgaSetTexBorderColor(t, tObj->_BorderChan);
+      break;
+
+   case GL_TEXTURE_BASE_LEVEL:
+   case GL_TEXTURE_MAX_LEVEL:
+   case GL_TEXTURE_MIN_LOD:
+   case GL_TEXTURE_MAX_LOD:
+      /* This isn't the most efficient solution but there doesn't appear to
+       * be a nice alternative.  Since there's no LOD clamping,
+       * we just have to rely on loading the right subset of mipmap levels
+       * to simulate a clamped LOD.
+       */
+      driSwapOutTextureObject( (driTextureObject *) t );
       break;
 
    default:
       return;
    }
-
-   mmesa->new_state |= MGA_NEW_TEXTURE;
 }
 
 
@@ -913,19 +492,11 @@ static void
 mgaDDBindTexture( GLcontext *ctx, GLenum target,
 		  struct gl_texture_object *tObj )
 {
-   mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-   int unit = ctx->Texture.CurrentUnit;
-
-   FLUSH_BATCH(mmesa);
-
-   if (mmesa->CurrentTexObj[unit]) {
-      mmesa->CurrentTexObj[unit]->bound &= ~(unit+1);
-      mmesa->CurrentTexObj[unit] = 0;
+   if ( target == GL_TEXTURE_2D ) {
+      if ( tObj->DriverData == NULL ) {
+	 mgaAllocTexObj( tObj );
+      }
    }
-
-   /* force the texture state to be updated
-    */
-   MGA_CONTEXT(ctx)->new_state |= MGA_NEW_TEXTURE;
 }
 
 
@@ -933,53 +504,48 @@ static void
 mgaDDDeleteTexture( GLcontext *ctx, struct gl_texture_object *tObj )
 {
    mgaContextPtr mmesa = MGA_CONTEXT( ctx );
-   mgaTextureObjectPtr t = (mgaTextureObjectPtr)tObj->DriverData;
+   driTextureObject * t = (driTextureObject *) tObj->DriverData;
 
    if ( t ) {
-      if (mmesa) {
-         if (t->bound) {
-            FLUSH_BATCH(mmesa);
-            if (t->bound & TEX_0) mmesa->CurrentTexObj[0] = 0;
-            if (t->bound & TEX_1) mmesa->CurrentTexObj[1] = 0;
-         }
-         mmesa->new_state |= MGA_NEW_TEXTURE;
+      if ( mmesa ) {
+	 FLUSH_BATCH( mmesa );
       }
 
-      mgaDestroyTexObj( mmesa, t );
+      driDestroyTextureObject( t );
    }
 }
 
 
-static GLboolean
-mgaDDIsTextureResident( GLcontext *ctx, struct gl_texture_object *t )
-{
-   mgaTextureObjectPtr mt = (mgaTextureObjectPtr)t->DriverData;
-   return mt && mt->MemBlock;
-}
-
-
 void
 mgaDDInitTextureFuncs( GLcontext *ctx )
 {
-   ctx->Driver.TexEnv = mgaDDTexEnv;
-
-   ctx->Driver.ChooseTextureFormat = _mesa_choose_tex_format;
-   ctx->Driver.TexImage1D = _mesa_store_teximage1d;
-   ctx->Driver.TexImage2D = mgaTexImage2D;
-   ctx->Driver.TexImage3D = _mesa_store_teximage3d;
-   ctx->Driver.TexSubImage1D = _mesa_store_texsubimage1d;
-   ctx->Driver.TexSubImage2D = mgaTexSubImage2D;
-   ctx->Driver.TexSubImage3D = _mesa_store_texsubimage3d;
-   ctx->Driver.CopyTexImage1D = _swrast_copy_teximage1d;
-   ctx->Driver.CopyTexImage2D = _swrast_copy_teximage2d;
-   ctx->Driver.CopyTexSubImage1D = _swrast_copy_texsubimage1d;
-   ctx->Driver.CopyTexSubImage2D = _swrast_copy_texsubimage2d;
-   ctx->Driver.CopyTexSubImage3D = _swrast_copy_texsubimage3d;
-   ctx->Driver.TestProxyTexImage = _mesa_test_proxy_teximage;
-
-   ctx->Driver.BindTexture = mgaDDBindTexture;
-   ctx->Driver.DeleteTexture = mgaDDDeleteTexture;
-   ctx->Driver.TexParameter = mgaDDTexParameter;
-   ctx->Driver.UpdateTexturePalette = 0;
-   ctx->Driver.IsTextureResident = mgaDDIsTextureResident;
+   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
+
+
+   ctx->Driver.ChooseTextureFormat	= mgaChooseTextureFormat;
+   ctx->Driver.TexImage1D		= _mesa_store_teximage1d;
+   ctx->Driver.TexImage2D		= mgaTexImage2D;
+   ctx->Driver.TexImage3D		= _mesa_store_teximage3d;
+   ctx->Driver.TexSubImage1D		= _mesa_store_texsubimage1d;
+   ctx->Driver.TexSubImage2D		= mgaTexSubImage2D;
+   ctx->Driver.TexSubImage3D		= _mesa_store_texsubimage3d;
+   ctx->Driver.CopyTexImage1D		= _swrast_copy_teximage1d;
+   ctx->Driver.CopyTexImage2D		= _swrast_copy_teximage2d;
+   ctx->Driver.CopyTexSubImage1D	= _swrast_copy_texsubimage1d;
+   ctx->Driver.CopyTexSubImage2D	= _swrast_copy_texsubimage2d;
+   ctx->Driver.CopyTexSubImage3D	= _swrast_copy_texsubimage3d;
+   ctx->Driver.TestProxyTexImage	= _mesa_test_proxy_teximage;
+
+   ctx->Driver.BindTexture		= mgaDDBindTexture;
+   ctx->Driver.CreateTexture		= NULL; /* FIXME: Is this used??? */
+   ctx->Driver.DeleteTexture		= mgaDDDeleteTexture;
+   ctx->Driver.IsTextureResident	= driIsTextureResident;
+   ctx->Driver.PrioritizeTexture	= NULL;
+   ctx->Driver.ActiveTexture		= NULL;
+   ctx->Driver.UpdateTexturePalette	= NULL;
+
+   ctx->Driver.TexEnv			= mgaDDTexEnv;
+   ctx->Driver.TexParameter		= mgaDDTexParameter;
+
+   driInitTextureObjects( ctx, & mmesa->swapped, DRI_TEXMGR_DO_TEXTURE_2D );
 }
diff --git a/src/mesa/drivers/dri/mga/mgatex.h b/src/mesa/drivers/dri/mga/mgatex.h
index c9f87d997e2..282577e9f80 100644
--- a/src/mesa/drivers/dri/mga/mgatex.h
+++ b/src/mesa/drivers/dri/mga/mgatex.h
@@ -40,23 +40,10 @@ typedef struct mga_texture_object_s *mgaTextureObjectPtr;
  */
 void mgaUpdateTextureState( GLcontext *ctx );
 
-void mgaConvertTexture( GLuint *dest, int texelBytes,
-			struct gl_texture_image *image,
-			int x, int y, int width, int height );
-
-
-void mgaUploadSubImageLocked( mgaContextPtr mmesa,
-			      mgaTextureObjectPtr t,
-			      int level,
-			      int x, int y, int width, int height );
-
 int mgaUploadTexImages( mgaContextPtr mmesa, mgaTextureObjectPtr t );
 
 void mgaDestroyTexObj( mgaContextPtr mmesa, mgaTextureObjectPtr t );
 
-void mgaAgeTextures( mgaContextPtr mmesa, int heap );
-
 void mgaDDInitTextureFuncs( GLcontext *ctx );
 
-
 #endif
diff --git a/src/mesa/drivers/dri/mga/mgatexmem.c b/src/mesa/drivers/dri/mga/mgatexmem.c
index 7dbdbc582fa..4e7fd6b24b3 100644
--- a/src/mesa/drivers/dri/mga/mgatexmem.c
+++ b/src/mesa/drivers/dri/mga/mgatexmem.c
@@ -26,539 +26,257 @@
  */
 /* $XFree86: xc/lib/GL/mesa/src/drv/mga/mgatexmem.c,v 1.7 2002/10/30 12:51:36 alanh Exp $ */
 
-#include <stdlib.h>
-#include <stdio.h>
-#include <GL/gl.h>
+#include "glheader.h"
 
 #include "mm.h"
 #include "mgacontext.h"
 #include "mgatex.h"
 #include "mgaregs.h"
 #include "mgaioctl.h"
+#include "mga_xmesa.h"
 
-/*#include "mem.h" */
+#include "imports.h"
 #include "simple_list.h"
 
-static void
-mgaSwapOutTexObj(mgaContextPtr mmesa, mgaTextureObjectPtr t)
-{
-   if (t->MemBlock) {
-      mmFreeMem(t->MemBlock);
-      t->MemBlock = 0;
-
-      if (t->age > mmesa->dirtyAge)
-	 mmesa->dirtyAge = t->age;
-   }
-
-   t->dirty_images = ~0;
-   move_to_tail(&(mmesa->SwappedOut), t);
-}
-
-static void
-mgaPrintLocalLRU( mgaContextPtr mmesa, int heap )
-{
-   mgaTextureObjectPtr t;
-   int sz = 1 << (mmesa->mgaScreen->logTextureGranularity[heap]);
-
-   fprintf(stderr, "\nLocal LRU, heap %d:\n", heap);
-
-   foreach( t, &(mmesa->TexObjList[heap]) ) {
-      if (!t->tObj)
-	 fprintf(stderr, "Placeholder %d at %x sz %x\n",
-		 t->MemBlock->ofs / sz,
-		 t->MemBlock->ofs,
-		 t->MemBlock->size);
-      else
-	 fprintf(stderr, "Texture (bound %d) at %x sz %x\n",
-		 t->bound,
-		 t->MemBlock->ofs,
-		 t->MemBlock->size);
-   }
-
-   fprintf(stderr, "\n\n");
-}
-
-static void
-mgaPrintGlobalLRU( mgaContextPtr mmesa, int heap )
+/**
+ * Destroy any device-dependent state associated with the texture.  This may
+ * include NULLing out hardware state that points to the texture.
+ */
+void
+mgaDestroyTexObj( mgaContextPtr mmesa, mgaTextureObjectPtr t )
 {
-   int i, j;
-   drmTextureRegion *list = mmesa->sarea->texList[heap];
+    unsigned   i;
 
-   fprintf(stderr, "\nGlobal LRU, heap %d list %p:\n", heap, list);
 
-   for (i = 0, j = MGA_NR_TEX_REGIONS ; i < MGA_NR_TEX_REGIONS ; i++) {
-      fprintf(stderr, "list[%d] age %d next %d prev %d\n",
-	      j, list[j].age, list[j].next, list[j].prev);
-      j = list[j].next;
-      if (j == MGA_NR_TEX_REGIONS) break;
-   }
+    /* See if it was the driver's current object.
+     */
 
-   if (j != MGA_NR_TEX_REGIONS) {
-      fprintf(stderr, "Loop detected in global LRU\n\n\n");
-      for (i = 0 ; i < MGA_NR_TEX_REGIONS ; i++) {
-	 fprintf(stderr, "list[%d] age %d next %d prev %d\n",
-		 i, list[i].age, list[i].next, list[i].prev);
-      }
-   }
+    if ( mmesa != NULL )
+    { 
+	if ( t->age > mmesa->dirtyAge )
+	    mmesa->dirtyAge = t->age;
 
-   fprintf(stderr, "\n\n");
+	for ( i = 0 ; i < mmesa->glCtx->Const.MaxTextureUnits ; i++ )
+	{
+	    if ( t == mmesa->CurrentTexObj[ i ] ) {
+		mmesa->CurrentTexObj[ i ] = NULL;
+	    }
+	}
+    }
 }
 
 
-static void mgaResetGlobalLRU( mgaContextPtr mmesa, GLuint heap )
+/**
+ * Upload a texture image from system memory to either on-card or AGP
+ * memory.  Uploads to on-card memory are performed using an ILOAD operation.
+ * This is used for both initial loading of the entire image, and texSubImage
+ * updates.
+ *
+ * Performed with the hardware lock held.
+ * 
+ * Even though this function is named "upload subimage," the entire image
+ * is uploaded.
+ * 
+ * \param mmesa  Driver context.
+ * \param t      Texture to be uploaded.
+ * \param hwlevel  Mipmap level of the texture to be uploaded.
+ * 
+ * \bug As mentioned above, this fuction actually copies the entier mipmap
+ *      level.  There should be a version of this function that performs
+ *      sub-rectangle uploads.  This will perform quite a bit better if only
+ *      a small portion of a larger texture has been updated.  Care would
+ *      need to be take with such an implementation once glCopyTexImage has
+ *      been hardware accelerated.
+ */
+static void mgaUploadSubImage( mgaContextPtr mmesa,
+			       mgaTextureObjectPtr t, GLint hwlevel )
 {
-   drmTextureRegion *list = mmesa->sarea->texList[heap];
-   int sz = 1 << mmesa->mgaScreen->logTextureGranularity[heap];
-   int i;
-
-   mmesa->texAge[heap] = ++mmesa->sarea->texAge[heap];
+   struct gl_texture_image * texImage;
+   unsigned     offset;
+   unsigned     texelBytes;
+   unsigned     length;
+   const int level = hwlevel + t->base.firstLevel;
 
-   if (0) fprintf(stderr, "mgaResetGlobalLRU %d\n", (int)heap);
 
-   /* (Re)initialize the global circular LRU list.  The last element
-    * in the array (MGA_NR_TEX_REGIONS) is the sentinal.  Keeping it
-    * at the end of the array allows it to be addressed rationally
-    * when looking up objects at a particular location in texture
-    * memory.
-    */
-   for (i = 0 ; (i+1) * sz <= mmesa->mgaScreen->textureSize[heap] ; i++) {
-      list[i].prev = i-1;
-      list[i].next = i+1;
-      list[i].age = mmesa->sarea->texAge[heap];
+   if ( (hwlevel < 0) 
+	|| (hwlevel >= (MGA_IS_G200(mmesa) 
+		      ? G200_TEX_MAXLEVELS : G400_TEX_MAXLEVELS)) ) {
+      fprintf( stderr, "[%s:%d] level = %d\n", __FILE__, __LINE__, level );
+      return;
    }
 
-   i--;
-   list[0].prev = MGA_NR_TEX_REGIONS;
-   list[i].prev = i-1;
-   list[i].next = MGA_NR_TEX_REGIONS;
-   list[MGA_NR_TEX_REGIONS].prev = i;
-   list[MGA_NR_TEX_REGIONS].next = 0;
-
-}
-
-
-static void mgaUpdateTexLRU( mgaContextPtr mmesa, mgaTextureObjectPtr t )
-{
-   int i;
-   int heap = t->heap;
-   int logsz = mmesa->mgaScreen->logTextureGranularity[heap];
-   int start = t->MemBlock->ofs >> logsz;
-   int end = (t->MemBlock->ofs + t->MemBlock->size - 1) >> logsz;
-   drmTextureRegion *list = mmesa->sarea->texList[heap];
-
-   mmesa->texAge[heap] = ++mmesa->sarea->texAge[heap];
-
-   if (!t->MemBlock) {
-      fprintf(stderr, "no memblock\n\n");
+   texImage = t->base.tObj->Image[level];
+   if ( texImage == NULL ) {
+      fprintf( stderr, "[%s:%d] Image[%d] = NULL\n", __FILE__, __LINE__,
+	       level );
       return;
    }
 
-   /* Update our local LRU
-    */
-   move_to_head( &(mmesa->TexObjList[heap]), t );
-
-
-   if (0)
-      fprintf(stderr, "mgaUpdateTexLRU heap %d list %p\n", heap, list);
-
 
-   /* Update the global LRU
-    */
-   for (i = start ; i <= end ; i++) {
-
-      list[i].in_use = 1;
-      list[i].age = mmesa->texAge[heap];
-
-      /* remove_from_list(i)
-       */
-      list[(unsigned)list[i].next].prev = list[i].prev;
-      list[(unsigned)list[i].prev].next = list[i].next;
-
-      /* insert_at_head(list, i)
-       */
-      list[i].prev = MGA_NR_TEX_REGIONS;
-      list[i].next = list[MGA_NR_TEX_REGIONS].next;
-      list[(unsigned)list[MGA_NR_TEX_REGIONS].next].prev = i;
-      list[MGA_NR_TEX_REGIONS].next = i;
-   }
-
-   if (0) {
-      mgaPrintGlobalLRU(mmesa, t->heap);
-      mgaPrintLocalLRU(mmesa, t->heap);
+   if (texImage->Data == NULL) {
+      fprintf(stderr, "null texture image data tObj %p level %d\n",
+	      t->base.tObj, level);
+      return;
    }
-}
-
-/* Called for every shared texture region which has increased in age
- * since we last held the lock.
- *
- * Figures out which of our textures have been ejected by other clients,
- * and pushes a placeholder texture onto the LRU list to represent
- * the other client's textures.
- */
-static void mgaTexturesGone( mgaContextPtr mmesa,
-			     GLuint heap,
-			     GLuint offset,
-			     GLuint size,
-			     GLuint in_use )
-{
-   mgaTextureObjectPtr t, tmp;
-
-
 
-   foreach_s ( t, tmp, &(mmesa->TexObjList[heap]) ) {
 
-      if (t->MemBlock->ofs >= offset + size ||
-	  t->MemBlock->ofs + t->MemBlock->size <= offset)
-	 continue;
-
-
-
-
-      /* It overlaps - kick it off.  Need to hold onto the currently bound
-       * objects, however.
-       */
-      if (t->bound)
-	 mgaSwapOutTexObj( mmesa, t );
-      else
-	 mgaDestroyTexObj( mmesa, t );
+   /* find the proper destination offset for this level */
+   if ( MGA_IS_G200(mmesa) ) {
+      offset = (t->base.memBlock->ofs + t->offsets[hwlevel]);
    }
+   else {
+      unsigned  i;
 
-
-   if (in_use) {
-      t = (mgaTextureObjectPtr) calloc(1, sizeof(*t));
-      if (!t) return;
-
-      t->heap = heap;
-      t->MemBlock = mmAllocMem( mmesa->texHeap[heap], size, 0, offset);
-      if (!t->MemBlock) {
-	 fprintf(stderr, "Couldn't alloc placeholder sz %x ofs %x\n",
-		 (int)size, (int)offset);
-	 mmDumpMemInfo( mmesa->texHeap[heap]);
-	 return;
+      offset = t->base.memBlock->ofs;
+      for ( i = 0 ; i < hwlevel ; i++ ) {
+	 offset += (t->offsets[1] >> (i * 2));
       }
-      insert_at_head( &(mmesa->TexObjList[heap]), t );
-   }
-}
-
-
-void mgaAgeTextures( mgaContextPtr mmesa, int heap )
-{
-   MGASAREAPrivPtr sarea = mmesa->sarea;
-   int sz = 1 << (mmesa->mgaScreen->logTextureGranularity[heap]);
-   int idx, nr = 0;
 
-   /* Have to go right round from the back to ensure stuff ends up
-    * LRU in our local list...  Fix with a cursor pointer.
-    */
-   for (idx = sarea->texList[heap][MGA_NR_TEX_REGIONS].prev ;
-	idx != MGA_NR_TEX_REGIONS && nr < MGA_NR_TEX_REGIONS ;
-	idx = sarea->texList[heap][idx].prev, nr++)
-   {
-      /* If switching texturing schemes, then the SAREA might not
-       * have been properly cleared, so we need to reset the
-       * global texture LRU.
+      /* Each mipmap must be DWORD aligned.
        */
-      if ( idx * sz > mmesa->mgaScreen->textureSize[heap] ) {
-         nr = MGA_NR_TEX_REGIONS;
-         break;
-      }
-
-      if (sarea->texList[heap][idx].age > mmesa->texAge[heap]) {
-        mgaTexturesGone(mmesa, heap, idx * sz, sz,
-                        sarea->texList[heap][idx].in_use);
-      }
-   }
-
-   if (nr == MGA_NR_TEX_REGIONS) {
-      mgaTexturesGone(mmesa, heap, 0,
-		      mmesa->mgaScreen->textureSize[heap], 0);
-      mgaResetGlobalLRU( mmesa, heap );
+      offset &= ~0x03;
    }
 
 
-   if (0) {
-      mgaPrintGlobalLRU( mmesa, heap );
-      mgaPrintLocalLRU( mmesa, heap );
-   }
-
-   mmesa->texAge[heap] = sarea->texAge[heap];
-   mmesa->dirty |= MGA_UPLOAD_TEX0IMAGE | MGA_UPLOAD_TEX1IMAGE;
-}
-
-/*
- * mgaUploadSubImageLocked
- *
- * Perform an iload based update of a resident buffer.  This is used for
- * both initial loading of the entire image, and texSubImage updates.
- *
- * Performed with the hardware lock held.
- */
-void mgaUploadSubImageLocked( mgaContextPtr mmesa,
-			      mgaTextureObjectPtr t,
-			      int level,
-			      int x, int y, int width, int height )
-{
-   int		x2;
-   int		dwords;
-   int		offset;
-   struct gl_texture_image *image;
-   int		texelBytes, texelsPerDword, texelMaccess, length;
-
-   if ( level < 0 || level >= MGA_TEX_MAXLEVELS )
-      return;
+   /* Copy the texture from system memory to a memory space that can be
+    * directly used by the hardware for texturing.
+    */
 
-   image = t->tObj->Image[level];
-   if ( !image ) return;
+   texelBytes = texImage->TexFormat->TexelBytes;
+   length = texImage->Width * texImage->Height * texelBytes;
+   if ( t->base.heap->heapId == MGA_CARD_HEAP ) {
+      unsigned  tex_offset = 0;
+      unsigned  to_copy;
 
 
-   if (image->Data == 0) {
-      fprintf(stderr, "null texture image data tObj %p level %d\n",
-	      t->tObj, level);
-      return;
-   }
+      /* We may not be able to upload the entire texture in one batch due to
+       * register limits or dma buffer limits.  Split the copy up into maximum
+       * sized chunks.
+       */
 
+      offset += mmesa->mgaScreen->textureOffset[ t->base.heap->heapId ];
+      while ( length != 0 ) {
+	 mgaGetILoadBufferLocked( mmesa );
 
-   /* find the proper destination offset for this level */
-   offset = (t->MemBlock->ofs +
-	     t->offsets[level]);
-
-
-   texelBytes = t->texelBytes;
-   switch( texelBytes ) {
-   case 1:
-      texelsPerDword = 4;
-      texelMaccess = 0;
-      break;
-   case 2:
-      texelsPerDword = 2;
-      texelMaccess = 1;
-      break;
-   case 4:
-      texelsPerDword = 1;
-      texelMaccess = 2;
-      break;
-   default:
-      return;
-   }
+	 /* The kernel ILOAD ioctl requires that the lenght be an even multiple
+	  * of MGA_ILOAD_ALIGN.
+	  */
+	 length = ((length) + MGA_ILOAD_MASK) & ~MGA_ILOAD_MASK;
 
+	 to_copy = MIN2( length, MGA_BUFFER_SIZE );
+	 (void) memcpy( mmesa->iload_buffer->address,
+			(GLubyte *) texImage->Data + tex_offset, to_copy );
 
-   /* We can't do a subimage update if pitch is < 32 texels due
-    * to hardware XY addressing limits, so we will need to
-    * linearly upload all modified rows.
-    */
-   if ( image->Width < 32 ) {
-      x = 0;
-      width = image->Width * height;
-      height = 1;
-
-      /* Assume that 1x1 textures aren't going to cause a
-       * bus error if we read up to four texels from that
-       * location:
-       */
-/*        if ( width < texelsPerDword ) { */
-/*  	 width = texelsPerDword; */
-/*        } */
-   } else {
-      /* pad the size out to dwords.  The image is a pointer
-	 to the entire image, so we can safely reference
-	 outside the x,y,width,height bounds if we need to */
-      x2 = x + width;
-      x2 = (x2 + (texelsPerDword-1)) & ~(texelsPerDword-1);
-      x = (x + (texelsPerDword-1)) & ~(texelsPerDword-1);
-      width = x2 - x;
-   }
+	 if ( MGA_DEBUG & DEBUG_VERBOSE_TEXTURE )
+	     fprintf(stderr, "[%s:%d] address/size = 0x%08lx/%d\n",
+		     __FILE__, __LINE__,
+		     (long) (offset + tex_offset),
+		     to_copy );
 
-   /* we may not be able to upload the entire texture in one
-      batch due to register limits or dma buffer limits.
-      Recursively split it up. */
-   while ( 1 ) {
-      dwords = height * width / texelsPerDword;
-      if ( dwords * 4 <= MGA_BUFFER_SIZE ) {
-	 break;
+	 mgaFireILoadLocked( mmesa, offset + tex_offset, to_copy );
+	 tex_offset += to_copy;
+	 length -= to_copy;
       }
-
-      mgaUploadSubImageLocked( mmesa, t, level, x, y,
-			       width, height >> 1 );
-      y += ( height >> 1 );
-      height -= ( height >> 1 );
-   }
-
-   length = dwords * 4;
-
-   /* Fill in the secondary buffer with properly converted texels
-    * from the mesa buffer. */
-   /* FIXME: the sync for direct copy reduces speed.. */
-   if(t->heap == MGA_CARD_HEAP  ) {
-      mgaGetILoadBufferLocked( mmesa );
-      mgaConvertTexture( (GLuint *)mmesa->iload_buffer->address,
-			 texelBytes, image, x, y, width, height );
-      if(length < 64) length = 64;
-
-      if (0)
-	 fprintf(stderr, "TexelBytes : %d, offset: %d, length : %d\n",
-		 texelBytes,
-		 mmesa->mgaScreen->textureOffset[t->heap] +
-		 offset +
-		 y * width * 4/texelsPerDword,
-		 length);
-
-      mgaFireILoadLocked( mmesa,
-			  mmesa->mgaScreen->textureOffset[t->heap] +
-			  offset +
-			  y * width * 4/texelsPerDword,
-			  length);
    } else {
+      /* FIXME: the sync for direct copy reduces speed.. */
       /* This works, is slower for uploads to card space and needs
        * additional synchronization with the dma stream.
        */
        
       UPDATE_LOCK(mmesa, DRM_LOCK_FLUSH | DRM_LOCK_QUIESCENT);
-      mgaConvertTexture( (GLuint *)
-			 (mmesa->mgaScreen->texVirtual[t->heap] +
-			  offset +
-			  y * width * 4/texelsPerDword),
-			 texelBytes, image, x, y, width, height );
-   }
-}
-
-
-static void mgaUploadTexLevel( mgaContextPtr mmesa,
-			       mgaTextureObjectPtr t,
-			       int l )
-{
-   mgaUploadSubImageLocked( mmesa,
-			    t,
-			    l,
-			    0, 0,
-			    t->tObj->Image[l]->Width,
-			    t->tObj->Image[l]->Height);
-}
-
-
-
-
-#if 0
-static void mgaMigrateTexture( mgaContextPtr mmesa, mgaTextureObjectPtr t )
-{
-	/* NOT DONE */
-}
-#endif
-
-
-static int mgaChooseTexHeap( mgaContextPtr mmesa, mgaTextureObjectPtr t )
-{
-   int freeagp, freecard;
-   int fitincard, fitinagp;
-   int totalcard, totalagp;
-   TMemBlock *b;
-
-   totalcard = totalagp = fitincard = fitinagp = freeagp = freecard = 0;
-
-   b = mmesa->texHeap[0];
-   while(b)
-   {
-     totalcard += b->size;
-     if(b->free) if(t->totalSize <= b->size)fitincard = 1;
-     b = b->next;
-   }
-
-   b = mmesa->texHeap[1];
-   while(b)
-   {
-     totalagp += b->size;
-     if(b->free)  if(t->totalSize <= b->size)fitinagp = 1;
-     b = b->next;
-   }
 
-   if(fitincard)return 0;
-   if(fitinagp)return 1;
+      memcpy( mmesa->mgaScreen->texVirtual[t->base.heap->heapId] + offset,
+	      texImage->Data, length );
 
-   if(totalcard && totalagp)
-   {
-     int ages;
-     int ratio = (totalcard > totalagp) ? totalcard / totalagp : totalagp / totalcard;
-     ages = mmesa->sarea->texAge[0] + mmesa->sarea->texAge[1];
-     if( (ages % ratio) == 0)return totalcard > totalagp ? 1 : 0;
-     else return totalcard > totalagp ? 0 : 1;
+      if ( MGA_DEBUG & DEBUG_VERBOSE_TEXTURE )
+	 fprintf(stderr, "[%s:%d] address/size = 0x%08lx/%d\n",
+		 __FILE__, __LINE__,
+		 (long) (mmesa->mgaScreen->texVirtual[t->base.heap->heapId] 
+			 + offset),
+		 length);
    }
-
-   if(totalagp) return 1;
-   return 0;
 }
 
 
+/**
+ * Upload the texture images associated with texture \a t.  This might
+ * require the allocation of texture memory.
+ * 
+ * \param mmesa Context pointer
+ * \param t Texture to be uploaded
+ */
+
 int mgaUploadTexImages( mgaContextPtr mmesa, mgaTextureObjectPtr t )
 {
-   int heap;
    int i;
    int ofs;
 
-   heap = t->heap = mgaChooseTexHeap( mmesa, t );
 
-   /* Do we need to eject LRU texture objects?
-    */
-   if (!t->MemBlock) {
-      while (1)
-      {
-	 mgaTextureObjectPtr tmp = mmesa->TexObjList[heap].prev;
-
-	 t->MemBlock = mmAllocMem( mmesa->texHeap[heap],
-				   t->totalSize,
-				   6, 0 );
-	 if (t->MemBlock)
-	    break;
-
-	 if (mmesa->TexObjList[heap].prev->bound) {
-	    fprintf(stderr, "Hit bound texture in upload\n");
-	    return -1;
-	 }
+   if ( (t == NULL) || (t->base.totalSize == 0) )
+      return 0;
 
-	 if (mmesa->TexObjList[heap].prev ==
-	     &(mmesa->TexObjList[heap]))
-	 {
-	    fprintf(stderr, "Failed to upload texture, sz %d\n", t->totalSize);
-	    mmDumpMemInfo( mmesa->texHeap[heap] );
-	    return -1;
-	 }
+   LOCK_HARDWARE( mmesa );
 
-	 mgaDestroyTexObj( mmesa, tmp );
+   if (t->base.memBlock == NULL ) {
+      int heap;
+
+      heap = driAllocateTexture( mmesa->texture_heaps, mmesa->nr_heaps,
+				 (driTextureObject *) t );
+      if ( heap == -1 ) {
+	 UNLOCK_HARDWARE( mmesa );
+	 return -1;
       }
 
-      ofs = t->MemBlock->ofs
-	 + mmesa->mgaScreen->textureOffset[heap]
-	 ;
+      ofs = mmesa->mgaScreen->textureOffset[ heap ]
+	   + t->base.memBlock->ofs;
 
-      t->setup.texorg  = ofs;
-      t->setup.texorg1 = ofs + t->offsets[1];
-      t->setup.texorg2 = ofs + t->offsets[2];
-      t->setup.texorg3 = ofs + t->offsets[3];
-      t->setup.texorg4 = ofs + t->offsets[4];
+      if ( MGA_IS_G200(mmesa) ) {
+	 t->setup.texorg  = ofs;
+	 t->setup.texorg1 = ofs + t->offsets[1];
+	 t->setup.texorg2 = ofs + t->offsets[2];
+	 t->setup.texorg3 = ofs + t->offsets[3];
+	 t->setup.texorg4 = ofs + t->offsets[4];
+      }
+      else {
+	 t->setup.texorg  = ofs | TO_texorgoffsetsel;
+	 t->setup.texorg1 = t->offsets[1];
+	 t->setup.texorg2 = 0;
+	 t->setup.texorg3 = 0;
+	 t->setup.texorg4 = 0;
+      }
 
       mmesa->dirty |= MGA_UPLOAD_CONTEXT;
    }
 
    /* Let the world know we've used this memory recently.
     */
-   mgaUpdateTexLRU( mmesa, t );
-
+   driUpdateTextureLRU( (driTextureObject *) t );
 
-   if (MGA_DEBUG&DEBUG_VERBOSE_LRU)
-      fprintf(stderr, "dispatch age: %d age freed memory: %d\n",
+   if (MGA_DEBUG&DEBUG_VERBOSE_TEXTURE)
+      fprintf(stderr, "[%s:%d] dispatch age: %d age freed memory: %d\n",
+	      __FILE__, __LINE__,
 	      GET_DISPATCH_AGE(mmesa), mmesa->dirtyAge);
 
    if (mmesa->dirtyAge >= GET_DISPATCH_AGE(mmesa))
       mgaWaitAgeLocked( mmesa, mmesa->dirtyAge );
 
-   if (t->dirty_images) {
-      if (MGA_DEBUG&DEBUG_VERBOSE_LRU)
-	 fprintf(stderr, "*");
+   if (t->base.dirty_images[0]) {
+      const int numLevels = t->base.lastLevel - t->base.firstLevel + 1;
 
-      for (i = 0 ; i <= t->lastLevel ; i++)
-	 if (t->dirty_images & (1<<i))
-	    mgaUploadTexLevel( mmesa, t, i );
+      if (MGA_DEBUG&DEBUG_VERBOSE_TEXTURE)
+	 fprintf(stderr, "[%s:%d] dirty_images[0] = 0x%04x\n",
+		 __FILE__, __LINE__, t->base.dirty_images[0] );
+
+      for (i = 0 ; i < numLevels ; i++) {
+	 if ( (t->base.dirty_images[0] & (1U << i)) != 0 ) {
+	    mgaUploadSubImage( mmesa, t, i );
+	 }
+      }
+      t->base.dirty_images[0] = 0;
    }
 
 
-   t->dirty_images = 0;
+   UNLOCK_HARDWARE( mmesa );
+
    return 0;
 }
diff --git a/src/mesa/drivers/dri/mga/mgatris.c b/src/mesa/drivers/dri/mga/mgatris.c
index e47cfc171fd..054c7f0635d 100644
--- a/src/mesa/drivers/dri/mga/mgatris.c
+++ b/src/mesa/drivers/dri/mga/mgatris.c
@@ -26,9 +26,6 @@
  */
 /* $XFree86: xc/lib/GL/mesa/src/drv/mga/mgatris.c,v 1.10 2002/10/30 12:51:36 alanh Exp $ */
 
-#include <stdio.h>
-#include <math.h>
-
 #include "mtypes.h"
 #include "macros.h"
 #include "colormac.h"
@@ -676,15 +673,6 @@ static void mgaFastRenderClippedPoly( GLcontext *ctx, const GLuint *elts,
 /**********************************************************************/
 
 
-
-#define _MGA_NEW_RENDERSTATE (_DD_NEW_LINE_STIPPLE |		\
-			       _DD_NEW_TRI_UNFILLED |		\
-			       _DD_NEW_TRI_LIGHT_TWOSIDE |	\
-			       _DD_NEW_TRI_OFFSET |		\
-			       _DD_NEW_TRI_STIPPLE |		\
-			       _NEW_POLYGONSTIPPLE)
-
-
 #define POINT_FALLBACK (DD_POINT_SMOOTH)
 #define LINE_FALLBACK (DD_LINE_SMOOTH | DD_LINE_STIPPLE)
 #define TRI_FALLBACK (DD_TRI_SMOOTH | DD_TRI_UNFILLED)
@@ -693,7 +681,7 @@ static void mgaFastRenderClippedPoly( GLcontext *ctx, const GLuint *elts,
 #define ANY_RASTER_FLAGS (DD_FLATSHADE|DD_TRI_LIGHT_TWOSIDE|DD_TRI_OFFSET| \
                           DD_TRI_UNFILLED)
 
-static void mgaChooseRenderState(GLcontext *ctx)
+void mgaChooseRenderState(GLcontext *ctx)
 {
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    mgaContextPtr mmesa = MGA_CONTEXT(ctx);
@@ -759,36 +747,6 @@ static void mgaChooseRenderState(GLcontext *ctx)
 /**********************************************************************/
 
 
-static void mgaRunPipeline( GLcontext *ctx )
-{
-   mgaContextPtr mmesa = MGA_CONTEXT(ctx);
-
-   if (mmesa->new_state) {
-      mgaDDUpdateHwState( ctx );
-   }
-
-   if (!mmesa->Fallback && mmesa->new_gl_state) {
-      if (mmesa->new_gl_state & _MGA_NEW_RASTERSETUP)
-	 mgaChooseVertexState( ctx );
-      
-      if (mmesa->new_gl_state & _MGA_NEW_RENDERSTATE)
-	 mgaChooseRenderState( ctx );
-      
-      mmesa->new_gl_state = 0;
-
-      /* Circularity: mgaDDUpdateHwState can affect mmesa->Fallback,
-       * but mgaChooseVertexState can affect mmesa->new_state.  Hence
-       * the second check.  (Fix this...)
-       */
-      if (mmesa->new_state) {
-	 mgaDDUpdateHwState( ctx );
-      }
-   }
-
-   _tnl_run_pipeline( ctx );
-}
-
-
 static GLenum reduced_prim[GL_POLYGON+1] = {
    GL_POINTS,
    GL_LINES,
@@ -812,10 +770,14 @@ void mgaRasterPrimitive( GLcontext *ctx, GLenum prim, GLuint hwprim )
    mgaContextPtr mmesa = MGA_CONTEXT( ctx );
 
    FLUSH_BATCH( mmesa );
+
+   /* Update culling */
+   if (mmesa->raster_primitive != prim)
+      mmesa->dirty |= MGA_UPLOAD_CONTEXT;
+
    mmesa->raster_primitive = prim;
 /*     mmesa->hw_primitive = hwprim; */
    mmesa->hw_primitive = MGA_WA_TRIANGLES; /* disable mgarender.c for now */
-   mgaUpdateCull(ctx);   
 
    if (ctx->Polygon.StippleFlag && mmesa->haveHwStipple)
    {
@@ -864,6 +826,28 @@ static void mgaRenderFinish( GLcontext *ctx )
 /*               Manage total rasterization fallbacks                 */
 /**********************************************************************/
 
+static const char * const fallbackStrings[] = {
+   "Texture mode",
+   "glDrawBuffer(GL_FRONT_AND_BACK)",
+   "read buffer",
+   "LogicOp != GL_COPY",
+   "glRenderMode(selection or feedback)",
+   "No hardware stencil",
+   "glDepthFunc( GL_NEVER )",
+   "Mixing GL_CLAMP_TO_EDGE and GL_CLAMP"
+};
+
+static const char *getFallbackString(GLuint bit)
+{
+   int i = 0;
+   while (bit > 1) {
+      i++;
+      bit >>= 1;
+   }
+   return fallbackStrings[i];
+}
+
+
 void mgaFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
 {
    TNLcontext *tnl = TNL_CONTEXT(ctx);
@@ -876,6 +860,10 @@ void mgaFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
 	 FLUSH_BATCH(mmesa);
 	 _swsetup_Wakeup( ctx );
 	 mmesa->RenderIndex = ~0;
+         if (MGA_DEBUG & DEBUG_VERBOSE_FALLBACK) {
+            fprintf(stderr, "MGA begin rasterization fallback: 0x%x %s\n",
+                    bit, getFallbackString(bit));
+         }
       }
    }
    else {
@@ -886,8 +874,12 @@ void mgaFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
 	 tnl->Driver.Render.PrimitiveNotify = mgaRenderPrimitive;
 	 tnl->Driver.Render.Finish = mgaRenderFinish;
 	 tnl->Driver.Render.BuildVertices = mgaBuildVertices;
-	 mmesa->new_gl_state |= (_MGA_NEW_RENDERSTATE |
-				 _MGA_NEW_RASTERSETUP);
+	 mmesa->NewGLState |= (_MGA_NEW_RENDERSTATE |
+			       _MGA_NEW_RASTERSETUP);
+         if (MGA_DEBUG & DEBUG_VERBOSE_FALLBACK) {
+            fprintf(stderr, "MGA end rasterization fallback: 0x%x %s\n",
+                    bit, getFallbackString(bit));
+         }
       }
    }
 }
@@ -905,7 +897,6 @@ void mgaDDInitTriFuncs( GLcontext *ctx )
 
    mmesa->RenderIndex = ~0;
 	
-   tnl->Driver.RunPipeline               = mgaRunPipeline;
    tnl->Driver.Render.Start              = mgaCheckTexSizes;
    tnl->Driver.Render.Finish             = mgaRenderFinish; 
    tnl->Driver.Render.PrimitiveNotify    = mgaRenderPrimitive;
diff --git a/src/mesa/drivers/dri/mga/mgatris.h b/src/mesa/drivers/dri/mga/mgatris.h
index 88eda91e13b..a8c0b8936b7 100644
--- a/src/mesa/drivers/dri/mga/mgatris.h
+++ b/src/mesa/drivers/dri/mga/mgatris.h
@@ -32,12 +32,17 @@
 #include "mtypes.h"
 
 extern void mgaDDInitTriFuncs( GLcontext *ctx );
-
+extern void mgaChooseRenderState( GLcontext *ctx );
 extern void mgaRasterPrimitive( GLcontext *ctx, GLenum prim, GLuint hwprim );
 
 extern void mgaFallback( GLcontext *ctx, GLuint bit, GLboolean mode );
 #define FALLBACK( ctx, bit, mode ) mgaFallback( ctx, bit, mode )
 
-
+#define _MGA_NEW_RENDERSTATE (_DD_NEW_LINE_STIPPLE |		\
+			       _DD_NEW_TRI_UNFILLED |		\
+			       _DD_NEW_TRI_LIGHT_TWOSIDE |	\
+			       _DD_NEW_TRI_OFFSET |		\
+			       _DD_NEW_TRI_STIPPLE |		\
+			       _NEW_POLYGONSTIPPLE)
 
 #endif
diff --git a/src/mesa/drivers/dri/mga/mgavb.c b/src/mesa/drivers/dri/mga/mgavb.c
index d354fa43d2e..34906865e8d 100644
--- a/src/mesa/drivers/dri/mga/mgavb.c
+++ b/src/mesa/drivers/dri/mga/mgavb.c
@@ -24,7 +24,7 @@
  * Authors:
  *    Keith Whitwell <[email protected]>
  */
-/* $XFree86: xc/lib/GL/mesa/src/drv/mga/mgavb.c,v 1.15 2003/03/26 20:43:49 tsi Exp $ */
+/* $XFree86: xc/lib/GL/mesa/src/drv/mga/mgavb.c,v 1.14 2002/10/30 12:51:36 alanh Exp $ */
 
 #include "mgacontext.h"
 #include "mgavb.h"
@@ -34,18 +34,14 @@
 
 #include "glheader.h"
 #include "mtypes.h"
-/*#include "mem.h" */
+#include "imports.h"
 #include "macros.h"
 #include "colormac.h"
-/*#include "mmath.h"*/
 
 #include "tnl/t_context.h"
 #include "swrast_setup/swrast_setup.h"
 #include "swrast/swrast.h"
 
-#include <stdio.h>
-#include <stdlib.h>
-
 
 #define MGA_TEX1_BIT       0x1
 #define MGA_TEX0_BIT       0x2	
@@ -93,7 +89,7 @@ static struct {
 #define GET_VIEWPORT_MAT() mmesa->hw_viewport
 #define GET_TEXSOURCE(n)  mmesa->tmu_source[n]
 #define GET_VERTEX_FORMAT() mmesa->vertex_format
-#define GET_VERTEX_STORE() ((GLubyte *)mmesa->verts)
+#define GET_VERTEX_STORE() mmesa->verts
 #define GET_VERTEX_STRIDE_SHIFT() mmesa->vertex_stride_shift
 #define GET_UBYTE_COLOR_STORE() &mmesa->UbyteColor
 #define GET_UBYTE_SPEC_COLOR_STORE() &mmesa->UbyteSecondaryColor
@@ -407,15 +403,18 @@ void mgaChooseVertexState( GLcontext *ctx )
    if (ctx->Fog.Enabled) 
       ind |= MGA_FOG_BIT;
    
-   if (ctx->Texture.Unit[1]._ReallyEnabled & (TEXTURE_1D_BIT|TEXTURE_2D_BIT)) {
-      if (ctx->Texture.Unit[0]._ReallyEnabled & (TEXTURE_1D_BIT|TEXTURE_2D_BIT)) {
+   if (ctx->Texture._EnabledUnits & 0x2) {
+      /* unit 1 enabled */
+      if (ctx->Texture._EnabledUnits & 0x1) {
+         /* unit 0 enabled */
 	 ind |= MGA_TEX1_BIT|MGA_TEX0_BIT;
       }
       else {
 	 ind |= MGA_TEX0_BIT;
       }
    }
-   else if (ctx->Texture.Unit[0]._ReallyEnabled & (TEXTURE_1D_BIT|TEXTURE_2D_BIT)) {
+   else if (ctx->Texture._EnabledUnits & 0x1) {
+      /* unit 0 enabled */
       ind |= MGA_TEX0_BIT;
    }
    
@@ -431,7 +430,6 @@ void mgaChooseVertexState( GLcontext *ctx )
 
    if (setup_tab[ind].vertex_format != mmesa->vertex_format) {
       FLUSH_BATCH(mmesa);      
-      mmesa->new_state |= MGA_NEW_WARP;
       mmesa->dirty |= MGA_UPLOAD_PIPE;
       mmesa->vertex_format = setup_tab[ind].vertex_format;
       mmesa->vertex_size = setup_tab[ind].vertex_size;
@@ -458,7 +456,7 @@ void mgaInitVB( GLcontext *ctx )
    mgaContextPtr mmesa = MGA_CONTEXT(ctx);
    GLuint size = TNL_CONTEXT(ctx)->vb.Size;
 
-   mmesa->verts = (char *)ALIGN_MALLOC(size * sizeof(mgaVertex), 32);
+   mmesa->verts = ALIGN_MALLOC(size * sizeof(mgaVertex), 32);
 
    {
       static int firsttime = 1;
@@ -468,7 +466,6 @@ void mgaInitVB( GLcontext *ctx )
       }
    }
 
-   mmesa->new_state |= MGA_NEW_WARP;
    mmesa->dirty |= MGA_UPLOAD_PIPE;
    mmesa->vertex_format = setup_tab[0].vertex_format;
    mmesa->vertex_size = setup_tab[0].vertex_size;
diff --git a/src/mesa/drivers/dri/mga/server/mga_common.h b/src/mesa/drivers/dri/mga/server/mga_common.h
index 90f6b37f4e6..dcc260a1a59 100644
--- a/src/mesa/drivers/dri/mga/server/mga_common.h
+++ b/src/mesa/drivers/dri/mga/server/mga_common.h
@@ -146,7 +146,7 @@ typedef struct {
 
 typedef struct {
 	int param;
-	int *value;
+	void *value;
 } drmMGAGetParam;
 
 #endif
diff --git a/src/mesa/drivers/dri/r128/Makefile.X11 b/src/mesa/drivers/dri/r128/Makefile.X11
new file mode 100644
index 00000000000..8ae146af915
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/Makefile.X11
@@ -0,0 +1,116 @@
+# $Id: Makefile.X11,v 1.1 2003/08/22 20:11:44 brianp Exp $
+
+# Mesa 3-D graphics library
+# Version:  5.0
+# Copyright (C) 1995-2002  Brian Paul
+
+TOP = ../../../../..
+
+default: linux-solo
+
+SHARED_INCLUDES = $(INCLUDE_DIRS) -I. -I../common -Iserver
+MINIGLX_INCLUDES = -I$(TOP)/src/glx/mini
+
+DEFINES += \
+	-D_HAVE_SWRAST=1 \
+	-D_HAVE_SWTNL=1 \
+	-D_HAVE_SANITY=1 \
+	-D_HAVE_CODEGEN=1 \
+	-D_HAVE_LIGHTING=1 \
+	-D_HAVE_TEXGEN=1 \
+	-D_HAVE_USERCLIP=1 \
+	-DGLX_DIRECT_RENDERING 
+
+MINIGLX_SOURCES = server/r128_dri.c 
+
+DRIVER_SOURCES = \
+		r128_context.c \
+		r128_lock.c \
+		r128_state.c \
+		r128_texstate.c \
+		r128_dd.c \
+		r128_screen.c \
+		r128_tex.c \
+		r128_tris.c \
+		r128_ioctl.c \
+		r128_span.c \
+		r128_texmem.c \
+		r128_vb.c \
+                ../common/mm.c \
+                ../common/utils.c \
+                ../common/texmem.c \
+                ../common/vblank.c
+
+INCLUDES = $(MINIGLX_INCLUDES) \
+	   $(SHARED_INCLUDES)
+
+
+C_SOURCES = $(DRIVER_SOURCES) \
+	    $(MINIGLX_SOURCES) 
+
+MESA_MODULES = $(TOP)/src/mesa/mesa.a
+
+
+ifeq ($(WINDOW_SYSTEM),dri)
+WINOBJ=$(MESABUILDDIR)/dri/dri.a
+WINLIB=
+else
+WINOBJ=
+WINLIB=-L$(MESA)/src/glx/mini
+endif
+
+ASM_SOURCES = 
+OBJECTS = $(C_SOURCES:.c=.o) \
+	  $(ASM_SOURCES:.S=.o) 
+
+### Include directories
+
+INCLUDE_DIRS = \
+	-I$(TOP)/include \
+	-I$(TOP)/src/mesa \
+	-I$(TOP)/src/mesa/main \
+	-I$(TOP)/src/mesa/glapi \
+	-I$(TOP)/src/mesa/math \
+	-I$(TOP)/src/mesa/transform \
+	-I$(TOP)/src/mesa/swrast \
+	-I$(TOP)/src/mesa/swrast_setup
+
+
+##### RULES #####
+
+.c.o:
+	$(CC) -c $(SHARED_INCLUDES) $(MINIGLX_INCLUDES) $(CFLAGS) $(DEFINES) $< -o $@
+
+.S.o:
+	$(CC) -c $(SHARED_INCLUDES) $(MINIGLX_INCLUDES) $(CFLAGS) $(DEFINES)  $< -o $@
+
+
+##### TARGETS #####
+
+targets: depend r128_dri.so
+
+r128_dri.so:  $(OBJECTS) $(MESA_MODULES) $(WINOBJ) Makefile.X11
+	rm -f $@ && gcc -o $@ -shared $(OBJECTS) $(MESA_MODULES) $(WINOBJ) $(WINLIB) -lc $(GL_LIB_DEPS)
+	rm -f $(TOP)/lib/r128_dri.so && \
+	install r128_dri.so $(TOP)/lib/r128_dri.so
+
+# Run 'make -f Makefile.X11 dep' to update the dependencies if you change
+# what's included by any source file.
+depend: $(C_SOURCES) $(ASM_SOURCES)
+	makedepend -fdepend -Y $(SHARED_INCLUDES) $(MINIGLX_INCLUDES) \
+		$(C_SOURCES) $(ASM_SOURCES)
+
+
+# Emacs tags
+tags:
+	etags `find . -name \*.[ch]` `find ../include`
+
+
+# Remove .o and backup files
+clean:
+	-rm -f *.o */*.o *~ *.o *~ *.so server/*.o
+
+
+include $(TOP)/Make-config
+
+include depend
diff --git a/src/mesa/drivers/dri/r128/r128_context.c b/src/mesa/drivers/dri/r128/r128_context.c
new file mode 100644
index 00000000000..f7f6d7141d2
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_context.c
@@ -0,0 +1,335 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_context.c,v 1.8 2002/10/30 12:51:38 alanh Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ *
+ */
+
+#include "glheader.h"
+#include "context.h"
+#include "simple_list.h"
+#include "imports.h"
+#include "matrix.h"
+#include "extensions.h"
+
+#include "swrast/swrast.h"
+#include "swrast_setup/swrast_setup.h"
+#include "array_cache/acache.h"
+
+#include "tnl/tnl.h"
+#include "tnl/t_pipeline.h"
+
+#include "r128_context.h"
+#include "r128_ioctl.h"
+#include "r128_dd.h"
+#include "r128_state.h"
+#include "r128_span.h"
+#include "r128_tex.h"
+#include "r128_tris.h"
+#include "r128_vb.h"
+
+#include "vblank.h"
+#include "utils.h"
+#include "texmem.h"
+
+#ifndef R128_DEBUG
+int R128_DEBUG = 0;
+#endif
+
+static const char * const card_extensions[] =
+{
+   "GL_ARB_multitexture",
+   "GL_ARB_texture_env_add",
+   "GL_ARB_texture_mirrored_repeat",
+   "GL_EXT_texture_edge_clamp",
+   "GL_EXT_texture_env_add",
+   "GL_IBM_texture_mirrored_repeat",
+   "GL_SGIS_generate_mipmap",
+   "GL_SGIS_texture_edge_clamp",
+   NULL
+};
+
+static const struct dri_debug_control debug_control[] =
+{
+    { "ioctl", DEBUG_VERBOSE_IOCTL },
+    { "verb",  DEBUG_VERBOSE_MSG },
+    { "dri",   DEBUG_VERBOSE_DRI },
+    { "2d",    DEBUG_VERBOSE_2D },
+    { "sync",  DEBUG_ALWAYS_SYNC },
+    { "api",   DEBUG_VERBOSE_API },
+    { NULL,    0 }
+};
+
+/* Create the device specific context.
+ */
+GLboolean r128CreateContext( const __GLcontextModes *glVisual,
+			     __DRIcontextPrivate *driContextPriv,
+                             void *sharedContextPrivate )
+{
+   GLcontext *ctx, *shareCtx;
+   __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv;
+   r128ContextPtr rmesa;
+   r128ScreenPtr r128scrn;
+   int i;
+
+   /* Allocate the r128 context */
+   rmesa = (r128ContextPtr) CALLOC( sizeof(*rmesa) );
+   if ( !rmesa )
+      return GL_FALSE;
+
+   /* Allocate the Mesa context */
+   if (sharedContextPrivate)
+      shareCtx = ((r128ContextPtr) sharedContextPrivate)->glCtx;
+   else 
+      shareCtx = NULL;
+   rmesa->glCtx = _mesa_create_context(glVisual, shareCtx, (void *) rmesa, GL_TRUE);
+   if (!rmesa->glCtx) {
+      FREE(rmesa);
+      return GL_FALSE;
+   }
+   driContextPriv->driverPrivate = rmesa;
+   ctx = rmesa->glCtx;
+
+   rmesa->driContext = driContextPriv;
+   rmesa->driScreen = sPriv;
+   rmesa->driDrawable = NULL;
+   rmesa->hHWContext = driContextPriv->hHWContext;
+   rmesa->driHwLock = &sPriv->pSAREA->lock;
+   rmesa->driFd = sPriv->fd;
+
+   r128scrn = rmesa->r128Screen = (r128ScreenPtr)(sPriv->private);
+
+   rmesa->sarea = (R128SAREAPrivPtr)((char *)sPriv->pSAREA +
+				     r128scrn->sarea_priv_offset);
+
+   rmesa->CurrentTexObj[0] = NULL;
+   rmesa->CurrentTexObj[1] = NULL;
+
+   (void) memset( rmesa->texture_heaps, 0, sizeof( rmesa->texture_heaps ) );
+   make_empty_list( & rmesa->swapped );
+
+   rmesa->nr_heaps = r128scrn->numTexHeaps;
+   for ( i = 0 ; i < rmesa->nr_heaps ; i++ ) {
+      rmesa->texture_heaps[i] = driCreateTextureHeap( i, rmesa,
+	    r128scrn->texSize[i],
+	    12,
+	    R128_NR_TEX_REGIONS,
+	    rmesa->sarea->texList[i],
+	    & rmesa->sarea->texAge[i],
+	    & rmesa->swapped,
+	    sizeof( r128TexObj ),
+	    (destroy_texture_object_t *) r128DestroyTexObj );
+
+      driSetTextureSwapCounterLocation( rmesa->texture_heaps[i],
+					& rmesa->c_textureSwaps );
+   }
+
+
+   rmesa->RenderIndex = -1;		/* Impossible value */
+   rmesa->vert_buf = NULL;
+   rmesa->num_verts = 0;
+
+   /* Set the maximum texture size small enough that we can guarentee that
+    * all texture units can bind a maximal texture and have them both in
+    * texturable memory at once.
+    */
+
+   ctx->Const.MaxTextureUnits = 2;
+
+   driCalculateMaxTextureLevels( rmesa->texture_heaps,
+				 rmesa->nr_heaps,
+				 & ctx->Const,
+				 4,
+				 10, /* max 2D texture size is 1024x1024 */
+				 0,  /* 3D textures unsupported. */
+				 0,  /* cube textures unsupported. */
+				 0,  /* texture rectangles unsupported. */
+				 11,
+				 GL_FALSE );
+
+   /* No wide points.
+    */
+   ctx->Const.MinPointSize = 1.0;
+   ctx->Const.MinPointSizeAA = 1.0;
+   ctx->Const.MaxPointSize = 1.0;
+   ctx->Const.MaxPointSizeAA = 1.0;
+
+   /* No wide lines.
+    */
+   ctx->Const.MinLineWidth = 1.0;
+   ctx->Const.MinLineWidthAA = 1.0;
+   ctx->Const.MaxLineWidth = 1.0;
+   ctx->Const.MaxLineWidthAA = 1.0;
+   ctx->Const.LineWidthGranularity = 1.0;
+
+#if ENABLE_PERF_BOXES
+   rmesa->boxes = (getenv( "LIBGL_PERFORMANCE_BOXES" ) != NULL);
+#endif
+
+   /* Initialize the software rasterizer and helper modules.
+    */
+   _swrast_CreateContext( ctx );
+   _ac_CreateContext( ctx );
+   _tnl_CreateContext( ctx );
+   _swsetup_CreateContext( ctx );
+
+   /* Install the customized pipeline:
+    */
+/*     _tnl_destroy_pipeline( ctx ); */
+/*     _tnl_install_pipeline( ctx, r128_pipeline ); */
+
+   /* Configure swrast to match hardware characteristics:
+    */
+   _swrast_allow_pixel_fog( ctx, GL_FALSE );
+   _swrast_allow_vertex_fog( ctx, GL_TRUE );
+
+   driInitExtensions( ctx, card_extensions, GL_TRUE );
+   if (sPriv->drmMinor >= 4)
+      _mesa_enable_extension( ctx, "GL_MESA_ycbcr_texture" );
+
+   r128InitVB( ctx );
+   r128InitTriFuncs( ctx );
+   r128DDInitDriverFuncs( ctx );
+   r128DDInitIoctlFuncs( ctx );
+   r128DDInitStateFuncs( ctx );
+   r128DDInitSpanFuncs( ctx );
+   r128DDInitTextureFuncs( ctx );
+   r128DDInitState( rmesa );
+
+   rmesa->do_irqs = (rmesa->r128Screen->irq && !getenv("R128_NO_IRQS"));
+
+   rmesa->vblank_flags = (rmesa->do_irqs)
+       ? driGetDefaultVBlankFlags() : VBLANK_FLAG_NO_IRQ;
+
+   driContextPriv->driverPrivate = (void *)rmesa;
+
+#if DO_DEBUG
+   R128_DEBUG = driParseDebugString( getenv( "R128_DEBUG" ),
+				     debug_control );
+#endif
+
+   return GL_TRUE;
+}
+
+/* Destroy the device specific context.
+ */
+void r128DestroyContext( __DRIcontextPrivate *driContextPriv  )
+{
+   r128ContextPtr rmesa = (r128ContextPtr) driContextPriv->driverPrivate;
+
+   assert(rmesa);  /* should never be null */
+   if ( rmesa ) {
+      GLboolean   release_texture_heaps;
+
+
+      release_texture_heaps = (rmesa->glCtx->Shared->RefCount == 1);
+
+      _swsetup_DestroyContext( rmesa->glCtx );
+      _tnl_DestroyContext( rmesa->glCtx );
+      _ac_DestroyContext( rmesa->glCtx );
+      _swrast_DestroyContext( rmesa->glCtx );
+
+      r128FreeVB( rmesa->glCtx );
+
+      /* free the Mesa context */
+      rmesa->glCtx->DriverCtx = NULL;
+      _mesa_destroy_context(rmesa->glCtx);
+
+      if ( release_texture_heaps ) {
+         /* This share group is about to go away, free our private
+          * texture object data.
+          */
+         int i;
+
+	 assert( is_empty_list( & rmesa->swapped ) );
+
+         for ( i = 0 ; i < rmesa->nr_heaps ; i++ ) {
+	    driDestroyTextureHeap( rmesa->texture_heaps[ i ] );
+	    rmesa->texture_heaps[ i ] = NULL;
+         }
+      }
+
+      FREE( rmesa );
+   }
+
+#if 0
+   /* Use this to force shared object profiling. */
+   glx_fini_prof();
+#endif
+}
+
+
+/* Force the context `c' to be the current context and associate with it
+ * buffer `b'.
+ */
+GLboolean
+r128MakeCurrent( __DRIcontextPrivate *driContextPriv,
+                 __DRIdrawablePrivate *driDrawPriv,
+                 __DRIdrawablePrivate *driReadPriv )
+{
+   if ( driContextPriv ) {
+      GET_CURRENT_CONTEXT(ctx);
+      r128ContextPtr oldR128Ctx = ctx ? R128_CONTEXT(ctx) : NULL;
+      r128ContextPtr newR128Ctx = (r128ContextPtr) driContextPriv->driverPrivate;
+
+      if ( newR128Ctx != oldR128Ctx ) {
+	 newR128Ctx->new_state |= R128_NEW_CONTEXT;
+	 newR128Ctx->dirty = R128_UPLOAD_ALL;
+      }
+
+      newR128Ctx->driDrawable = driDrawPriv;
+
+      _mesa_make_current2( newR128Ctx->glCtx,
+                           (GLframebuffer *) driDrawPriv->driverPrivate,
+                           (GLframebuffer *) driReadPriv->driverPrivate );
+
+
+      newR128Ctx->new_state |= R128_NEW_WINDOW | R128_NEW_CLIP;
+
+      if ( !newR128Ctx->glCtx->Viewport.Width ) {
+	 _mesa_set_viewport(newR128Ctx->glCtx, 0, 0,
+                            driDrawPriv->w, driDrawPriv->h);
+      }
+   } else {
+      _mesa_make_current( 0, 0 );
+   }
+
+   return GL_TRUE;
+}
+
+
+/* Force the context `c' to be unbound from its buffer.
+ */
+GLboolean
+r128UnbindContext( __DRIcontextPrivate *driContextPriv )
+{
+   return GL_TRUE;
+}
diff --git a/src/mesa/drivers/dri/r128/r128_context.h b/src/mesa/drivers/dri/r128/r128_context.h
new file mode 100644
index 00000000000..9c93012eb0f
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_context.h
@@ -0,0 +1,270 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_context.h,v 1.12 2002/12/16 16:18:52 dawes Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ *
+ */
+
+#ifndef __R128_CONTEXT_H__
+#define __R128_CONTEXT_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+#include "dri_util.h"
+
+#include "xf86drm.h"
+#include "r128_common.h"
+
+#include "mtypes.h"
+
+#include "r128_reg.h"
+
+#include "texmem.h"
+
+struct r128_context;
+typedef struct r128_context r128ContextRec;
+typedef struct r128_context *r128ContextPtr;
+
+#include "r128_lock.h"
+#include "r128_texobj.h"
+#include "r128_screen.h"
+
+/* Flags for what context state needs to be updated:
+ */
+#define R128_NEW_ALPHA		0x0001
+#define R128_NEW_DEPTH		0x0002
+#define R128_NEW_FOG		0x0004
+#define R128_NEW_CLIP		0x0008
+#define R128_NEW_CULL		0x0010
+#define R128_NEW_MASKS		0x0020
+#define R128_NEW_RENDER_NOT	0x0040
+#define R128_NEW_WINDOW		0x0080
+#define R128_NEW_CONTEXT	0x0100
+#define R128_NEW_ALL		0x01ff
+
+/* Flags for software fallback cases:
+ */
+#define R128_FALLBACK_TEXTURE		0x0001
+#define R128_FALLBACK_DRAW_BUFFER	0x0002
+#define R128_FALLBACK_READ_BUFFER	0x0004
+#define R128_FALLBACK_STENCIL		0x0008
+#define R128_FALLBACK_RENDER_MODE	0x0010
+#define R128_FALLBACK_MULTIDRAW		0x0020
+#define R128_FALLBACK_LOGICOP		0x0040
+#define R128_FALLBACK_SEP_SPECULAR	0x0080
+#define R128_FALLBACK_BLEND_EQ		0x0100
+#define R128_FALLBACK_BLEND_FUNC	0x0200
+
+
+/* Use the templated vertex format:
+ */
+#define TAG(x) r128##x
+#include "tnl_dd/t_dd_vertex.h"
+#undef TAG
+
+/* Reasons why the GL_BLEND fallback mightn't work:
+ */
+#define R128_BLEND_ENV_COLOR	0x1
+#define R128_BLEND_MULTITEX	0x2
+
+/* Subpixel offsets for window coordinates (triangles):
+ */
+#define SUBPIXEL_X  (0.0F)
+#define SUBPIXEL_Y  (0.125F)
+
+
+typedef void (*r128_tri_func)( r128ContextPtr, 
+				 r128Vertex *,
+				 r128Vertex *,
+				 r128Vertex * );
+
+typedef void (*r128_line_func)( r128ContextPtr, 
+				  r128Vertex *,
+				  r128Vertex * );
+
+typedef void (*r128_point_func)( r128ContextPtr,
+				   r128Vertex * );
+
+
+struct r128_context {
+   GLcontext *glCtx;			/* Mesa context */
+
+   /* Driver and hardware state management
+    */
+   GLuint new_state;
+   GLuint dirty;			/* Hardware state to be updated */
+   r128_context_regs_t setup;
+
+   /* Temporaries for translating away float colors:
+    */
+   struct gl_client_array UbyteColor;
+   struct gl_client_array UbyteSecondaryColor;
+
+   GLuint NewGLState;
+   GLuint Fallback;
+   GLuint SetupIndex;
+   GLuint SetupNewInputs;
+   GLuint RenderIndex;
+   GLfloat hw_viewport[16];
+   GLfloat depth_scale;
+   GLuint vertex_size;
+   GLuint vertex_stride_shift;
+   GLuint vertex_format;
+   GLuint num_verts;
+   GLubyte *verts;		
+
+   CARD32 ClearColor;			/* Color used to clear color buffer */
+   CARD32 ClearDepth;			/* Value used to clear depth buffer */
+   CARD32 ClearStencil;			/* Value used to clear stencil */
+
+   /* Map GL texture units onto hardware
+    */
+   GLint multitex;
+   GLint tmu_source[2];
+   GLuint tex_combine[2];
+   GLuint blend_flags;
+   GLuint env_color;
+
+   /* Texture object bookkeeping
+    */
+   unsigned              nr_heaps;
+   driTexHeap          * texture_heaps[ R128_NR_TEX_HEAPS ];
+   driTextureObject      swapped;
+
+   r128TexObjPtr CurrentTexObj[2];
+ 
+   /* Fallback rasterization functions 
+    */
+   r128_point_func draw_point;
+   r128_line_func draw_line;
+   r128_tri_func draw_tri;
+
+   /* Vertex buffers
+    */
+   drmBufPtr vert_buf;
+
+   GLuint hw_primitive;
+   GLenum render_primitive;
+
+   /* Page flipping
+    */
+   GLuint doPageFlip;
+
+   /* Busy waiting
+    */
+   GLuint do_irqs;
+
+   /* Drawable, cliprect and scissor information
+    */
+   GLint drawOffset, drawPitch;
+   GLint readOffset, readPitch;
+
+   GLuint numClipRects;			/* Cliprects for the draw buffer */
+   XF86DRIClipRectPtr pClipRects;
+
+   GLuint scissor;
+   XF86DRIClipRectRec ScissorRect;	/* Current software scissor */
+
+   /* Mirrors of some DRI state
+    */
+   __DRIcontextPrivate	*driContext;	/* DRI context */
+   __DRIscreenPrivate	*driScreen;	/* DRI screen */
+   __DRIdrawablePrivate	*driDrawable;	/* DRI drawable bound to this ctx */
+
+   unsigned int lastStamp;	        /* mirror driDrawable->lastStamp */
+
+   drmContext hHWContext;
+   drmLock *driHwLock;
+   int driFd;
+
+   r128ScreenPtr r128Screen;		/* Screen private DRI data */
+   R128SAREAPrivPtr sarea;		/* Private SAREA data */
+
+   /* Performance counters
+    */
+   GLuint boxes;			/* Draw performance boxes */
+   GLuint hardwareWentIdle;
+   GLuint c_clears;
+   GLuint c_drawWaits;
+   GLuint c_textureSwaps;
+   GLuint c_textureBytes;
+   GLuint c_vertexBuffers;
+
+   /* VBI
+    */
+   GLuint vbl_seq;
+   GLuint vblank_flags;
+};
+
+#define R128_CONTEXT(ctx)		((r128ContextPtr)(ctx->DriverCtx))
+
+#define R128_IS_PLAIN( rmesa ) \
+		(rmesa->r128Screen->chipset == R128_CARD_TYPE_R128)
+#define R128_IS_PRO( rmesa ) \
+		(rmesa->r128Screen->chipset == R128_CARD_TYPE_R128_PRO)
+#define R128_IS_MOBILITY( rmesa ) \
+		(rmesa->r128Screen->chipset == R128_CARD_TYPE_R128_MOBILITY)
+
+
+extern GLboolean r128CreateContext( const __GLcontextModes *glVisual,
+				    __DRIcontextPrivate *driContextPriv,
+                                    void *sharedContextPrivate );
+
+extern void r128DestroyContext( __DRIcontextPrivate * );
+
+extern GLboolean r128MakeCurrent( __DRIcontextPrivate *driContextPriv,
+                                  __DRIdrawablePrivate *driDrawPriv,
+                                  __DRIdrawablePrivate *driReadPriv );
+
+extern GLboolean r128UnbindContext( __DRIcontextPrivate *driContextPriv );
+
+/* ================================================================
+ * Debugging:
+ */
+#define DO_DEBUG		0
+#define ENABLE_PERF_BOXES	0
+
+#if DO_DEBUG
+extern int R128_DEBUG;
+#else
+#define R128_DEBUG		0
+#endif
+
+#define DEBUG_ALWAYS_SYNC	0x01
+#define DEBUG_VERBOSE_API	0x02
+#define DEBUG_VERBOSE_MSG	0x04
+#define DEBUG_VERBOSE_LRU	0x08
+#define DEBUG_VERBOSE_DRI	0x10
+#define DEBUG_VERBOSE_IOCTL	0x20
+#define DEBUG_VERBOSE_2D	0x40
+
+#endif
+#endif /* __R128_CONTEXT_H__ */
diff --git a/src/mesa/drivers/dri/r128/r128_dd.c b/src/mesa/drivers/dri/r128/r128_dd.c
new file mode 100644
index 00000000000..b64c26b4bdd
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_dd.c
@@ -0,0 +1,148 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_dd.c,v 1.15 2002/10/30 12:51:38 alanh Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Gareth Hughes <[email protected]>
+ *   Kevin E. Martin <[email protected]>
+ *
+ */
+
+#include "r128_context.h"
+#include "r128_ioctl.h"
+#include "r128_state.h"
+#include "r128_vb.h"
+#include "r128_dd.h"
+
+#include "context.h"
+
+#include "utils.h"
+
+#define DRIVER_DATE	"20030328"
+
+
+/* Return the width and height of the current color buffer.
+ */
+static void r128DDGetBufferSize( GLframebuffer *buffer,
+				 GLuint *width, GLuint *height )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   LOCK_HARDWARE( rmesa );
+   *width  = rmesa->driDrawable->w;
+   *height = rmesa->driDrawable->h;
+   UNLOCK_HARDWARE( rmesa );
+}
+
+/* Return various strings for glGetString().
+ */
+static const GLubyte *r128DDGetString( GLcontext *ctx, GLenum name )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   static char buffer[128];
+   unsigned   offset;
+   const char * card_name = "Rage 128";
+   GLuint agp_mode = rmesa->r128Screen->IsPCI ? 0 :
+      rmesa->r128Screen->AGPMode;
+
+   switch ( name ) {
+   case GL_VENDOR:
+      return (GLubyte *)"VA Linux Systems, Inc.";
+
+   case GL_RENDERER:
+      /* Select the spefic chipset.
+       */
+      if ( R128_IS_PRO( rmesa ) ) {
+	 card_name = "Rage 128 Pro";
+      }
+      else if ( R128_IS_MOBILITY( rmesa ) ) {
+	 card_name = "Rage 128 Mobility";
+      }
+
+      offset = driGetRendererString( buffer, card_name, DRIVER_DATE,
+				     agp_mode );
+
+      return (GLubyte *)buffer;
+
+   default:
+      return NULL;
+   }
+}
+
+/* Send all commands to the hardware.  If vertex buffers or indirect
+ * buffers are in use, then we need to make sure they are sent to the
+ * hardware.  All commands that are normally sent to the ring are
+ * already considered `flushed'.
+ */
+static void r128DDFlush( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   FLUSH_BATCH( rmesa );
+
+#if ENABLE_PERF_BOXES
+   if ( rmesa->boxes ) {
+      LOCK_HARDWARE( rmesa );
+      r128PerformanceBoxesLocked( rmesa );
+      UNLOCK_HARDWARE( rmesa );
+   }
+
+   /* Log the performance counters if necessary */
+   r128PerformanceCounters( rmesa );
+#endif
+}
+
+/* Make sure all commands have been sent to the hardware and have
+ * completed processing.
+ */
+static void r128DDFinish( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+#if ENABLE_PERF_BOXES
+   /* Bump the performance counter */
+   rmesa->c_drawWaits++;
+#endif
+
+   r128DDFlush( ctx );
+   r128WaitForIdle( rmesa );
+}
+
+
+/* Initialize the driver's misc functions.
+ */
+void r128DDInitDriverFuncs( GLcontext *ctx )
+{
+   ctx->Driver.GetBufferSize	= r128DDGetBufferSize;
+   ctx->Driver.ResizeBuffers    = _swrast_alloc_buffers;
+   ctx->Driver.GetString	= r128DDGetString;
+   ctx->Driver.Finish		= r128DDFinish;
+   ctx->Driver.Flush		= r128DDFlush;
+   ctx->Driver.Error		= NULL;
+}
diff --git a/src/mesa/drivers/dri/r128/r128_dd.h b/src/mesa/drivers/dri/r128/r128_dd.h
new file mode 100644
index 00000000000..b8fc7ad9bfd
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_dd.h
@@ -0,0 +1,44 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_dd.h,v 1.3 2001/01/08 01:07:20 martin Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Gareth Hughes <[email protected]>
+ *   Kevin E. Martin <[email protected]>
+ *
+ */
+
+#ifndef __R128_DD_H__
+#define __R128_DD_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+extern void r128DDInitDriverFuncs( GLcontext *ctx );
+
+#endif
+#endif
diff --git a/src/mesa/drivers/dri/r128/r128_ioctl.c b/src/mesa/drivers/dri/r128/r128_ioctl.c
new file mode 100644
index 00000000000..c5843473e12
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_ioctl.c
@@ -0,0 +1,812 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_ioctl.c,v 1.10 2002/12/16 16:18:53 dawes Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Gareth Hughes <[email protected]>
+ *
+ */
+#include <errno.h>
+
+#include "r128_context.h"
+#include "r128_state.h"
+#include "r128_ioctl.h"
+#include "r128_macros.h"
+
+#include "imports.h"
+#include "macros.h"
+
+#include "swrast/swrast.h"
+
+#include "vblank.h"
+
+#define R128_TIMEOUT        2048
+#define R128_IDLE_RETRY       32
+
+
+/* =============================================================
+ * Hardware vertex buffer handling
+ */
+
+/* Get a new VB from the pool of vertex buffers in AGP space.
+ */
+drmBufPtr r128GetBufferLocked( r128ContextPtr rmesa )
+{
+   int fd = rmesa->r128Screen->driScreen->fd;
+   int index = 0;
+   int size = 0;
+   drmDMAReq dma;
+   drmBufPtr buf = NULL;
+   int to = 0;
+   int ret;
+
+   dma.context = rmesa->hHWContext;
+   dma.send_count = 0;
+   dma.send_list = NULL;
+   dma.send_sizes = NULL;
+   dma.flags = 0;
+   dma.request_count = 1;
+   dma.request_size = R128_BUFFER_SIZE;
+   dma.request_list = &index;
+   dma.request_sizes = &size;
+   dma.granted_count = 0;
+
+   while ( !buf && ( to++ < R128_TIMEOUT ) ) {
+      ret = drmDMA( fd, &dma );
+
+      if ( ret == 0 ) {
+	 buf = &rmesa->r128Screen->buffers->list[index];
+	 buf->used = 0;
+#if ENABLE_PERF_BOXES
+	 /* Bump the performance counter */
+	 rmesa->c_vertexBuffers++;
+#endif
+	 return buf;
+      }
+   }
+
+   if ( !buf ) {
+      drmCommandNone( fd, DRM_R128_CCE_RESET);
+      UNLOCK_HARDWARE( rmesa );
+      fprintf( stderr, "Error: Could not get new VB... exiting\n" );
+      exit( -1 );
+   }
+
+   return buf;
+}
+
+void r128FlushVerticesLocked( r128ContextPtr rmesa )
+{
+   XF86DRIClipRectPtr pbox = rmesa->pClipRects;
+   int nbox = rmesa->numClipRects;
+   drmBufPtr buffer = rmesa->vert_buf;
+   int count = rmesa->num_verts;
+   int prim = rmesa->hw_primitive;
+   int fd = rmesa->driScreen->fd;
+   drmR128Vertex vertex;
+   int i;
+
+   rmesa->num_verts = 0;
+   rmesa->vert_buf = NULL;
+
+   if ( !buffer )
+      return;
+
+   if ( rmesa->dirty & ~R128_UPLOAD_CLIPRECTS )
+      r128EmitHwStateLocked( rmesa );
+
+   if ( !nbox )
+      count = 0;
+
+   if ( nbox >= R128_NR_SAREA_CLIPRECTS )
+      rmesa->dirty |= R128_UPLOAD_CLIPRECTS;
+
+   if ( !count || !(rmesa->dirty & R128_UPLOAD_CLIPRECTS) )
+   {
+      if ( nbox < 3 ) {
+	 rmesa->sarea->nbox = 0;
+      } else {
+	 rmesa->sarea->nbox = nbox;
+      }
+
+      vertex.prim = prim;
+      vertex.idx = buffer->idx;
+      vertex.count = count;
+      vertex.discard = 1;
+      drmCommandWrite( fd, DRM_R128_VERTEX, &vertex, sizeof(drmR128Vertex) );
+   }
+   else
+   {
+      for ( i = 0 ; i < nbox ; ) {
+	 int nr = MIN2( i + R128_NR_SAREA_CLIPRECTS, nbox );
+	 XF86DRIClipRectPtr b = rmesa->sarea->boxes;
+	 int discard = 0;
+
+	 rmesa->sarea->nbox = nr - i;
+	 for ( ; i < nr ; i++ ) {
+	    *b++ = pbox[i];
+	 }
+
+	 /* Finished with the buffer?
+	  */
+	 if ( nr == nbox ) {
+	    discard = 1;
+	 }
+
+	 rmesa->sarea->dirty |= R128_UPLOAD_CLIPRECTS;
+
+         vertex.prim = prim;
+         vertex.idx = buffer->idx;
+         vertex.count = count;
+         vertex.discard = discard;
+         drmCommandWrite( fd, DRM_R128_VERTEX, &vertex, sizeof(drmR128Vertex) );
+      }
+   }
+
+   rmesa->dirty &= ~R128_UPLOAD_CLIPRECTS;
+}
+
+
+
+
+
+/* ================================================================
+ * Texture uploads
+ */
+
+void r128FireBlitLocked( r128ContextPtr rmesa, drmBufPtr buffer,
+			 GLint offset, GLint pitch, GLint format,
+			 GLint x, GLint y, GLint width, GLint height )
+{
+   drmR128Blit blit;
+   GLint ret;
+
+   blit.idx = buffer->idx;
+   blit.offset = offset;
+   blit.pitch = pitch;
+   blit.format = format;
+   blit.x = x;
+   blit.y = y;
+   blit.width = width;
+   blit.height = height;
+
+   ret = drmCommandWrite( rmesa->driFd, DRM_R128_BLIT, 
+                          &blit, sizeof(drmR128Blit) );
+
+   if ( ret ) {
+      UNLOCK_HARDWARE( rmesa );
+      fprintf( stderr, "DRM_R128_BLIT: return = %d\n", ret );
+      exit( 1 );
+   }
+}
+
+
+/* ================================================================
+ * SwapBuffers with client-side throttling
+ */
+
+static void delay( void ) {
+/* Prevent an optimizing compiler from removing a spin loop */
+}
+
+#define R128_MAX_OUTSTANDING	2
+
+/* Throttle the frame rate -- only allow one pending swap buffers
+ * request at a time.
+ * GH: We probably don't want a timeout here, as we can wait as
+ * long as we want for a frame to complete.  If it never does, then
+ * the card has locked.
+ */
+static int r128WaitForFrameCompletion( r128ContextPtr rmesa )
+{
+   unsigned char *R128MMIO = rmesa->r128Screen->mmio.map;
+   CARD32 frame;
+   int i;
+   int wait = 0;
+
+   while ( 1 ) {
+      frame = INREG( R128_LAST_FRAME_REG );
+      if ( rmesa->sarea->last_frame - frame <= R128_MAX_OUTSTANDING ) {
+	 break;
+      }
+
+      /* Spin in place a bit so we aren't hammering the register */
+      wait++;
+      for ( i = 0 ; i < 1024 ; i++ ) {
+	 delay();
+      }
+   }
+
+   return wait;
+}
+
+/* Copy the back color buffer to the front color buffer.
+ */
+void r128CopyBuffer( const __DRIdrawablePrivate *dPriv )
+{
+   r128ContextPtr rmesa;
+   GLint nbox, i, ret;
+   GLboolean missed_target;
+
+   assert(dPriv);
+   assert(dPriv->driContextPriv);
+   assert(dPriv->driContextPriv->driverPrivate);
+
+   rmesa = (r128ContextPtr) dPriv->driContextPriv->driverPrivate;
+
+   if ( R128_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "\n********************************\n" );
+      fprintf( stderr, "\n%s( %p )\n\n",
+	       __FUNCTION__, rmesa->glCtx );
+      fflush( stderr );
+   }
+
+   FLUSH_BATCH( rmesa );
+
+   LOCK_HARDWARE( rmesa );
+
+   /* Throttle the frame rate -- only allow one pending swap buffers
+    * request at a time.
+    */
+   if ( !r128WaitForFrameCompletion( rmesa ) ) {
+      rmesa->hardwareWentIdle = 1;
+   } else {
+      rmesa->hardwareWentIdle = 0;
+   }
+
+   UNLOCK_HARDWARE( rmesa );
+   driWaitForVBlank( dPriv, &rmesa->vbl_seq, rmesa->vblank_flags, &missed_target );
+   LOCK_HARDWARE( rmesa );
+
+   nbox = dPriv->numClipRects;	/* must be in locked region */
+
+   for ( i = 0 ; i < nbox ; ) {
+      GLint nr = MIN2( i + R128_NR_SAREA_CLIPRECTS , nbox );
+      XF86DRIClipRectPtr box = dPriv->pClipRects;
+      XF86DRIClipRectPtr b = rmesa->sarea->boxes;
+      GLint n = 0;
+
+      for ( ; i < nr ; i++ ) {
+	 *b++ = *(XF86DRIClipRectRec *)&box[i];
+	 n++;
+      }
+      rmesa->sarea->nbox = n;
+
+      ret = drmCommandNone( rmesa->driFd, DRM_R128_SWAP );
+
+      if ( ret ) {
+	 UNLOCK_HARDWARE( rmesa );
+	 fprintf( stderr, "DRM_R128_SWAP: return = %d\n", ret );
+	 exit( 1 );
+      }
+   }
+
+   if ( R128_DEBUG & DEBUG_ALWAYS_SYNC ) {
+      i = 0;
+      do {
+         ret = drmCommandNone(rmesa->driFd, DRM_R128_CCE_IDLE);
+      } while ( ret && errno == EBUSY && i++ < R128_IDLE_RETRY );
+   }
+
+   UNLOCK_HARDWARE( rmesa );
+
+   rmesa->new_state |= R128_NEW_CONTEXT;
+   rmesa->dirty |= (R128_UPLOAD_CONTEXT |
+		    R128_UPLOAD_MASKS |
+		    R128_UPLOAD_CLIPRECTS);
+
+#if ENABLE_PERF_BOXES
+   /* Log the performance counters if necessary */
+   r128PerformanceCounters( rmesa );
+#endif
+}
+
+void r128PageFlip( const __DRIdrawablePrivate *dPriv )
+{
+   r128ContextPtr rmesa;
+   GLint ret;
+   GLboolean missed_target;
+
+   assert(dPriv);
+   assert(dPriv->driContextPriv);
+   assert(dPriv->driContextPriv->driverPrivate);
+
+   rmesa = (r128ContextPtr) dPriv->driContextPriv->driverPrivate;
+
+   if ( R128_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "\n%s( %p ): page=%d\n\n",
+	       __FUNCTION__, rmesa->glCtx, rmesa->sarea->pfCurrentPage );
+   }
+
+   FLUSH_BATCH( rmesa );
+
+   LOCK_HARDWARE( rmesa );
+
+   /* Throttle the frame rate -- only allow one pending swap buffers
+    * request at a time.
+    */
+   if ( !r128WaitForFrameCompletion( rmesa ) ) {
+      rmesa->hardwareWentIdle = 1;
+   } else {
+      rmesa->hardwareWentIdle = 0;
+   }
+
+   UNLOCK_HARDWARE( rmesa );
+   driWaitForVBlank( dPriv, &rmesa->vbl_seq, rmesa->vblank_flags, &missed_target );
+   LOCK_HARDWARE( rmesa );
+
+   /* The kernel will have been initialized to perform page flipping
+    * on a swapbuffers ioctl.
+    */
+   ret = drmCommandNone( rmesa->driFd, DRM_R128_FLIP );
+
+   UNLOCK_HARDWARE( rmesa );
+
+   if ( ret ) {
+      fprintf( stderr, "DRM_R128_FLIP: return = %d\n", ret );
+      exit( 1 );
+   }
+
+   if ( rmesa->sarea->pfCurrentPage == 1 ) {
+	 rmesa->drawOffset = rmesa->r128Screen->frontOffset;
+	 rmesa->drawPitch  = rmesa->r128Screen->frontPitch;
+   } else {
+	 rmesa->drawOffset = rmesa->r128Screen->backOffset;
+	 rmesa->drawPitch  = rmesa->r128Screen->backPitch;
+   }
+
+   rmesa->setup.dst_pitch_offset_c = (((rmesa->drawPitch/8) << 21) |
+				      (rmesa->drawOffset >> 5));
+   rmesa->new_state |= R128_NEW_WINDOW;
+
+   /* FIXME: Do we need this anymore? */
+   rmesa->new_state |= R128_NEW_CONTEXT;
+   rmesa->dirty |= (R128_UPLOAD_CONTEXT |
+		    R128_UPLOAD_MASKS |
+		    R128_UPLOAD_CLIPRECTS);
+
+#if ENABLE_PERF_BOXES
+   /* Log the performance counters if necessary */
+   r128PerformanceCounters( rmesa );
+#endif
+}
+
+
+/* ================================================================
+ * Buffer clear
+ */
+
+static void r128DDClear( GLcontext *ctx, GLbitfield mask, GLboolean all,
+			 GLint cx, GLint cy, GLint cw, GLint ch )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   __DRIdrawablePrivate *dPriv = rmesa->driDrawable;
+   drmR128Clear clear;
+   GLuint flags = 0;
+   GLint i;
+   GLint ret;
+
+   if ( R128_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s:\n", __FUNCTION__ );
+   }
+
+   FLUSH_BATCH( rmesa );
+
+   /* The only state change we care about here is the RGBA colormask
+    * We'll just update that state, if needed.  If we do more then
+    * there's some strange side-effects that the conformance tests find.
+    */
+   if ( rmesa->new_state & R128_NEW_MASKS) {
+      const GLuint save_state = rmesa->new_state;
+      rmesa->new_state = R128_NEW_MASKS;
+      r128DDUpdateHWState( ctx );
+      rmesa->new_state = save_state & ~R128_NEW_MASKS;
+   }
+
+   if ( mask & DD_FRONT_LEFT_BIT ) {
+      flags |= DRM_R128_FRONT_BUFFER;
+      mask &= ~DD_FRONT_LEFT_BIT;
+   }
+
+   if ( mask & DD_BACK_LEFT_BIT ) {
+      flags |= DRM_R128_BACK_BUFFER;
+      mask &= ~DD_BACK_LEFT_BIT;
+   }
+
+   if ( ( mask & DD_DEPTH_BIT ) && ctx->Depth.Mask ) {
+      flags |= DRM_R128_DEPTH_BUFFER;
+      mask &= ~DD_DEPTH_BIT;
+   }
+#if 0
+   /* FIXME: Add stencil support */
+   if ( mask & DD_STENCIL_BIT ) {
+      flags |= DRM_R128_DEPTH_BUFFER;
+      mask &= ~DD_STENCIL_BIT;
+   }
+#endif
+
+   if ( flags ) {
+
+      /* Flip top to bottom */
+      cx += dPriv->x;
+      cy  = dPriv->y + dPriv->h - cy - ch;
+
+      LOCK_HARDWARE( rmesa );
+
+      /* FIXME: Do we actually need this?
+       */
+      if ( rmesa->dirty & ~R128_UPLOAD_CLIPRECTS ) {
+	 r128EmitHwStateLocked( rmesa );
+      }
+
+      for ( i = 0 ; i < rmesa->numClipRects ; ) {
+	 GLint nr = MIN2( i + R128_NR_SAREA_CLIPRECTS , rmesa->numClipRects );
+	 XF86DRIClipRectPtr box = rmesa->pClipRects;
+	 XF86DRIClipRectPtr b = rmesa->sarea->boxes;
+	 GLint n = 0;
+
+	 if ( !all ) {
+	    for ( ; i < nr ; i++ ) {
+	       GLint x = box[i].x1;
+	       GLint y = box[i].y1;
+	       GLint w = box[i].x2 - x;
+	       GLint h = box[i].y2 - y;
+
+	       if ( x < cx ) w -= cx - x, x = cx;
+	       if ( y < cy ) h -= cy - y, y = cy;
+	       if ( x + w > cx + cw ) w = cx + cw - x;
+	       if ( y + h > cy + ch ) h = cy + ch - y;
+	       if ( w <= 0 ) continue;
+	       if ( h <= 0 ) continue;
+
+	       b->x1 = x;
+	       b->y1 = y;
+	       b->x2 = x + w;
+	       b->y2 = y + h;
+	       b++;
+	       n++;
+	    }
+	 } else {
+	    for ( ; i < nr ; i++ ) {
+	       *b++ = *(XF86DRIClipRectPtr)&box[i];
+	       n++;
+	    }
+	 }
+
+	 rmesa->sarea->nbox = n;
+
+	 if ( R128_DEBUG & DEBUG_VERBOSE_IOCTL ) {
+	    fprintf( stderr,
+		     "DRM_R128_CLEAR: flag 0x%x color %x depth %x nbox %d\n",
+		     flags,
+		     (GLuint)rmesa->ClearColor,
+		     (GLuint)rmesa->ClearDepth,
+		     rmesa->sarea->nbox );
+	 }
+
+         clear.flags = flags;
+         clear.clear_color = rmesa->ClearColor;
+         clear.clear_depth = rmesa->ClearDepth;
+         clear.color_mask = rmesa->setup.plane_3d_mask_c;
+         clear.depth_mask = ~0;
+
+         ret = drmCommandWrite( rmesa->driFd, DRM_R128_CLEAR,
+                                &clear, sizeof(drmR128Clear) );
+
+	 if ( ret ) {
+	    UNLOCK_HARDWARE( rmesa );
+	    fprintf( stderr, "DRM_R128_CLEAR: return = %d\n", ret );
+	    exit( 1 );
+	 }
+      }
+
+      UNLOCK_HARDWARE( rmesa );
+
+      rmesa->dirty |= R128_UPLOAD_CLIPRECTS;
+   }
+
+   if ( mask )
+      _swrast_Clear( ctx, mask, all, cx, cy, cw, ch );
+}
+
+
+/* ================================================================
+ * Depth spans, pixels
+ */
+
+void r128WriteDepthSpanLocked( r128ContextPtr rmesa,
+			       GLuint n, GLint x, GLint y,
+			       const GLdepth depth[],
+			       const GLubyte mask[] )
+{
+   XF86DRIClipRectPtr pbox = rmesa->pClipRects;
+   drmR128Depth d;
+   int nbox = rmesa->numClipRects;
+   int fd = rmesa->driScreen->fd;
+   int i;
+
+   if ( !nbox || !n ) {
+      return;
+   }
+   if ( nbox >= R128_NR_SAREA_CLIPRECTS ) {
+      rmesa->dirty |= R128_UPLOAD_CLIPRECTS;
+   }
+
+   if ( !(rmesa->dirty & R128_UPLOAD_CLIPRECTS) )
+   {
+      if ( nbox < 3 ) {
+	 rmesa->sarea->nbox = 0;
+      } else {
+	 rmesa->sarea->nbox = nbox;
+      }
+
+      d.func = DRM_R128_WRITE_SPAN;
+      d.n = n;
+      d.x = (int*)&x;
+      d.y = (int*)&y;
+      d.buffer = (unsigned int *)depth;
+      d.mask = (unsigned char *)mask;
+
+      drmCommandWrite( fd, DRM_R128_DEPTH, &d, sizeof(drmR128Depth));
+
+   }
+   else
+   {
+      for (i = 0 ; i < nbox ; ) {
+	 int nr = MIN2( i + R128_NR_SAREA_CLIPRECTS, nbox );
+	 XF86DRIClipRectPtr b = rmesa->sarea->boxes;
+
+	 rmesa->sarea->nbox = nr - i;
+	 for ( ; i < nr ; i++) {
+	    *b++ = pbox[i];
+	 }
+
+	 rmesa->sarea->dirty |= R128_UPLOAD_CLIPRECTS;
+
+         d.func = DRM_R128_WRITE_SPAN;
+         d.n = n;
+         d.x = (int*)&x;
+         d.y = (int*)&y;
+         d.buffer = (unsigned int *)depth;
+         d.mask = (unsigned char *)mask;
+
+         drmCommandWrite( fd, DRM_R128_DEPTH, &d, sizeof(drmR128Depth));
+      }
+   }
+
+   rmesa->dirty &= ~R128_UPLOAD_CLIPRECTS;
+}
+
+void r128WriteDepthPixelsLocked( r128ContextPtr rmesa, GLuint n,
+				 const GLint x[], const GLint y[],
+				 const GLdepth depth[],
+				 const GLubyte mask[] )
+{
+   XF86DRIClipRectPtr pbox = rmesa->pClipRects;
+   drmR128Depth d;
+   int nbox = rmesa->numClipRects;
+   int fd = rmesa->driScreen->fd;
+   int i;
+
+   if ( !nbox || !n ) {
+      return;
+   }
+   if ( nbox >= R128_NR_SAREA_CLIPRECTS ) {
+      rmesa->dirty |= R128_UPLOAD_CLIPRECTS;
+   }
+
+   if ( !(rmesa->dirty & R128_UPLOAD_CLIPRECTS) )
+   {
+      if ( nbox < 3 ) {
+	 rmesa->sarea->nbox = 0;
+      } else {
+	 rmesa->sarea->nbox = nbox;
+      }
+
+      d.func = DRM_R128_WRITE_PIXELS;
+      d.n = n;
+      d.x = (int*)&x;
+      d.y = (int*)&y;
+      d.buffer = (unsigned int *)depth;
+      d.mask = (unsigned char *)mask;
+
+      drmCommandWrite( fd, DRM_R128_DEPTH, &d, sizeof(drmR128Depth));
+   }
+   else
+   {
+      for (i = 0 ; i < nbox ; ) {
+	 int nr = MIN2( i + R128_NR_SAREA_CLIPRECTS, nbox );
+	 XF86DRIClipRectPtr b = rmesa->sarea->boxes;
+
+	 rmesa->sarea->nbox = nr - i;
+	 for ( ; i < nr ; i++) {
+	    *b++ = pbox[i];
+	 }
+
+	 rmesa->sarea->dirty |= R128_UPLOAD_CLIPRECTS;
+
+         d.func = DRM_R128_WRITE_PIXELS;
+         d.n = n;
+         d.x = (int*)&x;
+         d.y = (int*)&y;
+         d.buffer = (unsigned int *)depth;
+         d.mask = (unsigned char *)mask;
+
+         drmCommandWrite( fd, DRM_R128_DEPTH, &d, sizeof(drmR128Depth));
+      }
+   }
+
+   rmesa->dirty &= ~R128_UPLOAD_CLIPRECTS;
+}
+
+void r128ReadDepthSpanLocked( r128ContextPtr rmesa,
+			      GLuint n, GLint x, GLint y )
+{
+   XF86DRIClipRectPtr pbox = rmesa->pClipRects;
+   drmR128Depth d;
+   int nbox = rmesa->numClipRects;
+   int fd = rmesa->driScreen->fd;
+   int i;
+
+   if ( !nbox || !n ) {
+      return;
+   }
+   if ( nbox >= R128_NR_SAREA_CLIPRECTS ) {
+      rmesa->dirty |= R128_UPLOAD_CLIPRECTS;
+   }
+
+   if ( !(rmesa->dirty & R128_UPLOAD_CLIPRECTS) )
+   {
+      if ( nbox < 3 ) {
+	 rmesa->sarea->nbox = 0;
+      } else {
+	 rmesa->sarea->nbox = nbox;
+      }
+
+      d.func = DRM_R128_READ_SPAN;
+      d.n = n;
+      d.x = (int*)&x;
+      d.y = (int*)&y;
+      d.buffer = NULL;
+      d.mask = NULL;
+
+      drmCommandWrite( fd, DRM_R128_DEPTH, &d, sizeof(drmR128Depth));
+   }
+   else
+   {
+      for (i = 0 ; i < nbox ; ) {
+	 int nr = MIN2( i + R128_NR_SAREA_CLIPRECTS, nbox );
+	 XF86DRIClipRectPtr b = rmesa->sarea->boxes;
+
+	 rmesa->sarea->nbox = nr - i;
+	 for ( ; i < nr ; i++) {
+	    *b++ = pbox[i];
+	 }
+
+	 rmesa->sarea->dirty |= R128_UPLOAD_CLIPRECTS;
+
+         d.func = DRM_R128_READ_SPAN;
+         d.n = n;
+         d.x = (int*)&x;
+         d.y = (int*)&y;
+         d.buffer = NULL;
+         d.mask = NULL;
+
+         drmCommandWrite( fd, DRM_R128_DEPTH, &d, sizeof(drmR128Depth));
+      }
+   }
+
+   rmesa->dirty &= ~R128_UPLOAD_CLIPRECTS;
+}
+
+void r128ReadDepthPixelsLocked( r128ContextPtr rmesa, GLuint n,
+				const GLint x[], const GLint y[] )
+{
+   XF86DRIClipRectPtr pbox = rmesa->pClipRects;
+   drmR128Depth d;
+   int nbox = rmesa->numClipRects;
+   int fd = rmesa->driScreen->fd;
+   int i;
+
+   if ( !nbox || !n ) {
+      return;
+   }
+   if ( nbox >= R128_NR_SAREA_CLIPRECTS ) {
+      rmesa->dirty |= R128_UPLOAD_CLIPRECTS;
+   }
+
+   if ( !(rmesa->dirty & R128_UPLOAD_CLIPRECTS) )
+   {
+      if ( nbox < 3 ) {
+	 rmesa->sarea->nbox = 0;
+      } else {
+	 rmesa->sarea->nbox = nbox;
+      }
+
+      d.func = DRM_R128_READ_PIXELS;
+      d.n = n;
+      d.x = (int*)&x;
+      d.y = (int*)&y;
+      d.buffer = NULL;
+      d.mask = NULL;
+
+      drmCommandWrite( fd, DRM_R128_DEPTH, &d, sizeof(drmR128Depth));
+   }
+   else
+   {
+      for (i = 0 ; i < nbox ; ) {
+	 int nr = MIN2( i + R128_NR_SAREA_CLIPRECTS, nbox );
+	 XF86DRIClipRectPtr b = rmesa->sarea->boxes;
+
+	 rmesa->sarea->nbox = nr - i;
+	 for ( ; i < nr ; i++) {
+	    *b++ = pbox[i];
+	 }
+
+	 rmesa->sarea->dirty |= R128_UPLOAD_CLIPRECTS;
+
+         d.func = DRM_R128_READ_PIXELS;
+         d.n = n;
+         d.x = (int*)&x;
+         d.y = (int*)&y;
+         d.buffer = NULL;
+         d.mask = NULL;
+
+         drmCommandWrite( fd, DRM_R128_DEPTH, &d, sizeof(drmR128Depth));
+      }
+   }
+
+   rmesa->dirty &= ~R128_UPLOAD_CLIPRECTS;
+}
+
+
+void r128WaitForIdleLocked( r128ContextPtr rmesa )
+{
+    int fd = rmesa->r128Screen->driScreen->fd;
+    int to = 0;
+    int ret, i;
+
+    do {
+        i = 0;
+        do {
+            ret = drmCommandNone( fd, DRM_R128_CCE_IDLE);
+        } while ( ret && errno == EBUSY && i++ < R128_IDLE_RETRY );
+    } while ( ( ret == -EBUSY ) && ( to++ < R128_TIMEOUT ) );
+
+    if ( ret < 0 ) {
+        drmCommandNone( fd, DRM_R128_CCE_RESET);
+	UNLOCK_HARDWARE( rmesa );
+	fprintf( stderr, "Error: Rage 128 timed out... exiting\n" );
+	exit( -1 );
+    }
+}
+
+void r128DDInitIoctlFuncs( GLcontext *ctx )
+{
+    ctx->Driver.Clear = r128DDClear;
+}
diff --git a/src/mesa/drivers/dri/r128/r128_ioctl.h b/src/mesa/drivers/dri/r128/r128_ioctl.h
new file mode 100644
index 00000000000..b0f6b363a5f
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_ioctl.h
@@ -0,0 +1,146 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_ioctl.h,v 1.6 2002/12/16 16:18:53 dawes Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Gareth Hughes <[email protected]>
+ *
+ */
+
+#ifndef __R128_IOCTL_H__
+#define __R128_IOCTL_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+#include "r128_dri.h"
+#include "r128_reg.h"
+#include "r128_lock.h"
+
+#include "xf86drm.h"
+#include "r128_common.h"
+
+#define R128_BUFFER_MAX_DWORDS	(R128_BUFFER_SIZE / sizeof(CARD32))
+
+
+extern drmBufPtr r128GetBufferLocked( r128ContextPtr rmesa );
+extern void r128FlushVerticesLocked( r128ContextPtr rmesa );
+
+static __inline void *r128AllocDmaLow( r128ContextPtr rmesa, int bytes )
+{
+   CARD32 *head;
+
+   if ( !rmesa->vert_buf ) {
+      LOCK_HARDWARE( rmesa );
+      rmesa->vert_buf = r128GetBufferLocked( rmesa );
+      UNLOCK_HARDWARE( rmesa );
+   } else if ( rmesa->vert_buf->used + bytes > rmesa->vert_buf->total ) {
+      LOCK_HARDWARE( rmesa );
+      r128FlushVerticesLocked( rmesa );
+      rmesa->vert_buf = r128GetBufferLocked( rmesa );
+      UNLOCK_HARDWARE( rmesa );
+   }
+
+   head = (CARD32 *)((char *)rmesa->vert_buf->address + rmesa->vert_buf->used);
+   rmesa->vert_buf->used += bytes;
+   return head;
+}
+
+extern void r128FireBlitLocked( r128ContextPtr rmesa, drmBufPtr buffer,
+				GLint offset, GLint pitch, GLint format,
+				GLint x, GLint y, GLint width, GLint height );
+
+extern void r128WriteDepthSpanLocked( r128ContextPtr rmesa,
+				      GLuint n, GLint x, GLint y,
+				      const GLdepth depth[],
+				      const GLubyte mask[] );
+extern void r128WriteDepthPixelsLocked( r128ContextPtr rmesa, GLuint n,
+					const GLint x[], const GLint y[],
+					const GLdepth depth[],
+					const GLubyte mask[] );
+extern void r128ReadDepthSpanLocked( r128ContextPtr rmesa,
+				     GLuint n, GLint x, GLint y );
+extern void r128ReadDepthPixelsLocked( r128ContextPtr rmesa, GLuint n,
+				       const GLint x[], const GLint y[] );
+
+extern void r128CopyBuffer( const __DRIdrawablePrivate *dPriv );
+extern void r128PageFlip( const __DRIdrawablePrivate *dPriv );
+void r128WaitForVBlank( r128ContextPtr rmesa );
+
+extern void r128WaitForIdleLocked( r128ContextPtr rmesa );
+
+
+extern void r128DDInitIoctlFuncs( GLcontext *ctx );
+
+
+/* ================================================================
+ * Helper macros:
+ */
+
+#define FLUSH_BATCH( rmesa )						\
+do {									\
+   if ( R128_DEBUG & DEBUG_VERBOSE_IOCTL )				\
+      fprintf( stderr, "FLUSH_BATCH in %s\n", __FUNCTION__ );		\
+   if ( rmesa->vert_buf ) {						\
+      r128FlushVertices( rmesa );					\
+   }									\
+} while (0)
+
+/* 64-bit align the next element address, and then make room for the
+ * next indexed prim packet header.
+ */
+#define ALIGN_NEXT_ELT( rmesa )						\
+do {									\
+   rmesa->next_elt = (GLushort *)					\
+      (((GLuint)rmesa->next_elt + 7) & ~0x7);				\
+   rmesa->next_elt = (GLushort *)					\
+      ((GLubyte *)rmesa->next_elt + R128_INDEX_PRIM_OFFSET);		\
+} while (0)
+
+#define r128FlushVertices( rmesa )					\
+do {									\
+   LOCK_HARDWARE( rmesa );						\
+   r128FlushVerticesLocked( rmesa );					\
+   UNLOCK_HARDWARE( rmesa );						\
+} while (0)
+
+#define r128FlushElts( rmesa )						\
+do {									\
+   LOCK_HARDWARE( rmesa );						\
+   r128FlushEltsLocked( rmesa );					\
+   UNLOCK_HARDWARE( rmesa );						\
+} while (0)
+
+#define r128WaitForIdle( rmesa )					\
+   do {									\
+      LOCK_HARDWARE( rmesa );						\
+      r128WaitForIdleLocked( rmesa );					\
+      UNLOCK_HARDWARE( rmesa );						\
+   } while (0)
+
+#endif
+#endif /* __R128_IOCTL_H__ */
diff --git a/src/mesa/drivers/dri/r128/r128_lock.c b/src/mesa/drivers/dri/r128/r128_lock.c
new file mode 100644
index 00000000000..12eb0128b03
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_lock.c
@@ -0,0 +1,123 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_lock.c,v 1.5 2002/10/30 12:51:38 alanh Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Gareth Hughes <[email protected]>
+ *
+ */
+
+#include "r128_context.h"
+#include "r128_lock.h"
+#include "r128_tex.h"
+#include "r128_state.h"
+
+#if DEBUG_LOCKING
+char *prevLockFile = NULL;
+int prevLockLine = 0;
+#endif
+
+
+/* Turn on/off page flipping according to the flags in the sarea:
+ */
+static void
+r128UpdatePageFlipping( r128ContextPtr rmesa )
+{
+   int use_back;
+
+   rmesa->doPageFlip = rmesa->sarea->pfAllowPageFlip;
+
+   use_back = (rmesa->glCtx->Color._DrawDestMask == BACK_LEFT_BIT);
+   use_back ^= (rmesa->sarea->pfCurrentPage == 1);
+
+   if ( R128_DEBUG & DEBUG_VERBOSE_API )
+      fprintf(stderr, "%s allow %d current %d\n", __FUNCTION__, 
+	      rmesa->doPageFlip,
+	      rmesa->sarea->pfCurrentPage );
+
+   if ( use_back ) {
+	 rmesa->drawOffset = rmesa->r128Screen->backOffset;
+	 rmesa->drawPitch  = rmesa->r128Screen->backPitch;
+   } else {
+	 rmesa->drawOffset = rmesa->r128Screen->frontOffset;
+	 rmesa->drawPitch  = rmesa->r128Screen->frontPitch;
+   }
+
+   rmesa->setup.dst_pitch_offset_c = (((rmesa->drawPitch/8) << 21) |
+                                      (rmesa->drawOffset >> 5));
+   rmesa->new_state |= R128_NEW_WINDOW;
+}
+
+/* Update the hardware state.  This is called if another context has
+ * grabbed the hardware lock, which includes the X server.  This
+ * function also updates the driver's window state after the X server
+ * moves, resizes or restacks a window -- the change will be reflected
+ * in the drawable position and clip rects.  Since the X server grabs
+ * the hardware lock when it changes the window state, this routine will
+ * automatically be called after such a change.
+ */
+void r128GetLock( r128ContextPtr rmesa, GLuint flags )
+{
+   __DRIdrawablePrivate *dPriv = rmesa->driDrawable;
+   __DRIscreenPrivate *sPriv = rmesa->driScreen;
+   R128SAREAPrivPtr sarea = rmesa->sarea;
+   int i;
+
+   drmGetLock( rmesa->driFd, rmesa->hHWContext, flags );
+
+   /* The window might have moved, so we might need to get new clip
+    * rects.
+    *
+    * NOTE: This releases and regrabs the hw lock to allow the X server
+    * to respond to the DRI protocol request for new drawable info.
+    * Since the hardware state depends on having the latest drawable
+    * clip rects, all state checking must be done _after_ this call.
+    */
+   DRI_VALIDATE_DRAWABLE_INFO( sPriv, dPriv );
+
+   if ( rmesa->lastStamp != dPriv->lastStamp ) {
+      r128UpdatePageFlipping( rmesa );
+      rmesa->lastStamp = dPriv->lastStamp;
+      rmesa->new_state |= R128_NEW_CLIP;
+      rmesa->SetupNewInputs = ~0;
+   }
+
+   rmesa->dirty |= R128_UPLOAD_CONTEXT | R128_UPLOAD_CLIPRECTS;
+
+   rmesa->numClipRects = dPriv->numClipRects;
+   rmesa->pClipRects = dPriv->pClipRects;
+
+   if ( sarea->ctxOwner != rmesa->hHWContext ) {
+      sarea->ctxOwner = rmesa->hHWContext;
+      rmesa->dirty = R128_UPLOAD_ALL;
+   }
+
+   for ( i = 0 ; i < rmesa->nr_heaps ; i++ ) {
+      DRI_AGE_TEXTURES( rmesa->texture_heaps[i] );
+   }
+}
diff --git a/src/mesa/drivers/dri/r128/r128_lock.h b/src/mesa/drivers/dri/r128/r128_lock.h
new file mode 100644
index 00000000000..e734ba29def
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_lock.h
@@ -0,0 +1,111 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_lock.h,v 1.4 2001/01/08 01:07:21 martin Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ *
+ */
+
+#ifndef __R128_LOCK_H__
+#define __R128_LOCK_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+extern void r128GetLock( r128ContextPtr rmesa, GLuint flags );
+
+/* Turn DEBUG_LOCKING on to find locking conflicts.
+ */
+#define DEBUG_LOCKING	0
+
+#if DEBUG_LOCKING
+extern char *prevLockFile;
+extern int prevLockLine;
+
+#define DEBUG_LOCK()							\
+   do {									\
+      prevLockFile = (__FILE__);					\
+      prevLockLine = (__LINE__);					\
+   } while (0)
+
+#define DEBUG_RESET()							\
+   do {									\
+      prevLockFile = 0;							\
+      prevLockLine = 0;							\
+   } while (0)
+
+#define DEBUG_CHECK_LOCK()						\
+   do {									\
+      if ( prevLockFile ) {						\
+	 fprintf( stderr,						\
+		  "LOCK SET!\n\tPrevious %s:%d\n\tCurrent: %s:%d\n",	\
+		  prevLockFile, prevLockLine, __FILE__, __LINE__ );	\
+	 exit( 1 );							\
+      }									\
+   } while (0)
+
+#else
+
+#define DEBUG_LOCK()
+#define DEBUG_RESET()
+#define DEBUG_CHECK_LOCK()
+
+#endif
+
+/*
+ * !!! We may want to separate locks from locks with validation.  This
+ * could be used to improve performance for those things commands that
+ * do not do any drawing !!!
+ */
+
+/* Lock the hardware and validate our state.
+ */
+#define LOCK_HARDWARE( rmesa )						\
+   do {									\
+      char __ret = 0;							\
+      DEBUG_CHECK_LOCK();						\
+      DRM_CAS( rmesa->driHwLock, rmesa->hHWContext,			\
+	       (DRM_LOCK_HELD | rmesa->hHWContext), __ret );		\
+      if ( __ret )							\
+	 r128GetLock( rmesa, 0 );					\
+      DEBUG_LOCK();							\
+   } while (0)
+
+/* Unlock the hardware.
+ */
+#define UNLOCK_HARDWARE( rmesa )					\
+   do {									\
+      DRM_UNLOCK( rmesa->driFd,						\
+		  rmesa->driHwLock,					\
+		  rmesa->hHWContext );					\
+      DEBUG_RESET();							\
+   } while (0)
+
+#endif
+#endif /* __R128_LOCK_H__ */
diff --git a/src/mesa/drivers/dri/r128/r128_screen.c b/src/mesa/drivers/dri/r128/r128_screen.c
new file mode 100644
index 00000000000..8a8fc252177
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_screen.c
@@ -0,0 +1,357 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_screen.c,v 1.8 2002/12/16 16:18:53 dawes Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Gareth Hughes <[email protected]>
+ *   Kevin E. Martin <[email protected]>
+ *
+ */
+
+#include "r128_dri.h"
+
+#include "r128_context.h"
+#include "r128_ioctl.h"
+#include "r128_tris.h"
+#include "r128_vb.h"
+
+#include "context.h"
+#include "imports.h"
+
+#include "utils.h"
+#include "vblank.h"
+
+#ifndef _SOLO
+#include "glxextensions.h"
+#endif
+
+#if 1
+/* Including xf86PciInfo.h introduces a bunch of errors...
+ */
+#define PCI_CHIP_RAGE128LE	0x4C45
+#define PCI_CHIP_RAGE128LF	0x4C46
+#define PCI_CHIP_RAGE128PD	0x5044
+#define PCI_CHIP_RAGE128PF	0x5046
+#define PCI_CHIP_RAGE128PR	0x5052
+#define PCI_CHIP_RAGE128RE	0x5245
+#define PCI_CHIP_RAGE128RF	0x5246
+#define PCI_CHIP_RAGE128RK	0x524B
+#define PCI_CHIP_RAGE128RL	0x524C
+#endif
+
+
+/* Create the device specific screen private data struct.
+ */
+static r128ScreenPtr
+r128CreateScreen( __DRIscreenPrivate *sPriv )
+{
+   r128ScreenPtr r128Screen;
+   R128DRIPtr r128DRIPriv = (R128DRIPtr)sPriv->pDevPriv;
+
+   if ( ! driCheckDriDdxDrmVersions( sPriv, "Rage128", 4, 0, 4, 0, 2, 2 ) )
+      return NULL;
+
+   /* Allocate the private area */
+   r128Screen = (r128ScreenPtr) CALLOC( sizeof(*r128Screen) );
+   if ( !r128Screen ) return NULL;
+
+   /* This is first since which regions we map depends on whether or
+    * not we are using a PCI card.
+    */
+   r128Screen->IsPCI = r128DRIPriv->IsPCI;
+   r128Screen->sarea_priv_offset = r128DRIPriv->sarea_priv_offset;
+   
+   if (sPriv->drmMinor >= 3) {
+      drmR128GetParam gp;
+      int ret;
+
+      gp.param = R128_PARAM_IRQ_NR;
+      gp.value = &r128Screen->irq;
+
+      ret = drmCommandWriteRead( sPriv->fd, DRM_R128_GETPARAM,
+				    &gp, sizeof(gp));
+      if (ret) {
+         fprintf(stderr, "drmR128GetParam (R128_PARAM_IRQ_NR): %d\n", ret);
+         FREE( r128Screen );
+         return NULL;
+      }
+   }
+
+   r128Screen->mmio.handle = r128DRIPriv->registerHandle;
+   r128Screen->mmio.size   = r128DRIPriv->registerSize;
+   if ( drmMap( sPriv->fd,
+		r128Screen->mmio.handle,
+		r128Screen->mmio.size,
+		(drmAddressPtr)&r128Screen->mmio.map ) ) {
+      FREE( r128Screen );
+      return NULL;
+   }
+
+   r128Screen->buffers = drmMapBufs( sPriv->fd );
+   if ( !r128Screen->buffers ) {
+      drmUnmap( (drmAddress)r128Screen->mmio.map, r128Screen->mmio.size );
+      FREE( r128Screen );
+      return NULL;
+   }
+
+   if ( !r128Screen->IsPCI ) {
+      r128Screen->agpTextures.handle = r128DRIPriv->agpTexHandle;
+      r128Screen->agpTextures.size   = r128DRIPriv->agpTexMapSize;
+      if ( drmMap( sPriv->fd,
+		   r128Screen->agpTextures.handle,
+		   r128Screen->agpTextures.size,
+		   (drmAddressPtr)&r128Screen->agpTextures.map ) ) {
+	 drmUnmapBufs( r128Screen->buffers );
+	 drmUnmap( (drmAddress)r128Screen->mmio.map, r128Screen->mmio.size );
+	 FREE( r128Screen );
+	 return NULL;
+      }
+   }
+
+   switch ( r128DRIPriv->deviceID ) {
+   case PCI_CHIP_RAGE128RE:
+   case PCI_CHIP_RAGE128RF:
+   case PCI_CHIP_RAGE128RK:
+   case PCI_CHIP_RAGE128RL:
+      r128Screen->chipset = R128_CARD_TYPE_R128;
+      break;
+   case PCI_CHIP_RAGE128PD:
+   case PCI_CHIP_RAGE128PF:
+      r128Screen->chipset = R128_CARD_TYPE_R128_PRO;
+      break;
+   case PCI_CHIP_RAGE128LE:
+   case PCI_CHIP_RAGE128LF:
+      r128Screen->chipset = R128_CARD_TYPE_R128_MOBILITY;
+      break;
+   default:
+      r128Screen->chipset = R128_CARD_TYPE_R128;
+      break;
+   }
+
+   r128Screen->cpp = r128DRIPriv->bpp / 8;
+   r128Screen->AGPMode = r128DRIPriv->AGPMode;
+
+   r128Screen->frontOffset	= r128DRIPriv->frontOffset;
+   r128Screen->frontPitch	= r128DRIPriv->frontPitch;
+   r128Screen->backOffset	= r128DRIPriv->backOffset;
+   r128Screen->backPitch	= r128DRIPriv->backPitch;
+   r128Screen->depthOffset	= r128DRIPriv->depthOffset;
+   r128Screen->depthPitch	= r128DRIPriv->depthPitch;
+   r128Screen->spanOffset	= r128DRIPriv->spanOffset;
+
+   r128Screen->texOffset[R128_CARD_HEAP] = r128DRIPriv->textureOffset;
+   r128Screen->texSize[R128_CARD_HEAP] = r128DRIPriv->textureSize;
+   r128Screen->logTexGranularity[R128_CARD_HEAP] = r128DRIPriv->log2TexGran;
+
+   if ( r128Screen->IsPCI ) {
+      r128Screen->numTexHeaps = R128_NR_TEX_HEAPS - 1;
+      r128Screen->texOffset[R128_AGP_HEAP] = 0;
+      r128Screen->texSize[R128_AGP_HEAP] = 0;
+      r128Screen->logTexGranularity[R128_AGP_HEAP] = 0;
+   } else {
+      r128Screen->numTexHeaps = R128_NR_TEX_HEAPS;
+      r128Screen->texOffset[R128_AGP_HEAP] =
+	 r128DRIPriv->agpTexOffset + R128_AGP_TEX_OFFSET;
+      r128Screen->texSize[R128_AGP_HEAP] = r128DRIPriv->agpTexMapSize;
+      r128Screen->logTexGranularity[R128_AGP_HEAP] =
+	 r128DRIPriv->log2AGPTexGran;
+   }
+
+   r128Screen->driScreen = sPriv;
+
+   return r128Screen;
+}
+
+/* Destroy the device specific screen private data struct.
+ */
+static void
+r128DestroyScreen( __DRIscreenPrivate *sPriv )
+{
+   r128ScreenPtr r128Screen = (r128ScreenPtr)sPriv->private;
+
+   if ( !r128Screen )
+      return;
+
+   if ( !r128Screen->IsPCI ) {
+      drmUnmap( (drmAddress)r128Screen->agpTextures.map,
+		r128Screen->agpTextures.size );
+   }
+   drmUnmapBufs( r128Screen->buffers );
+   drmUnmap( (drmAddress)r128Screen->mmio.map, r128Screen->mmio.size );
+
+   FREE( r128Screen );
+   sPriv->private = NULL;
+}
+
+
+/* Initialize the fullscreen mode.
+ */
+static GLboolean
+r128OpenCloseFullScreen( __DRIcontextPrivate *driContextPriv )
+{
+   return GL_TRUE;
+}
+
+
+/* Create and initialize the Mesa and driver specific pixmap buffer
+ * data.
+ */
+static GLboolean
+r128CreateBuffer( __DRIscreenPrivate *driScrnPriv,
+                  __DRIdrawablePrivate *driDrawPriv,
+                  const __GLcontextModes *mesaVis,
+                  GLboolean isPixmap )
+{
+   if (isPixmap) {
+      return GL_FALSE; /* not implemented */
+   }
+   else {
+      driDrawPriv->driverPrivate = (void *) 
+         _mesa_create_framebuffer( mesaVis,
+                                   GL_FALSE,  /* software depth buffer? */
+                                   mesaVis->stencilBits > 0,
+                                   mesaVis->accumRedBits > 0,
+                                   mesaVis->alphaBits > 0 );
+      return (driDrawPriv->driverPrivate != NULL);
+   }
+}
+
+
+static void
+r128DestroyBuffer(__DRIdrawablePrivate *driDrawPriv)
+{
+   _mesa_destroy_framebuffer((GLframebuffer *) (driDrawPriv->driverPrivate));
+}
+
+
+/* Copy the back color buffer to the front color buffer */
+static void
+r128SwapBuffers(__DRIdrawablePrivate *dPriv)
+{
+   if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) {
+      r128ContextPtr rmesa;
+      GLcontext *ctx;
+      rmesa = (r128ContextPtr) dPriv->driContextPriv->driverPrivate;
+      ctx = rmesa->glCtx;
+      if (ctx->Visual.doubleBufferMode) {
+         _mesa_notifySwapBuffers( ctx );  /* flush pending rendering comands */
+         if ( rmesa->doPageFlip ) {
+            r128PageFlip( dPriv );
+         }
+         else {
+            r128CopyBuffer( dPriv );
+         }
+      }
+   }
+   else {
+      /* XXX this shouldn't be an error but we can't handle it for now */
+      _mesa_problem(NULL, "%s: drawable has no context!", __FUNCTION__);
+   }
+}
+
+
+/* Initialize the driver specific screen private data.
+ */
+static GLboolean
+r128InitDriver( __DRIscreenPrivate *sPriv )
+{
+   sPriv->private = (void *) r128CreateScreen( sPriv );
+
+   if ( !sPriv->private ) {
+      r128DestroyScreen( sPriv );
+      return GL_FALSE;
+   }
+
+   return GL_TRUE;
+}
+
+
+#ifndef _SOLO
+/* This function is called by libGL.so as soon as libGL.so is loaded.
+ * This is where we register new extension functions with the dispatcher.
+ */
+void __driRegisterExtensions( void )
+{
+   PFNGLXENABLEEXTENSIONPROC glx_enable_extension;
+
+   if ( driCompareGLXAPIVersion( 20030317 ) >= 0 ) {
+      glx_enable_extension = (PFNGLXENABLEEXTENSIONPROC)
+	  glXGetProcAddress( "__glXEnableExtension" );
+
+      if ( glx_enable_extension != NULL ) {
+	 glx_enable_extension( "GLX_SGI_swap_control", GL_FALSE );
+	 glx_enable_extension( "GLX_SGI_video_sync", GL_FALSE );
+	 glx_enable_extension( "GLX_MESA_swap_control", GL_FALSE );
+      }
+   }
+}
+#endif
+
+static struct __DriverAPIRec r128API = {
+   .InitDriver      = r128InitDriver,
+   .DestroyScreen   = r128DestroyScreen,
+   .CreateContext   = r128CreateContext,
+   .DestroyContext  = r128DestroyContext,
+   .CreateBuffer    = r128CreateBuffer,
+   .DestroyBuffer   = r128DestroyBuffer,
+   .SwapBuffers     = r128SwapBuffers,
+   .MakeCurrent     = r128MakeCurrent,
+   .UnbindContext   = r128UnbindContext,
+   .OpenFullScreen  = r128OpenCloseFullScreen,
+   .CloseFullScreen = r128OpenCloseFullScreen,
+   .GetSwapInfo     = NULL,
+   .GetMSC          = driGetMSC32,
+   .WaitForMSC      = driWaitForMSC32,
+   .WaitForSBC      = NULL,
+   .SwapBuffersMSC  = NULL
+
+};
+
+
+/*
+ * This is the bootstrap function for the driver.
+ * The __driCreateScreen name is the symbol that libGL.so fetches.
+ * Return:  pointer to a __DRIscreenPrivate.
+ */
+#ifndef _SOLO
+void *__driCreateScreen(Display *dpy, int scrn, __DRIscreen *psc,
+                        int numConfigs, __GLXvisualConfig *config)
+{
+   __DRIscreenPrivate *psp;
+   psp = __driUtilCreateScreen(dpy, scrn, psc, numConfigs, config, &r128API);
+   return (void *) psp;
+}
+#else
+void *__driCreateScreen(struct DRIDriverRec *driver,
+                        struct DRIDriverContextRec *driverContext)
+{
+   __DRIscreenPrivate *psp;
+   psp = __driUtilCreateScreen(driver, driverContext, &r128API);
+   return (void *) psp;
+}
+#endif
diff --git a/src/mesa/drivers/dri/r128/r128_screen.h b/src/mesa/drivers/dri/r128/r128_screen.h
new file mode 100644
index 00000000000..4b1a57292d9
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_screen.h
@@ -0,0 +1,84 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_screen.h,v 1.7 2002/12/16 16:18:53 dawes Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Gareth Hughes <[email protected]>
+ *   Kevin E. Martin <[email protected]>
+ *
+ */
+
+#ifndef __R128_SCREEN_H__
+#define __R128_SCREEN_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+#include "r128_sarea.h"
+
+typedef struct {
+   drmHandle handle;			/* Handle to the DRM region */
+   drmSize size;			/* Size of the DRM region */
+   unsigned char *map;			/* Mapping of the DRM region */
+} r128RegionRec, *r128RegionPtr;
+
+typedef struct {
+
+   GLint chipset;
+   GLint cpp;
+   GLint IsPCI;				/* Current card is a PCI card */
+   GLint AGPMode;
+   unsigned int irq;			/* IRQ number (0 means none) */
+
+   GLuint frontOffset;
+   GLuint frontPitch;
+   GLuint backOffset;
+   GLuint backPitch;
+
+   GLuint depthOffset;
+   GLuint depthPitch;
+   GLuint spanOffset;
+
+    /* Shared texture data */
+   GLint numTexHeaps;
+   GLint texOffset[R128_NR_TEX_HEAPS];
+   GLint texSize[R128_NR_TEX_HEAPS];
+   GLint logTexGranularity[R128_NR_TEX_HEAPS];
+
+   r128RegionRec mmio;
+   r128RegionRec agpTextures;
+
+   drmBufMapPtr buffers;
+
+   __DRIscreenPrivate *driScreen;
+   unsigned int sarea_priv_offset;
+
+} r128ScreenRec, *r128ScreenPtr;
+
+
+#endif
+#endif /* __R128_SCREEN_H__ */
diff --git a/src/mesa/drivers/dri/r128/r128_span.c b/src/mesa/drivers/dri/r128/r128_span.c
new file mode 100644
index 00000000000..c6f31959423
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_span.c
@@ -0,0 +1,475 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_span.c,v 1.8 2002/10/30 12:51:39 alanh Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Gareth Hughes <[email protected]>
+ *   Keith Whitwell <[email protected]>
+ *   Kevin E. Martin <[email protected]>
+ *
+ */
+
+#include "r128_context.h"
+#include "r128_ioctl.h"
+#include "r128_state.h"
+#include "r128_span.h"
+#include "r128_tex.h"
+
+#include "swrast/swrast.h"
+
+#define DBG 0
+
+#define HAVE_HW_DEPTH_SPANS	1
+#define HAVE_HW_DEPTH_PIXELS	1
+
+#define LOCAL_VARS							\
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);				\
+   r128ScreenPtr r128scrn = rmesa->r128Screen;				\
+   __DRIscreenPrivate *sPriv = rmesa->driScreen;			\
+   __DRIdrawablePrivate *dPriv = rmesa->driDrawable;			\
+   GLuint pitch = r128scrn->frontPitch * r128scrn->cpp;			\
+   GLuint height = dPriv->h;						\
+   char *buf = (char *)(sPriv->pFB +					\
+			rmesa->drawOffset +				\
+			(dPriv->x * r128scrn->cpp) +			\
+			(dPriv->y * pitch));				\
+   char *read_buf = (char *)(sPriv->pFB +				\
+			     rmesa->readOffset +			\
+			     (dPriv->x * r128scrn->cpp) +		\
+			     (dPriv->y * pitch));			\
+   GLuint p;								\
+   (void) read_buf; (void) buf; (void) p
+
+#define LOCAL_DEPTH_VARS						\
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);				\
+   r128ScreenPtr r128scrn = rmesa->r128Screen;				\
+   __DRIscreenPrivate *sPriv = rmesa->driScreen;			\
+   __DRIdrawablePrivate *dPriv = rmesa->driDrawable;			\
+   GLuint height = dPriv->h;						\
+   (void) r128scrn; (void) sPriv; (void) height
+
+#define LOCAL_STENCIL_VARS	LOCAL_DEPTH_VARS
+
+
+#define CLIPPIXEL( _x, _y )						\
+   ((_x >= minx) && (_x < maxx) && (_y >= miny) && (_y < maxy))
+
+
+#define CLIPSPAN( _x, _y, _n, _x1, _n1, _i )				\
+   if ( _y < miny || _y >= maxy ) {					\
+      _n1 = 0, _x1 = x;							\
+   } else {								\
+      _n1 = _n;								\
+      _x1 = _x;								\
+      if ( _x1 < minx ) _i += (minx-_x1), n1 -= (minx-_x1), _x1 = minx; \
+      if ( _x1 + _n1 >= maxx ) n1 -= (_x1 + n1 - maxx);		        \
+   }
+
+#define Y_FLIP( _y )		(height - _y - 1)
+
+
+#define HW_LOCK()							\
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);				\
+   FLUSH_BATCH( rmesa );						\
+   LOCK_HARDWARE( rmesa );						\
+   r128WaitForIdleLocked( rmesa );
+
+#define HW_CLIPLOOP()							\
+   do {									\
+      __DRIdrawablePrivate *dPriv = rmesa->driDrawable;			\
+      int _nc = dPriv->numClipRects;					\
+									\
+      while ( _nc-- ) {							\
+	 int minx = dPriv->pClipRects[_nc].x1 - dPriv->x;		\
+	 int miny = dPriv->pClipRects[_nc].y1 - dPriv->y;		\
+	 int maxx = dPriv->pClipRects[_nc].x2 - dPriv->x;		\
+	 int maxy = dPriv->pClipRects[_nc].y2 - dPriv->y;
+
+#define HW_ENDCLIPLOOP()						\
+      }									\
+   } while (0)
+
+#define HW_UNLOCK()							\
+   UNLOCK_HARDWARE( rmesa )
+
+
+
+/* ================================================================
+ * Color buffer
+ */
+
+/* 16 bit, RGB565 color spanline and pixel functions
+ */
+#undef INIT_MONO_PIXEL
+#define INIT_MONO_PIXEL(p, color) \
+  p = R128PACKCOLOR565( color[0], color[1], color[2] )
+
+#define WRITE_RGBA( _x, _y, r, g, b, a )				\
+   *(GLushort *)(buf + _x*2 + _y*pitch) = ((((int)r & 0xf8) << 8) |	\
+					   (((int)g & 0xfc) << 3) |	\
+					   (((int)b & 0xf8) >> 3))
+
+#define WRITE_PIXEL( _x, _y, p )					\
+   *(GLushort *)(buf + _x*2 + _y*pitch) = p
+
+#define READ_RGBA( rgba, _x, _y )					\
+   do {									\
+      GLushort p = *(GLushort *)(read_buf + _x*2 + _y*pitch);		\
+      rgba[0] = (p >> 8) & 0xf8;					\
+      rgba[1] = (p >> 3) & 0xfc;					\
+      rgba[2] = (p << 3) & 0xf8;					\
+      rgba[3] = 0xff;							\
+      if ( rgba[0] & 0x08 ) rgba[0] |= 0x07;				\
+      if ( rgba[1] & 0x04 ) rgba[1] |= 0x03;				\
+      if ( rgba[2] & 0x08 ) rgba[2] |= 0x07;				\
+   } while (0)
+
+#define TAG(x) r128##x##_RGB565
+#include "spantmp.h"
+
+#define READ_DEPTH(d, _x, _y)                                                 \
+    d = *(GLushort *)(buf + _x*2 + _y*pitch)
+
+/* 32 bit, ARGB8888 color spanline and pixel functions
+ */
+#undef INIT_MONO_PIXEL
+#define INIT_MONO_PIXEL(p, color) \
+  p = R128PACKCOLOR8888( color[0], color[1], color[2], color[3] )
+
+#define WRITE_RGBA( _x, _y, r, g, b, a )				\
+   *(GLuint *)(buf + _x*4 + _y*pitch) = ((b <<  0) |			\
+					 (g <<  8) |			\
+					 (r << 16) |			\
+					 (a << 24) )
+
+#define WRITE_PIXEL( _x, _y, p )					\
+   *(GLuint *)(buf + _x*4 + _y*pitch) = p
+
+#define READ_RGBA( rgba, _x, _y )					\
+do {									\
+   GLuint p = *(GLuint *)(read_buf + _x*4 + _y*pitch);			\
+   rgba[0] = (p >> 16) & 0xff;						\
+   rgba[1] = (p >>  8) & 0xff;						\
+   rgba[2] = (p >>  0) & 0xff;						\
+   rgba[3] = 0xff;/*(p >> 24) & 0xff;*/						\
+} while (0)
+
+#define TAG(x) r128##x##_ARGB8888
+#include "spantmp.h"
+
+
+/* 24 bit, RGB888 color spanline and pixel functions */
+#undef INIT_MONO_PIXEL
+#define INIT_MONO_PIXEL(p, color) \
+  p = R128PACKCOLOR888( color[0], color[1], color[2] )
+
+#define WRITE_RGBA(_x, _y, r, g, b, a)                                        \
+    *(GLuint *)(buf + _x*3 + _y*pitch) = ((r << 16) |                         \
+					  (g << 8)  |                         \
+					  (b << 0))
+
+#define WRITE_PIXEL(_x, _y, p)                                                \
+    *(GLuint *)(buf + _x*3 + _y*pitch) = p
+
+#define READ_RGBA(rgba, _x, _y)                                               \
+    do {                                                                      \
+	GLuint p = *(GLuint *)(read_buf + _x*3 + _y*pitch);                   \
+	rgba[0] = (p >> 16) & 0xff;                                           \
+	rgba[1] = (p >> 8)  & 0xff;                                           \
+	rgba[2] = (p >> 0)  & 0xff;                                           \
+	rgba[3] = 0xff;                                                       \
+    } while (0)
+
+/* ================================================================
+ * Depth buffer
+ */
+
+/* 16-bit depth buffer functions
+ */
+#define WRITE_DEPTH_SPAN()						\
+   r128WriteDepthSpanLocked( rmesa, n,					\
+			     x + dPriv->x,				\
+			     y + dPriv->y,				\
+			     depth, mask );
+
+#define WRITE_DEPTH_PIXELS()						\
+do {									\
+   GLint ox[MAX_WIDTH];							\
+   GLint oy[MAX_WIDTH];							\
+   for ( i = 0 ; i < n ; i++ ) {					\
+      ox[i] = x[i] + dPriv->x;						\
+   }									\
+   for ( i = 0 ; i < n ; i++ ) {					\
+      oy[i] = Y_FLIP( y[i] ) + dPriv->y;				\
+   }									\
+   r128WriteDepthPixelsLocked( rmesa, n, ox, oy, depth, mask );		\
+} while (0)
+
+#define READ_DEPTH_SPAN()						\
+do {									\
+   GLushort *buf = (GLushort *)((GLubyte *)sPriv->pFB +			\
+				r128scrn->spanOffset);			\
+   GLint i;								\
+									\
+   r128ReadDepthSpanLocked( rmesa, n,					\
+			    x + dPriv->x,				\
+			    y + dPriv->y );				\
+   r128WaitForIdleLocked( rmesa );					\
+									\
+   for ( i = 0 ; i < n ; i++ ) {					\
+      depth[i] = buf[i];						\
+   }									\
+} while (0)
+
+#define READ_DEPTH_PIXELS()						\
+do {									\
+   GLushort *buf = (GLushort *)((GLubyte *)sPriv->pFB +			\
+				r128scrn->spanOffset);			\
+   GLint i, remaining = n;						\
+									\
+   while ( remaining > 0 ) {						\
+      GLint ox[MAX_WIDTH];						\
+      GLint oy[MAX_WIDTH];						\
+      GLint count;							\
+									\
+      if ( remaining <= 128 ) {						\
+	 count = remaining;						\
+      } else {								\
+	 count = 128;							\
+      }									\
+      for ( i = 0 ; i < count ; i++ ) {					\
+	 ox[i] = x[i] + dPriv->x;					\
+      }									\
+      for ( i = 0 ; i < count ; i++ ) {					\
+	 oy[i] = Y_FLIP( y[i] ) + dPriv->y;				\
+      }									\
+									\
+      r128ReadDepthPixelsLocked( rmesa, count, ox, oy );		\
+      r128WaitForIdleLocked( rmesa );					\
+									\
+      for ( i = 0 ; i < count ; i++ ) {					\
+	 depth[i] = buf[i];						\
+      }									\
+      depth += count;							\
+      x += count;							\
+      y += count;							\
+      remaining -= count;						\
+   }									\
+} while (0)
+
+#define TAG(x) r128##x##_16
+#include "depthtmp.h"
+
+
+/* 24-bit depth, 8-bit stencil buffer functions
+ */
+#define WRITE_DEPTH_SPAN()						\
+   r128WriteDepthSpanLocked( rmesa, n,					\
+			     x + dPriv->x,				\
+			     y + dPriv->y,				\
+			     depth, mask );
+
+#define WRITE_DEPTH_PIXELS()						\
+do {									\
+   GLint ox[MAX_WIDTH];							\
+   GLint oy[MAX_WIDTH];							\
+   for ( i = 0 ; i < n ; i++ ) {					\
+      ox[i] = x[i] + dPriv->x;						\
+   }									\
+   for ( i = 0 ; i < n ; i++ ) {					\
+      oy[i] = Y_FLIP( y[i] ) + dPriv->y;				\
+   }									\
+   r128WriteDepthPixelsLocked( rmesa, n, ox, oy, depth, mask );		\
+} while (0)
+
+#define READ_DEPTH_SPAN()						\
+do {									\
+   GLuint *buf = (GLuint *)((GLubyte *)sPriv->pFB +			\
+			    r128scrn->spanOffset);			\
+   GLint i;								\
+									\
+   r128ReadDepthSpanLocked( rmesa, n,					\
+			    x + dPriv->x,				\
+			    y + dPriv->y );				\
+   r128WaitForIdleLocked( rmesa );					\
+									\
+   for ( i = 0 ; i < n ; i++ ) {					\
+      depth[i] = buf[i] & 0x00ffffff;					\
+   }									\
+} while (0)
+
+#define READ_DEPTH_PIXELS()						\
+do {									\
+   GLuint *buf = (GLuint *)((GLubyte *)sPriv->pFB +			\
+			    r128scrn->spanOffset);			\
+   GLint i, remaining = n;						\
+									\
+   while ( remaining > 0 ) {						\
+      GLint ox[MAX_WIDTH];						\
+      GLint oy[MAX_WIDTH];						\
+      GLint count;							\
+									\
+      if ( remaining <= 128 ) {						\
+	 count = remaining;						\
+      } else {								\
+	 count = 128;							\
+      }									\
+      for ( i = 0 ; i < count ; i++ ) {					\
+	 ox[i] = x[i] + dPriv->x;					\
+      }									\
+      for ( i = 0 ; i < count ; i++ ) {					\
+	 oy[i] = Y_FLIP( y[i] ) + dPriv->y;				\
+      }									\
+									\
+      r128ReadDepthPixelsLocked( rmesa, count, ox, oy );		\
+      r128WaitForIdleLocked( rmesa );					\
+									\
+      for ( i = 0 ; i < count ; i++ ) {					\
+	 depth[i] = buf[i] & 0x00ffffff;				\
+      }									\
+      depth += count;							\
+      x += count;							\
+      y += count;							\
+      remaining -= count;						\
+   }									\
+} while (0)
+
+#define TAG(x) r128##x##_24_8
+#include "depthtmp.h"
+
+
+
+/* ================================================================
+ * Stencil buffer
+ */
+
+/* FIXME: Add support for hardware stencil buffers.
+ */
+
+
+/* 32 bit depthbuffer functions */
+#define WRITE_DEPTH(_x, _y, d)                                                \
+    *(GLuint *)(buf + _x*4 + _y*pitch) = d
+
+
+
+/*
+ * This function is called to specify which buffer to read and write
+ * for software rasterization (swrast) fallbacks.  This doesn't necessarily
+ * correspond to glDrawBuffer() or glReadBuffer() calls.
+ */
+static void r128DDSetBuffer( GLcontext *ctx,
+                             GLframebuffer *colorBuffer,
+                             GLuint bufferBit )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   switch ( bufferBit ) {
+   case FRONT_LEFT_BIT:
+      if ( rmesa->sarea->pfCurrentPage == 1 ) {
+         rmesa->drawOffset = rmesa->readOffset = rmesa->r128Screen->backOffset;
+         rmesa->drawPitch  = rmesa->readPitch  = rmesa->r128Screen->backPitch;
+      } else {
+         rmesa->drawOffset = rmesa->readOffset = rmesa->r128Screen->frontOffset;
+         rmesa->drawPitch  = rmesa->readPitch  = rmesa->r128Screen->frontPitch;
+      }
+      break;
+   case BACK_LEFT_BIT:
+      if ( rmesa->sarea->pfCurrentPage == 1 ) {
+         rmesa->drawOffset = rmesa->readOffset = rmesa->r128Screen->frontOffset;
+         rmesa->drawPitch  = rmesa->readPitch  = rmesa->r128Screen->frontPitch;
+      } else {
+         rmesa->drawOffset = rmesa->readOffset = rmesa->r128Screen->backOffset;
+         rmesa->drawPitch  = rmesa->readPitch  = rmesa->r128Screen->backPitch;
+      }
+      break;
+   default:
+      break;
+   }
+}
+
+
+void r128DDInitSpanFuncs( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   struct swrast_device_driver *swdd = _swrast_GetDeviceDriverReference(ctx);
+
+   swdd->SetBuffer = r128DDSetBuffer;
+
+   switch ( rmesa->r128Screen->cpp ) {
+   case 2:
+      swdd->WriteRGBASpan	= r128WriteRGBASpan_RGB565;
+      swdd->WriteRGBSpan	= r128WriteRGBSpan_RGB565;
+      swdd->WriteMonoRGBASpan	= r128WriteMonoRGBASpan_RGB565;
+      swdd->WriteRGBAPixels	= r128WriteRGBAPixels_RGB565;
+      swdd->WriteMonoRGBAPixels	= r128WriteMonoRGBAPixels_RGB565;
+      swdd->ReadRGBASpan	= r128ReadRGBASpan_RGB565;
+      swdd->ReadRGBAPixels	= r128ReadRGBAPixels_RGB565;
+      break;
+
+   case 4:
+      swdd->WriteRGBASpan	= r128WriteRGBASpan_ARGB8888;
+      swdd->WriteRGBSpan	= r128WriteRGBSpan_ARGB8888;
+      swdd->WriteMonoRGBASpan	= r128WriteMonoRGBASpan_ARGB8888;
+      swdd->WriteRGBAPixels	= r128WriteRGBAPixels_ARGB8888;
+      swdd->WriteMonoRGBAPixels	= r128WriteMonoRGBAPixels_ARGB8888;
+      swdd->ReadRGBASpan	= r128ReadRGBASpan_ARGB8888;
+      swdd->ReadRGBAPixels	= r128ReadRGBAPixels_ARGB8888;
+      break;
+
+   default:
+      break;
+   }
+
+   switch ( rmesa->glCtx->Visual.depthBits ) {
+   case 16:
+      swdd->ReadDepthSpan	= r128ReadDepthSpan_16;
+      swdd->WriteDepthSpan	= r128WriteDepthSpan_16;
+      swdd->ReadDepthPixels	= r128ReadDepthPixels_16;
+      swdd->WriteDepthPixels	= r128WriteDepthPixels_16;
+      break;
+
+   case 24:
+      swdd->ReadDepthSpan	= r128ReadDepthSpan_24_8;
+      swdd->WriteDepthSpan	= r128WriteDepthSpan_24_8;
+      swdd->ReadDepthPixels	= r128ReadDepthPixels_24_8;
+      swdd->WriteDepthPixels	= r128WriteDepthPixels_24_8;
+      break;
+
+   default:
+      break;
+   }
+
+   swdd->WriteCI8Span		= NULL;
+   swdd->WriteCI32Span		= NULL;
+   swdd->WriteMonoCISpan	= NULL;
+   swdd->WriteCI32Pixels	= NULL;
+   swdd->WriteMonoCIPixels	= NULL;
+   swdd->ReadCI32Span		= NULL;
+   swdd->ReadCI32Pixels		= NULL;
+}
diff --git a/src/mesa/drivers/dri/r128/r128_span.h b/src/mesa/drivers/dri/r128/r128_span.h
new file mode 100644
index 00000000000..44356c5b241
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_span.h
@@ -0,0 +1,44 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_span.h,v 1.3 2001/01/08 01:07:21 martin Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Gareth Hughes <[email protected]>
+ *   Kevin E. Martin <[email protected]>
+ *
+ */
+
+#ifndef __R128_SPAN_H__
+#define __R128_SPAN_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+extern void r128DDInitSpanFuncs( GLcontext *ctx );
+
+#endif
+#endif
diff --git a/src/mesa/drivers/dri/r128/r128_state.c b/src/mesa/drivers/dri/r128/r128_state.c
new file mode 100644
index 00000000000..85b3b68402d
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_state.c
@@ -0,0 +1,1239 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_state.c,v 1.11 2002/10/30 12:51:39 alanh Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Gareth Hughes <[email protected]>
+ *   Kevin E. Martin <[email protected]>
+ *   Keith Whitwell <[email protected]>
+ *
+ */
+
+#include "r128_context.h"
+#include "r128_state.h"
+#include "r128_ioctl.h"
+#include "r128_tris.h"
+#include "r128_vb.h"
+#include "r128_tex.h"
+
+#include "context.h"
+#include "enums.h"
+#include "colormac.h"
+#include "swrast/swrast.h"
+#include "array_cache/acache.h"
+#include "tnl/tnl.h"
+#include "swrast_setup/swrast_setup.h"
+
+#include "tnl/t_pipeline.h"
+
+
+/* =============================================================
+ * Alpha blending
+ */
+
+static void r128UpdateAlphaMode( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   GLuint a = rmesa->setup.misc_3d_state_cntl_reg;
+   GLuint t = rmesa->setup.tex_cntl_c;
+
+   if ( ctx->Color.AlphaEnabled ) {
+      GLubyte ref;
+
+      CLAMPED_FLOAT_TO_UBYTE(ref, ctx->Color.AlphaRef);
+
+      a &= ~(R128_ALPHA_TEST_MASK | R128_REF_ALPHA_MASK);
+
+      switch ( ctx->Color.AlphaFunc ) {
+      case GL_NEVER:
+	 a |= R128_ALPHA_TEST_NEVER;
+	 break;
+      case GL_LESS:
+	 a |= R128_ALPHA_TEST_LESS;
+         break;
+      case GL_LEQUAL:
+	 a |= R128_ALPHA_TEST_LESSEQUAL;
+	 break;
+      case GL_EQUAL:
+	 a |= R128_ALPHA_TEST_EQUAL;
+	 break;
+      case GL_GEQUAL:
+	 a |= R128_ALPHA_TEST_GREATEREQUAL;
+	 break;
+      case GL_GREATER:
+	 a |= R128_ALPHA_TEST_GREATER;
+	 break;
+      case GL_NOTEQUAL:
+	 a |= R128_ALPHA_TEST_NEQUAL;
+	 break;
+      case GL_ALWAYS:
+	 a |= R128_ALPHA_TEST_ALWAYS;
+	 break;
+      }
+
+      a |= ref & R128_REF_ALPHA_MASK;
+      t |= R128_ALPHA_TEST_ENABLE;
+   } else {
+      t &= ~R128_ALPHA_TEST_ENABLE;
+   }
+
+   FALLBACK( rmesa, R128_FALLBACK_BLEND_FUNC, GL_FALSE );
+
+   if ( ctx->Color.BlendEnabled ) {
+      a &= ~(R128_ALPHA_BLEND_SRC_MASK | R128_ALPHA_BLEND_DST_MASK);
+
+      switch ( ctx->Color.BlendSrcRGB ) {
+      case GL_ZERO:
+	 a |= R128_ALPHA_BLEND_SRC_ZERO;
+	 break;
+      case GL_ONE:
+	 a |= R128_ALPHA_BLEND_SRC_ONE;
+	 break;
+      case GL_DST_COLOR:
+	 a |= R128_ALPHA_BLEND_SRC_DESTCOLOR;
+	 break;
+      case GL_ONE_MINUS_DST_COLOR:
+	 a |= R128_ALPHA_BLEND_SRC_INVDESTCOLOR;
+	 break;
+      case GL_SRC_ALPHA:
+	 a |= R128_ALPHA_BLEND_SRC_SRCALPHA;
+	 break;
+      case GL_ONE_MINUS_SRC_ALPHA:
+	 a |= R128_ALPHA_BLEND_SRC_INVSRCALPHA;
+	 break;
+      case GL_DST_ALPHA:
+	 a |= R128_ALPHA_BLEND_SRC_DESTALPHA;
+	 break;
+      case GL_ONE_MINUS_DST_ALPHA:
+	 a |= R128_ALPHA_BLEND_SRC_INVDESTALPHA;
+	 break;
+      case GL_SRC_ALPHA_SATURATE:
+	 a |= R128_ALPHA_BLEND_SRC_SRCALPHASAT;
+	 break;
+      default:
+         FALLBACK( rmesa, R128_FALLBACK_BLEND_FUNC, GL_TRUE );
+      }
+
+      switch ( ctx->Color.BlendDstRGB ) {
+      case GL_ZERO:
+	 a |= R128_ALPHA_BLEND_DST_ZERO;
+	 break;
+      case GL_ONE:
+	 a |= R128_ALPHA_BLEND_DST_ONE;
+	 break;
+      case GL_SRC_COLOR:
+	 a |= R128_ALPHA_BLEND_DST_SRCCOLOR;
+	 break;
+      case GL_ONE_MINUS_SRC_COLOR:
+	 a |= R128_ALPHA_BLEND_DST_INVSRCCOLOR;
+	 break;
+      case GL_SRC_ALPHA:
+	 a |= R128_ALPHA_BLEND_DST_SRCALPHA;
+	 break;
+      case GL_ONE_MINUS_SRC_ALPHA:
+	 a |= R128_ALPHA_BLEND_DST_INVSRCALPHA;
+	 break;
+      case GL_DST_ALPHA:
+	 a |= R128_ALPHA_BLEND_DST_DESTALPHA;
+	 break;
+      case GL_ONE_MINUS_DST_ALPHA:
+	 a |= R128_ALPHA_BLEND_DST_INVDESTALPHA;
+	 break;
+      default:
+         FALLBACK( rmesa, R128_FALLBACK_BLEND_FUNC, GL_TRUE );
+      }
+
+      t |=  R128_ALPHA_ENABLE;
+   } else {
+      t &= ~R128_ALPHA_ENABLE;
+   }
+
+   if ( rmesa->setup.misc_3d_state_cntl_reg != a ) {
+      rmesa->setup.misc_3d_state_cntl_reg = a;
+      rmesa->dirty |= R128_UPLOAD_CONTEXT | R128_UPLOAD_MASKS;
+   }
+   if ( rmesa->setup.tex_cntl_c != t ) {
+      rmesa->setup.tex_cntl_c = t;
+      rmesa->dirty |= R128_UPLOAD_CONTEXT | R128_UPLOAD_MASKS;
+   }
+}
+
+static void r128DDAlphaFunc( GLcontext *ctx, GLenum func, GLfloat ref )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   FLUSH_BATCH( rmesa );
+   rmesa->new_state |= R128_NEW_ALPHA;
+}
+
+static void r128DDBlendEquation( GLcontext *ctx, GLenum mode )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   FLUSH_BATCH( rmesa );
+
+   /* BlendEquation sets ColorLogicOpEnabled in an unexpected
+    * manner.
+    */
+   FALLBACK( R128_CONTEXT(ctx), R128_FALLBACK_LOGICOP,
+	     (ctx->Color.ColorLogicOpEnabled &&
+	      ctx->Color.LogicOp != GL_COPY));
+
+   /* Can only do blend addition, not min, max, subtract, etc. */
+   FALLBACK( R128_CONTEXT(ctx), R128_FALLBACK_BLEND_EQ,
+	     mode != GL_FUNC_ADD_EXT);
+
+   rmesa->new_state |= R128_NEW_ALPHA;
+}
+
+static void r128DDBlendFunc( GLcontext *ctx, GLenum sfactor, GLenum dfactor )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   FLUSH_BATCH( rmesa );
+   rmesa->new_state |= R128_NEW_ALPHA;
+}
+
+static void r128DDBlendFuncSeparate( GLcontext *ctx,
+				     GLenum sfactorRGB, GLenum dfactorRGB,
+				     GLenum sfactorA, GLenum dfactorA )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   FLUSH_BATCH( rmesa );
+   rmesa->new_state |= R128_NEW_ALPHA;
+}
+
+
+/* =============================================================
+ * Depth testing
+ */
+
+static void r128UpdateZMode( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   GLuint z = rmesa->setup.z_sten_cntl_c;
+   GLuint t = rmesa->setup.tex_cntl_c;
+
+   if ( ctx->Depth.Test ) {
+      z &= ~R128_Z_TEST_MASK;
+
+      switch ( ctx->Depth.Func ) {
+      case GL_NEVER:
+	 z |= R128_Z_TEST_NEVER;
+	 break;
+      case GL_ALWAYS:
+	 z |= R128_Z_TEST_ALWAYS;
+	 break;
+      case GL_LESS:
+	 z |= R128_Z_TEST_LESS;
+	 break;
+      case GL_LEQUAL:
+	 z |= R128_Z_TEST_LESSEQUAL;
+	 break;
+      case GL_EQUAL:
+	 z |= R128_Z_TEST_EQUAL;
+	 break;
+      case GL_GEQUAL:
+	 z |= R128_Z_TEST_GREATEREQUAL;
+	 break;
+      case GL_GREATER:
+	 z |= R128_Z_TEST_GREATER;
+	 break;
+      case GL_NOTEQUAL:
+	 z |= R128_Z_TEST_NEQUAL;
+	 break;
+      }
+
+      t |=  R128_Z_ENABLE;
+   } else {
+      t &= ~R128_Z_ENABLE;
+   }
+
+   if ( ctx->Depth.Mask ) {
+      t |=  R128_Z_WRITE_ENABLE;
+   } else {
+      t &= ~R128_Z_WRITE_ENABLE;
+   }
+
+   if ( rmesa->setup.z_sten_cntl_c != z ) {
+      rmesa->setup.z_sten_cntl_c = z;
+      rmesa->dirty |= R128_UPLOAD_CONTEXT;
+   }
+   if ( rmesa->setup.tex_cntl_c != t ) {
+      rmesa->setup.tex_cntl_c = t;
+      rmesa->dirty |= R128_UPLOAD_CONTEXT;
+   }
+}
+
+static void r128DDDepthFunc( GLcontext *ctx, GLenum func )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   FLUSH_BATCH( rmesa );
+   rmesa->new_state |= R128_NEW_DEPTH;
+}
+
+static void r128DDDepthMask( GLcontext *ctx, GLboolean flag )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   FLUSH_BATCH( rmesa );
+   rmesa->new_state |= R128_NEW_DEPTH;
+}
+
+static void r128DDClearDepth( GLcontext *ctx, GLclampd d )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   switch ( rmesa->setup.z_sten_cntl_c &  R128_Z_PIX_WIDTH_MASK ) {
+   case R128_Z_PIX_WIDTH_16:
+      rmesa->ClearDepth = d * 0x0000ffff;
+      break;
+   case R128_Z_PIX_WIDTH_24:
+      rmesa->ClearDepth = d * 0x00ffffff;
+      break;
+   case R128_Z_PIX_WIDTH_32:
+      rmesa->ClearDepth = d * 0xffffffff;
+      break;
+   }
+}
+
+
+/* =============================================================
+ * Fog
+ */
+
+static void r128UpdateFogAttrib( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   GLuint t = rmesa->setup.tex_cntl_c;
+   GLubyte c[4];
+   GLuint col;
+
+   if ( ctx->Fog.Enabled ) {
+      t |=  R128_FOG_ENABLE;
+   } else {
+      t &= ~R128_FOG_ENABLE;
+   }
+
+   c[0] = FLOAT_TO_UBYTE( ctx->Fog.Color[0] );
+   c[1] = FLOAT_TO_UBYTE( ctx->Fog.Color[1] );
+   c[2] = FLOAT_TO_UBYTE( ctx->Fog.Color[2] );
+
+   col = r128PackColor( 4, c[0], c[1], c[2], 0 );
+
+   if ( rmesa->setup.fog_color_c != col ) {
+      rmesa->setup.fog_color_c = col;
+      rmesa->dirty |= R128_UPLOAD_CONTEXT;
+   }
+   if ( rmesa->setup.tex_cntl_c != t ) {
+      rmesa->setup.tex_cntl_c = t;
+      rmesa->dirty |= R128_UPLOAD_CONTEXT;
+   }
+}
+
+static void r128DDFogfv( GLcontext *ctx, GLenum pname, const GLfloat *param )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   FLUSH_BATCH( rmesa );
+   rmesa->new_state |= R128_NEW_FOG;
+}
+
+
+/* =============================================================
+ * Clipping
+ */
+
+static void r128UpdateClipping( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   if ( rmesa->driDrawable ) {
+      __DRIdrawablePrivate *drawable = rmesa->driDrawable;
+      int x1 = 0;
+      int y1 = 0;
+      int x2 = drawable->w - 1;
+      int y2 = drawable->h - 1;
+
+      if ( ctx->Scissor.Enabled ) {
+	 if ( ctx->Scissor.X > x1 ) {
+	    x1 = ctx->Scissor.X;
+	 }
+	 if ( drawable->h - ctx->Scissor.Y - ctx->Scissor.Height > y1 ) {
+	    y1 = drawable->h - ctx->Scissor.Y - ctx->Scissor.Height;
+	 }
+	 if ( ctx->Scissor.X + ctx->Scissor.Width - 1 < x2 ) {
+	    x2 = ctx->Scissor.X + ctx->Scissor.Width - 1;
+	 }
+	 if ( drawable->h - ctx->Scissor.Y - 1 < y2 ) {
+	    y2 = drawable->h - ctx->Scissor.Y - 1;
+	 }
+      }
+
+      x1 += drawable->x;
+      y1 += drawable->y;
+      x2 += drawable->x;
+      y2 += drawable->y;
+
+      rmesa->setup.sc_top_left_c     = ((y1 << 16) | x1);
+      rmesa->setup.sc_bottom_right_c = ((y2 << 16) | x2);
+
+      rmesa->dirty |= R128_UPLOAD_CONTEXT;
+   }
+}
+
+static void r128DDScissor( GLcontext *ctx,
+			   GLint x, GLint y, GLsizei w, GLsizei h )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   FLUSH_BATCH( rmesa );
+   rmesa->new_state |= R128_NEW_CLIP;
+}
+
+
+/* =============================================================
+ * Culling
+ */
+
+static void r128UpdateCull( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   GLuint f = rmesa->setup.pm4_vc_fpu_setup;
+
+   f &= ~R128_FRONT_DIR_MASK;
+
+   switch ( ctx->Polygon.FrontFace ) {
+   case GL_CW:
+      f |= R128_FRONT_DIR_CW;
+      break;
+   case GL_CCW:
+      f |= R128_FRONT_DIR_CCW;
+      break;
+   }
+
+   f |= R128_BACKFACE_SOLID | R128_FRONTFACE_SOLID;
+
+   if ( ctx->Polygon.CullFlag ) {
+      switch ( ctx->Polygon.CullFaceMode ) {
+      case GL_FRONT:
+	 f &= ~R128_FRONTFACE_SOLID;
+	 break;
+      case GL_BACK:
+	 f &= ~R128_BACKFACE_SOLID;
+	 break;
+      case GL_FRONT_AND_BACK:
+	 f &= ~(R128_BACKFACE_SOLID |
+		R128_FRONTFACE_SOLID);
+	 break;
+      }
+   }
+
+   if ( 1 || rmesa->setup.pm4_vc_fpu_setup != f ) {
+      rmesa->setup.pm4_vc_fpu_setup = f;
+      rmesa->dirty |= R128_UPLOAD_CONTEXT | R128_UPLOAD_SETUP;
+   }
+}
+
+static void r128DDCullFace( GLcontext *ctx, GLenum mode )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   FLUSH_BATCH( rmesa );
+   rmesa->new_state |= R128_NEW_CULL;
+}
+
+static void r128DDFrontFace( GLcontext *ctx, GLenum mode )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   FLUSH_BATCH( rmesa );
+   rmesa->new_state |= R128_NEW_CULL;
+}
+
+
+/* =============================================================
+ * Masks
+ */
+
+static void r128UpdateMasks( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   GLuint mask = r128PackColor( rmesa->r128Screen->cpp,
+				ctx->Color.ColorMask[RCOMP],
+				ctx->Color.ColorMask[GCOMP],
+				ctx->Color.ColorMask[BCOMP],
+				ctx->Color.ColorMask[ACOMP] );
+
+   if ( rmesa->setup.plane_3d_mask_c != mask ) {
+      rmesa->setup.plane_3d_mask_c = mask;
+      rmesa->dirty |= R128_UPLOAD_CONTEXT | R128_UPLOAD_MASKS;
+   }
+}
+
+static void r128DDColorMask( GLcontext *ctx,
+			     GLboolean r, GLboolean g,
+			     GLboolean b, GLboolean a )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   FLUSH_BATCH( rmesa );
+   rmesa->new_state |= R128_NEW_MASKS;
+}
+
+
+/* =============================================================
+ * Rendering attributes
+ *
+ * We really don't want to recalculate all this every time we bind a
+ * texture.  These things shouldn't change all that often, so it makes
+ * sense to break them out of the core texture state update routines.
+ */
+
+static void updateSpecularLighting( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   GLuint t = rmesa->setup.tex_cntl_c;
+
+   if ( ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR &&
+        ctx->Light.Enabled) {
+      /* XXX separate specular color just doesn't seem to work as it should.
+       * For now, we fall back to s/w rendering whenever separate specular
+       * is enabled.
+       */
+#if 0
+      if (ctx->Light.ShadeModel == GL_FLAT) {
+         /* R128 can't do flat-shaded separate specular */
+         t &= ~R128_SPEC_LIGHT_ENABLE;
+         FALLBACK( rmesa, R128_FALLBACK_SEP_SPECULAR, GL_TRUE );
+         /*printf("%s fallback  sep spec\n", __FUNCTION__);*/
+      }
+      else {
+         t |= R128_SPEC_LIGHT_ENABLE;
+         FALLBACK( rmesa, R128_FALLBACK_SEP_SPECULAR, GL_FALSE );
+         /*printf("%s enable sep spec\n", __FUNCTION__);*/
+      }
+#else
+      t &= ~R128_SPEC_LIGHT_ENABLE;
+      FALLBACK( rmesa, R128_FALLBACK_SEP_SPECULAR, GL_TRUE );
+      /*printf("%s fallback  sep spec\n", __FUNCTION__);*/
+#endif
+   }
+   else {
+      t &= ~R128_SPEC_LIGHT_ENABLE;
+      FALLBACK( rmesa, R128_FALLBACK_SEP_SPECULAR, GL_FALSE );
+      /*printf("%s disable sep spec\n", __FUNCTION__);*/
+   }
+
+   if ( rmesa->setup.tex_cntl_c != t ) {
+      rmesa->setup.tex_cntl_c = t;
+      rmesa->dirty |= R128_UPLOAD_CONTEXT;
+      rmesa->dirty |= R128_UPLOAD_SETUP;
+      rmesa->new_state |= R128_NEW_CONTEXT;
+   }
+}
+
+
+static void r128DDLightModelfv( GLcontext *ctx, GLenum pname,
+				const GLfloat *param )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   if ( pname == GL_LIGHT_MODEL_COLOR_CONTROL ) {
+      FLUSH_BATCH( rmesa );
+      updateSpecularLighting(ctx);
+   }
+}
+
+static void r128DDShadeModel( GLcontext *ctx, GLenum mode )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   GLuint s = rmesa->setup.pm4_vc_fpu_setup;
+
+   s &= ~R128_FPU_COLOR_MASK;
+
+   switch ( mode ) {
+   case GL_FLAT:
+      s |= R128_FPU_COLOR_FLAT;
+      break;
+   case GL_SMOOTH:
+      s |= R128_FPU_COLOR_GOURAUD;
+      break;
+   default:
+      return;
+   }
+
+   updateSpecularLighting(ctx);
+
+   if ( rmesa->setup.pm4_vc_fpu_setup != s ) {
+      FLUSH_BATCH( rmesa );
+      rmesa->setup.pm4_vc_fpu_setup = s;
+
+      rmesa->new_state |= R128_NEW_CONTEXT;
+      rmesa->dirty |= R128_UPLOAD_SETUP;
+   }
+}
+
+
+/* =============================================================
+ * Window position
+ */
+
+void r128UpdateWindow( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   int x = rmesa->driDrawable->x;
+   int y = rmesa->driDrawable->y;
+
+   rmesa->setup.window_xy_offset = ((y << R128_WINDOW_Y_SHIFT) |
+				    (x << R128_WINDOW_X_SHIFT));
+
+   rmesa->dirty |= R128_UPLOAD_CONTEXT | R128_UPLOAD_WINDOW;
+}
+
+/* =============================================================
+ * Viewport
+ */
+
+
+static void r128CalcViewport( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   const GLfloat *v = ctx->Viewport._WindowMap.m;
+   GLfloat *m = rmesa->hw_viewport;
+
+   /* See also r128_translate_vertex.
+    */
+   m[MAT_SX] =   v[MAT_SX];
+   m[MAT_TX] =   v[MAT_TX] + SUBPIXEL_X;
+   m[MAT_SY] = - v[MAT_SY];
+   m[MAT_TY] = - v[MAT_TY] + rmesa->driDrawable->h + SUBPIXEL_Y;
+   m[MAT_SZ] =   v[MAT_SZ] * rmesa->depth_scale;
+   m[MAT_TZ] =   v[MAT_TZ] * rmesa->depth_scale;
+}
+
+static void r128Viewport( GLcontext *ctx,
+			  GLint x, GLint y,
+			  GLsizei width, GLsizei height )
+{
+   r128CalcViewport( ctx );
+}
+
+static void r128DepthRange( GLcontext *ctx,
+			    GLclampd nearval, GLclampd farval )
+{
+   r128CalcViewport( ctx );
+}
+
+
+/* =============================================================
+ * Miscellaneous
+ */
+
+static void r128DDClearColor( GLcontext *ctx,
+			      const GLfloat color[4] )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   GLubyte c[4];
+
+   CLAMPED_FLOAT_TO_UBYTE(c[0], color[0]);
+   CLAMPED_FLOAT_TO_UBYTE(c[1], color[1]);
+   CLAMPED_FLOAT_TO_UBYTE(c[2], color[2]);
+   CLAMPED_FLOAT_TO_UBYTE(c[3], color[3]);
+
+   rmesa->ClearColor = r128PackColor( rmesa->r128Screen->cpp,
+				      c[0], c[1], c[2], c[3] );
+}
+
+static void r128DDLogicOpCode( GLcontext *ctx, GLenum opcode )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   if ( ctx->Color.ColorLogicOpEnabled ) {
+      FLUSH_BATCH( rmesa );
+
+      FALLBACK( rmesa, R128_FALLBACK_LOGICOP, opcode != GL_COPY );
+   }
+}
+
+static void r128DDDrawBuffer( GLcontext *ctx, GLenum mode )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   FLUSH_BATCH( rmesa );
+
+   /*
+    * _DrawDestMask is easier to cope with than <mode>.
+    */
+   switch ( ctx->Color._DrawDestMask ) {
+   case FRONT_LEFT_BIT:
+      FALLBACK( rmesa, R128_FALLBACK_DRAW_BUFFER, GL_FALSE );
+      break;
+   case BACK_LEFT_BIT:
+      FALLBACK( rmesa, R128_FALLBACK_DRAW_BUFFER, GL_FALSE );
+      break;
+   default:
+      /* GL_NONE or GL_FRONT_AND_BACK or stereo left&right, etc */
+      FALLBACK( rmesa, R128_FALLBACK_DRAW_BUFFER, GL_TRUE );
+      break;
+   }
+
+   /* We want to update the s/w rast state too so that r200SetBuffer()
+    * gets called.
+    */
+   _swrast_DrawBuffer(ctx, mode);
+
+   rmesa->setup.dst_pitch_offset_c = (((rmesa->drawPitch/8) << 21) |
+                                      (rmesa->drawOffset >> 5));
+   rmesa->new_state |= R128_NEW_WINDOW;
+}
+
+static void r128DDReadBuffer( GLcontext *ctx, GLenum mode )
+{
+   /* nothing, until we implement h/w glRead/CopyPixels or CopyTexImage */
+}
+
+
+/* =============================================================
+ * Polygon stipple
+ */
+
+static void r128DDPolygonStipple( GLcontext *ctx, const GLubyte *mask )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   GLuint stipple[32], i;
+   drmR128Stipple stippleRec;
+
+   for (i = 0; i < 32; i++) {
+      stipple[31 - i] = ((mask[i*4+0] << 24) |
+                         (mask[i*4+1] << 16) |
+                         (mask[i*4+2] << 8)  |
+                         (mask[i*4+3]));
+   }
+
+   FLUSH_BATCH( rmesa );
+   LOCK_HARDWARE( rmesa );
+
+   stippleRec.mask = stipple;
+   drmCommandWrite( rmesa->driFd, DRM_R128_STIPPLE, 
+                    &stippleRec, sizeof(drmR128Stipple) );
+
+   UNLOCK_HARDWARE( rmesa );
+
+   rmesa->new_state |= R128_NEW_CONTEXT;
+   rmesa->dirty |= R128_UPLOAD_CONTEXT;
+}
+
+
+/* =============================================================
+ * Render mode
+ */
+
+static void r128DDRenderMode( GLcontext *ctx, GLenum mode )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   FALLBACK( rmesa, R128_FALLBACK_RENDER_MODE, (mode != GL_RENDER) );
+}
+
+
+
+/* =============================================================
+ * State enable/disable
+ */
+
+static void r128DDEnable( GLcontext *ctx, GLenum cap, GLboolean state )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   if ( R128_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s( %s = %s )\n",
+	       __FUNCTION__, _mesa_lookup_enum_by_nr( cap ),
+	       state ? "GL_TRUE" : "GL_FALSE" );
+   }
+
+   switch ( cap ) {
+   case GL_ALPHA_TEST:
+      FLUSH_BATCH( rmesa );
+      rmesa->new_state |= R128_NEW_ALPHA;
+      break;
+
+   case GL_BLEND:
+      FLUSH_BATCH( rmesa );
+      rmesa->new_state |= R128_NEW_ALPHA;
+
+      /* For some reason enable(GL_BLEND) affects ColorLogicOpEnabled.
+       */
+      FALLBACK( rmesa, R128_FALLBACK_LOGICOP,
+		(ctx->Color.ColorLogicOpEnabled &&
+		 ctx->Color.LogicOp != GL_COPY));
+      break;
+
+   case GL_CULL_FACE:
+      FLUSH_BATCH( rmesa );
+      rmesa->new_state |= R128_NEW_CULL;
+      break;
+
+   case GL_DEPTH_TEST:
+      FLUSH_BATCH( rmesa );
+      rmesa->new_state |= R128_NEW_DEPTH;
+      break;
+
+   case GL_DITHER:
+      do {
+	 GLuint t = rmesa->setup.tex_cntl_c;
+	 FLUSH_BATCH( rmesa );
+
+	 if ( ctx->Color.DitherFlag ) {
+	    t |=  R128_DITHER_ENABLE;
+	 } else {
+	    t &= ~R128_DITHER_ENABLE;
+	 }
+
+	 if ( rmesa->setup.tex_cntl_c != t ) {
+	    rmesa->setup.tex_cntl_c = t;
+	    rmesa->dirty |= R128_UPLOAD_CONTEXT;
+	 }
+      } while (0);
+      break;
+
+   case GL_FOG:
+      FLUSH_BATCH( rmesa );
+      rmesa->new_state |= R128_NEW_FOG;
+      break;
+
+   case GL_COLOR_LOGIC_OP:
+      FLUSH_BATCH( rmesa );
+      FALLBACK( rmesa, R128_FALLBACK_LOGICOP,
+		state && ctx->Color.LogicOp != GL_COPY );
+      break;
+
+   case GL_LIGHTING:
+      updateSpecularLighting(ctx);
+      break;
+
+   case GL_SCISSOR_TEST:
+      FLUSH_BATCH( rmesa );
+      rmesa->scissor = state;
+      rmesa->new_state |= R128_NEW_CLIP;
+      break;
+
+   case GL_STENCIL_TEST:
+      FLUSH_BATCH( rmesa );
+      FALLBACK( rmesa, R128_FALLBACK_STENCIL, state );
+      break;
+
+   case GL_TEXTURE_1D:
+   case GL_TEXTURE_2D:
+   case GL_TEXTURE_3D:
+      FLUSH_BATCH( rmesa );
+      break;
+
+   case GL_POLYGON_STIPPLE:
+      if ( rmesa->render_primitive == GL_TRIANGLES ) {
+	 FLUSH_BATCH( rmesa );
+	 rmesa->setup.dp_gui_master_cntl_c &= ~R128_GMC_BRUSH_NONE;
+	 if ( state ) {
+	    rmesa->setup.dp_gui_master_cntl_c |=
+	       R128_GMC_BRUSH_32x32_MONO_FG_LA;
+	 } else {
+	    rmesa->setup.dp_gui_master_cntl_c |=
+	       R128_GMC_BRUSH_SOLID_COLOR;
+	 }
+	 rmesa->new_state |= R128_NEW_CONTEXT;
+	 rmesa->dirty |= R128_UPLOAD_CONTEXT;
+      }
+      break;
+
+   default:
+      return;
+   }
+}
+
+
+/* =============================================================
+ * State initialization, management
+ */
+
+static void r128DDPrintDirty( const char *msg, GLuint state )
+{
+   fprintf( stderr,
+	    "%s: (0x%x) %s%s%s%s%s%s%s%s%s\n",
+	    msg,
+	    state,
+	    (state & R128_UPLOAD_CORE)		? "core, " : "",
+	    (state & R128_UPLOAD_CONTEXT)	? "context, " : "",
+	    (state & R128_UPLOAD_SETUP)		? "setup, " : "",
+	    (state & R128_UPLOAD_TEX0)		? "tex0, " : "",
+	    (state & R128_UPLOAD_TEX1)		? "tex1, " : "",
+	    (state & R128_UPLOAD_MASKS)		? "masks, " : "",
+	    (state & R128_UPLOAD_WINDOW)	? "window, " : "",
+	    (state & R128_UPLOAD_CLIPRECTS)	? "cliprects, " : "",
+	    (state & R128_REQUIRE_QUIESCENCE)	? "quiescence, " : "" );
+}
+
+/*
+ * Load the current context's state into the hardware.
+ *
+ * NOTE: Be VERY careful about ensuring the context state is marked for
+ * upload, the only place it shouldn't be uploaded is when the setup
+ * state has changed in ReducedPrimitiveChange as this comes right after
+ * a state update.
+ *
+ * Blits of any type should always upload the context and masks after
+ * they are done.
+ */
+void r128EmitHwStateLocked( r128ContextPtr rmesa )
+{
+   R128SAREAPrivPtr sarea = rmesa->sarea;
+   r128_context_regs_t *regs = &(rmesa->setup);
+   const r128TexObjPtr t0 = rmesa->CurrentTexObj[0];
+   const r128TexObjPtr t1 = rmesa->CurrentTexObj[1];
+
+   if ( R128_DEBUG & DEBUG_VERBOSE_MSG ) {
+      r128DDPrintDirty( "r128EmitHwStateLocked", rmesa->dirty );
+   }
+
+   if ( rmesa->dirty & (R128_UPLOAD_CONTEXT |
+			R128_UPLOAD_SETUP |
+			R128_UPLOAD_MASKS |
+			R128_UPLOAD_WINDOW |
+			R128_UPLOAD_CORE) ) {
+      memcpy( &sarea->ContextState, regs, sizeof(sarea->ContextState) );
+   }
+
+   if ( (rmesa->dirty & R128_UPLOAD_TEX0) && t0 ) {
+      r128_texture_regs_t *tex = &sarea->TexState[0];
+
+      tex->tex_cntl		= t0->setup.tex_cntl;
+      tex->tex_combine_cntl	= rmesa->tex_combine[0];
+      tex->tex_size_pitch	= t0->setup.tex_size_pitch;
+      memcpy( &tex->tex_offset[0], &t0->setup.tex_offset[0],
+	      sizeof(tex->tex_offset ) );
+      tex->tex_border_color	= t0->setup.tex_border_color;
+   }
+
+   if ( (rmesa->dirty & R128_UPLOAD_TEX1) && t1 ) {
+      r128_texture_regs_t *tex = &sarea->TexState[1];
+
+      tex->tex_cntl		= t1->setup.tex_cntl;
+      tex->tex_combine_cntl	= rmesa->tex_combine[1];
+      tex->tex_size_pitch	= t1->setup.tex_size_pitch;
+      memcpy( &tex->tex_offset[0], &t1->setup.tex_offset[0],
+	      sizeof(tex->tex_offset ) );
+      tex->tex_border_color	= t1->setup.tex_border_color;
+   }
+
+   sarea->vertsize = rmesa->vertex_size;
+   sarea->vc_format = rmesa->vertex_format;
+
+   /* Turn off the texture cache flushing */
+   rmesa->setup.tex_cntl_c &= ~R128_TEX_CACHE_FLUSH;
+
+   sarea->dirty |= rmesa->dirty;
+   rmesa->dirty &= R128_UPLOAD_CLIPRECTS;
+}
+
+static void r128DDPrintState( const char *msg, GLuint flags )
+{
+   fprintf( stderr,
+	    "%s: (0x%x) %s%s%s%s%s%s%s%s\n",
+	    msg,
+	    flags,
+	    (flags & R128_NEW_CONTEXT)	? "context, " : "",
+	    (flags & R128_NEW_ALPHA)	? "alpha, " : "",
+	    (flags & R128_NEW_DEPTH)	? "depth, " : "",
+	    (flags & R128_NEW_FOG)	? "fog, " : "",
+	    (flags & R128_NEW_CLIP)	? "clip, " : "",
+	    (flags & R128_NEW_CULL)	? "cull, " : "",
+	    (flags & R128_NEW_MASKS)	? "masks, " : "",
+	    (flags & R128_NEW_WINDOW)	? "window, " : "" );
+}
+
+void r128DDUpdateHWState( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   int new_state = rmesa->new_state;
+
+   if ( new_state || rmesa->NewGLState & _NEW_TEXTURE )
+   {
+      FLUSH_BATCH( rmesa );
+
+      rmesa->new_state = 0;
+
+      if ( R128_DEBUG & DEBUG_VERBOSE_MSG )
+	 r128DDPrintState( "r128UpdateHwState", new_state );
+
+      /* Update the various parts of the context's state.
+       */
+      if ( new_state & R128_NEW_ALPHA )
+	 r128UpdateAlphaMode( ctx );
+
+      if ( new_state & R128_NEW_DEPTH )
+	 r128UpdateZMode( ctx );
+
+      if ( new_state & R128_NEW_FOG )
+	 r128UpdateFogAttrib( ctx );
+
+      if ( new_state & R128_NEW_CLIP )
+	 r128UpdateClipping( ctx );
+
+      if ( new_state & R128_NEW_CULL )
+	 r128UpdateCull( ctx );
+
+      if ( new_state & R128_NEW_MASKS )
+	 r128UpdateMasks( ctx );
+
+      if ( new_state & R128_NEW_WINDOW )
+	 r128UpdateWindow( ctx );
+
+      if ( rmesa->NewGLState & _NEW_TEXTURE ) {
+	 r128UpdateTextureState( ctx );
+      }
+   }
+}
+
+
+static void r128DDInvalidateState( GLcontext *ctx, GLuint new_state )
+{
+   _swrast_InvalidateState( ctx, new_state );
+   _swsetup_InvalidateState( ctx, new_state );
+   _ac_InvalidateState( ctx, new_state );
+   _tnl_InvalidateState( ctx, new_state );
+   R128_CONTEXT(ctx)->NewGLState |= new_state;
+}
+
+
+
+/* Initialize the context's hardware state.
+ */
+void r128DDInitState( r128ContextPtr rmesa )
+{
+   int dst_bpp, depth_bpp;
+
+   switch ( rmesa->r128Screen->cpp ) {
+   case 2:
+      dst_bpp = R128_GMC_DST_16BPP;
+      break;
+   case 4:
+      dst_bpp = R128_GMC_DST_32BPP;
+      break;
+   default:
+      fprintf( stderr, "Error: Unsupported pixel depth... exiting\n" );
+      exit( -1 );
+   }
+
+   rmesa->ClearColor = 0x00000000;
+
+   switch ( rmesa->glCtx->Visual.depthBits ) {
+   case 16:
+      rmesa->ClearDepth = 0x0000ffff;
+      depth_bpp = R128_Z_PIX_WIDTH_16;
+      rmesa->depth_scale = 1.0 / (GLfloat)0xffff;
+      break;
+   case 24:
+      rmesa->ClearDepth = 0x00ffffff;
+      depth_bpp = R128_Z_PIX_WIDTH_24;
+      rmesa->depth_scale = 1.0 / (GLfloat)0xffffff;
+      break;
+   default:
+      fprintf( stderr, "Error: Unsupported depth %d... exiting\n",
+	       rmesa->glCtx->Visual.depthBits );
+      exit( -1 );
+   }
+
+   rmesa->Fallback = 0;
+
+   if ( rmesa->glCtx->Visual.doubleBufferMode && rmesa->sarea->pfCurrentPage == 0 ) {
+      rmesa->drawOffset = rmesa->readOffset = rmesa->r128Screen->backOffset;
+      rmesa->drawPitch  = rmesa->readPitch  = rmesa->r128Screen->backPitch;
+   } else {
+      rmesa->drawOffset = rmesa->readOffset = rmesa->r128Screen->frontOffset;
+      rmesa->drawPitch  = rmesa->readPitch  = rmesa->r128Screen->frontPitch;
+   }
+
+   /* Harware state:
+    */
+   rmesa->setup.dst_pitch_offset_c = (((rmesa->drawPitch/8) << 21) |
+				      (rmesa->drawOffset >> 5));
+
+   rmesa->setup.dp_gui_master_cntl_c = (R128_GMC_DST_PITCH_OFFSET_CNTL |
+					R128_GMC_DST_CLIPPING |
+					R128_GMC_BRUSH_SOLID_COLOR |
+					dst_bpp |
+					R128_GMC_SRC_DATATYPE_COLOR |
+					R128_GMC_BYTE_MSB_TO_LSB |
+					R128_GMC_CONVERSION_TEMP_6500 |
+					R128_ROP3_S |
+					R128_DP_SRC_SOURCE_MEMORY |
+					R128_GMC_3D_FCN_EN |
+					R128_GMC_CLR_CMP_CNTL_DIS |
+					R128_GMC_AUX_CLIP_DIS |
+					R128_GMC_WR_MSK_DIS);
+
+   rmesa->setup.sc_top_left_c     = 0x00000000;
+   rmesa->setup.sc_bottom_right_c = 0x1fff1fff;
+
+   rmesa->setup.z_offset_c = rmesa->r128Screen->depthOffset;
+   rmesa->setup.z_pitch_c = ((rmesa->r128Screen->depthPitch >> 3) |
+			     R128_Z_TILE);
+
+   rmesa->setup.z_sten_cntl_c = (depth_bpp |
+				 R128_Z_TEST_LESS |
+				 R128_STENCIL_TEST_ALWAYS |
+				 R128_STENCIL_S_FAIL_KEEP |
+				 R128_STENCIL_ZPASS_KEEP |
+				 R128_STENCIL_ZFAIL_KEEP);
+
+   rmesa->setup.tex_cntl_c = (R128_Z_WRITE_ENABLE |
+			      R128_SHADE_ENABLE |
+			      R128_DITHER_ENABLE |
+			      R128_ALPHA_IN_TEX_COMPLETE_A |
+			      R128_LIGHT_DIS |
+			      R128_ALPHA_LIGHT_DIS |
+			      R128_TEX_CACHE_FLUSH |
+			      (0x3f << R128_LOD_BIAS_SHIFT));
+
+   rmesa->setup.misc_3d_state_cntl_reg = (R128_MISC_SCALE_3D_TEXMAP_SHADE |
+					  R128_MISC_SCALE_PIX_REPLICATE |
+					  R128_ALPHA_COMB_ADD_CLAMP |
+					  R128_FOG_VERTEX |
+					  R128_ALPHA_BLEND_SRC_ONE |
+					  R128_ALPHA_BLEND_DST_ZERO |
+					  R128_ALPHA_TEST_ALWAYS);
+
+   rmesa->setup.texture_clr_cmp_clr_c = 0x00000000;
+   rmesa->setup.texture_clr_cmp_msk_c = 0xffffffff;
+
+   rmesa->setup.fog_color_c = 0x00000000;
+
+   rmesa->setup.pm4_vc_fpu_setup = (R128_FRONT_DIR_CCW |
+				    R128_BACKFACE_SOLID |
+				    R128_FRONTFACE_SOLID |
+				    R128_FPU_COLOR_GOURAUD |
+				    R128_FPU_SUB_PIX_4BITS |
+				    R128_FPU_MODE_3D |
+				    R128_TRAP_BITS_DISABLE |
+				    R128_XFACTOR_2 |
+				    R128_YFACTOR_2 |
+				    R128_FLAT_SHADE_VERTEX_OGL |
+				    R128_FPU_ROUND_TRUNCATE |
+				    R128_WM_SEL_8DW);
+
+   rmesa->setup.setup_cntl = (R128_COLOR_GOURAUD |
+			      R128_PRIM_TYPE_TRI |
+			      R128_TEXTURE_ST_MULT_W |
+			      R128_STARTING_VERTEX_1 |
+			      R128_ENDING_VERTEX_3 |
+			      R128_SU_POLY_LINE_NOT_LAST |
+			      R128_SUB_PIX_4BITS);
+
+   rmesa->setup.tex_size_pitch_c = 0x00000000;
+   rmesa->setup.constant_color_c = 0x00ffffff;
+
+   rmesa->setup.dp_write_mask   = 0xffffffff;
+   rmesa->setup.sten_ref_mask_c = 0xffff0000;
+   rmesa->setup.plane_3d_mask_c = 0xffffffff;
+
+   rmesa->setup.window_xy_offset = 0x00000000;
+
+   rmesa->setup.scale_3d_cntl = (R128_SCALE_DITHER_TABLE |
+				 R128_TEX_CACHE_SIZE_FULL |
+				 R128_DITHER_INIT_RESET |
+				 R128_SCALE_3D_TEXMAP_SHADE |
+				 R128_SCALE_PIX_REPLICATE |
+				 R128_ALPHA_COMB_ADD_CLAMP |
+				 R128_FOG_VERTEX |
+				 R128_ALPHA_BLEND_SRC_ONE |
+				 R128_ALPHA_BLEND_DST_ZERO |
+				 R128_ALPHA_TEST_ALWAYS |
+				 R128_COMPOSITE_SHADOW_CMP_EQUAL |
+				 R128_TEX_MAP_ALPHA_IN_TEXTURE |
+				 R128_TEX_CACHE_LINE_SIZE_4QW);
+
+   rmesa->new_state = R128_NEW_ALL;
+}
+
+/* Initialize the driver's state functions.
+ */
+void r128DDInitStateFuncs( GLcontext *ctx )
+{
+   ctx->Driver.UpdateState		= r128DDInvalidateState;
+
+   ctx->Driver.ClearIndex		= NULL;
+   ctx->Driver.ClearColor		= r128DDClearColor;
+   ctx->Driver.DrawBuffer		= r128DDDrawBuffer;
+   ctx->Driver.ReadBuffer		= r128DDReadBuffer;
+
+   ctx->Driver.IndexMask		= NULL;
+   ctx->Driver.ColorMask		= r128DDColorMask;
+   ctx->Driver.AlphaFunc		= r128DDAlphaFunc;
+   ctx->Driver.BlendEquation		= r128DDBlendEquation;
+   ctx->Driver.BlendFunc		= r128DDBlendFunc;
+   ctx->Driver.BlendFuncSeparate	= r128DDBlendFuncSeparate;
+   ctx->Driver.ClearDepth		= r128DDClearDepth;
+   ctx->Driver.CullFace			= r128DDCullFace;
+   ctx->Driver.FrontFace		= r128DDFrontFace;
+   ctx->Driver.DepthFunc		= r128DDDepthFunc;
+   ctx->Driver.DepthMask		= r128DDDepthMask;
+   ctx->Driver.Enable			= r128DDEnable;
+   ctx->Driver.Fogfv			= r128DDFogfv;
+   ctx->Driver.Hint			= NULL;
+   ctx->Driver.Lightfv			= NULL;
+   ctx->Driver.LightModelfv		= r128DDLightModelfv;
+   ctx->Driver.LogicOpcode		= r128DDLogicOpCode;
+   ctx->Driver.PolygonMode		= NULL;
+   ctx->Driver.PolygonStipple		= r128DDPolygonStipple;
+   ctx->Driver.RenderMode		= r128DDRenderMode;
+   ctx->Driver.Scissor			= r128DDScissor;
+   ctx->Driver.ShadeModel		= r128DDShadeModel;
+   ctx->Driver.ClearStencil		= NULL;
+   ctx->Driver.StencilFunc		= NULL;
+   ctx->Driver.StencilMask		= NULL;
+   ctx->Driver.StencilOp		= NULL;
+
+   ctx->Driver.DepthRange               = r128DepthRange;
+   ctx->Driver.Viewport                 = r128Viewport;
+
+   /* Pixel path fallbacks.
+    */
+   ctx->Driver.Accum = _swrast_Accum;
+   ctx->Driver.Bitmap = _swrast_Bitmap;
+   ctx->Driver.CopyPixels = _swrast_CopyPixels;
+   ctx->Driver.DrawPixels = _swrast_DrawPixels;
+   ctx->Driver.ReadPixels = _swrast_ReadPixels;
+
+   /* Swrast hooks for imaging extensions:
+    */
+   ctx->Driver.CopyColorTable = _swrast_CopyColorTable;
+   ctx->Driver.CopyColorSubTable = _swrast_CopyColorSubTable;
+   ctx->Driver.CopyConvolutionFilter1D = _swrast_CopyConvolutionFilter1D;
+   ctx->Driver.CopyConvolutionFilter2D = _swrast_CopyConvolutionFilter2D;
+}
diff --git a/src/mesa/drivers/dri/r128/r128_state.h b/src/mesa/drivers/dri/r128/r128_state.h
new file mode 100644
index 00000000000..c51ac239427
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_state.h
@@ -0,0 +1,54 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_state.h,v 1.3 2001/01/08 01:07:21 martin Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Gareth Hughes <[email protected]>
+ *   Kevin E. Martin <[email protected]>
+ *
+ */
+
+#ifndef __R128_STATE_H__
+#define __R128_STATE_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+#include "r128_context.h"
+
+extern void r128DDInitState( r128ContextPtr rmesa );
+extern void r128DDInitStateFuncs( GLcontext *ctx );
+
+extern void r128DDUpdateState( GLcontext *ctx );
+extern void r128DDUpdateHWState( GLcontext *ctx );
+
+extern void r128UpdateWindow( GLcontext *ctx );
+
+extern void r128EmitHwStateLocked( r128ContextPtr rmesa );
+
+#endif
+#endif
diff --git a/src/mesa/drivers/dri/r128/r128_tex.c b/src/mesa/drivers/dri/r128/r128_tex.c
new file mode 100644
index 00000000000..2793a793a0e
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_tex.c
@@ -0,0 +1,598 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_tex.c,v 1.14 2002/11/05 17:46:08 tsi Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Gareth Hughes <[email protected]>
+ *   Kevin E. Martin <[email protected]>
+ *   Brian Paul <[email protected]>
+ */
+
+#include "r128_context.h"
+#include "r128_state.h"
+#include "r128_ioctl.h"
+#include "r128_vb.h"
+#include "r128_tris.h"
+#include "r128_tex.h"
+#include "r128_texobj.h"
+
+#include "context.h"
+#include "macros.h"
+#include "simple_list.h"
+#include "enums.h"
+#include "texstore.h"
+#include "texformat.h"
+#include "teximage.h"
+#include "imports.h"
+#include "colormac.h"
+
+#define TEX_0	1
+#define TEX_1	2
+
+
+static void r128SetTexWrap( r128TexObjPtr t, GLenum swrap, GLenum twrap )
+{
+   t->setup.tex_cntl &= ~(R128_TEX_CLAMP_S_MASK | R128_TEX_CLAMP_T_MASK);
+
+   switch ( swrap ) {
+   case GL_CLAMP:
+      t->setup.tex_cntl |= R128_TEX_CLAMP_S_BORDER_COLOR;
+      break;
+   case GL_CLAMP_TO_EDGE:
+      t->setup.tex_cntl |= R128_TEX_CLAMP_S_CLAMP;
+      break;
+   case GL_REPEAT:
+      t->setup.tex_cntl |= R128_TEX_CLAMP_S_WRAP;
+      break;
+   case GL_CLAMP_TO_BORDER:
+      t->setup.tex_cntl |= R128_TEX_CLAMP_S_BORDER_COLOR;
+      break;
+   case GL_MIRRORED_REPEAT:
+      t->setup.tex_cntl |= R128_TEX_CLAMP_S_MIRROR;
+      break;
+   }
+
+   switch ( twrap ) {
+   case GL_CLAMP:
+      t->setup.tex_cntl |= R128_TEX_CLAMP_T_BORDER_COLOR;
+      break;
+   case GL_CLAMP_TO_EDGE:
+      t->setup.tex_cntl |= R128_TEX_CLAMP_T_CLAMP;
+      break;
+   case GL_REPEAT:
+      t->setup.tex_cntl |= R128_TEX_CLAMP_T_WRAP;
+      break;
+   case GL_CLAMP_TO_BORDER:
+      t->setup.tex_cntl |= R128_TEX_CLAMP_T_BORDER_COLOR;
+      break;
+   case GL_MIRRORED_REPEAT:
+      t->setup.tex_cntl |= R128_TEX_CLAMP_T_MIRROR;
+      break;
+   }
+}
+
+static void r128SetTexFilter( r128TexObjPtr t, GLenum minf, GLenum magf )
+{
+   t->setup.tex_cntl &= ~(R128_MIN_BLEND_MASK | R128_MAG_BLEND_MASK);
+
+   switch ( minf ) {
+   case GL_NEAREST:
+      t->setup.tex_cntl |= R128_MIN_BLEND_NEAREST;
+      break;
+   case GL_LINEAR:
+      t->setup.tex_cntl |= R128_MIN_BLEND_LINEAR;
+      break;
+   case GL_NEAREST_MIPMAP_NEAREST:
+      t->setup.tex_cntl |= R128_MIN_BLEND_MIPNEAREST;
+      break;
+   case GL_LINEAR_MIPMAP_NEAREST:
+      t->setup.tex_cntl |= R128_MIN_BLEND_MIPLINEAR;
+      break;
+   case GL_NEAREST_MIPMAP_LINEAR:
+      t->setup.tex_cntl |= R128_MIN_BLEND_LINEARMIPNEAREST;
+      break;
+   case GL_LINEAR_MIPMAP_LINEAR:
+      t->setup.tex_cntl |= R128_MIN_BLEND_LINEARMIPLINEAR;
+      break;
+   }
+
+   switch ( magf ) {
+   case GL_NEAREST:
+      t->setup.tex_cntl |= R128_MAG_BLEND_NEAREST;
+      break;
+   case GL_LINEAR:
+      t->setup.tex_cntl |= R128_MAG_BLEND_LINEAR;
+      break;
+   }
+}
+
+static void r128SetTexBorderColor( r128TexObjPtr t, GLubyte c[4] )
+{
+   t->setup.tex_border_color = r128PackColor( 4, c[0], c[1], c[2], c[3] );
+}
+
+
+static r128TexObjPtr r128AllocTexObj( struct gl_texture_object *texObj )
+{
+   r128TexObjPtr t;
+
+   if ( R128_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s( %p )\n", __FUNCTION__, texObj );
+   }
+
+   t = (r128TexObjPtr) CALLOC_STRUCT( r128_tex_obj );
+   texObj->DriverData = t;
+   if ( t != NULL ) {
+
+      /* Initialize non-image-dependent parts of the state:
+       */
+      t->base.tObj = texObj;
+
+      /* FIXME Something here to set initial values for other parts of
+       * FIXME t->setup?
+       */
+  
+      make_empty_list( (driTextureObject *) t );
+
+      r128SetTexWrap( t, texObj->WrapS, texObj->WrapT );
+      r128SetTexFilter( t, texObj->MinFilter, texObj->MagFilter );
+      r128SetTexBorderColor( t, texObj->_BorderChan );
+   }
+
+   return t;
+}
+
+
+/* Called by the _mesa_store_teximage[123]d() functions. */
+static const struct gl_texture_format *
+r128ChooseTextureFormat( GLcontext *ctx, GLint internalFormat,
+                         GLenum format, GLenum type )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   (void) format;
+   (void) type;
+
+   switch ( internalFormat ) {
+   case GL_ALPHA:
+   case GL_ALPHA4:
+   case GL_ALPHA8:
+   case GL_ALPHA12:
+   case GL_ALPHA16:
+   case GL_COMPRESSED_ALPHA:
+   case 2:
+   case GL_LUMINANCE_ALPHA:
+   case GL_LUMINANCE4_ALPHA4:
+   case GL_LUMINANCE6_ALPHA2:
+   case GL_LUMINANCE8_ALPHA8:
+   case GL_LUMINANCE12_ALPHA4:
+   case GL_LUMINANCE12_ALPHA12:
+   case GL_LUMINANCE16_ALPHA16:
+   case GL_COMPRESSED_LUMINANCE_ALPHA:
+   case 4:
+   case GL_RGBA:
+   case GL_COMPRESSED_RGBA:
+   case GL_RGBA2:
+   case GL_RGB5_A1:
+   case GL_RGBA8:
+   case GL_RGB10_A2:
+   case GL_RGBA12:
+   case GL_RGBA16:
+      if (rmesa->r128Screen->cpp == 4)
+         return &_mesa_texformat_argb8888;
+      else
+         return &_mesa_texformat_argb4444;
+   case GL_RGBA4:
+         return &_mesa_texformat_argb4444;
+
+   case 3:
+   case GL_RGB:
+   case GL_COMPRESSED_RGB:
+   case GL_R3_G3_B2:
+   case GL_RGB4:
+   case GL_RGB5:
+   case GL_RGB8:
+   case GL_RGB10:
+   case GL_RGB12:
+   case GL_RGB16:
+      if (rmesa->r128Screen->cpp == 4)
+         return &_mesa_texformat_argb8888;
+      else
+         return &_mesa_texformat_rgb565;
+
+   case 1:
+   case GL_LUMINANCE:
+   case GL_LUMINANCE4:
+   case GL_LUMINANCE8:
+   case GL_LUMINANCE12:
+   case GL_LUMINANCE16:
+   case GL_COMPRESSED_LUMINANCE:
+      if (rmesa->r128Screen->cpp == 4)
+         return &_mesa_texformat_argb8888; /* inefficient but accurate */
+      else
+         return &_mesa_texformat_rgb565;
+
+   case GL_INTENSITY4:
+      return &_mesa_texformat_argb4444;
+   case GL_INTENSITY:
+   case GL_INTENSITY8:
+   case GL_INTENSITY12:
+   case GL_INTENSITY16:
+   case GL_COMPRESSED_INTENSITY:
+      if (rmesa->r128Screen->cpp == 4)
+         return &_mesa_texformat_argb8888; /* inefficient but accurate */
+      else
+         return &_mesa_texformat_argb4444;
+
+   case GL_COLOR_INDEX:
+   case GL_COLOR_INDEX1_EXT:
+   case GL_COLOR_INDEX2_EXT:
+   case GL_COLOR_INDEX4_EXT:
+   case GL_COLOR_INDEX8_EXT:
+   case GL_COLOR_INDEX12_EXT:
+   case GL_COLOR_INDEX16_EXT:
+      return &_mesa_texformat_ci8;
+
+   case GL_YCBCR_MESA:
+      if (type == GL_UNSIGNED_SHORT_8_8_APPLE ||
+          type == GL_UNSIGNED_BYTE)
+         return &_mesa_texformat_ycbcr;
+      else
+         return &_mesa_texformat_ycbcr_rev;
+
+   default:
+      _mesa_problem( ctx, "unexpected format in %s", __FUNCTION__ );
+      return NULL;
+   }
+}
+
+
+static void r128TexImage1D( GLcontext *ctx, GLenum target, GLint level,
+			    GLint internalFormat,
+			    GLint width, GLint border,
+			    GLenum format, GLenum type, const GLvoid *pixels,
+			    const struct gl_pixelstore_attrib *packing,
+			    struct gl_texture_object *texObj,
+			    struct gl_texture_image *texImage )
+{
+   driTextureObject * t = (driTextureObject *) texObj->DriverData;
+
+   if ( t ) {
+      driSwapOutTextureObject( t );
+   }
+   else {
+      t = (driTextureObject *) r128AllocTexObj(texObj);
+      if (!t) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage1D");
+         return;
+      }
+   }
+
+   /* Note, this will call r128ChooseTextureFormat */
+   _mesa_store_teximage1d( ctx, target, level, internalFormat,
+			   width, border, format, type,
+			   pixels, packing, texObj, texImage );
+
+   t->dirty_images[0] |= (1 << level);
+}
+
+
+static void r128TexSubImage1D( GLcontext *ctx,
+			       GLenum target,
+			       GLint level,
+			       GLint xoffset,
+			       GLsizei width,
+			       GLenum format, GLenum type,
+			       const GLvoid *pixels,
+			       const struct gl_pixelstore_attrib *packing,
+			       struct gl_texture_object *texObj,
+			       struct gl_texture_image *texImage )
+{
+   driTextureObject * t = (driTextureObject *) texObj->DriverData;
+
+   assert( t ); /* this _should_ be true */
+   if ( t ) {
+      driSwapOutTextureObject( t );
+   }
+   else {
+      t = (driTextureObject *) r128AllocTexObj(texObj);
+      if (!t) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage1D");
+         return;
+      }
+   }
+
+   _mesa_store_texsubimage1d(ctx, target, level, xoffset, width,
+			     format, type, pixels, packing, texObj,
+			     texImage);
+
+   t->dirty_images[0] |= (1 << level);
+}
+
+
+static void r128TexImage2D( GLcontext *ctx, GLenum target, GLint level,
+			    GLint internalFormat,
+			    GLint width, GLint height, GLint border,
+			    GLenum format, GLenum type, const GLvoid *pixels,
+			    const struct gl_pixelstore_attrib *packing,
+			    struct gl_texture_object *texObj,
+			    struct gl_texture_image *texImage )
+{
+   driTextureObject * t = (driTextureObject *) texObj->DriverData;
+
+   if ( t ) {
+      driSwapOutTextureObject( (driTextureObject *) t );
+   }
+   else {
+      t = (driTextureObject *) r128AllocTexObj(texObj);
+      if (!t) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage2D");
+         return;
+      }
+   }
+
+   /* Note, this will call r128ChooseTextureFormat */
+   _mesa_store_teximage2d(ctx, target, level, internalFormat,
+                          width, height, border, format, type, pixels,
+                          &ctx->Unpack, texObj, texImage);
+
+   t->dirty_images[0] |= (1 << level);
+}
+
+
+static void r128TexSubImage2D( GLcontext *ctx,
+			       GLenum target,
+			       GLint level,
+			       GLint xoffset, GLint yoffset,
+			       GLsizei width, GLsizei height,
+			       GLenum format, GLenum type,
+			       const GLvoid *pixels,
+			       const struct gl_pixelstore_attrib *packing,
+			       struct gl_texture_object *texObj,
+			       struct gl_texture_image *texImage )
+{
+   driTextureObject * t = (driTextureObject *) texObj->DriverData;
+
+   assert( t ); /* this _should_ be true */
+   if ( t ) {
+      driSwapOutTextureObject( t );
+   }
+   else {
+      t = (driTextureObject *) r128AllocTexObj(texObj);
+      if (!t) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage2D");
+         return;
+      }
+   }
+
+   _mesa_store_texsubimage2d(ctx, target, level, xoffset, yoffset, width,
+			     height, format, type, pixels, packing, texObj,
+			     texImage);
+   t->dirty_images[0] |= (1 << level);
+}
+
+
+static void r128DDTexEnv( GLcontext *ctx, GLenum target,
+			  GLenum pname, const GLfloat *param )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   struct gl_texture_unit *texUnit;
+   GLubyte c[4];
+
+   if ( R128_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s( %s )\n",
+	       __FUNCTION__, _mesa_lookup_enum_by_nr( pname ) );
+   }
+
+   switch ( pname ) {
+   case GL_TEXTURE_ENV_MODE:
+      FLUSH_BATCH( rmesa );
+      rmesa->new_state |= R128_NEW_ALPHA;
+      break;
+
+   case GL_TEXTURE_ENV_COLOR:
+      texUnit = &ctx->Texture.Unit[ctx->Texture.CurrentUnit];
+      CLAMPED_FLOAT_TO_UBYTE( c[0], texUnit->EnvColor[0] );
+      CLAMPED_FLOAT_TO_UBYTE( c[1], texUnit->EnvColor[1] );
+      CLAMPED_FLOAT_TO_UBYTE( c[2], texUnit->EnvColor[2] );
+      CLAMPED_FLOAT_TO_UBYTE( c[3], texUnit->EnvColor[3] );
+      rmesa->env_color = r128PackColor( 4, c[0], c[1], c[2], c[3] );
+      if ( rmesa->setup.constant_color_c != rmesa->env_color ) {
+	 FLUSH_BATCH( rmesa );
+	 rmesa->setup.constant_color_c = rmesa->env_color;
+
+	 /* More complex multitexture/multipass fallbacks for GL_BLEND
+	  * can be done later, but this allows a single pass GL_BLEND
+	  * in some cases (ie. Performer town demo).  This is only
+	  * applicable to the regular Rage 128, as the Pro and M3 can
+	  * handle true single-pass GL_BLEND texturing.
+	  */
+	 rmesa->blend_flags &= ~R128_BLEND_ENV_COLOR;
+	 if ( R128_IS_PLAIN( rmesa ) &&
+	      rmesa->env_color != 0x00000000 &&
+	      rmesa->env_color != 0xff000000 &&
+	      rmesa->env_color != 0x00ffffff &&
+	      rmesa->env_color != 0xffffffff ) {
+	    rmesa->blend_flags |= R128_BLEND_ENV_COLOR;
+	 }
+      }
+      break;
+
+   case GL_TEXTURE_LOD_BIAS_EXT:
+      do {
+	 CARD32 t = rmesa->setup.tex_cntl_c;
+	 GLint bias;
+	 CARD32 b;
+
+	 /* GTH: This isn't exactly correct, but gives good results up to a
+	  * certain point.  It is better than completely ignoring the LOD
+	  * bias.  Unfortunately there isn't much range in the bias, the
+	  * spec mentions strides that vary between 0.5 and 2.0 but these
+	  * numbers don't seem to relate the the GL LOD bias value at all.
+	  */
+	 if ( param[0] >= 1.0 ) {
+	    bias = -128;
+	 } else if ( param[0] >= 0.5 ) {
+	    bias = -64;
+	 } else if ( param[0] >= 0.25 ) {
+	    bias = 0;
+	 } else if ( param[0] >= 0.0 ) {
+	    bias = 63;
+	 } else {
+	    bias = 127;
+	 }
+
+	 b = (CARD32)bias & 0xff;
+	 t &= ~R128_LOD_BIAS_MASK;
+	 t |= (b << R128_LOD_BIAS_SHIFT);
+
+	 if ( rmesa->setup.tex_cntl_c != t ) {
+	    FLUSH_BATCH( rmesa );
+	    rmesa->setup.tex_cntl_c = t;
+	    rmesa->dirty |= R128_UPLOAD_CONTEXT;
+	 }
+      } while (0);
+      break;
+
+   default:
+      return;
+   }
+}
+
+
+static void r128DDTexParameter( GLcontext *ctx, GLenum target,
+				struct gl_texture_object *tObj,
+				GLenum pname, const GLfloat *params )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   r128TexObjPtr t = (r128TexObjPtr)tObj->DriverData;
+
+   if ( R128_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s( %s )\n",
+	       __FUNCTION__, _mesa_lookup_enum_by_nr( pname ) );
+   }
+
+   if ( ( target != GL_TEXTURE_2D ) && ( target != GL_TEXTURE_1D ) )
+      return;
+
+   switch ( pname ) {
+   case GL_TEXTURE_MIN_FILTER:
+   case GL_TEXTURE_MAG_FILTER:
+      if ( t->base.bound ) FLUSH_BATCH( rmesa );
+      r128SetTexFilter( t, tObj->MinFilter, tObj->MagFilter );
+      break;
+
+   case GL_TEXTURE_WRAP_S:
+   case GL_TEXTURE_WRAP_T:
+      if ( t->base.bound ) FLUSH_BATCH( rmesa );
+      r128SetTexWrap( t, tObj->WrapS, tObj->WrapT );
+      break;
+
+   case GL_TEXTURE_BORDER_COLOR:
+      if ( t->base.bound ) FLUSH_BATCH( rmesa );
+      r128SetTexBorderColor( t, tObj->_BorderChan );
+      break;
+
+   case GL_TEXTURE_BASE_LEVEL:
+   case GL_TEXTURE_MAX_LEVEL:
+   case GL_TEXTURE_MIN_LOD:
+   case GL_TEXTURE_MAX_LOD:
+      /* This isn't the most efficient solution but there doesn't appear to
+       * be a nice alternative for R128.  Since there's no LOD clamping,
+       * we just have to rely on loading the right subset of mipmap levels
+       * to simulate a clamped LOD.
+       */
+      if ( t->base.bound ) FLUSH_BATCH( rmesa );
+      driSwapOutTextureObject( (driTextureObject *) t );
+      break;
+
+   default:
+      return;
+   }
+}
+
+static void r128DDBindTexture( GLcontext *ctx, GLenum target,
+			       struct gl_texture_object *tObj )
+{
+   if ( R128_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s( %p ) unit=%d\n", __FUNCTION__, tObj,
+	       ctx->Texture.CurrentUnit );
+   }
+
+   if ( target == GL_TEXTURE_2D || target == GL_TEXTURE_1D ) {
+      if ( tObj->DriverData == NULL ) {
+	 r128AllocTexObj( tObj );
+      }
+   }
+}
+
+static void r128DDDeleteTexture( GLcontext *ctx,
+				 struct gl_texture_object *tObj )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   driTextureObject * t = (driTextureObject *) tObj->DriverData;
+
+   if ( t ) {
+      if ( t->bound && rmesa ) {
+	 FLUSH_BATCH( rmesa );
+      }
+
+      driDestroyTextureObject( t );
+   }
+}
+
+void r128DDInitTextureFuncs( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+
+   ctx->Driver.TexEnv			= r128DDTexEnv;
+   ctx->Driver.ChooseTextureFormat	= r128ChooseTextureFormat;
+   ctx->Driver.TexImage1D		= r128TexImage1D;
+   ctx->Driver.TexSubImage1D		= r128TexSubImage1D;
+   ctx->Driver.TexImage2D		= r128TexImage2D;
+   ctx->Driver.TexSubImage2D		= r128TexSubImage2D;
+   ctx->Driver.TexImage3D               = _mesa_store_teximage3d;
+   ctx->Driver.TexSubImage3D            = _mesa_store_texsubimage3d;
+   ctx->Driver.CopyTexImage1D           = _swrast_copy_teximage1d;
+   ctx->Driver.CopyTexImage2D           = _swrast_copy_teximage2d;
+   ctx->Driver.CopyTexSubImage1D        = _swrast_copy_texsubimage1d;
+   ctx->Driver.CopyTexSubImage2D        = _swrast_copy_texsubimage2d;
+   ctx->Driver.CopyTexSubImage3D        = _swrast_copy_texsubimage3d;
+   ctx->Driver.TestProxyTexImage        = _mesa_test_proxy_teximage;
+   ctx->Driver.TexParameter		= r128DDTexParameter;
+   ctx->Driver.BindTexture		= r128DDBindTexture;
+   ctx->Driver.DeleteTexture		= r128DDDeleteTexture;
+   ctx->Driver.UpdateTexturePalette	= NULL;
+   ctx->Driver.ActiveTexture		= NULL;
+   ctx->Driver.IsTextureResident	= driIsTextureResident;
+   ctx->Driver.PrioritizeTexture	= NULL;
+
+   driInitTextureObjects( ctx, & rmesa->swapped,
+			  DRI_TEXMGR_DO_TEXTURE_1D
+			  | DRI_TEXMGR_DO_TEXTURE_2D );
+}
diff --git a/src/mesa/drivers/dri/r128/r128_tex.h b/src/mesa/drivers/dri/r128/r128_tex.h
new file mode 100644
index 00000000000..2e3d954fbb8
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_tex.h
@@ -0,0 +1,88 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_tex.h,v 1.7 2002/02/22 21:44:58 dawes Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Gareth Hughes <[email protected]>
+ *   Kevin E. Martin <[email protected]>
+ *
+ */
+
+#ifndef __R128_TEX_H__
+#define __R128_TEX_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+extern void r128UpdateTextureState( GLcontext *ctx );
+
+extern void r128UploadTexImages( r128ContextPtr rmesa, r128TexObjPtr t );
+
+extern void r128DestroyTexObj( r128ContextPtr rmesa, r128TexObjPtr t );
+
+extern void r128DDInitTextureFuncs( GLcontext *ctx );
+
+
+/* ================================================================
+ * Color conversion macros:
+ */
+
+#define R128PACKCOLOR332( r, g, b )					\
+   (((r) & 0xe0) | (((g) & 0xe0) >> 3) | (((b) & 0xc0) >> 6))
+
+#define R128PACKCOLOR1555( r, g, b, a )					\
+   ((((r) & 0xf8) << 7) | (((g) & 0xf8) << 2) | (((b) & 0xf8) >> 3) |	\
+    ((a) ? 0x8000 : 0))
+
+#define R128PACKCOLOR565( r, g, b )					\
+   ((((r) & 0xf8) << 8) | (((g) & 0xfc) << 3) | (((b) & 0xf8) >> 3))
+
+#define R128PACKCOLOR888( r, g, b )					\
+   (((r) << 16) | ((g) << 8) | (b))
+
+#define R128PACKCOLOR8888( r, g, b, a )					\
+   (((a) << 24) | ((r) << 16) | ((g) << 8) | (b))
+
+#define R128PACKCOLOR4444( r, g, b, a )					\
+   ((((a) & 0xf0) << 8) | (((r) & 0xf0) << 4) | ((g) & 0xf0) | ((b) >> 4))
+
+static __inline__ CARD32 r128PackColor( GLuint cpp,
+					GLubyte r, GLubyte g,
+					GLubyte b, GLubyte a )
+{
+    switch ( cpp ) {
+    case 2:
+       return R128PACKCOLOR565( r, g, b );
+    case 4:
+       return R128PACKCOLOR8888( r, g, b, a );
+    default:
+       return 0;
+    }
+}
+
+#endif
+#endif /* __R128_TEX_H__ */
diff --git a/src/mesa/drivers/dri/r128/r128_texmem.c b/src/mesa/drivers/dri/r128/r128_texmem.c
new file mode 100644
index 00000000000..f6a5dfbbcf3
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_texmem.c
@@ -0,0 +1,302 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_tex.c,v 1.7 2001/01/08 01:07:21 martin Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Gareth Hughes <[email protected]>
+ *   Kevin E. Martin <[email protected]>
+ *   Brian Paul <[email protected]>
+ */
+
+#include "r128_context.h"
+#include "r128_state.h"
+#include "r128_ioctl.h"
+#include "r128_vb.h"
+#include "r128_tris.h"
+#include "r128_tex.h"
+
+#include "context.h"
+#include "macros.h"
+#include "simple_list.h"
+#include "texformat.h"
+#include "imports.h"
+
+#define TEX_0	1
+#define TEX_1	2
+
+
+/* Destroy hardware state associated with texture `t'.
+ */
+void r128DestroyTexObj( r128ContextPtr rmesa, r128TexObjPtr t )
+{
+    unsigned   i;
+
+
+    /* See if it was the driver's current object.
+     */
+
+    if ( rmesa != NULL )
+    { 
+	for ( i = 0 ; i < rmesa->glCtx->Const.MaxTextureUnits ; i++ )
+	{
+	    if ( t == rmesa->CurrentTexObj[ i ] ) {
+		assert( t->base.bound & (1 << i) );
+		rmesa->CurrentTexObj[ i ] = NULL;
+	    }
+	}
+    }
+}
+
+
+/**
+ * Upload the texture image associated with texture \a t at the specified
+ * level at the address relative to \a start.
+ */
+static void uploadSubImage( r128ContextPtr rmesa, r128TexObjPtr t,
+			    GLint level,
+			    GLint x, GLint y, GLint width, GLint height )
+{
+   struct gl_texture_image *image;
+   int texelsPerDword = 0;
+   int imageWidth, imageHeight;
+   int remaining, rows;
+   int format, dwords;
+   CARD32 pitch, offset;
+   int i;
+
+   /* Ensure we have a valid texture to upload */
+   if ( ( level < 0 ) || ( level > R128_MAX_TEXTURE_LEVELS ) )
+      return;
+
+   image = t->base.tObj->Image[level];
+   if ( !image )
+      return;
+
+   switch ( image->TexFormat->TexelBytes ) {
+   case 1: texelsPerDword = 4; break;
+   case 2: texelsPerDword = 2; break;
+   case 4: texelsPerDword = 1; break;
+   }
+
+#if 1
+   /* FIXME: The subimage index calcs are wrong... */
+   x = 0;
+   y = 0;
+   width = image->Width;
+   height = image->Height;
+#endif
+
+   imageWidth  = image->Width;
+   imageHeight = image->Height;
+
+   format = t->textureFormat >> 16;
+
+   /* The texel upload routines have a minimum width, so force the size
+    * if needed.
+    */
+   if ( imageWidth < texelsPerDword ) {
+      int factor;
+
+      factor = texelsPerDword / imageWidth;
+      imageWidth = texelsPerDword;
+      imageHeight /= factor;
+      if ( imageHeight == 0 ) {
+	 /* In this case, the texel converter will actually walk a
+	  * texel or two off the end of the image, but normal malloc
+	  * alignment should prevent it from ever causing a fault.
+	  */
+	 imageHeight = 1;
+      }
+   }
+
+   /* We can't upload to a pitch less than 8 texels so we will need to
+    * linearly upload all modified rows for textures smaller than this.
+    * This makes the x/y/width/height different for the blitter and the
+    * texture walker.
+    */
+   if ( imageWidth >= 8 ) {
+      /* The texture walker and the blitter look identical */
+      pitch = imageWidth >> 3;
+   } else {
+      int factor;
+      int y2;
+      int start, end;
+
+      start = (y * imageWidth) & ~7;
+      end = (y + height) * imageWidth;
+
+      if ( end - start < 8 ) {
+	 /* Handle the case where the total number of texels
+	  * uploaded is < 8.
+	  */
+	 x = 0;
+	 y = start / 8;
+	 width = end - start;
+	 height = 1;
+      } else {
+	 /* Upload some number of full 8 texel blit rows */
+	 factor = 8 / imageWidth;
+
+	 y2 = y + height - 1;
+	 y /= factor;
+	 y2 /= factor;
+
+	 x = 0;
+	 width = 8;
+	 height = y2 - y + 1;
+      }
+
+      /* Fixed pitch of 8 */
+      pitch = 1;
+   }
+
+   dwords = width * height / texelsPerDword;
+   offset = t->bufAddr + t->image[level - t->base.firstLevel].offset;
+
+#if ENABLE_PERF_BOXES
+   /* Bump the performace counter */
+   rmesa->c_textureBytes += (dwords << 2);
+#endif
+
+   if ( R128_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "r128UploadSubImage: %d,%d of %d,%d at %d,%d\n",
+	       width, height, image->Width, image->Height, x, y );
+      fprintf( stderr, "          blit ofs: 0x%07x pitch: 0x%x dwords: %d "
+	       "level: %d format: %x\n",
+	       (GLuint)offset, (GLuint)pitch, dwords, level, format );
+   }
+
+   /* Subdivide the texture if required */
+   if ( dwords <= R128_BUFFER_MAX_DWORDS / 2 ) {
+      rows = height;
+   } else {
+      rows = (R128_BUFFER_MAX_DWORDS * texelsPerDword) / (2 * width);
+   }
+
+   for ( i = 0, remaining = height ;
+	 remaining > 0 ;
+	 remaining -= rows, y += rows, i++ )
+   {
+      CARD32 *dst;
+      drmBufPtr buffer;
+
+      assert(image->Data);
+
+      height = MIN2(remaining, rows);
+
+      /* Grab the indirect buffer for the texture blit */
+      LOCK_HARDWARE( rmesa );
+      buffer = r128GetBufferLocked( rmesa );
+
+      dst = (CARD32 *)((char *)buffer->address + R128_HOSTDATA_BLIT_OFFSET);
+
+      /* Copy the next chunck of the texture image into the blit buffer */
+      {
+         const GLubyte *src = (const GLubyte *) image->Data +
+            (y * image->Width + x) * image->TexFormat->TexelBytes;
+         const GLuint bytes = width * height * image->TexFormat->TexelBytes;
+         memcpy(dst, src, bytes);
+      }
+
+      r128FireBlitLocked( rmesa, buffer,
+			  offset, pitch, format,
+			  x, y, width, height );
+      UNLOCK_HARDWARE( rmesa );
+   }
+
+   rmesa->new_state |= R128_NEW_CONTEXT;
+   rmesa->dirty |= R128_UPLOAD_CONTEXT | R128_UPLOAD_MASKS;
+}
+
+
+/* Upload the texture images associated with texture `t'.  This might
+ * require removing our own and/or other client's texture objects to
+ * make room for these images.
+ */
+void r128UploadTexImages( r128ContextPtr rmesa, r128TexObjPtr t )
+{
+   const GLint numLevels = t->base.lastLevel - t->base.firstLevel + 1;
+   GLint i;
+
+   if ( R128_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s( %p, %p )\n",
+	       __FUNCTION__, rmesa->glCtx, t );
+   }
+
+   assert(t);
+
+   LOCK_HARDWARE( rmesa );
+
+   if ( !t->base.memBlock ) {
+      int heap;
+
+
+      heap = driAllocateTexture( rmesa->texture_heaps, rmesa->nr_heaps,
+				 (driTextureObject *) t );
+      if ( heap == -1 ) {
+	 UNLOCK_HARDWARE( rmesa );
+	 return;
+      }
+
+      /* Set the base offset of the texture image */
+      t->bufAddr = rmesa->r128Screen->texOffset[heap] 
+	   + t->base.memBlock->ofs;
+
+      /* Set texture offsets for each mipmap level */
+      if ( t->setup.tex_cntl & R128_MIP_MAP_DISABLE ) {
+	 for ( i = 0 ; i < R128_MAX_TEXTURE_LEVELS ; i++ ) {
+	    t->setup.tex_offset[i] = t->bufAddr;
+	 }
+      } else {
+         for ( i = 0; i < numLevels; i++ ) {
+            const int j = numLevels - i - 1;
+            t->setup.tex_offset[j] = t->bufAddr + t->image[i].offset;
+         }
+      }
+   }
+
+   /* Let the world know we've used this memory recently.
+    */
+   driUpdateTextureLRU( (driTextureObject *) t );
+   UNLOCK_HARDWARE( rmesa );
+
+   /* Upload any images that are new */
+   if ( t->base.dirty_images[0] ) {
+      for ( i = 0 ; i < numLevels; i++ ) {
+         const GLint j = t->base.firstLevel + i;  /* the texObj's level */
+	 if ( t->base.dirty_images[0] & (1 << j) ) {
+	    uploadSubImage( rmesa, t, j, 0, 0,
+			    t->image[i].width, t->image[i].height );
+	 }
+      }
+
+      rmesa->setup.tex_cntl_c |= R128_TEX_CACHE_FLUSH;
+      rmesa->dirty |= R128_UPLOAD_CONTEXT;
+      t->base.dirty_images[0] = 0;
+   }
+}
diff --git a/src/mesa/drivers/dri/r128/r128_texobj.h b/src/mesa/drivers/dri/r128/r128_texobj.h
new file mode 100644
index 00000000000..e2ff1ac24d3
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_texobj.h
@@ -0,0 +1,69 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_texobj.h,v 1.5 2002/02/22 21:44:58 dawes Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ *
+ */
+
+#ifndef _R128_TEXOBJ_H_
+#define _R128_TEXOBJ_H_
+
+#include "r128_sarea.h"
+#include "mm.h"
+
+/* Individual texture image information.
+ */
+typedef struct {
+    GLuint offset;			/* Relative to local texture space */
+    GLuint width;
+    GLuint height;
+} r128TexImage;
+
+typedef struct r128_tex_obj r128TexObj, *r128TexObjPtr;
+
+/* Texture object in locally shared texture space.
+ */
+struct r128_tex_obj {
+   driTextureObject   base;
+
+   CARD32 bufAddr;			/* Offset to start of locally
+					   shared texture block */
+
+   GLuint age;
+   r128TexImage image[R128_MAX_TEXTURE_LEVELS]; /* Image data for all
+						   mipmap levels */
+
+   CARD32 textureFormat;		/* Actual hardware format */
+
+   r128_texture_regs_t setup;		/* Setup regs for texture */
+};
+
+#endif /* _R128_TEXOBJ_H_ */
diff --git a/src/mesa/drivers/dri/r128/r128_texstate.c b/src/mesa/drivers/dri/r128/r128_texstate.c
new file mode 100644
index 00000000000..3f4347b02e8
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_texstate.c
@@ -0,0 +1,652 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_tex.c,v 1.7 2001/01/08 01:07:21 martin Exp $ */
+/**************************************************************************
+
+Copyright 1999, 2000 ATI Technologies Inc. and Precision Insight, 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
+ATI, PRECISION INSIGHT 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:
+ *   Gareth Hughes <[email protected]>
+ *   Kevin E. Martin <[email protected]>
+ *   Brian Paul <[email protected]>
+ */
+
+#include "glheader.h"
+#include "imports.h"
+#include "context.h"
+#include "macros.h"
+#include "texformat.h"
+
+#include "r128_context.h"
+#include "r128_state.h"
+#include "r128_ioctl.h"
+#include "r128_vb.h"
+#include "r128_tris.h"
+#include "r128_tex.h"
+
+
+static void r128SetTexImages( r128ContextPtr rmesa,
+                              const struct gl_texture_object *tObj )
+{
+   r128TexObjPtr t = (r128TexObjPtr) tObj->DriverData;
+   struct gl_texture_image *baseImage = tObj->Image[tObj->BaseLevel];
+   int log2Pitch, log2Height, log2Size, log2MinSize;
+   int totalSize;
+   int i;
+   GLint firstLevel, lastLevel;
+
+   assert(t);
+   assert(baseImage);
+
+   if ( R128_DEBUG & DEBUG_VERBOSE_API )
+      fprintf( stderr, "%s( %p )\n", __FUNCTION__, tObj );
+
+   switch (baseImage->TexFormat->MesaFormat) {
+   case MESA_FORMAT_ARGB8888:
+      t->textureFormat = R128_DATATYPE_ARGB8888;
+      break;
+   case MESA_FORMAT_ARGB4444:
+      t->textureFormat = R128_DATATYPE_ARGB4444;
+      break;
+   case MESA_FORMAT_RGB565:
+      t->textureFormat = R128_DATATYPE_RGB565;
+      break;
+   case MESA_FORMAT_RGB332:
+      t->textureFormat = R128_DATATYPE_RGB8;
+      break;
+   case MESA_FORMAT_CI8:
+      t->textureFormat = R128_DATATYPE_CI8;
+      break;
+   case MESA_FORMAT_YCBCR:
+      t->textureFormat = R128_DATATYPE_YVYU422;
+      break;
+   case MESA_FORMAT_YCBCR_REV:
+      t->textureFormat = R128_DATATYPE_VYUY422;
+      break;
+   default:
+      _mesa_problem(rmesa->glCtx, "Bad texture format in %s", __FUNCTION__);
+   };
+
+   /* Compute which mipmap levels we really want to send to the hardware.
+    * This depends on the base image size, GL_TEXTURE_MIN_LOD,
+    * GL_TEXTURE_MAX_LOD, GL_TEXTURE_BASE_LEVEL, and GL_TEXTURE_MAX_LEVEL.
+    * Yes, this looks overly complicated, but it's all needed.
+    */
+   firstLevel = tObj->BaseLevel + (GLint) (tObj->MinLod + 0.5);
+   firstLevel = MAX2(firstLevel, tObj->BaseLevel);
+   lastLevel = tObj->BaseLevel + (GLint) (tObj->MaxLod + 0.5);
+   lastLevel = MAX2(lastLevel, tObj->BaseLevel);
+   lastLevel = MIN2(lastLevel, tObj->BaseLevel + baseImage->MaxLog2);
+   lastLevel = MIN2(lastLevel, tObj->MaxLevel);
+   lastLevel = MAX2(firstLevel, lastLevel); /* need at least one level */
+
+   log2Pitch = tObj->Image[firstLevel]->WidthLog2;
+   log2Height = tObj->Image[firstLevel]->HeightLog2;
+   log2Size = MAX2(log2Pitch, log2Height);
+   log2MinSize = log2Size;
+
+   t->base.dirty_images[0] = 0;
+   totalSize = 0;
+   for ( i = firstLevel; i <= lastLevel; i++ ) {
+      const struct gl_texture_image *texImage;
+
+      texImage = tObj->Image[i];
+      if ( !texImage || !texImage->Data ) {
+         lastLevel = i - 1;
+	 break;
+      }
+
+      log2MinSize = texImage->MaxLog2;
+
+      t->image[i - firstLevel].offset = totalSize;
+      t->image[i - firstLevel].width  = tObj->Image[i]->Width;
+      t->image[i - firstLevel].height = tObj->Image[i]->Height;
+
+      t->base.dirty_images[0] |= (1 << i);
+
+      totalSize += (tObj->Image[i]->Height *
+		    tObj->Image[i]->Width *
+		    tObj->Image[i]->TexFormat->TexelBytes);
+
+      /* Offsets must be 32-byte aligned for host data blits and tiling */
+      totalSize = (totalSize + 31) & ~31;
+   }
+
+   t->base.totalSize = totalSize;
+   t->base.firstLevel = firstLevel;
+   t->base.lastLevel = lastLevel;
+
+   /* Set the texture format */
+   t->setup.tex_cntl &= ~(0xf << 16);
+   t->setup.tex_cntl |= t->textureFormat;
+
+   t->setup.tex_combine_cntl = 0x00000000;  /* XXX is this right? */
+
+   t->setup.tex_size_pitch = ((log2Pitch   << R128_TEX_PITCH_SHIFT) |
+			      (log2Size    << R128_TEX_SIZE_SHIFT) |
+			      (log2Height  << R128_TEX_HEIGHT_SHIFT) |
+			      (log2MinSize << R128_TEX_MIN_SIZE_SHIFT));
+
+   for ( i = 0 ; i < R128_MAX_TEXTURE_LEVELS ; i++ ) {
+      t->setup.tex_offset[i]  = 0x00000000;
+   }
+
+   if (firstLevel == lastLevel)
+      t->setup.tex_cntl |= R128_MIP_MAP_DISABLE;
+   else
+      t->setup.tex_cntl &= ~R128_MIP_MAP_DISABLE;
+
+   /* FYI: r128UploadTexImages( rmesa, t ); used to be called here */
+}
+
+
+/* ================================================================
+ * Texture combine functions
+ */
+
+#define COLOR_COMB_DISABLE		(R128_COMB_DIS |		\
+					 R128_COLOR_FACTOR_TEX)
+#define COLOR_COMB_COPY_INPUT		(R128_COMB_COPY_INP |		\
+					 R128_COLOR_FACTOR_TEX)
+#define COLOR_COMB_MODULATE		(R128_COMB_MODULATE |		\
+					 R128_COLOR_FACTOR_TEX)
+#define COLOR_COMB_MODULATE_NTEX	(R128_COMB_MODULATE |		\
+					 R128_COLOR_FACTOR_NTEX)
+#define COLOR_COMB_ADD			(R128_COMB_ADD |		\
+					 R128_COLOR_FACTOR_TEX)
+#define COLOR_COMB_BLEND_TEX		(R128_COMB_BLEND_TEXTURE |	\
+					 R128_COLOR_FACTOR_TEX)
+/* Rage 128 Pro/M3 only! */
+#define COLOR_COMB_BLEND_COLOR		(R128_COMB_MODULATE2X |		\
+					 R128_COMB_FCN_MSB |		\
+					 R128_COLOR_FACTOR_CONST_COLOR)
+
+#define ALPHA_COMB_DISABLE		(R128_COMB_ALPHA_DIS |		\
+					 R128_ALPHA_FACTOR_TEX_ALPHA)
+#define ALPHA_COMB_COPY_INPUT		(R128_COMB_ALPHA_COPY_INP |	\
+					 R128_ALPHA_FACTOR_TEX_ALPHA)
+#define ALPHA_COMB_MODULATE		(R128_COMB_ALPHA_MODULATE |	\
+					 R128_ALPHA_FACTOR_TEX_ALPHA)
+#define ALPHA_COMB_MODULATE_NTEX	(R128_COMB_ALPHA_MODULATE |	\
+					 R128_ALPHA_FACTOR_NTEX_ALPHA)
+#define ALPHA_COMB_ADD			(R128_COMB_ALPHA_ADD |		\
+					 R128_ALPHA_FACTOR_TEX_ALPHA)
+
+#define INPUT_INTERP			(R128_INPUT_FACTOR_INT_COLOR |	\
+					 R128_INP_FACTOR_A_INT_ALPHA)
+#define INPUT_PREVIOUS			(R128_INPUT_FACTOR_PREV_COLOR |	\
+					 R128_INP_FACTOR_A_PREV_ALPHA)
+
+static GLboolean r128UpdateTextureEnv( GLcontext *ctx, int unit )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   GLint source = rmesa->tmu_source[unit];
+   const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[source];
+   const struct gl_texture_object *tObj = texUnit->_Current;
+   const GLenum format = tObj->Image[tObj->BaseLevel]->Format;
+   GLuint combine;
+
+   if ( R128_DEBUG & DEBUG_VERBOSE_API ) {
+      fprintf( stderr, "%s( %p, %d )\n",
+	       __FUNCTION__, ctx, unit );
+   }
+
+   if ( unit == 0 ) {
+      combine = INPUT_INTERP;
+   } else {
+      combine = INPUT_PREVIOUS;
+   }
+
+   /* Set the texture environment state */
+   switch ( texUnit->EnvMode ) {
+   case GL_REPLACE:
+      switch ( format ) {
+      case GL_RGBA:
+      case GL_LUMINANCE_ALPHA:
+      case GL_INTENSITY:
+	 combine |= (COLOR_COMB_DISABLE |		/* C = Ct            */
+		     ALPHA_COMB_DISABLE);		/* A = At            */
+	 break;
+      case GL_RGB:
+      case GL_LUMINANCE:
+	 combine |= (COLOR_COMB_DISABLE |		/* C = Ct            */
+		     ALPHA_COMB_COPY_INPUT);		/* A = Af            */
+	 break;
+      case GL_ALPHA:
+	 combine |= (COLOR_COMB_COPY_INPUT |		/* C = Cf            */
+		     ALPHA_COMB_DISABLE);		/* A = At            */
+	 break;
+      case GL_COLOR_INDEX:
+      default:
+	 return GL_FALSE;
+      }
+      break;
+
+   case GL_MODULATE:
+      switch ( format ) {
+      case GL_RGBA:
+      case GL_LUMINANCE_ALPHA:
+      case GL_INTENSITY:
+	 combine |= (COLOR_COMB_MODULATE |		/* C = CfCt          */
+		     ALPHA_COMB_MODULATE);		/* A = AfAt          */
+	 break;
+      case GL_RGB:
+      case GL_LUMINANCE:
+	 combine |= (COLOR_COMB_MODULATE |		/* C = CfCt          */
+		     ALPHA_COMB_COPY_INPUT);		/* A = Af            */
+	 break;
+      case GL_ALPHA:
+	 combine |= (COLOR_COMB_COPY_INPUT |		/* C = Cf            */
+		     ALPHA_COMB_MODULATE);		/* A = AfAt          */
+	 break;
+      case GL_COLOR_INDEX:
+      default:
+	 return GL_FALSE;
+      }
+      break;
+
+   case GL_DECAL:
+      switch ( format ) {
+      case GL_RGBA:
+	 combine |= (COLOR_COMB_BLEND_TEX |		/* C = Cf(1-At)+CtAt */
+		     ALPHA_COMB_COPY_INPUT);		/* A = Af            */
+	 break;
+      case GL_RGB:
+	 combine |= (COLOR_COMB_DISABLE |		/* C = Ct            */
+		     ALPHA_COMB_COPY_INPUT);		/* A = Af            */
+	 break;
+      case GL_ALPHA:
+      case GL_LUMINANCE:
+      case GL_LUMINANCE_ALPHA:
+      case GL_INTENSITY:
+	 /* Undefined behaviour - just copy the incoming fragment */
+	 combine |= (COLOR_COMB_COPY_INPUT |		/* C = undefined     */
+		     ALPHA_COMB_COPY_INPUT);		/* A = undefined     */
+	 break;
+      case GL_COLOR_INDEX:
+      default:
+	 return GL_FALSE;
+      }
+      break;
+
+   case GL_BLEND:
+      /* Rage 128 Pro and M3 can handle GL_BLEND texturing.
+       */
+      if ( !R128_IS_PLAIN( rmesa ) ) {
+         /* XXX this hasn't been fully tested, I don't have a Pro card. -BP */
+	 switch ( format ) {
+	 case GL_RGBA:
+	 case GL_LUMINANCE_ALPHA:
+	    combine |= (COLOR_COMB_BLEND_COLOR |	/* C = Cf(1-Ct)+CcCt */
+			ALPHA_COMB_MODULATE);		/* A = AfAt          */
+	    break;
+
+	 case GL_RGB:
+	 case GL_LUMINANCE:
+	    combine |= (COLOR_COMB_BLEND_COLOR |	/* C = Cf(1-Ct)+CcCt */
+			ALPHA_COMB_COPY_INPUT);		/* A = Af            */
+	    break;
+
+	 case GL_ALPHA:
+	    combine |= (COLOR_COMB_COPY_INPUT |		/* C = Cf            */
+			ALPHA_COMB_MODULATE);		/* A = AfAt          */
+	    break;
+
+	 case GL_INTENSITY:
+	    /* GH: We could be smarter about this... */
+	    switch ( rmesa->env_color & 0xff000000 ) {
+	    case 0x00000000:
+	       combine |= (COLOR_COMB_BLEND_COLOR |	/* C = Cf(1-It)+CcIt */
+			   ALPHA_COMB_MODULATE_NTEX);	/* A = Af(1-It)      */
+	    default:
+	       combine |= (COLOR_COMB_MODULATE |	/* C = fallback      */
+			   ALPHA_COMB_MODULATE);	/* A = fallback      */
+	       return GL_FALSE;
+	    }
+	    break;
+
+	 case GL_COLOR_INDEX:
+	 default:
+	    return GL_FALSE;
+	 }
+	 break;
+      }
+
+      /* Rage 128 has to fake some cases of GL_BLEND, otherwise fallback
+       * to software rendering.
+       */
+      if ( rmesa->blend_flags ) {
+	 return GL_FALSE;
+      }
+      switch ( format ) {
+      case GL_RGBA:
+      case GL_LUMINANCE_ALPHA:
+	 switch ( rmesa->env_color & 0x00ffffff ) {
+	 case 0x00000000:
+	    combine |= (COLOR_COMB_MODULATE_NTEX |	/* C = Cf(1-Ct)      */
+			ALPHA_COMB_MODULATE);		/* A = AfAt          */
+	    break;
+#if 0
+         /* This isn't right - BP */
+	 case 0x00ffffff:
+	    if ( unit == 0 ) {
+	       combine |= (COLOR_COMB_MODULATE_NTEX |	/* C = Cf(1-Ct)      */
+			   ALPHA_COMB_MODULATE);	/* A = AfAt          */
+	    } else {
+	       combine |= (COLOR_COMB_ADD |		/* C = Cf+Ct         */
+			   ALPHA_COMB_COPY_INPUT);	/* A = Af            */
+	    }
+	    break;
+#endif
+	 default:
+	    combine |= (COLOR_COMB_MODULATE |		/* C = fallback      */
+			ALPHA_COMB_MODULATE);		/* A = fallback      */
+	    return GL_FALSE;
+	 }
+	 break;
+      case GL_RGB:
+      case GL_LUMINANCE:
+	 switch ( rmesa->env_color & 0x00ffffff ) {
+	 case 0x00000000:
+	    combine |= (COLOR_COMB_MODULATE_NTEX |	/* C = Cf(1-Ct)      */
+			ALPHA_COMB_COPY_INPUT);		/* A = Af            */
+	    break;
+#if 0
+         /* This isn't right - BP */
+	 case 0x00ffffff:
+	    if ( unit == 0 ) {
+	       combine |= (COLOR_COMB_MODULATE_NTEX |	/* C = Cf(1-Ct)      */
+			   ALPHA_COMB_COPY_INPUT);	/* A = Af            */
+	    } else {
+	       combine |= (COLOR_COMB_ADD |		/* C = Cf+Ct         */
+			   ALPHA_COMB_COPY_INPUT);	/* A = Af            */
+	    }
+	    break;
+#endif
+	 default:
+	    combine |= (COLOR_COMB_MODULATE |		/* C = fallback      */
+			ALPHA_COMB_COPY_INPUT);		/* A = fallback      */
+	    return GL_FALSE;
+	 }
+	 break;
+      case GL_ALPHA:
+	 if ( unit == 0 ) {
+	    combine |= (COLOR_COMB_COPY_INPUT |		/* C = Cf            */
+			ALPHA_COMB_MODULATE);		/* A = AfAt          */
+	 } else {
+	    combine |= (COLOR_COMB_COPY_INPUT |		/* C = Cf            */
+			ALPHA_COMB_COPY_INPUT);		/* A = Af            */
+	 }
+	 break;
+      case GL_INTENSITY:
+	 switch ( rmesa->env_color & 0x00ffffff ) {
+	 case 0x00000000:
+	    combine |= COLOR_COMB_MODULATE_NTEX;	/* C = Cf(1-It)      */
+	    break;
+#if 0
+         /* This isn't right - BP */
+	 case 0x00ffffff:
+	    if ( unit == 0 ) {
+	       combine |= COLOR_COMB_MODULATE_NTEX;	/* C = Cf(1-It)      */
+	    } else {
+	       combine |= COLOR_COMB_ADD;		/* C = Cf+It         */
+	    }
+	    break;
+#endif
+	 default:
+	    combine |= (COLOR_COMB_MODULATE |		/* C = fallback      */
+			ALPHA_COMB_MODULATE);		/* A = fallback      */
+	    return GL_FALSE;
+	 }
+	 switch ( rmesa->env_color & 0xff000000 ) {
+	 case 0x00000000:
+	    combine |= ALPHA_COMB_MODULATE_NTEX;	/* A = Af(1-It)      */
+	    break;
+#if 0
+         /* This isn't right - BP */
+	 case 0xff000000:
+	    if ( unit == 0 ) {
+	       combine |= ALPHA_COMB_MODULATE_NTEX;	/* A = Af(1-It)      */
+	    } else {
+	       combine |= ALPHA_COMB_ADD;		/* A = Af+It         */
+	    }
+	    break;
+#endif
+	 default:
+	    combine |= (COLOR_COMB_MODULATE |		/* C = fallback      */
+			ALPHA_COMB_MODULATE);		/* A = fallback      */
+	    return GL_FALSE;
+	 }
+	 break;
+      case GL_COLOR_INDEX:
+      default:
+	 return GL_FALSE;
+      }
+      break;
+
+   case GL_ADD:
+      switch ( format ) {
+      case GL_RGBA:
+      case GL_LUMINANCE_ALPHA:
+	 combine |= (COLOR_COMB_ADD |			/* C = Cf+Ct         */
+		     ALPHA_COMB_MODULATE);		/* A = AfAt          */
+	 break;
+      case GL_RGB:
+      case GL_LUMINANCE:
+	 combine |= (COLOR_COMB_ADD |			/* C = Cf+Ct         */
+		     ALPHA_COMB_COPY_INPUT);		/* A = Af            */
+	 break;
+      case GL_ALPHA:
+	 combine |= (COLOR_COMB_COPY_INPUT |		/* C = Cf            */
+		     ALPHA_COMB_MODULATE);		/* A = AfAt          */
+	 break;
+      case GL_INTENSITY:
+	 combine |= (COLOR_COMB_ADD |			/* C = Cf+Ct         */
+		     ALPHA_COMB_ADD);			/* A = Af+At         */
+	 break;
+      case GL_COLOR_INDEX:
+      default:
+	 return GL_FALSE;
+      }
+      break;
+
+   default:
+      return GL_FALSE;
+   }
+
+   if ( rmesa->tex_combine[unit] != combine ) {
+     rmesa->tex_combine[unit] = combine;
+     rmesa->dirty |= R128_UPLOAD_TEX0 << unit;
+   }
+   return GL_TRUE;
+}
+
+static void disable_tex( GLcontext *ctx, int unit )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   FLUSH_BATCH( rmesa );
+
+   if ( rmesa->CurrentTexObj[unit] ) {
+      rmesa->CurrentTexObj[unit]->base.bound &= ~(1 << unit);
+      rmesa->CurrentTexObj[unit] = NULL;
+   }
+
+   rmesa->setup.tex_cntl_c &= ~(R128_TEXMAP_ENABLE << unit);
+   rmesa->setup.tex_size_pitch_c &= ~(R128_TEX_SIZE_PITCH_MASK << 
+				      (R128_SEC_TEX_SIZE_PITCH_SHIFT * unit));
+   rmesa->dirty |= R128_UPLOAD_CONTEXT;
+
+   /* If either texture unit is disabled, then multitexturing is not
+    * happening.
+    */
+
+   rmesa->blend_flags &= ~R128_BLEND_MULTITEX;
+}
+
+static GLboolean enable_tex_2d( GLcontext *ctx, int unit )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   const int source = rmesa->tmu_source[unit];
+   const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[source];
+   const struct gl_texture_object *tObj = texUnit->_Current;
+   r128TexObjPtr t = (r128TexObjPtr) tObj->DriverData;
+
+   /* Need to load the 2d images associated with this unit.
+    */
+   if ( t->base.dirty_images[0] ) {
+      /* FIXME: For Radeon, RADEON_FIREVERTICES is called here.  Should
+       * FIXME: something similar be done for R128?
+       */
+      /*  FLUSH_BATCH( rmesa ); */
+
+      r128SetTexImages( rmesa, tObj );
+      r128UploadTexImages( rmesa, t );
+      if ( !t->base.memBlock ) 
+	  return GL_FALSE;
+   }
+
+   return GL_TRUE;
+}
+
+static GLboolean update_tex_common( GLcontext *ctx, int unit )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   const int source = rmesa->tmu_source[unit];
+   const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[source];
+   const struct gl_texture_object *tObj = texUnit->_Current;
+   r128TexObjPtr t = (r128TexObjPtr) tObj->DriverData;
+
+
+   /* Fallback if there's a texture border */
+   if ( tObj->Image[tObj->BaseLevel]->Border > 0 ) {
+      return GL_FALSE;
+   }
+
+
+   /* Update state if this is a different texture object to last
+    * time.
+    */
+   if ( rmesa->CurrentTexObj[unit] != t ) {
+      if ( rmesa->CurrentTexObj[unit] != NULL ) {
+	 /* The old texture is no longer bound to this texture unit.
+	  * Mark it as such.
+	  */
+
+	 rmesa->CurrentTexObj[unit]->base.bound &= 
+	     ~(1UL << unit);
+      }
+
+      rmesa->CurrentTexObj[unit] = t;
+      t->base.bound |= (1UL << unit);
+      rmesa->dirty |= R128_UPLOAD_TEX0 << unit;
+
+      driUpdateTextureLRU( (driTextureObject *) t ); /* XXX: should be locked! */
+   }
+
+   /* FIXME: We need to update the texture unit if any texture parameters have
+    * changed, but this texture was already bound.  This could be changed to
+    * work like the Radeon driver where the texture object has it's own
+    * dirty state flags
+    */
+   rmesa->dirty |= R128_UPLOAD_TEX0 << unit;
+
+   /* register setup */
+   rmesa->setup.tex_size_pitch_c &= ~(R128_TEX_SIZE_PITCH_MASK << 
+				      (R128_SEC_TEX_SIZE_PITCH_SHIFT * unit));
+
+   if ( unit == 0 ) {
+      rmesa->setup.tex_cntl_c       |= R128_TEXMAP_ENABLE;
+      rmesa->setup.tex_size_pitch_c |= t->setup.tex_size_pitch << 0;
+      rmesa->setup.scale_3d_cntl    &= ~R128_TEX_CACHE_SPLIT;
+      t->setup.tex_cntl             &= ~R128_SEC_SELECT_SEC_ST;
+   }
+   else {
+      rmesa->setup.tex_cntl_c       |= R128_SEC_TEXMAP_ENABLE;
+      rmesa->setup.tex_size_pitch_c |= t->setup.tex_size_pitch << 16;
+      rmesa->setup.scale_3d_cntl    |= R128_TEX_CACHE_SPLIT;
+      t->setup.tex_cntl             |=  R128_SEC_SELECT_SEC_ST;
+
+      /* If the second TMU is enabled, then multitexturing is happening.
+       */
+      if ( R128_IS_PLAIN( rmesa ) )
+	  rmesa->blend_flags            |= R128_BLEND_MULTITEX;
+   }
+
+   rmesa->dirty |= R128_UPLOAD_CONTEXT;
+
+
+   /* FIXME: The Radeon has some cached state so that it can avoid calling
+    * FIXME: UpdateTextureEnv in some cases.  Is that possible here?
+    */
+   return r128UpdateTextureEnv( ctx, unit );
+}
+
+static GLboolean updateTextureUnit( GLcontext *ctx, int unit )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   const int source = rmesa->tmu_source[unit];
+   const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[source];
+
+
+   if (texUnit->_ReallyEnabled & (TEXTURE_1D_BIT | TEXTURE_2D_BIT)) {
+      return (enable_tex_2d( ctx, unit ) &&
+	      update_tex_common( ctx, unit ));
+   }
+   else if ( texUnit->_ReallyEnabled ) {
+      return GL_FALSE;
+   }
+   else {
+      disable_tex( ctx, unit );
+      return GL_TRUE;
+   }
+}
+
+
+void r128UpdateTextureState( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   GLboolean ok;
+
+
+   /* This works around a quirk with the R128 hardware.  If only OpenGL 
+    * TEXTURE1 is enabled, then the hardware TEXTURE0 must be used.  The
+    * hardware TEXTURE1 can ONLY be used when hardware TEXTURE0 is also used.
+    */
+
+   rmesa->tmu_source[0] = 0;
+   rmesa->tmu_source[1] = 1;
+
+   if ((ctx->Texture._EnabledUnits & 0x03) == 0x02) {
+      /* only texture 1 enabled */
+      rmesa->tmu_source[0] = 1;
+      rmesa->tmu_source[1] = 0;
+   }
+
+   ok = (updateTextureUnit( ctx, 0 ) &&
+	 updateTextureUnit( ctx, 1 ));
+
+   FALLBACK( rmesa, R128_FALLBACK_TEXTURE, !ok );
+}
diff --git a/src/mesa/drivers/dri/r128/r128_tris.c b/src/mesa/drivers/dri/r128/r128_tris.c
new file mode 100644
index 00000000000..4e2860288e4
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_tris.c
@@ -0,0 +1,721 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_tris.c,v 1.8 2002/10/30 12:51:43 alanh Exp $ */ /* -*- c-basic-offset: 3 -*- */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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
+ATI, VA LINUX SYSTEMS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell <[email protected]>
+ *
+ */
+
+#include "glheader.h"
+#include "mtypes.h"
+#include "colormac.h"
+#include "macros.h"
+
+#include "swrast/swrast.h"
+#include "swrast_setup/swrast_setup.h"
+#include "tnl/tnl.h"
+#include "tnl/t_context.h"
+#include "tnl/t_pipeline.h"
+
+#include "r128_tris.h"
+#include "r128_state.h"
+#include "r128_tex.h"
+#include "r128_vb.h"
+#include "r128_ioctl.h"
+
+static const GLuint hw_prim[GL_POLYGON+1] = {
+   R128_CCE_VC_CNTL_PRIM_TYPE_POINT,
+   R128_CCE_VC_CNTL_PRIM_TYPE_LINE,
+   R128_CCE_VC_CNTL_PRIM_TYPE_LINE,
+   R128_CCE_VC_CNTL_PRIM_TYPE_LINE,
+   R128_CCE_VC_CNTL_PRIM_TYPE_TRI_LIST,
+   R128_CCE_VC_CNTL_PRIM_TYPE_TRI_LIST,
+   R128_CCE_VC_CNTL_PRIM_TYPE_TRI_LIST,
+   R128_CCE_VC_CNTL_PRIM_TYPE_TRI_LIST,
+   R128_CCE_VC_CNTL_PRIM_TYPE_TRI_LIST,
+   R128_CCE_VC_CNTL_PRIM_TYPE_TRI_LIST,
+};
+
+static void r128RasterPrimitive( GLcontext *ctx, GLuint hwprim );
+static void r128RenderPrimitive( GLcontext *ctx, GLenum prim );
+
+
+/***********************************************************************
+ *                    Emit primitives as inline vertices               *
+ ***********************************************************************/
+
+#if defined(USE_X86_ASM)
+#define COPY_DWORDS( j, vb, vertsize, v )				\
+do {									\
+	int __tmp;							\
+	__asm__ __volatile__( "rep ; movsl"				\
+			      : "=%c" (j), "=D" (vb), "=S" (__tmp)	\
+			      : "0" (vertsize),				\
+			        "D" ((long)vb),				\
+			        "S" ((long)v) );			\
+} while (0)
+#else
+#define COPY_DWORDS( j, vb, vertsize, v )				\
+do {									\
+   for ( j = 0 ; j < vertsize ; j++ )					\
+      vb[j] = CPU_TO_LE32(((GLuint *)v)[j]);				\
+   vb += vertsize;							\
+} while (0)
+#endif
+
+static __inline void r128_draw_quad( r128ContextPtr rmesa,
+				     r128VertexPtr v0,
+				     r128VertexPtr v1,
+				     r128VertexPtr v2,
+				     r128VertexPtr v3 )
+{
+   GLuint vertsize = rmesa->vertex_size;
+   GLuint *vb = (GLuint *)r128AllocDmaLow( rmesa, 6 * vertsize * 4 );
+   GLuint j;
+
+   rmesa->num_verts += 6;
+   COPY_DWORDS( j, vb, vertsize, v0 );
+   COPY_DWORDS( j, vb, vertsize, v1 );
+   COPY_DWORDS( j, vb, vertsize, v3 );
+   COPY_DWORDS( j, vb, vertsize, v1 );
+   COPY_DWORDS( j, vb, vertsize, v2 );
+   COPY_DWORDS( j, vb, vertsize, v3 );
+}
+
+
+static __inline void r128_draw_triangle( r128ContextPtr rmesa,
+					 r128VertexPtr v0,
+					 r128VertexPtr v1,
+					 r128VertexPtr v2 )
+{
+   GLuint vertsize = rmesa->vertex_size;
+   GLuint *vb = (GLuint *)r128AllocDmaLow( rmesa, 3 * vertsize * 4 );
+   GLuint j;
+
+   rmesa->num_verts += 3;
+   COPY_DWORDS( j, vb, vertsize, v0 );
+   COPY_DWORDS( j, vb, vertsize, v1 );
+   COPY_DWORDS( j, vb, vertsize, v2 );
+}
+
+static __inline void r128_draw_line( r128ContextPtr rmesa,
+				     r128VertexPtr v0,
+				     r128VertexPtr v1 )
+{
+   GLuint vertsize = rmesa->vertex_size;
+   GLuint *vb = (GLuint *)r128AllocDmaLow( rmesa, 2 * vertsize * 4 );
+   GLuint j;
+
+   rmesa->num_verts += 2;
+   COPY_DWORDS( j, vb, vertsize, v0 );
+   COPY_DWORDS( j, vb, vertsize, v1 );
+}
+
+static __inline void r128_draw_point( r128ContextPtr rmesa,
+				      r128VertexPtr v0 )
+{
+   int vertsize = rmesa->vertex_size;
+   GLuint *vb = (GLuint *)r128AllocDmaLow( rmesa, vertsize * 4 );
+   int j;
+
+   rmesa->num_verts += 1;
+   COPY_DWORDS( j, vb, vertsize, v0 );
+}
+
+/***********************************************************************
+ *          Macros for t_dd_tritmp.h to draw basic primitives          *
+ ***********************************************************************/
+
+#define TRI( a, b, c )				\
+do {						\
+   if (DO_FALLBACK)				\
+      rmesa->draw_tri( rmesa, a, b, c );	\
+   else						\
+      r128_draw_triangle( rmesa, a, b, c );	\
+} while (0)
+
+#define QUAD( a, b, c, d )			\
+do {						\
+   if (DO_FALLBACK) {				\
+      rmesa->draw_tri( rmesa, a, b, d );	\
+      rmesa->draw_tri( rmesa, b, c, d );	\
+   } else 					\
+      r128_draw_quad( rmesa, a, b, c, d );	\
+} while (0)
+
+#define LINE( v0, v1 )				\
+do {						\
+   if (DO_FALLBACK)				\
+      rmesa->draw_line( rmesa, v0, v1 );	\
+   else 					\
+      r128_draw_line( rmesa, v0, v1 );	\
+} while (0)
+
+#define POINT( v0 )				\
+do {						\
+   if (DO_FALLBACK)				\
+      rmesa->draw_point( rmesa, v0 );		\
+   else 					\
+      r128_draw_point( rmesa, v0 );		\
+} while (0)
+
+
+/***********************************************************************
+ *              Build render functions from dd templates               *
+ ***********************************************************************/
+
+#define R128_OFFSET_BIT	0x01
+#define R128_TWOSIDE_BIT	0x02
+#define R128_UNFILLED_BIT	0x04
+#define R128_FALLBACK_BIT	0x08
+#define R128_MAX_TRIFUNC	0x10
+
+
+static struct {
+   points_func	        points;
+   line_func		line;
+   triangle_func	triangle;
+   quad_func		quad;
+} rast_tab[R128_MAX_TRIFUNC];
+
+
+#define DO_FALLBACK (IND & R128_FALLBACK_BIT)
+#define DO_OFFSET   (IND & R128_OFFSET_BIT)
+#define DO_UNFILLED (IND & R128_UNFILLED_BIT)
+#define DO_TWOSIDE  (IND & R128_TWOSIDE_BIT)
+#define DO_FLAT      0
+#define DO_TRI       1
+#define DO_QUAD      1
+#define DO_LINE      1
+#define DO_POINTS    1
+#define DO_FULL_QUAD 1
+
+#define HAVE_RGBA   1
+#define HAVE_SPEC   1
+#define HAVE_BACK_COLORS  0
+#define HAVE_HW_FLATSHADE 1
+#define VERTEX r128Vertex
+#define TAB rast_tab
+
+#define DEPTH_SCALE 1.0
+#define UNFILLED_TRI unfilled_tri
+#define UNFILLED_QUAD unfilled_quad
+#define VERT_X(_v) _v->v.x
+#define VERT_Y(_v) _v->v.y
+#define VERT_Z(_v) _v->v.z
+#define AREA_IS_CCW( a ) (a > 0)
+#define GET_VERTEX(e) (rmesa->verts + (e<<rmesa->vertex_stride_shift))
+
+#define VERT_SET_RGBA( v, c )	do {				\
+					r128_color_t *vc = (r128_color_t *)&(v)->ui[coloroffset];	\
+					vc->blue  = (c)[2];	\
+					vc->green = (c)[1];	\
+					vc->red   = (c)[0];	\
+					vc->alpha = (c)[3];	\
+				} while (0)
+#define VERT_COPY_RGBA( v0, v1 ) v0->ui[coloroffset] = v1->ui[coloroffset]
+#define VERT_SAVE_RGBA( idx )    color[idx] = v[idx]->ui[coloroffset]
+#define VERT_RESTORE_RGBA( idx ) v[idx]->ui[coloroffset] = color[idx]
+
+#define VERT_SET_SPEC( v0, c )   if (havespec) {			\
+					(v0)->v.specular.red   = (c)[0];\
+					(v0)->v.specular.green = (c)[1];\
+					(v0)->v.specular.blue  = (c)[2]; }
+#define VERT_COPY_SPEC( v0, v1 ) if (havespec) {					\
+					(v0)->v.specular.red   = v1->v.specular.red;	\
+					(v0)->v.specular.green = v1->v.specular.green;	\
+					(v0)->v.specular.blue  = v1->v.specular.blue; }
+
+#define VERT_SAVE_SPEC( idx )    if (havespec) spec[idx] = v[idx]->ui[5]
+#define VERT_RESTORE_SPEC( idx ) if (havespec) v[idx]->ui[5] = spec[idx]
+
+#define LOCAL_VARS(n)						\
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);			\
+   GLuint color[n], spec[n];					\
+   GLuint coloroffset = (rmesa->vertex_size == 4 ? 3 : 4);	\
+   GLboolean havespec = (rmesa->vertex_size == 4 ? 0 : 1);	\
+   (void) color; (void) spec; (void) coloroffset; (void) havespec;
+
+/***********************************************************************
+ *                Helpers for rendering unfilled primitives            *
+ ***********************************************************************/
+
+#define RASTERIZE(x) if (rmesa->hw_primitive != hw_prim[x]) \
+                        r128RasterPrimitive( ctx, hw_prim[x] )
+#define RENDER_PRIMITIVE rmesa->render_primitive
+#define IND R128_FALLBACK_BIT
+#define TAG(x) x
+#include "tnl_dd/t_dd_unfilled.h"
+#undef IND
+
+
+/***********************************************************************
+ *                      Generate GL render functions                   *
+ ***********************************************************************/
+
+
+#define IND (0)
+#define TAG(x) x
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (R128_OFFSET_BIT)
+#define TAG(x) x##_offset
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (R128_TWOSIDE_BIT)
+#define TAG(x) x##_twoside
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (R128_TWOSIDE_BIT|R128_OFFSET_BIT)
+#define TAG(x) x##_twoside_offset
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (R128_UNFILLED_BIT)
+#define TAG(x) x##_unfilled
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (R128_OFFSET_BIT|R128_UNFILLED_BIT)
+#define TAG(x) x##_offset_unfilled
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (R128_TWOSIDE_BIT|R128_UNFILLED_BIT)
+#define TAG(x) x##_twoside_unfilled
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (R128_TWOSIDE_BIT|R128_OFFSET_BIT|R128_UNFILLED_BIT)
+#define TAG(x) x##_twoside_offset_unfilled
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (R128_FALLBACK_BIT)
+#define TAG(x) x##_fallback
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (R128_OFFSET_BIT|R128_FALLBACK_BIT)
+#define TAG(x) x##_offset_fallback
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (R128_TWOSIDE_BIT|R128_FALLBACK_BIT)
+#define TAG(x) x##_twoside_fallback
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (R128_TWOSIDE_BIT|R128_OFFSET_BIT|R128_FALLBACK_BIT)
+#define TAG(x) x##_twoside_offset_fallback
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (R128_UNFILLED_BIT|R128_FALLBACK_BIT)
+#define TAG(x) x##_unfilled_fallback
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (R128_OFFSET_BIT|R128_UNFILLED_BIT|R128_FALLBACK_BIT)
+#define TAG(x) x##_offset_unfilled_fallback
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (R128_TWOSIDE_BIT|R128_UNFILLED_BIT|R128_FALLBACK_BIT)
+#define TAG(x) x##_twoside_unfilled_fallback
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (R128_TWOSIDE_BIT|R128_OFFSET_BIT|R128_UNFILLED_BIT| \
+	     R128_FALLBACK_BIT)
+#define TAG(x) x##_twoside_offset_unfilled_fallback
+#include "tnl_dd/t_dd_tritmp.h"
+
+
+static void init_rast_tab( void )
+{
+   init();
+   init_offset();
+   init_twoside();
+   init_twoside_offset();
+   init_unfilled();
+   init_offset_unfilled();
+   init_twoside_unfilled();
+   init_twoside_offset_unfilled();
+   init_fallback();
+   init_offset_fallback();
+   init_twoside_fallback();
+   init_twoside_offset_fallback();
+   init_unfilled_fallback();
+   init_offset_unfilled_fallback();
+   init_twoside_unfilled_fallback();
+   init_twoside_offset_unfilled_fallback();
+}
+
+
+
+/***********************************************************************
+ *                    Rasterization fallback helpers                   *
+ ***********************************************************************/
+
+
+/* This code is hit only when a mix of accelerated and unaccelerated
+ * primitives are being drawn, and only for the unaccelerated
+ * primitives.
+ */
+static void
+r128_fallback_tri( r128ContextPtr rmesa,
+		     r128Vertex *v0,
+		     r128Vertex *v1,
+		     r128Vertex *v2 )
+{
+   GLcontext *ctx = rmesa->glCtx;
+   SWvertex v[3];
+   r128_translate_vertex( ctx, v0, &v[0] );
+   r128_translate_vertex( ctx, v1, &v[1] );
+   r128_translate_vertex( ctx, v2, &v[2] );
+   _swrast_Triangle( ctx, &v[0], &v[1], &v[2] );
+}
+
+
+static void
+r128_fallback_line( r128ContextPtr rmesa,
+		    r128Vertex *v0,
+		    r128Vertex *v1 )
+{
+   GLcontext *ctx = rmesa->glCtx;
+   SWvertex v[2];
+   r128_translate_vertex( ctx, v0, &v[0] );
+   r128_translate_vertex( ctx, v1, &v[1] );
+   _swrast_Line( ctx, &v[0], &v[1] );
+}
+
+
+static void
+r128_fallback_point( r128ContextPtr rmesa,
+		     r128Vertex *v0 )
+{
+   GLcontext *ctx = rmesa->glCtx;
+   SWvertex v[1];
+   r128_translate_vertex( ctx, v0, &v[0] );
+   _swrast_Point( ctx, &v[0] );
+}
+
+
+
+/**********************************************************************/
+/*               Render unclipped begin/end objects                   */
+/**********************************************************************/
+
+#define VERT(x) (r128Vertex *)(r128verts + (x << shift))
+#define RENDER_POINTS( start, count )		\
+   for ( ; start < count ; start++)		\
+      r128_draw_point( rmesa, VERT(start) )
+#define RENDER_LINE( v0, v1 ) \
+   r128_draw_line( rmesa, VERT(v0), VERT(v1) )
+#define RENDER_TRI( v0, v1, v2 )  \
+   r128_draw_triangle( rmesa, VERT(v0), VERT(v1), VERT(v2) )
+#define RENDER_QUAD( v0, v1, v2, v3 ) \
+   r128_draw_quad( rmesa, VERT(v0), VERT(v1), VERT(v2), VERT(v3) )
+#define INIT(x) do {					\
+   if (0) fprintf(stderr, "%s\n", __FUNCTION__);	\
+   r128RenderPrimitive( ctx, x );			\
+} while (0)
+#undef LOCAL_VARS
+#define LOCAL_VARS						\
+    r128ContextPtr rmesa = R128_CONTEXT(ctx);		\
+    const GLuint shift = rmesa->vertex_stride_shift;		\
+    const char *r128verts = (char *)rmesa->verts;		\
+    const GLuint * const elt = TNL_CONTEXT(ctx)->vb.Elts;	\
+    (void) elt;
+#define RESET_STIPPLE
+#define RESET_OCCLUSION
+#define PRESERVE_VB_DEFS
+#define ELT(x) (x)
+#define TAG(x) r128_##x##_verts
+#include "tnl/t_vb_rendertmp.h"
+#undef ELT
+#undef TAG
+#define TAG(x) r128_##x##_elts
+#define ELT(x) elt[x]
+#include "tnl/t_vb_rendertmp.h"
+
+
+/**********************************************************************/
+/*                    Render clipped primitives                       */
+/**********************************************************************/
+
+static void r128RenderClippedPoly( GLcontext *ctx, const GLuint *elts,
+				     GLuint n )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+
+   /* Render the new vertices as an unclipped polygon.
+    */
+   {
+      GLuint *tmp = VB->Elts;
+      VB->Elts = (GLuint *)elts;
+      tnl->Driver.Render.PrimTabElts[GL_POLYGON]( ctx, 0, n, PRIM_BEGIN|PRIM_END );
+      VB->Elts = tmp;
+   }
+}
+
+static void r128RenderClippedLine( GLcontext *ctx, GLuint ii, GLuint jj )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   tnl->Driver.Render.Line( ctx, ii, jj );
+}
+
+static void r128FastRenderClippedPoly( GLcontext *ctx, const GLuint *elts,
+					 GLuint n )
+{
+   r128ContextPtr rmesa = R128_CONTEXT( ctx );
+   GLuint vertsize = rmesa->vertex_size;
+   GLuint *vb = r128AllocDmaLow( rmesa, (n-2) * 3 * 4 * vertsize );
+   GLubyte *r128verts = (GLubyte *)rmesa->verts;
+   const GLuint shift = rmesa->vertex_stride_shift;
+   const GLuint *start = (const GLuint *)VERT(elts[0]);
+   int i,j;
+
+   rmesa->num_verts += (n-2) * 3;
+
+   for (i = 2 ; i < n ; i++) {
+      COPY_DWORDS( j, vb, vertsize, (r128VertexPtr) VERT(elts[i-1]) );
+      COPY_DWORDS( j, vb, vertsize, (r128VertexPtr) VERT(elts[i]) );
+      COPY_DWORDS( j, vb, vertsize, (r128VertexPtr) start );
+   }
+}
+
+
+
+
+/**********************************************************************/
+/*                    Choose render functions                         */
+/**********************************************************************/
+
+#define _R128_NEW_RENDER_STATE (_DD_NEW_LINE_STIPPLE |	\
+			          _DD_NEW_LINE_SMOOTH |		\
+			          _DD_NEW_POINT_SMOOTH |	\
+			          _DD_NEW_TRI_SMOOTH |		\
+			          _DD_NEW_TRI_UNFILLED |	\
+			          _DD_NEW_TRI_LIGHT_TWOSIDE |	\
+			          _DD_NEW_TRI_OFFSET)		\
+
+
+#define POINT_FALLBACK (DD_POINT_SMOOTH)
+#define LINE_FALLBACK (DD_LINE_STIPPLE|DD_LINE_SMOOTH)
+#define TRI_FALLBACK (DD_TRI_SMOOTH)
+#define ANY_FALLBACK_FLAGS (POINT_FALLBACK|LINE_FALLBACK|TRI_FALLBACK)
+#define ANY_RASTER_FLAGS (DD_TRI_LIGHT_TWOSIDE|DD_TRI_OFFSET|DD_TRI_UNFILLED)
+
+
+static void r128ChooseRenderState(GLcontext *ctx)
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   GLuint flags = ctx->_TriangleCaps;
+   GLuint index = 0;
+
+   if (flags & (ANY_RASTER_FLAGS|ANY_FALLBACK_FLAGS)) {
+      rmesa->draw_point = r128_draw_point;
+      rmesa->draw_line = r128_draw_line;
+      rmesa->draw_tri = r128_draw_triangle;
+
+      if (flags & ANY_RASTER_FLAGS) {
+	 if (flags & DD_TRI_LIGHT_TWOSIDE) index |= R128_TWOSIDE_BIT;
+	 if (flags & DD_TRI_OFFSET)        index |= R128_OFFSET_BIT;
+	 if (flags & DD_TRI_UNFILLED)      index |= R128_UNFILLED_BIT;
+      }
+
+      /* Hook in fallbacks for specific primitives.
+       */
+      if (flags & (POINT_FALLBACK|LINE_FALLBACK|TRI_FALLBACK)) {
+	 if (flags & POINT_FALLBACK) rmesa->draw_point = r128_fallback_point;
+	 if (flags & LINE_FALLBACK)  rmesa->draw_line = r128_fallback_line;
+	 if (flags & TRI_FALLBACK)   rmesa->draw_tri = r128_fallback_tri;
+	 index |= R128_FALLBACK_BIT;
+      }
+   }
+
+   if (index != rmesa->RenderIndex) {
+      TNLcontext *tnl = TNL_CONTEXT(ctx);
+      tnl->Driver.Render.Points = rast_tab[index].points;
+      tnl->Driver.Render.Line = rast_tab[index].line;
+      tnl->Driver.Render.Triangle = rast_tab[index].triangle;
+      tnl->Driver.Render.Quad = rast_tab[index].quad;
+
+      if (index == 0) {
+	 tnl->Driver.Render.PrimTabVerts = r128_render_tab_verts;
+	 tnl->Driver.Render.PrimTabElts = r128_render_tab_elts;
+	 tnl->Driver.Render.ClippedLine = rast_tab[index].line;
+	 tnl->Driver.Render.ClippedPolygon = r128FastRenderClippedPoly;
+      } else {
+	 tnl->Driver.Render.PrimTabVerts = _tnl_render_tab_verts;
+	 tnl->Driver.Render.PrimTabElts = _tnl_render_tab_elts;
+	 tnl->Driver.Render.ClippedLine = r128RenderClippedLine;
+	 tnl->Driver.Render.ClippedPolygon = r128RenderClippedPoly;
+      }
+
+      rmesa->RenderIndex = index;
+   }
+}
+
+/**********************************************************************/
+/*                 Validate state at pipeline start                   */
+/**********************************************************************/
+
+static void r128RunPipeline( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   if (rmesa->new_state || rmesa->NewGLState & _NEW_TEXTURE)
+      r128DDUpdateHWState( ctx );
+
+   if (!rmesa->Fallback && rmesa->NewGLState) {
+      if (rmesa->NewGLState & _R128_NEW_VERTEX_STATE)
+	 r128ChooseVertexState( ctx );
+
+      if (rmesa->NewGLState & _R128_NEW_RENDER_STATE)
+	 r128ChooseRenderState( ctx );
+
+      rmesa->NewGLState = 0;
+   }
+
+   _tnl_run_pipeline( ctx );
+}
+
+/**********************************************************************/
+/*                 High level hooks for t_vb_render.c                 */
+/**********************************************************************/
+
+/* This is called when Mesa switches between rendering triangle
+ * primitives (such as GL_POLYGON, GL_QUADS, GL_TRIANGLE_STRIP, etc),
+ * and lines, points and bitmaps.
+ *
+ * As the r128 uses triangles to render lines and points, it is
+ * necessary to turn off hardware culling when rendering these
+ * primitives.
+ */
+
+static void r128RasterPrimitive( GLcontext *ctx, GLuint hwprim )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+
+   rmesa->setup.dp_gui_master_cntl_c &= ~R128_GMC_BRUSH_NONE;
+
+   if ( ctx->Polygon.StippleFlag && hwprim == GL_TRIANGLES ) {
+      rmesa->setup.dp_gui_master_cntl_c |= R128_GMC_BRUSH_32x32_MONO_FG_LA;
+   }
+   else {
+      rmesa->setup.dp_gui_master_cntl_c |= R128_GMC_BRUSH_SOLID_COLOR;
+   }
+
+   rmesa->new_state |= R128_NEW_CONTEXT;
+   rmesa->dirty |= R128_UPLOAD_CONTEXT;
+
+   if (rmesa->hw_primitive != hwprim) {
+      FLUSH_BATCH( rmesa );
+      rmesa->hw_primitive = hwprim;
+   }
+}
+
+static void r128RenderPrimitive( GLcontext *ctx, GLenum prim )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   GLuint hw = hw_prim[prim];
+   rmesa->render_primitive = prim;
+   if (prim >= GL_TRIANGLES && (ctx->_TriangleCaps & DD_TRI_UNFILLED))
+      return;
+   r128RasterPrimitive( ctx, hw );
+}
+
+
+static void r128RenderStart( GLcontext *ctx )
+{
+   /* Check for projective texturing.  Make sure all texcoord
+    * pointers point to something.  (fix in mesa?)
+    */
+   r128CheckTexSizes( ctx );
+}
+
+static void r128RenderFinish( GLcontext *ctx )
+{
+   if (R128_CONTEXT(ctx)->RenderIndex & R128_FALLBACK_BIT)
+      _swrast_flush( ctx );
+}
+
+
+/**********************************************************************/
+/*           Transition to/from hardware rasterization.               */
+/**********************************************************************/
+
+void r128Fallback( GLcontext *ctx, GLuint bit, GLboolean mode )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   GLuint oldfallback = rmesa->Fallback;
+
+   if (mode) {
+      rmesa->Fallback |= bit;
+      if (oldfallback == 0) {
+	 FLUSH_BATCH( rmesa );
+	 _swsetup_Wakeup( ctx );
+	 rmesa->RenderIndex = ~0;
+      }
+   }
+   else {
+      rmesa->Fallback &= ~bit;
+      if (oldfallback == bit) {
+	 _swrast_flush( ctx );
+	 tnl->Driver.Render.Start = r128RenderStart;
+	 tnl->Driver.Render.PrimitiveNotify = r128RenderPrimitive;
+	 tnl->Driver.Render.Finish = r128RenderFinish;
+	 tnl->Driver.Render.BuildVertices = r128BuildVertices;
+	 rmesa->NewGLState |= (_R128_NEW_RENDER_STATE|
+			       _R128_NEW_VERTEX_STATE);
+      }
+   }
+}
+
+
+/**********************************************************************/
+/*                            Initialization.                         */
+/**********************************************************************/
+
+void r128InitTriFuncs( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   static int firsttime = 1;
+
+   if (firsttime) {
+      init_rast_tab();
+      firsttime = 0;
+   }
+
+   tnl->Driver.RunPipeline = r128RunPipeline;
+   tnl->Driver.Render.Start = r128RenderStart;
+   tnl->Driver.Render.Finish = r128RenderFinish;
+   tnl->Driver.Render.PrimitiveNotify = r128RenderPrimitive;
+   tnl->Driver.Render.ResetLineStipple = _swrast_ResetLineStipple;
+   tnl->Driver.Render.BuildVertices = r128BuildVertices;
+   rmesa->NewGLState |= (_R128_NEW_RENDER_STATE|
+			 _R128_NEW_VERTEX_STATE);
+
+/*     r128Fallback( ctx, 0x100000, 1 ); */
+}
diff --git a/src/mesa/drivers/dri/r128/r128_tris.h b/src/mesa/drivers/dri/r128/r128_tris.h
new file mode 100644
index 00000000000..755d3320b03
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_tris.h
@@ -0,0 +1,48 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_tris.h,v 1.8 2002/10/30 12:51:43 alanh Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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
+ATI, VA LINUX SYSTEMS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell <[email protected]>
+ *
+ */
+
+#ifndef __R128_TRIS_H__
+#define __R128_TRIS_H__
+
+#include "mtypes.h"
+
+extern void r128InitTriFuncs( GLcontext *ctx );
+
+
+extern void r128Fallback( GLcontext *ctx, GLuint bit, GLboolean mode );
+#define FALLBACK( rmesa, bit, mode ) r128Fallback( rmesa->glCtx, bit, mode )
+
+
+#endif /* __R128_TRIS_H__ */
diff --git a/src/mesa/drivers/dri/r128/r128_vb.c b/src/mesa/drivers/dri/r128/r128_vb.c
new file mode 100644
index 00000000000..77389016486
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_vb.c
@@ -0,0 +1,522 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_vb.c,v 1.15 2002/10/30 12:51:43 alanh Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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
+ATI, VA LINUX SYSTEMS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell <[email protected]>
+ *
+ */
+
+#include "glheader.h"
+#include "mtypes.h"
+#include "imports.h"
+#include "macros.h"
+#include "colormac.h"
+
+#include "swrast_setup/swrast_setup.h"
+#include "tnl/t_context.h"
+
+#include "r128_context.h"
+#include "r128_vb.h"
+#include "r128_ioctl.h"
+#include "r128_tris.h"
+#include "r128_state.h"
+
+
+#define R128_TEX1_BIT       0x1
+#define R128_TEX0_BIT       0x2
+#define R128_RGBA_BIT       0x4
+#define R128_SPEC_BIT       0x8
+#define R128_FOG_BIT	    0x10
+#define R128_XYZW_BIT       0x20
+#define R128_PTEX_BIT       0x40
+#define R128_MAX_SETUP      0x80
+
+static struct {
+   void                (*emit)( GLcontext *, GLuint, GLuint, void *, GLuint );
+   interp_func		interp;
+   copy_pv_func	        copy_pv;
+   GLboolean           (*check_tex_sizes)( GLcontext *ctx );
+   GLuint               vertex_size;
+   GLuint               vertex_stride_shift;
+   GLuint               vertex_format;
+} setup_tab[R128_MAX_SETUP];
+
+#define TINY_VERTEX_FORMAT		(R128_CCE_VC_FRMT_DIFFUSE_ARGB)
+
+#define NOTEX_VERTEX_FORMAT		(R128_CCE_VC_FRMT_RHW |		\
+					 R128_CCE_VC_FRMT_DIFFUSE_ARGB |\
+					 R128_CCE_VC_FRMT_SPEC_FRGB)
+
+#define TEX0_VERTEX_FORMAT		(R128_CCE_VC_FRMT_RHW |		\
+					 R128_CCE_VC_FRMT_DIFFUSE_ARGB |\
+					 R128_CCE_VC_FRMT_SPEC_FRGB |	\
+					 R128_CCE_VC_FRMT_S_T)
+
+#define TEX1_VERTEX_FORMAT		(R128_CCE_VC_FRMT_RHW |		\
+					 R128_CCE_VC_FRMT_DIFFUSE_ARGB |\
+					 R128_CCE_VC_FRMT_SPEC_FRGB |	\
+					 R128_CCE_VC_FRMT_S_T |		\
+					 R128_CCE_VC_FRMT_S2_T2)
+
+
+#define PROJ_TEX1_VERTEX_FORMAT 0
+#define TEX2_VERTEX_FORMAT      0
+#define TEX3_VERTEX_FORMAT      0
+#define PROJ_TEX3_VERTEX_FORMAT 0
+
+#define DO_XYZW (IND & R128_XYZW_BIT)
+#define DO_RGBA (IND & R128_RGBA_BIT)
+#define DO_SPEC (IND & R128_SPEC_BIT)
+#define DO_FOG  (IND & R128_FOG_BIT)
+#define DO_TEX0 (IND & R128_TEX0_BIT)
+#define DO_TEX1 (IND & R128_TEX1_BIT)
+#define DO_TEX2 0
+#define DO_TEX3 0
+#define DO_PTEX (IND & R128_PTEX_BIT)
+
+#define VERTEX r128Vertex
+#define VERTEX_COLOR r128_color_t
+#define LOCALVARS r128ContextPtr rmesa = R128_CONTEXT(ctx);
+#define GET_VIEWPORT_MAT() rmesa->hw_viewport
+#define GET_TEXSOURCE(n)  rmesa->tmu_source[n]
+#define GET_VERTEX_FORMAT() rmesa->vertex_format
+#define GET_VERTEX_STORE() rmesa->verts
+#define GET_VERTEX_STRIDE_SHIFT() rmesa->vertex_stride_shift
+#define INVALIDATE_STORED_VERTICES()
+#define GET_UBYTE_COLOR_STORE() &rmesa->UbyteColor
+#define GET_UBYTE_SPEC_COLOR_STORE() &rmesa->UbyteSecondaryColor
+
+#define HAVE_HW_VIEWPORT    0
+#define HAVE_HW_DIVIDE      0
+#define HAVE_RGBA_COLOR     0
+#define HAVE_TINY_VERTICES  1
+#define HAVE_NOTEX_VERTICES 1
+#define HAVE_TEX0_VERTICES  1
+#define HAVE_TEX1_VERTICES  1
+#define HAVE_TEX2_VERTICES  0
+#define HAVE_TEX3_VERTICES  0
+#define HAVE_PTEX_VERTICES  0	/* r128 rhw2 not supported by template */
+
+#define UNVIEWPORT_VARS  GLfloat h = R128_CONTEXT(ctx)->driDrawable->h
+#define UNVIEWPORT_X(x)  x - SUBPIXEL_X
+#define UNVIEWPORT_Y(y)  - y + h + SUBPIXEL_Y
+#define UNVIEWPORT_Z(z)  z / rmesa->depth_scale
+
+#define PTEX_FALLBACK() FALLBACK(R128_CONTEXT(ctx), R128_FALLBACK_TEXTURE, 1)
+
+#define IMPORT_FLOAT_COLORS r128_import_float_colors
+#define IMPORT_FLOAT_SPEC_COLORS r128_import_float_spec_colors
+
+#define INTERP_VERTEX setup_tab[rmesa->SetupIndex].interp
+#define COPY_PV_VERTEX setup_tab[rmesa->SetupIndex].copy_pv
+
+/***********************************************************************
+ *         Generate  pv-copying and translation functions              *
+ ***********************************************************************/
+
+#define TAG(x) r128_##x
+#include "tnl_dd/t_dd_vb.c"
+
+/***********************************************************************
+ *             Generate vertex emit and interp functions               *
+ ***********************************************************************/
+
+
+#define IND (R128_XYZW_BIT|R128_RGBA_BIT)
+#define TAG(x) x##_wg
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_XYZW_BIT|R128_RGBA_BIT|R128_SPEC_BIT)
+#define TAG(x) x##_wgs
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_XYZW_BIT|R128_RGBA_BIT|R128_TEX0_BIT)
+#define TAG(x) x##_wgt0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_XYZW_BIT|R128_RGBA_BIT|R128_TEX0_BIT|R128_TEX1_BIT)
+#define TAG(x) x##_wgt0t1
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_XYZW_BIT|R128_RGBA_BIT|R128_TEX0_BIT|R128_PTEX_BIT)
+#define TAG(x) x##_wgpt0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_XYZW_BIT|R128_RGBA_BIT|R128_SPEC_BIT|R128_TEX0_BIT)
+#define TAG(x) x##_wgst0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_XYZW_BIT|R128_RGBA_BIT|R128_SPEC_BIT|R128_TEX0_BIT|\
+             R128_TEX1_BIT)
+#define TAG(x) x##_wgst0t1
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_XYZW_BIT|R128_RGBA_BIT|R128_SPEC_BIT|R128_TEX0_BIT|\
+             R128_PTEX_BIT)
+#define TAG(x) x##_wgspt0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_XYZW_BIT|R128_RGBA_BIT|R128_FOG_BIT)
+#define TAG(x) x##_wgf
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_XYZW_BIT|R128_RGBA_BIT|R128_FOG_BIT|R128_SPEC_BIT)
+#define TAG(x) x##_wgfs
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_XYZW_BIT|R128_RGBA_BIT|R128_FOG_BIT|R128_TEX0_BIT)
+#define TAG(x) x##_wgft0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_XYZW_BIT|R128_RGBA_BIT|R128_FOG_BIT|R128_TEX0_BIT|\
+             R128_TEX1_BIT)
+#define TAG(x) x##_wgft0t1
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_XYZW_BIT|R128_RGBA_BIT|R128_FOG_BIT|R128_TEX0_BIT|\
+             R128_PTEX_BIT)
+#define TAG(x) x##_wgfpt0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_XYZW_BIT|R128_RGBA_BIT|R128_FOG_BIT|R128_SPEC_BIT|\
+             R128_TEX0_BIT)
+#define TAG(x) x##_wgfst0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_XYZW_BIT|R128_RGBA_BIT|R128_FOG_BIT|R128_SPEC_BIT|\
+             R128_TEX0_BIT|R128_TEX1_BIT)
+#define TAG(x) x##_wgfst0t1
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_XYZW_BIT|R128_RGBA_BIT|R128_FOG_BIT|R128_SPEC_BIT|\
+             R128_TEX0_BIT|R128_PTEX_BIT)
+#define TAG(x) x##_wgfspt0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_TEX0_BIT)
+#define TAG(x) x##_t0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_TEX0_BIT|R128_TEX1_BIT)
+#define TAG(x) x##_t0t1
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_FOG_BIT)
+#define TAG(x) x##_f
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_FOG_BIT|R128_TEX0_BIT)
+#define TAG(x) x##_ft0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_FOG_BIT|R128_TEX0_BIT|R128_TEX1_BIT)
+#define TAG(x) x##_ft0t1
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_RGBA_BIT)
+#define TAG(x) x##_g
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_RGBA_BIT|R128_SPEC_BIT)
+#define TAG(x) x##_gs
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_RGBA_BIT|R128_TEX0_BIT)
+#define TAG(x) x##_gt0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_RGBA_BIT|R128_TEX0_BIT|R128_TEX1_BIT)
+#define TAG(x) x##_gt0t1
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_RGBA_BIT|R128_SPEC_BIT|R128_TEX0_BIT)
+#define TAG(x) x##_gst0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_RGBA_BIT|R128_SPEC_BIT|R128_TEX0_BIT|R128_TEX1_BIT)
+#define TAG(x) x##_gst0t1
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_RGBA_BIT|R128_FOG_BIT)
+#define TAG(x) x##_gf
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_RGBA_BIT|R128_FOG_BIT|R128_SPEC_BIT)
+#define TAG(x) x##_gfs
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_RGBA_BIT|R128_FOG_BIT|R128_TEX0_BIT)
+#define TAG(x) x##_gft0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_RGBA_BIT|R128_FOG_BIT|R128_TEX0_BIT|R128_TEX1_BIT)
+#define TAG(x) x##_gft0t1
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_RGBA_BIT|R128_FOG_BIT|R128_SPEC_BIT|R128_TEX0_BIT)
+#define TAG(x) x##_gfst0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (R128_RGBA_BIT|R128_FOG_BIT|R128_SPEC_BIT|R128_TEX0_BIT|\
+             R128_TEX1_BIT)
+#define TAG(x) x##_gfst0t1
+#include "tnl_dd/t_dd_vbtmp.h"
+
+
+static void init_setup_tab( void )
+{
+   init_wg();
+   init_wgs();
+   init_wgt0();
+   init_wgt0t1();
+   init_wgpt0();
+   init_wgst0();
+   init_wgst0t1();
+   init_wgspt0();
+   init_wgf();
+   init_wgfs();
+   init_wgft0();
+   init_wgft0t1();
+   init_wgfpt0();
+   init_wgfst0();
+   init_wgfst0t1();
+   init_wgfspt0();
+   init_t0();
+   init_t0t1();
+   init_f();
+   init_ft0();
+   init_ft0t1();
+   init_g();
+   init_gs();
+   init_gt0();
+   init_gt0t1();
+   init_gst0();
+   init_gst0t1();
+   init_gf();
+   init_gfs();
+   init_gft0();
+   init_gft0t1();
+   init_gfst0();
+   init_gfst0t1();
+}
+
+
+
+void r128PrintSetupFlags(char *msg, GLuint flags )
+{
+   fprintf(stderr, "%s(%x): %s%s%s%s%s%s\n",
+	   msg,
+	   (int)flags,
+	   (flags & R128_XYZW_BIT)     ? " xyzw," : "",
+	   (flags & R128_RGBA_BIT)     ? " rgba," : "",
+	   (flags & R128_SPEC_BIT)     ? " spec," : "",
+	   (flags & R128_FOG_BIT)      ? " fog," : "",
+	   (flags & R128_TEX0_BIT)     ? " tex-0," : "",
+	   (flags & R128_TEX1_BIT)     ? " tex-1," : "");
+}
+
+
+
+void r128CheckTexSizes( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT( ctx );
+
+   if (!setup_tab[rmesa->SetupIndex].check_tex_sizes(ctx)) {
+      TNLcontext *tnl = TNL_CONTEXT(ctx);
+
+      /* Invalidate stored verts
+       */
+      rmesa->SetupNewInputs = ~0;
+      rmesa->SetupIndex |= R128_PTEX_BIT;
+
+      if (!rmesa->Fallback &&
+	  !(ctx->_TriangleCaps & (DD_TRI_LIGHT_TWOSIDE|DD_TRI_UNFILLED))) {
+	 tnl->Driver.Render.Interp = setup_tab[rmesa->SetupIndex].interp;
+	 tnl->Driver.Render.CopyPV = setup_tab[rmesa->SetupIndex].copy_pv;
+      }
+   }
+}
+
+void r128BuildVertices( GLcontext *ctx,
+			GLuint start,
+			GLuint count,
+			GLuint newinputs )
+{
+   r128ContextPtr rmesa = R128_CONTEXT( ctx );
+   GLubyte *v = ((GLubyte *)rmesa->verts + (start<<rmesa->vertex_stride_shift));
+   GLuint stride = 1<<rmesa->vertex_stride_shift;
+
+   newinputs |= rmesa->SetupNewInputs;
+   rmesa->SetupNewInputs = 0;
+
+   if (!newinputs)
+      return;
+
+   if (newinputs & VERT_BIT_CLIP) {
+      setup_tab[rmesa->SetupIndex].emit( ctx, start, count, v, stride );
+   } else {
+      GLuint ind = 0;
+
+      if (newinputs & VERT_BIT_COLOR0)
+	 ind |= R128_RGBA_BIT;
+
+      if (newinputs & VERT_BIT_COLOR1)
+	 ind |= R128_SPEC_BIT;
+
+      if (newinputs & VERT_BIT_TEX0)
+	 ind |= R128_TEX0_BIT;
+
+      if (newinputs & VERT_BIT_TEX1)
+	 ind |= R128_TEX1_BIT;
+
+      if (newinputs & VERT_BIT_FOG)
+	 ind |= R128_FOG_BIT;
+
+      if (rmesa->SetupIndex & R128_PTEX_BIT)
+	 ind = ~0;
+
+      ind &= rmesa->SetupIndex;
+
+      if (ind) {
+	 setup_tab[ind].emit( ctx, start, count, v, stride );
+      }
+   }
+}
+
+void r128ChooseVertexState( GLcontext *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   r128ContextPtr rmesa = R128_CONTEXT( ctx );
+   GLuint ind = R128_XYZW_BIT|R128_RGBA_BIT;
+
+   if (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR)
+      ind |= R128_SPEC_BIT;
+
+   if (ctx->Fog.Enabled)
+      ind |= R128_FOG_BIT;
+
+   if (ctx->Texture._EnabledUnits) {
+      ind |= R128_TEX0_BIT;
+      if (ctx->Texture.Unit[0]._ReallyEnabled &&
+	  ctx->Texture.Unit[1]._ReallyEnabled)
+	 ind |= R128_TEX1_BIT;
+   }
+
+   rmesa->SetupIndex = ind;
+
+   if (ctx->_TriangleCaps & (DD_TRI_LIGHT_TWOSIDE|DD_TRI_UNFILLED)) {
+      tnl->Driver.Render.Interp = r128_interp_extras;
+      tnl->Driver.Render.CopyPV = r128_copy_pv_extras;
+   } else {
+      tnl->Driver.Render.Interp = setup_tab[ind].interp;
+      tnl->Driver.Render.CopyPV = setup_tab[ind].copy_pv;
+   }
+
+   if (setup_tab[ind].vertex_format != rmesa->vertex_format) {
+      FLUSH_BATCH(rmesa);
+      rmesa->vertex_format = setup_tab[ind].vertex_format;
+      rmesa->vertex_size = setup_tab[ind].vertex_size;
+      rmesa->vertex_stride_shift = setup_tab[ind].vertex_stride_shift;
+   }
+}
+
+
+
+void r128_emit_contiguous_verts( GLcontext *ctx,
+				 GLuint start,
+				 GLuint count )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   GLuint vertex_size = rmesa->vertex_size * 4;
+   GLuint *dest = r128AllocDmaLow( rmesa, (count-start) * vertex_size);
+   setup_tab[rmesa->SetupIndex].emit( ctx, start, count, dest, vertex_size );
+}
+
+
+#if 0
+void r128_emit_indexed_verts( GLcontext *ctx, GLuint start, GLuint count )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   GLuint vertex_size = rmesa->vertex_size * 4;
+   GLuint bufsz = (count-start) * vertex_size;
+   CARD32 *dest;
+
+   rmesa->vertex_low = (rmesa->vertex_low + 63) & ~63; /* alignment */
+   rmesa->vertex_last_prim = rmesa->vertex_low;
+
+   dest = r128AllocDmaLow( rmesa, bufsz, __FUNCTION__);
+   setup_tab[rmesa->SetupIndex].emit( ctx, start, count, dest, vertex_size );
+
+   rmesa->retained_buffer = rmesa->vertex_buffer;
+   rmesa->vb_offset = (rmesa->vertex_buffer->idx * R128_BUFFER_SIZE +
+		       rmesa->vertex_low - bufsz);
+
+   rmesa->vertex_low = (rmesa->vertex_low + 0x7) & ~0x7;  /* alignment */
+   rmesa->vertex_last_prim = rmesa->vertex_low;
+}
+#endif
+
+
+void r128InitVB( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   GLuint size = TNL_CONTEXT(ctx)->vb.Size;
+
+   rmesa->verts = ALIGN_MALLOC(size * 4 * 16, 32);
+
+   {
+      static int firsttime = 1;
+      if (firsttime) {
+	 init_setup_tab();
+	 firsttime = 0;
+      }
+   }
+}
+
+
+void r128FreeVB( GLcontext *ctx )
+{
+   r128ContextPtr rmesa = R128_CONTEXT(ctx);
+   if (rmesa->verts) {
+      ALIGN_FREE(rmesa->verts);
+      rmesa->verts = 0;
+   }
+
+
+   if (rmesa->UbyteSecondaryColor.Ptr) {
+      ALIGN_FREE(rmesa->UbyteSecondaryColor.Ptr);
+      rmesa->UbyteSecondaryColor.Ptr = 0;
+   }
+
+   if (rmesa->UbyteColor.Ptr) {
+      ALIGN_FREE(rmesa->UbyteColor.Ptr);
+      rmesa->UbyteColor.Ptr = 0;
+   }
+}
diff --git a/src/mesa/drivers/dri/r128/r128_vb.h b/src/mesa/drivers/dri/r128/r128_vb.h
new file mode 100644
index 00000000000..31afa74f1cf
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/r128_vb.h
@@ -0,0 +1,74 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r128/r128_vb.h,v 1.8 2002/10/30 12:51:46 alanh Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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
+ATI, VA LINUX SYSTEMS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell <[email protected]>
+ *
+ */
+
+#ifndef R128VB_INC
+#define R128VB_INC
+
+#include "mtypes.h"
+#include "swrast/swrast.h"
+#include "r128_context.h"
+
+#define _R128_NEW_VERTEX_STATE (_DD_NEW_SEPARATE_SPECULAR |		\
+				_DD_NEW_TRI_LIGHT_TWOSIDE |		\
+				_DD_NEW_TRI_UNFILLED |			\
+				_NEW_TEXTURE |				\
+				_NEW_FOG)
+
+extern void r128CheckTexSizes( GLcontext *ctx );
+extern void r128ChooseVertexState( GLcontext *ctx );
+
+extern void r128BuildVertices( GLcontext *ctx, GLuint start, GLuint count,
+				 GLuint newinputs );
+
+extern void r128PrintSetupFlags(char *msg, GLuint flags );
+
+extern void r128InitVB( GLcontext *ctx );
+extern void r128FreeVB( GLcontext *ctx );
+
+extern void r128_emit_contiguous_verts( GLcontext *ctx,
+					  GLuint start,
+					  GLuint count );
+
+extern void r128_emit_indexed_verts( GLcontext *ctx,
+				       GLuint start,
+				       GLuint count );
+
+extern void r128_translate_vertex( GLcontext *ctx,
+				     const r128Vertex *src,
+				     SWvertex *dst );
+
+extern void r128_print_vertex( GLcontext *ctx, const r128Vertex *v );
+
+#endif
diff --git a/src/mesa/drivers/dri/r128/server/r128.h b/src/mesa/drivers/dri/r128/server/r128.h
new file mode 100644
index 00000000000..fba01d6d604
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/server/r128.h
@@ -0,0 +1,566 @@
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/ati/r128.h,v 1.24 2002/12/16 16:19:10 dawes Exp $ */
+/*
+ * Copyright 1999, 2000 ATI Technologies Inc., Markham, Ontario,
+ *                      Precision Insight, Inc., Cedar Park, Texas, and
+ *                      VA Linux Systems Inc., Fremont, California.
+ *
+ * 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, 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
+ * NON-INFRINGEMENT.  IN NO EVENT SHALL ATI, PRECISION INSIGHT, VA LINUX
+ * SYSTEMS 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:
+ *   Rickard E. Faith <[email protected]>
+ *   Kevin E. Martin <[email protected]>
+ *
+ */
+
+#ifndef _R128_H_
+#define _R128_H_
+
+#include "dri_util.h"
+#ifndef _SOLO
+#include "xf86str.h"
+
+				/* PCI support */
+#include "xf86Pci.h"
+
+				/* XAA and Cursor Support */
+#include "xaa.h"
+#include "xf86Cursor.h"
+
+				/* DDC support */
+#include "xf86DDC.h"
+
+				/* Xv support */
+#include "xf86xv.h"
+
+				/* DRI support */
+#ifdef XF86DRI
+#define _XF86DRI_SERVER_
+#include "r128_dripriv.h"
+#include "dri.h"
+#include "GL/glxint.h"
+#endif
+#endif
+#ifdef _SOLO
+#define XF86DRI
+#endif
+
+#define R128_DEBUG          0   /* Turn off debugging output               */
+#define R128_IDLE_RETRY    32   /* Fall out of idle loops after this count */
+#define R128_TIMEOUT  2000000   /* Fall out of wait loops after this count */
+#define R128_MMIOSIZE  0x4000
+
+#define R128_VBIOS_SIZE 0x00010000
+
+#if R128_DEBUG
+#define R128TRACE(x)                                          \
+    do {                                                      \
+	ErrorF("(**) %s(%d): ", R128_NAME, pScrn->scrnIndex); \
+	ErrorF x;                                             \
+    } while (0);
+#else
+#define R128TRACE(x)
+#endif
+
+
+/* Other macros */
+#define R128_ARRAY_SIZE(x)  (sizeof(x)/sizeof(x[0]))
+#define R128_ALIGN(x,bytes) (((x) + ((bytes) - 1)) & ~((bytes) - 1))
+#define R128PTR(pScrn) ((R128InfoPtr)(pScrn)->driverPrivate)
+    
+/**
+ * \brief Chip families.
+ */
+typedef enum {
+    CHIP_FAMILY_UNKNOWN,
+    CHIP_FAMILY_R128_PCI,
+    CHIP_FAMILY_R128_AGP,
+} R128ChipFamily;
+
+typedef struct {        /* All values in XCLKS    */
+    int  ML;            /* Memory Read Latency    */
+    int  MB;            /* Memory Burst Length    */
+    int  Trcd;          /* RAS to CAS delay       */
+    int  Trp;           /* RAS percentage         */
+    int  Twr;           /* Write Recovery         */
+    int  CL;            /* CAS Latency            */
+    int  Tr2w;          /* Read to Write Delay    */
+    int  Rloop;         /* Loop Latency           */
+    int  Rloop_fudge;   /* Add to ML to get Rloop */
+    char *name;
+} R128RAMRec, *R128RAMPtr;
+
+typedef struct {
+				/* Common registers */
+    CARD32     ovr_clr;
+    CARD32     ovr_wid_left_right;
+    CARD32     ovr_wid_top_bottom;
+    CARD32     ov0_scale_cntl;
+    CARD32     mpp_tb_config;
+    CARD32     mpp_gp_config;
+    CARD32     subpic_cntl;
+    CARD32     viph_control;
+    CARD32     i2c_cntl_1;
+    CARD32     gen_int_cntl;
+    CARD32     cap0_trig_cntl;
+    CARD32     cap1_trig_cntl;
+    CARD32     bus_cntl;
+    CARD32     config_cntl;
+
+				/* Other registers to save for VT switches */
+    CARD32     dp_datatype;
+    CARD32     gen_reset_cntl;
+    CARD32     clock_cntl_index;
+    CARD32     amcgpio_en_reg;
+    CARD32     amcgpio_mask;
+
+				/* CRTC registers */
+    CARD32     crtc_gen_cntl;
+    CARD32     crtc_ext_cntl;
+    CARD32     dac_cntl;
+    CARD32     crtc_h_total_disp;
+    CARD32     crtc_h_sync_strt_wid;
+    CARD32     crtc_v_total_disp;
+    CARD32     crtc_v_sync_strt_wid;
+    CARD32     crtc_offset;
+    CARD32     crtc_offset_cntl;
+    CARD32     crtc_pitch;
+
+				/* CRTC2 registers */
+    CARD32     crtc2_gen_cntl;
+
+				/* Flat panel registers */
+    CARD32     fp_crtc_h_total_disp;
+    CARD32     fp_crtc_v_total_disp;
+    CARD32     fp_gen_cntl;
+    CARD32     fp_h_sync_strt_wid;
+    CARD32     fp_horz_stretch;
+    CARD32     fp_panel_cntl;
+    CARD32     fp_v_sync_strt_wid;
+    CARD32     fp_vert_stretch;
+    CARD32     lvds_gen_cntl;
+    CARD32     tmds_crc;
+    CARD32     tmds_transmitter_cntl;
+
+				/* Computed values for PLL */
+    CARD32     dot_clock_freq;
+    CARD32     pll_output_freq;
+    int        feedback_div;
+    int        post_div;
+
+				/* PLL registers */
+    CARD32     ppll_ref_div;
+    CARD32     ppll_div_3;
+    CARD32     htotal_cntl;
+
+				/* DDA register */
+    CARD32     dda_config;
+    CARD32     dda_on_off;
+
+				/* Pallet */
+    GLboolean  palette_valid;
+    CARD32     palette[256];
+} R128SaveRec, *R128SavePtr;
+
+#ifndef _SOLO
+typedef struct {
+    CARD16        reference_freq;
+    CARD16        reference_div;
+    CARD32        min_pll_freq;
+    CARD32        max_pll_freq;
+    CARD16        xclk;
+} R128PLLRec, *R128PLLPtr;
+
+typedef struct {
+    int                bitsPerPixel;
+    int                depth;
+    int                displayWidth;
+    int                pixel_code;
+    int                pixel_bytes;
+    DisplayModePtr     mode;
+} R128FBLayout;
+#endif
+
+typedef struct {
+#ifndef _SOLO    
+    EntityInfoPtr     pEnt;
+    pciVideoPtr       PciInfo;
+    PCITAG            PciTag;
+#endif    
+    int               Chipset;
+    GLboolean              Primary;
+
+    GLboolean              FBDev;
+
+    unsigned long     LinearAddr;   /* Frame buffer physical address         */
+    unsigned long     BIOSAddr;     /* BIOS physical address                 */
+
+    unsigned char     *MMIO;        /* Map of MMIO region                    */
+    unsigned char     *FB;          /* Map of frame buffer                   */
+
+    CARD32            MemCntl;
+    CARD32            BusCntl;
+    unsigned long     FbMapSize;    /* Size of frame buffer, in bytes        */
+    int               Flags;        /* Saved copy of mode flags              */
+
+#ifndef _SOLO    
+    CARD8             BIOSDisplay;  /* Device the BIOS is set to display to  */
+
+    GLboolean         HasPanelRegs; /* Current chip can connect to a FP      */
+    CARD8             *VBIOS;       /* Video BIOS for mode validation on FPs */
+    int               FPBIOSstart;  /* Start of the flat panel info          */
+#endif
+				/* Computed values for FPs */
+    int               PanelXRes;
+    int               PanelYRes;
+    int               HOverPlus;
+    int               HSyncWidth;
+    int               HBlank;
+    int               VOverPlus;
+    int               VSyncWidth;
+    int               VBlank;
+    int               PanelPwrDly;
+#ifndef _SOLO
+    R128PLLRec        pll;
+    R128RAMPtr        ram;
+
+    R128SaveRec       SavedReg;     /* Original (text) mode                  */
+    R128SaveRec       ModeReg;      /* Current mode                          */
+    GLboolean              (*CloseScreen)(int, ScreenPtr);
+    void              (*BlockHandler)(int, pointer, pointer, pointer);
+
+    GLboolean              PaletteSavedOnVT; /* Palette saved on last VT switch   */
+
+    XAAInfoRecPtr     accel;
+    GLboolean              accelOn;
+    xf86CursorInfoPtr cursor;
+#endif    
+    unsigned long     cursor_start;
+    unsigned long     cursor_end;
+
+    /*
+     * XAAForceTransBlit is used to change the behavior of the XAA
+     * SetupForScreenToScreenCopy function, to make it DGA-friendly.
+     */
+    GLboolean              XAAForceTransBlit;
+
+    int               fifo_slots;   /* Free slots in the FIFO (64 max)       */
+    int               pix24bpp;     /* Depth of pixmap for 24bpp framebuffer */
+    GLboolean              dac6bits;     /* Use 6 bit DAC?                        */
+
+				/* Computed values for Rage 128 */
+    int               pitch;
+    int               datatype;
+    CARD32            dp_gui_master_cntl;
+
+				/* Saved values for ScreenToScreenCopy */
+    int               xdir;
+    int               ydir;
+
+				/* ScanlineScreenToScreenColorExpand support */
+    unsigned char     *scratch_buffer[1];
+    unsigned char     *scratch_save;
+    int               scanline_x;
+    int               scanline_y;
+    int               scanline_w;
+    int               scanline_h;
+#ifdef XF86DRI
+    int               scanline_hpass;
+    int               scanline_x1clip;
+    int               scanline_x2clip;
+    int               scanline_rop;
+    int               scanline_fg;
+    int               scanline_bg;
+#endif /* XF86DRI */
+    int               scanline_words;
+    int               scanline_direct;
+    int               scanline_bpp; /* Only used for ImageWrite */
+
+#ifndef _SOLO    
+    DGAModePtr        DGAModes;
+    int               numDGAModes;
+    GLboolean         DGAactive;
+    int               DGAViewportStatus;
+    DGAFunctionRec    DGAFuncs;
+
+    R128FBLayout      CurrentLayout;
+#endif    
+#ifdef XF86DRI
+    drmContext        drmCtx;
+#ifndef _SOLO    
+    DRIInfoPtr        pDRIInfo;
+    int               numVisualConfigs;
+    __GLXvisualConfig *pVisualConfigs;
+    R128ConfigPrivPtr pVisualConfigsPriv;
+#endif    
+
+    drmSize           registerSize;
+    drmHandle         registerHandle;
+
+    GLboolean         IsPCI;            /* Current card is a PCI card */
+    drmSize           pciSize;
+    drmHandle         pciMemHandle;
+    unsigned char     *PCI;             /* Map */
+
+    GLboolean         allowPageFlip;    /* Enable 3d page flipping */
+    GLboolean         have3DWindows;    /* Are there any 3d clients? */
+    int               drmMinor;
+
+    drmSize           agpSize;
+    drmHandle         agpMemHandle;     /* Handle from drmAgpAlloc */
+    unsigned long     agpOffset;
+    unsigned char     *AGP;             /* Map */
+    int               agpMode;
+
+    GLboolean         CCEInUse;         /* CCE is currently active */
+    int               CCEMode;          /* CCE mode that server/clients use */
+    int               CCEFifoSize;      /* Size of the CCE command FIFO */
+    GLboolean         CCESecure;        /* CCE security enabled */
+    int               CCEusecTimeout;   /* CCE timeout in usecs */
+
+				/* CCE ring buffer data */
+    unsigned long     ringStart;        /* Offset into AGP space */
+    drmHandle         ringHandle;       /* Handle from drmAddMap */
+    drmSize           ringMapSize;      /* Size of map */
+    int               ringSize;         /* Size of ring (in MB) */
+    unsigned char     *ring;            /* Map */
+    int               ringSizeLog2QW;
+
+    unsigned long     ringReadOffset;   /* Offset into AGP space */
+    drmHandle         ringReadPtrHandle; /* Handle from drmAddMap */
+    drmSize           ringReadMapSize;  /* Size of map */
+    unsigned char     *ringReadPtr;     /* Map */
+
+				/* CCE vertex/indirect buffer data */
+    unsigned long     bufStart;        /* Offset into AGP space */
+    drmHandle         bufHandle;       /* Handle from drmAddMap */
+    drmSize           bufMapSize;      /* Size of map */
+    int               bufSize;         /* Size of buffers (in MB) */
+    unsigned char     *buf;            /* Map */
+    int               bufNumBufs;      /* Number of buffers */
+    drmBufMapPtr      buffers;         /* Buffer map */
+
+				/* CCE AGP Texture data */
+    unsigned long     agpTexStart;      /* Offset into AGP space */
+    drmHandle         agpTexHandle;     /* Handle from drmAddMap */
+    drmSize           agpTexMapSize;    /* Size of map */
+    int               agpTexSize;       /* Size of AGP tex space (in MB) */
+    unsigned char     *agpTex;          /* Map */
+    int               log2AGPTexGran;
+
+				/* CCE 2D accleration */
+    drmBufPtr         indirectBuffer;
+    int               indirectStart;
+
+				/* DRI screen private data */
+    int               fbX;
+    int               fbY;
+    int               backX;
+    int               backY;
+    int               depthX;
+    int               depthY;
+
+    int               frontOffset;
+    int               frontPitch;
+    int               backOffset;
+    int               backPitch;
+    int               depthOffset;
+    int               depthPitch;
+    int               spanOffset;
+    int               textureOffset;
+    int               textureSize;
+    int               log2TexGran;
+
+				/* Saved scissor values */
+    CARD32            sc_left;
+    CARD32            sc_right;
+    CARD32            sc_top;
+    CARD32            sc_bottom;
+
+    CARD32            re_top_left;
+    CARD32            re_width_height;
+
+    CARD32            aux_sc_cntl;
+
+    int               irq;
+    CARD32            gen_int_cntl;
+
+    GLboolean              DMAForXv;
+#endif
+
+#ifndef _SOLO    
+    XF86VideoAdaptorPtr adaptor;
+    void              (*VideoTimerCallback)(ScrnInfoPtr, Time);
+    int               videoKey;
+    GLboolean              showCache;
+    OptionInfoPtr     Options;
+
+    GLboolean              isDFP;
+    GLboolean              isPro2;
+    I2CBusPtr         pI2CBus;
+    CARD32            DDCReg;
+#endif
+} R128InfoRec, *R128InfoPtr;
+
+#define R128WaitForFifo(pScrn, entries)                                      \
+do {                                                                         \
+    if (info->fifo_slots < entries) R128WaitForFifoFunction(pScrn, entries); \
+    info->fifo_slots -= entries;                                             \
+} while (0)
+
+extern void        r128WaitForFifoFunction(const DRIDriverContext *ctx, int entries);
+extern void        r128WaitForIdle(const DRIDriverContext *ctx);
+
+extern void        r128WaitForVerticalSync(const DRIDriverContext *ctx);
+
+extern GLboolean        r128AccelInit(const DRIDriverContext *ctx);
+extern void        r128EngineInit(const DRIDriverContext *ctx);
+extern GLboolean        r128CursorInit(const DRIDriverContext *ctx);
+extern GLboolean        r128DGAInit(const DRIDriverContext *ctx);
+
+extern void        r128InitVideo(const DRIDriverContext *ctx);
+
+extern GLboolean        r128DRIScreenInit(const DRIDriverContext *ctx);
+extern void        r128DRICloseScreen(const DRIDriverContext *ctx);
+extern GLboolean        r128DRIFinishScreenInit(const DRIDriverContext *ctx);
+
+#define R128CCE_START(ctx, info)					\
+do {									\
+    int _ret = drmCommandNone(ctx->drmFD, DRM_R128_CCE_START);		\
+    if (_ret) {								\
+	   fprintf(stderr,				\
+		   "%s: CCE start %d\n", __FUNCTION__, _ret);		\
+    }									\
+} while (0)
+
+#define R128CCE_STOP(ctx, info)					\
+do {									\
+    int _ret = R128CCEStop(ctx);					\
+    if (_ret) {								\
+	   fprintf(stderr,				\
+		   "%s: CCE stop %d\n", __FUNCTION__, _ret);		\
+    }									\
+} while (0)
+
+#define R128CCE_RESET(ctx, info)					\
+do {									\
+    if (info->directRenderingEnabled					\
+	&& R128CCE_USE_RING_BUFFER(info->CCEMode)) {			\
+	int _ret = drmCommandNone(info->drmFD, DRM_R128_CCE_RESET);	\
+	if (_ret) {							\
+	       fprintf(stderr,			\
+		       "%s: CCE reset %d\n", __FUNCTION__, _ret);	\
+	}								\
+    }									\
+} while (0)
+
+    
+#define CCE_PACKET0( reg, n )						\
+	(R128_CCE_PACKET0 | ((n) << 16) | ((reg) >> 2))
+#define CCE_PACKET1( reg0, reg1 )					\
+	(R128_CCE_PACKET1 | (((reg1) >> 2) << 11) | ((reg0) >> 2))
+#define CCE_PACKET2()							\
+	(R128_CCE_PACKET2)
+#define CCE_PACKET3( pkt, n )						\
+	(R128_CCE_PACKET3 | (pkt) | ((n) << 16))
+
+
+#define R128_VERBOSE	0
+
+#define RING_LOCALS	CARD32 *__head; int __count;
+
+#define R128CCE_REFRESH(pScrn, info)					\
+do {									\
+   if ( R128_VERBOSE ) {						\
+         fprintf(stderr, "REFRESH( %d ) in %s\n",	\
+		  !info->CCEInUse , __FUNCTION__ );			\
+   }									\
+   if ( !info->CCEInUse ) {						\
+      R128CCEWaitForIdle(pScrn);       					\
+      BEGIN_RING( 6 );							\
+      OUT_RING_REG( R128_RE_TOP_LEFT,     info->re_top_left );		\
+      OUT_RING_REG( R128_RE_WIDTH_HEIGHT, info->re_width_height );	\
+      OUT_RING_REG( R128_AUX_SC_CNTL,     info->aux_sc_cntl );		\
+      ADVANCE_RING();							\
+      info->CCEInUse = TRUE;						\
+   }									\
+} while (0)
+
+#define BEGIN_RING( n ) do {						\
+   if ( R128_VERBOSE ) {						\
+         fprintf(stderr,				\
+		  "BEGIN_RING( %d ) in %s\n", n, __FUNCTION__ );	\
+   }									\
+   if ( !info->indirectBuffer ) {					\
+      info->indirectBuffer = R128CCEGetBuffer( pScrn );			\
+      info->indirectStart = 0;						\
+   } else if ( (info->indirectBuffer->used + 4*(n)) >			\
+                info->indirectBuffer->total ) {				\
+      R128CCEFlushIndirect( pScrn, 1 );					\
+   }									\
+   __head = (pointer)((char *)info->indirectBuffer->address +		\
+		       info->indirectBuffer->used);			\
+   __count = 0;								\
+} while (0)
+
+#define ADVANCE_RING() do {						\
+   if ( R128_VERBOSE ) {						\
+         fprintf(stderr,				\
+		  "ADVANCE_RING() used: %d+%d=%d/%d\n",			\
+		  info->indirectBuffer->used - info->indirectStart,	\
+		  __count * sizeof(CARD32),				\
+		  info->indirectBuffer->used - info->indirectStart +	\
+		  __count * sizeof(CARD32),				\
+		  info->indirectBuffer->total - info->indirectStart );	\
+   }									\
+   info->indirectBuffer->used += __count * (int)sizeof(CARD32);		\
+} while (0)
+
+#define OUT_RING( x ) do {						\
+   if ( R128_VERBOSE ) {						\
+         fprintf(stderr,				\
+		  "   OUT_RING( 0x%08x )\n", (unsigned int)(x) );	\
+   }									\
+   MMIO_OUT32(&__head[__count++], 0, (x));				\
+} while (0)
+
+#define OUT_RING_REG( reg, val )					\
+do {									\
+   OUT_RING( CCE_PACKET0( reg, 0 ) );					\
+   OUT_RING( val );							\
+} while (0)
+
+#define FLUSH_RING()							\
+do {									\
+   if ( R128_VERBOSE )							\
+         fprintf(stderr,				\
+		  "FLUSH_RING in %s\n", __FUNCTION__ );			\
+   if ( info->indirectBuffer ) {					\
+      R128CCEFlushIndirect( pScrn, 0 );					\
+   }									\
+} while (0)
+
+    
+#endif
diff --git a/src/mesa/drivers/dri/r128/server/r128_common.h b/src/mesa/drivers/dri/r128/server/r128_common.h
new file mode 100644
index 00000000000..fa81360ea58
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/server/r128_common.h
@@ -0,0 +1,169 @@
+/* r128_common.h -- common header definitions for R128 2D/3D/DRM suite
+ * Created: Sun Apr  9 18:16:28 2000 by [email protected]
+ *
+ * Copyright 1999, 2000 Precision Insight, Inc., Cedar Park, Texas.
+ * 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
+ * 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
+ * PRECISION INSIGHT 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.
+ *
+ * Author:
+ *   Gareth Hughes <[email protected]>
+ *   Kevin E. Martin <[email protected]>
+ *
+ * Converted to common header format:
+ *   Jens Owen <[email protected]>
+ *
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/xf86drmR128.h,v 3.11 2001/04/16 15:02:13 tsi Exp $
+ *
+ */
+
+#ifndef _R128_COMMON_H_
+#define _R128_COMMON_H_
+
+/*
+ * WARNING: If you change any of these defines, make sure to change
+ * the kernel include file as well (r128_drm.h)
+ */
+
+/* Driver specific DRM command indices
+ * NOTE: these are not OS specific, but they are driver specific
+ */
+#define DRM_R128_INIT           0x00
+#define DRM_R128_CCE_START      0x01
+#define DRM_R128_CCE_STOP       0x02
+#define DRM_R128_CCE_RESET      0x03
+#define DRM_R128_CCE_IDLE       0x04
+#define DRM_R128_UNDEFINED1     0x05
+#define DRM_R128_RESET          0x06
+#define DRM_R128_SWAP           0x07
+#define DRM_R128_CLEAR          0x08
+#define DRM_R128_VERTEX         0x09
+#define DRM_R128_INDICES        0x0a
+#define DRM_R128_BLIT           0x0b
+#define DRM_R128_DEPTH          0x0c
+#define DRM_R128_STIPPLE        0x0d
+#define DRM_R128_UNDEFINED2     0x0e
+#define DRM_R128_INDIRECT       0x0f
+#define DRM_R128_FULLSCREEN     0x10
+#define DRM_R128_CLEAR2         0x11
+#define DRM_R128_GETPARAM       0x12
+#define DRM_R128_FLIP           0x13
+
+#define DRM_R128_FRONT_BUFFER	0x1
+#define DRM_R128_BACK_BUFFER	0x2
+#define DRM_R128_DEPTH_BUFFER	0x4
+
+typedef struct {
+   enum {
+      DRM_R128_INIT_CCE    = 0x01,
+      DRM_R128_CLEANUP_CCE = 0x02
+   } func;
+   unsigned long sarea_priv_offset;
+   int is_pci;
+   int cce_mode;
+   int cce_secure;		/* FIXME: Deprecated, we should remove this */
+   int ring_size;
+   int usec_timeout;
+
+   unsigned int fb_bpp;
+   unsigned int front_offset, front_pitch;
+   unsigned int back_offset, back_pitch;
+   unsigned int depth_bpp;
+   unsigned int depth_offset, depth_pitch;
+   unsigned int span_offset;
+
+   unsigned long fb_offset;
+   unsigned long mmio_offset;
+   unsigned long ring_offset;
+   unsigned long ring_rptr_offset;
+   unsigned long buffers_offset;
+   unsigned long agp_textures_offset;
+} drmR128Init;
+
+typedef struct {
+   int flush;
+   int idle;
+} drmR128CCEStop;
+
+typedef struct {
+   int idx;
+   int start;
+   int end;
+   int discard;
+} drmR128Indirect;
+
+typedef struct {
+   int idx;
+   int pitch;
+   int offset;
+   int format;
+   unsigned short x, y;
+   unsigned short width, height;
+} drmR128Blit;
+
+typedef struct {
+   enum {
+      DRM_R128_WRITE_SPAN         = 0x01,
+      DRM_R128_WRITE_PIXELS       = 0x02,
+      DRM_R128_READ_SPAN          = 0x03,
+      DRM_R128_READ_PIXELS        = 0x04
+   } func;
+   int n;
+   int *x;
+   int *y;
+   unsigned int *buffer;
+   unsigned char *mask;
+} drmR128Depth;
+
+typedef struct {
+   int prim;
+   int idx;                        /* Index of vertex buffer */
+   int count;                      /* Number of vertices in buffer */
+   int discard;                    /* Client finished with buffer? */
+} drmR128Vertex;
+
+typedef struct {
+   unsigned int *mask;
+} drmR128Stipple;
+
+typedef struct {
+   unsigned int flags;
+   unsigned int clear_color;
+   unsigned int clear_depth;
+   unsigned int color_mask;
+   unsigned int depth_mask;
+} drmR128Clear;
+
+typedef struct {
+   enum {
+      DRM_R128_INIT_FULLSCREEN    = 0x01,
+      DRM_R128_CLEANUP_FULLSCREEN = 0x02
+   } func;
+} drmR128Fullscreen;
+
+typedef struct drm_r128_getparam {
+	int param;
+	void *value;
+} drmR128GetParam;
+
+#define R128_PARAM_IRQ_NR            1
+
+#endif
diff --git a/src/mesa/drivers/dri/r128/server/r128_dri.c b/src/mesa/drivers/dri/r128/server/r128_dri.c
new file mode 100644
index 00000000000..aa8f49f9c7e
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/server/r128_dri.c
@@ -0,0 +1,1165 @@
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/ati/r128_dri.c,v 1.28 2003/02/07 20:41:14 martin Exp $ */
+/*
+ * Copyright 1999, 2000 ATI Technologies Inc., Markham, Ontario,
+ *                      Precision Insight, Inc., Cedar Park, Texas, and
+ *                      VA Linux Systems Inc., Fremont, California.
+ *
+ * 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, 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
+ * NON-INFRINGEMENT.  IN NO EVENT SHALL ATI, PRECISION INSIGHT, VA LINUX
+ * SYSTEMS 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:
+ *   Kevin E. Martin <[email protected]>
+ *   Rickard E. Faith <[email protected]>
+ *   Daryll Strauss <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ *
+ */
+
+#include "sarea.h"
+#include <stdio.h>
+#include <stdlib.h>
+#include <string.h>
+#include <errno.h>
+#include <unistd.h>
+#include <linux/pci_ids.h>
+
+#include "driver.h"
+#include "drm.h"
+
+#include "sarea.h"
+#include "r128.h"
+#include "r128_dri.h"
+#include "r128_macros.h"
+#include "r128_reg.h"
+#include "r128_sarea.h"
+#include "r128_version.h"
+
+
+/* ?? HACK - for now, put this here... */
+/* ?? Alpha - this may need to be a variable to handle UP1x00 vs TITAN */
+#if defined(__alpha__)
+# define DRM_PAGE_SIZE 8192
+#elif defined(__ia64__)
+# define DRM_PAGE_SIZE getpagesize()
+#else
+# define DRM_PAGE_SIZE 4096
+#endif
+
+/**
+ * \brief Establish the set of modes available for the display.
+ *
+ * \param ctx display handle.
+ * \param numModes will receive the number of supported modes.
+ * \param modes will point to the list of supported modes.
+ *
+ * \return one on success, or zero on failure.
+ * 
+ * Allocates a single visual and fills it with information according to the
+ * display bit depth. Supports only 16 and 32 bpp bit depths, aborting
+ * otherwise.
+ */
+const __GLcontextModes __glModes[] = {
+    
+    /* 32 bit, RGBA Depth=24 Stencil=8 */
+    {.rgbMode = GL_TRUE, .colorIndexMode = GL_FALSE, .doubleBufferMode = GL_TRUE, .stereoMode = GL_FALSE,
+     .haveAccumBuffer = GL_FALSE, .haveDepthBuffer = GL_TRUE, .haveStencilBuffer = GL_TRUE,
+     .redBits = 8, .greenBits = 8, .blueBits = 8, .alphaBits = 8,
+     .redMask = 0xff0000, .greenMask = 0xff00, .blueMask = 0xff, .alphaMask = 0xff000000,
+     .rgbBits = 32, .indexBits = 0,
+     .accumRedBits = 0, .accumGreenBits = 0, .accumBlueBits = 0, .accumAlphaBits = 0,
+     .depthBits = 24, .stencilBits = 8,
+     .numAuxBuffers= 0, .level = 0, .pixmapMode = GL_FALSE, },
+
+    /* 16 bit, RGB Depth=16 */
+    {.rgbMode = GL_TRUE, .colorIndexMode = GL_FALSE, .doubleBufferMode = GL_TRUE, .stereoMode = GL_FALSE,
+     .haveAccumBuffer = GL_FALSE, .haveDepthBuffer = GL_TRUE, .haveStencilBuffer = GL_FALSE,
+     .redBits = 5, .greenBits = 6, .blueBits = 5, .alphaBits = 0,
+     .redMask = 0xf800, .greenMask = 0x07e0, .blueMask = 0x001f, .alphaMask = 0x0,
+     .rgbBits = 16, .indexBits = 0,
+     .accumRedBits = 0, .accumGreenBits = 0, .accumBlueBits = 0, .accumAlphaBits = 0,
+     .depthBits = 16, .stencilBits = 0,
+     .numAuxBuffers= 0, .level = 0, .pixmapMode = GL_FALSE, },
+};
+static int R128InitContextModes( const DRIDriverContext *ctx,
+				   int *numModes, const __GLcontextModes **modes)
+{
+   *numModes = sizeof(__glModes)/sizeof(__GLcontextModes *);
+   *modes = &__glModes[0];
+   return 1;
+}
+
+/* Compute log base 2 of val. */
+static int R128MinBits(int val)
+{
+    int bits;
+
+    if (!val) return 1;
+    for (bits = 0; val; val >>= 1, ++bits);
+    return bits;
+}
+
+/* Initialize the AGP state.  Request memory for use in AGP space, and
+   initialize the Rage 128 registers to point to that memory. */
+static GLboolean R128DRIAgpInit(const DRIDriverContext *ctx)
+{
+    unsigned char *R128MMIO = ctx->MMIOAddress;
+    R128InfoPtr info = ctx->driverPrivate;
+    unsigned long mode;
+    unsigned int  vendor, device;
+    int           ret;
+    unsigned long cntl, chunk;
+    int           s, l;
+    int           flags;
+    unsigned long agpBase;
+
+    if (drmAgpAcquire(ctx->drmFD) < 0) {
+	fprintf(stderr, "[agp] AGP not available\n");
+	return GL_FALSE;
+    }
+
+				/* Modify the mode if the default mode is
+				   not appropriate for this particular
+				   combination of graphics card and AGP
+				   chipset. */
+
+    mode   = drmAgpGetMode(ctx->drmFD);        /* Default mode */
+    vendor = drmAgpVendorId(ctx->drmFD);
+    device = drmAgpDeviceId(ctx->drmFD);
+
+    mode &= ~R128_AGP_MODE_MASK;
+    switch (info->agpMode) {
+    case 4:          mode |= R128_AGP_4X_MODE;
+    case 2:          mode |= R128_AGP_2X_MODE;
+    case 1: default: mode |= R128_AGP_1X_MODE;
+    }
+
+    fprintf(stderr,
+	       "[agp] Mode 0x%08lx [AGP 0x%04x/0x%04x; Card 0x%04x/0x%04x]\n",
+	       mode, vendor, device,
+	       0x1002,
+	       info->Chipset);
+
+    if (drmAgpEnable(ctx->drmFD, mode) < 0) {
+	fprintf(stderr, "[agp] AGP not enabled\n");
+	drmAgpRelease(ctx->drmFD);
+	return GL_FALSE;
+    }
+
+    info->agpOffset = 0;
+
+    if ((ret = drmAgpAlloc(ctx->drmFD, info->agpSize*1024*1024, 0, NULL,
+			   &info->agpMemHandle)) < 0) {
+	fprintf(stderr, "[agp] Out of memory (%d)\n", ret);
+	drmAgpRelease(ctx->drmFD);
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[agp] %d kB allocated with handle 0x%08x\n",
+	       info->agpSize*1024, info->agpMemHandle);
+
+    if (drmAgpBind(ctx->drmFD, info->agpMemHandle, info->agpOffset) < 0) {
+	fprintf(stderr, "[agp] Could not bind\n");
+	drmAgpFree(ctx->drmFD, info->agpMemHandle);
+	drmAgpRelease(ctx->drmFD);
+	return GL_FALSE;
+    }
+
+				/* Initialize the CCE ring buffer data */
+    info->ringStart       = info->agpOffset;
+    info->ringMapSize     = info->ringSize*1024*1024 + DRM_PAGE_SIZE;
+    info->ringSizeLog2QW  = R128MinBits(info->ringSize*1024*1024/8) - 1;
+
+    info->ringReadOffset  = info->ringStart + info->ringMapSize;
+    info->ringReadMapSize = DRM_PAGE_SIZE;
+
+				/* Reserve space for vertex/indirect buffers */
+    info->bufStart        = info->ringReadOffset + info->ringReadMapSize;
+    info->bufMapSize      = info->bufSize*1024*1024;
+
+				/* Reserve the rest for AGP textures */
+    info->agpTexStart     = info->bufStart + info->bufMapSize;
+    s = (info->agpSize*1024*1024 - info->agpTexStart);
+    l = R128MinBits((s-1) / R128_NR_TEX_REGIONS);
+    if (l < R128_LOG_TEX_GRANULARITY) l = R128_LOG_TEX_GRANULARITY;
+    info->agpTexMapSize   = (s >> l) << l;
+    info->log2AGPTexGran  = l;
+
+    if (info->CCESecure) flags = DRM_READ_ONLY;
+    else                  flags = 0;
+
+    if (drmAddMap(ctx->drmFD, info->ringStart, info->ringMapSize,
+		  DRM_AGP, flags, &info->ringHandle) < 0) {
+	fprintf(stderr,
+		   "[agp] Could not add ring mapping\n");
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[agp] ring handle = 0x%08lx\n", info->ringHandle);
+
+    if (drmMap(ctx->drmFD, info->ringHandle, info->ringMapSize,
+	       (drmAddressPtr)&info->ring) < 0) {
+	fprintf(stderr, "[agp] Could not map ring\n");
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[agp] Ring mapped at 0x%08lx\n",
+	       (unsigned long)info->ring);
+
+    if (drmAddMap(ctx->drmFD, info->ringReadOffset, info->ringReadMapSize,
+		  DRM_AGP, flags, &info->ringReadPtrHandle) < 0) {
+	fprintf(stderr,
+		   "[agp] Could not add ring read ptr mapping\n");
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[agp] ring read ptr handle = 0x%08lx\n",
+	       info->ringReadPtrHandle);
+
+    if (drmMap(ctx->drmFD, info->ringReadPtrHandle, info->ringReadMapSize,
+	       (drmAddressPtr)&info->ringReadPtr) < 0) {
+	fprintf(stderr,
+		   "[agp] Could not map ring read ptr\n");
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[agp] Ring read ptr mapped at 0x%08lx\n",
+	       (unsigned long)info->ringReadPtr);
+
+    if (drmAddMap(ctx->drmFD, info->bufStart, info->bufMapSize,
+		  DRM_AGP, 0, &info->bufHandle) < 0) {
+	fprintf(stderr,
+		   "[agp] Could not add vertex/indirect buffers mapping\n");
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[agp] vertex/indirect buffers handle = 0x%08lx\n",
+	       info->bufHandle);
+
+    if (drmMap(ctx->drmFD, info->bufHandle, info->bufMapSize,
+	       (drmAddressPtr)&info->buf) < 0) {
+	fprintf(stderr,
+		   "[agp] Could not map vertex/indirect buffers\n");
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[agp] Vertex/indirect buffers mapped at 0x%08lx\n",
+	       (unsigned long)info->buf);
+
+    if (drmAddMap(ctx->drmFD, info->agpTexStart, info->agpTexMapSize,
+		  DRM_AGP, 0, &info->agpTexHandle) < 0) {
+	fprintf(stderr,
+		   "[agp] Could not add AGP texture map mapping\n");
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[agp] AGP texture map handle = 0x%08lx\n",
+	       info->agpTexHandle);
+
+    if (drmMap(ctx->drmFD, info->agpTexHandle, info->agpTexMapSize,
+	       (drmAddressPtr)&info->agpTex) < 0) {
+	fprintf(stderr,
+		   "[agp] Could not map AGP texture map\n");
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[agp] AGP Texture map mapped at 0x%08lx\n",
+	       (unsigned long)info->agpTex);
+
+				/* Initialize Rage 128's AGP registers */
+    cntl  = INREG(R128_AGP_CNTL);
+    cntl &= ~R128_AGP_APER_SIZE_MASK;
+    switch (info->agpSize) {
+    case 256: cntl |= R128_AGP_APER_SIZE_256MB; break;
+    case 128: cntl |= R128_AGP_APER_SIZE_128MB; break;
+    case  64: cntl |= R128_AGP_APER_SIZE_64MB;  break;
+    case  32: cntl |= R128_AGP_APER_SIZE_32MB;  break;
+    case  16: cntl |= R128_AGP_APER_SIZE_16MB;  break;
+    case   8: cntl |= R128_AGP_APER_SIZE_8MB;   break;
+    case   4: cntl |= R128_AGP_APER_SIZE_4MB;   break;
+    default:
+	fprintf(stderr,
+		   "[agp] Illegal aperture size %d kB\n",
+		   info->agpSize*1024);
+	return GL_FALSE;
+    }
+    agpBase = drmAgpBase(ctx->drmFD);
+    OUTREG(R128_AGP_BASE, agpBase); 
+    OUTREG(R128_AGP_CNTL, cntl);
+
+				/* Disable Rage 128's PCIGART registers */
+    chunk = INREG(R128_BM_CHUNK_0_VAL);
+    chunk &= ~(R128_BM_PTR_FORCE_TO_PCI |
+	       R128_BM_PM4_RD_FORCE_TO_PCI |
+	       R128_BM_GLOBAL_FORCE_TO_PCI);
+    OUTREG(R128_BM_CHUNK_0_VAL, chunk);
+
+    OUTREG(R128_PCI_GART_PAGE, 1); /* Ensure AGP GART is used (for now) */
+
+    return GL_TRUE;
+}
+
+static GLboolean R128DRIPciInit(const DRIDriverContext *ctx)
+{
+    R128InfoPtr info = ctx->driverPrivate;
+    unsigned char *R128MMIO = ctx->MMIOAddress;
+    CARD32 chunk;
+    int ret;
+    int flags;
+
+    info->agpOffset = 0;
+
+    ret = drmScatterGatherAlloc(ctx->drmFD, info->agpSize*1024*1024,
+				&info->pciMemHandle);
+    if (ret < 0) {
+	fprintf(stderr, "[pci] Out of memory (%d)\n", ret);
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[pci] %d kB allocated with handle 0x%08x\n",
+	       info->agpSize*1024, info->pciMemHandle);
+
+				/* Initialize the CCE ring buffer data */
+    info->ringStart       = info->agpOffset;
+    info->ringMapSize     = info->ringSize*1024*1024 + DRM_PAGE_SIZE;
+    info->ringSizeLog2QW  = R128MinBits(info->ringSize*1024*1024/8) - 1;
+
+    info->ringReadOffset  = info->ringStart + info->ringMapSize;
+    info->ringReadMapSize = DRM_PAGE_SIZE;
+
+				/* Reserve space for vertex/indirect buffers */
+    info->bufStart        = info->ringReadOffset + info->ringReadMapSize;
+    info->bufMapSize      = info->bufSize*1024*1024;
+
+    flags = DRM_READ_ONLY | DRM_LOCKED | DRM_KERNEL;
+
+    if (drmAddMap(ctx->drmFD, info->ringStart, info->ringMapSize,
+		  DRM_SCATTER_GATHER, flags, &info->ringHandle) < 0) {
+	fprintf(stderr,
+		   "[pci] Could not add ring mapping\n");
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[pci] ring handle = 0x%08lx\n", info->ringHandle);
+
+    if (drmMap(ctx->drmFD, info->ringHandle, info->ringMapSize,
+	       (drmAddressPtr)&info->ring) < 0) {
+	fprintf(stderr, "[pci] Could not map ring\n");
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[pci] Ring mapped at 0x%08lx\n",
+	       (unsigned long)info->ring);
+    fprintf(stderr,
+	       "[pci] Ring contents 0x%08lx\n",
+	       *(unsigned long *)info->ring);
+
+    if (drmAddMap(ctx->drmFD, info->ringReadOffset, info->ringReadMapSize,
+		  DRM_SCATTER_GATHER, flags, &info->ringReadPtrHandle) < 0) {
+	fprintf(stderr,
+		   "[pci] Could not add ring read ptr mapping\n");
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[pci] ring read ptr handle = 0x%08lx\n",
+	       info->ringReadPtrHandle);
+
+    if (drmMap(ctx->drmFD, info->ringReadPtrHandle, info->ringReadMapSize,
+	       (drmAddressPtr)&info->ringReadPtr) < 0) {
+	fprintf(stderr,
+		   "[pci] Could not map ring read ptr\n");
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[pci] Ring read ptr mapped at 0x%08lx\n",
+	       (unsigned long)info->ringReadPtr);
+    fprintf(stderr,
+	       "[pci] Ring read ptr contents 0x%08lx\n",
+	       *(unsigned long *)info->ringReadPtr);
+
+    if (drmAddMap(ctx->drmFD, info->bufStart, info->bufMapSize,
+		  DRM_SCATTER_GATHER, 0, &info->bufHandle) < 0) {
+	fprintf(stderr,
+		   "[pci] Could not add vertex/indirect buffers mapping\n");
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[pci] vertex/indirect buffers handle = 0x%08lx\n",
+	       info->bufHandle);
+
+    if (drmMap(ctx->drmFD, info->bufHandle, info->bufMapSize,
+	       (drmAddressPtr)&info->buf) < 0) {
+	fprintf(stderr,
+		   "[pci] Could not map vertex/indirect buffers\n");
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[pci] Vertex/indirect buffers mapped at 0x%08lx\n",
+	       (unsigned long)info->buf);
+    fprintf(stderr,
+	       "[pci] Vertex/indirect buffers contents 0x%08lx\n",
+	       *(unsigned long *)info->buf);
+
+    if (!info->IsPCI) {
+	/* This is really an AGP card, force PCI GART mode */
+        chunk = INREG(R128_BM_CHUNK_0_VAL);
+        chunk |= (R128_BM_PTR_FORCE_TO_PCI |
+		  R128_BM_PM4_RD_FORCE_TO_PCI |
+		  R128_BM_GLOBAL_FORCE_TO_PCI);
+        OUTREG(R128_BM_CHUNK_0_VAL, chunk);
+        OUTREG(R128_PCI_GART_PAGE, 0); /* Ensure PCI GART is used */
+    }
+
+    return GL_TRUE;
+}
+
+/* Add a map for the MMIO registers that will be accessed by any
+   DRI-based clients. */
+static GLboolean R128DRIMapInit(const DRIDriverContext *ctx)
+{
+    R128InfoPtr info = ctx->driverPrivate;
+    int flags;
+
+    if (info->CCESecure) flags = DRM_READ_ONLY;
+    else                 flags = 0;
+
+				/* Map registers */
+    if (drmAddMap(ctx->drmFD, ctx->MMIOStart, ctx->MMIOSize,
+		  DRM_REGISTERS, flags, &info->registerHandle) < 0) {
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[drm] register handle = 0x%08lx\n", info->registerHandle);
+
+    return GL_TRUE;
+}
+
+/* Initialize the kernel data structures. */
+static int R128DRIKernelInit(const DRIDriverContext *ctx)
+{
+    R128InfoPtr info = ctx->driverPrivate;
+    drmR128Init drmInfo;
+
+    memset( &drmInfo, 0, sizeof(drmR128Init) );
+
+    drmInfo.func                = DRM_R128_INIT_CCE;
+    drmInfo.sarea_priv_offset   = sizeof(XF86DRISAREARec);
+    drmInfo.is_pci              = info->IsPCI;
+    drmInfo.cce_mode            = info->CCEMode;
+    drmInfo.cce_secure          = info->CCESecure;
+    drmInfo.ring_size           = info->ringSize*1024*1024;
+    drmInfo.usec_timeout        = info->CCEusecTimeout;
+
+    drmInfo.fb_bpp              = ctx->bpp;
+    drmInfo.depth_bpp           = ctx->bpp;
+
+    drmInfo.front_offset        = info->frontOffset;
+    drmInfo.front_pitch         = info->frontPitch;
+
+    drmInfo.back_offset         = info->backOffset;
+    drmInfo.back_pitch          = info->backPitch;
+
+    drmInfo.depth_offset        = info->depthOffset;
+    drmInfo.depth_pitch         = info->depthPitch;
+    drmInfo.span_offset         = info->spanOffset;
+
+    drmInfo.fb_offset           = info->LinearAddr;
+    drmInfo.mmio_offset         = info->registerHandle;
+    drmInfo.ring_offset         = info->ringHandle;
+    drmInfo.ring_rptr_offset    = info->ringReadPtrHandle;
+    drmInfo.buffers_offset      = info->bufHandle;
+    drmInfo.agp_textures_offset = info->agpTexHandle;
+
+    if (drmCommandWrite(ctx->drmFD, DRM_R128_INIT,
+                        &drmInfo, sizeof(drmR128Init)) < 0)
+        return GL_FALSE;
+
+    return GL_TRUE;
+}
+
+/* Add a map for the vertex buffers that will be accessed by any
+   DRI-based clients. */
+static GLboolean R128DRIBufInit(const DRIDriverContext *ctx)
+{
+    R128InfoPtr info = ctx->driverPrivate;
+				/* Initialize vertex buffers */
+    if (info->IsPCI) {
+	info->bufNumBufs = drmAddBufs(ctx->drmFD,
+				      info->bufMapSize / R128_BUFFER_SIZE,
+				      R128_BUFFER_SIZE,
+				      DRM_SG_BUFFER,
+				      info->bufStart);
+    } else {
+	info->bufNumBufs = drmAddBufs(ctx->drmFD,
+				      info->bufMapSize / R128_BUFFER_SIZE,
+				      R128_BUFFER_SIZE,
+				      DRM_AGP_BUFFER,
+				      info->bufStart);
+    }
+    if (info->bufNumBufs <= 0) {
+	fprintf(stderr,
+		   "[drm] Could not create vertex/indirect buffers list\n");
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[drm] Added %d %d byte vertex/indirect buffers\n",
+	       info->bufNumBufs, R128_BUFFER_SIZE);
+
+    if (!(info->buffers = drmMapBufs(ctx->drmFD))) {
+	fprintf(stderr,
+		   "[drm] Failed to map vertex/indirect buffers list\n");
+	return GL_FALSE;
+    }
+    fprintf(stderr,
+	       "[drm] Mapped %d vertex/indirect buffers\n",
+	       info->buffers->count);
+
+    return GL_TRUE;
+}
+
+static void R128DRIIrqInit(const DRIDriverContext *ctx)
+{
+   R128InfoPtr info = ctx->driverPrivate;
+   unsigned char *R128MMIO = ctx->MMIOAddress;
+   
+   if (!info->irq) {
+       info->irq = drmGetInterruptFromBusID(
+	   ctx->drmFD,
+	   ctx->pciBus,
+	   ctx->pciDevice,
+	   ctx->pciFunc);
+
+      if((drmCtlInstHandler(ctx->drmFD, info->irq)) != 0) {
+	 fprintf(stderr,
+		    "[drm] failure adding irq handler, "
+		    "there is a device already using that irq\n"
+		    "[drm] falling back to irq-free operation\n");
+	 info->irq = 0;
+      } else {
+          info->gen_int_cntl = INREG( R128_GEN_INT_CNTL );
+      }
+   }
+
+   if (info->irq)
+      fprintf(stderr,
+		 "[drm] dma control initialized, using IRQ %d\n",
+		 info->irq);
+}
+
+static int R128CCEStop(const DRIDriverContext *ctx)
+{
+    R128InfoPtr info = ctx->driverPrivate;
+    drmR128CCEStop stop;
+    int            ret, i;
+
+    stop.flush = 1;
+    stop.idle  = 1;
+
+    ret = drmCommandWrite( ctx->drmFD, DRM_R128_CCE_STOP,
+                           &stop, sizeof(drmR128CCEStop) );
+
+    if ( ret == 0 ) {
+        return 0;
+    } else if ( errno != EBUSY ) {
+        return -errno;
+    }
+
+    stop.flush = 0;
+
+    i = 0;
+    do {
+        ret = drmCommandWrite( ctx->drmFD, DRM_R128_CCE_STOP,
+                               &stop, sizeof(drmR128CCEStop) );
+    } while ( ret && errno == EBUSY && i++ < R128_IDLE_RETRY );
+
+    if ( ret == 0 ) {
+        return 0;
+    } else if ( errno != EBUSY ) {
+        return -errno;
+    }
+
+    stop.idle = 0;
+
+    if ( drmCommandWrite( ctx->drmFD, DRM_R128_CCE_STOP,
+                          &stop, sizeof(drmR128CCEStop) )) {
+        return -errno;
+    } else {
+        return 0;
+    }
+}
+
+/* Initialize the CCE state, and start the CCE (if used by the X server) */
+static void R128DRICCEInit(const DRIDriverContext *ctx)
+{
+   R128InfoPtr info = ctx->driverPrivate;
+
+				/* Turn on bus mastering */
+    info->BusCntl &= ~R128_BUS_MASTER_DIS;
+
+				/* CCEMode is initialized in r128_driver.c */
+    switch (info->CCEMode) {
+    case R128_PM4_NONPM4:                 info->CCEFifoSize = 0;   break;
+    case R128_PM4_192PIO:                 info->CCEFifoSize = 192; break;
+    case R128_PM4_192BM:                  info->CCEFifoSize = 192; break;
+    case R128_PM4_128PIO_64INDBM:         info->CCEFifoSize = 128; break;
+    case R128_PM4_128BM_64INDBM:          info->CCEFifoSize = 128; break;
+    case R128_PM4_64PIO_128INDBM:         info->CCEFifoSize = 64;  break;
+    case R128_PM4_64BM_128INDBM:          info->CCEFifoSize = 64;  break;
+    case R128_PM4_64PIO_64VCBM_64INDBM:   info->CCEFifoSize = 64;  break;
+    case R128_PM4_64BM_64VCBM_64INDBM:    info->CCEFifoSize = 64;  break;
+    case R128_PM4_64PIO_64VCPIO_64INDPIO: info->CCEFifoSize = 64;  break;
+    }
+
+    /* Make sure the CCE is on for the X server */
+    R128CCE_START(ctx, info);
+}
+
+
+static int R128MemoryInit(const DRIDriverContext *ctx)
+{
+   R128InfoPtr info = ctx->driverPrivate;
+   int        width_bytes = ctx->shared.virtualWidth * ctx->cpp;
+   int        cpp         = ctx->cpp;
+   int        bufferSize  = ((ctx->shared.virtualHeight * width_bytes
+			      + R128_BUFFER_ALIGN)
+			     & ~R128_BUFFER_ALIGN);
+   int        depthSize   = ((((ctx->shared.virtualHeight+15) & ~15) * width_bytes
+			      + R128_BUFFER_ALIGN)
+			     & ~R128_BUFFER_ALIGN);
+   int        l;
+
+   info->frontOffset = 0;
+   info->frontPitch = ctx->shared.virtualWidth;
+
+   fprintf(stderr, 
+	   "Using %d MB AGP aperture\n", info->agpSize);
+   fprintf(stderr, 
+	   "Using %d MB for the ring buffer\n", info->ringSize);
+   fprintf(stderr, 
+	   "Using %d MB for vertex/indirect buffers\n", info->bufSize);
+   fprintf(stderr, 
+	   "Using %d MB for AGP textures\n", info->agpTexSize);
+
+   /* Front, back and depth buffers - everything else texture??
+    */
+   info->textureSize = ctx->shared.fbSize - 2 * bufferSize - depthSize;
+
+   if (info->textureSize < 0) 
+      return 0;
+
+   l = R128MinBits((info->textureSize-1) / R128_NR_TEX_REGIONS);
+   if (l < R128_LOG_TEX_GRANULARITY) l = R128_LOG_TEX_GRANULARITY;
+
+   /* Round the texture size up to the nearest whole number of
+    * texture regions.  Again, be greedy about this, don't
+    * round down.
+    */
+   info->log2TexGran = l;
+   info->textureSize = (info->textureSize >> l) << l;
+
+   /* Set a minimum usable local texture heap size.  This will fit
+    * two 256x256x32bpp textures.
+    */
+   if (info->textureSize < 512 * 1024) {
+      info->textureOffset = 0;
+      info->textureSize = 0;
+   }
+
+   /* Reserve space for textures */
+   info->textureOffset = ((ctx->shared.fbSize - info->textureSize +
+			   R128_BUFFER_ALIGN) &
+			  ~R128_BUFFER_ALIGN);
+
+   /* Reserve space for the shared depth
+    * buffer.
+    */
+   info->depthOffset = ((info->textureOffset - depthSize +
+			 R128_BUFFER_ALIGN) &
+			~R128_BUFFER_ALIGN);
+   info->depthPitch = ctx->shared.virtualWidth;
+
+   info->backOffset = ((info->depthOffset - bufferSize +
+			R128_BUFFER_ALIGN) &
+		       ~R128_BUFFER_ALIGN);
+   info->backPitch = ctx->shared.virtualWidth;
+
+
+   fprintf(stderr, 
+	   "Will use back buffer at offset 0x%x\n",
+	   info->backOffset);
+   fprintf(stderr, 
+	   "Will use depth buffer at offset 0x%x\n",
+	   info->depthOffset);
+   fprintf(stderr, 
+	   "Will use %d kb for textures at offset 0x%x\n",
+	   info->textureSize/1024, info->textureOffset);
+
+   return 1;
+} 
+
+
+/* Initialize the screen-specific data structures for the DRI and the
+   Rage 128.  This is the main entry point to the device-specific
+   initialization code.  It calls device-independent DRI functions to
+   create the DRI data structures and initialize the DRI state. */
+static GLboolean R128DRIScreenInit(DRIDriverContext *ctx)
+{
+    R128InfoPtr info = ctx->driverPrivate;
+    R128DRIPtr    pR128DRI;
+    int           err, major, minor, patch;
+    drmVersionPtr version;
+
+    switch (ctx->bpp) {
+    case 8:
+	/* These modes are not supported (yet). */
+    case 15:
+    case 24:
+	fprintf(stderr,
+		   "[dri] R128DRIScreenInit failed (depth %d not supported).  "
+		   "[dri] Disabling DRI.\n", ctx->bpp);
+	return GL_FALSE;
+
+	/* Only 16 and 32 color depths are supports currently. */
+    case 16:
+    case 32:
+	break;
+    }
+
+    info->registerSize = ctx->MMIOSize;
+    ctx->shared.SAREASize = DRM_PAGE_SIZE;
+
+    /* Note that drmOpen will try to load the kernel module, if needed. */
+    ctx->drmFD = drmOpen("r128", NULL );
+    if (ctx->drmFD < 0) {
+	fprintf(stderr, "[drm] drmOpen failed\n");
+	return 0;
+    }
+    
+    /* Check the r128 DRM version */
+    version = drmGetVersion(ctx->drmFD);
+    if (version) {
+	if (version->version_major != 2 ||
+	    version->version_minor < 2) {
+	    /* incompatible drm version */
+	    fprintf(stderr,
+		"[dri] R128DRIScreenInit failed because of a version mismatch.\n"
+		"[dri] r128.o kernel module version is %d.%d.%d but version 2.2 or greater is needed.\n"
+		"[dri] Disabling the DRI.\n",
+		version->version_major,
+		version->version_minor,
+		version->version_patchlevel);
+	    drmFreeVersion(version);
+	    return GL_FALSE;
+	}
+	info->drmMinor = version->version_minor;
+	drmFreeVersion(version);
+    }
+    
+    if ((err = drmSetBusid(ctx->drmFD, ctx->pciBusID)) < 0) {
+	fprintf(stderr, "[drm] drmSetBusid failed (%d, %s), %s\n",
+		ctx->drmFD, ctx->pciBusID, strerror(-err));
+	return 0;
+    }
+    
+   if (drmAddMap( ctx->drmFD,
+		  0,
+		  ctx->shared.SAREASize,
+		  DRM_SHM,
+		  DRM_CONTAINS_LOCK,
+		  &ctx->shared.hSAREA) < 0)
+   {
+      fprintf(stderr, "[drm] drmAddMap failed\n");
+      return 0;
+   }
+   fprintf(stderr, "[drm] added %d byte SAREA at 0x%08lx\n",
+	   ctx->shared.SAREASize, ctx->shared.hSAREA);
+
+   if (drmMap( ctx->drmFD,
+	       ctx->shared.hSAREA,
+	       ctx->shared.SAREASize,
+	       (drmAddressPtr)(&ctx->pSAREA)) < 0)
+   {
+      fprintf(stderr, "[drm] drmMap failed\n");
+      return 0;
+   }
+   memset(ctx->pSAREA, 0, ctx->shared.SAREASize);
+   fprintf(stderr, "[drm] mapped SAREA 0x%08lx to %p, size %d\n",
+	   ctx->shared.hSAREA, ctx->pSAREA, ctx->shared.SAREASize);
+   
+   /* Need to AddMap the framebuffer and mmio regions here:
+    */
+   if (drmAddMap( ctx->drmFD,
+		  (drmHandle)ctx->FBStart,
+		  ctx->FBSize,
+		  DRM_FRAME_BUFFER,
+		  0,
+		  &ctx->shared.hFrameBuffer) < 0)
+   {
+      fprintf(stderr, "[drm] drmAddMap framebuffer failed\n");
+      return 0;
+   }
+
+   fprintf(stderr, "[drm] framebuffer handle = 0x%08lx\n",
+	   ctx->shared.hFrameBuffer);
+
+   if (!R128MemoryInit(ctx))
+	return GL_FALSE;
+   
+				/* Initialize AGP */
+    if (!info->IsPCI && !R128DRIAgpInit(ctx)) {
+	info->IsPCI = GL_TRUE;
+	fprintf(stderr,
+		   "[agp] AGP failed to initialize -- falling back to PCI mode.\n");
+	fprintf(stderr,
+		   "[agp] Make sure you have the agpgart kernel module loaded.\n");
+    }
+
+				/* Initialize PCIGART */
+    if (info->IsPCI && !R128DRIPciInit(ctx)) {
+	return GL_FALSE;
+    }
+
+				/* DRIScreenInit doesn't add all the
+				   common mappings.  Add additional
+				   mappings here. */
+    if (!R128DRIMapInit(ctx)) {
+	return GL_FALSE;
+    }
+
+   /* Create a 'server' context so we can grab the lock for
+    * initialization ioctls.
+    */
+   if ((err = drmCreateContext(ctx->drmFD, &ctx->serverContext)) != 0) {
+      fprintf(stderr, "%s: drmCreateContext failed %d\n", __FUNCTION__, err);
+      return 0;
+   }
+
+   DRM_LOCK(ctx->drmFD, ctx->pSAREA, ctx->serverContext, 0); 
+
+    /* Initialize the kernel data structures */
+    if (!R128DRIKernelInit(ctx)) {
+	return GL_FALSE;
+    }
+
+    /* Initialize the vertex buffers list */
+    if (!R128DRIBufInit(ctx)) {
+	return GL_FALSE;
+    }
+
+    /* Initialize IRQ */
+    R128DRIIrqInit(ctx);
+
+    /* Initialize and start the CCE if required */
+    R128DRICCEInit(ctx);
+
+   /* Quick hack to clear the front & back buffers.  Could also use
+    * the clear ioctl to do this, but would need to setup hw state
+    * first.
+    */
+   memset((char *)ctx->FBAddress + info->frontOffset,
+	  0,
+	  info->frontPitch * ctx->cpp * ctx->shared.virtualHeight );
+
+   memset((char *)ctx->FBAddress + info->backOffset,
+	  0,
+	  info->backPitch * ctx->cpp * ctx->shared.virtualHeight );
+    
+    R128SAREAPrivPtr pSAREAPriv;
+    pSAREAPriv = (R128SAREAPrivPtr)(((char*)ctx->pSAREA) + 
+					sizeof(XF86DRISAREARec));
+    memset(pSAREAPriv, 0, sizeof(*pSAREAPriv));
+
+   /* This is the struct passed to radeon_dri.so for its initialization */
+   ctx->driverClientMsg = malloc(sizeof(R128DRIRec));
+   ctx->driverClientMsgSize = sizeof(R128DRIRec);
+   
+    pR128DRI                    = (R128DRIPtr)ctx->driverClientMsg;
+    pR128DRI->deviceID          = info->Chipset;
+    pR128DRI->width             = ctx->shared.virtualWidth;
+    pR128DRI->height            = ctx->shared.virtualHeight;
+    pR128DRI->depth             = ctx->bpp;
+    pR128DRI->bpp               = ctx->bpp;
+
+    pR128DRI->IsPCI             = info->IsPCI;
+    pR128DRI->AGPMode           = info->agpMode;
+
+    pR128DRI->frontOffset       = info->frontOffset;
+    pR128DRI->frontPitch        = info->frontPitch;
+    pR128DRI->backOffset        = info->backOffset;
+    pR128DRI->backPitch         = info->backPitch;
+    pR128DRI->depthOffset       = info->depthOffset;
+    pR128DRI->depthPitch        = info->depthPitch;
+    pR128DRI->spanOffset        = info->spanOffset;
+    pR128DRI->textureOffset     = info->textureOffset;
+    pR128DRI->textureSize       = info->textureSize;
+    pR128DRI->log2TexGran       = info->log2TexGran;
+
+    pR128DRI->registerHandle    = info->registerHandle;
+    pR128DRI->registerSize      = info->registerSize;
+
+    pR128DRI->agpTexHandle      = info->agpTexHandle;
+    pR128DRI->agpTexMapSize     = info->agpTexMapSize;
+    pR128DRI->log2AGPTexGran    = info->log2AGPTexGran;
+    pR128DRI->agpTexOffset      = info->agpTexStart;
+    pR128DRI->sarea_priv_offset = sizeof(XF86DRISAREARec);
+
+    return GL_TRUE;
+}
+
+/* The screen is being closed, so clean up any state and free any
+   resources used by the DRI. */
+void R128DRICloseScreen(const DRIDriverContext *ctx)
+{
+    R128InfoPtr info = ctx->driverPrivate;
+    drmR128Init drmInfo;
+
+    /* Stop the CCE if it is still in use */
+    R128CCE_STOP(ctx, info);
+
+    if (info->irq) {
+	drmCtlUninstHandler(ctx->drmFD);
+	info->irq = 0;
+    }
+
+    /* De-allocate vertex buffers */
+    if (info->buffers) {
+	drmUnmapBufs(info->buffers);
+	info->buffers = NULL;
+    }
+
+    /* De-allocate all kernel resources */
+    memset(&drmInfo, 0, sizeof(drmR128Init));
+    drmInfo.func = DRM_R128_CLEANUP_CCE;
+    drmCommandWrite(ctx->drmFD, DRM_R128_INIT,
+                    &drmInfo, sizeof(drmR128Init));
+
+    /* De-allocate all AGP resources */
+    if (info->agpTex) {
+	drmUnmap(info->agpTex, info->agpTexMapSize);
+	info->agpTex = NULL;
+    }
+    if (info->buf) {
+	drmUnmap(info->buf, info->bufMapSize);
+	info->buf = NULL;
+    }
+    if (info->ringReadPtr) {
+	drmUnmap(info->ringReadPtr, info->ringReadMapSize);
+	info->ringReadPtr = NULL;
+    }
+    if (info->ring) {
+	drmUnmap(info->ring, info->ringMapSize);
+	info->ring = NULL;
+    }
+    if (info->agpMemHandle != DRM_AGP_NO_HANDLE) {
+	drmAgpUnbind(ctx->drmFD, info->agpMemHandle);
+	drmAgpFree(ctx->drmFD, info->agpMemHandle);
+	info->agpMemHandle = 0;
+	drmAgpRelease(ctx->drmFD);
+    }
+    if (info->pciMemHandle) {
+	drmScatterGatherFree(ctx->drmFD, info->pciMemHandle);
+	info->pciMemHandle = 0;
+    }
+}
+
+static GLboolean R128PreInitDRI(const DRIDriverContext *ctx)
+{
+    R128InfoPtr info = ctx->driverPrivate;
+
+    /*info->CCEMode = R128_DEFAULT_CCE_PIO_MODE;*/
+    info->CCEMode = R128_DEFAULT_CCE_BM_MODE;
+    info->CCESecure = GL_TRUE;
+
+    info->agpMode        = R128_DEFAULT_AGP_MODE;
+    info->agpSize        = R128_DEFAULT_AGP_SIZE;
+    info->ringSize       = R128_DEFAULT_RING_SIZE;
+    info->bufSize        = R128_DEFAULT_BUFFER_SIZE;
+    info->agpTexSize     = R128_DEFAULT_AGP_TEX_SIZE;
+
+    info->CCEusecTimeout = R128_DEFAULT_CCE_TIMEOUT;
+
+    return GL_TRUE;
+}
+
+/**
+ * \brief Initialize the framebuffer device mode
+ *
+ * \param ctx display handle.
+ *
+ * \return one on success, or zero on failure.
+ *
+ * Fills in \p info with some default values and some information from \p ctx
+ * and then calls R128ScreenInit() for the screen initialization.
+ * 
+ * Before exiting clears the framebuffer memory accessing it directly.
+ */
+static int R128InitFBDev( DRIDriverContext *ctx )
+{
+   R128InfoPtr info = calloc(1, sizeof(*info));
+
+   {
+      int  dummy = ctx->shared.virtualWidth;
+
+      switch (ctx->bpp / 8) {
+      case 1: dummy = (ctx->shared.virtualWidth + 127) & ~127; break;
+      case 2: dummy = (ctx->shared.virtualWidth +  31) &  ~31; break;
+      case 3:
+      case 4: dummy = (ctx->shared.virtualWidth +  15) &  ~15; break;
+      }
+
+      ctx->shared.virtualWidth = dummy;
+   }
+
+   ctx->driverPrivate = (void *)info;
+   
+   info->Chipset = ctx->chipset;
+
+   switch (info->Chipset) {
+   case PCI_DEVICE_ID_ATI_RAGE128_LE:
+   case PCI_DEVICE_ID_ATI_RAGE128_RE:
+   case PCI_DEVICE_ID_ATI_RAGE128_RK:
+   case PCI_DEVICE_ID_ATI_RAGE128_PD:
+   case PCI_DEVICE_ID_ATI_RAGE128_PP:
+   case PCI_DEVICE_ID_ATI_RAGE128_PR:
+       /* This is a PCI card */
+       info->IsPCI = GL_TRUE;
+       break;
+   default:
+       /* This is an AGP card */
+       info->IsPCI = GL_FALSE;
+       break;
+   }
+
+   info->frontPitch = ctx->shared.virtualWidth;
+   info->LinearAddr = ctx->FBStart & 0xfc000000;
+   
+   if (!R128PreInitDRI(ctx))
+       return 0;
+
+   if (!R128DRIScreenInit(ctx))
+      return 0;
+
+   return 1;
+}
+
+
+/**
+ * \brief The screen is being closed, so clean up any state and free any
+ * resources used by the DRI.
+ *
+ * \param ctx display handle.
+ *
+ * Unmaps the SAREA, closes the DRM device file descriptor and frees the driver
+ * private data.
+ */
+static void R128HaltFBDev( DRIDriverContext *ctx )
+{
+    drmUnmap( ctx->pSAREA, ctx->shared.SAREASize );
+    drmClose(ctx->drmFD);
+
+    if (ctx->driverPrivate) {
+       free(ctx->driverPrivate);
+       ctx->driverPrivate = 0;
+    }
+}
+
+
+/**
+ * \brief Validate the fbdev mode.
+ * 
+ * \param ctx display handle.
+ *
+ * \return one on success, or zero on failure.
+ *
+ * Saves some registers and returns 1.
+ *
+ * \sa R128PostValidateMode().
+ */
+static int R128ValidateMode( const DRIDriverContext *ctx )
+{
+   return 1;
+}
+
+
+/**
+ * \brief Examine mode returned by fbdev.
+ * 
+ * \param ctx display handle.
+ *
+ * \return one on success, or zero on failure.
+ *
+ * Restores registers that fbdev has clobbered and returns 1.
+ *
+ * \sa R128ValidateMode().
+ */
+static int R128PostValidateMode( const DRIDriverContext *ctx )
+{
+   return 1;
+}
+
+
+/**
+ * \brief Shutdown the drawing engine.
+ *
+ * \param ctx display handle
+ *
+ * Turns off the command processor engine & restores the graphics card
+ * to a state that fbdev understands.
+ */
+static int R128EngineShutdown( const DRIDriverContext *ctx )
+{
+    return 1;
+}
+
+/**
+ * \brief Restore the drawing engine.
+ *
+ * \param ctx display handle
+ *
+ * Resets the graphics card and sets initial values for several registers of
+ * the card's drawing engine.
+ *
+ * Turns on the R128 command processor engine (i.e., the ringbuffer).
+ */
+static int R128EngineRestore( const DRIDriverContext *ctx )
+{
+   return 1;
+}
+
+
+/**
+ * \brief Exported driver interface for Mini GLX.
+ *
+ * \sa DRIDriverRec.
+ */
+const struct DRIDriverRec __driDriver = {
+   R128InitContextModes,
+   R128ValidateMode,
+   R128PostValidateMode,
+   R128InitFBDev,
+   R128HaltFBDev,
+   R128EngineShutdown,
+   R128EngineRestore,  
+   0,
+};
diff --git a/src/mesa/drivers/dri/r128/server/r128_dri.h b/src/mesa/drivers/dri/r128/server/r128_dri.h
new file mode 100644
index 00000000000..80f94ded68e
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/server/r128_dri.h
@@ -0,0 +1,103 @@
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/ati/r128_dri.h,v 1.7 2002/10/30 12:52:12 alanh Exp $ */
+/*
+ * Copyright 1999, 2000 ATI Technologies Inc., Markham, Ontario,
+ *                      Precision Insight, Inc., Cedar Park, Texas, and
+ *                      VA Linux Systems Inc., Fremont, California.
+ *
+ * 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, 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
+ * NON-INFRINGEMENT.  IN NO EVENT SHALL ATI, PRECISION INSIGHT, VA LINUX
+ * SYSTEMS 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:
+ *   Kevin E. Martin <[email protected]>
+ *   Rickard E. Faith <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ *
+ */
+
+#ifndef _R128_DRI_
+#define _R128_DRI_
+
+#include "xf86drm.h"
+#include "r128_common.h"
+
+/* DRI Driver defaults */
+#define R128_DEFAULT_CCE_PIO_MODE R128_PM4_64PIO_64VCBM_64INDBM
+#define R128_DEFAULT_CCE_BM_MODE  R128_PM4_64BM_64VCBM_64INDBM
+#define R128_DEFAULT_AGP_MODE     1
+#define R128_DEFAULT_AGP_SIZE     8 /* MB (must be a power of 2 and > 4MB) */
+#define R128_DEFAULT_RING_SIZE    1 /* MB (must be page aligned) */
+#define R128_DEFAULT_BUFFER_SIZE  2 /* MB (must be page aligned) */
+#define R128_DEFAULT_AGP_TEX_SIZE 1 /* MB (must be page aligned) */
+
+#define R128_DEFAULT_CCE_TIMEOUT  10000  /* usecs */
+
+#define R128_AGP_MAX_MODE         4
+#define R128_BUFFER_ALIGN         0x00000fff
+
+#define R128_CARD_TYPE_R128          1
+#define R128_CARD_TYPE_R128_PRO      2
+#define R128_CARD_TYPE_R128_MOBILITY 3
+
+#define R128CCE_USE_RING_BUFFER(m)                                        \
+(((m) == R128_PM4_192BM) ||                                               \
+ ((m) == R128_PM4_128BM_64INDBM) ||                                       \
+ ((m) == R128_PM4_64BM_128INDBM) ||                                       \
+ ((m) == R128_PM4_64BM_64VCBM_64INDBM))
+
+typedef struct {
+    /* DRI screen private data */
+    int           deviceID;     /* PCI device ID */
+    int           width;        /* Width in pixels of display */
+    int           height;       /* Height in scanlines of display */
+    int           depth;        /* Depth of display (8, 15, 16, 24) */
+    int           bpp;          /* Bit depth of display (8, 16, 24, 32) */
+
+    int           IsPCI;        /* Current card is a PCI card */
+    int           AGPMode;
+
+    int           frontOffset;  /* Start of front buffer */
+    int           frontPitch;
+    int           backOffset;   /* Start of shared back buffer */
+    int           backPitch;
+    int           depthOffset;  /* Start of shared depth buffer */
+    int           depthPitch;
+    int           spanOffset;   /* Start of scratch spanline */
+    int           textureOffset;/* Start of texture data in frame buffer */
+    int           textureSize;
+    int           log2TexGran;
+
+    /* MMIO register data */
+    drmHandle     registerHandle;
+    drmSize       registerSize;
+
+    /* CCE AGP Texture data */
+    drmHandle     agpTexHandle;
+    drmSize       agpTexMapSize;
+    int           log2AGPTexGran;
+    int           agpTexOffset;
+    unsigned int  sarea_priv_offset;
+} R128DRIRec, *R128DRIPtr;
+
+#endif
diff --git a/src/mesa/drivers/dri/r128/server/r128_macros.h b/src/mesa/drivers/dri/r128/server/r128_macros.h
new file mode 100644
index 00000000000..93b7feb02c7
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/server/r128_macros.h
@@ -0,0 +1,135 @@
+/**
+ * \file server/R128_macros.h
+ * \brief Macros for R128 MMIO operation.
+ *
+ * \authors Kevin E. Martin <[email protected]>
+ * \authors Rickard E. Faith <[email protected]>
+ * \authors Alan Hourihane <[email protected]>
+ */
+
+/*
+ * Copyright 2000 ATI Technologies Inc., Markham, Ontario, and
+ *                VA Linux Systems Inc., Fremont, California.
+ *
+ * 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, 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
+ * NON-INFRINGEMENT.  IN NO EVENT SHALL ATI, VA LINUX SYSTEMS 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.
+ */
+
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/ati/R128_reg.h,v 1.20 2002/10/12 01:38:07 martin Exp $ */
+
+#ifndef _R128_MACROS_H_
+#define _R128_MACROS_H_
+
+
+
+#  define MMIO_IN8(base, offset) \
+	*(volatile unsigned char *)(((unsigned char*)(base)) + (offset))
+#  define MMIO_IN16(base, offset) \
+	*(volatile unsigned short *)(void *)(((unsigned char*)(base)) + (offset))
+#  define MMIO_IN32(base, offset) \
+	*(volatile unsigned int *)(void *)(((unsigned char*)(base)) + (offset))
+#  define MMIO_OUT8(base, offset, val) \
+	*(volatile unsigned char *)(((unsigned char*)(base)) + (offset)) = (val)
+#  define MMIO_OUT16(base, offset, val) \
+	*(volatile unsigned short *)(void *)(((unsigned char*)(base)) + (offset)) = (val)
+#  define MMIO_OUT32(base, offset, val) \
+	*(volatile unsigned int *)(void *)(((unsigned char*)(base)) + (offset)) = (val)
+
+
+				/* Memory mapped register access macros */
+#define INREG8(addr)        MMIO_IN8(R128MMIO, addr)
+#define INREG16(addr)       MMIO_IN16(R128MMIO, addr)
+#define INREG(addr)         MMIO_IN32(R128MMIO, addr)
+#define OUTREG8(addr, val)  MMIO_OUT8(R128MMIO, addr, val)
+#define OUTREG16(addr, val) MMIO_OUT16(R128MMIO, addr, val)
+#define OUTREG(addr, val)   MMIO_OUT32(R128MMIO, addr, val)
+
+#define ADDRREG(addr)       ((volatile GLuint *)(pointer)(R128MMIO + (addr)))
+
+
+#define OUTREGP(addr, val, mask)					\
+do {									\
+    GLuint tmp = INREG(addr);						\
+    tmp &= (mask);							\
+    tmp |= (val);							\
+    OUTREG(addr, tmp);							\
+} while (0)
+
+#define INPLL(dpy, addr) r128INPLL(dpy, addr)
+
+#define OUTPLL(addr, val)						\
+do {									\
+    OUTREG8(R128_CLOCK_CNTL_INDEX, (((addr) & 0x3f) |			\
+				      R128_PLL_WR_EN));		\
+    OUTREG(R128_CLOCK_CNTL_DATA, val);				\
+} while (0)
+
+#define OUTPLLP(dpy, addr, val, mask)					\
+do {									\
+    GLuint tmp = INPLL(dpy, addr);					\
+    tmp &= (mask);							\
+    tmp |= (val);							\
+    OUTPLL(addr, tmp);							\
+} while (0)
+
+#define OUTPAL_START(idx)						\
+do {									\
+    OUTREG8(R128_PALETTE_INDEX, (idx));				\
+} while (0)
+
+#define OUTPAL_NEXT(r, g, b)						\
+do {									\
+    OUTREG(R128_PALETTE_DATA, ((r) << 16) | ((g) << 8) | (b));	\
+} while (0)
+
+#define OUTPAL_NEXT_CARD32(v)						\
+do {									\
+    OUTREG(R128_PALETTE_DATA, (v & 0x00ffffff));			\
+} while (0)
+
+#define OUTPAL(idx, r, g, b)						\
+do {									\
+    OUTPAL_START((idx));						\
+    OUTPAL_NEXT((r), (g), (b));						\
+} while (0)
+
+#define INPAL_START(idx)						\
+do {									\
+    OUTREG(R128_PALETTE_INDEX, (idx) << 16);				\
+} while (0)
+
+#define INPAL_NEXT() INREG(R128_PALETTE_DATA)
+
+#define PAL_SELECT(idx)							\
+do {									\
+    if (!idx) {								\
+	OUTREG(R128_DAC_CNTL2, INREG(R128_DAC_CNTL2) &		\
+	       (GLuint)~R128_DAC2_PALETTE_ACC_CTL);			\
+    } else {								\
+	OUTREG(R128_DAC_CNTL2, INREG(R128_DAC_CNTL2) |		\
+	       R128_DAC2_PALETTE_ACC_CTL);				\
+    }									\
+} while (0)
+
+
+#endif
diff --git a/src/mesa/drivers/dri/r128/server/r128_reg.h b/src/mesa/drivers/dri/r128/server/r128_reg.h
new file mode 100644
index 00000000000..7ec4004492c
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/server/r128_reg.h
@@ -0,0 +1,1431 @@
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/ati/r128_reg.h,v 1.15 2002/12/16 16:19:11 dawes Exp $ */
+/*
+ * Copyright 1999, 2000 ATI Technologies Inc., Markham, Ontario,
+ *                      Precision Insight, Inc., Cedar Park, Texas, and
+ *                      VA Linux Systems Inc., Fremont, California.
+ *
+ * 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, 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
+ * NON-INFRINGEMENT.  IN NO EVENT SHALL ATI, PRECISION INSIGHT, VA LINUX
+ * SYSTEMS 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:
+ *   Rickard E. Faith <[email protected]>
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ *
+ * References:
+ *
+ *   RAGE 128 VR/ RAGE 128 GL Register Reference Manual (Technical
+ *   Reference Manual P/N RRG-G04100-C Rev. 0.04), ATI Technologies: April
+ *   1999.
+ *
+ *   RAGE 128 Software Development Manual (Technical Reference Manual P/N
+ *   SDK-G04000 Rev. 0.01), ATI Technologies: June 1999.
+ *
+ */
+
+#ifndef _R128_REG_H_
+#define _R128_REG_H_
+
+#define R128_ADAPTER_ID                   0x0f2c /* PCI */
+#define R128_AGP_APER_OFFSET              0x0178
+#define R128_AGP_BASE                     0x0170
+#define R128_AGP_CNTL                     0x0174
+#       define R128_AGP_APER_SIZE_256MB   (0x00 << 0)
+#       define R128_AGP_APER_SIZE_128MB   (0x20 << 0)
+#       define R128_AGP_APER_SIZE_64MB    (0x30 << 0)
+#       define R128_AGP_APER_SIZE_32MB    (0x38 << 0)
+#       define R128_AGP_APER_SIZE_16MB    (0x3c << 0)
+#       define R128_AGP_APER_SIZE_8MB     (0x3e << 0)
+#       define R128_AGP_APER_SIZE_4MB     (0x3f << 0)
+#       define R128_AGP_APER_SIZE_MASK    (0x3f << 0)
+#define R128_AGP_CNTL_B                   0x0b44
+#define R128_AGP_COMMAND                  0x0f58 /* PCI */
+#define R128_AGP_PLL_CNTL                 0x0010 /* PLL */
+#define R128_AGP_STATUS                   0x0f54 /* PCI */
+#       define R128_AGP_1X_MODE           0x01
+#       define R128_AGP_2X_MODE           0x02
+#       define R128_AGP_4X_MODE           0x04
+#       define R128_AGP_MODE_MASK         0x07
+#define R128_AMCGPIO_A_REG                0x01a0
+#define R128_AMCGPIO_EN_REG               0x01a8
+#define R128_AMCGPIO_MASK                 0x0194
+#define R128_AMCGPIO_Y_REG                0x01a4
+#define R128_ATTRDR                       0x03c1 /* VGA */
+#define R128_ATTRDW                       0x03c0 /* VGA */
+#define R128_ATTRX                        0x03c0 /* VGA */
+#define R128_AUX_SC_CNTL                  0x1660
+#       define R128_AUX1_SC_EN            (1 << 0)
+#       define R128_AUX1_SC_MODE_OR       (0 << 1)
+#       define R128_AUX1_SC_MODE_NAND     (1 << 1)
+#       define R128_AUX2_SC_EN            (1 << 2)
+#       define R128_AUX2_SC_MODE_OR       (0 << 3)
+#       define R128_AUX2_SC_MODE_NAND     (1 << 3)
+#       define R128_AUX3_SC_EN            (1 << 4)
+#       define R128_AUX3_SC_MODE_OR       (0 << 5)
+#       define R128_AUX3_SC_MODE_NAND     (1 << 5)
+#define R128_AUX1_SC_BOTTOM               0x1670
+#define R128_AUX1_SC_LEFT                 0x1664
+#define R128_AUX1_SC_RIGHT                0x1668
+#define R128_AUX1_SC_TOP                  0x166c
+#define R128_AUX2_SC_BOTTOM               0x1680
+#define R128_AUX2_SC_LEFT                 0x1674
+#define R128_AUX2_SC_RIGHT                0x1678
+#define R128_AUX2_SC_TOP                  0x167c
+#define R128_AUX3_SC_BOTTOM               0x1690
+#define R128_AUX3_SC_LEFT                 0x1684
+#define R128_AUX3_SC_RIGHT                0x1688
+#define R128_AUX3_SC_TOP                  0x168c
+#define R128_AUX_WINDOW_HORZ_CNTL         0x02d8
+#define R128_AUX_WINDOW_VERT_CNTL         0x02dc
+
+#define R128_BASE_CODE                    0x0f0b
+#define R128_BIOS_0_SCRATCH               0x0010
+#define R128_BIOS_1_SCRATCH               0x0014
+#define R128_BIOS_2_SCRATCH               0x0018
+#define R128_BIOS_3_SCRATCH               0x001c
+#define R128_BIOS_4_SCRATCH               0x0020
+#define R128_BIOS_5_SCRATCH               0x0024
+#       define R128_BIOS_DISPLAY_FP       (1 << 0)
+#       define R128_BIOS_DISPLAY_CRT      (2 << 0)
+#       define R128_BIOS_DISPLAY_FP_CRT   (3 << 0)
+#define R128_BIOS_6_SCRATCH               0x0028
+#define R128_BIOS_7_SCRATCH               0x002c
+#define R128_BIOS_ROM                     0x0f30 /* PCI */
+#define R128_BIST                         0x0f0f /* PCI */
+#define R128_BM_CHUNK_0_VAL               0x0a18
+#       define R128_BM_PTR_FORCE_TO_PCI    (1 << 21)
+#       define R128_BM_PM4_RD_FORCE_TO_PCI (1 << 22)
+#       define R128_BM_GLOBAL_FORCE_TO_PCI (1 << 23)
+#define R128_BRUSH_DATA0                  0x1480
+#define R128_BRUSH_DATA1                  0x1484
+#define R128_BRUSH_DATA10                 0x14a8
+#define R128_BRUSH_DATA11                 0x14ac
+#define R128_BRUSH_DATA12                 0x14b0
+#define R128_BRUSH_DATA13                 0x14b4
+#define R128_BRUSH_DATA14                 0x14b8
+#define R128_BRUSH_DATA15                 0x14bc
+#define R128_BRUSH_DATA16                 0x14c0
+#define R128_BRUSH_DATA17                 0x14c4
+#define R128_BRUSH_DATA18                 0x14c8
+#define R128_BRUSH_DATA19                 0x14cc
+#define R128_BRUSH_DATA2                  0x1488
+#define R128_BRUSH_DATA20                 0x14d0
+#define R128_BRUSH_DATA21                 0x14d4
+#define R128_BRUSH_DATA22                 0x14d8
+#define R128_BRUSH_DATA23                 0x14dc
+#define R128_BRUSH_DATA24                 0x14e0
+#define R128_BRUSH_DATA25                 0x14e4
+#define R128_BRUSH_DATA26                 0x14e8
+#define R128_BRUSH_DATA27                 0x14ec
+#define R128_BRUSH_DATA28                 0x14f0
+#define R128_BRUSH_DATA29                 0x14f4
+#define R128_BRUSH_DATA3                  0x148c
+#define R128_BRUSH_DATA30                 0x14f8
+#define R128_BRUSH_DATA31                 0x14fc
+#define R128_BRUSH_DATA32                 0x1500
+#define R128_BRUSH_DATA33                 0x1504
+#define R128_BRUSH_DATA34                 0x1508
+#define R128_BRUSH_DATA35                 0x150c
+#define R128_BRUSH_DATA36                 0x1510
+#define R128_BRUSH_DATA37                 0x1514
+#define R128_BRUSH_DATA38                 0x1518
+#define R128_BRUSH_DATA39                 0x151c
+#define R128_BRUSH_DATA4                  0x1490
+#define R128_BRUSH_DATA40                 0x1520
+#define R128_BRUSH_DATA41                 0x1524
+#define R128_BRUSH_DATA42                 0x1528
+#define R128_BRUSH_DATA43                 0x152c
+#define R128_BRUSH_DATA44                 0x1530
+#define R128_BRUSH_DATA45                 0x1534
+#define R128_BRUSH_DATA46                 0x1538
+#define R128_BRUSH_DATA47                 0x153c
+#define R128_BRUSH_DATA48                 0x1540
+#define R128_BRUSH_DATA49                 0x1544
+#define R128_BRUSH_DATA5                  0x1494
+#define R128_BRUSH_DATA50                 0x1548
+#define R128_BRUSH_DATA51                 0x154c
+#define R128_BRUSH_DATA52                 0x1550
+#define R128_BRUSH_DATA53                 0x1554
+#define R128_BRUSH_DATA54                 0x1558
+#define R128_BRUSH_DATA55                 0x155c
+#define R128_BRUSH_DATA56                 0x1560
+#define R128_BRUSH_DATA57                 0x1564
+#define R128_BRUSH_DATA58                 0x1568
+#define R128_BRUSH_DATA59                 0x156c
+#define R128_BRUSH_DATA6                  0x1498
+#define R128_BRUSH_DATA60                 0x1570
+#define R128_BRUSH_DATA61                 0x1574
+#define R128_BRUSH_DATA62                 0x1578
+#define R128_BRUSH_DATA63                 0x157c
+#define R128_BRUSH_DATA7                  0x149c
+#define R128_BRUSH_DATA8                  0x14a0
+#define R128_BRUSH_DATA9                  0x14a4
+#define R128_BRUSH_SCALE                  0x1470
+#define R128_BRUSH_Y_X                    0x1474
+#define R128_BUS_CNTL                     0x0030
+#       define R128_BUS_MASTER_DIS         (1 << 6)
+#       define R128_BUS_RD_DISCARD_EN      (1 << 24)
+#       define R128_BUS_RD_ABORT_EN        (1 << 25)
+#       define R128_BUS_MSTR_DISCONNECT_EN (1 << 28)
+#       define R128_BUS_WRT_BURST          (1 << 29)
+#       define R128_BUS_READ_BURST         (1 << 30)
+#define R128_BUS_CNTL1                    0x0034
+#       define R128_BUS_WAIT_ON_LOCK_EN    (1 << 4)
+
+#define R128_CACHE_CNTL                   0x1724
+#define R128_CACHE_LINE                   0x0f0c /* PCI */
+#define R128_CAP0_TRIG_CNTL               0x0950 /* ? */
+#define R128_CAP1_TRIG_CNTL               0x09c0 /* ? */
+#define R128_CAPABILITIES_ID              0x0f50 /* PCI */
+#define R128_CAPABILITIES_PTR             0x0f34 /* PCI */
+#define R128_CLK_PIN_CNTL                 0x0001 /* PLL */
+#define R128_CLOCK_CNTL_DATA              0x000c
+#define R128_CLOCK_CNTL_INDEX             0x0008
+#       define R128_PLL_WR_EN             (1 << 7)
+#       define R128_PLL_DIV_SEL           (3 << 8)
+#define R128_CLR_CMP_CLR_3D               0x1a24
+#define R128_CLR_CMP_CLR_DST              0x15c8
+#define R128_CLR_CMP_CLR_SRC              0x15c4
+#define R128_CLR_CMP_CNTL                 0x15c0
+#       define R128_SRC_CMP_EQ_COLOR      (4 <<  0)
+#       define R128_SRC_CMP_NEQ_COLOR     (5 <<  0)
+#       define R128_CLR_CMP_SRC_SOURCE    (1 << 24)
+#define R128_CLR_CMP_MASK                 0x15cc
+#       define R128_CLR_CMP_MSK           0xffffffff
+#define R128_CLR_CMP_MASK_3D              0x1A28
+#define R128_COMMAND                      0x0f04 /* PCI */
+#define R128_COMPOSITE_SHADOW_ID          0x1a0c
+#define R128_CONFIG_APER_0_BASE           0x0100
+#define R128_CONFIG_APER_1_BASE           0x0104
+#define R128_CONFIG_APER_SIZE             0x0108
+#define R128_CONFIG_BONDS                 0x00e8
+#define R128_CONFIG_CNTL                  0x00e0
+#       define APER_0_BIG_ENDIAN_16BPP_SWAP (1 << 0)
+#       define APER_0_BIG_ENDIAN_32BPP_SWAP (2 << 0)
+#define R128_CONFIG_MEMSIZE               0x00f8
+#define R128_CONFIG_MEMSIZE_EMBEDDED      0x0114
+#define R128_CONFIG_REG_1_BASE            0x010c
+#define R128_CONFIG_REG_APER_SIZE         0x0110
+#define R128_CONFIG_XSTRAP                0x00e4
+#define R128_CONSTANT_COLOR_C             0x1d34
+#       define R128_CONSTANT_COLOR_MASK   0x00ffffff
+#       define R128_CONSTANT_COLOR_ONE    0x00ffffff
+#       define R128_CONSTANT_COLOR_ZERO   0x00000000
+#define R128_CRC_CMDFIFO_ADDR             0x0740
+#define R128_CRC_CMDFIFO_DOUT             0x0744
+#define R128_CRTC_CRNT_FRAME              0x0214
+#define R128_CRTC_DEBUG                   0x021c
+#define R128_CRTC_EXT_CNTL                0x0054
+#       define R128_CRTC_VGA_XOVERSCAN    (1 <<  0)
+#       define R128_VGA_ATI_LINEAR        (1 <<  3)
+#       define R128_XCRT_CNT_EN           (1 <<  6)
+#       define R128_CRTC_HSYNC_DIS        (1 <<  8)
+#       define R128_CRTC_VSYNC_DIS        (1 <<  9)
+#       define R128_CRTC_DISPLAY_DIS      (1 << 10)
+#       define R128_CRTC_CRT_ON           (1 << 15)
+#       define R128_FP_OUT_EN             (1 << 22)
+#       define R128_FP_ACTIVE             (1 << 23)
+#define R128_CRTC_EXT_CNTL_DPMS_BYTE      0x0055
+#       define R128_CRTC_HSYNC_DIS_BYTE   (1 <<  0)
+#       define R128_CRTC_VSYNC_DIS_BYTE   (1 <<  1)
+#       define R128_CRTC_DISPLAY_DIS_BYTE (1 <<  2)
+#define R128_CRTC_GEN_CNTL                0x0050
+#       define R128_CRTC_DBL_SCAN_EN      (1 <<  0)
+#       define R128_CRTC_INTERLACE_EN     (1 <<  1)
+#       define R128_CRTC_CSYNC_EN         (1 <<  4)
+#       define R128_CRTC_CUR_EN           (1 << 16)
+#       define R128_CRTC_CUR_MODE_MASK    (7 << 17)
+#       define R128_CRTC_ICON_EN          (1 << 20)
+#       define R128_CRTC_EXT_DISP_EN      (1 << 24)
+#       define R128_CRTC_EN               (1 << 25)
+#       define R128_CRTC_DISP_REQ_EN_B    (1 << 26)
+#define R128_CRTC_GUI_TRIG_VLINE          0x0218
+#define R128_CRTC_H_SYNC_STRT_WID         0x0204
+#       define R128_CRTC_H_SYNC_STRT_PIX        (0x07  <<  0)
+#       define R128_CRTC_H_SYNC_STRT_CHAR       (0x1ff <<  3)
+#       define R128_CRTC_H_SYNC_STRT_CHAR_SHIFT 3
+#       define R128_CRTC_H_SYNC_WID             (0x3f  << 16)
+#       define R128_CRTC_H_SYNC_WID_SHIFT       16
+#       define R128_CRTC_H_SYNC_POL             (1     << 23)
+#define R128_CRTC_H_TOTAL_DISP            0x0200
+#       define R128_CRTC_H_TOTAL          (0x01ff << 0)
+#       define R128_CRTC_H_TOTAL_SHIFT    0
+#       define R128_CRTC_H_DISP           (0x00ff << 16)
+#       define R128_CRTC_H_DISP_SHIFT     16
+#define R128_CRTC_OFFSET                  0x0224
+#define R128_CRTC_OFFSET_CNTL             0x0228
+#define R128_CRTC_PITCH                   0x022c
+#define R128_CRTC_STATUS                  0x005c
+#       define R128_CRTC_VBLANK_SAVE      (1 <<  1)
+#define R128_CRTC_V_SYNC_STRT_WID         0x020c
+#       define R128_CRTC_V_SYNC_STRT       (0x7ff <<  0)
+#       define R128_CRTC_V_SYNC_STRT_SHIFT 0
+#       define R128_CRTC_V_SYNC_WID        (0x1f  << 16)
+#       define R128_CRTC_V_SYNC_WID_SHIFT  16
+#       define R128_CRTC_V_SYNC_POL        (1     << 23)
+#define R128_CRTC_V_TOTAL_DISP            0x0208
+#       define R128_CRTC_V_TOTAL          (0x07ff << 0)
+#       define R128_CRTC_V_TOTAL_SHIFT    0
+#       define R128_CRTC_V_DISP           (0x07ff << 16)
+#       define R128_CRTC_V_DISP_SHIFT     16
+#define R128_CRTC_VLINE_CRNT_VLINE        0x0210
+#       define R128_CRTC_CRNT_VLINE_MASK  (0x7ff << 16)
+#define R128_CRTC2_CRNT_FRAME             0x0314
+#define R128_CRTC2_DEBUG                  0x031c
+#define R128_CRTC2_GEN_CNTL               0x03f8
+#define R128_CRTC2_GUI_TRIG_VLINE         0x0318
+#define R128_CRTC2_H_SYNC_STRT_WID        0x0304
+#define R128_CRTC2_H_TOTAL_DISP           0x0300
+#define R128_CRTC2_OFFSET                 0x0324
+#define R128_CRTC2_OFFSET_CNTL            0x0328
+#define R128_CRTC2_PITCH                  0x032c
+#define R128_CRTC2_STATUS                 0x03fc
+#define R128_CRTC2_V_SYNC_STRT_WID        0x030c
+#define R128_CRTC2_V_TOTAL_DISP           0x0308
+#define R128_CRTC2_VLINE_CRNT_VLINE       0x0310
+#define R128_CRTC8_DATA                   0x03d5 /* VGA, 0x3b5 */
+#define R128_CRTC8_IDX                    0x03d4 /* VGA, 0x3b4 */
+#define R128_CUR_CLR0                     0x026c
+#define R128_CUR_CLR1                     0x0270
+#define R128_CUR_HORZ_VERT_OFF            0x0268
+#define R128_CUR_HORZ_VERT_POSN           0x0264
+#define R128_CUR_OFFSET                   0x0260
+#       define R128_CUR_LOCK              (1 << 31)
+
+#define R128_DAC_CNTL                     0x0058
+#       define R128_DAC_RANGE_CNTL        (3 <<  0)
+#       define R128_DAC_BLANKING          (1 <<  2)
+#       define R128_DAC_CRT_SEL_CRTC2     (1 <<  4)
+#       define R128_DAC_PALETTE_ACC_CTL   (1 <<  5)
+#       define R128_DAC_8BIT_EN           (1 <<  8)
+#       define R128_DAC_VGA_ADR_EN        (1 << 13)
+#       define R128_DAC_MASK_ALL          (0xff << 24)
+#define R128_DAC_CRC_SIG                  0x02cc
+#define R128_DAC_DATA                     0x03c9 /* VGA */
+#define R128_DAC_MASK                     0x03c6 /* VGA */
+#define R128_DAC_R_INDEX                  0x03c7 /* VGA */
+#define R128_DAC_W_INDEX                  0x03c8 /* VGA */
+#define R128_DDA_CONFIG                   0x02e0
+#define R128_DDA_ON_OFF                   0x02e4
+#define R128_DEFAULT_OFFSET               0x16e0
+#define R128_DEFAULT_PITCH                0x16e4
+#define R128_DEFAULT_SC_BOTTOM_RIGHT      0x16e8
+#       define R128_DEFAULT_SC_RIGHT_MAX  (0x1fff <<  0)
+#       define R128_DEFAULT_SC_BOTTOM_MAX (0x1fff << 16)
+#define R128_DESTINATION_3D_CLR_CMP_VAL   0x1820
+#define R128_DESTINATION_3D_CLR_CMP_MSK   0x1824
+#define R128_DEVICE_ID                    0x0f02 /* PCI */
+#define R128_DP_BRUSH_BKGD_CLR            0x1478
+#define R128_DP_BRUSH_FRGD_CLR            0x147c
+#define R128_DP_CNTL                      0x16c0
+#       define R128_DST_X_LEFT_TO_RIGHT   (1 <<  0)
+#       define R128_DST_Y_TOP_TO_BOTTOM   (1 <<  1)
+#define R128_DP_CNTL_XDIR_YDIR_YMAJOR     0x16d0
+#       define R128_DST_Y_MAJOR             (1 <<  2)
+#       define R128_DST_Y_DIR_TOP_TO_BOTTOM (1 << 15)
+#       define R128_DST_X_DIR_LEFT_TO_RIGHT (1 << 31)
+#define R128_DP_DATATYPE                  0x16c4
+#       define R128_HOST_BIG_ENDIAN_EN    (1 << 29)
+#define R128_DP_GUI_MASTER_CNTL           0x146c
+#       define R128_GMC_SRC_PITCH_OFFSET_CNTL (1    <<  0)
+#       define R128_GMC_DST_PITCH_OFFSET_CNTL (1    <<  1)
+#       define R128_GMC_SRC_CLIPPING          (1    <<  2)
+#       define R128_GMC_DST_CLIPPING          (1    <<  3)
+#       define R128_GMC_BRUSH_DATATYPE_MASK   (0x0f <<  4)
+#       define R128_GMC_BRUSH_8X8_MONO_FG_BG  (0    <<  4)
+#       define R128_GMC_BRUSH_8X8_MONO_FG_LA  (1    <<  4)
+#       define R128_GMC_BRUSH_1X8_MONO_FG_BG  (4    <<  4)
+#       define R128_GMC_BRUSH_1X8_MONO_FG_LA  (5    <<  4)
+#       define R128_GMC_BRUSH_32x1_MONO_FG_BG (6    <<  4)
+#       define R128_GMC_BRUSH_32x1_MONO_FG_LA (7    <<  4)
+#       define R128_GMC_BRUSH_32x32_MONO_FG_BG (8    <<  4)
+#       define R128_GMC_BRUSH_32x32_MONO_FG_LA (9    <<  4)
+#       define R128_GMC_BRUSH_8x8_COLOR       (10   <<  4)
+#       define R128_GMC_BRUSH_1X8_COLOR       (12   <<  4)
+#       define R128_GMC_BRUSH_SOLID_COLOR     (13   <<  4)
+#       define R128_GMC_BRUSH_NONE            (15   <<  4)
+#       define R128_GMC_DST_8BPP_CI           (2    <<  8)
+#       define R128_GMC_DST_15BPP             (3    <<  8)
+#       define R128_GMC_DST_16BPP             (4    <<  8)
+#       define R128_GMC_DST_24BPP             (5    <<  8)
+#       define R128_GMC_DST_32BPP             (6    <<  8)
+#       define R128_GMC_DST_8BPP_RGB          (7    <<  8)
+#       define R128_GMC_DST_Y8                (8    <<  8)
+#       define R128_GMC_DST_RGB8              (9    <<  8)
+#       define R128_GMC_DST_VYUY              (11   <<  8)
+#       define R128_GMC_DST_YVYU              (12   <<  8)
+#       define R128_GMC_DST_AYUV444           (14   <<  8)
+#       define R128_GMC_DST_ARGB4444          (15   <<  8)
+#       define R128_GMC_DST_DATATYPE_MASK     (0x0f <<  8)
+#       define R128_GMC_DST_DATATYPE_SHIFT    8
+#       define R128_GMC_SRC_DATATYPE_MASK       (3    << 12)
+#       define R128_GMC_SRC_DATATYPE_MONO_FG_BG (0    << 12)
+#       define R128_GMC_SRC_DATATYPE_MONO_FG_LA (1    << 12)
+#       define R128_GMC_SRC_DATATYPE_COLOR      (3    << 12)
+#       define R128_GMC_BYTE_PIX_ORDER        (1    << 14)
+#       define R128_GMC_BYTE_MSB_TO_LSB       (0    << 14)
+#       define R128_GMC_BYTE_LSB_TO_MSB       (1    << 14)
+#       define R128_GMC_CONVERSION_TEMP       (1    << 15)
+#       define R128_GMC_CONVERSION_TEMP_6500  (0    << 15)
+#       define R128_GMC_CONVERSION_TEMP_9300  (1    << 15)
+#       define R128_GMC_ROP3_MASK             (0xff << 16)
+#       define R128_DP_SRC_SOURCE_MASK        (7    << 24)
+#       define R128_DP_SRC_SOURCE_MEMORY      (2    << 24)
+#       define R128_DP_SRC_SOURCE_HOST_DATA   (3    << 24)
+#       define R128_GMC_3D_FCN_EN             (1    << 27)
+#       define R128_GMC_CLR_CMP_CNTL_DIS      (1    << 28)
+#       define R128_GMC_AUX_CLIP_DIS          (1    << 29)
+#       define R128_GMC_WR_MSK_DIS            (1    << 30)
+#       define R128_GMC_LD_BRUSH_Y_X          (1    << 31)
+#       define R128_ROP3_ZERO             0x00000000
+#       define R128_ROP3_DSa              0x00880000
+#       define R128_ROP3_SDna             0x00440000
+#       define R128_ROP3_S                0x00cc0000
+#       define R128_ROP3_DSna             0x00220000
+#       define R128_ROP3_D                0x00aa0000
+#       define R128_ROP3_DSx              0x00660000
+#       define R128_ROP3_DSo              0x00ee0000
+#       define R128_ROP3_DSon             0x00110000
+#       define R128_ROP3_DSxn             0x00990000
+#       define R128_ROP3_Dn               0x00550000
+#       define R128_ROP3_SDno             0x00dd0000
+#       define R128_ROP3_Sn               0x00330000
+#       define R128_ROP3_DSno             0x00bb0000
+#       define R128_ROP3_DSan             0x00770000
+#       define R128_ROP3_ONE              0x00ff0000
+#       define R128_ROP3_DPa              0x00a00000
+#       define R128_ROP3_PDna             0x00500000
+#       define R128_ROP3_P                0x00f00000
+#       define R128_ROP3_DPna             0x000a0000
+#       define R128_ROP3_D                0x00aa0000
+#       define R128_ROP3_DPx              0x005a0000
+#       define R128_ROP3_DPo              0x00fa0000
+#       define R128_ROP3_DPon             0x00050000
+#       define R128_ROP3_PDxn             0x00a50000
+#       define R128_ROP3_PDno             0x00f50000
+#       define R128_ROP3_Pn               0x000f0000
+#       define R128_ROP3_DPno             0x00af0000
+#       define R128_ROP3_DPan             0x005f0000
+
+
+#define R128_DP_GUI_MASTER_CNTL_C         0x1c84
+#define R128_DP_MIX                       0x16c8
+#define R128_DP_SRC_BKGD_CLR              0x15dc
+#define R128_DP_SRC_FRGD_CLR              0x15d8
+#define R128_DP_WRITE_MASK                0x16cc
+#define R128_DST_BRES_DEC                 0x1630
+#define R128_DST_BRES_ERR                 0x1628
+#define R128_DST_BRES_INC                 0x162c
+#define R128_DST_BRES_LNTH                0x1634
+#define R128_DST_BRES_LNTH_SUB            0x1638
+#define R128_DST_HEIGHT                   0x1410
+#define R128_DST_HEIGHT_WIDTH             0x143c
+#define R128_DST_HEIGHT_WIDTH_8           0x158c
+#define R128_DST_HEIGHT_WIDTH_BW          0x15b4
+#define R128_DST_HEIGHT_Y                 0x15a0
+#define R128_DST_OFFSET                   0x1404
+#define R128_DST_PITCH                    0x1408
+#define R128_DST_PITCH_OFFSET             0x142c
+#define R128_DST_PITCH_OFFSET_C           0x1c80
+#       define R128_PITCH_SHIFT               21
+#       define R128_DST_TILE                 (1 << 31)
+#define R128_DST_WIDTH                    0x140c
+#define R128_DST_WIDTH_HEIGHT             0x1598
+#define R128_DST_WIDTH_X                  0x1588
+#define R128_DST_WIDTH_X_INCY             0x159c
+#define R128_DST_X                        0x141c
+#define R128_DST_X_SUB                    0x15a4
+#define R128_DST_X_Y                      0x1594
+#define R128_DST_Y                        0x1420
+#define R128_DST_Y_SUB                    0x15a8
+#define R128_DST_Y_X                      0x1438
+
+#define R128_EXT_MEM_CNTL                 0x0144
+
+#define R128_FCP_CNTL                     0x0012 /* PLL */
+#define R128_FLUSH_1                      0x1704
+#define R128_FLUSH_2                      0x1708
+#define R128_FLUSH_3                      0x170c
+#define R128_FLUSH_4                      0x1710
+#define R128_FLUSH_5                      0x1714
+#define R128_FLUSH_6                      0x1718
+#define R128_FLUSH_7                      0x171c
+#define R128_FOG_3D_TABLE_START           0x1810
+#define R128_FOG_3D_TABLE_END             0x1814
+#define R128_FOG_3D_TABLE_DENSITY         0x181c
+#define R128_FOG_TABLE_INDEX              0x1a14
+#define R128_FOG_TABLE_DATA               0x1a18
+#define R128_FP_CRTC_H_TOTAL_DISP         0x0250
+#define R128_FP_CRTC_V_TOTAL_DISP         0x0254
+#define R128_FP_GEN_CNTL                  0x0284
+#       define R128_FP_FPON                  (1 << 0)
+#       define R128_FP_BLANK_DIS             (1 << 1)
+#       define R128_FP_TDMS_EN               (1 <<  2)
+#       define R128_FP_DETECT_SENSE          (1 <<  8)
+#       define R128_FP_SEL_CRTC2             (1 << 13)
+#       define R128_FP_CRTC_DONT_SHADOW_VPAR (1 << 16)
+#       define R128_FP_CRTC_DONT_SHADOW_HEND (1 << 17)
+#       define R128_FP_CRTC_USE_SHADOW_VEND  (1 << 18)
+#       define R128_FP_CRTC_USE_SHADOW_ROWCUR (1 << 19)
+#       define R128_FP_CRTC_HORZ_DIV2_EN     (1 << 20)
+#       define R128_FP_CRTC_HOR_CRT_DIV2_DIS (1 << 21)
+#       define R128_FP_CRT_SYNC_SEL          (1 << 23)
+#       define R128_FP_USE_SHADOW_EN         (1 << 24)
+#define R128_FP_H_SYNC_STRT_WID           0x02c4
+#define R128_FP_HORZ_STRETCH              0x028c
+#       define R128_HORZ_STRETCH_RATIO_MASK  0xffff
+#       define R128_HORZ_STRETCH_RATIO_SHIFT 0
+#       define R128_HORZ_STRETCH_RATIO_MAX   4096
+#       define R128_HORZ_PANEL_SIZE          (0xff   << 16)
+#       define R128_HORZ_PANEL_SHIFT         16
+#       define R128_AUTO_HORZ_RATIO          (0      << 24)
+#       define R128_HORZ_STRETCH_PIXREP      (0      << 25)
+#       define R128_HORZ_STRETCH_BLEND       (1      << 25)
+#       define R128_HORZ_STRETCH_ENABLE      (1      << 26)
+#       define R128_HORZ_FP_LOOP_STRETCH     (0x7    << 27)
+#       define R128_HORZ_STRETCH_RESERVED    (1      << 30)
+#       define R128_HORZ_AUTO_RATIO_FIX_EN   (1      << 31)
+
+#define R128_FP_PANEL_CNTL                0x0288
+#       define R128_FP_DIGON              (1 << 0)
+#       define R128_FP_BLON               (1 << 1)
+#define R128_FP_V_SYNC_STRT_WID           0x02c8
+#define R128_FP_VERT_STRETCH              0x0290
+#       define R128_VERT_PANEL_SIZE          (0x7ff <<  0)
+#       define R128_VERT_PANEL_SHIFT         0
+#       define R128_VERT_STRETCH_RATIO_MASK  0x3ff
+#       define R128_VERT_STRETCH_RATIO_SHIFT 11
+#       define R128_VERT_STRETCH_RATIO_MAX   1024
+#       define R128_VERT_STRETCH_ENABLE      (1     << 24)
+#       define R128_VERT_STRETCH_LINEREP     (0     << 25)
+#       define R128_VERT_STRETCH_BLEND       (1     << 25)
+#       define R128_VERT_AUTO_RATIO_EN       (1     << 26)
+#       define R128_VERT_STRETCH_RESERVED    0xf8e00000
+
+#define R128_GEN_INT_CNTL                 0x0040
+#define R128_GEN_INT_STATUS               0x0044
+#       define R128_VSYNC_INT_AK          (1 <<  2)
+#       define R128_VSYNC_INT             (1 <<  2)
+#define R128_GEN_RESET_CNTL               0x00f0
+#       define R128_SOFT_RESET_GUI          (1 <<  0)
+#       define R128_SOFT_RESET_VCLK         (1 <<  8)
+#       define R128_SOFT_RESET_PCLK         (1 <<  9)
+#       define R128_SOFT_RESET_DISPENG_XCLK (1 << 11)
+#       define R128_SOFT_RESET_MEMCTLR_XCLK (1 << 12)
+#define R128_GENENB                       0x03c3 /* VGA */
+#define R128_GENFC_RD                     0x03ca /* VGA */
+#define R128_GENFC_WT                     0x03da /* VGA, 0x03ba */
+#define R128_GENMO_RD                     0x03cc /* VGA */
+#define R128_GENMO_WT                     0x03c2 /* VGA */
+#define R128_GENS0                        0x03c2 /* VGA */
+#define R128_GENS1                        0x03da /* VGA, 0x03ba */
+#define R128_GPIO_MONID                   0x0068
+#       define R128_GPIO_MONID_A_0        (1 <<  0)
+#       define R128_GPIO_MONID_A_1        (1 <<  1)
+#       define R128_GPIO_MONID_A_2        (1 <<  2)
+#       define R128_GPIO_MONID_A_3        (1 <<  3)
+#       define R128_GPIO_MONID_Y_0        (1 <<  8)
+#       define R128_GPIO_MONID_Y_1        (1 <<  9)
+#       define R128_GPIO_MONID_Y_2        (1 << 10)
+#       define R128_GPIO_MONID_Y_3        (1 << 11)
+#       define R128_GPIO_MONID_EN_0       (1 << 16)
+#       define R128_GPIO_MONID_EN_1       (1 << 17)
+#       define R128_GPIO_MONID_EN_2       (1 << 18)
+#       define R128_GPIO_MONID_EN_3       (1 << 19)
+#       define R128_GPIO_MONID_MASK_0     (1 << 24)
+#       define R128_GPIO_MONID_MASK_1     (1 << 25)
+#       define R128_GPIO_MONID_MASK_2     (1 << 26)
+#       define R128_GPIO_MONID_MASK_3     (1 << 27)
+#define R128_GPIO_MONIDB                  0x006c
+#define R128_GRPH8_DATA                   0x03cf /* VGA */
+#define R128_GRPH8_IDX                    0x03ce /* VGA */
+#define R128_GUI_DEBUG0                   0x16a0
+#define R128_GUI_DEBUG1                   0x16a4
+#define R128_GUI_DEBUG2                   0x16a8
+#define R128_GUI_DEBUG3                   0x16ac
+#define R128_GUI_DEBUG4                   0x16b0
+#define R128_GUI_DEBUG5                   0x16b4
+#define R128_GUI_DEBUG6                   0x16b8
+#define R128_GUI_PROBE                    0x16bc
+#define R128_GUI_SCRATCH_REG0             0x15e0
+#define R128_GUI_SCRATCH_REG1             0x15e4
+#define R128_GUI_SCRATCH_REG2             0x15e8
+#define R128_GUI_SCRATCH_REG3             0x15ec
+#define R128_GUI_SCRATCH_REG4             0x15f0
+#define R128_GUI_SCRATCH_REG5             0x15f4
+#define R128_GUI_STAT                     0x1740
+#       define R128_GUI_FIFOCNT_MASK      0x0fff
+#       define R128_GUI_ACTIVE            (1 << 31)
+
+#define R128_HEADER                       0x0f0e /* PCI */
+#define R128_HOST_DATA0                   0x17c0
+#define R128_HOST_DATA1                   0x17c4
+#define R128_HOST_DATA2                   0x17c8
+#define R128_HOST_DATA3                   0x17cc
+#define R128_HOST_DATA4                   0x17d0
+#define R128_HOST_DATA5                   0x17d4
+#define R128_HOST_DATA6                   0x17d8
+#define R128_HOST_DATA7                   0x17dc
+#define R128_HOST_DATA_LAST               0x17e0
+#define R128_HOST_PATH_CNTL               0x0130
+#define R128_HTOTAL_CNTL                  0x0009 /* PLL */
+#define R128_HW_DEBUG                     0x0128
+#define R128_HW_DEBUG2                    0x011c
+
+#define R128_I2C_CNTL_1                   0x0094 /* ? */
+#define R128_INTERRUPT_LINE               0x0f3c /* PCI */
+#define R128_INTERRUPT_PIN                0x0f3d /* PCI */
+#define R128_IO_BASE                      0x0f14 /* PCI */
+
+#define R128_LATENCY                      0x0f0d /* PCI */
+#define R128_LEAD_BRES_DEC                0x1608
+#define R128_LEAD_BRES_ERR                0x1600
+#define R128_LEAD_BRES_INC                0x1604
+#define R128_LEAD_BRES_LNTH               0x161c
+#define R128_LEAD_BRES_LNTH_SUB           0x1624
+#define R128_LVDS_GEN_CNTL                0x02d0
+#       define R128_LVDS_ON               (1   <<  0)
+#       define R128_LVDS_DISPLAY_DIS      (1   <<  1)
+#       define R128_LVDS_EN               (1   <<  7)
+#       define R128_LVDS_DIGON            (1   << 18)
+#       define R128_LVDS_BLON             (1   << 19)
+#       define R128_LVDS_SEL_CRTC2        (1   << 23)
+#       define R128_HSYNC_DELAY_SHIFT     28
+#       define R128_HSYNC_DELAY_MASK      (0xf << 28)
+
+#define R128_MAX_LATENCY                  0x0f3f /* PCI */
+#define R128_MCLK_CNTL                    0x000f /* PLL */
+#       define R128_FORCE_GCP             (1 << 16)
+#       define R128_FORCE_PIPE3D_CP       (1 << 17)
+#       define R128_FORCE_RCP             (1 << 18)
+#define R128_MDGPIO_A_REG                 0x01ac
+#define R128_MDGPIO_EN_REG                0x01b0
+#define R128_MDGPIO_MASK                  0x0198
+#define R128_MDGPIO_Y_REG                 0x01b4
+#define R128_MEM_ADDR_CONFIG              0x0148
+#define R128_MEM_BASE                     0x0f10 /* PCI */
+#define R128_MEM_CNTL                     0x0140
+#define R128_MEM_INIT_LAT_TIMER           0x0154
+#define R128_MEM_INTF_CNTL                0x014c
+#define R128_MEM_SDRAM_MODE_REG           0x0158
+#define R128_MEM_STR_CNTL                 0x0150
+#define R128_MEM_VGA_RP_SEL               0x003c
+#define R128_MEM_VGA_WP_SEL               0x0038
+#define R128_MIN_GRANT                    0x0f3e /* PCI */
+#define R128_MM_DATA                      0x0004
+#define R128_MM_INDEX                     0x0000
+#define R128_MPLL_CNTL                    0x000e /* PLL */
+#define R128_MPP_TB_CONFIG                0x01c0 /* ? */
+#define R128_MPP_GP_CONFIG                0x01c8 /* ? */
+
+#define R128_N_VIF_COUNT                  0x0248
+
+#define R128_OVR_CLR                      0x0230
+#define R128_OVR_WID_LEFT_RIGHT           0x0234
+#define R128_OVR_WID_TOP_BOTTOM           0x0238
+
+/* first overlay unit (there is only one) */
+
+#define R128_OV0_Y_X_START                0x0400
+#define R128_OV0_Y_X_END                  0x0404
+#define R128_OV0_EXCLUSIVE_HORZ           0x0408
+#       define  R128_EXCL_HORZ_START_MASK        0x000000ff
+#       define  R128_EXCL_HORZ_END_MASK          0x0000ff00
+#       define  R128_EXCL_HORZ_BACK_PORCH_MASK   0x00ff0000
+#       define  R128_EXCL_HORZ_EXCLUSIVE_EN      0x80000000
+#define R128_OV0_EXCLUSIVE_VERT           0x040C
+#       define  R128_EXCL_VERT_START_MASK        0x000003ff
+#       define  R128_EXCL_VERT_END_MASK          0x03ff0000
+#define R128_OV0_REG_LOAD_CNTL            0x0410
+#       define  R128_REG_LD_CTL_LOCK                 0x00000001L
+#       define  R128_REG_LD_CTL_VBLANK_DURING_LOCK   0x00000002L
+#       define  R128_REG_LD_CTL_STALL_GUI_UNTIL_FLIP 0x00000004L
+#       define  R128_REG_LD_CTL_LOCK_READBACK        0x00000008L
+#define R128_OV0_SCALE_CNTL               0x0420
+#       define  R128_SCALER_PIX_EXPAND           0x00000001L
+#       define  R128_SCALER_Y2R_TEMP             0x00000002L
+#       define  R128_SCALER_HORZ_PICK_NEAREST    0x00000003L
+#       define  R128_SCALER_VERT_PICK_NEAREST    0x00000004L
+#       define  R128_SCALER_SIGNED_UV            0x00000010L
+#       define  R128_SCALER_GAMMA_SEL_MASK       0x00000060L
+#       define  R128_SCALER_GAMMA_SEL_BRIGHT     0x00000000L
+#       define  R128_SCALER_GAMMA_SEL_G22        0x00000020L
+#       define  R128_SCALER_GAMMA_SEL_G18        0x00000040L
+#       define  R128_SCALER_GAMMA_SEL_G14        0x00000060L
+#       define  R128_SCALER_COMCORE_SHIFT_UP_ONE 0x00000080L
+#       define  R128_SCALER_SURFAC_FORMAT        0x00000f00L
+#       define  R128_SCALER_SOURCE_15BPP         0x00000300L
+#       define  R128_SCALER_SOURCE_16BPP         0x00000400L
+#       define  R128_SCALER_SOURCE_32BPP         0x00000600L
+#       define  R128_SCALER_SOURCE_YUV9          0x00000900L
+#       define  R128_SCALER_SOURCE_YUV12         0x00000A00L
+#       define  R128_SCALER_SOURCE_VYUY422       0x00000B00L
+#       define  R128_SCALER_SOURCE_YVYU422       0x00000C00L
+#       define  R128_SCALER_SMART_SWITCH         0x00008000L
+#       define  R128_SCALER_BURST_PER_PLANE      0x00ff0000L
+#       define  R128_SCALER_DOUBLE_BUFFER        0x01000000L
+#       define  R128_SCALER_DIS_LIMIT            0x08000000L
+#       define  R128_SCALER_PRG_LOAD_START       0x10000000L
+#       define  R128_SCALER_INT_EMU              0x20000000L
+#       define  R128_SCALER_ENABLE               0x40000000L
+#       define  R128_SCALER_SOFT_RESET           0x80000000L
+#define R128_OV0_V_INC                    0x0424
+#define R128_OV0_P1_V_ACCUM_INIT          0x0428
+#       define  R128_OV0_P1_MAX_LN_IN_PER_LN_OUT        0x00000003L
+#       define  R128_OV0_P1_V_ACCUM_INIT_MASK           0x01ff8000L
+#define R128_OV0_P23_V_ACCUM_INIT         0x042C
+#define R128_OV0_P1_BLANK_LINES_AT_TOP    0x0430
+#       define  R128_P1_BLNK_LN_AT_TOP_M1_MASK   0x00000fffL
+#       define  R128_P1_ACTIVE_LINES_M1          0x0fff0000L
+#define R128_OV0_P23_BLANK_LINES_AT_TOP   0x0434
+#       define  R128_P23_BLNK_LN_AT_TOP_M1_MASK  0x000007ffL
+#       define  R128_P23_ACTIVE_LINES_M1         0x07ff0000L
+#define R128_OV0_VID_BUF0_BASE_ADRS       0x0440
+#       define  R128_VIF_BUF0_PITCH_SEL          0x00000001L
+#       define  R128_VIF_BUF0_TILE_ADRS          0x00000002L
+#       define  R128_VIF_BUF0_BASE_ADRS_MASK     0x03fffff0L
+#       define  R128_VIF_BUF0_1ST_LINE_LSBS_MASK 0x48000000L
+#define R128_OV0_VID_BUF1_BASE_ADRS       0x0444
+#       define  R128_VIF_BUF1_PITCH_SEL          0x00000001L
+#       define  R128_VIF_BUF1_TILE_ADRS          0x00000002L
+#       define  R128_VIF_BUF1_BASE_ADRS_MASK     0x03fffff0L
+#       define  R128_VIF_BUF1_1ST_LINE_LSBS_MASK 0x48000000L
+#define R128_OV0_VID_BUF2_BASE_ADRS       0x0448
+#       define  R128_VIF_BUF2_PITCH_SEL          0x00000001L
+#       define  R128_VIF_BUF2_TILE_ADRS          0x00000002L
+#       define  R128_VIF_BUF2_BASE_ADRS_MASK     0x03fffff0L
+#       define  R128_VIF_BUF2_1ST_LINE_LSBS_MASK 0x48000000L
+#define R128_OV0_VID_BUF3_BASE_ADRS       0x044C
+#define R128_OV0_VID_BUF4_BASE_ADRS       0x0450
+#define R128_OV0_VID_BUF5_BASE_ADRS       0x0454
+#define R128_OV0_VID_BUF_PITCH0_VALUE     0x0460
+#define R128_OV0_VID_BUF_PITCH1_VALUE     0x0464
+#define R128_OV0_AUTO_FLIP_CNTL           0x0470
+#define R128_OV0_DEINTERLACE_PATTERN      0x0474
+#define R128_OV0_H_INC                    0x0480
+#define R128_OV0_STEP_BY                  0x0484
+#define R128_OV0_P1_H_ACCUM_INIT          0x0488
+#define R128_OV0_P23_H_ACCUM_INIT         0x048C
+#define R128_OV0_P1_X_START_END           0x0494
+#define R128_OV0_P2_X_START_END           0x0498
+#define R128_OV0_P3_X_START_END           0x049C
+#define R128_OV0_FILTER_CNTL              0x04A0
+#define R128_OV0_FOUR_TAP_COEF_0          0x04B0
+#define R128_OV0_FOUR_TAP_COEF_1          0x04B4
+#define R128_OV0_FOUR_TAP_COEF_2          0x04B8
+#define R128_OV0_FOUR_TAP_COEF_3          0x04BC
+#define R128_OV0_FOUR_TAP_COEF_4          0x04C0
+#define R128_OV0_COLOUR_CNTL              0x04E0
+#define R128_OV0_VIDEO_KEY_CLR            0x04E4
+#define R128_OV0_VIDEO_KEY_MSK            0x04E8
+#define R128_OV0_GRAPHICS_KEY_CLR         0x04EC
+#define R128_OV0_GRAPHICS_KEY_MSK         0x04F0
+#define R128_OV0_KEY_CNTL                 0x04F4
+#       define  R128_VIDEO_KEY_FN_MASK           0x00000007L
+#       define  R128_VIDEO_KEY_FN_FALSE          0x00000000L
+#       define  R128_VIDEO_KEY_FN_TRUE           0x00000001L
+#       define  R128_VIDEO_KEY_FN_EQ             0x00000004L
+#       define  R128_VIDEO_KEY_FN_NE             0x00000005L
+#       define  R128_GRAPHIC_KEY_FN_MASK         0x00000070L
+#       define  R128_GRAPHIC_KEY_FN_FALSE        0x00000000L
+#       define  R128_GRAPHIC_KEY_FN_TRUE         0x00000010L
+#       define  R128_GRAPHIC_KEY_FN_EQ           0x00000040L
+#       define  R128_GRAPHIC_KEY_FN_NE           0x00000050L
+#       define  R128_CMP_MIX_MASK                0x00000100L
+#       define  R128_CMP_MIX_OR                  0x00000000L
+#       define  R128_CMP_MIX_AND                 0x00000100L
+#define R128_OV0_TEST                     0x04F8
+
+
+#define R128_PALETTE_DATA                 0x00b4
+#define R128_PALETTE_INDEX                0x00b0
+#define R128_PC_DEBUG_MODE                0x1760
+#define R128_PC_GUI_CTLSTAT               0x1748
+#define R128_PC_GUI_MODE                  0x1744
+#       define R128_PC_IGNORE_UNIFY       (1 << 5)
+#define R128_PC_MISC_CNTL                 0x0188
+#define R128_PC_NGUI_CTLSTAT              0x0184
+#       define R128_PC_FLUSH_GUI          (3 << 0)
+#       define R128_PC_RI_GUI             (1 << 2)
+#       define R128_PC_FLUSH_ALL          0x00ff
+#       define R128_PC_BUSY               (1 << 31)
+#define R128_PC_NGUI_MODE                 0x0180
+#define R128_PCI_GART_PAGE                0x017c
+#define R128_PLANE_3D_MASK_C              0x1d44
+#define R128_PLL_TEST_CNTL                0x0013 /* PLL */
+#define R128_PMI_CAP_ID                   0x0f5c /* PCI */
+#define R128_PMI_DATA                     0x0f63 /* PCI */
+#define R128_PMI_NXT_CAP_PTR              0x0f5d /* PCI */
+#define R128_PMI_PMC_REG                  0x0f5e /* PCI */
+#define R128_PMI_PMCSR_REG                0x0f60 /* PCI */
+#define R128_PMI_REGISTER                 0x0f5c /* PCI */
+#define R128_PPLL_CNTL                    0x0002 /* PLL */
+#       define R128_PPLL_RESET                (1 <<  0)
+#       define R128_PPLL_SLEEP                (1 <<  1)
+#       define R128_PPLL_ATOMIC_UPDATE_EN     (1 << 16)
+#       define R128_PPLL_VGA_ATOMIC_UPDATE_EN (1 << 17)
+#define R128_PPLL_DIV_0                   0x0004 /* PLL */
+#define R128_PPLL_DIV_1                   0x0005 /* PLL */
+#define R128_PPLL_DIV_2                   0x0006 /* PLL */
+#define R128_PPLL_DIV_3                   0x0007 /* PLL */
+#       define R128_PPLL_FB3_DIV_MASK     0x07ff
+#       define R128_PPLL_POST3_DIV_MASK   0x00070000
+#define R128_PPLL_REF_DIV                 0x0003 /* PLL */
+#       define R128_PPLL_REF_DIV_MASK     0x03ff
+#       define R128_PPLL_ATOMIC_UPDATE_R  (1 << 15) /* same as _W */
+#       define R128_PPLL_ATOMIC_UPDATE_W  (1 << 15) /* same as _R */
+#define R128_PWR_MNGMT_CNTL_STATUS        0x0f60 /* PCI */
+#define R128_REG_BASE                     0x0f18 /* PCI */
+#define R128_REGPROG_INF                  0x0f09 /* PCI */
+#define R128_REVISION_ID                  0x0f08 /* PCI */
+
+#define R128_SC_BOTTOM                    0x164c
+#define R128_SC_BOTTOM_RIGHT              0x16f0
+#define R128_SC_BOTTOM_RIGHT_C            0x1c8c
+#define R128_SC_LEFT                      0x1640
+#define R128_SC_RIGHT                     0x1644
+#define R128_SC_TOP                       0x1648
+#define R128_SC_TOP_LEFT                  0x16ec
+#define R128_SC_TOP_LEFT_C                0x1c88
+#define R128_SEQ8_DATA                    0x03c5 /* VGA */
+#define R128_SEQ8_IDX                     0x03c4 /* VGA */
+#define R128_SNAPSHOT_F_COUNT             0x0244
+#define R128_SNAPSHOT_VH_COUNTS           0x0240
+#define R128_SNAPSHOT_VIF_COUNT           0x024c
+#define R128_SRC_OFFSET                   0x15ac
+#define R128_SRC_PITCH                    0x15b0
+#define R128_SRC_PITCH_OFFSET             0x1428
+#define R128_SRC_SC_BOTTOM                0x165c
+#define R128_SRC_SC_BOTTOM_RIGHT          0x16f4
+#define R128_SRC_SC_RIGHT                 0x1654
+#define R128_SRC_X                        0x1414
+#define R128_SRC_X_Y                      0x1590
+#define R128_SRC_Y                        0x1418
+#define R128_SRC_Y_X                      0x1434
+#define R128_STATUS                       0x0f06 /* PCI */
+#define R128_SUBPIC_CNTL                  0x0540 /* ? */
+#define R128_SUB_CLASS                    0x0f0a /* PCI */
+#define R128_SURFACE_DELAY                0x0b00
+#define R128_SURFACE0_INFO                0x0b0c
+#define R128_SURFACE0_LOWER_BOUND         0x0b04
+#define R128_SURFACE0_UPPER_BOUND         0x0b08
+#define R128_SURFACE1_INFO                0x0b1c
+#define R128_SURFACE1_LOWER_BOUND         0x0b14
+#define R128_SURFACE1_UPPER_BOUND         0x0b18
+#define R128_SURFACE2_INFO                0x0b2c
+#define R128_SURFACE2_LOWER_BOUND         0x0b24
+#define R128_SURFACE2_UPPER_BOUND         0x0b28
+#define R128_SURFACE3_INFO                0x0b3c
+#define R128_SURFACE3_LOWER_BOUND         0x0b34
+#define R128_SURFACE3_UPPER_BOUND         0x0b38
+#define R128_SW_SEMAPHORE                 0x013c
+
+#define R128_TEST_DEBUG_CNTL              0x0120
+#define R128_TEST_DEBUG_MUX               0x0124
+#define R128_TEST_DEBUG_OUT               0x012c
+#define R128_TMDS_CRC                     0x02a0
+#define R128_TMDS_TRANSMITTER_CNTL        0x02a4
+#       define R128_TMDS_PLLEN            (1 << 0)
+#       define R128_TMDS_PLLRST           (1 << 1)
+#define R128_TRAIL_BRES_DEC               0x1614
+#define R128_TRAIL_BRES_ERR               0x160c
+#define R128_TRAIL_BRES_INC               0x1610
+#define R128_TRAIL_X                      0x1618
+#define R128_TRAIL_X_SUB                  0x1620
+
+#define R128_VCLK_ECP_CNTL                0x0008 /* PLL */
+#define R128_VENDOR_ID                    0x0f00 /* PCI */
+#define R128_VGA_DDA_CONFIG               0x02e8
+#define R128_VGA_DDA_ON_OFF               0x02ec
+#define R128_VID_BUFFER_CONTROL           0x0900
+#define R128_VIDEOMUX_CNTL                0x0190
+#define R128_VIPH_CONTROL                 0x01D0 /* ? */
+
+#define R128_WAIT_UNTIL                   0x1720
+
+#define R128_X_MPLL_REF_FB_DIV            0x000a /* PLL */
+#define R128_XCLK_CNTL                    0x000d /* PLL */
+#define R128_XDLL_CNTL                    0x000c /* PLL */
+#define R128_XPLL_CNTL                    0x000b /* PLL */
+
+				/* Registers for CCE and Microcode Engine */
+#define R128_PM4_MICROCODE_ADDR           0x07d4
+#define R128_PM4_MICROCODE_RADDR          0x07d8
+#define R128_PM4_MICROCODE_DATAH          0x07dc
+#define R128_PM4_MICROCODE_DATAL          0x07e0
+
+#define R128_PM4_BUFFER_OFFSET            0x0700
+#define R128_PM4_BUFFER_CNTL              0x0704
+#       define R128_PM4_NONPM4                 (0  << 28)
+#       define R128_PM4_192PIO                 (1  << 28)
+#       define R128_PM4_192BM                  (2  << 28)
+#       define R128_PM4_128PIO_64INDBM         (3  << 28)
+#       define R128_PM4_128BM_64INDBM          (4  << 28)
+#       define R128_PM4_64PIO_128INDBM         (5  << 28)
+#       define R128_PM4_64BM_128INDBM          (6  << 28)
+#       define R128_PM4_64PIO_64VCBM_64INDBM   (7  << 28)
+#       define R128_PM4_64BM_64VCBM_64INDBM    (8  << 28)
+#       define R128_PM4_64PIO_64VCPIO_64INDPIO (15 << 28)
+#define R128_PM4_BUFFER_WM_CNTL           0x0708
+#       define R128_WMA_SHIFT                  0
+#       define R128_WMB_SHIFT                  8
+#       define R128_WMC_SHIFT                 16
+#       define R128_WB_WM_SHIFT               24
+#define R128_PM4_BUFFER_DL_RPTR_ADDR      0x070c
+#define R128_PM4_BUFFER_DL_RPTR           0x0710
+#define R128_PM4_BUFFER_DL_WPTR           0x0714
+#       define R128_PM4_BUFFER_DL_DONE    (1 << 31)
+#define R128_PM4_BUFFER_DL_WPTR_DELAY     0x0718
+#       define R128_PRE_WRITE_TIMER_SHIFT      0
+#       define R128_PRE_WRITE_LIMIT_SHIFT     23
+#define R128_PM4_VC_FPU_SETUP             0x071c
+#       define R128_FRONT_DIR_CW          (0 <<  0)
+#       define R128_FRONT_DIR_CCW         (1 <<  0)
+#       define R128_FRONT_DIR_MASK        (1 <<  0)
+#       define R128_BACKFACE_CULL         (0 <<  1)
+#       define R128_BACKFACE_POINTS       (1 <<  1)
+#       define R128_BACKFACE_LINES        (2 <<  1)
+#       define R128_BACKFACE_SOLID        (3 <<  1)
+#       define R128_BACKFACE_MASK         (3 <<  1)
+#       define R128_FRONTFACE_CULL        (0 <<  3)
+#       define R128_FRONTFACE_POINTS      (1 <<  3)
+#       define R128_FRONTFACE_LINES       (2 <<  3)
+#       define R128_FRONTFACE_SOLID       (3 <<  3)
+#       define R128_FRONTFACE_MASK        (3 <<  3)
+#       define R128_FPU_COLOR_SOLID       (0 <<  5)
+#       define R128_FPU_COLOR_FLAT        (1 <<  5)
+#       define R128_FPU_COLOR_GOURAUD     (2 <<  5)
+#       define R128_FPU_COLOR_GOURAUD2    (3 <<  5)
+#       define R128_FPU_COLOR_MASK        (3 <<  5)
+#       define R128_FPU_SUB_PIX_2BITS     (0 <<  7)
+#       define R128_FPU_SUB_PIX_4BITS     (1 <<  7)
+#       define R128_FPU_MODE_2D           (0 <<  8)
+#       define R128_FPU_MODE_3D           (1 <<  8)
+#       define R128_TRAP_BITS_DISABLE     (1 <<  9)
+#       define R128_EDGE_ANTIALIAS        (1 << 10)
+#       define R128_SUPERSAMPLE           (1 << 11)
+#       define R128_XFACTOR_2             (0 << 12)
+#       define R128_XFACTOR_4             (1 << 12)
+#       define R128_YFACTOR_2             (0 << 13)
+#       define R128_YFACTOR_4             (1 << 13)
+#       define R128_FLAT_SHADE_VERTEX_D3D (0 << 14)
+#       define R128_FLAT_SHADE_VERTEX_OGL (1 << 14)
+#       define R128_FPU_ROUND_TRUNCATE    (0 << 15)
+#       define R128_FPU_ROUND_NEAREST     (1 << 15)
+#       define R128_WM_SEL_8DW            (0 << 16)
+#       define R128_WM_SEL_16DW           (1 << 16)
+#       define R128_WM_SEL_32DW           (2 << 16)
+#define R128_PM4_VC_DEBUG_CONFIG          0x07a4
+#define R128_PM4_VC_STAT                  0x07a8
+#define R128_PM4_VC_TIMESTAMP0            0x07b0
+#define R128_PM4_VC_TIMESTAMP1            0x07b4
+#define R128_PM4_STAT                     0x07b8
+#       define R128_PM4_FIFOCNT_MASK      0x0fff
+#       define R128_PM4_BUSY              (1 << 16)
+#       define R128_PM4_GUI_ACTIVE        (1 << 31)
+#define R128_PM4_BUFFER_ADDR              0x07f0
+#define R128_PM4_MICRO_CNTL               0x07fc
+#       define R128_PM4_MICRO_FREERUN     (1 << 30)
+#define R128_PM4_FIFO_DATA_EVEN           0x1000
+#define R128_PM4_FIFO_DATA_ODD            0x1004
+
+#define R128_SCALE_3D_CNTL                0x1a00
+#       define R128_SCALE_DITHER_ERR_DIFF         (0  <<  1)
+#       define R128_SCALE_DITHER_TABLE            (1  <<  1)
+#       define R128_TEX_CACHE_SIZE_FULL           (0  <<  2)
+#       define R128_TEX_CACHE_SIZE_HALF           (1  <<  2)
+#       define R128_DITHER_INIT_CURR              (0  <<  3)
+#       define R128_DITHER_INIT_RESET             (1  <<  3)
+#       define R128_ROUND_24BIT                   (1  <<  4)
+#       define R128_TEX_CACHE_DISABLE             (1  <<  5)
+#       define R128_SCALE_3D_NOOP                 (0  <<  6)
+#       define R128_SCALE_3D_SCALE                (1  <<  6)
+#       define R128_SCALE_3D_TEXMAP_SHADE         (2  <<  6)
+#       define R128_SCALE_PIX_BLEND               (0  <<  8)
+#       define R128_SCALE_PIX_REPLICATE           (1  <<  8)
+#       define R128_TEX_CACHE_SPLIT               (1  <<  9)
+#       define R128_APPLE_YUV_MODE                (1  << 10)
+#       define R128_TEX_CACHE_PALLETE_MODE        (1  << 11)
+#       define R128_ALPHA_COMB_ADD_CLAMP          (0  << 12)
+#       define R128_ALPHA_COMB_ADD_NCLAMP         (1  << 12)
+#       define R128_ALPHA_COMB_SUB_DST_SRC_CLAMP  (2  << 12)
+#       define R128_ALPHA_COMB_SUB_DST_SRC_NCLAMP (3  << 12)
+#       define R128_FOG_TABLE                     (1  << 14)
+#       define R128_SIGNED_DST_CLAMP              (1  << 15)
+#       define R128_ALPHA_BLEND_SRC_ZERO          (0  << 16)
+#       define R128_ALPHA_BLEND_SRC_ONE           (1  << 16)
+#       define R128_ALPHA_BLEND_SRC_SRCCOLOR      (2  << 16)
+#       define R128_ALPHA_BLEND_SRC_INVSRCCOLOR   (3  << 16)
+#       define R128_ALPHA_BLEND_SRC_SRCALPHA      (4  << 16)
+#       define R128_ALPHA_BLEND_SRC_INVSRCALPHA   (5  << 16)
+#       define R128_ALPHA_BLEND_SRC_DSTALPHA      (6  << 16)
+#       define R128_ALPHA_BLEND_SRC_INVDSTALPHA   (7  << 16)
+#       define R128_ALPHA_BLEND_SRC_DSTCOLOR      (8  << 16)
+#       define R128_ALPHA_BLEND_SRC_INVDSTCOLOR   (9  << 16)
+#       define R128_ALPHA_BLEND_SRC_SAT           (10 << 16)
+#       define R128_ALPHA_BLEND_SRC_BLEND         (11 << 16)
+#       define R128_ALPHA_BLEND_SRC_INVBLEND      (12 << 16)
+#       define R128_ALPHA_BLEND_DST_ZERO          (0  << 20)
+#       define R128_ALPHA_BLEND_DST_ONE           (1  << 20)
+#       define R128_ALPHA_BLEND_DST_SRCCOLOR      (2  << 20)
+#       define R128_ALPHA_BLEND_DST_INVSRCCOLOR   (3  << 20)
+#       define R128_ALPHA_BLEND_DST_SRCALPHA      (4  << 20)
+#       define R128_ALPHA_BLEND_DST_INVSRCALPHA   (5  << 20)
+#       define R128_ALPHA_BLEND_DST_DSTALPHA      (6  << 20)
+#       define R128_ALPHA_BLEND_DST_INVDSTALPHA   (7  << 20)
+#       define R128_ALPHA_BLEND_DST_DSTCOLOR      (8  << 20)
+#       define R128_ALPHA_BLEND_DST_INVDSTCOLOR   (9  << 20)
+#       define R128_ALPHA_TEST_NEVER              (0  << 24)
+#       define R128_ALPHA_TEST_LESS               (1  << 24)
+#       define R128_ALPHA_TEST_LESSEQUAL          (2  << 24)
+#       define R128_ALPHA_TEST_EQUAL              (3  << 24)
+#       define R128_ALPHA_TEST_GREATEREQUAL       (4  << 24)
+#       define R128_ALPHA_TEST_GREATER            (5  << 24)
+#       define R128_ALPHA_TEST_NEQUAL             (6  << 24)
+#       define R128_ALPHA_TEST_ALWAYS             (7  << 24)
+#       define R128_COMPOSITE_SHADOW_CMP_EQUAL    (0  << 28)
+#       define R128_COMPOSITE_SHADOW_CMP_NEQUAL   (1  << 28)
+#       define R128_COMPOSITE_SHADOW              (1  << 29)
+#       define R128_TEX_MAP_ALPHA_IN_TEXTURE      (1  << 30)
+#       define R128_TEX_CACHE_LINE_SIZE_8QW       (0  << 31)
+#       define R128_TEX_CACHE_LINE_SIZE_4QW       (1  << 31)
+#define R128_SCALE_3D_DATATYPE            0x1a20
+
+#define R128_SETUP_CNTL                   0x1bc4
+#       define R128_DONT_START_TRIANGLE   (1 <<  0)
+#       define R128_Z_BIAS                (0 <<  1)
+#       define R128_DONT_START_ANY_ON     (1 <<  2)
+#       define R128_COLOR_SOLID_COLOR     (0 <<  3)
+#       define R128_COLOR_FLAT_VERT_1     (1 <<  3)
+#       define R128_COLOR_FLAT_VERT_2     (2 <<  3)
+#       define R128_COLOR_FLAT_VERT_3     (3 <<  3)
+#       define R128_COLOR_GOURAUD         (4 <<  3)
+#       define R128_PRIM_TYPE_TRI         (0 <<  7)
+#       define R128_PRIM_TYPE_LINE        (1 <<  7)
+#       define R128_PRIM_TYPE_POINT       (2 <<  7)
+#       define R128_PRIM_TYPE_POLY_EDGE   (3 <<  7)
+#       define R128_TEXTURE_ST_MULT_W     (0 <<  9)
+#       define R128_TEXTURE_ST_DIRECT     (1 <<  9)
+#       define R128_STARTING_VERTEX_1     (1 << 14)
+#       define R128_STARTING_VERTEX_2     (2 << 14)
+#       define R128_STARTING_VERTEX_3     (3 << 14)
+#       define R128_ENDING_VERTEX_1       (1 << 16)
+#       define R128_ENDING_VERTEX_2       (2 << 16)
+#       define R128_ENDING_VERTEX_3       (3 << 16)
+#       define R128_SU_POLY_LINE_LAST     (0 << 18)
+#       define R128_SU_POLY_LINE_NOT_LAST (1 << 18)
+#       define R128_SUB_PIX_2BITS         (0 << 19)
+#       define R128_SUB_PIX_4BITS         (1 << 19)
+#       define R128_SET_UP_CONTINUE       (1 << 31)
+
+#define R128_WINDOW_XY_OFFSET             0x1bcc
+#       define R128_WINDOW_Y_SHIFT        4
+#       define R128_WINDOW_X_SHIFT        20
+
+#define R128_Z_OFFSET_C                   0x1c90
+#define R128_Z_PITCH_C                    0x1c94
+#       define R128_Z_TILE                    (1 << 16)
+#define R128_Z_STEN_CNTL_C                0x1c98
+#       define R128_Z_PIX_WIDTH_16            (0 <<  1)
+#       define R128_Z_PIX_WIDTH_24            (1 <<  1)
+#       define R128_Z_PIX_WIDTH_32            (2 <<  1)
+#       define R128_Z_PIX_WIDTH_MASK          (3 <<  1)
+#       define R128_Z_TEST_NEVER              (0 <<  4)
+#       define R128_Z_TEST_LESS               (1 <<  4)
+#       define R128_Z_TEST_LESSEQUAL          (2 <<  4)
+#       define R128_Z_TEST_EQUAL              (3 <<  4)
+#       define R128_Z_TEST_GREATEREQUAL       (4 <<  4)
+#       define R128_Z_TEST_GREATER            (5 <<  4)
+#       define R128_Z_TEST_NEQUAL             (6 <<  4)
+#       define R128_Z_TEST_ALWAYS             (7 <<  4)
+#       define R128_Z_TEST_MASK               (7 <<  4)
+#       define R128_STENCIL_TEST_NEVER        (0 << 12)
+#       define R128_STENCIL_TEST_LESS         (1 << 12)
+#       define R128_STENCIL_TEST_LESSEQUAL    (2 << 12)
+#       define R128_STENCIL_TEST_EQUAL        (3 << 12)
+#       define R128_STENCIL_TEST_GREATEREQUAL (4 << 12)
+#       define R128_STENCIL_TEST_GREATER      (5 << 12)
+#       define R128_STENCIL_TEST_NEQUAL       (6 << 12)
+#       define R128_STENCIL_TEST_ALWAYS       (7 << 12)
+#       define R128_STENCIL_S_FAIL_KEEP       (0 << 16)
+#       define R128_STENCIL_S_FAIL_ZERO       (1 << 16)
+#       define R128_STENCIL_S_FAIL_REPLACE    (2 << 16)
+#       define R128_STENCIL_S_FAIL_INC        (3 << 16)
+#       define R128_STENCIL_S_FAIL_DEC        (4 << 16)
+#       define R128_STENCIL_S_FAIL_INV        (5 << 16)
+#       define R128_STENCIL_ZPASS_KEEP        (0 << 20)
+#       define R128_STENCIL_ZPASS_ZERO        (1 << 20)
+#       define R128_STENCIL_ZPASS_REPLACE     (2 << 20)
+#       define R128_STENCIL_ZPASS_INC         (3 << 20)
+#       define R128_STENCIL_ZPASS_DEC         (4 << 20)
+#       define R128_STENCIL_ZPASS_INV         (5 << 20)
+#       define R128_STENCIL_ZFAIL_KEEP        (0 << 24)
+#       define R128_STENCIL_ZFAIL_ZERO        (1 << 24)
+#       define R128_STENCIL_ZFAIL_REPLACE     (2 << 24)
+#       define R128_STENCIL_ZFAIL_INC         (3 << 24)
+#       define R128_STENCIL_ZFAIL_DEC         (4 << 24)
+#       define R128_STENCIL_ZFAIL_INV         (5 << 24)
+#define R128_TEX_CNTL_C                   0x1c9c
+#       define R128_Z_ENABLE                   (1 <<  0)
+#       define R128_Z_WRITE_ENABLE             (1 <<  1)
+#       define R128_STENCIL_ENABLE             (1 <<  3)
+#       define R128_SHADE_ENABLE               (0 <<  4)
+#       define R128_TEXMAP_ENABLE              (1 <<  4)
+#       define R128_SEC_TEXMAP_ENABLE          (1 <<  5)
+#       define R128_FOG_ENABLE                 (1 <<  7)
+#       define R128_DITHER_ENABLE              (1 <<  8)
+#       define R128_ALPHA_ENABLE               (1 <<  9)
+#       define R128_ALPHA_TEST_ENABLE          (1 << 10)
+#       define R128_SPEC_LIGHT_ENABLE          (1 << 11)
+#       define R128_TEX_CHROMA_KEY_ENABLE      (1 << 12)
+#       define R128_ALPHA_IN_TEX_COMPLETE_A    (0 << 13)
+#       define R128_ALPHA_IN_TEX_LSB_A         (1 << 13)
+#       define R128_LIGHT_DIS                  (0 << 14)
+#       define R128_LIGHT_COPY                 (1 << 14)
+#       define R128_LIGHT_MODULATE             (2 << 14)
+#       define R128_LIGHT_ADD                  (3 << 14)
+#       define R128_LIGHT_BLEND_CONSTANT       (4 << 14)
+#       define R128_LIGHT_BLEND_TEXTURE        (5 << 14)
+#       define R128_LIGHT_BLEND_VERTEX         (6 << 14)
+#       define R128_LIGHT_BLEND_CONST_COLOR    (7 << 14)
+#       define R128_ALPHA_LIGHT_DIS            (0 << 18)
+#       define R128_ALPHA_LIGHT_COPY           (1 << 18)
+#       define R128_ALPHA_LIGHT_MODULATE       (2 << 18)
+#       define R128_ALPHA_LIGHT_ADD            (3 << 18)
+#       define R128_ANTI_ALIAS                 (1 << 21)
+#       define R128_TEX_CACHE_FLUSH            (1 << 23)
+#       define R128_LOD_BIAS_SHIFT             24
+#       define R128_LOD_BIAS_MASK              (0xff << 24)
+#define R128_MISC_3D_STATE_CNTL_REG       0x1ca0
+#       define R128_REF_ALPHA_MASK                  0xff
+#       define R128_MISC_SCALE_3D_NOOP              (0  <<  8)
+#       define R128_MISC_SCALE_3D_SCALE             (1  <<  8)
+#       define R128_MISC_SCALE_3D_TEXMAP_SHADE      (2  <<  8)
+#       define R128_MISC_SCALE_PIX_BLEND            (0  << 10)
+#       define R128_MISC_SCALE_PIX_REPLICATE        (1  << 10)
+#       define R128_ALPHA_COMB_ADD_CLAMP            (0  << 12)
+#       define R128_ALPHA_COMB_ADD_NO_CLAMP         (1  << 12)
+#       define R128_ALPHA_COMB_SUB_SRC_DST_CLAMP    (2  << 12)
+#       define R128_ALPHA_COMB_SUB_SRC_DST_NO_CLAMP (3  << 12)
+#       define R128_FOG_VERTEX                      (0  << 14)
+#       define R128_FOG_TABLE                       (1  << 14)
+#       define R128_ALPHA_BLEND_SRC_ZERO            (0  << 16)
+#       define R128_ALPHA_BLEND_SRC_ONE             (1  << 16)
+#       define R128_ALPHA_BLEND_SRC_SRCCOLOR        (2  << 16)
+#       define R128_ALPHA_BLEND_SRC_INVSRCCOLOR     (3  << 16)
+#       define R128_ALPHA_BLEND_SRC_SRCALPHA        (4  << 16)
+#       define R128_ALPHA_BLEND_SRC_INVSRCALPHA     (5  << 16)
+#       define R128_ALPHA_BLEND_SRC_DESTALPHA       (6  << 16)
+#       define R128_ALPHA_BLEND_SRC_INVDESTALPHA    (7  << 16)
+#       define R128_ALPHA_BLEND_SRC_DESTCOLOR       (8  << 16)
+#       define R128_ALPHA_BLEND_SRC_INVDESTCOLOR    (9  << 16)
+#       define R128_ALPHA_BLEND_SRC_SRCALPHASAT     (10 << 16)
+#       define R128_ALPHA_BLEND_SRC_BOTHSRCALPHA    (11 << 16)
+#       define R128_ALPHA_BLEND_SRC_BOTHINVSRCALPHA (12 << 16)
+#       define R128_ALPHA_BLEND_SRC_MASK            (15 << 16)
+#       define R128_ALPHA_BLEND_DST_ZERO            (0  << 20)
+#       define R128_ALPHA_BLEND_DST_ONE             (1  << 20)
+#       define R128_ALPHA_BLEND_DST_SRCCOLOR        (2  << 20)
+#       define R128_ALPHA_BLEND_DST_INVSRCCOLOR     (3  << 20)
+#       define R128_ALPHA_BLEND_DST_SRCALPHA        (4  << 20)
+#       define R128_ALPHA_BLEND_DST_INVSRCALPHA     (5  << 20)
+#       define R128_ALPHA_BLEND_DST_DESTALPHA       (6  << 20)
+#       define R128_ALPHA_BLEND_DST_INVDESTALPHA    (7  << 20)
+#       define R128_ALPHA_BLEND_DST_DESTCOLOR       (8  << 20)
+#       define R128_ALPHA_BLEND_DST_INVDESTCOLOR    (9  << 20)
+#       define R128_ALPHA_BLEND_DST_SRCALPHASAT     (10 << 20)
+#       define R128_ALPHA_BLEND_DST_MASK            (15 << 20)
+#       define R128_ALPHA_TEST_NEVER                (0  << 24)
+#       define R128_ALPHA_TEST_LESS                 (1  << 24)
+#       define R128_ALPHA_TEST_LESSEQUAL            (2  << 24)
+#       define R128_ALPHA_TEST_EQUAL                (3  << 24)
+#       define R128_ALPHA_TEST_GREATEREQUAL         (4  << 24)
+#       define R128_ALPHA_TEST_GREATER              (5  << 24)
+#       define R128_ALPHA_TEST_NEQUAL               (6  << 24)
+#       define R128_ALPHA_TEST_ALWAYS               (7  << 24)
+#       define R128_ALPHA_TEST_MASK                 (7  << 24)
+#define R128_TEXTURE_CLR_CMP_CLR_C        0x1ca4
+#define R128_TEXTURE_CLR_CMP_MSK_C        0x1ca8
+#define R128_FOG_COLOR_C                  0x1cac
+#       define R128_FOG_BLUE_SHIFT             0
+#       define R128_FOG_GREEN_SHIFT            8
+#       define R128_FOG_RED_SHIFT             16
+#define R128_PRIM_TEX_CNTL_C              0x1cb0
+#       define R128_MIN_BLEND_NEAREST          (0  <<  1)
+#       define R128_MIN_BLEND_LINEAR           (1  <<  1)
+#       define R128_MIN_BLEND_MIPNEAREST       (2  <<  1)
+#       define R128_MIN_BLEND_MIPLINEAR        (3  <<  1)
+#       define R128_MIN_BLEND_LINEARMIPNEAREST (4  <<  1)
+#       define R128_MIN_BLEND_LINEARMIPLINEAR  (5  <<  1)
+#       define R128_MIN_BLEND_MASK             (7  <<  1)
+#       define R128_MAG_BLEND_NEAREST          (0  <<  4)
+#       define R128_MAG_BLEND_LINEAR           (1  <<  4)
+#       define R128_MAG_BLEND_MASK             (7  <<  4)
+#       define R128_MIP_MAP_DISABLE            (1  <<  7)
+#       define R128_TEX_CLAMP_S_WRAP           (0  <<  8)
+#       define R128_TEX_CLAMP_S_MIRROR         (1  <<  8)
+#       define R128_TEX_CLAMP_S_CLAMP          (2  <<  8)
+#       define R128_TEX_CLAMP_S_BORDER_COLOR   (3  <<  8)
+#       define R128_TEX_CLAMP_S_MASK           (3  <<  8)
+#       define R128_TEX_WRAP_S                 (1  << 10)
+#       define R128_TEX_CLAMP_T_WRAP           (0  << 11)
+#       define R128_TEX_CLAMP_T_MIRROR         (1  << 11)
+#       define R128_TEX_CLAMP_T_CLAMP          (2  << 11)
+#       define R128_TEX_CLAMP_T_BORDER_COLOR   (3  << 11)
+#       define R128_TEX_CLAMP_T_MASK           (3  << 11)
+#       define R128_TEX_WRAP_T                 (1  << 13)
+#       define R128_TEX_PERSPECTIVE_DISABLE    (1  << 14)
+#       define R128_DATATYPE_VQ                (0  << 16)
+#       define R128_DATATYPE_CI4               (1  << 16)
+#       define R128_DATATYPE_CI8               (2  << 16)
+#       define R128_DATATYPE_ARGB1555          (3  << 16)
+#       define R128_DATATYPE_RGB565            (4  << 16)
+#       define R128_DATATYPE_RGB888            (5  << 16)
+#       define R128_DATATYPE_ARGB8888          (6  << 16)
+#       define R128_DATATYPE_RGB332            (7  << 16)
+#       define R128_DATATYPE_Y8                (8  << 16)
+#       define R128_DATATYPE_RGB8              (9  << 16)
+#       define R128_DATATYPE_CI16              (10 << 16)
+#       define R128_DATATYPE_YVYU422           (11 << 16)
+#       define R128_DATATYPE_VYUY422           (12 << 16)
+#       define R128_DATATYPE_AYUV444           (14 << 16)
+#       define R128_DATATYPE_ARGB4444          (15 << 16)
+#       define R128_PALLETE_EITHER             (0  << 20)
+#       define R128_PALLETE_1                  (1  << 20)
+#       define R128_PALLETE_2                  (2  << 20)
+#       define R128_PSEUDOCOLOR_DT_RGB565      (0  << 24)
+#       define R128_PSEUDOCOLOR_DT_ARGB1555    (1  << 24)
+#       define R128_PSEUDOCOLOR_DT_ARGB4444    (2  << 24)
+#define R128_PRIM_TEXTURE_COMBINE_CNTL_C  0x1cb4
+#       define R128_COMB_DIS                   (0  <<  0)
+#       define R128_COMB_COPY                  (1  <<  0)
+#       define R128_COMB_COPY_INP              (2  <<  0)
+#       define R128_COMB_MODULATE              (3  <<  0)
+#       define R128_COMB_MODULATE2X            (4  <<  0)
+#       define R128_COMB_MODULATE4X            (5  <<  0)
+#       define R128_COMB_ADD                   (6  <<  0)
+#       define R128_COMB_ADD_SIGNED            (7  <<  0)
+#       define R128_COMB_BLEND_VERTEX          (8  <<  0)
+#       define R128_COMB_BLEND_TEXTURE         (9  <<  0)
+#       define R128_COMB_BLEND_CONST           (10 <<  0)
+#       define R128_COMB_BLEND_PREMULT         (11 <<  0)
+#       define R128_COMB_BLEND_PREV            (12 <<  0)
+#       define R128_COMB_BLEND_PREMULT_INV     (13 <<  0)
+#       define R128_COMB_ADD_SIGNED2X          (14 <<  0)
+#       define R128_COMB_BLEND_CONST_COLOR     (15 <<  0)
+#       define R128_COMB_MASK                  (15 <<  0)
+#       define R128_COLOR_FACTOR_CONST_COLOR   (0  <<  4)
+#       define R128_COLOR_FACTOR_NCONST_COLOR  (1  <<  4)
+#       define R128_COLOR_FACTOR_TEX           (4  <<  4)
+#       define R128_COLOR_FACTOR_NTEX          (5  <<  4)
+#       define R128_COLOR_FACTOR_ALPHA         (6  <<  4)
+#       define R128_COLOR_FACTOR_NALPHA        (7  <<  4)
+#       define R128_COLOR_FACTOR_PREV_COLOR    (8  <<  4)
+#       define R128_COLOR_FACTOR_MASK          (15 <<  4)
+#       define R128_COMB_FCN_MSB               (1  <<  8)
+#       define R128_INPUT_FACTOR_CONST_COLOR   (2  << 10)
+#       define R128_INPUT_FACTOR_CONST_ALPHA   (3  << 10)
+#       define R128_INPUT_FACTOR_INT_COLOR     (4  << 10)
+#       define R128_INPUT_FACTOR_INT_ALPHA     (5  << 10)
+#       define R128_INPUT_FACTOR_MASK          (15 << 10)
+#       define R128_COMB_ALPHA_DIS             (0  << 14)
+#       define R128_COMB_ALPHA_COPY            (1  << 14)
+#       define R128_COMB_ALPHA_COPY_INP        (2  << 14)
+#       define R128_COMB_ALPHA_MODULATE        (3  << 14)
+#       define R128_COMB_ALPHA_MODULATE2X      (4  << 14)
+#       define R128_COMB_ALPHA_MODULATE4X      (5  << 14)
+#       define R128_COMB_ALPHA_ADD             (6  << 14)
+#       define R128_COMB_ALPHA_ADD_SIGNED      (7  << 14)
+#       define R128_COMB_ALPHA_ADD_SIGNED2X    (14 << 14)
+#       define R128_COMB_ALPHA_MASK            (15 << 14)
+#       define R128_ALPHA_FACTOR_TEX_ALPHA     (6  << 18)
+#       define R128_ALPHA_FACTOR_NTEX_ALPHA    (7  << 18)
+#       define R128_ALPHA_FACTOR_MASK          (15 << 18)
+#       define R128_INP_FACTOR_A_CONST_ALPHA   (1  << 25)
+#       define R128_INP_FACTOR_A_INT_ALPHA     (2  << 25)
+#       define R128_INP_FACTOR_A_MASK          (7  << 25)
+#define R128_TEX_SIZE_PITCH_C             0x1cb8
+#       define R128_TEX_PITCH_SHIFT           0
+#       define R128_TEX_SIZE_SHIFT            4
+#       define R128_TEX_HEIGHT_SHIFT          8
+#       define R128_TEX_MIN_SIZE_SHIFT       12
+#       define R128_SEC_TEX_PITCH_SHIFT      16
+#       define R128_SEC_TEX_SIZE_SHIFT       20
+#       define R128_SEC_TEX_HEIGHT_SHIFT     24
+#       define R128_SEC_TEX_MIN_SIZE_SHIFT   28
+#       define R128_TEX_PITCH_MASK           (0x0f <<  0)
+#       define R128_TEX_SIZE_MASK            (0x0f <<  4)
+#       define R128_TEX_HEIGHT_MASK          (0x0f <<  8)
+#       define R128_TEX_MIN_SIZE_MASK        (0x0f << 12)
+#       define R128_SEC_TEX_PITCH_MASK       (0x0f << 16)
+#       define R128_SEC_TEX_SIZE_MASK        (0x0f << 20)
+#       define R128_SEC_TEX_HEIGHT_MASK      (0x0f << 24)
+#       define R128_SEC_TEX_MIN_SIZE_MASK    (0x0f << 28)
+#       define R128_TEX_SIZE_PITCH_SHIFT      0
+#       define R128_SEC_TEX_SIZE_PITCH_SHIFT 16
+#       define R128_TEX_SIZE_PITCH_MASK      (0xffff <<  0)
+#       define R128_SEC_TEX_SIZE_PITCH_MASK  (0xffff << 16)
+#define R128_PRIM_TEX_0_OFFSET_C          0x1cbc
+#define R128_PRIM_TEX_1_OFFSET_C          0x1cc0
+#define R128_PRIM_TEX_2_OFFSET_C          0x1cc4
+#define R128_PRIM_TEX_3_OFFSET_C          0x1cc8
+#define R128_PRIM_TEX_4_OFFSET_C          0x1ccc
+#define R128_PRIM_TEX_5_OFFSET_C          0x1cd0
+#define R128_PRIM_TEX_6_OFFSET_C          0x1cd4
+#define R128_PRIM_TEX_7_OFFSET_C          0x1cd8
+#define R128_PRIM_TEX_8_OFFSET_C          0x1cdc
+#define R128_PRIM_TEX_9_OFFSET_C          0x1ce0
+#define R128_PRIM_TEX_10_OFFSET_C         0x1ce4
+#       define R128_TEX_NO_TILE           (0 << 30)
+#       define R128_TEX_TILED_BY_HOST     (1 << 30)
+#       define R128_TEX_TILED_BY_STORAGE  (2 << 30)
+#       define R128_TEX_TILED_BY_STORAGE2 (3 << 30)
+
+#define R128_SEC_TEX_CNTL_C               0x1d00
+#       define R128_SEC_SELECT_PRIM_ST    (0  <<  0)
+#       define R128_SEC_SELECT_SEC_ST     (1  <<  0)
+#define R128_SEC_TEX_COMBINE_CNTL_C       0x1d04
+#       define R128_INPUT_FACTOR_PREV_COLOR (8  << 10)
+#       define R128_INPUT_FACTOR_PREV_ALPHA (9  << 10)
+#       define R128_INP_FACTOR_A_PREV_ALPHA (4  << 25)
+#define R128_SEC_TEX_0_OFFSET_C           0x1d08
+#define R128_SEC_TEX_1_OFFSET_C           0x1d0c
+#define R128_SEC_TEX_2_OFFSET_C           0x1d10
+#define R128_SEC_TEX_3_OFFSET_C           0x1d14
+#define R128_SEC_TEX_4_OFFSET_C           0x1d18
+#define R128_SEC_TEX_5_OFFSET_C           0x1d1c
+#define R128_SEC_TEX_6_OFFSET_C           0x1d20
+#define R128_SEC_TEX_7_OFFSET_C           0x1d24
+#define R128_SEC_TEX_8_OFFSET_C           0x1d28
+#define R128_SEC_TEX_9_OFFSET_C           0x1d2c
+#define R128_SEC_TEX_10_OFFSET_C          0x1d30
+#define R128_CONSTANT_COLOR_C             0x1d34
+#       define R128_CONSTANT_BLUE_SHIFT        0
+#       define R128_CONSTANT_GREEN_SHIFT       8
+#       define R128_CONSTANT_RED_SHIFT        16
+#       define R128_CONSTANT_ALPHA_SHIFT      24
+#define R128_PRIM_TEXTURE_BORDER_COLOR_C  0x1d38
+#       define R128_PRIM_TEX_BORDER_BLUE_SHIFT   0
+#       define R128_PRIM_TEX_BORDER_GREEN_SHIFT  8
+#       define R128_PRIM_TEX_BORDER_RED_SHIFT   16
+#       define R128_PRIM_TEX_BORDER_ALPHA_SHIFT 24
+#define R128_SEC_TEXTURE_BORDER_COLOR_C   0x1d3c
+#       define R128_SEC_TEX_BORDER_BLUE_SHIFT   0
+#       define R128_SEC_TEX_BORDER_GREEN_SHIFT  8
+#       define R128_SEC_TEX_BORDER_RED_SHIFT   16
+#       define R128_SEC_TEX_BORDER_ALPHA_SHIFT 24
+#define R128_STEN_REF_MASK_C              0x1d40
+#       define R128_STEN_REFERENCE_SHIFT       0
+#       define R128_STEN_MASK_SHIFT           16
+#       define R128_STEN_WRITE_MASK_SHIFT     24
+#define R128_PLANE_3D_MASK_C              0x1d44
+#define R128_TEX_CACHE_STAT_COUNT         0x1974
+
+
+				/* Constants */
+#define R128_AGP_TEX_OFFSET               0x02000000
+
+#define R128_LAST_FRAME_REG               R128_GUI_SCRATCH_REG0
+
+				/* CCE packet types */
+#define R128_CCE_PACKET0                         0x00000000
+#define R128_CCE_PACKET0_ONE_REG_WR              0x00008000
+#define R128_CCE_PACKET1                         0x40000000
+#define R128_CCE_PACKET2                         0x80000000
+#define R128_CCE_PACKET3                         0xC0000000
+#define R128_CCE_PACKET3_NOP                     0xC0001000
+#define R128_CCE_PACKET3_PAINT                   0xC0001100
+#define R128_CCE_PACKET3_BITBLT                  0xC0001200
+#define R128_CCE_PACKET3_SMALLTEXT               0xC0001300
+#define R128_CCE_PACKET3_HOSTDATA_BLT            0xC0001400
+#define R128_CCE_PACKET3_POLYLINE                0xC0001500
+#define R128_CCE_PACKET3_SCALING                 0xC0001600
+#define R128_CCE_PACKET3_TRANS_SCALING           0xC0001700
+#define R128_CCE_PACKET3_POLYSCANLINES           0xC0001800
+#define R128_CCE_PACKET3_NEXT_CHAR               0xC0001900
+#define R128_CCE_PACKET3_PAINT_MULTI             0xC0001A00
+#define R128_CCE_PACKET3_BITBLT_MULTI            0xC0001B00
+#define R128_CCE_PACKET3_PLY_NEXTSCAN            0xC0001D00
+#define R128_CCE_PACKET3_SET_SCISSORS            0xC0001E00
+#define R128_CCE_PACKET3_SET_MODE24BPP           0xC0001F00
+#define R128_CCE_PACKET3_CNTL_PAINT              0xC0009100
+#define R128_CCE_PACKET3_CNTL_BITBLT             0xC0009200
+#define R128_CCE_PACKET3_CNTL_SMALLTEXT          0xC0009300
+#define R128_CCE_PACKET3_CNTL_HOSTDATA_BLT       0xC0009400
+#define R128_CCE_PACKET3_CNTL_POLYLINE           0xC0009500
+#define R128_CCE_PACKET3_CNTL_SCALING            0xC0009600
+#define R128_CCE_PACKET3_CNTL_TRANS_SCALING      0xC0009700
+#define R128_CCE_PACKET3_CNTL_POLYSCANLINES      0xC0009800
+#define R128_CCE_PACKET3_CNTL_NEXT_CHAR          0xC0009900
+#define R128_CCE_PACKET3_CNTL_PAINT_MULTI        0xC0009A00
+#define R128_CCE_PACKET3_CNTL_BITBLT_MULTI       0xC0009B00
+#define R128_CCE_PACKET3_CNTL_TRANS_BITBLT       0xC0009C00
+#define R128_CCE_PACKET3_3D_SAVE_CONTEXT         0xC0002000
+#define R128_CCE_PACKET3_3D_PLAY_CONTEXT         0xC0002100
+#define R128_CCE_PACKET3_3D_RNDR_GEN_INDX_PRIM   0xC0002300
+#define R128_CCE_PACKET3_3D_RNDR_GEN_PRIM        0xC0002500
+#define R128_CCE_PACKET3_LOAD_PALETTE            0xC0002C00
+#define R128_CCE_PACKET3_PURGE                   0xC0002D00
+#define R128_CCE_PACKET3_NEXT_VERTEX_BUNDLE      0xC0002E00
+#       define R128_CCE_PACKET_MASK              0xC0000000
+#       define R128_CCE_PACKET_COUNT_MASK        0x3fff0000
+#       define R128_CCE_PACKET_MAX_DWORDS        (1 << 12)
+#       define R128_CCE_PACKET0_REG_MASK         0x000007ff
+#       define R128_CCE_PACKET1_REG0_MASK        0x000007ff
+#       define R128_CCE_PACKET1_REG1_MASK        0x003ff800
+
+#define R128_CCE_VC_FRMT_RHW                     0x00000001
+#define R128_CCE_VC_FRMT_DIFFUSE_BGR             0x00000002
+#define R128_CCE_VC_FRMT_DIFFUSE_A               0x00000004
+#define R128_CCE_VC_FRMT_DIFFUSE_ARGB            0x00000008
+#define R128_CCE_VC_FRMT_SPEC_BGR                0x00000010
+#define R128_CCE_VC_FRMT_SPEC_F                  0x00000020
+#define R128_CCE_VC_FRMT_SPEC_FRGB               0x00000040
+#define R128_CCE_VC_FRMT_S_T                     0x00000080
+#define R128_CCE_VC_FRMT_S2_T2                   0x00000100
+#define R128_CCE_VC_FRMT_RHW2                    0x00000200
+
+#define R128_CCE_VC_CNTL_PRIM_TYPE_NONE          0x00000000
+#define R128_CCE_VC_CNTL_PRIM_TYPE_POINT         0x00000001
+#define R128_CCE_VC_CNTL_PRIM_TYPE_LINE          0x00000002
+#define R128_CCE_VC_CNTL_PRIM_TYPE_POLY_LINE     0x00000003
+#define R128_CCE_VC_CNTL_PRIM_TYPE_TRI_LIST      0x00000004
+#define R128_CCE_VC_CNTL_PRIM_TYPE_TRI_FAN       0x00000005
+#define R128_CCE_VC_CNTL_PRIM_TYPE_TRI_STRIP     0x00000006
+#define R128_CCE_VC_CNTL_PRIM_TYPE_TRI_TYPE2     0x00000007
+#define R128_CCE_VC_CNTL_PRIM_WALK_IND           0x00000010
+#define R128_CCE_VC_CNTL_PRIM_WALK_LIST          0x00000020
+#define R128_CCE_VC_CNTL_PRIM_WALK_RING          0x00000030
+#define R128_CCE_VC_CNTL_NUM_SHIFT               16
+
+/* hmm copyed blindly (no specs) from radeon.h ... */
+#define R128_RE_TOP_LEFT                  0x26c0
+#       define R128_RE_LEFT_SHIFT         0
+#       define R128_RE_TOP_SHIFT          16
+#define R128_RE_WIDTH_HEIGHT              0x1c44
+#       define R128_RE_WIDTH_SHIFT        0
+#       define R128_RE_HEIGHT_SHIFT       16
+
+#endif
diff --git a/src/mesa/drivers/dri/r128/server/r128_sarea.h b/src/mesa/drivers/dri/r128/server/r128_sarea.h
new file mode 100644
index 00000000000..8a9f3a41769
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/server/r128_sarea.h
@@ -0,0 +1,195 @@
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/ati/r128_sarea.h,v 1.7 2002/02/16 21:26:35 herrb Exp $ */
+/*
+ * Copyright 1999, 2000 ATI Technologies Inc., Markham, Ontario,
+ *                      Precision Insight, Inc., Cedar Park, Texas, and
+ *                      VA Linux Systems Inc., Fremont, California.
+ *
+ * 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, 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
+ * NON-INFRINGEMENT.  IN NO EVENT SHALL ATI, PRECISION INSIGHT, VA LINUX
+ * SYSTEMS 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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ *
+ */
+
+#ifndef _R128_SAREA_H_
+#define _R128_SAREA_H_
+
+/* WARNING: If you change any of these defines, make sure to change the
+ * defines in the kernel file (r128_drm.h)
+ */
+#ifndef __R128_SAREA_DEFINES__
+#define __R128_SAREA_DEFINES__
+
+/* What needs to be changed for the current vertex buffer?
+ */
+#define R128_UPLOAD_CONTEXT		0x001
+#define R128_UPLOAD_SETUP		0x002
+#define R128_UPLOAD_TEX0		0x004
+#define R128_UPLOAD_TEX1		0x008
+#define R128_UPLOAD_TEX0IMAGES		0x010
+#define R128_UPLOAD_TEX1IMAGES		0x020
+#define R128_UPLOAD_CORE		0x040
+#define R128_UPLOAD_MASKS		0x080
+#define R128_UPLOAD_WINDOW		0x100
+#define R128_UPLOAD_CLIPRECTS		0x200	/* handled client-side */
+#define R128_REQUIRE_QUIESCENCE		0x400
+#define R128_UPLOAD_ALL			0x7ff
+
+#define R128_FRONT			0x1
+#define R128_BACK			0x2
+#define R128_DEPTH			0x4
+
+/* Primitive types
+ */
+#define R128_POINTS			0x1
+#define R128_LINES			0x2
+#define R128_LINE_STRIP			0x3
+#define R128_TRIANGLES			0x4
+#define R128_TRIANGLE_FAN		0x5
+#define R128_TRIANGLE_STRIP		0x6
+
+/* Vertex/indirect buffer size
+ */
+#define R128_BUFFER_SIZE		16384
+
+/* Byte offsets for indirect buffer data
+ */
+#define R128_INDEX_PRIM_OFFSET		20
+#define R128_HOSTDATA_BLIT_OFFSET	32
+
+/* Keep these small for testing
+ */
+#define R128_NR_SAREA_CLIPRECTS		12
+
+/* There are 2 heaps (local/AGP).  Each region within a heap is a
+ * minimum of 64k, and there are at most 64 of them per heap.
+ */
+#define R128_CARD_HEAP			0
+#define R128_AGP_HEAP			1
+#define R128_NR_TEX_HEAPS		2
+#define R128_NR_TEX_REGIONS		64
+#define R128_LOG_TEX_GRANULARITY	16
+
+#define R128_NR_CONTEXT_REGS		12
+
+#define R128_MAX_TEXTURE_LEVELS		11
+#define R128_MAX_TEXTURE_UNITS		2
+
+#endif /* __R128_SAREA_DEFINES__ */
+
+typedef struct {
+    /* Context state - can be written in one large chunk */
+    unsigned int dst_pitch_offset_c;
+    unsigned int dp_gui_master_cntl_c;
+    unsigned int sc_top_left_c;
+    unsigned int sc_bottom_right_c;
+    unsigned int z_offset_c;
+    unsigned int z_pitch_c;
+    unsigned int z_sten_cntl_c;
+    unsigned int tex_cntl_c;
+    unsigned int misc_3d_state_cntl_reg;
+    unsigned int texture_clr_cmp_clr_c;
+    unsigned int texture_clr_cmp_msk_c;
+    unsigned int fog_color_c;
+
+    /* Texture state */
+    unsigned int tex_size_pitch_c;
+    unsigned int constant_color_c;
+
+    /* Setup state */
+    unsigned int pm4_vc_fpu_setup;
+    unsigned int setup_cntl;
+
+    /* Mask state */
+    unsigned int dp_write_mask;
+    unsigned int sten_ref_mask_c;
+    unsigned int plane_3d_mask_c;
+
+    /* Window state */
+    unsigned int window_xy_offset;
+
+    /* Core state */
+    unsigned int scale_3d_cntl;
+} r128_context_regs_t;
+
+/* Setup registers for each texture unit
+ */
+typedef struct {
+    unsigned int tex_cntl;
+    unsigned int tex_combine_cntl;
+    unsigned int tex_size_pitch;
+    unsigned int tex_offset[R128_MAX_TEXTURE_LEVELS];
+    unsigned int tex_border_color;
+} r128_texture_regs_t;
+
+typedef struct {
+    /* The channel for communication of state information to the kernel
+     * on firing a vertex buffer.
+     */
+    r128_context_regs_t	ContextState;
+    r128_texture_regs_t	TexState[R128_MAX_TEXTURE_UNITS];
+    unsigned int dirty;
+    unsigned int vertsize;
+    unsigned int vc_format;
+
+#if defined(XF86DRI) | defined(_SOLO)
+    /* The current cliprects, or a subset thereof.
+     */
+    XF86DRIClipRectRec boxes[R128_NR_SAREA_CLIPRECTS];
+    unsigned int nbox;
+#endif
+
+    /* Counters for throttling of rendering clients.
+     */
+    unsigned int last_frame;
+    unsigned int last_dispatch;
+
+    /* Maintain an LRU of contiguous regions of texture space.  If you
+     * think you own a region of texture memory, and it has an age
+     * different to the one you set, then you are mistaken and it has
+     * been stolen by another client.  If global texAge hasn't changed,
+     * there is no need to walk the list.
+     *
+     * These regions can be used as a proxy for the fine-grained texture
+     * information of other clients - by maintaining them in the same
+     * lru which is used to age their own textures, clients have an
+     * approximate lru for the whole of global texture space, and can
+     * make informed decisions as to which areas to kick out.  There is
+     * no need to choose whether to kick out your own texture or someone
+     * else's - simply eject them all in LRU order.
+     */
+				/* Last elt is sentinal */
+    drmTextureRegion texList[R128_NR_TEX_HEAPS][R128_NR_TEX_REGIONS+1];
+				/* last time texture was uploaded */
+    unsigned int texAge[R128_NR_TEX_HEAPS];
+
+    int ctxOwner;		/* last context to upload state */
+    int pfAllowPageFlip;	/* set by the 2d driver, read by the client */
+    int pfCurrentPage;		/* set by kernel, read by others */
+} R128SAREAPriv, *R128SAREAPrivPtr;
+
+#endif
diff --git a/src/mesa/drivers/dri/r128/server/r128_version.h b/src/mesa/drivers/dri/r128/server/r128_version.h
new file mode 100644
index 00000000000..589d8d40bce
--- /dev/null
+++ b/src/mesa/drivers/dri/r128/server/r128_version.h
@@ -0,0 +1,60 @@
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/ati/r128_version.h,v 1.6 2003/01/01 19:16:35 tsi Exp $ */
+/*
+ * Copyright 2000 through 2003 by Marc Aurele La France (TSI @ UQV), [email protected]
+ *
+ * Permission to use, copy, modify, distribute, and sell this software and its
+ * documentation for any purpose is hereby granted without fee, provided that
+ * the above copyright notice appear in all copies and that both that copyright
+ * notice and this permission notice appear in supporting documentation, and
+ * that the name of Marc Aurele La France not be used in advertising or
+ * publicity pertaining to distribution of the software without specific,
+ * written prior permission.  Marc Aurele La France makes no representations
+ * about the suitability of this software for any purpose.  It is provided
+ * "as-is" without express or implied warranty.
+ *
+ * MARC AURELE LA FRANCE DISCLAIMS ALL WARRANTIES WITH REGARD TO THIS SOFTWARE,
+ * INCLUDING ALL IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS.  IN NO
+ * EVENT SHALL MARC AURELE LA FRANCE BE LIABLE FOR ANY SPECIAL, INDIRECT OR
+ * CONSEQUENTIAL DAMAGES OR ANY DAMAGES WHATSOEVER RESULTING FROM LOSS OF USE,
+ * DATA OR PROFITS, WHETHER IN AN ACTION OF CONTRACT, NEGLIGENCE OR OTHER
+ * TORTIOUS ACTION, ARISING OUT OF OR IN CONNECTION WITH THE USE OR
+ * PERFORMANCE OF THIS SOFTWARE.
+ */
+
+#ifndef _R128_VERSION_H_
+#define _R128_VERSION_H_ 1
+
+#undef  R128_NAME
+#undef  R128_DRIVER_NAME
+#undef  R128_VERSION_MAJOR
+#undef  R128_VERSION_MINOR
+#undef  R128_VERSION_PATCH
+#undef  R128_VERSION_CURRENT
+#undef  R128_VERSION_EVALUATE
+#undef  R128_VERSION_STRINGIFY
+#undef  R128_VERSION_NAME
+
+#define R128_NAME          "R128"
+#define R128_DRIVER_NAME   "r128"
+
+#define R128_VERSION_MAJOR 4
+#define R128_VERSION_MINOR 0
+#define R128_VERSION_PATCH 1
+
+#ifndef R128_VERSION_EXTRA
+#define R128_VERSION_EXTRA ""
+#endif
+
+#define R128_VERSION_CURRENT \
+    ((R128_VERSION_MAJOR << 20) | \
+     (R128_VERSION_MINOR << 10) | \
+     (R128_VERSION_PATCH))
+
+#define R128_VERSION_EVALUATE(__x) #__x
+#define R128_VERSION_STRINGIFY(_x) R128_VERSION_EVALUATE(_x)
+#define R128_VERSION_NAME                                         \
+    R128_VERSION_STRINGIFY(R128_VERSION_MAJOR) "."                \
+    R128_VERSION_STRINGIFY(R128_VERSION_MINOR) "."                \
+    R128_VERSION_STRINGIFY(R128_VERSION_MINOR) R128_VERSION_EXTRA
+
+#endif /* _R128_VERSION_H_ */
diff --git a/src/mesa/drivers/dri/r200/Makefile.X11 b/src/mesa/drivers/dri/r200/Makefile.X11
index b182ba559aa..f1887b59144 100644
--- a/src/mesa/drivers/dri/r200/Makefile.X11
+++ b/src/mesa/drivers/dri/r200/Makefile.X11
@@ -1,4 +1,4 @@
-# $Id: Makefile.X11,v 1.1 2003/08/06 17:59:57 keithw Exp $
+# $Id: Makefile.X11,v 1.2 2003/08/22 20:11:45 brianp Exp $
 
 # Mesa 3-D graphics library
 # Version:  5.0
@@ -6,8 +6,10 @@
 
 TOP = ../../../../..
 
+default: linux-solo
+
 SHARED_INCLUDES = $(INCLUDE_DIRS) -I. -I../common -Iserver
-MINIGLX_INCLUDES = -I$(TOP)/src/miniglx 
+MINIGLX_INCLUDES = -I$(TOP)/src/glx/mini
 
 DEFINES += \
 	-D_HAVE_SWRAST=1 \
@@ -62,7 +64,7 @@ WINOBJ=$(MESABUILDDIR)/dri/dri.a
 WINLIB=
 else
 WINOBJ=
-WINLIB=-L$(MESA)/src/miniglx
+WINLIB=-L$(MESA)/src/glx/mini
 endif
 
 ASM_SOURCES = 
@@ -107,7 +109,7 @@ INCLUDE_DIRS = \
 
 ##### TARGETS #####
 
-targets: r200_dri.so
+targets: depend r200_dri.so
 
 r200_dri.so:  $(SYMLINKS) $(OBJECTS) $(MESA_MODULES) $(WINOBJ) Makefile.X11
 	rm -f $@ && gcc -o $@ -shared $(OBJECTS) $(MESA_MODULES) $(WINOBJ) $(WINLIB) -lc $(GL_LIB_DEPS)
@@ -116,8 +118,8 @@ r200_dri.so:  $(SYMLINKS) $(OBJECTS) $(MESA_MODULES) $(WINOBJ) Makefile.X11
 
 # Run 'make -f Makefile.X11 dep' to update the dependencies if you change
 # what's included by any source file.
-dep: $(C_SOURCES) $(ASM_SOURCES)
-	makedepend -fdepend -Y $(SHARED_INCLUDES) \
+depend: $(C_SOURCES) $(ASM_SOURCES)
+	makedepend -fdepend -Y $(SHARED_INCLUDES) $(MINIGLX_INCLUDES) \
 		$(C_SOURCES) $(ASM_SOURCES)
 
 
@@ -128,7 +130,7 @@ tags:
 
 # Remove .o and backup files
 clean:
-	-rm -f *.o *~ *.o *~ *.so
+	-rm -f *.o */*.o *~ *.o *~ *.so server/*.o
 
 
 include $(TOP)/Make-config
diff --git a/src/mesa/drivers/dri/r200/r200_context.c b/src/mesa/drivers/dri/r200/r200_context.c
index c5f23effabb..ec95f05e6ef 100644
--- a/src/mesa/drivers/dri/r200/r200_context.c
+++ b/src/mesa/drivers/dri/r200/r200_context.c
@@ -492,8 +492,9 @@ void r200DestroyContext( __DRIcontextPrivate *driContextPriv )
           * texture object data.
           */
          int i;
-
-	 assert( is_empty_list( & rmesa->swapped ) );
+	 
+	 /* this assert is wrong. The default textures are always on swap list
+	 assert( is_empty_list( & rmesa->swapped ) ); */
 
          for ( i = 0 ; i < rmesa->nr_heaps ; i++ ) {
 	    driDestroyTextureHeap( rmesa->texture_heaps[ i ] );
diff --git a/src/mesa/drivers/dri/r200/r200_context.h b/src/mesa/drivers/dri/r200/r200_context.h
index 3802deead98..9e00d7a3585 100644
--- a/src/mesa/drivers/dri/r200/r200_context.h
+++ b/src/mesa/drivers/dri/r200/r200_context.h
@@ -615,7 +615,7 @@ struct r200_swtcl_info {
    GLuint vertex_size;
    GLuint vertex_stride_shift;
    GLuint vertex_format;
-   char *verts;
+   GLubyte *verts;
 
    /* Fallback rasterization functions
     */
diff --git a/src/mesa/drivers/dri/r200/r200_ioctl.c b/src/mesa/drivers/dri/r200/r200_ioctl.c
index ea672162387..480121a094d 100644
--- a/src/mesa/drivers/dri/r200/r200_ioctl.c
+++ b/src/mesa/drivers/dri/r200/r200_ioctl.c
@@ -320,7 +320,7 @@ static CARD32 r200GetLastFrame(r200ContextPtr rmesa)
    CARD32 frame;
 
    gp.param = RADEON_PARAM_LAST_FRAME;
-   gp.value = (int *)&frame;
+   gp.value = &frame;
    ret = drmCommandWriteRead( rmesa->dri.fd, DRM_RADEON_GETPARAM,
 			      &gp, sizeof(gp) );
    if ( ret ) {
@@ -404,7 +404,7 @@ void r200CopyBuffer( const __DRIdrawablePrivate *dPriv )
    r200ContextPtr rmesa;
    GLint nbox, i, ret;
    GLboolean   missed_target;
-   int64_t     ust;
+   uint64_t     ust;
 
    assert(dPriv);
    assert(dPriv->driContextPriv);
@@ -616,7 +616,7 @@ static void r200Clear( GLcontext *ctx, GLbitfield mask, GLboolean all,
       int clear;
 
       gp.param = RADEON_PARAM_LAST_CLEAR;
-      gp.value = (int *)&clear;
+      gp.value = &clear;
       ret = drmCommandWriteRead( rmesa->dri.fd,
 		      DRM_RADEON_GETPARAM, &gp, sizeof(gp) );
 
diff --git a/src/mesa/drivers/dri/r200/r200_swtcl.c b/src/mesa/drivers/dri/r200/r200_swtcl.c
index 6a436509a01..dd4025235b6 100644
--- a/src/mesa/drivers/dri/r200/r200_swtcl.c
+++ b/src/mesa/drivers/dri/r200/r200_swtcl.c
@@ -446,7 +446,7 @@ static __inline void *r200AllocDmaLowVerts( r200ContextPtr rmesa,
 
 
    {
-      GLubyte *head = rmesa->dma.current.address + rmesa->dma.current.ptr;
+      char *head = rmesa->dma.current.address + rmesa->dma.current.ptr;
       rmesa->dma.current.ptr += bytes;
       rmesa->swtcl.numverts += nverts;
       return head;
@@ -1260,7 +1260,7 @@ void r200InitSwtcl( GLcontext *ctx )
    tnl->Driver.Render.ResetLineStipple = r200ResetLineStipple;
    tnl->Driver.Render.BuildVertices = r200BuildVertices;
 
-   rmesa->swtcl.verts = (char *)ALIGN_MALLOC( size * 16 * 4, 32 );
+   rmesa->swtcl.verts = (GLubyte *)ALIGN_MALLOC( size * 16 * 4, 32 );
    rmesa->swtcl.RenderIndex = ~0;
    rmesa->swtcl.render_primitive = GL_TRIANGLES;
    rmesa->swtcl.hw_primitive = 0;
diff --git a/src/mesa/drivers/dri/r200/r200_tcl.c b/src/mesa/drivers/dri/r200/r200_tcl.c
index 994221d80cf..9869ad61963 100644
--- a/src/mesa/drivers/dri/r200/r200_tcl.c
+++ b/src/mesa/drivers/dri/r200/r200_tcl.c
@@ -102,33 +102,23 @@ static GLboolean discrete_prim[0x10] = {
 };
    
 
-#define LOCAL_VARS r200ContextPtr rmesa = R200_CONTEXT(ctx);rmesa = rmesa
-#define ELTS_VARS  GLushort *dest
+#define LOCAL_VARS r200ContextPtr rmesa = R200_CONTEXT(ctx)
+#define ELT_TYPE  GLushort
 
 #define ELT_INIT(prim, hw_prim) \
    r200TclPrimitive( ctx, prim, hw_prim | R200_VF_PRIM_WALK_IND )
 
-#define GET_ELTS() rmesa->tcl.Elts
+#define GET_MESA_ELTS() rmesa->tcl.Elts
 
 
-#define NEW_PRIMITIVE()  R200_NEWPRIM( rmesa )
-#define NEW_BUFFER()  r200RefillCurrentDmaRegion( rmesa )
-
 /* Don't really know how many elts will fit in what's left of cmdbuf,
  * as there is state to emit, etc:
  */
 
-#if 0
-#define GET_CURRENT_VB_MAX_ELTS() \
-   ((R200_CMD_BUF_SZ - (rmesa->store.cmd_used + 16)) / 2) 
-#define GET_SUBSEQUENT_VB_MAX_ELTS() ((R200_CMD_BUF_SZ - 16) / 2) 
-#else
 /* Testing on isosurf shows a maximum around here.  Don't know if it's
  * the card or driver or kernel module that is causing the behaviour.
  */
-#define GET_CURRENT_VB_MAX_ELTS() 300
-#define GET_SUBSEQUENT_VB_MAX_ELTS() 300
-#endif
+#define GET_MAX_HW_ELTS() 300
 
 #define RESET_STIPPLE() do {			\
    R200_STATECHANGE( rmesa, lin );		\
@@ -147,32 +137,22 @@ static GLboolean discrete_prim[0x10] = {
 } while (0)
 
 
-/* How do you extend an existing primitive?
- */
-#define ALLOC_ELTS(nr)							\
-do {									\
-   if (rmesa->dma.flush == r200FlushElts &&				\
-       rmesa->store.cmd_used + nr*2 < R200_CMD_BUF_SZ) {		\
-									\
-      dest = (GLushort *)(rmesa->store.cmd_buf + 			\
-			  rmesa->store.cmd_used);			\
-      rmesa->store.cmd_used += nr*2;					\
-   }									\
-   else {								\
-      if (rmesa->dma.flush)						\
-	 rmesa->dma.flush( rmesa );					\
-									\
-      r200EmitAOS( rmesa,						\
-	  	     rmesa->tcl.aos_components,				\
-		     rmesa->tcl.nr_aos_components,			\
-		     0 );						\
-									\
-      dest = r200AllocEltsOpenEnded( rmesa,				\
-				       rmesa->tcl.hw_primitive,		\
-				       nr );				\
-   }									\
-} while (0) 
+#define ALLOC_ELTS(nr)	r200AllocElts( rmesa, nr )
 
+static GLushort *r200AllocElts( r200ContextPtr rmesa, GLuint nr ) 
+{
+   if (rmesa->dma.flush)
+      rmesa->dma.flush( rmesa );
+
+   r200EmitAOS( rmesa,
+		rmesa->tcl.aos_components,
+		rmesa->tcl.nr_aos_components, 0 );
+
+   return r200AllocEltsOpenEnded( rmesa, rmesa->tcl.hw_primitive, nr );
+}
+
+
+#define CLOSE_ELTS()  R200_NEWPRIM( rmesa )
 
 
 /* TODO: Try to extend existing primitive if both are identical,
@@ -217,17 +197,15 @@ static void EMIT_PRIM( GLcontext *ctx,
 
 #ifdef MESA_BIG_ENDIAN
 /* We could do without (most of) this ugliness if dest was always 32 bit word aligned... */
-#define EMIT_ELT(offset, x) do {                                \
+#define EMIT_ELT(dest, offset, x) do {                                \
         int off = offset + ( ( (GLuint)dest & 0x2 ) >> 1 );     \
         GLushort *des = (GLushort *)( (GLuint)dest & ~0x2 );    \
         (des)[ off + 1 - 2 * ( off & 1 ) ] = (GLushort)(x); } while (0)
 #else
-#define EMIT_ELT(offset, x) (dest)[offset] = (GLushort) (x)
+#define EMIT_ELT(dest, offset, x) (dest)[offset] = (GLushort) (x)
 #endif
-#define EMIT_TWO_ELTS(offset, x, y)  *(GLuint *)(dest+offset) = ((y)<<16)|(x);
-#define INCR_ELTS( nr ) dest += nr
-#define RELEASE_ELT_VERTS() \
-   r200ReleaseArrays( ctx, ~0 )
+
+#define EMIT_TWO_ELTS(dest, offset, x, y)  *(GLuint *)((dest)+offset) = ((y)<<16)|(x);
 
 
 
diff --git a/src/mesa/drivers/dri/radeon/Makefile.X11 b/src/mesa/drivers/dri/radeon/Makefile.X11
new file mode 100644
index 00000000000..b73abe8f51b
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/Makefile.X11
@@ -0,0 +1,154 @@
+# $Id: Makefile.X11,v 1.1 2003/08/22 20:11:45 brianp Exp $
+
+# Mesa 3-D graphics library
+# Version:  5.0
+# Copyright (C) 1995-2002  Brian Paul
+
+TOP = ../../../../..
+
+default: linux-solo
+
+SHARED_INCLUDES = $(INCLUDE_DIRS) -I. -I../common -Iserver
+MINIGLX_INCLUDES = -I$(TOP)/src/glx/mini
+
+ifeq ($(EMBEDDED),true)
+TARGET = radeon_es_dri.so
+DEFINES += \
+              -D_EMBEDDED \
+	-D_HAVE_SWRAST=0 \
+	-D_HAVE_SWTNL=0 \
+	-D_HAVE_SANITY=0 \
+	-D_HAVE_CODEGEN=0 \
+	-D_HAVE_LIGHTING=0 \
+	-D_HAVE_TEXGEN=0 \
+	-D_HAVE_USERCLIP=0 \
+	-DGLX_DIRECT_RENDERING
+else
+TARGET = radeon_dri.so
+DEFINES += \
+	-D_HAVE_SWRAST=1 \
+	-D_HAVE_SWTNL=1 \
+	-D_HAVE_SANITY=1 \
+	-D_HAVE_CODEGEN=1 \
+	-D_HAVE_LIGHTING=1 \
+	-D_HAVE_TEXGEN=1 \
+	-D_HAVE_USERCLIP=1 \
+	-DGLX_DIRECT_RENDERING
+endif
+
+MESA_MODULES = $(TOP)/src/mesa/mesa.a
+
+MINIGLX_SOURCES = server/radeon_dri.c 
+
+DRIVER_SOURCES = radeon_context.c \
+		 radeon_ioctl.c \
+		 radeon_lock.c \
+		 radeon_screen.c \
+		 radeon_state.c \
+		 radeon_state_init.c \
+		 ../common/mm.c \
+		 ../common/utils.c \
+		 ../common/texmem.c \
+		 ../common/vblank.c
+
+SUBSET_DRIVER_SOURCES = \
+		radeon_subset_bitmap.c \
+		radeon_subset_readpix.c \
+		radeon_subset_select.c \
+		radeon_subset_tex.c \
+		radeon_subset_vtx.c 
+
+FULL_DRIVER_SOURCES = 	\
+		 radeon_tex.c \
+		 radeon_texmem.c \
+		 radeon_texstate.c \
+		 radeon_tcl.c \
+		 radeon_swtcl.c \
+		 radeon_span.c \
+		 radeon_maos.c \
+		 radeon_sanity.c \
+	  	 radeon_compat.c \
+		 radeon_vtxfmt.c \
+		 radeon_vtxfmt_c.c \
+		 radeon_vtxfmt_sse.c \
+		 radeon_vtxfmt_x86.c 
+
+
+INCLUDES = $(MINIGLX_INCLUDES) \
+	   $(SHARED_INCLUDES)
+
+
+ifeq ($(EMBEDDED),true)
+C_SOURCES = $(DRIVER_SOURCES) \
+	    $(SUBSET_DRIVER_SOURCES) \
+	    $(MINIGLX_SOURCES) 
+else
+C_SOURCES = $(DRIVER_SOURCES) \
+	    $(FULL_DRIVER_SOURCES) \
+	    $(MINIGLX_SOURCES) 
+endif
+
+
+ifeq ($(WINDOW_SYSTEM),dri)
+WINOBJ=$(MESABUILDDIR)/dri/dri.a
+WINLIB=
+else
+WINOBJ=
+WINLIB=-L$(MESA)/src/glx/mini
+endif
+
+ASM_SOURCES = 
+OBJECTS = $(C_SOURCES:.c=.o) \
+	  $(ASM_SOURCES:.S=.o) 
+
+### Include directories
+
+INCLUDE_DIRS = \
+	-I$(TOP)/include \
+	-I$(TOP)/src/mesa \
+	-I$(TOP)/src/mesa/main \
+	-I$(TOP)/src/mesa/glapi \
+	-I$(TOP)/src/mesa/math \
+	-I$(TOP)/src/mesa/transform \
+	-I$(TOP)/src/mesa/swrast \
+	-I$(TOP)/src/mesa/swrast_setup
+
+
+##### RULES #####
+
+.c.o:
+	$(CC) -c $(SHARED_INCLUDES) $(MINIGLX_INCLUDES) $(CFLAGS) $(DEFINES) $< -o $@
+
+.S.o:
+	$(CC) -c $(SHARED_INCLUDES) $(MINIGLX_INCLUDES) $(CFLAGS) $(DEFINES)  $< -o $@
+
+
+##### TARGETS #####
+
+targets: depend $(TARGET)
+
+$(TARGET):  $(OBJECTS) $(MESA_MODULES) $(WINOBJ) Makefile.X11
+	rm -f $@ && gcc -o $@ -shared $(OBJECTS) $(MESA_MODULES) $(WINOBJ) $(WINLIB) -lc -lm
+	rm -f $(TOP)/lib/$(TARGET) && \
+	install $(TARGET) $(TOP)/lib/$(TARGET)
+
+# Run 'make -f Makefile.X11 dep' to update the dependencies if you change
+# what's included by any source file.
+depend: $(C_SOURCES) $(ASM_SOURCES)
+	makedepend -fdepend -Y $(SHARED_INCLUDES) $(MINIGLX_INCLUDES)\
+		$(C_SOURCES) $(ASM_SOURCES)
+
+
+# Emacs tags
+tags:
+	etags `find . -name \*.[ch]` `find ../include`
+
+
+# Remove .o and backup files
+clean:
+	-rm -f *.o *~ *.o *~ *.so server/*.o
+
+
+include $(TOP)/Make-config
+
+include depend
diff --git a/src/mesa/drivers/dri/radeon/radeon_compat.c b/src/mesa/drivers/dri/radeon/radeon_compat.c
new file mode 100644
index 00000000000..0c32641530b
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_compat.c
@@ -0,0 +1,304 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 2002 ATI Technologies Inc., Ontario, Canada, and
+               Tungsten Graphics Inc., Austin, 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
+ATI, TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell <[email protected]>
+ *
+ */
+
+#include "glheader.h"
+#include "imports.h"
+
+#include "radeon_context.h"
+#include "radeon_state.h"
+#include "radeon_ioctl.h"
+
+
+static struct { 
+	int start; 
+	int len; 
+	const char *name;
+} packet[RADEON_MAX_STATE_PACKETS] = {
+	{ RADEON_PP_MISC,7,"RADEON_PP_MISC" },
+	{ RADEON_PP_CNTL,3,"RADEON_PP_CNTL" },
+	{ RADEON_RB3D_COLORPITCH,1,"RADEON_RB3D_COLORPITCH" },
+	{ RADEON_RE_LINE_PATTERN,2,"RADEON_RE_LINE_PATTERN" },
+	{ RADEON_SE_LINE_WIDTH,1,"RADEON_SE_LINE_WIDTH" },
+	{ RADEON_PP_LUM_MATRIX,1,"RADEON_PP_LUM_MATRIX" },
+	{ RADEON_PP_ROT_MATRIX_0,2,"RADEON_PP_ROT_MATRIX_0" },
+	{ RADEON_RB3D_STENCILREFMASK,3,"RADEON_RB3D_STENCILREFMASK" },
+	{ RADEON_SE_VPORT_XSCALE,6,"RADEON_SE_VPORT_XSCALE" },
+	{ RADEON_SE_CNTL,2,"RADEON_SE_CNTL" },
+	{ RADEON_SE_CNTL_STATUS,1,"RADEON_SE_CNTL_STATUS" },
+	{ RADEON_RE_MISC,1,"RADEON_RE_MISC" },
+	{ RADEON_PP_TXFILTER_0,6,"RADEON_PP_TXFILTER_0" },
+	{ RADEON_PP_BORDER_COLOR_0,1,"RADEON_PP_BORDER_COLOR_0" },
+	{ RADEON_PP_TXFILTER_1,6,"RADEON_PP_TXFILTER_1" },
+	{ RADEON_PP_BORDER_COLOR_1,1,"RADEON_PP_BORDER_COLOR_1" },
+	{ RADEON_PP_TXFILTER_2,6,"RADEON_PP_TXFILTER_2" },
+	{ RADEON_PP_BORDER_COLOR_2,1,"RADEON_PP_BORDER_COLOR_2" },
+	{ RADEON_SE_ZBIAS_FACTOR,2,"RADEON_SE_ZBIAS_FACTOR" },
+	{ RADEON_SE_TCL_OUTPUT_VTX_FMT,11,"RADEON_SE_TCL_OUTPUT_VTX_FMT" },
+	{ RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED,17,"RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED" },
+};
+
+
+static void radeonCompatEmitPacket( radeonContextPtr rmesa, 
+				    struct radeon_state_atom *state )
+{
+   RADEONSAREAPrivPtr sarea = rmesa->sarea;
+   radeon_context_regs_t *ctx = &sarea->ContextState;
+   radeon_texture_regs_t *tex0 = &sarea->TexState[0];
+   radeon_texture_regs_t *tex1 = &sarea->TexState[1];
+   int i;
+   int *buf = state->cmd;
+
+   for ( i = 0 ; i < state->cmd_size ; ) {
+      drmRadeonCmdHeader *header = (drmRadeonCmdHeader *)&buf[i++];
+
+      if (RADEON_DEBUG & DEBUG_STATE)
+	 fprintf(stderr, "%s %d: %s\n", __FUNCTION__, header->packet.packet_id,
+		 packet[(int)header->packet.packet_id].name);
+
+      switch (header->packet.packet_id) {
+      case RADEON_EMIT_PP_MISC:
+	 ctx->pp_misc = buf[i++]; 
+	 ctx->pp_fog_color = buf[i++];
+	 ctx->re_solid_color = buf[i++];
+	 ctx->rb3d_blendcntl = buf[i++];
+	 ctx->rb3d_depthoffset = buf[i++];
+	 ctx->rb3d_depthpitch = buf[i++];
+	 ctx->rb3d_zstencilcntl = buf[i++];
+	 sarea->dirty |= RADEON_UPLOAD_CONTEXT;
+	 break;
+      case RADEON_EMIT_PP_CNTL:
+	 ctx->pp_cntl = buf[i++];
+	 ctx->rb3d_cntl = buf[i++];
+	 ctx->rb3d_coloroffset = buf[i++];
+	 sarea->dirty |= RADEON_UPLOAD_CONTEXT;
+	 break;
+      case RADEON_EMIT_RB3D_COLORPITCH:
+	 ctx->rb3d_colorpitch = buf[i++];
+	 sarea->dirty |= RADEON_UPLOAD_CONTEXT;
+	 break;
+      case RADEON_EMIT_RE_LINE_PATTERN:
+	 ctx->re_line_pattern = buf[i++];
+	 ctx->re_line_state = buf[i++];
+	 sarea->dirty |= RADEON_UPLOAD_LINE;
+	 break;
+      case RADEON_EMIT_SE_LINE_WIDTH:
+	 ctx->se_line_width = buf[i++];
+	 sarea->dirty |= RADEON_UPLOAD_LINE;
+	 break;
+      case RADEON_EMIT_PP_LUM_MATRIX:
+	 ctx->pp_lum_matrix = buf[i++];
+	 sarea->dirty |= RADEON_UPLOAD_BUMPMAP;
+	 break;
+      case RADEON_EMIT_PP_ROT_MATRIX_0:
+	 ctx->pp_rot_matrix_0 = buf[i++];
+	 ctx->pp_rot_matrix_1 = buf[i++];
+	 sarea->dirty |= RADEON_UPLOAD_BUMPMAP;
+	 break;
+      case RADEON_EMIT_RB3D_STENCILREFMASK:
+	 ctx->rb3d_stencilrefmask = buf[i++];
+	 ctx->rb3d_ropcntl = buf[i++];
+	 ctx->rb3d_planemask = buf[i++];
+	 sarea->dirty |= RADEON_UPLOAD_MASKS;
+	 break;
+      case RADEON_EMIT_SE_VPORT_XSCALE:
+	 ctx->se_vport_xscale = buf[i++];
+	 ctx->se_vport_xoffset = buf[i++];
+	 ctx->se_vport_yscale = buf[i++];
+	 ctx->se_vport_yoffset = buf[i++];
+	 ctx->se_vport_zscale = buf[i++];
+	 ctx->se_vport_zoffset = buf[i++];
+	 sarea->dirty |= RADEON_UPLOAD_VIEWPORT;
+	 break;
+      case RADEON_EMIT_SE_CNTL:
+	 ctx->se_cntl = buf[i++];
+	 ctx->se_coord_fmt = buf[i++];
+	 sarea->dirty |= RADEON_UPLOAD_CONTEXT | RADEON_UPLOAD_VERTFMT;
+	 break;
+      case RADEON_EMIT_SE_CNTL_STATUS:
+	 ctx->se_cntl_status = buf[i++];
+	 sarea->dirty |= RADEON_UPLOAD_SETUP;
+	 break;
+      case RADEON_EMIT_RE_MISC:
+	 ctx->re_misc = buf[i++];
+	 sarea->dirty |= RADEON_UPLOAD_MISC;
+	 break;
+      case RADEON_EMIT_PP_TXFILTER_0:
+	 tex0->pp_txfilter = buf[i++];
+	 tex0->pp_txformat = buf[i++];
+	 tex0->pp_txoffset = buf[i++];
+	 tex0->pp_txcblend = buf[i++];
+	 tex0->pp_txablend = buf[i++];
+	 tex0->pp_tfactor = buf[i++];
+	 sarea->dirty |= RADEON_UPLOAD_TEX0;
+	 break;
+      case RADEON_EMIT_PP_BORDER_COLOR_0:
+	 tex0->pp_border_color = buf[i++];
+	 sarea->dirty |= RADEON_UPLOAD_TEX0;
+	 break;
+      case RADEON_EMIT_PP_TXFILTER_1:
+	 tex1->pp_txfilter = buf[i++];
+	 tex1->pp_txformat = buf[i++];
+	 tex1->pp_txoffset = buf[i++];
+	 tex1->pp_txcblend = buf[i++];
+	 tex1->pp_txablend = buf[i++];
+	 tex1->pp_tfactor = buf[i++];
+	 sarea->dirty |= RADEON_UPLOAD_TEX1;
+	 break;
+      case RADEON_EMIT_PP_BORDER_COLOR_1:
+	 tex1->pp_border_color = buf[i++];
+	 sarea->dirty |= RADEON_UPLOAD_TEX1;
+	 break;
+
+      case RADEON_EMIT_SE_ZBIAS_FACTOR:
+	 i++;
+	 i++;
+	 break;
+
+      case RADEON_EMIT_PP_TXFILTER_2:
+      case RADEON_EMIT_PP_BORDER_COLOR_2:
+      case RADEON_EMIT_SE_TCL_OUTPUT_VTX_FMT:
+      case RADEON_EMIT_SE_TCL_MATERIAL_EMMISSIVE_RED:
+      default:
+	 /* These states aren't understood by radeon drm 1.1 */
+	 fprintf(stderr, "Tried to emit unsupported state\n");
+	 return;
+      }
+   }
+}
+
+
+
+static void radeonCompatEmitStateLocked( radeonContextPtr rmesa )
+{
+   struct radeon_state_atom *state, *tmp;
+
+   if (RADEON_DEBUG & (DEBUG_STATE|DEBUG_PRIMS))
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   if (rmesa->lost_context) {
+      if (RADEON_DEBUG & (DEBUG_STATE|DEBUG_PRIMS|DEBUG_IOCTL))
+	 fprintf(stderr, "%s - lost context\n", __FUNCTION__); 
+
+      foreach_s( state, tmp, &(rmesa->hw.clean) ) 
+	 move_to_tail(&(rmesa->hw.dirty), state );
+
+      rmesa->lost_context = 0;
+   }
+
+   foreach_s( state, tmp, &(rmesa->hw.dirty) ) {
+      if (!state->is_tcl)
+	 radeonCompatEmitPacket( rmesa, state );
+      move_to_head( &(rmesa->hw.clean), state );
+   }
+}
+
+
+
+static void radeonCompatEmitPrimitiveLocked( radeonContextPtr rmesa,
+					     GLuint hw_primitive,
+					     GLuint nverts,
+					     XF86DRIClipRectPtr pbox,
+					     GLuint nbox )
+{
+   int i;
+
+   for ( i = 0 ; i < nbox ; ) {
+      int nr = MIN2( i + RADEON_NR_SAREA_CLIPRECTS, nbox );
+      XF86DRIClipRectPtr b = rmesa->sarea->boxes;
+      drmRadeonVertex vtx;
+      
+      rmesa->sarea->dirty |= RADEON_UPLOAD_CLIPRECTS;
+      rmesa->sarea->nbox = nr - i;
+
+      for ( ; i < nr ; i++) 
+	 *b++ = pbox[i];
+      
+      if (RADEON_DEBUG & DEBUG_IOCTL)
+	 fprintf(stderr, 
+		 "RadeonFlushVertexBuffer: prim %x buf %d verts %d "
+		 "disc %d nbox %d\n",
+		 hw_primitive, 
+		 rmesa->dma.current.buf->buf->idx, 
+		 nverts, 
+		 nr == nbox,
+		 rmesa->sarea->nbox );
+
+      vtx.prim = hw_primitive;
+      vtx.idx = rmesa->dma.current.buf->buf->idx;
+      vtx.count = nverts;
+      vtx.discard = (nr == nbox);      
+
+      drmCommandWrite( rmesa->dri.fd, 
+		       DRM_RADEON_VERTEX,
+		       &vtx, sizeof(vtx));
+   }
+}
+
+
+
+/* No 'start' for 1.1 vertices ioctl: only one vertex prim/buffer!  
+ */
+void radeonCompatEmitPrimitive( radeonContextPtr rmesa,
+				GLuint vertex_format,
+				GLuint hw_primitive,
+				GLuint nrverts )
+{
+   if (RADEON_DEBUG & DEBUG_IOCTL)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   LOCK_HARDWARE( rmesa );
+
+   radeonCompatEmitStateLocked( rmesa );
+   rmesa->sarea->vc_format = vertex_format;
+   
+   if (rmesa->state.scissor.enabled) {
+      radeonCompatEmitPrimitiveLocked( rmesa, 
+				       hw_primitive,
+				       nrverts,
+				       rmesa->state.scissor.pClipRects,
+				       rmesa->state.scissor.numClipRects );
+   }
+   else {
+      radeonCompatEmitPrimitiveLocked( rmesa, 
+				       hw_primitive,
+				       nrverts,
+				       rmesa->pClipRects,
+				       rmesa->numClipRects );
+   }
+
+
+   UNLOCK_HARDWARE( rmesa );
+}
+
diff --git a/src/mesa/drivers/dri/radeon/radeon_context.c b/src/mesa/drivers/dri/radeon/radeon_context.c
new file mode 100644
index 00000000000..835cecbc3a0
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_context.c
@@ -0,0 +1,598 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_context.c,v 1.7 2003/02/08 21:26:45 dawes Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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 (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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ *   Keith Whitwell <[email protected]>
+ */
+
+#include "glheader.h"
+#include "api_arrayelt.h"
+#include "context.h"
+#include "simple_list.h"
+#include "imports.h"
+#include "matrix.h"
+#include "extensions.h"
+
+#include "swrast/swrast.h"
+#include "swrast_setup/swrast_setup.h"
+#include "array_cache/acache.h"
+
+#include "tnl/tnl.h"
+#include "tnl/t_pipeline.h"
+
+#include "radeon_context.h"
+#include "radeon_ioctl.h"
+#include "radeon_state.h"
+#include "radeon_span.h"
+#include "radeon_tex.h"
+#include "radeon_swtcl.h"
+#include "radeon_tcl.h"
+#include "radeon_vtxfmt.h"
+#include "radeon_maos.h"
+
+#define DRIVER_DATE	"20030328"
+
+#include "vblank.h"
+#include "utils.h"
+#ifndef RADEON_DEBUG
+int RADEON_DEBUG = (0);
+#endif
+
+
+
+/* Return the width and height of the given buffer.
+ */
+static void radeonGetBufferSize( GLframebuffer *buffer,
+				 GLuint *width, GLuint *height )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   LOCK_HARDWARE( rmesa );
+   *width  = rmesa->dri.drawable->w;
+   *height = rmesa->dri.drawable->h;
+   UNLOCK_HARDWARE( rmesa );
+}
+
+/* Return various strings for glGetString().
+ */
+static const GLubyte *radeonGetString( GLcontext *ctx, GLenum name )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   static char buffer[128];
+   unsigned   offset;
+   GLuint agp_mode = rmesa->radeonScreen->IsPCI ? 0 :
+      rmesa->radeonScreen->AGPMode;
+
+   switch ( name ) {
+   case GL_VENDOR:
+      return (GLubyte *)"Tungsten Graphics, Inc.";
+
+   case GL_RENDERER:
+      offset = driGetRendererString( buffer, "Radeon", DRIVER_DATE,
+				     agp_mode );
+
+      sprintf( & buffer[ offset ], "%s %sTCL",
+	       ( rmesa->dri.drmMinor < 3 ) ? " DRM-COMPAT" : "",
+	       !(rmesa->TclFallback & RADEON_TCL_FALLBACK_TCL_DISABLE)
+	       ? "" : "NO-" );
+
+      return (GLubyte *)buffer;
+
+   default:
+      return NULL;
+   }
+}
+
+
+/* Extension strings exported by the R100 driver.
+ */
+static const char * const card_extensions[] =
+{
+    "GL_ARB_multisample",
+    "GL_ARB_multitexture",
+    "GL_ARB_texture_border_clamp",
+    "GL_ARB_texture_compression",
+    "GL_ARB_texture_env_add",
+    "GL_ARB_texture_env_combine",
+    "GL_ARB_texture_env_dot3",
+    "GL_ARB_texture_mirrored_repeat",
+    "GL_EXT_blend_logic_op",
+    "GL_EXT_blend_subtract",
+    "GL_EXT_secondary_color",
+    "GL_EXT_texture_edge_clamp",
+    "GL_EXT_texture_env_add",
+    "GL_EXT_texture_env_combine",
+    "GL_EXT_texture_env_dot3",
+    "GL_EXT_texture_filter_anisotropic",
+    "GL_EXT_texture_lod_bias",
+    "GL_ATI_texture_env_combine3",
+    "GL_ATI_texture_mirror_once",
+    "GL_IBM_texture_mirrored_repeat",
+    "GL_MESA_ycbcr_texture",
+    "GL_NV_blend_square",
+    "GL_SGIS_generate_mipmap",
+    "GL_SGIS_texture_border_clamp",
+    "GL_SGIS_texture_edge_clamp",
+    NULL
+};
+
+extern const struct gl_pipeline_stage _radeon_texrect_stage;
+extern const struct gl_pipeline_stage _radeon_render_stage;
+extern const struct gl_pipeline_stage _radeon_tcl_stage;
+
+static const struct gl_pipeline_stage *radeon_pipeline[] = {
+
+   /* Try and go straight to t&l
+    */
+   &_radeon_tcl_stage,  
+
+   /* Catch any t&l fallbacks
+    */
+   &_tnl_vertex_transform_stage,
+   &_tnl_normal_transform_stage,
+   &_tnl_lighting_stage,
+   &_tnl_fog_coordinate_stage,
+   &_tnl_texgen_stage,
+   &_tnl_texture_transform_stage,
+
+   /* Scale texture rectangle to 0..1.
+    */
+   &_radeon_texrect_stage,
+
+   &_radeon_render_stage,
+   &_tnl_render_stage,		/* FALLBACK:  */
+   0,
+};
+
+
+
+/* Initialize the driver's misc functions.
+ */
+static void radeonInitDriverFuncs( GLcontext *ctx )
+{
+    ctx->Driver.GetBufferSize		= radeonGetBufferSize;
+    ctx->Driver.ResizeBuffers           = _swrast_alloc_buffers;
+    ctx->Driver.GetString		= radeonGetString;
+
+    ctx->Driver.Error			= NULL;
+    ctx->Driver.DrawPixels		= NULL;
+    ctx->Driver.Bitmap			= NULL;
+}
+
+static const struct dri_debug_control debug_control[] =
+{
+    { "fall",  DEBUG_FALLBACKS },
+    { "tex",   DEBUG_TEXTURE },
+    { "ioctl", DEBUG_IOCTL },
+    { "prim",  DEBUG_PRIMS },
+    { "vert",  DEBUG_VERTS },
+    { "state", DEBUG_STATE },
+    { "code",  DEBUG_CODEGEN },
+    { "vfmt",  DEBUG_VFMT },
+    { "vtxf",  DEBUG_VFMT },
+    { "verb",  DEBUG_VERBOSE },
+    { "dri",   DEBUG_DRI },
+    { "dma",   DEBUG_DMA },
+    { "san",   DEBUG_SANITY },
+    { NULL,    0 }
+};
+
+
+static int
+get_ust_nop( uint64_t * ust )
+{
+   *ust = 1;
+   return 0;
+}
+
+
+/* Create the device specific context.
+ */
+GLboolean
+radeonCreateContext( const __GLcontextModes *glVisual,
+                     __DRIcontextPrivate *driContextPriv,
+                     void *sharedContextPrivate)
+{
+   __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv;
+   radeonScreenPtr screen = (radeonScreenPtr)(sPriv->private);
+   radeonContextPtr rmesa;
+   GLcontext *ctx, *shareCtx;
+   int i;
+
+   assert(glVisual);
+   assert(driContextPriv);
+   assert(screen);
+
+   /* Allocate the Radeon context */
+   rmesa = (radeonContextPtr) CALLOC( sizeof(*rmesa) );
+   if ( !rmesa )
+      return GL_FALSE;
+
+   /* Allocate the Mesa context */
+   if (sharedContextPrivate)
+      shareCtx = ((radeonContextPtr) sharedContextPrivate)->glCtx;
+   else
+      shareCtx = NULL;
+   rmesa->glCtx = _mesa_create_context(glVisual, shareCtx, (void *) rmesa, GL_TRUE);
+   if (!rmesa->glCtx) {
+      FREE(rmesa);
+      return GL_FALSE;
+   }
+   driContextPriv->driverPrivate = rmesa;
+
+   /* Init radeon context data */
+   rmesa->dri.context = driContextPriv;
+   rmesa->dri.screen = sPriv;
+   rmesa->dri.drawable = NULL; /* Set by XMesaMakeCurrent */
+   rmesa->dri.hwContext = driContextPriv->hHWContext;
+   rmesa->dri.hwLock = &sPriv->pSAREA->lock;
+   rmesa->dri.fd = sPriv->fd;
+
+   /* If we don't have 1.3, fallback to the 1.1 interfaces.
+    */
+   if (getenv("RADEON_COMPAT") || sPriv->drmMinor < 3 ) 
+      rmesa->dri.drmMinor = 1;
+   else
+      rmesa->dri.drmMinor = sPriv->drmMinor;
+
+   rmesa->radeonScreen = screen;
+   rmesa->sarea = (RADEONSAREAPrivPtr)((GLubyte *)sPriv->pSAREA +
+				       screen->sarea_priv_offset);
+
+
+   rmesa->dma.buf0_address = rmesa->radeonScreen->buffers->list[0].address;
+
+   (void) memset( rmesa->texture_heaps, 0, sizeof( rmesa->texture_heaps ) );
+   make_empty_list( & rmesa->swapped );
+
+   rmesa->nr_heaps = screen->numTexHeaps;
+   for ( i = 0 ; i < rmesa->nr_heaps ; i++ ) {
+      rmesa->texture_heaps[i] = driCreateTextureHeap( i, rmesa,
+	    screen->texSize[i],
+	    12,
+	    RADEON_NR_TEX_REGIONS,
+	    rmesa->sarea->texList[i],
+	    & rmesa->sarea->texAge[i],
+	    & rmesa->swapped,
+	    sizeof( radeonTexObj ),
+	    (destroy_texture_object_t *) radeonDestroyTexObj );
+
+      driSetTextureSwapCounterLocation( rmesa->texture_heaps[i],
+					& rmesa->c_textureSwaps );
+   }
+
+   rmesa->swtcl.RenderIndex = ~0;
+   rmesa->lost_context = 1;
+
+   /* Set the maximum texture size small enough that we can guarentee that
+    * all texture units can bind a maximal texture and have them both in
+    * texturable memory at once.
+    */
+
+   ctx = rmesa->glCtx;
+   ctx->Const.MaxTextureUnits = 2;
+
+   driCalculateMaxTextureLevels( rmesa->texture_heaps,
+				 rmesa->nr_heaps,
+				 & ctx->Const,
+				 4,
+				 11, /* max 2D texture size is 2048x2048 */
+				 0,  /* 3D textures unsupported. */
+				 0,  /* cube textures unsupported. */
+				 11, /* max rect texture size is 2048x2048. */
+				 12,
+				 GL_FALSE );
+
+   ctx->Const.MaxTextureMaxAnisotropy = 16.0;
+
+   /* No wide points.
+    */
+   ctx->Const.MinPointSize = 1.0;
+   ctx->Const.MinPointSizeAA = 1.0;
+   ctx->Const.MaxPointSize = 1.0;
+   ctx->Const.MaxPointSizeAA = 1.0;
+
+   ctx->Const.MinLineWidth = 1.0;
+   ctx->Const.MinLineWidthAA = 1.0;
+   ctx->Const.MaxLineWidth = 10.0;
+   ctx->Const.MaxLineWidthAA = 10.0;
+   ctx->Const.LineWidthGranularity = 0.0625;
+
+   /* Set maxlocksize (and hence vb size) small enough to avoid
+    * fallbacks in radeon_tcl.c.  ie. guarentee that all vertices can
+    * fit in a single dma buffer for indexed rendering of quad strips,
+    * etc.
+    */
+   ctx->Const.MaxArrayLockSize = 
+      MIN2( ctx->Const.MaxArrayLockSize, 
+ 	    RADEON_BUFFER_SIZE / RADEON_MAX_TCL_VERTSIZE ); 
+
+   rmesa->boxes = (getenv("LIBGL_PERFORMANCE_BOXES") != NULL);
+
+   /* Initialize the software rasterizer and helper modules.
+    */
+   _swrast_CreateContext( ctx );
+   _ac_CreateContext( ctx );
+   _tnl_CreateContext( ctx );
+   _swsetup_CreateContext( ctx );
+   _ae_create_context( ctx );
+
+   /* Install the customized pipeline:
+    */
+   _tnl_destroy_pipeline( ctx );
+   _tnl_install_pipeline( ctx, radeon_pipeline );
+   ctx->Driver.FlushVertices = radeonFlushVertices;
+
+   /* Try and keep materials and vertices separate:
+    */
+   _tnl_isolate_materials( ctx, GL_TRUE );
+
+
+/*     _mesa_allow_light_in_model( ctx, GL_FALSE ); */
+
+   /* Try and keep materials and vertices separate:
+    */
+   _tnl_isolate_materials( ctx, GL_TRUE );
+
+
+   /* Configure swrast to match hardware characteristics:
+    */
+   _swrast_allow_pixel_fog( ctx, GL_FALSE );
+   _swrast_allow_vertex_fog( ctx, GL_TRUE );
+
+
+   _math_matrix_ctr( &rmesa->TexGenMatrix[0] );
+   _math_matrix_ctr( &rmesa->TexGenMatrix[1] );
+   _math_matrix_ctr( &rmesa->tmpmat );
+   _math_matrix_set_identity( &rmesa->TexGenMatrix[0] );
+   _math_matrix_set_identity( &rmesa->TexGenMatrix[1] );
+   _math_matrix_set_identity( &rmesa->tmpmat );
+
+   driInitExtensions( ctx, card_extensions, GL_TRUE );
+
+   if (rmesa->dri.drmMinor >= 9)
+      _mesa_enable_extension( ctx, "GL_NV_texture_rectangle");
+
+   radeonInitDriverFuncs( ctx );
+   radeonInitIoctlFuncs( ctx );
+   radeonInitStateFuncs( ctx );
+   radeonInitSpanFuncs( ctx );
+   radeonInitTextureFuncs( ctx );
+   radeonInitState( rmesa );
+   radeonInitSwtcl( ctx );
+
+   rmesa->iw.irq_seq = -1;
+   rmesa->irqsEmitted = 0;
+   rmesa->do_irqs = (rmesa->radeonScreen->irq && !getenv("RADEON_NO_IRQS"));
+
+   rmesa->do_usleeps = !getenv("RADEON_NO_USLEEPS");
+
+   rmesa->vblank_flags = (rmesa->do_irqs)
+       ? driGetDefaultVBlankFlags() : VBLANK_FLAG_NO_IRQ;
+
+#ifndef _SOLO
+   rmesa->get_ust = (PFNGLXGETUSTPROC) glXGetProcAddress( "__glXGetUST" );
+   if ( rmesa->get_ust == NULL ) 
+#endif
+   {
+      rmesa->get_ust = get_ust_nop;
+   }
+
+   (*rmesa->get_ust)( & rmesa->swap_ust );
+
+
+#if DO_DEBUG
+   RADEON_DEBUG = driParseDebugString( getenv( "RADEON_DEBUG" ),
+				       debug_control );
+#endif
+
+   if (getenv("RADEON_NO_RAST")) {
+      fprintf(stderr, "disabling 3D acceleration\n");
+      FALLBACK(rmesa, RADEON_FALLBACK_DISABLE, 1); 
+   }
+   else if (getenv("RADEON_TCL_FORCE_ENABLE")) {
+      fprintf(stderr, "Enabling TCL support...  this will probably crash\n");
+      fprintf(stderr, "         your card if it isn't capable of TCL!\n");
+      rmesa->radeonScreen->chipset |= RADEON_CHIPSET_TCL;
+   } else if (getenv("RADEON_TCL_FORCE_DISABLE") ||
+	    rmesa->dri.drmMinor < 3 ||
+	    !(rmesa->radeonScreen->chipset & RADEON_CHIPSET_TCL)) {
+      rmesa->radeonScreen->chipset &= ~RADEON_CHIPSET_TCL;
+      fprintf(stderr, "disabling TCL support\n");
+      TCL_FALLBACK(rmesa->glCtx, RADEON_TCL_FALLBACK_TCL_DISABLE, 1); 
+   }
+
+   if (rmesa->radeonScreen->chipset & RADEON_CHIPSET_TCL) {
+      if (!getenv("RADEON_NO_VTXFMT"))
+	 radeonVtxfmtInit( ctx );
+
+      _tnl_need_dlist_norm_lengths( ctx, GL_FALSE );
+   }
+   return GL_TRUE;
+}
+
+
+/* Destroy the device specific context.
+ */
+/* Destroy the Mesa and driver specific context data.
+ */
+void radeonDestroyContext( __DRIcontextPrivate *driContextPriv )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = (radeonContextPtr) driContextPriv->driverPrivate;
+   radeonContextPtr current = ctx ? RADEON_CONTEXT(ctx) : NULL;
+
+   /* check if we're deleting the currently bound context */
+   if (rmesa == current) {
+      RADEON_FIREVERTICES( rmesa );
+      _mesa_make_current2(NULL, NULL, NULL);
+   }
+
+   /* Free radeon context resources */
+   assert(rmesa); /* should never be null */
+   if ( rmesa ) {
+      GLboolean   release_texture_heaps;
+
+
+      release_texture_heaps = (rmesa->glCtx->Shared->RefCount == 1);
+      _swsetup_DestroyContext( rmesa->glCtx );
+      _tnl_DestroyContext( rmesa->glCtx );
+      _ac_DestroyContext( rmesa->glCtx );
+      _swrast_DestroyContext( rmesa->glCtx );
+
+      radeonDestroySwtcl( rmesa->glCtx );
+      radeonReleaseArrays( rmesa->glCtx, ~0 );
+      if (rmesa->dma.current.buf) {
+	 radeonReleaseDmaRegion( rmesa, &rmesa->dma.current, __FUNCTION__ );
+	 radeonFlushCmdBuf( rmesa, __FUNCTION__ );
+      }
+
+      if (!rmesa->TclFallback & RADEON_TCL_FALLBACK_TCL_DISABLE)
+	 if (!getenv("RADEON_NO_VTXFMT"))
+	    radeonVtxfmtDestroy( rmesa->glCtx );
+
+      /* free the Mesa context */
+      rmesa->glCtx->DriverCtx = NULL;
+      _mesa_destroy_context( rmesa->glCtx );
+
+      if (rmesa->state.scissor.pClipRects) {
+	 FREE(rmesa->state.scissor.pClipRects);
+	 rmesa->state.scissor.pClipRects = 0;
+      }
+
+      if ( release_texture_heaps ) {
+         /* This share group is about to go away, free our private
+          * texture object data.
+          */
+         int i;
+
+	 /* this assert is not correct, default textures are always on swap list
+	 assert( is_empty_list( & rmesa->swapped ) ); */
+
+         for ( i = 0 ; i < rmesa->nr_heaps ; i++ ) {
+	    driDestroyTextureHeap( rmesa->texture_heaps[ i ] );
+	    rmesa->texture_heaps[ i ] = NULL;
+         }
+      }
+
+      FREE( rmesa );
+   }
+}
+
+
+
+
+void
+radeonSwapBuffers( __DRIdrawablePrivate *dPriv )
+{
+
+   if (dPriv->driContextPriv && dPriv->driContextPriv->driverPrivate) {
+      radeonContextPtr rmesa;
+      GLcontext *ctx;
+      rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
+      ctx = rmesa->glCtx;
+      if (ctx->Visual.doubleBufferMode) {
+         _mesa_notifySwapBuffers( ctx );  /* flush pending rendering comands */
+
+         if ( rmesa->doPageFlip ) {
+            radeonPageFlip( dPriv );
+         }
+         else {
+            radeonCopyBuffer( dPriv );
+         }
+      }
+   }
+   else {
+      /* XXX this shouldn't be an error but we can't handle it for now */
+      _mesa_problem(NULL, "%s: drawable has no context!", __FUNCTION__);
+   }
+}
+
+
+/* Force the context `c' to be the current context and associate with it
+ * buffer `b'.
+ */
+GLboolean
+radeonMakeCurrent( __DRIcontextPrivate *driContextPriv,
+                   __DRIdrawablePrivate *driDrawPriv,
+                   __DRIdrawablePrivate *driReadPriv )
+{
+   if ( driContextPriv ) {
+      radeonContextPtr newCtx = 
+	 (radeonContextPtr) driContextPriv->driverPrivate;
+
+      if (RADEON_DEBUG & DEBUG_DRI)
+	 fprintf(stderr, "%s ctx %p\n", __FUNCTION__, newCtx->glCtx);
+
+      if ( newCtx->dri.drawable != driDrawPriv ) {
+	 newCtx->dri.drawable = driDrawPriv;
+	 radeonUpdateWindow( newCtx->glCtx );
+	 radeonUpdateViewportOffset( newCtx->glCtx );
+      }
+
+      _mesa_make_current2( newCtx->glCtx,
+			   (GLframebuffer *) driDrawPriv->driverPrivate,
+			   (GLframebuffer *) driReadPriv->driverPrivate );
+
+      if ( !newCtx->glCtx->Viewport.Width ) {
+	 _mesa_set_viewport( newCtx->glCtx, 0, 0,
+			     driDrawPriv->w, driDrawPriv->h );
+      }
+
+      if (newCtx->vb.enabled)
+	 radeonVtxfmtMakeCurrent( newCtx->glCtx );
+
+   } else {
+      if (RADEON_DEBUG & DEBUG_DRI)
+	 fprintf(stderr, "%s ctx is null\n", __FUNCTION__);
+      _mesa_make_current( 0, 0 );
+   }
+
+   if (RADEON_DEBUG & DEBUG_DRI)
+      fprintf(stderr, "End %s\n", __FUNCTION__);
+   return GL_TRUE;
+}
+
+/* Force the context `c' to be unbound from its buffer.
+ */
+GLboolean
+radeonUnbindContext( __DRIcontextPrivate *driContextPriv )
+{
+   radeonContextPtr rmesa = (radeonContextPtr) driContextPriv->driverPrivate;
+
+   if (RADEON_DEBUG & DEBUG_DRI)
+      fprintf(stderr, "%s ctx %p\n", __FUNCTION__, rmesa->glCtx);
+
+   return GL_TRUE;
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_context.h b/src/mesa/drivers/dri/radeon/radeon_context.h
new file mode 100644
index 00000000000..5f1f9659e84
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_context.h
@@ -0,0 +1,839 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_context.h,v 1.6 2002/12/16 16:18:58 dawes Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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 (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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ *   Keith Whitwell <[email protected]>
+ */
+
+#ifndef __RADEON_CONTEXT_H__
+#define __RADEON_CONTEXT_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+#include <inttypes.h>
+#include "dri_util.h"
+#include "radeon_common.h"
+#include "texmem.h"
+
+#include "macros.h"
+#include "mtypes.h"
+#include "colormac.h"
+
+struct radeon_context;
+typedef struct radeon_context radeonContextRec;
+typedef struct radeon_context *radeonContextPtr;
+
+#include "radeon_lock.h"
+#include "radeon_screen.h"
+#include "mm.h"
+
+/* Flags for software fallback cases */
+/* See correponding strings in radeon_swtcl.c */
+#define RADEON_FALLBACK_TEXTURE		0x0001
+#define RADEON_FALLBACK_DRAW_BUFFER	0x0002
+#define RADEON_FALLBACK_STENCIL		0x0004
+#define RADEON_FALLBACK_RENDER_MODE	0x0008
+#define RADEON_FALLBACK_BLEND_EQ	0x0010
+#define RADEON_FALLBACK_BLEND_FUNC	0x0020
+#define RADEON_FALLBACK_DISABLE 	0x0040
+#define RADEON_FALLBACK_BORDER_MODE	0x0080
+
+/* The blit width for texture uploads
+ */
+#define BLIT_WIDTH_BYTES 1024
+
+/* Use the templated vertex format:
+ */
+#define COLOR_IS_RGBA
+#define TAG(x) radeon##x
+#include "tnl_dd/t_dd_vertex.h"
+#undef TAG
+
+typedef void (*radeon_tri_func)( radeonContextPtr,
+				 radeonVertex *,
+				 radeonVertex *,
+				 radeonVertex * );
+
+typedef void (*radeon_line_func)( radeonContextPtr,
+				  radeonVertex *,
+				  radeonVertex * );
+
+typedef void (*radeon_point_func)( radeonContextPtr,
+				   radeonVertex * );
+
+
+struct radeon_colorbuffer_state {
+   GLuint clear;
+   GLint drawOffset, drawPitch;
+};
+
+
+struct radeon_depthbuffer_state {
+   GLuint clear;
+   GLfloat scale;
+};
+
+struct radeon_pixel_state {
+   GLint readOffset, readPitch;
+};
+
+struct radeon_scissor_state {
+   XF86DRIClipRectRec rect;
+   GLboolean enabled;
+
+   GLuint numClipRects;			/* Cliprects active */
+   GLuint numAllocedClipRects;		/* Cliprects available */
+   XF86DRIClipRectPtr pClipRects;
+};
+
+struct radeon_stencilbuffer_state {
+   GLboolean hwBuffer;
+   GLuint clear;			/* rb3d_stencilrefmask value */
+};
+
+struct radeon_stipple_state {
+   GLuint mask[32];
+};
+
+
+
+#define TEX_0   0x1
+#define TEX_1   0x2
+#define TEX_ALL 0x3
+
+typedef struct radeon_tex_obj radeonTexObj, *radeonTexObjPtr;
+
+/* Texture object in locally shared texture space.
+ */
+struct radeon_tex_obj {
+   driTextureObject   base;
+
+   GLuint bufAddr;			/* Offset to start of locally
+					   shared texture block */
+
+   GLuint dirty_state;		        /* Flags (1 per texunit) for
+					   whether or not this texobj
+					   has dirty hardware state
+					   (pp_*) that needs to be
+					   brought into the
+					   texunit. */
+
+   drmRadeonTexImage image[6][RADEON_MAX_TEXTURE_LEVELS];
+					/* Six, for the cube faces */
+
+   GLuint pp_txfilter;		        /* hardware register values */
+   GLuint pp_txformat;
+   GLuint pp_txoffset;		        /* Image location in texmem.
+					   All cube faces follow. */
+   GLuint pp_txsize;		        /* npot only */
+   GLuint pp_txpitch;		        /* npot only */
+   GLuint pp_border_color;
+   GLuint pp_cubic_faces;	        /* cube face 1,2,3,4 log2 sizes */
+
+   GLboolean  border_fallback;
+};
+
+
+struct radeon_texture_env_state {
+   radeonTexObjPtr texobj;
+   GLenum format;
+   GLenum envMode;
+};
+
+struct radeon_texture_state {
+   struct radeon_texture_env_state unit[RADEON_MAX_TEXTURE_UNITS];
+};
+
+
+struct radeon_state_atom {
+   struct radeon_state_atom *next, *prev;
+   const char *name;		         /* for debug */
+   int cmd_size;		         /* size in bytes */
+   GLuint is_tcl;
+   int *cmd;			         /* one or more cmd's */
+   int *lastcmd;			 /* one or more cmd's */
+   GLboolean (*check)( GLcontext * );    /* is this state active? */
+};
+   
+
+
+/* Trying to keep these relatively short as the variables are becoming
+ * extravagently long.  Drop the driver name prefix off the front of
+ * everything - I think we know which driver we're in by now, and keep the
+ * prefix to 3 letters unless absolutely impossible.  
+ */
+
+#define CTX_CMD_0             0
+#define CTX_PP_MISC           1
+#define CTX_PP_FOG_COLOR      2
+#define CTX_RE_SOLID_COLOR    3
+#define CTX_RB3D_BLENDCNTL    4
+#define CTX_RB3D_DEPTHOFFSET  5
+#define CTX_RB3D_DEPTHPITCH   6
+#define CTX_RB3D_ZSTENCILCNTL 7
+#define CTX_CMD_1             8
+#define CTX_PP_CNTL           9
+#define CTX_RB3D_CNTL         10
+#define CTX_RB3D_COLOROFFSET  11
+#define CTX_CMD_2             12
+#define CTX_RB3D_COLORPITCH   13
+#define CTX_STATE_SIZE        14
+
+#define SET_CMD_0               0
+#define SET_SE_CNTL             1
+#define SET_SE_COORDFMT         2
+#define SET_CMD_1               3
+#define SET_SE_CNTL_STATUS      4
+#define SET_STATE_SIZE          5
+
+#define LIN_CMD_0               0
+#define LIN_RE_LINE_PATTERN     1
+#define LIN_RE_LINE_STATE       2
+#define LIN_CMD_1               3
+#define LIN_SE_LINE_WIDTH       4
+#define LIN_STATE_SIZE          5
+
+#define MSK_CMD_0               0
+#define MSK_RB3D_STENCILREFMASK 1
+#define MSK_RB3D_ROPCNTL        2
+#define MSK_RB3D_PLANEMASK      3
+#define MSK_STATE_SIZE          4
+
+#define VPT_CMD_0           0
+#define VPT_SE_VPORT_XSCALE          1
+#define VPT_SE_VPORT_XOFFSET         2
+#define VPT_SE_VPORT_YSCALE          3
+#define VPT_SE_VPORT_YOFFSET         4
+#define VPT_SE_VPORT_ZSCALE          5
+#define VPT_SE_VPORT_ZOFFSET         6
+#define VPT_STATE_SIZE      7
+
+#define MSC_CMD_0               0
+#define MSC_RE_MISC             1
+#define MSC_STATE_SIZE          2
+
+#define TEX_CMD_0                   0
+#define TEX_PP_TXFILTER             1
+#define TEX_PP_TXFORMAT             2
+#define TEX_PP_TXOFFSET             3
+#define TEX_PP_TXCBLEND             4
+#define TEX_PP_TXABLEND             5
+#define TEX_PP_TFACTOR              6
+#define TEX_CMD_1                   7
+#define TEX_PP_BORDER_COLOR         8
+#define TEX_STATE_SIZE              9
+
+#define TXR_CMD_0                   0 /* rectangle textures */
+#define TXR_PP_TEX_SIZE             1 /* 0x1d04, 0x1d0c for NPOT! */
+#define TXR_PP_TEX_PITCH            2 /* 0x1d08, 0x1d10 for NPOT! */
+#define TXR_STATE_SIZE              3
+
+#define ZBS_CMD_0              0
+#define ZBS_SE_ZBIAS_FACTOR             1
+#define ZBS_SE_ZBIAS_CONSTANT           2
+#define ZBS_STATE_SIZE         3
+
+#define TCL_CMD_0                        0
+#define TCL_OUTPUT_VTXFMT         1
+#define TCL_OUTPUT_VTXSEL         2
+#define TCL_MATRIX_SELECT_0       3
+#define TCL_MATRIX_SELECT_1       4
+#define TCL_UCP_VERT_BLEND_CTL    5
+#define TCL_TEXTURE_PROC_CTL      6
+#define TCL_LIGHT_MODEL_CTL       7
+#define TCL_PER_LIGHT_CTL_0       8
+#define TCL_PER_LIGHT_CTL_1       9
+#define TCL_PER_LIGHT_CTL_2       10
+#define TCL_PER_LIGHT_CTL_3       11
+#define TCL_STATE_SIZE                   12
+
+#define MTL_CMD_0            0	
+#define MTL_EMMISSIVE_RED    1	
+#define MTL_EMMISSIVE_GREEN  2	
+#define MTL_EMMISSIVE_BLUE   3	
+#define MTL_EMMISSIVE_ALPHA  4	
+#define MTL_AMBIENT_RED      5
+#define MTL_AMBIENT_GREEN    6
+#define MTL_AMBIENT_BLUE     7
+#define MTL_AMBIENT_ALPHA    8
+#define MTL_DIFFUSE_RED      9
+#define MTL_DIFFUSE_GREEN    10
+#define MTL_DIFFUSE_BLUE     11
+#define MTL_DIFFUSE_ALPHA    12
+#define MTL_SPECULAR_RED     13
+#define MTL_SPECULAR_GREEN   14
+#define MTL_SPECULAR_BLUE    15
+#define MTL_SPECULAR_ALPHA   16
+#define MTL_SHININESS        17
+#define MTL_STATE_SIZE       18
+
+#define VTX_CMD_0              0
+#define VTX_SE_COORD_FMT       1
+#define VTX_STATE_SIZE         2
+
+#define MAT_CMD_0              0
+#define MAT_ELT_0              1
+#define MAT_STATE_SIZE         17
+
+#define GRD_CMD_0                  0
+#define GRD_VERT_GUARD_CLIP_ADJ    1
+#define GRD_VERT_GUARD_DISCARD_ADJ 2
+#define GRD_HORZ_GUARD_CLIP_ADJ    3
+#define GRD_HORZ_GUARD_DISCARD_ADJ 4
+#define GRD_STATE_SIZE             5
+
+/* position changes frequently when lighting in modelpos - separate
+ * out to new state item?  
+ */
+#define LIT_CMD_0                  0
+#define LIT_AMBIENT_RED            1
+#define LIT_AMBIENT_GREEN          2
+#define LIT_AMBIENT_BLUE           3
+#define LIT_AMBIENT_ALPHA          4
+#define LIT_DIFFUSE_RED            5
+#define LIT_DIFFUSE_GREEN          6
+#define LIT_DIFFUSE_BLUE           7
+#define LIT_DIFFUSE_ALPHA          8
+#define LIT_SPECULAR_RED           9
+#define LIT_SPECULAR_GREEN         10
+#define LIT_SPECULAR_BLUE          11
+#define LIT_SPECULAR_ALPHA         12
+#define LIT_POSITION_X             13
+#define LIT_POSITION_Y             14
+#define LIT_POSITION_Z             15
+#define LIT_POSITION_W             16
+#define LIT_DIRECTION_X            17
+#define LIT_DIRECTION_Y            18
+#define LIT_DIRECTION_Z            19
+#define LIT_DIRECTION_W            20
+#define LIT_ATTEN_CONST            21
+#define LIT_ATTEN_LINEAR           22
+#define LIT_ATTEN_QUADRATIC        23
+#define LIT_ATTEN_XXX              24
+#define LIT_CMD_1                  25
+#define LIT_SPOT_DCD               26
+#define LIT_SPOT_EXPONENT          27
+#define LIT_SPOT_CUTOFF            28
+#define LIT_SPECULAR_THRESH        29
+#define LIT_RANGE_CUTOFF           30 /* ? */
+#define LIT_RANGE_ATTEN            31 /* ? */
+#define LIT_STATE_SIZE             32
+
+/* Fog
+ */
+#define FOG_CMD_0      0
+#define FOG_R          1
+#define FOG_C          2
+#define FOG_D          3
+#define FOG_PAD        4
+#define FOG_STATE_SIZE 5
+
+/* UCP
+ */
+#define UCP_CMD_0      0
+#define UCP_X          1
+#define UCP_Y          2
+#define UCP_Z          3
+#define UCP_W          4
+#define UCP_STATE_SIZE 5
+
+/* GLT - Global ambient
+ */
+#define GLT_CMD_0      0
+#define GLT_RED        1
+#define GLT_GREEN      2
+#define GLT_BLUE       3
+#define GLT_ALPHA      4
+#define GLT_STATE_SIZE 5
+
+/* EYE
+ */
+#define EYE_CMD_0          0
+#define EYE_X              1
+#define EYE_Y              2
+#define EYE_Z              3
+#define EYE_RESCALE_FACTOR 4
+#define EYE_STATE_SIZE     5
+
+#define SHN_CMD_0          0
+#define SHN_SHININESS      1
+#define SHN_STATE_SIZE     2
+
+
+
+
+
+struct radeon_hw_state {
+   /* All state should be on one of these lists:
+    */
+   struct radeon_state_atom dirty; /* dirty list head placeholder */
+   struct radeon_state_atom clean; /* clean list head placeholder */
+
+   /* Hardware state, stored as cmdbuf commands:  
+    *   -- Need to doublebuffer for
+    *           - reviving state after loss of context
+    *           - eliding noop statechange loops? (except line stipple count)
+    */
+   struct radeon_state_atom ctx;
+   struct radeon_state_atom set;
+   struct radeon_state_atom lin;
+   struct radeon_state_atom msk;
+   struct radeon_state_atom vpt;
+   struct radeon_state_atom tcl;
+   struct radeon_state_atom msc;
+   struct radeon_state_atom tex[2];
+   struct radeon_state_atom zbs;
+   struct radeon_state_atom mtl; 
+   struct radeon_state_atom mat[5]; 
+   struct radeon_state_atom lit[8]; /* includes vec, scl commands */
+   struct radeon_state_atom ucp[6];
+   struct radeon_state_atom eye; /* eye pos */
+   struct radeon_state_atom grd; /* guard band clipping */
+   struct radeon_state_atom fog; 
+   struct radeon_state_atom glt; 
+   struct radeon_state_atom txr[2]; /* for NPOT */
+};
+
+struct radeon_state {
+   /* Derived state for internal purposes:
+    */
+   struct radeon_colorbuffer_state color;
+   struct radeon_depthbuffer_state depth;
+   struct radeon_pixel_state pixel;
+   struct radeon_scissor_state scissor;
+   struct radeon_stencilbuffer_state stencil;
+   struct radeon_stipple_state stipple;
+   struct radeon_texture_state texture;
+};
+
+
+/* Need refcounting on dma buffers:
+ */
+struct radeon_dma_buffer {
+   int refcount;		/* the number of retained regions in buf */
+   drmBufPtr buf;
+};
+
+#define GET_START(rvb) (rmesa->radeonScreen->agp_buffer_offset +			\
+			(rvb)->address - rmesa->dma.buf0_address +	\
+			(rvb)->start)
+
+/* A retained region, eg vertices for indexed vertices.
+ */
+struct radeon_dma_region {
+   struct radeon_dma_buffer *buf;
+   char *address;		/* == buf->address */
+   int start, end, ptr;		/* offsets from start of buf */
+   int aos_start;
+   int aos_stride;
+   int aos_size;
+};
+
+
+struct radeon_dma {
+   /* Active dma region.  Allocations for vertices and retained
+    * regions come from here.  Also used for emitting random vertices,
+    * these may be flushed by calling flush_current();
+    */
+   struct radeon_dma_region current;
+   
+   void (*flush)( radeonContextPtr );
+
+   char *buf0_address;		/* start of buf[0], for index calcs */
+   GLuint nr_released_bufs;	/* flush after so many buffers released */
+};
+
+struct radeon_dri_mirror {
+   __DRIcontextPrivate	*context;	/* DRI context */
+   __DRIscreenPrivate	*screen;	/* DRI screen */
+   __DRIdrawablePrivate	*drawable;	/* DRI drawable bound to this ctx */
+
+   drmContext hwContext;
+   drmLock *hwLock;
+   int fd;
+   int drmMinor;
+};
+
+
+#define RADEON_CMD_BUF_SZ  (8*1024) 
+
+struct radeon_store {
+   GLuint statenr;
+   GLuint primnr;
+   char cmd_buf[RADEON_CMD_BUF_SZ];
+   int cmd_used;   
+   int elts_start;
+};
+
+
+/* radeon_tcl.c
+ */
+struct radeon_tcl_info {
+   GLuint vertex_format;
+   GLint last_offset;
+   GLuint hw_primitive;
+
+   struct radeon_dma_region *aos_components[8];
+   GLuint nr_aos_components;
+
+   GLuint *Elts;
+
+   struct radeon_dma_region indexed_verts;
+   struct radeon_dma_region obj;
+   struct radeon_dma_region rgba;
+   struct radeon_dma_region spec;
+   struct radeon_dma_region fog;
+   struct radeon_dma_region tex[RADEON_MAX_TEXTURE_UNITS];
+   struct radeon_dma_region norm;
+};
+
+
+/* radeon_swtcl.c
+ */
+struct radeon_swtcl_info {
+   GLuint SetupIndex;
+   GLuint SetupNewInputs;
+   GLuint RenderIndex;
+   GLuint vertex_size;
+   GLuint vertex_stride_shift;
+   GLuint vertex_format;
+   GLubyte *verts;
+
+   /* Fallback rasterization functions
+    */
+   radeon_point_func draw_point;
+   radeon_line_func draw_line;
+   radeon_tri_func draw_tri;
+
+   GLuint hw_primitive;
+   GLenum render_primitive;
+   GLuint numverts;
+
+   struct radeon_dma_region indexed_verts;
+};
+
+
+struct radeon_ioctl {
+   GLuint vertex_offset;
+   GLuint vertex_size;
+};
+
+
+
+#define RADEON_MAX_PRIMS 64
+
+
+/* Want to keep a cache of these around.  Each is parameterized by
+ * only a single value which has only a small range.  Only expect a
+ * few, so just rescan the list each time?
+ */
+struct dynfn {
+   struct dynfn *next, *prev;
+   int key;
+   char *code;
+};
+
+struct dfn_lists {
+   struct dynfn Vertex2f;
+   struct dynfn Vertex2fv;
+   struct dynfn Vertex3f;
+   struct dynfn Vertex3fv;
+   struct dynfn Color4ub;
+   struct dynfn Color4ubv;
+   struct dynfn Color3ub;
+   struct dynfn Color3ubv;
+   struct dynfn Color4f;
+   struct dynfn Color4fv;
+   struct dynfn Color3f;
+   struct dynfn Color3fv;
+   struct dynfn SecondaryColor3ubEXT;
+   struct dynfn SecondaryColor3ubvEXT;
+   struct dynfn SecondaryColor3fEXT;
+   struct dynfn SecondaryColor3fvEXT;
+   struct dynfn Normal3f;
+   struct dynfn Normal3fv;
+   struct dynfn TexCoord2f;
+   struct dynfn TexCoord2fv;
+   struct dynfn TexCoord1f;
+   struct dynfn TexCoord1fv;
+   struct dynfn MultiTexCoord2fARB;
+   struct dynfn MultiTexCoord2fvARB;
+   struct dynfn MultiTexCoord1fARB;
+   struct dynfn MultiTexCoord1fvARB;
+};
+
+struct dfn_generators {
+   struct dynfn *(*Vertex2f)( GLcontext *, int );
+   struct dynfn *(*Vertex2fv)( GLcontext *, int );
+   struct dynfn *(*Vertex3f)( GLcontext *, int );
+   struct dynfn *(*Vertex3fv)( GLcontext *, int );
+   struct dynfn *(*Color4ub)( GLcontext *, int );
+   struct dynfn *(*Color4ubv)( GLcontext *, int );
+   struct dynfn *(*Color3ub)( GLcontext *, int );
+   struct dynfn *(*Color3ubv)( GLcontext *, int );
+   struct dynfn *(*Color4f)( GLcontext *, int );
+   struct dynfn *(*Color4fv)( GLcontext *, int );
+   struct dynfn *(*Color3f)( GLcontext *, int );
+   struct dynfn *(*Color3fv)( GLcontext *, int );
+   struct dynfn *(*SecondaryColor3ubEXT)( GLcontext *, int );
+   struct dynfn *(*SecondaryColor3ubvEXT)( GLcontext *, int );
+   struct dynfn *(*SecondaryColor3fEXT)( GLcontext *, int );
+   struct dynfn *(*SecondaryColor3fvEXT)( GLcontext *, int );
+   struct dynfn *(*Normal3f)( GLcontext *, int );
+   struct dynfn *(*Normal3fv)( GLcontext *, int );
+   struct dynfn *(*TexCoord2f)( GLcontext *, int );
+   struct dynfn *(*TexCoord2fv)( GLcontext *, int );
+   struct dynfn *(*TexCoord1f)( GLcontext *, int );
+   struct dynfn *(*TexCoord1fv)( GLcontext *, int );
+   struct dynfn *(*MultiTexCoord2fARB)( GLcontext *, int );
+   struct dynfn *(*MultiTexCoord2fvARB)( GLcontext *, int );
+   struct dynfn *(*MultiTexCoord1fARB)( GLcontext *, int );
+   struct dynfn *(*MultiTexCoord1fvARB)( GLcontext *, int );
+};
+
+
+
+struct radeon_prim {
+   GLuint start;
+   GLuint end;
+   GLuint prim;
+};
+
+struct radeon_vbinfo {
+   GLint counter, initial_counter;
+   GLint *dmaptr;
+   void (*notify)( void );
+   GLint vertex_size;
+
+   /* A maximum total of 15 elements per vertex:  3 floats for position, 3
+    * floats for normal, 4 floats for color, 4 bytes for secondary color,
+    * 2 floats for each texture unit (4 floats total).
+    * 
+    * As soon as the 3rd TMU is supported or cube maps (or 3D textures) are
+    * supported, this value will grow.
+    * 
+    * The position data is never actually stored here, so 3 elements could be
+    * trimmed out of the buffer.
+    */
+   union { float f; int i; radeon_color_t color; } vertex[15];
+
+   GLfloat *normalptr;
+   GLfloat *floatcolorptr;
+   radeon_color_t *colorptr;
+   GLfloat *floatspecptr;
+   radeon_color_t *specptr;
+   GLfloat *texcoordptr[2];
+
+   GLenum *prim;		/* &ctx->Driver.CurrentExecPrimitive */
+   GLuint primflags;
+   GLboolean enabled;		/* *_NO_VTXFMT / *_NO_TCL env vars */
+   GLboolean installed;
+   GLboolean fell_back;
+   GLboolean recheck;
+   GLint nrverts;
+   GLuint vertex_format;
+
+   GLuint installed_vertex_format;
+   GLuint installed_color_3f_sz;
+
+   struct radeon_prim primlist[RADEON_MAX_PRIMS];
+   int nrprims;
+
+   struct dfn_lists dfn_cache;
+   struct dfn_generators codegen;
+   GLvertexformat vtxfmt;
+};
+
+
+
+
+struct radeon_context {
+   GLcontext *glCtx;			/* Mesa context */
+
+   /* Driver and hardware state management
+    */
+   struct radeon_hw_state hw;
+   struct radeon_state state;
+
+   /* Texture object bookkeeping
+    */
+   unsigned              nr_heaps;
+   driTexHeap          * texture_heaps[ RADEON_NR_TEX_HEAPS ];
+   driTextureObject      swapped;
+
+
+   /* Rasterization and vertex state:
+    */
+   GLuint TclFallback;
+   GLuint Fallback;
+   GLuint NewGLState;
+
+   
+   /* Temporaries for translating away float colors:
+    */
+   struct gl_client_array UbyteColor;
+   struct gl_client_array UbyteSecondaryColor;
+
+   /* Vertex buffers
+    */
+   struct radeon_ioctl ioctl;
+   struct radeon_dma dma;
+   struct radeon_store store;
+
+   /* Page flipping
+    */
+   GLuint doPageFlip;
+
+   /* Busy waiting
+    */
+   GLuint do_usleeps;
+   GLuint do_irqs;
+   GLuint irqsEmitted;
+   drmRadeonIrqWait iw;
+
+   /* Drawable, cliprect and scissor information
+    */
+   GLuint numClipRects;			/* Cliprects for the draw buffer */
+   XF86DRIClipRectPtr pClipRects;
+   unsigned int lastStamp;
+   GLboolean lost_context;
+   radeonScreenPtr radeonScreen;	/* Screen private DRI data */
+   RADEONSAREAPrivPtr sarea;		/* Private SAREA data */
+
+   /* TCL stuff
+    */
+   GLmatrix TexGenMatrix[RADEON_MAX_TEXTURE_UNITS];
+   GLboolean recheck_texgen[RADEON_MAX_TEXTURE_UNITS];
+   GLboolean TexGenNeedNormals[RADEON_MAX_TEXTURE_UNITS];
+   GLuint TexMatEnabled;
+   GLuint TexGenEnabled;
+   GLmatrix tmpmat;
+   GLuint last_ReallyEnabled;
+
+   /* VBI
+    */
+   GLuint vbl_seq;
+   GLuint vblank_flags;
+
+   uint64_t swap_ust;
+   uint64_t swap_missed_ust;
+
+   GLuint swap_count;
+   GLuint swap_missed_count;
+
+   PFNGLXGETUSTPROC get_ust;
+
+   /* radeon_tcl.c
+    */
+   struct radeon_tcl_info tcl;
+
+   /* radeon_swtcl.c
+    */
+   struct radeon_swtcl_info swtcl;
+
+   /* radeon_vtxfmt.c
+    */
+   struct radeon_vbinfo vb;
+
+   /* Mirrors of some DRI state
+    */
+   struct radeon_dri_mirror dri;
+
+ 
+   /* Performance counters
+    */
+   GLuint boxes;			/* Draw performance boxes */
+   GLuint hardwareWentIdle;
+   GLuint c_clears;
+   GLuint c_drawWaits;
+   GLuint c_textureSwaps;
+   GLuint c_textureBytes;
+   GLuint c_vertexBuffers;
+};
+
+#define RADEON_CONTEXT(ctx)		((radeonContextPtr)(ctx->DriverCtx))
+
+
+static __inline GLuint radeonPackColor( GLuint cpp,
+					GLubyte r, GLubyte g,
+					GLubyte b, GLubyte a )
+{
+   switch ( cpp ) {
+   case 2:
+      return PACK_COLOR_565( r, g, b );
+   case 4:
+      return PACK_COLOR_8888( a, r, g, b );
+   default:
+      return 0;
+   }
+}
+
+#define RADEON_OLD_PACKETS 1
+
+
+extern void radeonDestroyContext( __DRIcontextPrivate *driContextPriv );
+extern GLboolean radeonCreateContext(const __GLcontextModes *glVisual,
+				     __DRIcontextPrivate *driContextPriv,
+				     void *sharedContextPrivate);
+extern void radeonSwapBuffers( __DRIdrawablePrivate *dPriv );
+extern GLboolean radeonMakeCurrent( __DRIcontextPrivate *driContextPriv,
+				    __DRIdrawablePrivate *driDrawPriv,
+				    __DRIdrawablePrivate *driReadPriv );
+extern GLboolean radeonUnbindContext( __DRIcontextPrivate *driContextPriv );
+
+/* ================================================================
+ * Debugging:
+ */
+#define DO_DEBUG		1
+
+#if DO_DEBUG
+extern int RADEON_DEBUG;
+#else
+#define RADEON_DEBUG		0
+#endif
+
+#define DEBUG_TEXTURE	0x001
+#define DEBUG_STATE	0x002
+#define DEBUG_IOCTL	0x004
+#define DEBUG_PRIMS	0x008
+#define DEBUG_VERTS	0x010
+#define DEBUG_FALLBACKS	0x020
+#define DEBUG_VFMT	0x040
+#define DEBUG_CODEGEN	0x080
+#define DEBUG_VERBOSE	0x100
+#define DEBUG_DRI       0x200
+#define DEBUG_DMA       0x400
+#define DEBUG_SANITY    0x800
+
+#endif
+#endif /* __RADEON_CONTEXT_H__ */
diff --git a/src/mesa/drivers/dri/radeon/radeon_ioctl.c b/src/mesa/drivers/dri/radeon/radeon_ioctl.c
new file mode 100644
index 00000000000..a388720aba1
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_ioctl.c
@@ -0,0 +1,1191 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_ioctl.c,v 1.11 2003/01/29 22:04:59 dawes Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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 (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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ *   Keith Whitwell <[email protected]>
+ */
+#include <sched.h>
+#include <errno.h>
+
+#include "glheader.h"
+#include "imports.h"
+#include "simple_list.h"
+#include "swrast/swrast.h"
+
+#include "radeon_context.h"
+#include "radeon_state.h"
+#include "radeon_ioctl.h"
+#include "radeon_tcl.h"
+#include "radeon_sanity.h"
+
+#include "radeon_macros.h"  /* for INREG() */
+
+#include "vblank.h"
+
+#define RADEON_TIMEOUT             512
+#define RADEON_IDLE_RETRY           16
+
+
+static void radeonWaitForIdle( radeonContextPtr rmesa );
+
+/* =============================================================
+ * Kernel command buffer handling
+ */
+
+static void print_state_atom( struct radeon_state_atom *state )
+{
+   int i;
+
+   fprintf(stderr, "emit %s/%d\n", state->name, state->cmd_size);
+
+   if (RADEON_DEBUG & DEBUG_VERBOSE) 
+      for (i = 0 ; i < state->cmd_size ; i++) 
+	 fprintf(stderr, "\t%s[%d]: %x\n", state->name, i, state->cmd[i]);
+
+}
+
+static void radeon_emit_state_list( radeonContextPtr rmesa, 
+				    struct radeon_state_atom *list )
+{
+   struct radeon_state_atom *state, *tmp;
+   char *dest;
+
+   /* From Felix Kuhling: similar to some other lockups, glaxium will
+    * lock with what we believe to be a normal command stream, but
+    * sprinkling some magic waits arounds allows it to run
+    * uninterrupted.  This has a slight effect on q3 framerates, but
+    * it might now be possible to remove the zbs hack, below.
+    *
+    * Felix reports that this can be narrowed down to just
+    * tcl,tex0,tex1 state, but that's pretty much every statechange,
+    * so let's just put the wait in always (unless Felix wants to
+    * narrow it down further...)
+    */
+   if (1) {
+      drmRadeonCmdHeader *cmd;
+      cmd = (drmRadeonCmdHeader *)radeonAllocCmdBuf( rmesa, sizeof(*cmd), 
+						     __FUNCTION__ );
+      cmd->wait.cmd_type = RADEON_CMD_WAIT;
+      cmd->wait.flags = RADEON_WAIT_3D;
+   }
+
+   foreach_s( state, tmp, list ) {
+      if (state->check( rmesa->glCtx )) {
+	 dest = radeonAllocCmdBuf( rmesa, state->cmd_size * 4, __FUNCTION__);
+	 memcpy( dest, state->cmd, state->cmd_size * 4);
+	 move_to_head( &(rmesa->hw.clean), state );
+	 if (RADEON_DEBUG & DEBUG_STATE) 
+	    print_state_atom( state );
+      }
+      else if (RADEON_DEBUG & DEBUG_STATE)
+	 fprintf(stderr, "skip state %s\n", state->name);
+   }
+}
+
+
+void radeonEmitState( radeonContextPtr rmesa )
+{
+   struct radeon_state_atom *state, *tmp;
+
+   if (RADEON_DEBUG & (DEBUG_STATE|DEBUG_PRIMS))
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   /* Somewhat overkill:
+    */
+   if (rmesa->lost_context) {
+      if (RADEON_DEBUG & (DEBUG_STATE|DEBUG_PRIMS|DEBUG_IOCTL))
+	 fprintf(stderr, "%s - lost context\n", __FUNCTION__); 
+
+      foreach_s( state, tmp, &(rmesa->hw.clean) ) 
+	 move_to_tail(&(rmesa->hw.dirty), state );
+
+      rmesa->lost_context = 0;
+   }
+   else if (1) {
+      /* This is a darstardly kludge to work around a lockup that I
+       * haven't otherwise figured out.
+       */
+      move_to_tail(&(rmesa->hw.dirty), &(rmesa->hw.zbs) );
+   }
+
+   if (!(rmesa->radeonScreen->chipset & RADEON_CHIPSET_TCL)) {
+     foreach_s( state, tmp, &(rmesa->hw.dirty) ) {
+       if (state->is_tcl) {
+	 move_to_head( &(rmesa->hw.clean), state );
+       }
+     }
+   }
+
+   radeon_emit_state_list( rmesa, &rmesa->hw.dirty );
+}
+
+
+
+/* Fire a section of the retained (indexed_verts) buffer as a regular
+ * primtive.  
+ */
+extern void radeonEmitVbufPrim( radeonContextPtr rmesa,
+				GLuint vertex_format,
+				GLuint primitive,
+				GLuint vertex_nr )
+{
+   drmRadeonCmdHeader *cmd;
+
+
+   assert(rmesa->dri.drmMinor >= 3); 
+   assert(!(primitive & RADEON_CP_VC_CNTL_PRIM_WALK_IND));
+   
+   radeonEmitState( rmesa );
+
+   if (RADEON_DEBUG & DEBUG_IOCTL)
+      fprintf(stderr, "%s cmd_used/4: %d\n", __FUNCTION__,
+	      rmesa->store.cmd_used/4);
+   
+#if RADEON_OLD_PACKETS
+   cmd = (drmRadeonCmdHeader *)radeonAllocCmdBuf( rmesa, 6 * sizeof(*cmd),
+						  __FUNCTION__ );
+   cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP;
+   cmd[1].i = RADEON_CP_PACKET3_3D_RNDR_GEN_INDX_PRIM | (3 << 16);
+   cmd[2].i = rmesa->ioctl.vertex_offset;
+   cmd[3].i = vertex_nr;
+   cmd[4].i = vertex_format;
+   cmd[5].i = (primitive | 
+	       RADEON_CP_VC_CNTL_PRIM_WALK_LIST |
+	       RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA |
+	       RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE |
+	       (vertex_nr << RADEON_CP_VC_CNTL_NUM_SHIFT));
+
+   if (RADEON_DEBUG & DEBUG_PRIMS)
+      fprintf(stderr, "%s: header 0x%x offt 0x%x vfmt 0x%x vfcntl %x \n",
+	      __FUNCTION__,
+	      cmd[1].i, cmd[2].i, cmd[4].i, cmd[5].i);
+#else
+   cmd = (drmRadeonCmdHeader *)radeonAllocCmdBuf( rmesa, 4 * sizeof(*cmd),
+						  __FUNCTION__ );
+   cmd[0].i = 0;
+   cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP;
+   cmd[1].i = RADEON_CP_PACKET3_3D_DRAW_VBUF | (1 << 16);
+   cmd[2].i = vertex_format;
+   cmd[3].i = (primitive | 
+	       RADEON_CP_VC_CNTL_PRIM_WALK_LIST |
+	       RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA |
+	       RADEON_CP_VC_CNTL_MAOS_ENABLE |
+	       RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE |
+	       (vertex_nr << RADEON_CP_VC_CNTL_NUM_SHIFT));
+
+
+   if (RADEON_DEBUG & DEBUG_PRIMS)
+      fprintf(stderr, "%s: header 0x%x vfmt 0x%x vfcntl %x \n",
+	      __FUNCTION__,
+	      cmd[1].i, cmd[2].i, cmd[3].i);
+#endif
+}
+
+
+void radeonFlushElts( radeonContextPtr rmesa )
+{
+   int *cmd = (int *)(rmesa->store.cmd_buf + rmesa->store.elts_start);
+   int dwords;
+#if RADEON_OLD_PACKETS
+   int nr = (rmesa->store.cmd_used - (rmesa->store.elts_start + 24)) / 2;
+#else
+   int nr = (rmesa->store.cmd_used - (rmesa->store.elts_start + 16)) / 2;
+#endif
+
+   if (RADEON_DEBUG & DEBUG_IOCTL)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   assert( rmesa->dma.flush == radeonFlushElts );
+   rmesa->dma.flush = 0;
+
+   /* Cope with odd number of elts:
+    */
+   rmesa->store.cmd_used = (rmesa->store.cmd_used + 2) & ~2;
+   dwords = (rmesa->store.cmd_used - rmesa->store.elts_start) / 4;
+
+#if RADEON_OLD_PACKETS
+   cmd[1] |= (dwords - 3) << 16;
+   cmd[5] |= nr << RADEON_CP_VC_CNTL_NUM_SHIFT;
+#else
+   cmd[1] |= (dwords - 3) << 16;
+   cmd[3] |= nr << RADEON_CP_VC_CNTL_NUM_SHIFT;
+#endif
+}
+
+
+GLushort *radeonAllocEltsOpenEnded( radeonContextPtr rmesa,
+				    GLuint vertex_format,
+				    GLuint primitive,
+				    GLuint min_nr )
+{
+   drmRadeonCmdHeader *cmd;
+   GLushort *retval;
+
+   if (RADEON_DEBUG & DEBUG_IOCTL)
+      fprintf(stderr, "%s %d\n", __FUNCTION__, min_nr);
+
+   assert(rmesa->dri.drmMinor >= 3); 
+   assert((primitive & RADEON_CP_VC_CNTL_PRIM_WALK_IND));
+   
+   radeonEmitState( rmesa );
+   
+#if RADEON_OLD_PACKETS
+   cmd = (drmRadeonCmdHeader *)radeonAllocCmdBuf( rmesa, 
+						  24 + min_nr*2,
+						  __FUNCTION__ );
+   cmd[0].i = 0;
+   cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP;
+   cmd[1].i = RADEON_CP_PACKET3_3D_RNDR_GEN_INDX_PRIM;
+   cmd[2].i = rmesa->ioctl.vertex_offset;
+   cmd[3].i = 0xffff;
+   cmd[4].i = vertex_format;
+   cmd[5].i = (primitive | 
+	       RADEON_CP_VC_CNTL_PRIM_WALK_IND |
+	       RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA |
+	       RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE);
+
+   retval = (GLushort *)(cmd+6);
+#else   
+   cmd = (drmRadeonCmdHeader *)radeonAllocCmdBuf( rmesa, 
+						  16 + min_nr*2,
+						  __FUNCTION__ );
+   cmd[0].i = 0;
+   cmd[0].header.cmd_type = RADEON_CMD_PACKET3_CLIP;
+   cmd[1].i = RADEON_CP_PACKET3_3D_DRAW_INDX;
+   cmd[2].i = vertex_format;
+   cmd[3].i = (primitive | 
+	       RADEON_CP_VC_CNTL_PRIM_WALK_IND |
+	       RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA |
+	       RADEON_CP_VC_CNTL_MAOS_ENABLE |
+	       RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE);
+
+   retval = (GLushort *)(cmd+4);
+#endif
+
+   if (RADEON_DEBUG & DEBUG_PRIMS)
+      fprintf(stderr, "%s: header 0x%x vfmt 0x%x prim %x \n",
+	      __FUNCTION__,
+	      cmd[1].i, vertex_format, primitive);
+
+   assert(!rmesa->dma.flush);
+   rmesa->glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES;
+   rmesa->dma.flush = radeonFlushElts;
+
+   rmesa->store.elts_start = ((char *)cmd) - rmesa->store.cmd_buf;
+
+   return retval;
+}
+
+
+
+void radeonEmitVertexAOS( radeonContextPtr rmesa,
+			  GLuint vertex_size,
+			  GLuint offset )
+{
+#if RADEON_OLD_PACKETS
+   rmesa->ioctl.vertex_size = vertex_size;
+   rmesa->ioctl.vertex_offset = offset;
+#else
+   drmRadeonCmdHeader *cmd;
+   assert(rmesa->dri.drmMinor >= 3); 
+
+   if (RADEON_DEBUG & (DEBUG_PRIMS|DEBUG_IOCTL))
+      fprintf(stderr, "%s:  vertex_size 0x%x offset 0x%x \n",
+	      __FUNCTION__, vertex_size, offset);
+
+   cmd = (drmRadeonCmdHeader *)radeonAllocCmdBuf( rmesa, 5 * sizeof(int),
+						  __FUNCTION__ );
+
+   cmd[0].i = 0;
+   cmd[0].header.cmd_type = RADEON_CMD_PACKET3;
+   cmd[1].i = RADEON_CP_PACKET3_3D_LOAD_VBPNTR | (2 << 16);
+   cmd[2].i = 1;
+   cmd[3].i = vertex_size | (vertex_size << 8);
+   cmd[4].i = offset;
+#endif
+}
+		       
+
+void radeonEmitAOS( radeonContextPtr rmesa,
+		    struct radeon_dma_region **component,
+		    GLuint nr,
+		    GLuint offset )
+{
+#if RADEON_OLD_PACKETS
+   assert( nr == 1 );
+   assert( component[0]->aos_size == component[0]->aos_stride );
+   rmesa->ioctl.vertex_size = component[0]->aos_size;
+   rmesa->ioctl.vertex_offset = 
+      (component[0]->aos_start + offset * component[0]->aos_stride * 4);
+#else
+   drmRadeonCmdHeader *cmd;
+   int sz = 3 + (nr/2 * 3) + (nr & 1) * 2;
+   int i;
+   int *tmp;
+
+   if (RADEON_DEBUG & DEBUG_IOCTL)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   assert(rmesa->dri.drmMinor >= 3); 
+
+   cmd = (drmRadeonCmdHeader *)radeonAllocCmdBuf( rmesa, sz * sizeof(int),
+						  __FUNCTION__ );
+   cmd[0].i = 0;
+   cmd[0].header.cmd_type = RADEON_CMD_PACKET3;
+   cmd[1].i = RADEON_CP_PACKET3_3D_LOAD_VBPNTR | ((sz-3) << 16);
+   cmd[2].i = nr;
+   tmp = &cmd[0].i;
+   cmd += 3;
+
+   for (i = 0 ; i < nr ; i++) {
+      if (i & 1) {
+	 cmd[0].i |= ((component[i]->aos_stride << 24) | 
+		      (component[i]->aos_size << 16));
+	 cmd[2].i = (component[i]->aos_start + 
+		     offset * component[i]->aos_stride * 4);
+	 cmd += 3;
+      }
+      else {
+	 cmd[0].i = ((component[i]->aos_stride << 8) | 
+		     (component[i]->aos_size << 0));
+	 cmd[1].i = (component[i]->aos_start + 
+		     offset * component[i]->aos_stride * 4);
+      }
+   }
+
+   if (RADEON_DEBUG & DEBUG_VERTS) {
+      fprintf(stderr, "%s:\n", __FUNCTION__);
+      for (i = 0 ; i < sz ; i++)
+	 fprintf(stderr, "   %d: %x\n", i, tmp[i]);
+   }
+#endif
+}
+
+/* using already shifted color_fmt! */
+void radeonEmitBlit( radeonContextPtr rmesa, /* FIXME: which drmMinor is required? */
+		   GLuint color_fmt,
+		   GLuint src_pitch,
+		   GLuint src_offset,
+		   GLuint dst_pitch,
+		   GLuint dst_offset,
+		   GLint srcx, GLint srcy,
+		   GLint dstx, GLint dsty,
+		   GLuint w, GLuint h )
+{
+   drmRadeonCmdHeader *cmd;
+
+   if (RADEON_DEBUG & DEBUG_IOCTL)
+      fprintf(stderr, "%s src %x/%x %d,%d dst: %x/%x %d,%d sz: %dx%d\n",
+	      __FUNCTION__, 
+	      src_pitch, src_offset, srcx, srcy,
+	      dst_pitch, dst_offset, dstx, dsty,
+	      w, h);
+
+   assert( (src_pitch & 63) == 0 );
+   assert( (dst_pitch & 63) == 0 );
+   assert( (src_offset & 1023) == 0 ); 
+   assert( (dst_offset & 1023) == 0 ); 
+   assert( w < (1<<16) );
+   assert( h < (1<<16) );
+
+   cmd = (drmRadeonCmdHeader *)radeonAllocCmdBuf( rmesa, 8 * sizeof(int),
+						  __FUNCTION__ );
+
+
+   cmd[0].i = 0;
+   cmd[0].header.cmd_type = RADEON_CMD_PACKET3;
+   cmd[1].i = RADEON_CP_PACKET3_CNTL_BITBLT_MULTI | (5 << 16);
+   cmd[2].i = (RADEON_GMC_SRC_PITCH_OFFSET_CNTL |
+	       RADEON_GMC_DST_PITCH_OFFSET_CNTL |
+	       RADEON_GMC_BRUSH_NONE |
+	       color_fmt |
+	       RADEON_GMC_SRC_DATATYPE_COLOR |
+	       RADEON_ROP3_S |
+	       RADEON_DP_SRC_SOURCE_MEMORY |
+	       RADEON_GMC_CLR_CMP_CNTL_DIS |
+	       RADEON_GMC_WR_MSK_DIS );
+
+   cmd[3].i = ((src_pitch/64)<<22) | (src_offset >> 10);
+   cmd[4].i = ((dst_pitch/64)<<22) | (dst_offset >> 10);
+   cmd[5].i = (srcx << 16) | srcy;
+   cmd[6].i = (dstx << 16) | dsty; /* dst */
+   cmd[7].i = (w << 16) | h;
+}
+
+
+void radeonEmitWait( radeonContextPtr rmesa, GLuint flags )
+{
+   if (rmesa->dri.drmMinor >= 6) {
+      drmRadeonCmdHeader *cmd;
+
+      assert( !(flags & ~(RADEON_WAIT_2D|RADEON_WAIT_3D)) );
+      
+      cmd = (drmRadeonCmdHeader *)radeonAllocCmdBuf( rmesa, 1 * sizeof(int),
+						   __FUNCTION__ );
+      cmd[0].i = 0;
+      cmd[0].wait.cmd_type = RADEON_CMD_WAIT;
+      cmd[0].wait.flags = flags;
+   }
+}
+
+
+static int radeonFlushCmdBufLocked( radeonContextPtr rmesa, 
+				    const char * caller )
+{
+   int ret, i;
+   drmRadeonCmdBuffer cmd;
+
+   if (RADEON_DEBUG & DEBUG_IOCTL) {
+      fprintf(stderr, "%s from %s\n", __FUNCTION__, caller); 
+
+      if (RADEON_DEBUG & DEBUG_VERBOSE) 
+	 for (i = 0 ; i < rmesa->store.cmd_used ; i += 4 )
+	    fprintf(stderr, "%d: %x\n", i/4, 
+		    *(int *)(&rmesa->store.cmd_buf[i]));
+   }
+
+   if (RADEON_DEBUG & DEBUG_DMA)
+      fprintf(stderr, "%s: Releasing %d buffers\n", __FUNCTION__,
+	      rmesa->dma.nr_released_bufs);
+
+
+   if (RADEON_DEBUG & DEBUG_SANITY) {
+      if (rmesa->state.scissor.enabled) 
+	 ret = radeonSanityCmdBuffer( rmesa, 
+				      rmesa->state.scissor.numClipRects,
+				      rmesa->state.scissor.pClipRects);
+      else
+	 ret = radeonSanityCmdBuffer( rmesa, 
+				      rmesa->numClipRects,
+				      rmesa->pClipRects);
+      if (ret) {
+	 fprintf(stderr, "drmSanityCommandWrite: %d\n", ret);	 
+	 goto out;
+      }
+   }
+
+
+   cmd.bufsz = rmesa->store.cmd_used;
+   cmd.buf = rmesa->store.cmd_buf;
+
+   if (rmesa->state.scissor.enabled) {
+      cmd.nbox = rmesa->state.scissor.numClipRects;
+      cmd.boxes = (drmClipRect *)rmesa->state.scissor.pClipRects;
+   } else {
+      cmd.nbox = rmesa->numClipRects;
+      cmd.boxes = (drmClipRect *)rmesa->pClipRects;
+   }
+
+   ret = drmCommandWrite( rmesa->dri.fd,
+			  DRM_RADEON_CMDBUF,
+			  &cmd, sizeof(cmd) );
+
+   if (ret)
+      fprintf(stderr, "drmCommandWrite: %d\n", ret);
+
+ out:
+   rmesa->store.primnr = 0;
+   rmesa->store.statenr = 0;
+   rmesa->store.cmd_used = 0;
+   rmesa->dma.nr_released_bufs = 0;
+   rmesa->lost_context = 1;	
+   return ret;
+}
+
+
+/* Note: does not emit any commands to avoid recursion on
+ * radeonAllocCmdBuf.
+ */
+void radeonFlushCmdBuf( radeonContextPtr rmesa, const char *caller )
+{
+   int ret;
+
+	      
+   assert (rmesa->dri.drmMinor >= 3);
+
+   LOCK_HARDWARE( rmesa );
+
+   ret = radeonFlushCmdBufLocked( rmesa, caller );
+
+   UNLOCK_HARDWARE( rmesa );
+
+   if (ret) {
+      fprintf(stderr, "drmRadeonCmdBuffer: %d (exiting)\n", ret);
+      exit(ret);
+   }
+}
+
+/* =============================================================
+ * Hardware vertex buffer handling
+ */
+
+
+void radeonRefillCurrentDmaRegion( radeonContextPtr rmesa )
+{
+   struct radeon_dma_buffer *dmabuf;
+   int fd = rmesa->dri.fd;
+   int index = 0;
+   int size = 0;
+   drmDMAReq dma;
+   int ret;
+
+   if (RADEON_DEBUG & (DEBUG_IOCTL|DEBUG_DMA))
+      fprintf(stderr, "%s\n", __FUNCTION__);  
+
+   if (rmesa->dma.flush) {
+      rmesa->dma.flush( rmesa );
+   }
+
+   if (rmesa->dma.current.buf)
+      radeonReleaseDmaRegion( rmesa, &rmesa->dma.current, __FUNCTION__ );
+
+   if (rmesa->dma.nr_released_bufs > 4)
+      radeonFlushCmdBuf( rmesa, __FUNCTION__ );
+
+   dma.context = rmesa->dri.hwContext;
+   dma.send_count = 0;
+   dma.send_list = NULL;
+   dma.send_sizes = NULL;
+   dma.flags = 0;
+   dma.request_count = 1;
+   dma.request_size = RADEON_BUFFER_SIZE;
+   dma.request_list = &index;
+   dma.request_sizes = &size;
+   dma.granted_count = 0;
+
+   LOCK_HARDWARE(rmesa);	/* no need to validate */
+
+   ret = drmDMA( fd, &dma );
+      
+   if (ret != 0) {
+      /* Free some up this way?
+       */
+      if (rmesa->dma.nr_released_bufs) {
+	 radeonFlushCmdBufLocked( rmesa, __FUNCTION__ );
+      }
+      
+      if (RADEON_DEBUG & DEBUG_DMA)
+	 fprintf(stderr, "Waiting for buffers\n");
+
+      radeonWaitForIdleLocked( rmesa );
+      ret = drmDMA( fd, &dma );
+
+      if ( ret != 0 ) {
+	 UNLOCK_HARDWARE( rmesa );
+	 fprintf( stderr, "Error: Could not get dma buffer... exiting\n" );
+	 exit( -1 );
+      }
+   }
+
+   UNLOCK_HARDWARE(rmesa);
+
+   if (RADEON_DEBUG & DEBUG_DMA)
+      fprintf(stderr, "Allocated buffer %d\n", index);
+
+   dmabuf = CALLOC_STRUCT( radeon_dma_buffer );
+   dmabuf->buf = &rmesa->radeonScreen->buffers->list[index];
+   dmabuf->refcount = 1;
+
+   rmesa->dma.current.buf = dmabuf;
+   rmesa->dma.current.address = dmabuf->buf->address;
+   rmesa->dma.current.end = dmabuf->buf->total;
+   rmesa->dma.current.start = 0;
+   rmesa->dma.current.ptr = 0;
+
+   rmesa->c_vertexBuffers++;
+}
+
+void radeonReleaseDmaRegion( radeonContextPtr rmesa,
+			     struct radeon_dma_region *region,
+			     const char *caller )
+{
+   if (RADEON_DEBUG & DEBUG_IOCTL)
+      fprintf(stderr, "%s from %s\n", __FUNCTION__, caller); 
+   
+   if (!region->buf)
+      return;
+
+   if (rmesa->dma.flush)
+      rmesa->dma.flush( rmesa );
+
+   if (--region->buf->refcount == 0) {
+      drmRadeonCmdHeader *cmd;
+
+      if (RADEON_DEBUG & (DEBUG_IOCTL|DEBUG_DMA))
+	 fprintf(stderr, "%s -- DISCARD BUF %d\n", __FUNCTION__,
+		 region->buf->buf->idx);  
+      
+      cmd = (drmRadeonCmdHeader *)radeonAllocCmdBuf( rmesa, sizeof(*cmd), 
+						     __FUNCTION__ );
+      cmd->dma.cmd_type = RADEON_CMD_DMA_DISCARD;
+      cmd->dma.buf_idx = region->buf->buf->idx;
+      FREE(region->buf);
+      rmesa->dma.nr_released_bufs++;
+   }
+
+   region->buf = 0;
+   region->start = 0;
+}
+
+/* Allocates a region from rmesa->dma.current.  If there isn't enough
+ * space in current, grab a new buffer (and discard what was left of current)
+ */
+void radeonAllocDmaRegion( radeonContextPtr rmesa, 
+			   struct radeon_dma_region *region,
+			   int bytes,
+			   int alignment )
+{
+   if (RADEON_DEBUG & DEBUG_IOCTL)
+      fprintf(stderr, "%s %d\n", __FUNCTION__, bytes);
+
+   if (rmesa->dma.flush)
+      rmesa->dma.flush( rmesa );
+
+   if (region->buf)
+      radeonReleaseDmaRegion( rmesa, region, __FUNCTION__ );
+
+   alignment--;
+   rmesa->dma.current.start = rmesa->dma.current.ptr = 
+      (rmesa->dma.current.ptr + alignment) & ~alignment;
+
+   if ( rmesa->dma.current.ptr + bytes > rmesa->dma.current.end ) 
+      radeonRefillCurrentDmaRegion( rmesa );
+
+   region->start = rmesa->dma.current.start;
+   region->ptr = rmesa->dma.current.start;
+   region->end = rmesa->dma.current.start + bytes;
+   region->address = rmesa->dma.current.address;
+   region->buf = rmesa->dma.current.buf;
+   region->buf->refcount++;
+
+   rmesa->dma.current.ptr += bytes; /* bug - if alignment > 7 */
+   rmesa->dma.current.start = 
+      rmesa->dma.current.ptr = (rmesa->dma.current.ptr + 0x7) & ~0x7;  
+
+   if ( rmesa->dri.drmMinor < 3 ) 
+      radeonRefillCurrentDmaRegion( rmesa );
+}
+
+void radeonAllocDmaRegionVerts( radeonContextPtr rmesa, 
+				struct radeon_dma_region *region,
+				int numverts,
+				int vertsize,
+				int alignment )
+{
+   radeonAllocDmaRegion( rmesa, region, vertsize * numverts, alignment );
+}
+
+/* ================================================================
+ * SwapBuffers with client-side throttling
+ */
+
+static CARD32 radeonGetLastFrame (radeonContextPtr rmesa) 
+{
+   unsigned char *RADEONMMIO = rmesa->radeonScreen->mmio.map;
+   int ret;
+   CARD32 frame;
+
+   if (rmesa->dri.screen->drmMinor >= 4) {
+      drmRadeonGetParam gp;
+
+      gp.param = RADEON_PARAM_LAST_FRAME;
+      gp.value = &frame;
+      ret = drmCommandWriteRead( rmesa->dri.fd, DRM_RADEON_GETPARAM,
+				 &gp, sizeof(gp) );
+   } 
+   else
+      ret = -EINVAL;
+
+#ifndef __alpha__
+   if ( ret == -EINVAL ) {
+      frame = INREG( RADEON_LAST_FRAME_REG );
+      ret = 0;
+   } 
+#endif
+   if ( ret ) {
+      fprintf( stderr, "%s: drmRadeonGetParam: %d\n", __FUNCTION__, ret );
+      exit(1);
+   }
+
+   return frame;
+}
+
+static void radeonEmitIrqLocked( radeonContextPtr rmesa )
+{
+   drmRadeonIrqEmit ie;
+   int ret;
+
+   ie.irq_seq = &rmesa->iw.irq_seq;
+   ret = drmCommandWriteRead( rmesa->dri.fd, DRM_RADEON_IRQ_EMIT, 
+			      &ie, sizeof(ie) );
+   if ( ret ) {
+      fprintf( stderr, "%s: drmRadeonIrqEmit: %d\n", __FUNCTION__, ret );
+      exit(1);
+   }
+}
+
+
+static void radeonWaitIrq( radeonContextPtr rmesa )
+{
+   int ret;
+
+   do {
+      ret = drmCommandWrite( rmesa->dri.fd, DRM_RADEON_IRQ_WAIT,
+			     &rmesa->iw, sizeof(rmesa->iw) );
+   } while (ret && (errno == EINTR || errno == EAGAIN));
+
+   if ( ret ) {
+      fprintf( stderr, "%s: drmRadeonIrqWait: %d\n", __FUNCTION__, ret );
+      exit(1);
+   }
+}
+
+
+static void radeonWaitForFrameCompletion( radeonContextPtr rmesa )
+{
+   RADEONSAREAPrivPtr sarea = rmesa->sarea;
+
+   if (rmesa->do_irqs) {
+      if (radeonGetLastFrame(rmesa) < sarea->last_frame) {
+	 if (!rmesa->irqsEmitted) {
+	    while (radeonGetLastFrame (rmesa) < sarea->last_frame)
+	       ;
+	 }
+	 else {
+	    UNLOCK_HARDWARE( rmesa ); 
+	    radeonWaitIrq( rmesa );	
+	    LOCK_HARDWARE( rmesa ); 
+	 }
+	 rmesa->irqsEmitted = 10;
+      }
+
+      if (rmesa->irqsEmitted) {
+	 radeonEmitIrqLocked( rmesa );
+	 rmesa->irqsEmitted--;
+      }
+   } 
+   else {
+      while (radeonGetLastFrame (rmesa) < sarea->last_frame) {
+	 UNLOCK_HARDWARE( rmesa ); 
+	 if (rmesa->do_usleeps) 
+	    DO_USLEEP( 1 );
+	 LOCK_HARDWARE( rmesa ); 
+      }
+   }
+}
+
+/* Copy the back color buffer to the front color buffer.
+ */
+void radeonCopyBuffer( const __DRIdrawablePrivate *dPriv )
+{
+   radeonContextPtr rmesa;
+   GLint nbox, i, ret;
+   GLboolean   missed_target;
+   uint64_t     ust;
+
+   assert(dPriv);
+   assert(dPriv->driContextPriv);
+   assert(dPriv->driContextPriv->driverPrivate);
+
+   rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
+
+   if ( RADEON_DEBUG & DEBUG_IOCTL ) {
+      fprintf( stderr, "\n%s( %p )\n\n", __FUNCTION__, rmesa->glCtx );
+   }
+
+   RADEON_FIREVERTICES( rmesa );
+   LOCK_HARDWARE( rmesa );
+
+   /* Throttle the frame rate -- only allow one pending swap buffers
+    * request at a time.
+    */
+   radeonWaitForFrameCompletion( rmesa );
+   UNLOCK_HARDWARE( rmesa );
+   driWaitForVBlank( dPriv, & rmesa->vbl_seq, rmesa->vblank_flags, & missed_target );
+   LOCK_HARDWARE( rmesa );
+
+   nbox = dPriv->numClipRects; /* must be in locked region */
+
+   for ( i = 0 ; i < nbox ; ) {
+      GLint nr = MIN2( i + RADEON_NR_SAREA_CLIPRECTS , nbox );
+      XF86DRIClipRectPtr box = dPriv->pClipRects;
+      XF86DRIClipRectPtr b = rmesa->sarea->boxes;
+      GLint n = 0;
+
+      for ( ; i < nr ; i++ ) {
+	 *b++ = box[i];
+	 n++;
+      }
+      rmesa->sarea->nbox = n;
+
+      ret = drmCommandNone( rmesa->dri.fd, DRM_RADEON_SWAP );
+
+      if ( ret ) {
+	 fprintf( stderr, "DRM_RADEON_SWAP_BUFFERS: return = %d\n", ret );
+	 UNLOCK_HARDWARE( rmesa );
+	 exit( 1 );
+      }
+   }
+
+   UNLOCK_HARDWARE( rmesa );
+   rmesa->swap_count++;
+   (*rmesa->get_ust)( & ust );
+   if ( missed_target ) {
+      rmesa->swap_missed_count++;
+      rmesa->swap_missed_ust = ust - rmesa->swap_ust;
+   }
+
+   rmesa->swap_ust = ust;
+}
+
+void radeonPageFlip( const __DRIdrawablePrivate *dPriv )
+{
+   radeonContextPtr rmesa;
+   GLint ret;
+   GLboolean   missed_target;
+
+   assert(dPriv);
+   assert(dPriv->driContextPriv);
+   assert(dPriv->driContextPriv->driverPrivate);
+
+   rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
+
+   if ( RADEON_DEBUG & DEBUG_IOCTL ) {
+      fprintf(stderr, "%s: pfCurrentPage: %d\n", __FUNCTION__,
+	      rmesa->sarea->pfCurrentPage);
+   }
+
+   RADEON_FIREVERTICES( rmesa );
+   LOCK_HARDWARE( rmesa );
+
+   /* Need to do this for the perf box placement:
+    */
+   if (dPriv->numClipRects)
+   {
+      XF86DRIClipRectPtr box = dPriv->pClipRects;
+      XF86DRIClipRectPtr b = rmesa->sarea->boxes;
+      b[0] = box[0];
+      rmesa->sarea->nbox = 1;
+   }
+
+   /* Throttle the frame rate -- only allow a few pending swap buffers
+    * request at a time.
+    */
+   radeonWaitForFrameCompletion( rmesa );
+   UNLOCK_HARDWARE( rmesa );
+   driWaitForVBlank( dPriv, & rmesa->vbl_seq, rmesa->vblank_flags, & missed_target );
+   if ( missed_target ) {
+      rmesa->swap_missed_count++;
+      (void) (*rmesa->get_ust)( & rmesa->swap_missed_ust );
+   }
+   LOCK_HARDWARE( rmesa );
+
+   ret = drmCommandNone( rmesa->dri.fd, DRM_RADEON_FLIP );
+
+   UNLOCK_HARDWARE( rmesa );
+
+   if ( ret ) {
+      fprintf( stderr, "DRM_RADEON_FLIP: return = %d\n", ret );
+      exit( 1 );
+   }
+
+   rmesa->swap_count++;
+   (void) (*rmesa->get_ust)( & rmesa->swap_ust );
+
+   if ( rmesa->sarea->pfCurrentPage == 1 ) {
+	 rmesa->state.color.drawOffset = rmesa->radeonScreen->frontOffset;
+	 rmesa->state.color.drawPitch  = rmesa->radeonScreen->frontPitch;
+   } else {
+	 rmesa->state.color.drawOffset = rmesa->radeonScreen->backOffset;
+	 rmesa->state.color.drawPitch  = rmesa->radeonScreen->backPitch;
+   }
+
+   RADEON_STATECHANGE( rmesa, ctx );
+   rmesa->hw.ctx.cmd[CTX_RB3D_COLOROFFSET] = rmesa->state.color.drawOffset;
+   rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH]  = rmesa->state.color.drawPitch;
+}
+
+
+/* ================================================================
+ * Buffer clear
+ */
+#define RADEON_MAX_CLEARS	256
+
+static void radeonClear( GLcontext *ctx, GLbitfield mask, GLboolean all,
+			 GLint cx, GLint cy, GLint cw, GLint ch )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
+   RADEONSAREAPrivPtr sarea = rmesa->sarea;
+   unsigned char *RADEONMMIO = rmesa->radeonScreen->mmio.map;
+   CARD32 clear;
+   GLuint flags = 0;
+   GLuint color_mask = 0;
+   GLint ret, i;
+
+   if ( RADEON_DEBUG & DEBUG_IOCTL ) {
+      fprintf( stderr, "%s:  all=%d cx=%d cy=%d cw=%d ch=%d\n",
+	       __FUNCTION__, all, cx, cy, cw, ch );
+   }
+
+   radeonEmitState( rmesa );
+
+   /* Need to cope with lostcontext here as kernel relies on
+    * some residual state:
+    */
+   RADEON_FIREVERTICES( rmesa ); 
+
+   if ( mask & DD_FRONT_LEFT_BIT ) {
+      flags |= RADEON_FRONT;
+      color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK];
+      mask &= ~DD_FRONT_LEFT_BIT;
+   }
+
+   if ( mask & DD_BACK_LEFT_BIT ) {
+      flags |= RADEON_BACK;
+      color_mask = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK];
+      mask &= ~DD_BACK_LEFT_BIT;
+   }
+
+   if ( mask & DD_DEPTH_BIT ) {
+      if ( ctx->Depth.Mask ) flags |= RADEON_DEPTH; /* FIXME: ??? */
+      mask &= ~DD_DEPTH_BIT;
+   }
+
+   if ( (mask & DD_STENCIL_BIT) && rmesa->state.stencil.hwBuffer ) {
+      flags |= RADEON_STENCIL;
+      mask &= ~DD_STENCIL_BIT;
+   }
+
+   if ( mask ) {
+      if (RADEON_DEBUG & DEBUG_FALLBACKS)
+	 fprintf(stderr, "%s: swrast clear, mask: %x\n", __FUNCTION__, mask);
+      _swrast_Clear( ctx, mask, all, cx, cy, cw, ch );
+   }
+
+   if ( !flags ) 
+      return;
+
+
+   /* Flip top to bottom */
+   cx += dPriv->x;
+   cy  = dPriv->y + dPriv->h - cy - ch;
+
+   LOCK_HARDWARE( rmesa );
+
+   /* Throttle the number of clear ioctls we do.
+    */
+   while ( 1 ) {
+      int ret;
+
+      if (rmesa->dri.screen->drmMinor >= 4) {
+	drmRadeonGetParam gp;
+
+	gp.param = RADEON_PARAM_LAST_CLEAR;
+	gp.value = &clear;
+	ret = drmCommandWriteRead( rmesa->dri.fd,
+				   DRM_RADEON_GETPARAM, &gp, sizeof(gp) );
+      } else
+	ret = -EINVAL;
+
+#ifndef __alpha__
+      if ( ret == -EINVAL ) {
+	 clear = INREG( RADEON_LAST_CLEAR_REG );
+	 ret = 0;
+      }
+#endif
+      if ( ret ) {
+	 fprintf( stderr, "%s: drmRadeonGetParam: %d\n", __FUNCTION__, ret );
+	 exit(1);
+      }
+      if ( RADEON_DEBUG & DEBUG_IOCTL ) {
+	 fprintf( stderr, "%s( %d )\n", __FUNCTION__, (int)clear );
+	 if ( ret ) fprintf( stderr, " ( RADEON_LAST_CLEAR register read directly )\n" );
+      }
+
+      if ( sarea->last_clear - clear <= RADEON_MAX_CLEARS ) {
+	 break;
+      }
+
+      if ( rmesa->do_usleeps ) {
+	 UNLOCK_HARDWARE( rmesa );
+	 DO_USLEEP( 1 );
+	 LOCK_HARDWARE( rmesa );
+      }
+   }
+
+   for ( i = 0 ; i < dPriv->numClipRects ; ) {
+      GLint nr = MIN2( i + RADEON_NR_SAREA_CLIPRECTS, dPriv->numClipRects );
+      XF86DRIClipRectPtr box = dPriv->pClipRects;
+      XF86DRIClipRectPtr b = rmesa->sarea->boxes;
+      drmRadeonClearType clear;
+      drmRadeonClearRect depth_boxes[RADEON_NR_SAREA_CLIPRECTS];
+      GLint n = 0;
+
+      if ( !all ) {
+	 for ( ; i < nr ; i++ ) {
+	    GLint x = box[i].x1;
+	    GLint y = box[i].y1;
+	    GLint w = box[i].x2 - x;
+	    GLint h = box[i].y2 - y;
+
+	    if ( x < cx ) w -= cx - x, x = cx;
+	    if ( y < cy ) h -= cy - y, y = cy;
+	    if ( x + w > cx + cw ) w = cx + cw - x;
+	    if ( y + h > cy + ch ) h = cy + ch - y;
+	    if ( w <= 0 ) continue;
+	    if ( h <= 0 ) continue;
+
+	    b->x1 = x;
+	    b->y1 = y;
+	    b->x2 = x + w;
+	    b->y2 = y + h;
+	    b++;
+	    n++;
+	 }
+      } else {
+	 for ( ; i < nr ; i++ ) {
+	    *b++ = box[i];
+	    n++;
+	 }
+      }
+
+      rmesa->sarea->nbox = n;
+
+      clear.flags       = flags;
+      clear.clear_color = rmesa->state.color.clear;
+      clear.clear_depth = rmesa->state.depth.clear;
+      clear.color_mask  = rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK];
+      clear.depth_mask  = rmesa->state.stencil.clear;
+      clear.depth_boxes = depth_boxes;
+
+      n--;
+      b = rmesa->sarea->boxes;
+      for ( ; n >= 0 ; n-- ) {
+	 depth_boxes[n].f[RADEON_CLEAR_X1] = (float)b[n].x1;
+	 depth_boxes[n].f[RADEON_CLEAR_Y1] = (float)b[n].y1;
+	 depth_boxes[n].f[RADEON_CLEAR_X2] = (float)b[n].x2;
+	 depth_boxes[n].f[RADEON_CLEAR_Y2] = (float)b[n].y2;
+	 depth_boxes[n].f[RADEON_CLEAR_DEPTH] = 
+	    (float)rmesa->state.depth.clear;
+      }
+
+      ret = drmCommandWrite( rmesa->dri.fd, DRM_RADEON_CLEAR,
+			     &clear, sizeof(drmRadeonClearType));
+
+      if ( ret ) {
+	 UNLOCK_HARDWARE( rmesa );
+	 fprintf( stderr, "DRM_RADEON_CLEAR: return = %d\n", ret );
+	 exit( 1 );
+      }
+   }
+
+   UNLOCK_HARDWARE( rmesa );
+}
+
+
+void radeonWaitForIdleLocked( radeonContextPtr rmesa )
+{
+    int fd = rmesa->dri.fd;
+    int to = 0;
+    int ret, i = 0;
+
+    rmesa->c_drawWaits++;
+
+    do {
+        do {
+            ret = drmCommandNone( fd, DRM_RADEON_CP_IDLE);
+        } while ( ret && errno == EBUSY && i++ < RADEON_IDLE_RETRY );
+    } while ( ( ret == -EBUSY ) && ( to++ < RADEON_TIMEOUT ) );
+
+    if ( ret < 0 ) {
+	UNLOCK_HARDWARE( rmesa );
+	fprintf( stderr, "Error: Radeon timed out... exiting\n" );
+	exit( -1 );
+    }
+}
+
+
+static void radeonWaitForIdle( radeonContextPtr rmesa )
+{
+   LOCK_HARDWARE(rmesa);
+   radeonWaitForIdleLocked( rmesa );
+   UNLOCK_HARDWARE(rmesa);
+}
+
+
+void radeonFlush( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+
+   if (RADEON_DEBUG & DEBUG_IOCTL)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   if (rmesa->dma.flush)
+      rmesa->dma.flush( rmesa );
+
+   if (rmesa->dri.drmMinor >= 3) {
+      if (!is_empty_list(&rmesa->hw.dirty)) 
+	 radeonEmitState( rmesa );
+   
+      if (rmesa->store.cmd_used)
+	 radeonFlushCmdBuf( rmesa, __FUNCTION__ );
+   }
+}
+
+/* Make sure all commands have been sent to the hardware and have
+ * completed processing.
+ */
+void radeonFinish( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeonFlush( ctx );
+
+   if (rmesa->do_irqs) {
+      LOCK_HARDWARE( rmesa );
+      radeonEmitIrqLocked( rmesa );
+      UNLOCK_HARDWARE( rmesa );
+      radeonWaitIrq( rmesa );
+   }
+   else
+      radeonWaitForIdle( rmesa );
+}
+
+
+void radeonInitIoctlFuncs( GLcontext *ctx )
+{
+    ctx->Driver.Clear = radeonClear;
+    ctx->Driver.Finish = radeonFinish;
+    ctx->Driver.Flush = radeonFlush;
+}
+
diff --git a/src/mesa/drivers/dri/radeon/radeon_ioctl.h b/src/mesa/drivers/dri/radeon/radeon_ioctl.h
new file mode 100644
index 00000000000..3f6e1751cf6
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_ioctl.h
@@ -0,0 +1,188 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_ioctl.h,v 1.6 2002/12/16 16:18:58 dawes Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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 (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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ */
+
+#ifndef __RADEON_IOCTL_H__
+#define __RADEON_IOCTL_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+#include "simple_list.h"
+#include "radeon_lock.h"
+
+
+extern void radeonEmitState( radeonContextPtr rmesa );
+extern void radeonEmitVertexAOS( radeonContextPtr rmesa,
+				 GLuint vertex_size,
+				 GLuint offset );
+
+extern void radeonEmitVbufPrim( radeonContextPtr rmesa,
+				GLuint vertex_format,
+				GLuint primitive,
+				GLuint vertex_nr );
+
+extern void radeonFlushElts( radeonContextPtr rmesa );
+
+extern GLushort *radeonAllocEltsOpenEnded( radeonContextPtr rmesa,
+					   GLuint vertex_format,
+					   GLuint primitive,
+					   GLuint min_nr );
+
+extern void radeonEmitAOS( radeonContextPtr rmesa,
+			   struct radeon_dma_region **regions,
+			   GLuint n,
+			   GLuint offset );
+
+extern void radeonEmitBlit( radeonContextPtr rmesa,
+			    GLuint color_fmt,
+			    GLuint src_pitch,
+			    GLuint src_offset,
+			    GLuint dst_pitch,
+			    GLuint dst_offset,
+			    GLint srcx, GLint srcy,
+			    GLint dstx, GLint dsty,
+			    GLuint w, GLuint h );
+
+extern void radeonEmitWait( radeonContextPtr rmesa, GLuint flags );
+
+extern void radeonFlushCmdBuf( radeonContextPtr rmesa, const char * );
+extern void radeonRefillCurrentDmaRegion( radeonContextPtr rmesa );
+
+extern void radeonAllocDmaRegion( radeonContextPtr rmesa,
+				  struct radeon_dma_region *region,
+				  int bytes, 
+				  int alignment );
+
+extern void radeonAllocDmaRegionVerts( radeonContextPtr rmesa,
+				       struct radeon_dma_region *region,
+				       int numverts,
+				       int vertsize, 
+				       int alignment );
+
+extern void radeonReleaseDmaRegion( radeonContextPtr rmesa,
+				    struct radeon_dma_region *region,
+				    const char *caller );
+
+extern void radeonCopyBuffer( const __DRIdrawablePrivate *drawable );
+extern void radeonPageFlip( const __DRIdrawablePrivate *drawable );
+extern void radeonFlush( GLcontext *ctx );
+extern void radeonFinish( GLcontext *ctx );
+extern void radeonWaitForIdleLocked( radeonContextPtr rmesa );
+extern void radeonWaitForVBlank( radeonContextPtr rmesa );
+extern void radeonInitIoctlFuncs( GLcontext *ctx );
+extern void radeonGetAllParams( radeonContextPtr rmesa );
+
+/* radeon_compat.c:
+ */
+extern void radeonCompatEmitPrimitive( radeonContextPtr rmesa,
+				       GLuint vertex_format,
+				       GLuint hw_primitive,
+				       GLuint nrverts );
+
+
+/* ================================================================
+ * Helper macros:
+ */
+
+/* Close off the last primitive, if it exists.
+ */
+#define RADEON_NEWPRIM( rmesa )			\
+do {						\
+   if ( rmesa->dma.flush )			\
+      rmesa->dma.flush( rmesa );	\
+} while (0)
+
+/* Can accomodate several state changes and primitive changes without
+ * actually firing the buffer.
+ */
+#define RADEON_STATECHANGE( rmesa, ATOM )			\
+do {								\
+   RADEON_NEWPRIM( rmesa );					\
+   move_to_head( &(rmesa->hw.dirty), &(rmesa->hw.ATOM));	\
+} while (0)
+
+#define RADEON_DB_STATE( ATOM )			        \
+   memcpy( rmesa->hw.ATOM.lastcmd, rmesa->hw.ATOM.cmd,	\
+	   rmesa->hw.ATOM.cmd_size * 4)
+
+static __inline int RADEON_DB_STATECHANGE( 
+   radeonContextPtr rmesa,
+   struct radeon_state_atom *atom )
+{
+   if (memcmp(atom->cmd, atom->lastcmd, atom->cmd_size*4)) {
+      int *tmp;
+      RADEON_NEWPRIM( rmesa );
+      move_to_head( &(rmesa->hw.dirty), atom );
+      tmp = atom->cmd; 
+      atom->cmd = atom->lastcmd;
+      atom->lastcmd = tmp;
+      return 1;
+   }
+   else
+      return 0;
+}
+
+
+/* Fire the buffered vertices no matter what.
+ */
+#define RADEON_FIREVERTICES( rmesa )			\
+do {							\
+   if ( rmesa->store.cmd_used || rmesa->dma.flush ) {	\
+      radeonFlush( rmesa->glCtx );			\
+   }							\
+} while (0)
+
+/* Alloc space in the command buffer
+ */
+static __inline char *radeonAllocCmdBuf( radeonContextPtr rmesa,
+					 int bytes, const char *where )
+{
+   if (rmesa->store.cmd_used + bytes > RADEON_CMD_BUF_SZ)
+      radeonFlushCmdBuf( rmesa, __FUNCTION__ );
+   
+   assert(rmesa->dri.drmMinor >= 3);
+
+   {
+      char *head = rmesa->store.cmd_buf + rmesa->store.cmd_used;
+      rmesa->store.cmd_used += bytes;
+      return head;
+   }
+}
+
+
+
+
+#endif
+#endif /* __RADEON_IOCTL_H__ */
diff --git a/src/mesa/drivers/dri/radeon/radeon_lighting.c b/src/mesa/drivers/dri/radeon/radeon_lighting.c
new file mode 100644
index 00000000000..b00c9cb6ded
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_lighting.c
@@ -0,0 +1,682 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_state.c,v 1.5 2002/09/16 18:05:20 eich Exp $ */
+/*
+ * Copyright 2000, 2001 VA Linux Systems Inc., Fremont, California.
+ *
+ * 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
+ * VA LINUX SYSTEMS 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:
+ *    Gareth Hughes <[email protected]>
+ *    Keith Whitwell <[email protected]>
+ */
+
+#include "glheader.h"
+#include "imports.h"
+#include "api_arrayelt.h"
+/* #include "mmath.h" */
+#include "enums.h"
+#include "colormac.h"
+
+
+#include "radeon_context.h"
+#include "radeon_ioctl.h"
+#include "radeon_state.h"
+#include "radeon_tcl.h"
+#include "radeon_tex.h"
+#include "radeon_vtxfmt.h"
+
+
+
+/* =============================================================
+ * Materials
+ */
+
+
+/* Update on colormaterial, material emmissive/ambient, 
+ * lightmodel.globalambient
+ */
+void update_global_ambient( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   float *fcmd = (float *)RADEON_DB_STATE( glt );
+
+   /* Need to do more if both emmissive & ambient are PREMULT:
+    */
+   if ((rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &
+       ((3 << RADEON_EMISSIVE_SOURCE_SHIFT) |
+	(3 << RADEON_AMBIENT_SOURCE_SHIFT))) == 0) 
+   {
+      COPY_3V( &fcmd[GLT_RED], 
+	       ctx->Light.Material[0].Emission);
+      ACC_SCALE_3V( &fcmd[GLT_RED],
+		   ctx->Light.Model.Ambient,
+		   ctx->Light.Material[0].Ambient);
+   } 
+   else
+   {
+      COPY_3V( &fcmd[GLT_RED], ctx->Light.Model.Ambient );
+   }
+   
+   RADEON_DB_STATECHANGE(rmesa, &rmesa->hw.glt);
+}
+
+/* Update on change to 
+ *    - light[p].colors
+ *    - light[p].enabled
+ *    - material,
+ *    - colormaterial enabled
+ *    - colormaterial bitmask
+ */
+void update_light_colors( GLcontext *ctx, GLuint p )
+{
+   struct gl_light *l = &ctx->Light.Light[p];
+
+/*     fprintf(stderr, "%s\n", __FUNCTION__); */
+
+   if (l->Enabled) {
+      radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+      float *fcmd = (float *)RADEON_DB_STATE( lit[p] );
+      GLuint bitmask = ctx->Light.ColorMaterialBitmask;
+      struct gl_material *mat = &ctx->Light.Material[0];
+
+      COPY_4V( &fcmd[LIT_AMBIENT_RED], l->Ambient );	 
+      COPY_4V( &fcmd[LIT_DIFFUSE_RED], l->Diffuse );
+      COPY_4V( &fcmd[LIT_SPECULAR_RED], l->Specular );
+      
+      if (!ctx->Light.ColorMaterialEnabled)
+	 bitmask = 0;
+
+      if ((bitmask & FRONT_AMBIENT_BIT) == 0) 
+	 SELF_SCALE_3V( &fcmd[LIT_AMBIENT_RED], mat->Ambient );
+
+      if ((bitmask & FRONT_DIFFUSE_BIT) == 0) 
+	 SELF_SCALE_3V( &fcmd[LIT_DIFFUSE_RED], mat->Diffuse );
+      
+      if ((bitmask & FRONT_SPECULAR_BIT) == 0) 
+	 SELF_SCALE_3V( &fcmd[LIT_SPECULAR_RED], mat->Specular );
+
+      RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.lit[p] );
+   }
+}
+
+/* Also fallback for asym colormaterial mode in twoside lighting...
+ */
+void check_twoside_fallback( GLcontext *ctx )
+{
+   GLboolean fallback = GL_FALSE;
+
+   if (ctx->Light.Enabled && ctx->Light.Model.TwoSide) {
+      if (memcmp( &ctx->Light.Material[0],
+		  &ctx->Light.Material[1],
+		  sizeof(struct gl_material)) != 0)
+	 fallback = GL_TRUE;  
+      else if (ctx->Light.ColorMaterialEnabled &&
+	       (ctx->Light.ColorMaterialBitmask & BACK_MATERIAL_BITS) != 
+	       ((ctx->Light.ColorMaterialBitmask & FRONT_MATERIAL_BITS)<<1))
+	 fallback = GL_TRUE;
+   }
+
+   TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_LIGHT_TWOSIDE, fallback );
+}
+
+void radeonColorMaterial( GLcontext *ctx, GLenum face, GLenum mode )
+{
+   if (ctx->Light.ColorMaterialEnabled) {
+      radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+      GLuint light_model_ctl = rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL];
+      GLuint mask = ctx->Light.ColorMaterialBitmask;
+
+      /* Default to PREMULT:
+       */
+      light_model_ctl &= ~((3 << RADEON_EMISSIVE_SOURCE_SHIFT) |
+			   (3 << RADEON_AMBIENT_SOURCE_SHIFT) |
+			   (3 << RADEON_DIFFUSE_SOURCE_SHIFT) |
+			   (3 << RADEON_SPECULAR_SOURCE_SHIFT)); 
+   
+      if (mask & FRONT_EMISSION_BIT) {
+	 light_model_ctl |= (RADEON_LM_SOURCE_VERTEX_DIFFUSE <<
+			     RADEON_EMISSIVE_SOURCE_SHIFT);
+      }
+
+      if (mask & FRONT_AMBIENT_BIT) {
+	 light_model_ctl |= (RADEON_LM_SOURCE_VERTEX_DIFFUSE <<
+			     RADEON_AMBIENT_SOURCE_SHIFT);
+      }
+	 
+      if (mask & FRONT_DIFFUSE_BIT) {
+	 light_model_ctl |= (RADEON_LM_SOURCE_VERTEX_DIFFUSE <<
+			     RADEON_DIFFUSE_SOURCE_SHIFT);
+      }
+   
+      if (mask & FRONT_SPECULAR_BIT) {
+	 light_model_ctl |= (RADEON_LM_SOURCE_VERTEX_DIFFUSE <<
+			     RADEON_SPECULAR_SOURCE_SHIFT);
+      }
+   
+      if (light_model_ctl != rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL]) {
+	 GLuint p;
+
+	 RADEON_STATECHANGE( rmesa, tcl );
+	 rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] = light_model_ctl;      
+
+	 for (p = 0 ; p < MAX_LIGHTS; p++) 
+	    update_light_colors( ctx, p );
+	 update_global_ambient( ctx );
+      }
+   }
+   
+   check_twoside_fallback( ctx );
+}
+
+void radeonUpdateMaterial( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *fcmd = (GLfloat *)RADEON_DB_STATE( mtl );
+   GLuint p;
+   GLuint mask = ~0;
+   
+   if (ctx->Light.ColorMaterialEnabled)
+      mask &= ~ctx->Light.ColorMaterialBitmask;
+
+   if (RADEON_DEBUG & DEBUG_STATE)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+      
+   if (mask & FRONT_EMISSION_BIT) {
+      fcmd[MTL_EMMISSIVE_RED]   = ctx->Light.Material[0].Emission[0];
+      fcmd[MTL_EMMISSIVE_GREEN] = ctx->Light.Material[0].Emission[1];
+      fcmd[MTL_EMMISSIVE_BLUE]  = ctx->Light.Material[0].Emission[2];
+      fcmd[MTL_EMMISSIVE_ALPHA] = ctx->Light.Material[0].Emission[3];
+   }
+   if (mask & FRONT_AMBIENT_BIT) {
+      fcmd[MTL_AMBIENT_RED]     = ctx->Light.Material[0].Ambient[0];
+      fcmd[MTL_AMBIENT_GREEN]   = ctx->Light.Material[0].Ambient[1];
+      fcmd[MTL_AMBIENT_BLUE]    = ctx->Light.Material[0].Ambient[2];
+      fcmd[MTL_AMBIENT_ALPHA]   = ctx->Light.Material[0].Ambient[3];
+   }
+   if (mask & FRONT_DIFFUSE_BIT) {
+      fcmd[MTL_DIFFUSE_RED]     = ctx->Light.Material[0].Diffuse[0];
+      fcmd[MTL_DIFFUSE_GREEN]   = ctx->Light.Material[0].Diffuse[1];
+      fcmd[MTL_DIFFUSE_BLUE]    = ctx->Light.Material[0].Diffuse[2];
+      fcmd[MTL_DIFFUSE_ALPHA]   = ctx->Light.Material[0].Diffuse[3];
+   }
+   if (mask & FRONT_SPECULAR_BIT) {
+      fcmd[MTL_SPECULAR_RED]    = ctx->Light.Material[0].Specular[0];
+      fcmd[MTL_SPECULAR_GREEN]  = ctx->Light.Material[0].Specular[1];
+      fcmd[MTL_SPECULAR_BLUE]   = ctx->Light.Material[0].Specular[2];
+      fcmd[MTL_SPECULAR_ALPHA]  = ctx->Light.Material[0].Specular[3];
+   }
+   if (mask & FRONT_SHININESS_BIT) {
+      fcmd[MTL_SHININESS]       = ctx->Light.Material[0].Shininess;
+   }
+
+   if (RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.mtl )) {
+      for (p = 0 ; p < MAX_LIGHTS; p++) 
+	 update_light_colors( ctx, p );
+
+      check_twoside_fallback( ctx );
+      update_global_ambient( ctx );
+   }
+   else if (RADEON_DEBUG & (DEBUG_PRIMS|DEBUG_STATE))
+      fprintf(stderr, "%s: Elided noop material call\n", __FUNCTION__);
+}
+
+/* _NEW_LIGHT
+ * _NEW_MODELVIEW
+ * _MESA_NEW_NEED_EYE_COORDS
+ *
+ * Uses derived state from mesa:
+ *       _VP_inf_norm
+ *       _h_inf_norm
+ *       _Position
+ *       _NormDirection
+ *       _ModelViewInvScale
+ *       _NeedEyeCoords
+ *       _EyeZDir
+ *
+ * which are calculated in light.c and are correct for the current
+ * lighting space (model or eye), hence dependencies on _NEW_MODELVIEW
+ * and _MESA_NEW_NEED_EYE_COORDS.  
+ */
+void radeonUpdateLighting( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   /* Have to check these, or have an automatic shortcircuit mechanism
+    * to remove noop statechanges. (Or just do a better job on the
+    * front end).
+    */
+   {
+      GLuint tmp = rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL];
+
+      if (ctx->_NeedEyeCoords)
+	 tmp &= ~RADEON_LIGHT_IN_MODELSPACE;
+      else
+	 tmp |= RADEON_LIGHT_IN_MODELSPACE;
+      
+
+      /* Leave this test disabled: (unexplained q3 lockup) (even with
+         new packets)
+      */
+      if (tmp != rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL]) 
+      {
+	 RADEON_STATECHANGE( rmesa, tcl );
+	 rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] = tmp;
+      }
+   }
+
+   {
+      GLfloat *fcmd = (GLfloat *)RADEON_DB_STATE( eye );
+      fcmd[EYE_X] = ctx->_EyeZDir[0];
+      fcmd[EYE_Y] = ctx->_EyeZDir[1];
+      fcmd[EYE_Z] = - ctx->_EyeZDir[2];
+      fcmd[EYE_RESCALE_FACTOR] = ctx->_ModelViewInvScale;
+      RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.eye );
+   }
+
+
+/*     RADEON_STATECHANGE( rmesa, glt ); */
+
+   if (ctx->Light.Enabled) {
+      GLint p;
+      for (p = 0 ; p < MAX_LIGHTS; p++) {
+	 if (ctx->Light.Light[p].Enabled) {
+	    struct gl_light *l = &ctx->Light.Light[p];
+	    GLfloat *fcmd = (GLfloat *)RADEON_DB_STATE( lit[p] );
+	    
+	    if (l->EyePosition[3] == 0.0) {
+	       COPY_3FV( &fcmd[LIT_POSITION_X], l->_VP_inf_norm ); 
+	       COPY_3FV( &fcmd[LIT_DIRECTION_X], l->_h_inf_norm ); 
+	       fcmd[LIT_POSITION_W] = 0;
+	       fcmd[LIT_DIRECTION_W] = 0;
+	    } else {
+	       COPY_4V( &fcmd[LIT_POSITION_X], l->_Position );
+	       fcmd[LIT_DIRECTION_X] = -l->_NormDirection[0];
+	       fcmd[LIT_DIRECTION_Y] = -l->_NormDirection[1];
+	       fcmd[LIT_DIRECTION_Z] = -l->_NormDirection[2];
+	       fcmd[LIT_DIRECTION_W] = 0;
+	    }
+
+	    RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.lit[p] );
+	 }
+      }
+   }
+}
+
+
+void radeonLightfv( GLcontext *ctx, GLenum light,
+		    GLenum pname, const GLfloat *params )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLint p = light - GL_LIGHT0;
+   struct gl_light *l = &ctx->Light.Light[p];
+   GLfloat *fcmd = (GLfloat *)rmesa->hw.lit[p].cmd;
+   
+
+   switch (pname) {
+   case GL_AMBIENT:		
+   case GL_DIFFUSE:
+   case GL_SPECULAR:
+      update_light_colors( ctx, p );
+      break;
+
+   case GL_SPOT_DIRECTION: 
+      /* picked up in update_light */	
+      break;
+
+   case GL_POSITION: {
+      /* positions picked up in update_light, but can do flag here */	
+      GLuint flag = (p&1)? RADEON_LIGHT_1_IS_LOCAL : RADEON_LIGHT_0_IS_LOCAL;
+      GLuint idx = TCL_PER_LIGHT_CTL_0 + p/2;
+
+      RADEON_STATECHANGE(rmesa, tcl);
+      if (l->EyePosition[3] != 0.0F)
+	 rmesa->hw.tcl.cmd[idx] |= flag;
+      else
+	 rmesa->hw.tcl.cmd[idx] &= ~flag;
+      break;
+   }
+
+   case GL_SPOT_EXPONENT:
+      RADEON_STATECHANGE(rmesa, lit[p]);
+      fcmd[LIT_SPOT_EXPONENT] = params[0];
+      break;
+
+   case GL_SPOT_CUTOFF: {
+      GLuint flag = (p&1) ? RADEON_LIGHT_1_IS_SPOT : RADEON_LIGHT_0_IS_SPOT;
+      GLuint idx = TCL_PER_LIGHT_CTL_0 + p/2;
+
+      RADEON_STATECHANGE(rmesa, lit[p]);
+      fcmd[LIT_SPOT_CUTOFF] = l->_CosCutoff;
+
+      RADEON_STATECHANGE(rmesa, tcl);
+      if (l->SpotCutoff != 180.0F)
+	 rmesa->hw.tcl.cmd[idx] |= flag;
+      else
+	 rmesa->hw.tcl.cmd[idx] &= ~flag;
+      break;
+   }
+
+   case GL_CONSTANT_ATTENUATION:
+      RADEON_STATECHANGE(rmesa, lit[p]);
+      fcmd[LIT_ATTEN_CONST] = params[0];
+      break;
+   case GL_LINEAR_ATTENUATION:
+      RADEON_STATECHANGE(rmesa, lit[p]);
+      fcmd[LIT_ATTEN_LINEAR] = params[0];
+      break;
+   case GL_QUADRATIC_ATTENUATION:
+      RADEON_STATECHANGE(rmesa, lit[p]);
+      fcmd[LIT_ATTEN_QUADRATIC] = params[0];
+      break;
+   default:
+      return;
+   }
+
+}
+
+		  
+
+
+void radeonLightModelfv( GLcontext *ctx, GLenum pname,
+			 const GLfloat *param )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   switch (pname) {
+      case GL_LIGHT_MODEL_AMBIENT: 
+	 update_global_ambient( ctx );
+	 break;
+
+      case GL_LIGHT_MODEL_LOCAL_VIEWER:
+	 RADEON_STATECHANGE( rmesa, tcl );
+	 if (ctx->Light.Model.LocalViewer)
+	    rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] |= RADEON_LOCAL_VIEWER;
+	 else
+	    rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &= ~RADEON_LOCAL_VIEWER;
+         break;
+
+      case GL_LIGHT_MODEL_TWO_SIDE:
+	 RADEON_STATECHANGE( rmesa, tcl );
+	 if (ctx->Light.Model.TwoSide)
+	    rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] |= RADEON_LIGHT_TWOSIDE;
+	 else
+	    rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] &= ~RADEON_LIGHT_TWOSIDE;
+
+	 check_twoside_fallback( ctx );
+
+#if _HAVE_SWTNL
+	 if (rmesa->TclFallback) {
+	    radeonChooseRenderState( ctx );
+	    radeonChooseVertexState( ctx );
+	 }
+#endif
+         break;
+
+      case GL_LIGHT_MODEL_COLOR_CONTROL:
+	 radeonUpdateSpecular(ctx);
+
+	 RADEON_STATECHANGE( rmesa, tcl );
+	 if (ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR) 
+	    rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &= 
+	       ~RADEON_DIFFUSE_SPECULAR_COMBINE;
+	 else
+	    rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] |= 
+	       RADEON_DIFFUSE_SPECULAR_COMBINE;
+         break;
+
+      default:
+         break;
+   }
+}
+
+
+/* =============================================================
+ * Fog
+ */
+
+
+static void radeonFogfv( GLcontext *ctx, GLenum pname, const GLfloat *param )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   union { int i; float f; } c, d;
+   GLchan col[4];
+
+   c.i = rmesa->hw.fog.cmd[FOG_C];
+   d.i = rmesa->hw.fog.cmd[FOG_D];
+
+   switch (pname) {
+   case GL_FOG_MODE:
+      if (!ctx->Fog.Enabled)
+	 return;
+      RADEON_STATECHANGE(rmesa, tcl);
+      rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] &= ~RADEON_TCL_FOG_MASK;
+      switch (ctx->Fog.Mode) {
+      case GL_LINEAR:
+	 rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] |= RADEON_TCL_FOG_LINEAR;
+	 if (ctx->Fog.Start == ctx->Fog.End) {
+	    c.f = 1.0F;
+	    d.f = 1.0F;
+	 }
+	 else {
+	    c.f = ctx->Fog.End/(ctx->Fog.End-ctx->Fog.Start);
+	    d.f = 1.0/(ctx->Fog.End-ctx->Fog.Start);
+	 }
+	 break;
+      case GL_EXP:
+	 rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] |= RADEON_TCL_FOG_EXP;
+	 c.f = 0.0;
+	 d.f = ctx->Fog.Density;
+	 break;
+      case GL_EXP2:
+	 rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] |= RADEON_TCL_FOG_EXP2;
+	 c.f = 0.0;
+	 d.f = -(ctx->Fog.Density * ctx->Fog.Density);
+	 break;
+      default:
+	 return;
+      }
+      break;
+   case GL_FOG_DENSITY:
+      switch (ctx->Fog.Mode) {
+      case GL_EXP:
+	 c.f = 0.0;
+	 d.f = ctx->Fog.Density;
+	 break;
+      case GL_EXP2:
+	 c.f = 0.0;
+	 d.f = -(ctx->Fog.Density * ctx->Fog.Density);
+	 break;
+      default:
+	 break;
+      }
+      break;
+   case GL_FOG_START:
+   case GL_FOG_END:
+      if (ctx->Fog.Mode == GL_LINEAR) {
+	 if (ctx->Fog.Start == ctx->Fog.End) {
+	    c.f = 1.0F;
+	    d.f = 1.0F;
+	 } else {
+	    c.f = ctx->Fog.End/(ctx->Fog.End-ctx->Fog.Start);
+	    d.f = 1.0/(ctx->Fog.End-ctx->Fog.Start);
+	 }
+      }
+      break;
+   case GL_FOG_COLOR: 
+      RADEON_STATECHANGE( rmesa, ctx );
+      UNCLAMPED_FLOAT_TO_RGB_CHAN( col, ctx->Fog.Color );
+      rmesa->hw.ctx.cmd[CTX_PP_FOG_COLOR] =
+	 radeonPackColor( 4, col[0], col[1], col[2], 0 );
+      break;
+   case GL_FOG_COORDINATE_SOURCE_EXT: 
+      /* What to do?
+       */
+      break;
+   default:
+      return;
+   }
+
+   if (c.i != rmesa->hw.fog.cmd[FOG_C] || d.i != rmesa->hw.fog.cmd[FOG_D]) {
+      RADEON_STATECHANGE( rmesa, fog );
+      rmesa->hw.fog.cmd[FOG_C] = c.i;
+      rmesa->hw.fog.cmd[FOG_D] = d.i;
+   }
+}
+
+/* Examine lighting and texture state to determine if separate specular
+ * should be enabled.
+ */
+void radeonUpdateSpecular( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint p = rmesa->hw.ctx.cmd[CTX_PP_CNTL];
+
+   if ( ctx->_TriangleCaps & DD_SEPARATE_SPECULAR ) {
+      p |=  RADEON_SPECULAR_ENABLE;
+   } else {
+      p &= ~RADEON_SPECULAR_ENABLE;
+   }
+
+   if ( rmesa->hw.ctx.cmd[CTX_PP_CNTL] != p ) {
+      RADEON_STATECHANGE( rmesa, ctx );
+      rmesa->hw.ctx.cmd[CTX_PP_CNTL] = p;
+   }
+
+   /* Bizzare: have to leave lighting enabled to get fog.
+    */
+   RADEON_STATECHANGE( rmesa, tcl );
+   if ((ctx->Light.Enabled &&
+	ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR)) {
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] |= RADEON_TCL_COMPUTE_SPECULAR;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] |= RADEON_TCL_COMPUTE_DIFFUSE;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_SPEC;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_DIFFUSE;
+      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] |= RADEON_LIGHTING_ENABLE;
+   }
+   else if (ctx->Fog.Enabled) {
+      if (ctx->Light.Enabled) {
+	 rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] |= RADEON_TCL_COMPUTE_SPECULAR;
+	 rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] |= RADEON_TCL_COMPUTE_DIFFUSE;
+	 rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_SPEC;
+	 rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_DIFFUSE;
+	 rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] |= RADEON_LIGHTING_ENABLE;
+      } else {
+	 rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] |= RADEON_TCL_COMPUTE_SPECULAR;
+	 rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] &= ~RADEON_TCL_COMPUTE_DIFFUSE;
+	 rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_SPEC;
+	 rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_DIFFUSE;
+	 rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] |= RADEON_LIGHTING_ENABLE;
+      }
+   }
+   else if (ctx->Light.Enabled) {
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] &= ~RADEON_TCL_COMPUTE_SPECULAR;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] |= RADEON_TCL_COMPUTE_DIFFUSE;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] &= ~RADEON_TCL_VTX_PK_SPEC;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_DIFFUSE;
+      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] |= RADEON_LIGHTING_ENABLE;
+   } else if (ctx->Fog.ColorSumEnabled ) {
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] &= ~RADEON_TCL_COMPUTE_SPECULAR;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] &= ~RADEON_TCL_COMPUTE_DIFFUSE;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_SPEC;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_DIFFUSE;
+      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &= ~RADEON_LIGHTING_ENABLE;
+   } else {
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] &= ~RADEON_TCL_COMPUTE_SPECULAR;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] &= ~RADEON_TCL_COMPUTE_DIFFUSE;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] &= ~RADEON_TCL_VTX_PK_SPEC;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_DIFFUSE;
+      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &= ~RADEON_LIGHTING_ENABLE;
+   }
+
+#if _HAVE_SWTNL
+   /* Update vertex/render formats
+    */
+   if (rmesa->TclFallback) { 
+      radeonChooseRenderState( ctx );
+      radeonChooseVertexState( ctx );
+   }
+#endif
+}
+
+
+
+static void radeonLightingSpaceChange( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLboolean tmp;
+   RADEON_STATECHANGE( rmesa, tcl );
+
+   if (RADEON_DEBUG & DEBUG_STATE)
+      fprintf(stderr, "%s %d\n", __FUNCTION__, ctx->_NeedEyeCoords);
+
+   if (ctx->_NeedEyeCoords)
+      tmp = ctx->Transform.RescaleNormals;
+   else
+      tmp = !ctx->Transform.RescaleNormals;
+
+   if ( tmp ) {
+      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] |=  RADEON_RESCALE_NORMALS;
+   } else {
+      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &= ~RADEON_RESCALE_NORMALS;
+   }
+}
+
+void radeonInitLightStateFuncs( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   int i;
+
+   ctx->Driver.LightModelfv		= radeonLightModelfv; 
+   ctx->Driver.Lightfv			= radeonLightfv; 
+   ctx->Driver.Fogfv			= radeonFogfv;
+   ctx->Driver.LightingSpaceChange      = radeonLightingSpaceChange;
+
+   for (i = 0 ; i < 8; i++) {
+      struct gl_light *l = &ctx->Light.Light[i];
+      GLenum p = GL_LIGHT0 + i;
+      *(float *)&(rmesa->hw.lit[i].cmd[LIT_RANGE_CUTOFF]) = FLT_MAX;
+
+      ctx->Driver.Lightfv( ctx, p, GL_AMBIENT, l->Ambient );
+      ctx->Driver.Lightfv( ctx, p, GL_DIFFUSE, l->Diffuse );
+      ctx->Driver.Lightfv( ctx, p, GL_SPECULAR, l->Specular );
+      ctx->Driver.Lightfv( ctx, p, GL_POSITION, 0 );
+      ctx->Driver.Lightfv( ctx, p, GL_SPOT_DIRECTION, 0 );
+      ctx->Driver.Lightfv( ctx, p, GL_SPOT_EXPONENT, &l->SpotExponent );
+      ctx->Driver.Lightfv( ctx, p, GL_SPOT_CUTOFF, &l->SpotCutoff );
+      ctx->Driver.Lightfv( ctx, p, GL_CONSTANT_ATTENUATION,
+			   &l->ConstantAttenuation );
+      ctx->Driver.Lightfv( ctx, p, GL_LINEAR_ATTENUATION, 
+			   &l->LinearAttenuation );
+      ctx->Driver.Lightfv( ctx, p, GL_QUADRATIC_ATTENUATION, 
+		     &l->QuadraticAttenuation );
+   }
+
+   ctx->Driver.LightModelfv( ctx, GL_LIGHT_MODEL_AMBIENT, 
+			     ctx->Light.Model.Ambient );
+
+   ctx->Driver.Fogfv( ctx, GL_FOG_MODE, 0 );
+   ctx->Driver.Fogfv( ctx, GL_FOG_DENSITY, &ctx->Fog.Density );
+   ctx->Driver.Fogfv( ctx, GL_FOG_START, &ctx->Fog.Start );
+   ctx->Driver.Fogfv( ctx, GL_FOG_END, &ctx->Fog.End );
+   ctx->Driver.Fogfv( ctx, GL_FOG_COLOR, ctx->Fog.Color );
+   ctx->Driver.Fogfv( ctx, GL_FOG_COORDINATE_SOURCE_EXT, 0 );
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_lock.c b/src/mesa/drivers/dri/radeon/radeon_lock.c
new file mode 100644
index 00000000000..96a4f9d1123
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_lock.c
@@ -0,0 +1,128 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_lock.c,v 1.5 2002/10/30 12:51:55 alanh Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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 (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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ */
+
+#include "glheader.h"
+#include "radeon_context.h"
+#include "radeon_lock.h"
+#include "radeon_tex.h"
+#include "radeon_state.h"
+#include "radeon_ioctl.h"
+
+#if DEBUG_LOCKING
+char *prevLockFile = NULL;
+int prevLockLine = 0;
+#endif
+
+/* Turn on/off page flipping according to the flags in the sarea:
+ */
+static void
+radeonUpdatePageFlipping( radeonContextPtr rmesa )
+{
+   int use_back;
+
+   if (rmesa->dri.drmMinor < 3)
+      return;
+
+   rmesa->doPageFlip = rmesa->sarea->pfAllowPageFlip;
+
+   use_back = (rmesa->glCtx->Color._DrawDestMask == BACK_LEFT_BIT);
+   use_back ^= (rmesa->sarea->pfCurrentPage == 1);
+
+   if ( RADEON_DEBUG & DEBUG_VERBOSE )
+      fprintf(stderr, "%s allow %d current %d\n", __FUNCTION__, 
+	      rmesa->doPageFlip,
+	      rmesa->sarea->pfCurrentPage );
+
+   if ( use_back ) {
+	 rmesa->state.color.drawOffset = rmesa->radeonScreen->backOffset;
+	 rmesa->state.color.drawPitch  = rmesa->radeonScreen->backPitch;
+   } else {
+	 rmesa->state.color.drawOffset = rmesa->radeonScreen->frontOffset;
+	 rmesa->state.color.drawPitch  = rmesa->radeonScreen->frontPitch;
+   }
+
+   RADEON_STATECHANGE( rmesa, ctx );
+   rmesa->hw.ctx.cmd[CTX_RB3D_COLOROFFSET] = rmesa->state.color.drawOffset;
+   rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH]  = rmesa->state.color.drawPitch;
+}
+
+
+
+/* Update the hardware state.  This is called if another context has
+ * grabbed the hardware lock, which includes the X server.  This
+ * function also updates the driver's window state after the X server
+ * moves, resizes or restacks a window -- the change will be reflected
+ * in the drawable position and clip rects.  Since the X server grabs
+ * the hardware lock when it changes the window state, this routine will
+ * automatically be called after such a change.
+ */
+void radeonGetLock( radeonContextPtr rmesa, GLuint flags )
+{
+   __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
+   __DRIscreenPrivate *sPriv = rmesa->dri.screen;
+   RADEONSAREAPrivPtr sarea = rmesa->sarea;
+
+   drmGetLock( rmesa->dri.fd, rmesa->dri.hwContext, flags );
+
+   /* The window might have moved, so we might need to get new clip
+    * rects.
+    *
+    * NOTE: This releases and regrabs the hw lock to allow the X server
+    * to respond to the DRI protocol request for new drawable info.
+    * Since the hardware state depends on having the latest drawable
+    * clip rects, all state checking must be done _after_ this call.
+    */
+   DRI_VALIDATE_DRAWABLE_INFO( sPriv, dPriv );
+
+   if ( rmesa->lastStamp != dPriv->lastStamp ) {
+      radeonUpdatePageFlipping( rmesa );
+      if (rmesa->glCtx->Color._DrawDestMask == BACK_LEFT_BIT)
+         radeonSetCliprects( rmesa, GL_BACK_LEFT );
+      else
+         radeonSetCliprects( rmesa, GL_FRONT_LEFT );
+      radeonUpdateViewportOffset( rmesa->glCtx );
+      rmesa->lastStamp = dPriv->lastStamp;
+   }
+
+   if ( sarea->ctxOwner != rmesa->dri.hwContext ) {
+      int i;
+      sarea->ctxOwner = rmesa->dri.hwContext;
+
+      for ( i = 0 ; i < rmesa->nr_heaps ; i++ ) {
+	 DRI_AGE_TEXTURES( rmesa->texture_heaps[ i ] );
+      }
+   }
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_lock.h b/src/mesa/drivers/dri/radeon/radeon_lock.h
new file mode 100644
index 00000000000..783db7e92a8
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_lock.h
@@ -0,0 +1,113 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_lock.h,v 1.3 2002/10/30 12:51:55 alanh Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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 (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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ */
+
+#ifndef __RADEON_LOCK_H__
+#define __RADEON_LOCK_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+extern void radeonGetLock( radeonContextPtr rmesa, GLuint flags );
+
+/* Turn DEBUG_LOCKING on to find locking conflicts.
+ */
+#define DEBUG_LOCKING	0
+
+#if DEBUG_LOCKING
+extern char *prevLockFile;
+extern int prevLockLine;
+
+#define DEBUG_LOCK()							\
+   do {									\
+      prevLockFile = (__FILE__);					\
+      prevLockLine = (__LINE__);					\
+   } while (0)
+
+#define DEBUG_RESET()							\
+   do {									\
+      prevLockFile = 0;							\
+      prevLockLine = 0;							\
+   } while (0)
+
+#define DEBUG_CHECK_LOCK()						\
+   do {									\
+      if ( prevLockFile ) {						\
+	 fprintf( stderr,						\
+		  "LOCK SET!\n\tPrevious %s:%d\n\tCurrent: %s:%d\n",	\
+		  prevLockFile, prevLockLine, __FILE__, __LINE__ );	\
+	 exit( 1 );							\
+      }									\
+   } while (0)
+
+#else
+
+#define DEBUG_LOCK()
+#define DEBUG_RESET()
+#define DEBUG_CHECK_LOCK()
+
+#endif
+
+/*
+ * !!! We may want to separate locks from locks with validation.  This
+ * could be used to improve performance for those things commands that
+ * do not do any drawing !!!
+ */
+
+
+/* Lock the hardware and validate our state.
+ */
+#define LOCK_HARDWARE( rmesa )					\
+   do {								\
+      char __ret = 0;						\
+      DEBUG_CHECK_LOCK();					\
+      DRM_CAS( rmesa->dri.hwLock, rmesa->dri.hwContext,		\
+	       (DRM_LOCK_HELD | rmesa->dri.hwContext), __ret );	\
+      if ( __ret )						\
+	 radeonGetLock( rmesa, 0 );				\
+      DEBUG_LOCK();						\
+   } while (0)
+
+/* Unlock the hardware.
+ */
+#define UNLOCK_HARDWARE( rmesa )					\
+   do {									\
+      DRM_UNLOCK( rmesa->dri.fd,					\
+		  rmesa->dri.hwLock,					\
+		  rmesa->dri.hwContext );				\
+      DEBUG_RESET();							\
+   } while (0)
+
+#endif
+#endif /* __RADEON_LOCK_H__ */
diff --git a/src/mesa/drivers/dri/radeon/radeon_maos.c b/src/mesa/drivers/dri/radeon/radeon_maos.c
new file mode 100644
index 00000000000..c62edd715c7
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_maos.c
@@ -0,0 +1,12 @@
+
+
+/* If using new packets, can choose either verts or arrays.
+ * Otherwise, must use verts.
+ */
+#include "radeon_context.h"
+#define RADEON_MAOS_VERTS 1
+#if (RADEON_MAOS_VERTS) || (RADEON_OLD_PACKETS)
+#include "radeon_maos_verts.c"
+#else
+#include "radeon_maos_arrays.c"
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_maos.h b/src/mesa/drivers/dri/radeon/radeon_maos.h
new file mode 100644
index 00000000000..7e2bd643d34
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_maos.h
@@ -0,0 +1,47 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     Tungsten Grahpics Inc., Austin, 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, 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 <[email protected]>
+ */
+
+#ifndef __RADEON_MAOS_H__
+#define __RADEON_MAOS_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+#include "radeon_context.h"
+
+extern void radeonEmitArrays( GLcontext *ctx, GLuint inputs );
+extern void radeonReleaseArrays( GLcontext *ctx, GLuint newinputs );
+
+#endif
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_maos_arrays.c b/src/mesa/drivers/dri/radeon/radeon_maos_arrays.c
new file mode 100644
index 00000000000..cec05a89d7e
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_maos_arrays.c
@@ -0,0 +1,591 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     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, 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 <[email protected]>
+ */
+
+#include "glheader.h"
+#include "imports.h"
+#include "mtypes.h"
+#include "mmath.h"
+#include "macros.h"
+
+#include "swrast_setup/swrast_setup.h"
+#include "math/m_translate.h"
+#include "tnl/tnl.h"
+#include "tnl/t_context.h"
+#include "tnl/t_imm_debug.h"
+
+#include "radeon_context.h"
+#include "radeon_ioctl.h"
+#include "radeon_state.h"
+#include "radeon_swtcl.h"
+#include "radeon_maos.h"
+
+/* Usage:
+ *   - from radeon_tcl_render
+ *   - call radeonEmitArrays to ensure uptodate arrays in dma
+ *   - emit primitives (new type?) which reference the data
+ *       -- need to use elts for lineloop, quads, quadstrip/flat
+ *       -- other primitives are all well-formed (need tristrip-1,fake-poly)
+ *
+ */
+static void emit_ubyte_rgba3( GLcontext *ctx,
+		       struct radeon_dma_region *rvb,
+		       char *data,
+		       int stride,
+		       int count )
+{
+   int i;
+   radeon_color_t *out = (radeon_color_t *)(rvb->start + rvb->address);
+
+   if (RADEON_DEBUG & DEBUG_VERTS)
+      fprintf(stderr, "%s count %d stride %d out %p\n",
+	      __FUNCTION__, count, stride, out);
+
+   for (i = 0; i < count; i++) {
+      out->red   = *data;
+      out->green = *(data+1);
+      out->blue  = *(data+2);
+      out->alpha = 0xFF;
+      out++;
+      data += stride;
+   }
+}
+
+
+#if defined(USE_X86_ASM)
+#define COPY_DWORDS( dst, src, nr )					\
+do {									\
+	int __tmp;							\
+	__asm__ __volatile__( "rep ; movsl"				\
+			      : "=%c" (__tmp), "=D" (dst), "=S" (__tmp)	\
+			      : "0" (nr),				\
+			        "D" ((long)dst),			\
+			        "S" ((long)src) );			\
+} while (0)
+#else
+#define COPY_DWORDS( dst, src, nr )		\
+do {						\
+   int j;					\
+   for ( j = 0 ; j < nr ; j++ )			\
+      dst[j] = ((int *)src)[j];			\
+   dst += nr;					\
+} while (0)
+#endif
+
+
+
+static void emit_ubyte_rgba4( GLcontext *ctx,
+			      struct radeon_dma_region *rvb,
+			      char *data,
+			      int stride,
+			      int count )
+{
+   int i;
+   int *out = (int *)(rvb->address + rvb->start);
+
+   if (RADEON_DEBUG & DEBUG_VERTS)
+      fprintf(stderr, "%s count %d stride %d\n",
+	      __FUNCTION__, count, stride);
+
+   if (stride == 4)
+       COPY_DWORDS( out, data, count );
+   else
+      for (i = 0; i < count; i++) {
+	 *out++ = LE32_TO_CPU(*(int *)data);
+	 data += stride;
+      }
+}
+
+
+static void emit_ubyte_rgba( GLcontext *ctx,
+			     struct radeon_dma_region *rvb,
+			     char *data,
+			     int size,
+			     int stride,
+			     int count )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if (RADEON_DEBUG & DEBUG_VERTS)
+      fprintf(stderr, "%s %d/%d\n", __FUNCTION__, count, size);
+
+   assert (!rvb->buf);
+
+   if (stride == 0) {
+      radeonAllocDmaRegion( rmesa, rvb, 4, 4 );
+      count = 1;
+      rvb->aos_start = GET_START(rvb);
+      rvb->aos_stride = 0;
+      rvb->aos_size = 1;
+   }
+   else {
+      radeonAllocDmaRegion( rmesa, rvb, 4 * count, 4 );	/* alignment? */
+      rvb->aos_start = GET_START(rvb);
+      rvb->aos_stride = 1;
+      rvb->aos_size = 1;
+   }
+
+   /* Emit the data
+    */
+   switch (size) {
+   case 3:
+      emit_ubyte_rgba3( ctx, rvb, data, stride, count );
+      break;
+   case 4:
+      emit_ubyte_rgba4( ctx, rvb, data, stride, count );
+      break;
+   default:
+      assert(0);
+      exit(1);
+      break;
+   }
+}
+
+
+
+
+static void emit_vec8( GLcontext *ctx,
+		       struct radeon_dma_region *rvb,
+		       char *data,
+		       int stride,
+		       int count )
+{
+   int i;
+   int *out = (int *)(rvb->address + rvb->start);
+
+   if (RADEON_DEBUG & DEBUG_VERTS)
+      fprintf(stderr, "%s count %d stride %d\n",
+	      __FUNCTION__, count, stride);
+
+   if (stride == 8)
+      COPY_DWORDS( out, data, count*2 );
+   else
+      for (i = 0; i < count; i++) {
+	 out[0] = *(int *)data;
+	 out[1] = *(int *)(data+4);
+	 out += 2;
+	 data += stride;
+      }
+}
+
+static void emit_vec12( GLcontext *ctx,
+		       struct radeon_dma_region *rvb,
+		       char *data,
+		       int stride,
+		       int count )
+{
+   int i;
+   int *out = (int *)(rvb->address + rvb->start);
+
+   if (RADEON_DEBUG & DEBUG_VERTS)
+      fprintf(stderr, "%s count %d stride %d out %p data %p\n",
+	      __FUNCTION__, count, stride, out, data);
+
+   if (stride == 12)
+      COPY_DWORDS( out, data, count*3 );
+   else
+      for (i = 0; i < count; i++) {
+	 out[0] = *(int *)data;
+	 out[1] = *(int *)(data+4);
+	 out[2] = *(int *)(data+8);
+	 out += 3;
+	 data += stride;
+      }
+}
+
+static void emit_vec16( GLcontext *ctx,
+			struct radeon_dma_region *rvb,
+			char *data,
+			int stride,
+			int count )
+{
+   int i;
+   int *out = (int *)(rvb->address + rvb->start);
+
+   if (RADEON_DEBUG & DEBUG_VERTS)
+      fprintf(stderr, "%s count %d stride %d\n",
+	      __FUNCTION__, count, stride);
+
+   if (stride == 16)
+      COPY_DWORDS( out, data, count*4 );
+   else
+      for (i = 0; i < count; i++) {
+	 out[0] = *(int *)data;
+	 out[1] = *(int *)(data+4);
+	 out[2] = *(int *)(data+8);
+	 out[3] = *(int *)(data+12);
+	 out += 4;
+	 data += stride;
+      }
+}
+
+
+static void emit_vector( GLcontext *ctx,
+			 struct radeon_dma_region *rvb,
+			 char *data,
+			 int size,
+			 int stride,
+			 int count )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if (RADEON_DEBUG & DEBUG_VERTS)
+      fprintf(stderr, "%s count %d size %d stride %d\n",
+	      __FUNCTION__, count, size, stride);
+
+   assert (!rvb->buf);
+
+   if (stride == 0) {
+      radeonAllocDmaRegion( rmesa, rvb, size * 4, 4 );
+      count = 1;
+      rvb->aos_start = GET_START(rvb);
+      rvb->aos_stride = 0;
+      rvb->aos_size = size;
+   }
+   else {
+      radeonAllocDmaRegion( rmesa, rvb, size * count * 4, 4 );	/* alignment? */
+      rvb->aos_start = GET_START(rvb);
+      rvb->aos_stride = size;
+      rvb->aos_size = size;
+   }
+
+   /* Emit the data
+    */
+   switch (size) {
+   case 2:
+      emit_vec8( ctx, rvb, data, stride, count );
+      break;
+   case 3:
+      emit_vec12( ctx, rvb, data, stride, count );
+      break;
+   case 4:
+      emit_vec16( ctx, rvb, data, stride, count );
+      break;
+   default:
+      assert(0);
+      exit(1);
+      break;
+   }
+
+}
+
+
+
+static void emit_s0_vec( GLcontext *ctx,
+			 struct radeon_dma_region *rvb,
+			 char *data,
+			 int stride,
+			 int count )
+{
+   int i;
+   int *out = (int *)(rvb->address + rvb->start);
+
+   if (RADEON_DEBUG & DEBUG_VERTS)
+      fprintf(stderr, "%s count %d stride %d\n",
+	      __FUNCTION__, count, stride);
+
+   for (i = 0; i < count; i++) {
+      out[0] = *(int *)data;
+      out[1] = 0;
+      out += 2;
+      data += stride;
+   }
+}
+
+static void emit_stq_vec( GLcontext *ctx,
+			 struct radeon_dma_region *rvb,
+			 char *data,
+			 int stride,
+			 int count )
+{
+   int i;
+   int *out = (int *)(rvb->address + rvb->start);
+
+   if (RADEON_DEBUG & DEBUG_VERTS)
+      fprintf(stderr, "%s count %d stride %d\n",
+	      __FUNCTION__, count, stride);
+
+   for (i = 0; i < count; i++) {
+      out[0] = *(int *)data;
+      out[1] = *(int *)(data+4);
+      out[2] = *(int *)(data+12);
+      out += 3;
+      data += stride;
+   }
+}
+
+
+
+
+static void emit_tex_vector( GLcontext *ctx,
+			     struct radeon_dma_region *rvb,
+			     char *data,
+			     int size,
+			     int stride,
+			     int count )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   int emitsize;
+
+   if (RADEON_DEBUG & DEBUG_VERTS)
+      fprintf(stderr, "%s %d/%d\n", __FUNCTION__, count, size);
+
+   assert (!rvb->buf);
+
+   switch (size) {
+   case 4: emitsize = 3; break;
+   default: emitsize = 2; break;
+   }
+
+
+   if (stride == 0) {
+      radeonAllocDmaRegion( rmesa, rvb, 4 * emitsize, 4 );
+      count = 1;
+      rvb->aos_start = GET_START(rvb);
+      rvb->aos_stride = 0;
+      rvb->aos_size = emitsize;
+   }
+   else {
+      radeonAllocDmaRegion( rmesa, rvb, 4 * emitsize * count, 4 );
+      rvb->aos_start = GET_START(rvb);
+      rvb->aos_stride = emitsize;
+      rvb->aos_size = emitsize;
+   }
+
+
+   /* Emit the data
+    */
+   switch (size) {
+   case 1:
+      emit_s0_vec( ctx, rvb, data, stride, count ); 
+      break;
+   case 2:
+      emit_vec8( ctx, rvb, data, stride, count );
+      break;
+   case 3:
+      emit_vec8( ctx, rvb, data, stride, count );
+      break;
+   case 4:
+      emit_stq_vec( ctx, rvb, data, stride, count );
+      break;
+   default:
+      assert(0);
+      exit(1);
+      break;
+   }
+}
+
+
+
+
+/* Emit any changed arrays to new agp memory, re-emit a packet to
+ * update the arrays.  
+ */
+void radeonEmitArrays( GLcontext *ctx, GLuint inputs )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+   struct vertex_buffer *VB = &TNL_CONTEXT( ctx )->vb;
+   struct radeon_dma_region **component = rmesa->tcl.aos_components;
+   GLuint nr = 0;
+   GLuint vfmt = 0;
+   GLuint count = VB->Count;
+   GLuint vtx;
+   
+   if (RADEON_DEBUG & DEBUG_VERTS) 
+      _tnl_print_vert_flags( __FUNCTION__, inputs );
+
+   if (1) {
+      if (!rmesa->tcl.obj.buf) 
+	 emit_vector( ctx, 
+		      &rmesa->tcl.obj, 
+		      (char *)VB->ObjPtr->data,
+		      VB->ObjPtr->size,
+		      VB->ObjPtr->stride,
+		      count);
+
+      switch( VB->ObjPtr->size ) {
+      case 4: vfmt |= RADEON_CP_VC_FRMT_W0;
+      case 3: vfmt |= RADEON_CP_VC_FRMT_Z;
+      case 2: vfmt |= RADEON_CP_VC_FRMT_XY;
+      default:
+      }
+      component[nr++] = &rmesa->tcl.obj;
+   }
+   
+
+   if (inputs & VERT_BIT_NORMAL) {
+      if (!rmesa->tcl.norm.buf)
+	 emit_vector( ctx, 
+		      &(rmesa->tcl.norm), 
+		      (char *)VB->NormalPtr->data,
+		      3,
+		      VB->NormalPtr->stride,
+		      count);
+
+      vfmt |= RADEON_CP_VC_FRMT_N0;
+      component[nr++] = &rmesa->tcl.norm;
+   }
+
+   if (inputs & VERT_BIT_COLOR0) {
+      if (VB->ColorPtr[0]->Type == GL_UNSIGNED_BYTE) {
+	 if (!rmesa->tcl.rgba.buf)
+	    emit_ubyte_rgba( ctx, 
+			     &rmesa->tcl.rgba, 
+			     (char *)VB->ColorPtr[0]->Ptr,
+			     VB->ColorPtr[0]->Size,
+			     VB->ColorPtr[0]->StrideB,
+			     count);
+
+	 vfmt |= RADEON_CP_VC_FRMT_PKCOLOR; 
+      }
+      else {
+	 int emitsize;
+
+	 if (VB->ColorPtr[0]->Size == 4 &&
+	     (VB->ColorPtr[0]->StrideB != 0 ||
+	      ((GLfloat *)VB->ColorPtr[0]->Ptr)[3] != 1.0)) { 
+	    vfmt |= RADEON_CP_VC_FRMT_FPCOLOR | RADEON_CP_VC_FRMT_FPALPHA;
+	    emitsize = 4;
+	 }
+	 else { 
+	    vfmt |= RADEON_CP_VC_FRMT_FPCOLOR;
+	    emitsize = 3;
+	 }
+
+	 if (!rmesa->tcl.rgba.buf)
+	    emit_vector( ctx, 
+			 &(rmesa->tcl.rgba), 
+			 (char *)VB->ColorPtr[0]->Ptr,
+			 emitsize,
+			 VB->ColorPtr[0]->StrideB,
+			 count);
+      }
+
+      component[nr++] = &rmesa->tcl.rgba;
+   }
+
+
+   if (inputs & VERT_BIT_COLOR1) {
+      if (!rmesa->tcl.spec.buf) {
+	 if (VB->SecondaryColorPtr[0]->Type != GL_UNSIGNED_BYTE)
+	    radeon_import_float_spec_colors( ctx );
+
+	 emit_ubyte_rgba( ctx, 
+			  &rmesa->tcl.spec, 
+			  (char *)VB->SecondaryColorPtr[0]->Ptr,
+			  3,
+			  VB->SecondaryColorPtr[0]->StrideB,
+			  count);
+      }
+
+      vfmt |= RADEON_CP_VC_FRMT_PKSPEC; 
+      component[nr++] = &rmesa->tcl.spec;
+   }
+
+   vtx = (rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] &
+	  ~(RADEON_TCL_VTX_Q0|RADEON_TCL_VTX_Q1));
+      
+   if (inputs & VERT_BIT_TEX0) {
+      if (!rmesa->tcl.tex[0].buf)
+	 emit_tex_vector( ctx, 
+			  &(rmesa->tcl.tex[0]), 
+			  (char *)VB->TexCoordPtr[0]->data,
+			  VB->TexCoordPtr[0]->size,
+			  VB->TexCoordPtr[0]->stride,
+			  count );
+
+      switch( VB->TexCoordPtr[0]->size ) {
+      case 4:
+	 vtx |= RADEON_TCL_VTX_Q0; 
+	 vfmt |= RADEON_CP_VC_FRMT_Q0;
+      default: 
+	 vfmt |= RADEON_CP_VC_FRMT_ST0;
+      }
+      component[nr++] = &rmesa->tcl.tex[0];
+   }
+
+   if (inputs & VERT_BIT_TEX1) {
+      if (!rmesa->tcl.tex[1].buf)
+	 emit_tex_vector( ctx, 
+			  &(rmesa->tcl.tex[1]), 
+			  (char *)VB->TexCoordPtr[1]->data,
+			  VB->TexCoordPtr[1]->size,
+			  VB->TexCoordPtr[1]->stride,
+			  count );
+	 
+      switch( VB->TexCoordPtr[1]->size ) {
+      case 4: 
+	 vtx |= RADEON_TCL_VTX_Q1;
+	 vfmt |= RADEON_CP_VC_FRMT_Q1;
+      default: 
+	 vfmt |= RADEON_CP_VC_FRMT_ST1;
+      }
+      component[nr++] = &rmesa->tcl.tex[1];
+   }
+
+   if (vtx != rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT]) {
+      RADEON_STATECHANGE( rmesa, tcl );
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] = vtx;
+   }
+
+   rmesa->tcl.nr_aos_components = nr;
+   rmesa->tcl.vertex_format = vfmt;
+}
+
+
+void radeonReleaseArrays( GLcontext *ctx, GLuint newinputs )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+
+   if (RADEON_DEBUG & DEBUG_VERTS) 
+      _tnl_print_vert_flags( __FUNCTION__, newinputs );
+
+   if (newinputs & VERT_BIT_POS) 
+     radeonReleaseDmaRegion( rmesa, &rmesa->tcl.obj, __FUNCTION__ );
+
+   if (newinputs & VERT_BIT_NORMAL) 
+      radeonReleaseDmaRegion( rmesa, &rmesa->tcl.norm, __FUNCTION__ );
+
+   if (newinputs & VERT_BIT_COLOR0) 
+      radeonReleaseDmaRegion( rmesa, &rmesa->tcl.rgba, __FUNCTION__ );
+
+   if (newinputs & VERT_BIT_COLOR1) 
+      radeonReleaseDmaRegion( rmesa, &rmesa->tcl.spec, __FUNCTION__ );
+
+   if (newinputs & VERT_BIT_TEX0)
+      radeonReleaseDmaRegion( rmesa, &rmesa->tcl.tex[0], __FUNCTION__ );
+
+   if (newinputs & VERT_BIT_TEX1)
+      radeonReleaseDmaRegion( rmesa, &rmesa->tcl.tex[1], __FUNCTION__ );
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_maos_vbtmp.h b/src/mesa/drivers/dri/radeon/radeon_maos_vbtmp.h
new file mode 100644
index 00000000000..b379bad985f
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_maos_vbtmp.h
@@ -0,0 +1,368 @@
+/*
+ * Mesa 3-D graphics library
+ * Version:  4.1
+ *
+ * Copyright (C) 1999-2002  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 <[email protected]>
+ */
+
+#ifndef LOCALVARS
+#define LOCALVARS
+#endif
+
+#undef TCL_DEBUG
+#ifndef TCL_DEBUG
+#define TCL_DEBUG 0
+#endif
+
+static void TAG(emit)( GLcontext *ctx,
+		       GLuint start, GLuint end,
+		       void *dest )
+{
+   LOCALVARS
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+   GLuint (*tc0)[4], (*tc1)[4], (*tc2)[4];
+   GLfloat (*fog)[4];
+   GLuint (*norm)[4];
+   GLubyte (*col)[4], (*spec)[4];
+   GLuint tc0_stride, tc1_stride, col_stride, spec_stride, fog_stride;
+   GLuint tc2_stride, norm_stride;
+   GLuint (*coord)[4];
+   GLuint coord_stride; /* object coordinates */
+   GLubyte dummy[4];
+   int i;
+
+   union emit_union *v = (union emit_union *)dest;
+
+   if (RADEON_DEBUG & DEBUG_VERTS)
+      fprintf(stderr, "%s\n", __FUNCTION__); 
+
+   /* The vertex code expects Obj to be clean to element 3.  To fix
+    * this, add more vertex code (for obj-2, obj-3) or preferably move
+    * to maos.  
+    */
+   if (VB->ObjPtr->size < 3) {
+      if (VB->ObjPtr->flags & VEC_NOT_WRITEABLE) {
+	 VB->import_data( ctx, VERT_BIT_POS, VEC_NOT_WRITEABLE );
+      }
+      _mesa_vector4f_clean_elem( VB->ObjPtr, VB->Count, 2 );
+   }
+
+   if (DO_W && VB->ObjPtr->size < 4) {
+      if (VB->ObjPtr->flags & VEC_NOT_WRITEABLE) {
+	 VB->import_data( ctx, VERT_BIT_POS, VEC_NOT_WRITEABLE );
+      }
+      _mesa_vector4f_clean_elem( VB->ObjPtr, VB->Count, 3 );
+   }
+
+   coord = (GLuint (*)[4])VB->ObjPtr->data;
+   coord_stride = VB->ObjPtr->stride;
+
+   if (DO_TEX2) {
+      const GLuint t2 = GET_TEXSOURCE(2);
+      tc2 = (GLuint (*)[4])VB->TexCoordPtr[t2]->data;
+      tc2_stride = VB->TexCoordPtr[t2]->stride;
+      if (DO_PTEX && VB->TexCoordPtr[t2]->size < 4) {
+	 if (VB->TexCoordPtr[t2]->flags & VEC_NOT_WRITEABLE) {
+	    VB->import_data( ctx, VERT_BIT_TEX2, VEC_NOT_WRITEABLE );
+	 }
+	 _mesa_vector4f_clean_elem( VB->TexCoordPtr[t2], VB->Count, 3 );
+      }
+   }
+
+   if (DO_TEX1) {
+      if (VB->TexCoordPtr[1]) {
+	 const GLuint t1 = GET_TEXSOURCE(1);
+	 tc1 = (GLuint (*)[4])VB->TexCoordPtr[t1]->data;
+	 tc1_stride = VB->TexCoordPtr[t1]->stride;
+	 if (DO_PTEX && VB->TexCoordPtr[t1]->size < 4) {
+	    if (VB->TexCoordPtr[t1]->flags & VEC_NOT_WRITEABLE) {
+	       VB->import_data( ctx, VERT_BIT_TEX1, VEC_NOT_WRITEABLE );
+	    }
+	    _mesa_vector4f_clean_elem( VB->TexCoordPtr[t1], VB->Count, 3 );
+	 }
+      } else {
+	 tc1 = (GLuint (*)[4])&ctx->Current.Attrib[VERT_ATTRIB_TEX1]; /* could be anything, really */
+	 tc1_stride = 0;
+      }
+   }
+
+   if (DO_TEX0) {
+      if (VB->TexCoordPtr[0]) {
+	 const GLuint t0 = GET_TEXSOURCE(0);
+	 tc0_stride = VB->TexCoordPtr[t0]->stride;
+	 tc0 = (GLuint (*)[4])VB->TexCoordPtr[t0]->data;
+	 if (DO_PTEX && VB->TexCoordPtr[t0]->size < 4) {
+	    if (VB->TexCoordPtr[t0]->flags & VEC_NOT_WRITEABLE) {
+	       VB->import_data( ctx, VERT_BIT_TEX0, VEC_NOT_WRITEABLE );
+	    }
+	    _mesa_vector4f_clean_elem( VB->TexCoordPtr[t0], VB->Count, 3 );
+	 }
+      } else {
+	 tc0 = (GLuint (*)[4])&ctx->Current.Attrib[VERT_ATTRIB_TEX0]; /* could be anything, really */
+	 tc0_stride = 0;
+      }
+	 
+   }
+
+   if (DO_NORM) {
+      if (VB->NormalPtr) {
+	 norm_stride = VB->NormalPtr->stride;
+	 norm = (GLuint (*)[4])VB->NormalPtr->data;
+      } else {
+	 norm_stride = 0;
+	 norm = (GLuint (*)[4])&ctx->Current.Attrib[VERT_ATTRIB_NORMAL];
+      }
+   }
+
+   if (DO_RGBA) {
+      if (VB->ColorPtr[0]) {
+	 /* This is incorrect when colormaterial is enabled:
+	  */
+	 if (VB->ColorPtr[0]->Type != GL_UNSIGNED_BYTE) {
+	    if (0) fprintf(stderr, "IMPORTING FLOAT COLORS\n");
+	    IMPORT_FLOAT_COLORS( ctx );
+	 }
+	 col = (GLubyte (*)[4])VB->ColorPtr[0]->Ptr;
+	 col_stride = VB->ColorPtr[0]->StrideB;
+      } else {
+	 col = &dummy; /* any old memory is fine */
+	 col_stride = 0;
+      }
+   }
+
+   if (DO_SPEC) {
+      if (VB->SecondaryColorPtr[0]) {
+	 if (VB->SecondaryColorPtr[0]->Type != GL_UNSIGNED_BYTE)
+	    IMPORT_FLOAT_SPEC_COLORS( ctx );
+	 spec = (GLubyte (*)[4])VB->SecondaryColorPtr[0]->Ptr;
+	 spec_stride = VB->SecondaryColorPtr[0]->StrideB;
+      } else {
+	 spec = &dummy;
+	 spec_stride = 0;
+      }
+   }
+
+   if (DO_FOG) {
+      if (VB->FogCoordPtr) {
+	 fog = VB->FogCoordPtr->data;
+	 fog_stride = VB->FogCoordPtr->stride;
+      } else {
+	 fog = (GLfloat (*)[4])&dummy; fog[0][0] = 0.0F;
+	 fog_stride = 0;
+      }
+   }
+   
+   
+   if (VB->importable_data) {
+      if (start) {
+	 coord =  (GLuint (*)[4])((GLubyte *)coord + start * coord_stride);
+	 if (DO_TEX0)
+	    tc0 =  (GLuint (*)[4])((GLubyte *)tc0 + start * tc0_stride);
+	 if (DO_TEX1) 
+	    tc1 =  (GLuint (*)[4])((GLubyte *)tc1 + start * tc1_stride);
+	 if (DO_TEX2) 
+	    tc2 =  (GLuint (*)[4])((GLubyte *)tc2 + start * tc2_stride);
+	 if (DO_NORM) 
+	    norm =  (GLuint (*)[4])((GLubyte *)norm + start * norm_stride);
+	 if (DO_RGBA) 
+	    STRIDE_4UB(col, start * col_stride);
+	 if (DO_SPEC)
+	    STRIDE_4UB(spec, start * spec_stride);
+	 if (DO_FOG)
+	    fog =  (GLfloat (*)[4])((GLubyte *)fog + start * fog_stride);
+      }
+
+      for (i=start; i < end; i++) {
+	 v[0].ui = coord[0][0];
+	 v[1].ui = coord[0][1];
+	 v[2].ui = coord[0][2];
+	 if (TCL_DEBUG) fprintf(stderr, "%d: %.2f %.2f %.2f ", i, v[0].f, v[1].f, v[2].f);
+	 if (DO_W) {
+	    v[3].ui = coord[0][3];
+	    if (TCL_DEBUG) fprintf(stderr, "%.2f ", v[3].f);
+	    v += 4;
+	 } 
+	 else
+	    v += 3;
+	 coord =  (GLuint (*)[4])((GLubyte *)coord +  coord_stride);
+
+	 if (DO_NORM) {
+	    v[0].ui = norm[0][0];
+	    v[1].ui = norm[0][1];
+	    v[2].ui = norm[0][2];
+	    if (TCL_DEBUG) fprintf(stderr, "norm: %.2f %.2f %.2f ", v[0].f, v[1].f, v[2].f);
+	    v += 3;
+	    norm =  (GLuint (*)[4])((GLubyte *)norm +  norm_stride);
+	 }
+	 if (DO_RGBA) {
+	    v[0].ui = LE32_TO_CPU(*(GLuint *)&col[0]);
+	    STRIDE_4UB(col, col_stride);
+	    if (TCL_DEBUG) fprintf(stderr, "%x ", v[0].ui);
+	    v++;
+	 }
+	 if (DO_SPEC || DO_FOG) {
+	    if (DO_SPEC) {
+	       v[0].specular.red   = spec[0][0];
+	       v[0].specular.green = spec[0][1];
+	       v[0].specular.blue  = spec[0][2];
+	       STRIDE_4UB(spec, spec_stride);
+	    }
+	    if (DO_FOG) {
+	       v[0].specular.alpha = fog[0][0] * 255.0;
+               fog = (GLfloat (*)[4])((GLubyte *)fog + fog_stride);
+	    }
+	    if (TCL_DEBUG) fprintf(stderr, "%x ", v[0].ui);
+	    v++;
+	 }
+	 if (DO_TEX0) {
+	    v[0].ui = tc0[0][0];
+	    v[1].ui = tc0[0][1];
+	    if (TCL_DEBUG) fprintf(stderr, "t0: %.2f %.2f ", v[0].f, v[1].f);
+	    if (DO_PTEX) {
+	       v[2].ui = tc0[0][3];
+	       if (TCL_DEBUG) fprintf(stderr, "%.2f ", v[2].f);
+	       v += 3;
+	    } 
+	    else
+	       v += 2;
+	    tc0 =  (GLuint (*)[4])((GLubyte *)tc0 +  tc0_stride);
+	 }
+	 if (DO_TEX1) {
+	    v[0].ui = tc1[0][0];
+	    v[1].ui = tc1[0][1];
+	    if (TCL_DEBUG) fprintf(stderr, "t1: %.2f %.2f ", v[0].f, v[1].f);
+	    if (DO_PTEX) {
+	       v[2].ui = tc1[0][3];
+	       if (TCL_DEBUG) fprintf(stderr, "%.2f ", v[2].f);
+	       v += 3;
+	    } 
+	    else
+	       v += 2;
+	    tc1 =  (GLuint (*)[4])((GLubyte *)tc1 +  tc1_stride);
+	 } 
+	 if (DO_TEX2) {
+	    v[0].ui = tc2[0][0];
+	    v[1].ui = tc2[0][1];
+	    if (DO_PTEX) {
+	       v[2].ui = tc2[0][3];
+	       v += 3;
+	    } 
+	    else
+	       v += 2;
+	    tc2 =  (GLuint (*)[4])((GLubyte *)tc2 +  tc2_stride);
+	 } 
+	 if (TCL_DEBUG) fprintf(stderr, "\n");
+      }
+   } else {
+      for (i=start; i < end; i++) {
+	 v[0].ui = coord[i][0];
+	 v[1].ui = coord[i][1];
+	 v[2].ui = coord[i][2];
+	 if (DO_W) {
+	    v[3].ui = coord[i][3];
+	    v += 4;
+	 } 
+	 else
+	    v += 3;
+
+	 if (DO_NORM) {
+	    v[0].ui = norm[i][0];
+	    v[1].ui = norm[i][1];
+	    v[2].ui = norm[i][2];
+	    v += 3;
+	 }
+	 if (DO_RGBA) {
+	    v[0].ui = LE32_TO_CPU(*(GLuint *)&col[i]);
+	    v++;
+	 }
+	 if (DO_SPEC || DO_FOG) {
+	    if (DO_SPEC) {
+	       v[0].specular.red   = spec[i][0];
+	       v[0].specular.green = spec[i][1];
+	       v[0].specular.blue  = spec[i][2];
+	    }
+	    if (DO_FOG) {
+               GLfloat *f = (GLfloat *) ((GLubyte *)fog + fog_stride);
+               v[0].specular.alpha = *f * 255.0;
+	    }
+	    v++;
+	 }
+	 if (DO_TEX0) {
+	    v[0].ui = tc0[i][0];
+	    v[1].ui = tc0[i][1];
+	    if (DO_PTEX) {
+	       v[2].ui = tc0[i][3];
+	       v += 3;
+	    } 
+	    else
+	       v += 2;
+	 }
+	 if (DO_TEX1) {
+	    v[0].ui = tc1[i][0];
+	    v[1].ui = tc1[i][1];
+	    if (DO_PTEX) {
+	       v[2].ui = tc1[i][3];
+	       v += 3;
+	    } 
+	    else
+	       v += 2;
+	 } 
+	 if (DO_TEX2) {
+	    v[0].ui = tc2[i][0];
+	    v[1].ui = tc2[i][1];
+	    if (DO_PTEX) {
+	       v[2].ui = tc2[i][3];
+	       v += 3;
+	    } 
+	    else
+	       v += 2;
+	 } 
+      }
+   }
+}
+
+
+
+static void TAG(init)( void )
+{
+   int sz = 3;
+   if (DO_W) sz++;
+   if (DO_NORM) sz += 3;
+   if (DO_RGBA) sz++;
+   if (DO_SPEC || DO_FOG) sz++;
+   if (DO_TEX0) sz += 2;
+   if (DO_TEX0 && DO_PTEX) sz++;
+   if (DO_TEX1) sz += 2;
+   if (DO_TEX1 && DO_PTEX) sz++;
+   if (DO_TEX2) sz += 2;
+   if (DO_TEX2 && DO_PTEX) sz++;
+
+   setup_tab[IDX].emit = TAG(emit);
+   setup_tab[IDX].vertex_format = IND;
+   setup_tab[IDX].vertex_size = sz;
+}
+
+
+#undef IND
+#undef TAG
+#undef IDX
diff --git a/src/mesa/drivers/dri/radeon/radeon_maos_verts.c b/src/mesa/drivers/dri/radeon/radeon_maos_verts.c
new file mode 100644
index 00000000000..39b1f575074
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_maos_verts.c
@@ -0,0 +1,335 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     Tungsten Graphics Inc., Austin, 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, 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 <[email protected]>
+ */
+
+#include "glheader.h"
+#include "imports.h"
+#include "mtypes.h"
+
+#include "array_cache/acache.h"
+#include "tnl/tnl.h"
+#include "tnl/t_pipeline.h"
+#include "tnl/t_imm_debug.h"
+
+#include "radeon_context.h"
+#include "radeon_state.h"
+#include "radeon_ioctl.h"
+#include "radeon_tex.h"
+#include "radeon_tcl.h"
+#include "radeon_swtcl.h"
+#include "radeon_maos.h"
+
+
+#define RADEON_TCL_MAX_SETUP 13
+
+union emit_union { float f; GLuint ui; radeon_color_t specular; };
+
+static struct {
+   void   (*emit)( GLcontext *, GLuint, GLuint, void * );
+   GLuint vertex_size;
+   GLuint vertex_format;
+} setup_tab[RADEON_TCL_MAX_SETUP];
+
+#define DO_W    (IND & RADEON_CP_VC_FRMT_W0)
+#define DO_RGBA (IND & RADEON_CP_VC_FRMT_PKCOLOR)
+#define DO_SPEC (IND & RADEON_CP_VC_FRMT_PKSPEC)
+#define DO_FOG  (IND & RADEON_CP_VC_FRMT_PKSPEC)
+#define DO_TEX0 (IND & RADEON_CP_VC_FRMT_ST0)
+#define DO_TEX1 (IND & RADEON_CP_VC_FRMT_ST1)
+#define DO_PTEX (IND & RADEON_CP_VC_FRMT_Q0)
+#define DO_NORM (IND & RADEON_CP_VC_FRMT_N0)
+
+#define DO_TEX2 0
+#define DO_TEX3 0
+
+#define GET_TEXSOURCE(n)  n
+#define GET_UBYTE_COLOR_STORE() &RADEON_CONTEXT(ctx)->UbyteColor
+#define GET_UBYTE_SPEC_COLOR_STORE() &RADEON_CONTEXT(ctx)->UbyteSecondaryColor
+
+#define IMPORT_FLOAT_COLORS radeon_import_float_colors
+#define IMPORT_FLOAT_SPEC_COLORS radeon_import_float_spec_colors
+
+/***********************************************************************
+ *             Generate vertex emit functions               *
+ ***********************************************************************/
+
+
+/* Defined in order of increasing vertex size:
+ */
+#define IDX 0
+#define IND (RADEON_CP_VC_FRMT_XY|		\
+	     RADEON_CP_VC_FRMT_Z|		\
+	     RADEON_CP_VC_FRMT_PKCOLOR)
+#define TAG(x) x##_rgba
+#include "radeon_maos_vbtmp.h"
+
+#define IDX 1
+#define IND (RADEON_CP_VC_FRMT_XY|		\
+	     RADEON_CP_VC_FRMT_Z|		\
+	     RADEON_CP_VC_FRMT_N0)
+#define TAG(x) x##_n
+#include "radeon_maos_vbtmp.h"
+
+#define IDX 2
+#define IND (RADEON_CP_VC_FRMT_XY|		\
+	     RADEON_CP_VC_FRMT_Z|		\
+	     RADEON_CP_VC_FRMT_PKCOLOR|		\
+	     RADEON_CP_VC_FRMT_ST0)
+#define TAG(x) x##_rgba_st
+#include "radeon_maos_vbtmp.h"
+
+#define IDX 3
+#define IND (RADEON_CP_VC_FRMT_XY|		\
+	     RADEON_CP_VC_FRMT_Z|		\
+	     RADEON_CP_VC_FRMT_PKCOLOR|		\
+	     RADEON_CP_VC_FRMT_N0)
+#define TAG(x) x##_rgba_n
+#include "radeon_maos_vbtmp.h"
+
+#define IDX 4
+#define IND (RADEON_CP_VC_FRMT_XY|		\
+	     RADEON_CP_VC_FRMT_Z|		\
+	     RADEON_CP_VC_FRMT_ST0|		\
+	     RADEON_CP_VC_FRMT_N0)
+#define TAG(x) x##_st_n
+#include "radeon_maos_vbtmp.h"
+
+#define IDX 5
+#define IND (RADEON_CP_VC_FRMT_XY|		\
+	     RADEON_CP_VC_FRMT_Z|		\
+	     RADEON_CP_VC_FRMT_PKCOLOR|		\
+	     RADEON_CP_VC_FRMT_ST0|		\
+	     RADEON_CP_VC_FRMT_ST1)
+#define TAG(x) x##_rgba_st_st
+#include "radeon_maos_vbtmp.h"
+
+#define IDX 6
+#define IND (RADEON_CP_VC_FRMT_XY|		\
+	     RADEON_CP_VC_FRMT_Z|		\
+	     RADEON_CP_VC_FRMT_PKCOLOR|		\
+	     RADEON_CP_VC_FRMT_ST0|		\
+	     RADEON_CP_VC_FRMT_N0)
+#define TAG(x) x##_rgba_st_n
+#include "radeon_maos_vbtmp.h"
+
+#define IDX 7
+#define IND (RADEON_CP_VC_FRMT_XY|		\
+	     RADEON_CP_VC_FRMT_Z|		\
+	     RADEON_CP_VC_FRMT_PKCOLOR|		\
+	     RADEON_CP_VC_FRMT_PKSPEC|		\
+	     RADEON_CP_VC_FRMT_ST0|		\
+	     RADEON_CP_VC_FRMT_ST1)
+#define TAG(x) x##_rgba_spec_st_st
+#include "radeon_maos_vbtmp.h"
+
+#define IDX 8
+#define IND (RADEON_CP_VC_FRMT_XY|		\
+	     RADEON_CP_VC_FRMT_Z|		\
+	     RADEON_CP_VC_FRMT_ST0|		\
+	     RADEON_CP_VC_FRMT_ST1|		\
+	     RADEON_CP_VC_FRMT_N0)
+#define TAG(x) x##_st_st_n
+#include "radeon_maos_vbtmp.h"
+
+#define IDX 9
+#define IND (RADEON_CP_VC_FRMT_XY|		\
+	     RADEON_CP_VC_FRMT_Z|		\
+	     RADEON_CP_VC_FRMT_PKCOLOR|		\
+	     RADEON_CP_VC_FRMT_PKSPEC|		\
+	     RADEON_CP_VC_FRMT_ST0|		\
+	     RADEON_CP_VC_FRMT_ST1|		\
+	     RADEON_CP_VC_FRMT_N0)
+#define TAG(x) x##_rgpa_spec_st_st_n
+#include "radeon_maos_vbtmp.h"
+
+#define IDX 10
+#define IND (RADEON_CP_VC_FRMT_XY|		\
+	     RADEON_CP_VC_FRMT_Z|		\
+	     RADEON_CP_VC_FRMT_PKCOLOR|		\
+	     RADEON_CP_VC_FRMT_ST0|		\
+	     RADEON_CP_VC_FRMT_Q0)
+#define TAG(x) x##_rgba_stq
+#include "radeon_maos_vbtmp.h"
+
+#define IDX 11
+#define IND (RADEON_CP_VC_FRMT_XY|		\
+	     RADEON_CP_VC_FRMT_Z|		\
+	     RADEON_CP_VC_FRMT_PKCOLOR|		\
+	     RADEON_CP_VC_FRMT_ST1|		\
+	     RADEON_CP_VC_FRMT_Q1|		\
+	     RADEON_CP_VC_FRMT_ST0|		\
+	     RADEON_CP_VC_FRMT_Q0)
+#define TAG(x) x##_rgba_stq_stq
+#include "radeon_maos_vbtmp.h"
+
+#define IDX 12
+#define IND (RADEON_CP_VC_FRMT_XY|		\
+	     RADEON_CP_VC_FRMT_Z|		\
+	     RADEON_CP_VC_FRMT_W0|		\
+	     RADEON_CP_VC_FRMT_PKCOLOR|		\
+	     RADEON_CP_VC_FRMT_PKSPEC|		\
+	     RADEON_CP_VC_FRMT_ST0|		\
+	     RADEON_CP_VC_FRMT_Q0|		\
+	     RADEON_CP_VC_FRMT_ST1|		\
+	     RADEON_CP_VC_FRMT_Q1|		\
+	     RADEON_CP_VC_FRMT_N0)
+#define TAG(x) x##_w_rgpa_spec_stq_stq_n
+#include "radeon_maos_vbtmp.h"
+
+
+
+
+
+/***********************************************************************
+ *                         Initialization 
+ ***********************************************************************/
+
+
+static void init_tcl_verts( void )
+{
+   init_rgba();
+   init_n();
+   init_rgba_n();
+   init_rgba_st();
+   init_st_n();
+   init_rgba_st_st();
+   init_rgba_st_n();
+   init_rgba_spec_st_st();
+   init_st_st_n();
+   init_rgpa_spec_st_st_n();
+   init_rgba_stq();
+   init_rgba_stq_stq();
+   init_w_rgpa_spec_stq_stq_n();
+}
+
+
+void radeonEmitArrays( GLcontext *ctx, GLuint inputs )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+   GLuint req = 0;
+   GLuint vtx = (rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] &
+		 ~(RADEON_TCL_VTX_Q0|RADEON_TCL_VTX_Q1));
+   int i;
+   static int firsttime = 1;
+
+   if (firsttime) {
+      init_tcl_verts();
+      firsttime = 0;
+   }
+		     
+   if (1) {
+      req |= RADEON_CP_VC_FRMT_Z;
+      if (VB->ObjPtr->size == 4) {
+	 req |= RADEON_CP_VC_FRMT_W0;
+      }
+   }
+
+   if (inputs & VERT_BIT_NORMAL) {
+      req |= RADEON_CP_VC_FRMT_N0;
+   }
+   
+   if (inputs & VERT_BIT_COLOR0) {
+      req |= RADEON_CP_VC_FRMT_PKCOLOR;
+   }
+
+   if (inputs & VERT_BIT_COLOR1) {
+      req |= RADEON_CP_VC_FRMT_PKSPEC;
+   }
+
+   if (inputs & VERT_BIT_TEX0) {
+      req |= RADEON_CP_VC_FRMT_ST0;
+
+      if (VB->TexCoordPtr[0]->size == 4) {
+	 req |= RADEON_CP_VC_FRMT_Q0;
+	 vtx |= RADEON_TCL_VTX_Q0;
+      }
+   }
+
+   if (inputs & VERT_BIT_TEX1) {
+      req |= RADEON_CP_VC_FRMT_ST1;
+
+      if (VB->TexCoordPtr[1]->size == 4) {
+	 req |= RADEON_CP_VC_FRMT_Q1;
+	 vtx |= RADEON_TCL_VTX_Q1;
+      }
+   }
+
+   if (vtx != rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT]) {
+      RADEON_STATECHANGE( rmesa, tcl );
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] = vtx;
+   }
+
+   for (i = 0 ; i < RADEON_TCL_MAX_SETUP ; i++) 
+      if ((setup_tab[i].vertex_format & req) == req) 
+	 break;
+
+   if (rmesa->tcl.vertex_format == setup_tab[i].vertex_format &&
+       rmesa->tcl.indexed_verts.buf)
+      return;
+
+   if (rmesa->tcl.indexed_verts.buf)
+      radeonReleaseArrays( ctx, ~0 );
+
+   radeonAllocDmaRegionVerts( rmesa, 
+			      &rmesa->tcl.indexed_verts, 
+			      VB->Count,
+			      setup_tab[i].vertex_size * 4, 
+			      4);
+
+   setup_tab[i].emit( ctx, 0, VB->Count, 
+		      rmesa->tcl.indexed_verts.address + 
+		      rmesa->tcl.indexed_verts.start );
+
+   rmesa->tcl.vertex_format = setup_tab[i].vertex_format;
+   rmesa->tcl.indexed_verts.aos_start = GET_START( &rmesa->tcl.indexed_verts );
+   rmesa->tcl.indexed_verts.aos_size = setup_tab[i].vertex_size;
+   rmesa->tcl.indexed_verts.aos_stride = setup_tab[i].vertex_size;
+
+   rmesa->tcl.aos_components[0] = &rmesa->tcl.indexed_verts;
+   rmesa->tcl.nr_aos_components = 1;
+}
+
+
+
+void radeonReleaseArrays( GLcontext *ctx, GLuint newinputs )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+
+   if (RADEON_DEBUG & DEBUG_VERTS) 
+      _tnl_print_vert_flags( __FUNCTION__, newinputs );
+
+   if (newinputs) 
+     radeonReleaseDmaRegion( rmesa, &rmesa->tcl.indexed_verts, __FUNCTION__ );
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_sanity.c b/src/mesa/drivers/dri/radeon/radeon_sanity.c
new file mode 100644
index 00000000000..e3b37bf3de9
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_sanity.c
@@ -0,0 +1,1043 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 2002 ATI Technologies Inc., Ontario, Canada, and
+                     Tungsten Graphics Inc, Cedar Park, TX.
+
+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
+ATI, TUNGSTEN GRAPHICS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell <[email protected]>
+ *
+ */
+#include <errno.h>
+
+#include "glheader.h"
+
+#include "radeon_context.h"
+#include "radeon_ioctl.h"
+#include "radeon_sanity.h"
+
+/* Set this '1' to get more verbiage.
+ */
+#define MORE_VERBOSE 1
+
+#if MORE_VERBOSE
+#define VERBOSE (RADEON_DEBUG & DEBUG_VERBOSE)
+#define NORMAL  (1)
+#else
+#define VERBOSE 0
+#define NORMAL  (RADEON_DEBUG & DEBUG_VERBOSE)
+#endif
+
+
+/* New (1.3) state mechanism.  3 commands (packet, scalar, vector) in
+ * 1.3 cmdbuffers allow all previous state to be updated as well as
+ * the tcl scalar and vector areas.  
+ */
+static struct { 
+   int start; 
+   int len; 
+   const char *name;
+} packet[RADEON_MAX_STATE_PACKETS] = {
+   { RADEON_PP_MISC,7,"RADEON_PP_MISC" },
+   { RADEON_PP_CNTL,3,"RADEON_PP_CNTL" },
+   { RADEON_RB3D_COLORPITCH,1,"RADEON_RB3D_COLORPITCH" },
+   { RADEON_RE_LINE_PATTERN,2,"RADEON_RE_LINE_PATTERN" },
+   { RADEON_SE_LINE_WIDTH,1,"RADEON_SE_LINE_WIDTH" },
+   { RADEON_PP_LUM_MATRIX,1,"RADEON_PP_LUM_MATRIX" },
+   { RADEON_PP_ROT_MATRIX_0,2,"RADEON_PP_ROT_MATRIX_0" },
+   { RADEON_RB3D_STENCILREFMASK,3,"RADEON_RB3D_STENCILREFMASK" },
+   { RADEON_SE_VPORT_XSCALE,6,"RADEON_SE_VPORT_XSCALE" },
+   { RADEON_SE_CNTL,2,"RADEON_SE_CNTL" },
+   { RADEON_SE_CNTL_STATUS,1,"RADEON_SE_CNTL_STATUS" },
+   { RADEON_RE_MISC,1,"RADEON_RE_MISC" },
+   { RADEON_PP_TXFILTER_0,6,"RADEON_PP_TXFILTER_0" },
+   { RADEON_PP_BORDER_COLOR_0,1,"RADEON_PP_BORDER_COLOR_0" },
+   { RADEON_PP_TXFILTER_1,6,"RADEON_PP_TXFILTER_1" },
+   { RADEON_PP_BORDER_COLOR_1,1,"RADEON_PP_BORDER_COLOR_1" },
+   { RADEON_PP_TXFILTER_2,6,"RADEON_PP_TXFILTER_2" },
+   { RADEON_PP_BORDER_COLOR_2,1,"RADEON_PP_BORDER_COLOR_2" },
+   { RADEON_SE_ZBIAS_FACTOR,2,"RADEON_SE_ZBIAS_FACTOR" },
+   { RADEON_SE_TCL_OUTPUT_VTX_FMT,11,"RADEON_SE_TCL_OUTPUT_VTX_FMT" },
+   { RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED,17,"RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED" },
+	{ 0, 4, "R200_PP_TXCBLEND_0" },
+	{ 0, 4, "R200_PP_TXCBLEND_1" },
+	{ 0, 4, "R200_PP_TXCBLEND_2" },
+	{ 0, 4, "R200_PP_TXCBLEND_3" },
+	{ 0, 4, "R200_PP_TXCBLEND_4" },
+	{ 0, 4, "R200_PP_TXCBLEND_5" },
+	{ 0, 4, "R200_PP_TXCBLEND_6" },
+	{ 0, 4, "R200_PP_TXCBLEND_7" },
+	{ 0, 6, "R200_SE_TCL_LIGHT_MODEL_CTL_0" },
+	{ 0, 6, "R200_PP_TFACTOR_0" },
+	{ 0, 4, "R200_SE_VTX_FMT_0" },
+	{ 0, 1, "R200_SE_VAP_CNTL" },
+	{ 0, 5, "R200_SE_TCL_MATRIX_SEL_0" },
+	{ 0, 5, "R200_SE_TCL_TEX_PROC_CTL_2" },
+	{ 0, 1, "R200_SE_TCL_UCP_VERT_BLEND_CTL" },
+	{ 0, 6, "R200_PP_TXFILTER_0" },
+	{ 0, 6, "R200_PP_TXFILTER_1" },
+	{ 0, 6, "R200_PP_TXFILTER_2" },
+	{ 0, 6, "R200_PP_TXFILTER_3" },
+	{ 0, 6, "R200_PP_TXFILTER_4" },
+	{ 0, 6, "R200_PP_TXFILTER_5" },
+	{ 0, 1, "R200_PP_TXOFFSET_0" },
+	{ 0, 1, "R200_PP_TXOFFSET_1" },
+	{ 0, 1, "R200_PP_TXOFFSET_2" },
+	{ 0, 1, "R200_PP_TXOFFSET_3" },
+	{ 0, 1, "R200_PP_TXOFFSET_4" },
+	{ 0, 1, "R200_PP_TXOFFSET_5" },
+	{ 0, 1, "R200_SE_VTE_CNTL" },
+	{ 0, 1, "R200_SE_TCL_OUTPUT_VTX_COMP_SEL" },
+	{ 0, 1, "R200_PP_TAM_DEBUG3" },
+	{ 0, 1, "R200_PP_CNTL_X" }, 
+	{ 0, 1, "R200_RB3D_DEPTHXY_OFFSET" }, 
+	{ 0, 1, "R200_RE_AUX_SCISSOR_CNTL" }, 
+	{ 0, 2, "R200_RE_SCISSOR_TL_0" }, 
+	{ 0, 2, "R200_RE_SCISSOR_TL_1" }, 
+	{ 0, 2, "R200_RE_SCISSOR_TL_2" }, 
+	{ 0, 1, "R200_SE_VAP_CNTL_STATUS" }, 
+	{ 0, 1, "R200_SE_VTX_STATE_CNTL" }, 
+	{ 0, 1, "R200_RE_POINTSIZE" }, 
+	{ 0, 4, "R200_SE_TCL_INPUT_VTX_VECTOR_ADDR_0" },
+	{ 0, 1, "R200_PP_CUBIC_FACES_0" }, /* 61 */
+	{ 0, 5, "R200_PP_CUBIC_OFFSET_F1_0" }, /* 62 */
+	{ 0, 1, "R200_PP_CUBIC_FACES_1" },
+	{ 0, 5, "R200_PP_CUBIC_OFFSET_F1_1" },
+	{ 0, 1, "R200_PP_CUBIC_FACES_2" },
+	{ 0, 5, "R200_PP_CUBIC_OFFSET_F1_2" },
+	{ 0, 1, "R200_PP_CUBIC_FACES_3" },
+	{ 0, 5, "R200_PP_CUBIC_OFFSET_F1_3" },
+	{ 0, 1, "R200_PP_CUBIC_FACES_4" },
+	{ 0, 5, "R200_PP_CUBIC_OFFSET_F1_4" },
+	{ 0, 1, "R200_PP_CUBIC_FACES_5" },
+	{ 0, 5, "R200_PP_CUBIC_OFFSET_F1_5" },
+   { RADEON_PP_TEX_SIZE_0, 2, "RADEON_PP_TEX_SIZE_0" },
+   { RADEON_PP_TEX_SIZE_1, 2, "RADEON_PP_TEX_SIZE_1" },
+   { RADEON_PP_TEX_SIZE_2, 2, "RADEON_PP_TEX_SIZE_1" },
+};
+
+struct reg_names {
+   int idx;
+   const char *name;
+};
+
+static struct reg_names reg_names[] = {
+   { RADEON_PP_MISC, "RADEON_PP_MISC" },
+   { RADEON_PP_FOG_COLOR, "RADEON_PP_FOG_COLOR" },
+   { RADEON_RE_SOLID_COLOR, "RADEON_RE_SOLID_COLOR" },
+   { RADEON_RB3D_BLENDCNTL, "RADEON_RB3D_BLENDCNTL" },
+   { RADEON_RB3D_DEPTHOFFSET, "RADEON_RB3D_DEPTHOFFSET" },
+   { RADEON_RB3D_DEPTHPITCH, "RADEON_RB3D_DEPTHPITCH" },
+   { RADEON_RB3D_ZSTENCILCNTL, "RADEON_RB3D_ZSTENCILCNTL" },
+   { RADEON_PP_CNTL, "RADEON_PP_CNTL" },
+   { RADEON_RB3D_CNTL, "RADEON_RB3D_CNTL" },
+   { RADEON_RB3D_COLOROFFSET, "RADEON_RB3D_COLOROFFSET" },
+   { RADEON_RB3D_COLORPITCH, "RADEON_RB3D_COLORPITCH" },
+   { RADEON_SE_CNTL, "RADEON_SE_CNTL" },
+   { RADEON_SE_COORD_FMT, "RADEON_SE_COORDFMT" },
+   { RADEON_SE_CNTL_STATUS, "RADEON_SE_CNTL_STATUS" },
+   { RADEON_RE_LINE_PATTERN, "RADEON_RE_LINE_PATTERN" },
+   { RADEON_RE_LINE_STATE, "RADEON_RE_LINE_STATE" },
+   { RADEON_SE_LINE_WIDTH, "RADEON_SE_LINE_WIDTH" },
+   { RADEON_RB3D_STENCILREFMASK, "RADEON_RB3D_STENCILREFMASK" },
+   { RADEON_RB3D_ROPCNTL, "RADEON_RB3D_ROPCNTL" },
+   { RADEON_RB3D_PLANEMASK, "RADEON_RB3D_PLANEMASK" },
+   { RADEON_SE_VPORT_XSCALE, "RADEON_SE_VPORT_XSCALE" },
+   { RADEON_SE_VPORT_XOFFSET, "RADEON_SE_VPORT_XOFFSET" },
+   { RADEON_SE_VPORT_YSCALE, "RADEON_SE_VPORT_YSCALE" },
+   { RADEON_SE_VPORT_YOFFSET, "RADEON_SE_VPORT_YOFFSET" },
+   { RADEON_SE_VPORT_ZSCALE, "RADEON_SE_VPORT_ZSCALE" },
+   { RADEON_SE_VPORT_ZOFFSET, "RADEON_SE_VPORT_ZOFFSET" },
+   { RADEON_RE_MISC, "RADEON_RE_MISC" },
+   { RADEON_PP_TXFILTER_0, "RADEON_PP_TXFILTER_0" },
+   { RADEON_PP_TXFILTER_1, "RADEON_PP_TXFILTER_1" },
+   { RADEON_PP_TXFILTER_2, "RADEON_PP_TXFILTER_2" },
+   { RADEON_PP_TXFORMAT_0, "RADEON_PP_TXFORMAT_0" },
+   { RADEON_PP_TXFORMAT_1, "RADEON_PP_TXFORMAT_1" },
+   { RADEON_PP_TXFORMAT_2, "RADEON_PP_TXFORMAT_3" },
+   { RADEON_PP_TXOFFSET_0, "RADEON_PP_TXOFFSET_0" },
+   { RADEON_PP_TXOFFSET_1, "RADEON_PP_TXOFFSET_1" },
+   { RADEON_PP_TXOFFSET_2, "RADEON_PP_TXOFFSET_3" },
+   { RADEON_PP_TXCBLEND_0, "RADEON_PP_TXCBLEND_0" },
+   { RADEON_PP_TXCBLEND_1, "RADEON_PP_TXCBLEND_1" },
+   { RADEON_PP_TXCBLEND_2, "RADEON_PP_TXCBLEND_3" },
+   { RADEON_PP_TXABLEND_0, "RADEON_PP_TXABLEND_0" },
+   { RADEON_PP_TXABLEND_1, "RADEON_PP_TXABLEND_1" },
+   { RADEON_PP_TXABLEND_2, "RADEON_PP_TXABLEND_3" },
+   { RADEON_PP_TFACTOR_0, "RADEON_PP_TFACTOR_0" },
+   { RADEON_PP_TFACTOR_1, "RADEON_PP_TFACTOR_1" },
+   { RADEON_PP_TFACTOR_2, "RADEON_PP_TFACTOR_3" },
+   { RADEON_PP_BORDER_COLOR_0, "RADEON_PP_BORDER_COLOR_0" },
+   { RADEON_PP_BORDER_COLOR_1, "RADEON_PP_BORDER_COLOR_1" },
+   { RADEON_PP_BORDER_COLOR_2, "RADEON_PP_BORDER_COLOR_3" },
+   { RADEON_SE_ZBIAS_FACTOR, "RADEON_SE_ZBIAS_FACTOR" },
+   { RADEON_SE_ZBIAS_CONSTANT, "RADEON_SE_ZBIAS_CONSTANT" },
+   { RADEON_SE_TCL_OUTPUT_VTX_FMT, "RADEON_SE_TCL_OUTPUT_VTXFMT" },
+   { RADEON_SE_TCL_OUTPUT_VTX_SEL, "RADEON_SE_TCL_OUTPUT_VTXSEL" },
+   { RADEON_SE_TCL_MATRIX_SELECT_0, "RADEON_SE_TCL_MATRIX_SELECT_0" },
+   { RADEON_SE_TCL_MATRIX_SELECT_1, "RADEON_SE_TCL_MATRIX_SELECT_1" },
+   { RADEON_SE_TCL_UCP_VERT_BLEND_CTL, "RADEON_SE_TCL_UCP_VERT_BLEND_CTL" },
+   { RADEON_SE_TCL_TEXTURE_PROC_CTL, "RADEON_SE_TCL_TEXTURE_PROC_CTL" },
+   { RADEON_SE_TCL_LIGHT_MODEL_CTL, "RADEON_SE_TCL_LIGHT_MODEL_CTL" },
+   { RADEON_SE_TCL_PER_LIGHT_CTL_0, "RADEON_SE_TCL_PER_LIGHT_CTL_0" },
+   { RADEON_SE_TCL_PER_LIGHT_CTL_1, "RADEON_SE_TCL_PER_LIGHT_CTL_1" },
+   { RADEON_SE_TCL_PER_LIGHT_CTL_2, "RADEON_SE_TCL_PER_LIGHT_CTL_2" },
+   { RADEON_SE_TCL_PER_LIGHT_CTL_3, "RADEON_SE_TCL_PER_LIGHT_CTL_3" },
+   { RADEON_SE_TCL_MATERIAL_EMMISSIVE_RED, "RADEON_SE_TCL_EMMISSIVE_RED" },
+   { RADEON_SE_TCL_MATERIAL_EMMISSIVE_GREEN, "RADEON_SE_TCL_EMMISSIVE_GREEN" },
+   { RADEON_SE_TCL_MATERIAL_EMMISSIVE_BLUE, "RADEON_SE_TCL_EMMISSIVE_BLUE" },
+   { RADEON_SE_TCL_MATERIAL_EMMISSIVE_ALPHA, "RADEON_SE_TCL_EMMISSIVE_ALPHA" },
+   { RADEON_SE_TCL_MATERIAL_AMBIENT_RED, "RADEON_SE_TCL_AMBIENT_RED" },
+   { RADEON_SE_TCL_MATERIAL_AMBIENT_GREEN, "RADEON_SE_TCL_AMBIENT_GREEN" },
+   { RADEON_SE_TCL_MATERIAL_AMBIENT_BLUE, "RADEON_SE_TCL_AMBIENT_BLUE" },
+   { RADEON_SE_TCL_MATERIAL_AMBIENT_ALPHA, "RADEON_SE_TCL_AMBIENT_ALPHA" },
+   { RADEON_SE_TCL_MATERIAL_DIFFUSE_RED, "RADEON_SE_TCL_DIFFUSE_RED" },
+   { RADEON_SE_TCL_MATERIAL_DIFFUSE_GREEN, "RADEON_SE_TCL_DIFFUSE_GREEN" },
+   { RADEON_SE_TCL_MATERIAL_DIFFUSE_BLUE, "RADEON_SE_TCL_DIFFUSE_BLUE" },
+   { RADEON_SE_TCL_MATERIAL_DIFFUSE_ALPHA, "RADEON_SE_TCL_DIFFUSE_ALPHA" },
+   { RADEON_SE_TCL_MATERIAL_SPECULAR_RED, "RADEON_SE_TCL_SPECULAR_RED" },
+   { RADEON_SE_TCL_MATERIAL_SPECULAR_GREEN, "RADEON_SE_TCL_SPECULAR_GREEN" },
+   { RADEON_SE_TCL_MATERIAL_SPECULAR_BLUE, "RADEON_SE_TCL_SPECULAR_BLUE" },
+   { RADEON_SE_TCL_MATERIAL_SPECULAR_ALPHA, "RADEON_SE_TCL_SPECULAR_ALPHA" },
+   { RADEON_SE_TCL_SHININESS, "RADEON_SE_TCL_SHININESS" },
+   { RADEON_SE_COORD_FMT, "RADEON_SE_COORD_FMT" },
+   { RADEON_PP_TEX_SIZE_0, "RADEON_PP_TEX_SIZE_0" },
+   { RADEON_PP_TEX_SIZE_1, "RADEON_PP_TEX_SIZE_1" },
+   { RADEON_PP_TEX_SIZE_2, "RADEON_PP_TEX_SIZE_2" },
+   { RADEON_PP_TEX_SIZE_0+4, "RADEON_PP_TEX_PITCH_0" },
+   { RADEON_PP_TEX_SIZE_1+4, "RADEON_PP_TEX_PITCH_1" },
+   { RADEON_PP_TEX_SIZE_2+4, "RADEON_PP_TEX_PITCH_2" },
+};
+
+static struct reg_names scalar_names[] = {
+   { RADEON_SS_LIGHT_DCD_ADDR, "LIGHT_DCD" },
+   { RADEON_SS_LIGHT_SPOT_EXPONENT_ADDR, "LIGHT_SPOT_EXPONENT" },
+   { RADEON_SS_LIGHT_SPOT_CUTOFF_ADDR, "LIGHT_SPOT_CUTOFF" },
+   { RADEON_SS_LIGHT_SPECULAR_THRESH_ADDR, "LIGHT_SPECULAR_THRESH" },
+   { RADEON_SS_LIGHT_RANGE_CUTOFF_ADDR, "LIGHT_RANGE_CUTOFF" },
+   { RADEON_SS_VERT_GUARD_CLIP_ADJ_ADDR, "VERT_GUARD_CLIP" },
+   { RADEON_SS_VERT_GUARD_DISCARD_ADJ_ADDR, "VERT_GUARD_DISCARD" },
+   { RADEON_SS_HORZ_GUARD_CLIP_ADJ_ADDR, "HORZ_GUARD_CLIP" },
+   { RADEON_SS_HORZ_GUARD_DISCARD_ADJ_ADDR, "HORZ_GUARD_DISCARD" },
+   { RADEON_SS_SHININESS, "SHININESS" },
+   { 1000, "" },
+};
+
+/* Puff these out to make them look like normal (dword) registers.
+ */
+static struct reg_names vector_names[] = {
+   { RADEON_VS_MATRIX_0_ADDR * 4, "MATRIX_0" },
+   { RADEON_VS_MATRIX_1_ADDR * 4, "MATRIX_1" },
+   { RADEON_VS_MATRIX_2_ADDR * 4, "MATRIX_2" },
+   { RADEON_VS_MATRIX_3_ADDR * 4, "MATRIX_3" },
+   { RADEON_VS_MATRIX_4_ADDR * 4, "MATRIX_4" },
+   { RADEON_VS_MATRIX_5_ADDR * 4, "MATRIX_5" },
+   { RADEON_VS_MATRIX_6_ADDR * 4, "MATRIX_6" },
+   { RADEON_VS_MATRIX_7_ADDR * 4, "MATRIX_7" },
+   { RADEON_VS_MATRIX_8_ADDR * 4, "MATRIX_8" },
+   { RADEON_VS_MATRIX_9_ADDR * 4, "MATRIX_9" },
+   { RADEON_VS_MATRIX_10_ADDR * 4, "MATRIX_10" },
+   { RADEON_VS_MATRIX_11_ADDR * 4, "MATRIX_11" },
+   { RADEON_VS_MATRIX_12_ADDR * 4, "MATRIX_12" },
+   { RADEON_VS_MATRIX_13_ADDR * 4, "MATRIX_13" },
+   { RADEON_VS_MATRIX_14_ADDR * 4, "MATRIX_14" },
+   { RADEON_VS_MATRIX_15_ADDR * 4, "MATRIX_15" },
+   { RADEON_VS_LIGHT_AMBIENT_ADDR * 4, "LIGHT_AMBIENT" },
+   { RADEON_VS_LIGHT_DIFFUSE_ADDR * 4, "LIGHT_DIFFUSE" },
+   { RADEON_VS_LIGHT_SPECULAR_ADDR * 4, "LIGHT_SPECULAR" },
+   { RADEON_VS_LIGHT_DIRPOS_ADDR * 4, "LIGHT_DIRPOS" },
+   { RADEON_VS_LIGHT_HWVSPOT_ADDR * 4, "LIGHT_HWVSPOT" },
+   { RADEON_VS_LIGHT_ATTENUATION_ADDR * 4, "LIGHT_ATTENUATION" },
+   { RADEON_VS_MATRIX_EYE2CLIP_ADDR * 4, "MATRIX_EYE2CLIP" },
+   { RADEON_VS_UCP_ADDR * 4, "UCP" },
+   { RADEON_VS_GLOBAL_AMBIENT_ADDR * 4, "GLOBAL_AMBIENT" },
+   { RADEON_VS_FOG_PARAM_ADDR * 4, "FOG_PARAM" },
+   { RADEON_VS_EYE_VECTOR_ADDR * 4, "EYE_VECTOR" },
+   { 1000, "" },
+};
+
+union fi { float f; int i; };
+
+#define ISVEC   1
+#define ISFLOAT 2
+#define TOUCHED 4
+
+struct reg {
+   int idx; 
+   struct reg_names *closest;
+   int flags;
+   union fi current;
+   union fi *values;
+   int nvalues;
+   int nalloc;
+   float vmin, vmax;
+};
+
+
+static struct reg regs[Elements(reg_names)+1];
+static struct reg scalars[512+1];
+static struct reg vectors[512*4+1];
+
+static int total, total_changed, bufs;
+
+static void init_regs( void )
+{
+   struct reg_names *tmp;
+   int i;
+
+   for (i = 0 ; i < Elements(regs) ; i++) {
+      regs[i].idx = reg_names[i].idx;
+      regs[i].closest = &reg_names[i];
+      regs[i].flags = 0;
+   }
+
+   for (i = 0, tmp = scalar_names ; i < Elements(scalars) ; i++) {
+      if (tmp[1].idx == i) tmp++;
+      scalars[i].idx = i;
+      scalars[i].closest = tmp;
+      scalars[i].flags = ISFLOAT;
+   }
+
+   for (i = 0, tmp = vector_names ; i < Elements(vectors) ; i++) {
+      if (tmp[1].idx*4 == i) tmp++;
+      vectors[i].idx = i;
+      vectors[i].closest = tmp;
+      vectors[i].flags = ISFLOAT|ISVEC;
+   }
+
+   regs[Elements(regs)-1].idx = -1;
+   scalars[Elements(scalars)-1].idx = -1;
+   vectors[Elements(vectors)-1].idx = -1;
+}
+
+static int find_or_add_value( struct reg *reg, int val )
+{
+   int j;
+
+   for ( j = 0 ; j < reg->nvalues ; j++)
+      if ( val == reg->values[j].i )
+	 return 1;
+
+   if (j == reg->nalloc) {
+      reg->nalloc += 5;
+      reg->nalloc *= 2;
+      reg->values = (union fi *) realloc( reg->values, 
+					  reg->nalloc * sizeof(union fi) );
+   }
+
+   reg->values[reg->nvalues++].i = val;
+   return 0;
+}
+
+static struct reg *lookup_reg( struct reg *tab, int reg )
+{
+   int i;
+
+   for (i = 0 ; tab[i].idx != -1 ; i++) {
+      if (tab[i].idx == reg)
+	 return &tab[i];
+   }
+
+   fprintf(stderr, "*** unknown reg 0x%x\n", reg);
+   return 0;
+}
+
+
+static const char *get_reg_name( struct reg *reg )
+{
+   static char tmp[80];
+
+   if (reg->idx == reg->closest->idx) 
+      return reg->closest->name;
+
+   
+   if (reg->flags & ISVEC) {
+      if (reg->idx/4 != reg->closest->idx)
+	 sprintf(tmp, "%s+%d[%d]", 
+		 reg->closest->name, 
+		 (reg->idx/4) - reg->closest->idx,
+		 reg->idx%4);
+      else
+	 sprintf(tmp, "%s[%d]", reg->closest->name, reg->idx%4);
+   }
+   else {
+      if (reg->idx != reg->closest->idx)
+	 sprintf(tmp, "%s+%d", reg->closest->name, reg->idx - reg->closest->idx);
+      else
+	 sprintf(tmp, "%s", reg->closest->name);
+   }
+
+   return tmp;
+}
+
+static int print_int_reg_assignment( struct reg *reg, int data )
+{
+   int changed = (reg->current.i != data);
+   int ever_seen = find_or_add_value( reg, data );
+   
+   if (VERBOSE || (NORMAL && (changed || !ever_seen)))
+       fprintf(stderr, "   %s <-- 0x%x", get_reg_name(reg), data);
+       
+   if (NORMAL) {
+      if (!ever_seen) 
+	 fprintf(stderr, " *** BRAND NEW VALUE");
+      else if (changed) 
+	 fprintf(stderr, " *** CHANGED"); 
+   }
+   
+   reg->current.i = data;
+
+   if (VERBOSE || (NORMAL && (changed || !ever_seen)))
+      fprintf(stderr, "\n");
+
+   return changed;
+}
+
+
+static int print_float_reg_assignment( struct reg *reg, float data )
+{
+   int changed = (reg->current.f != data);
+   int newmin = (data < reg->vmin);
+   int newmax = (data > reg->vmax);
+
+   if (VERBOSE || (NORMAL && (newmin || newmax || changed)))
+      fprintf(stderr, "   %s <-- %.3f", get_reg_name(reg), data);
+
+   if (NORMAL) {
+      if (newmin) {
+	 fprintf(stderr, " *** NEW MIN (prev %.3f)", reg->vmin);
+	 reg->vmin = data;
+      }
+      else if (newmax) {
+	 fprintf(stderr, " *** NEW MAX (prev %.3f)", reg->vmax);
+	 reg->vmax = data;
+      }
+      else if (changed) {
+	 fprintf(stderr, " *** CHANGED");
+      }
+   }
+
+   reg->current.f = data;
+
+   if (VERBOSE || (NORMAL && (newmin || newmax || changed)))
+      fprintf(stderr, "\n");
+
+   return changed;
+}
+
+static int print_reg_assignment( struct reg *reg, int data )
+{
+   reg->flags |= TOUCHED;
+   if (reg->flags & ISFLOAT)
+      return print_float_reg_assignment( reg, *(float *)&data );
+   else
+      return print_int_reg_assignment( reg, data );
+}
+
+static void print_reg( struct reg *reg )
+{
+   if (reg->flags & TOUCHED) {
+      if (reg->flags & ISFLOAT) {
+	 fprintf(stderr, "   %s == %f\n", get_reg_name(reg), reg->current.f);
+      } else {
+	 fprintf(stderr, "   %s == 0x%x\n", get_reg_name(reg), reg->current.i);
+      }
+   }
+}
+
+
+static void dump_state( void )
+{
+   int i;
+
+   for (i = 0 ; i < Elements(regs) ; i++) 
+      print_reg( &regs[i] );
+
+   for (i = 0 ; i < Elements(scalars) ; i++) 
+      print_reg( &scalars[i] );
+
+   for (i = 0 ; i < Elements(vectors) ; i++) 
+      print_reg( &vectors[i] );
+}
+
+
+
+static int radeon_emit_packets( 
+   drmRadeonCmdHeader header,
+   drmRadeonCmdBuffer *cmdbuf )
+{
+   int id = (int)header.packet.packet_id;
+   int sz = packet[id].len;
+   int *data = (int *)cmdbuf->buf;
+   int i;
+   
+   if (sz * sizeof(int) > cmdbuf->bufsz) {
+      fprintf(stderr, "Packet overflows cmdbuf\n");      
+      return -EINVAL;
+   }
+
+   if (!packet[id].name) {
+      fprintf(stderr, "*** Unknown packet 0 nr %d\n", id );
+      return -EINVAL;
+   }
+
+   
+   if (VERBOSE) 
+      fprintf(stderr, "Packet 0 reg %s nr %d\n", packet[id].name, sz );
+
+   for ( i = 0 ; i < sz ; i++) {
+      struct reg *reg = lookup_reg( regs, packet[id].start + i*4 );
+      if (print_reg_assignment( reg, data[i] ))
+	 total_changed++;
+      total++;
+   }
+
+   cmdbuf->buf += sz * sizeof(int);
+   cmdbuf->bufsz -= sz * sizeof(int);
+   return 0;
+}
+
+
+static int radeon_emit_scalars( 
+   drmRadeonCmdHeader header,
+   drmRadeonCmdBuffer *cmdbuf )
+{
+   int sz = header.scalars.count;
+   int *data = (int *)cmdbuf->buf;
+   int start = header.scalars.offset;
+   int stride = header.scalars.stride;
+   int i;
+
+   if (VERBOSE)
+      fprintf(stderr, "emit scalars, start %d stride %d nr %d (end %d)\n",
+	      start, stride, sz, start + stride * sz);
+
+
+   for (i = 0 ; i < sz ; i++, start += stride) {
+      struct reg *reg = lookup_reg( scalars, start );
+      if (print_reg_assignment( reg, data[i] ))
+	 total_changed++;
+      total++;
+   }
+	 
+   cmdbuf->buf += sz * sizeof(int);
+   cmdbuf->bufsz -= sz * sizeof(int);
+   return 0;
+}
+
+
+static int radeon_emit_scalars2( 
+   drmRadeonCmdHeader header,
+   drmRadeonCmdBuffer *cmdbuf )
+{
+   int sz = header.scalars.count;
+   int *data = (int *)cmdbuf->buf;
+   int start = header.scalars.offset + 0x100;
+   int stride = header.scalars.stride;
+   int i;
+
+   if (VERBOSE)
+      fprintf(stderr, "emit scalars2, start %d stride %d nr %d (end %d)\n",
+	      start, stride, sz, start + stride * sz);
+
+   if (start + stride * sz > 257) {
+      fprintf(stderr, "emit scalars OVERFLOW %d/%d/%d\n", start, stride, sz);
+      return -1;
+   }
+
+   for (i = 0 ; i < sz ; i++, start += stride) {
+      struct reg *reg = lookup_reg( scalars, start );
+      if (print_reg_assignment( reg, data[i] ))
+	 total_changed++;
+      total++;
+   }
+	 
+   cmdbuf->buf += sz * sizeof(int);
+   cmdbuf->bufsz -= sz * sizeof(int);
+   return 0;
+}
+
+/* Check: inf/nan/extreme-size?
+ * Check: table start, end, nr, etc.
+ */
+static int radeon_emit_vectors( 
+   drmRadeonCmdHeader header,
+   drmRadeonCmdBuffer *cmdbuf )
+{
+   int sz = header.vectors.count;
+   int *data = (int *)cmdbuf->buf;
+   int start = header.vectors.offset;
+   int stride = header.vectors.stride;
+   int i,j;
+
+   if (VERBOSE)
+      fprintf(stderr, "emit vectors, start %d stride %d nr %d (end %d) (0x%x)\n",
+	      start, stride, sz, start + stride * sz, header.i);
+
+/*    if (start + stride * (sz/4) > 128) { */
+/*       fprintf(stderr, "emit vectors OVERFLOW %d/%d/%d\n", start, stride, sz); */
+/*       return -1; */
+/*    } */
+
+   for (i = 0 ; i < sz ;  start += stride) {
+      int changed = 0;
+      for (j = 0 ; j < 4 ; i++,j++) {
+	 struct reg *reg = lookup_reg( vectors, start*4+j );
+	 if (print_reg_assignment( reg, data[i] ))
+	    changed = 1;
+      }
+      if (changed)
+	 total_changed += 4;
+      total += 4;
+   }
+	 
+
+   cmdbuf->buf += sz * sizeof(int);
+   cmdbuf->bufsz -= sz * sizeof(int);
+   return 0;
+}
+
+
+static int print_vertex_format( int vfmt )
+{
+   if (NORMAL) {
+      fprintf(stderr, "   %s(%x): %s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s%s",
+	      "vertex format",
+	      vfmt,
+	      "xy,",
+	      (vfmt & RADEON_CP_VC_FRMT_Z) ? "z," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_W0) ? "w0," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_FPCOLOR) ? "fpcolor," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_FPALPHA) ? "fpalpha," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_PKCOLOR) ? "pkcolor," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_FPSPEC) ? "fpspec," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_FPFOG) ? "fpfog," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_PKSPEC) ? "pkspec," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_ST0) ? "st0," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_ST1) ? "st1," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_Q1) ? "q1," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_ST2) ? "st2," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_Q2) ? "q2," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_ST3) ? "st3," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_Q3) ? "q3," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_Q0) ? "q0," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_N0) ? "n0," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_XY1) ? "xy1," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_Z1) ? "z1," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_W1) ? "w1," : "",
+	      (vfmt & RADEON_CP_VC_FRMT_N1) ? "n1," : "");
+
+   
+/*       if (!find_or_add_value( &others[V_VTXFMT], vfmt )) */
+/* 	 fprintf(stderr, " *** NEW VALUE"); */
+
+      fprintf(stderr, "\n");
+   }
+
+   return 0;
+}
+
+static char *primname[0xf] = {
+   "NONE",
+   "POINTS",
+   "LINES",
+   "LINE_STRIP",
+   "TRIANGLES",
+   "TRIANGLE_FAN",
+   "TRIANGLE_STRIP",
+   "TRI_TYPE_2",
+   "RECT_LIST",
+   "3VRT_POINTS",
+   "3VRT_LINES",
+};
+
+static int print_prim_and_flags( int prim )
+{
+   int numverts;
+   
+   if (NORMAL)
+      fprintf(stderr, "   %s(%x): %s%s%s%s%s%s%s\n",
+	      "prim flags",
+	      prim,
+	      ((prim & 0x30) == RADEON_CP_VC_CNTL_PRIM_WALK_IND) ? "IND," : "",
+	      ((prim & 0x30) == RADEON_CP_VC_CNTL_PRIM_WALK_LIST) ? "LIST," : "",
+	      ((prim & 0x30) == RADEON_CP_VC_CNTL_PRIM_WALK_RING) ? "RING," : "",
+	      (prim & RADEON_CP_VC_CNTL_COLOR_ORDER_RGBA) ? "RGBA," : "BGRA, ",
+	      (prim & RADEON_CP_VC_CNTL_MAOS_ENABLE) ? "MAOS," : "",
+	      (prim & RADEON_CP_VC_CNTL_VTX_FMT_RADEON_MODE) ? "RADEON," : "",
+	      (prim & RADEON_CP_VC_CNTL_TCL_ENABLE) ? "TCL," : "");
+
+   if ((prim & 0xf) > RADEON_CP_VC_CNTL_PRIM_TYPE_3VRT_LINE_LIST) {
+      fprintf(stderr, "   *** Bad primitive: %x\n", prim & 0xf);
+      return -1;
+   }
+
+   numverts = prim>>16;
+   
+   if (NORMAL)
+      fprintf(stderr, "   prim: %s numverts %d\n", primname[prim&0xf], numverts);
+
+   switch (prim & 0xf) {
+   case RADEON_CP_VC_CNTL_PRIM_TYPE_NONE:
+   case RADEON_CP_VC_CNTL_PRIM_TYPE_POINT:
+      if (numverts < 1) {
+	 fprintf(stderr, "Bad nr verts for line %d\n", numverts);
+	 return -1;
+      }
+      break;
+   case RADEON_CP_VC_CNTL_PRIM_TYPE_LINE:
+      if ((numverts & 1) || numverts == 0) {
+	 fprintf(stderr, "Bad nr verts for line %d\n", numverts);
+	 return -1;
+      }
+      break;
+   case RADEON_CP_VC_CNTL_PRIM_TYPE_LINE_STRIP:
+      if (numverts < 2) {
+	 fprintf(stderr, "Bad nr verts for line_strip %d\n", numverts);
+	 return -1;
+      }
+      break;
+   case RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST:
+   case RADEON_CP_VC_CNTL_PRIM_TYPE_3VRT_POINT_LIST:
+   case RADEON_CP_VC_CNTL_PRIM_TYPE_3VRT_LINE_LIST:
+   case RADEON_CP_VC_CNTL_PRIM_TYPE_RECT_LIST:
+      if (numverts % 3 || numverts == 0) {
+	 fprintf(stderr, "Bad nr verts for tri %d\n", numverts);
+	 return -1;
+      }
+      break;
+   case RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_FAN:
+   case RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_STRIP:
+      if (numverts < 3) {
+	 fprintf(stderr, "Bad nr verts for strip/fan %d\n", numverts);
+	 return -1;
+      }
+      break;
+   default:
+      fprintf(stderr, "Bad primitive\n");
+      return -1;
+   }	
+   return 0;
+}
+
+/* build in knowledge about each packet type
+ */
+static int radeon_emit_packet3( drmRadeonCmdBuffer *cmdbuf )
+{
+   int cmdsz;
+   int *cmd = (int *)cmdbuf->buf;
+   int *tmp;
+   int i, stride, size, start;
+
+   cmdsz = 2 + ((cmd[0] & RADEON_CP_PACKET_COUNT_MASK) >> 16);
+
+   if ((cmd[0] & RADEON_CP_PACKET_MASK) != RADEON_CP_PACKET3 ||
+       cmdsz * 4 > cmdbuf->bufsz ||
+       cmdsz > RADEON_CP_PACKET_MAX_DWORDS) {
+      fprintf(stderr, "Bad packet\n");
+      return -EINVAL;
+   }
+
+   switch( cmd[0] & ~RADEON_CP_PACKET_COUNT_MASK ) {
+   case RADEON_CP_PACKET3_NOP:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_NOP, %d dwords\n", cmdsz);
+      break;
+   case RADEON_CP_PACKET3_NEXT_CHAR:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_NEXT_CHAR, %d dwords\n", cmdsz);
+      break;
+   case RADEON_CP_PACKET3_PLY_NEXTSCAN:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_PLY_NEXTSCAN, %d dwords\n", cmdsz);
+      break;
+   case RADEON_CP_PACKET3_SET_SCISSORS:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_SET_SCISSORS, %d dwords\n", cmdsz);
+      break;
+   case RADEON_CP_PACKET3_3D_RNDR_GEN_INDX_PRIM:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_3D_RNDR_GEN_INDX_PRIM, %d dwords\n",
+	      cmdsz);
+      break;
+   case RADEON_CP_PACKET3_LOAD_MICROCODE:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_LOAD_MICROCODE, %d dwords\n", cmdsz);
+      break;
+   case RADEON_CP_PACKET3_WAIT_FOR_IDLE:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_WAIT_FOR_IDLE, %d dwords\n", cmdsz);
+      break;
+
+   case RADEON_CP_PACKET3_3D_DRAW_VBUF:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_3D_DRAW_VBUF, %d dwords\n", cmdsz);
+      print_vertex_format(cmd[1]);
+      print_prim_and_flags(cmd[2]);
+      break;
+
+   case RADEON_CP_PACKET3_3D_DRAW_IMMD:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_3D_DRAW_IMMD, %d dwords\n", cmdsz);
+      break;
+   case RADEON_CP_PACKET3_3D_DRAW_INDX: {
+      int neltdwords;
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_3D_DRAW_INDX, %d dwords\n", cmdsz);
+      print_vertex_format(cmd[1]);
+      print_prim_and_flags(cmd[2]);
+      neltdwords = cmd[2]>>16;
+      neltdwords += neltdwords & 1;
+      neltdwords /= 2;
+      if (neltdwords + 3 != cmdsz)
+	 fprintf(stderr, "Mismatch in DRAW_INDX, %d vs cmdsz %d\n",
+		 neltdwords, cmdsz);
+      break;
+   }
+   case RADEON_CP_PACKET3_LOAD_PALETTE:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_LOAD_PALETTE, %d dwords\n", cmdsz);
+      break;
+   case RADEON_CP_PACKET3_3D_LOAD_VBPNTR:
+      if (NORMAL) {
+	 fprintf(stderr, "PACKET3_3D_LOAD_VBPNTR, %d dwords\n", cmdsz);
+	 fprintf(stderr, "   nr arrays: %d\n", cmd[1]);
+      }
+
+      if (cmd[1]/2 + cmd[1]%2 != cmdsz - 3) {
+	 fprintf(stderr, "  ****** MISMATCH %d/%d *******\n",
+		 cmd[1]/2 + cmd[1]%2 + 3, cmdsz);
+	 return -EINVAL;
+      }
+
+      if (NORMAL) {
+	 tmp = cmd+2;
+	 for (i = 0 ; i < cmd[1] ; i++) {
+	    if (i & 1) {
+	       stride = (tmp[0]>>24) & 0xff;
+	       size = (tmp[0]>>16) & 0xff;
+	       start = tmp[2];
+	       tmp += 3;
+	    }
+	    else {
+	       stride = (tmp[0]>>8) & 0xff;
+	       size = (tmp[0]) & 0xff;
+	       start = tmp[1];
+	    }
+	    fprintf(stderr, "   array %d: start 0x%x vsize %d vstride %d\n",
+		    i, start, size, stride );
+	 }
+      }
+      break;
+   case RADEON_CP_PACKET3_CNTL_PAINT:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_CNTL_PAINT, %d dwords\n", cmdsz);
+      break;
+   case RADEON_CP_PACKET3_CNTL_BITBLT:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_CNTL_BITBLT, %d dwords\n", cmdsz);
+      break;
+   case RADEON_CP_PACKET3_CNTL_SMALLTEXT:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_CNTL_SMALLTEXT, %d dwords\n", cmdsz);
+      break;
+   case RADEON_CP_PACKET3_CNTL_HOSTDATA_BLT:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_CNTL_HOSTDATA_BLT, %d dwords\n", 
+	      cmdsz);
+      break;
+   case RADEON_CP_PACKET3_CNTL_POLYLINE:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_CNTL_POLYLINE, %d dwords\n", cmdsz);
+      break;
+   case RADEON_CP_PACKET3_CNTL_POLYSCANLINES:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_CNTL_POLYSCANLINES, %d dwords\n", 
+	      cmdsz);
+      break;
+   case RADEON_CP_PACKET3_CNTL_PAINT_MULTI:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_CNTL_PAINT_MULTI, %d dwords\n", 
+	      cmdsz);
+      break;
+   case RADEON_CP_PACKET3_CNTL_BITBLT_MULTI:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_CNTL_BITBLT_MULTI, %d dwords\n", 
+	      cmdsz);
+      break;
+   case RADEON_CP_PACKET3_CNTL_TRANS_BITBLT:
+      if (NORMAL)
+	 fprintf(stderr, "PACKET3_CNTL_TRANS_BITBLT, %d dwords\n", 
+	      cmdsz);
+      break;
+   default:
+      fprintf(stderr, "UNKNOWN PACKET, %d dwords\n", cmdsz);
+      break;
+   }
+      
+   cmdbuf->buf += cmdsz * 4;
+   cmdbuf->bufsz -= cmdsz * 4;
+   return 0;
+}
+
+
+/* Check cliprects for bounds, then pass on to above:
+ */
+static int radeon_emit_packet3_cliprect( drmRadeonCmdBuffer *cmdbuf )
+{   
+   XF86DRIClipRectRec *boxes = (XF86DRIClipRectRec *)cmdbuf->boxes;
+   int i = 0;
+
+   if (VERBOSE && total_changed) {
+      dump_state();
+      total_changed = 0;
+   }
+   else fprintf(stderr, "total_changed zero\n");
+
+   if (NORMAL) {
+      do {
+	 if ( i < cmdbuf->nbox ) {
+	    fprintf(stderr, "Emit box %d/%d %d,%d %d,%d\n",
+		    i, cmdbuf->nbox,
+		    boxes[i].x1, boxes[i].y1, boxes[i].x2, boxes[i].y2);
+	 }
+      } while ( ++i < cmdbuf->nbox );
+   }
+
+   if (cmdbuf->nbox == 1)
+      cmdbuf->nbox = 0;
+
+   return radeon_emit_packet3( cmdbuf );
+}
+
+
+int radeonSanityCmdBuffer( radeonContextPtr rmesa,
+			   int nbox,
+			   XF86DRIClipRectRec *boxes )
+{
+   int idx;
+   drmRadeonCmdBuffer cmdbuf;
+   drmRadeonCmdHeader header;
+   static int inited = 0;
+
+   if (!inited) {
+      init_regs();
+      inited = 1;
+   }
+
+   cmdbuf.buf = rmesa->store.cmd_buf;
+   cmdbuf.bufsz = rmesa->store.cmd_used;
+   cmdbuf.boxes = (drmClipRect *)boxes;
+   cmdbuf.nbox = nbox;
+
+   while ( cmdbuf.bufsz >= sizeof(header) ) {
+		
+      header.i = *(int *)cmdbuf.buf;
+      cmdbuf.buf += sizeof(header);
+      cmdbuf.bufsz -= sizeof(header);
+
+      switch (header.header.cmd_type) {
+      case RADEON_CMD_PACKET: 
+	 if (radeon_emit_packets( header, &cmdbuf )) {
+	    fprintf(stderr,"radeon_emit_packets failed\n");
+	    return -EINVAL;
+	 }
+	 break;
+
+      case RADEON_CMD_SCALARS:
+	 if (radeon_emit_scalars( header, &cmdbuf )) {
+	    fprintf(stderr,"radeon_emit_scalars failed\n");
+	    return -EINVAL;
+	 }
+	 break;
+
+      case RADEON_CMD_SCALARS2:
+	 if (radeon_emit_scalars2( header, &cmdbuf )) {
+	    fprintf(stderr,"radeon_emit_scalars failed\n");
+	    return -EINVAL;
+	 }
+	 break;
+
+      case RADEON_CMD_VECTORS:
+	 if (radeon_emit_vectors( header, &cmdbuf )) {
+	    fprintf(stderr,"radeon_emit_vectors failed\n");
+	    return -EINVAL;
+	 }
+	 break;
+
+      case RADEON_CMD_DMA_DISCARD:
+	 idx = header.dma.buf_idx;
+	 if (NORMAL)
+	    fprintf(stderr, "RADEON_CMD_DMA_DISCARD buf %d\n", idx);
+	 bufs++;
+	 break;
+
+      case RADEON_CMD_PACKET3:
+	 if (radeon_emit_packet3( &cmdbuf )) {
+	    fprintf(stderr,"radeon_emit_packet3 failed\n");
+	    return -EINVAL;
+	 }
+	 break;
+
+      case RADEON_CMD_PACKET3_CLIP:
+	 if (radeon_emit_packet3_cliprect( &cmdbuf )) {
+	    fprintf(stderr,"radeon_emit_packet3_clip failed\n");
+	    return -EINVAL;
+	 }
+	 break;
+
+      case RADEON_CMD_WAIT:
+	 break;
+
+      default:
+	 fprintf(stderr,"bad cmd_type %d at %p\n", 
+		   header.header.cmd_type,
+		   cmdbuf.buf - sizeof(header));
+	 return -EINVAL;
+      }
+   }
+
+   if (0)
+   {
+      static int n = 0;
+      n++;
+      if (n == 10) {
+	 fprintf(stderr, "Bufs %d Total emitted %d real changes %d (%.2f%%)\n",
+		 bufs,
+		 total, total_changed, 
+		 ((float)total_changed/(float)total*100.0));
+	 fprintf(stderr, "Total emitted per buf: %.2f\n",
+		 (float)total/(float)bufs);
+	 fprintf(stderr, "Real changes per buf: %.2f\n",
+		 (float)total_changed/(float)bufs);
+
+	 bufs = n = total = total_changed = 0;
+      }
+   }
+
+   return 0;
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_sanity.h b/src/mesa/drivers/dri/radeon/radeon_sanity.h
new file mode 100644
index 00000000000..58e8335dd60
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_sanity.h
@@ -0,0 +1,8 @@
+#ifndef RADEON_SANITY_H
+#define RADEON_SANITY_H
+
+extern int radeonSanityCmdBuffer( radeonContextPtr rmesa,
+				  int nbox,
+				  XF86DRIClipRectRec *boxes );
+
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_screen.c b/src/mesa/drivers/dri/radeon/radeon_screen.c
new file mode 100644
index 00000000000..ec8ed42d649
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_screen.c
@@ -0,0 +1,421 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_screen.c,v 1.6 2002/12/16 16:18:58 dawes Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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 (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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ *
+ */
+
+#include "glheader.h"
+#include "imports.h"
+
+#include "radeon_context.h"
+#include "radeon_screen.h"
+#include "radeon_macros.h"
+
+#include "utils.h"
+#include "context.h"
+#include "vblank.h"
+
+#ifndef _SOLO
+#include "glxextensions.h"
+#endif 
+
+#if 1
+/* Including xf86PciInfo.h introduces a bunch of errors...
+ */
+#define PCI_CHIP_RADEON_QD	0x5144
+#define PCI_CHIP_RADEON_QE	0x5145
+#define PCI_CHIP_RADEON_QF	0x5146
+#define PCI_CHIP_RADEON_QG	0x5147
+
+#define PCI_CHIP_RADEON_QY	0x5159
+#define PCI_CHIP_RADEON_QZ	0x515A
+
+#define PCI_CHIP_RADEON_LW	0x4C57 /* mobility 7 - has tcl */
+
+#define PCI_CHIP_RADEON_LY	0x4C59
+#define PCI_CHIP_RADEON_LZ	0x4C5A
+
+#define PCI_CHIP_RV200_QW	0x5157 /* Radeon 7500 - not an R200 at all */
+#endif
+
+static int getSwapInfo( __DRIdrawablePrivate *dPriv, __DRIswapInfo * sInfo );
+
+/* Create the device specific screen private data struct.
+ */
+radeonScreenPtr radeonCreateScreen( __DRIscreenPrivate *sPriv )
+{
+   radeonScreenPtr screen;
+   RADEONDRIPtr dri_priv = (RADEONDRIPtr)sPriv->pDevPriv;
+
+   if ( ! driCheckDriDdxDrmVersions( sPriv, "Radeon", 4, 0, 4, 0, 1, 3 ) )
+      return NULL;
+
+   /* Allocate the private area */
+   screen = (radeonScreenPtr) CALLOC( sizeof(*screen) );
+   if ( !screen ) {
+      __driUtilMessage("%s: Could not allocate memory for screen structure",
+		       __FUNCTION__);
+      return NULL;
+   }
+
+   if ( sPriv->drmMinor < 3 ||
+        getenv("RADEON_COMPAT")) {
+	   fprintf( stderr, "Radeon DRI driver:\n\t"
+		    "Compatibility mode for DRM driver version %d.%d.%d\n\t"
+		    "TCL will be disabled, expect reduced performance\n\t"
+		    "(prefer DRM radeon.o 1.3.x or newer)\n\t", 
+		    sPriv->drmMajor, sPriv->drmMinor, sPriv->drmPatch ); 
+   }
+
+
+   /* This is first since which regions we map depends on whether or
+    * not we are using a PCI card.
+    */
+   screen->IsPCI = dri_priv->IsPCI;
+
+   if (sPriv->drmMinor >= 3) {
+      int ret;
+      drmRadeonGetParam gp;
+
+      gp.param = RADEON_PARAM_AGP_BUFFER_OFFSET;
+      gp.value = &screen->agp_buffer_offset;
+
+      ret = drmCommandWriteRead( sPriv->fd, DRM_RADEON_GETPARAM,
+				 &gp, sizeof(gp));
+      if (ret) {
+	 fprintf(stderr, "drmRadeonGetParam (RADEON_PARAM_AGP_BUFFER_OFFSET): %d\n", ret);
+	 return NULL;
+      }
+
+      if (sPriv->drmMinor >= 6) {
+	 gp.param = RADEON_PARAM_IRQ_NR;
+	 gp.value = &screen->irq;
+
+	 ret = drmCommandWriteRead( sPriv->fd, DRM_RADEON_GETPARAM,
+				    &gp, sizeof(gp));
+	 if (ret) {
+	    FREE( screen );
+	    fprintf(stderr, "drmRadeonGetParam (RADEON_PARAM_IRQ_NR): %d\n", ret);
+	    return NULL;
+	 }
+      }
+   }
+
+   screen->mmio.handle = dri_priv->registerHandle;
+   screen->mmio.size   = dri_priv->registerSize;
+   if ( drmMap( sPriv->fd,
+		screen->mmio.handle,
+		screen->mmio.size,
+		&screen->mmio.map ) ) {
+      FREE( screen );
+      __driUtilMessage("%s: drmMap failed\n", __FUNCTION__ );
+      return NULL;
+   }
+
+   screen->status.handle = dri_priv->statusHandle;
+   screen->status.size   = dri_priv->statusSize;
+   if ( drmMap( sPriv->fd,
+		screen->status.handle,
+		screen->status.size,
+		&screen->status.map ) ) {
+      drmUnmap( screen->mmio.map, screen->mmio.size );
+      FREE( screen );
+      __driUtilMessage("%s: drmMap (2) failed\n", __FUNCTION__ );
+      return NULL;
+   }
+   screen->scratch = (__volatile__ CARD32 *)
+      ((GLubyte *)screen->status.map + RADEON_SCRATCH_REG_OFFSET);
+
+   screen->buffers = drmMapBufs( sPriv->fd );
+   if ( !screen->buffers ) {
+      drmUnmap( screen->status.map, screen->status.size );
+      drmUnmap( screen->mmio.map, screen->mmio.size );
+      FREE( screen );
+      __driUtilMessage("%s: drmMapBufs failed\n", __FUNCTION__ );
+      return NULL;
+   }
+
+   if ( !screen->IsPCI ) {
+      screen->agpTextures.handle = dri_priv->agpTexHandle;
+      screen->agpTextures.size   = dri_priv->agpTexMapSize;
+      if ( drmMap( sPriv->fd,
+		   screen->agpTextures.handle,
+		   screen->agpTextures.size,
+		   (drmAddressPtr)&screen->agpTextures.map ) ) {
+	 drmUnmapBufs( screen->buffers );
+	 drmUnmap( screen->status.map, screen->status.size );
+	 drmUnmap( screen->mmio.map, screen->mmio.size );
+	 FREE( screen );
+         __driUtilMessage("%s: IsPCI failed\n", __FUNCTION__);
+	 return NULL;
+      }
+   }
+
+   screen->chipset = 0;
+   switch ( dri_priv->deviceID ) {
+   default:
+      fprintf(stderr, "unknown chip id, assuming full radeon support\n");
+   case PCI_CHIP_RADEON_QD:
+   case PCI_CHIP_RADEON_QE:
+   case PCI_CHIP_RADEON_QF:
+   case PCI_CHIP_RADEON_QG:
+   case PCI_CHIP_RV200_QW:
+   case PCI_CHIP_RADEON_LW:
+      screen->chipset |= RADEON_CHIPSET_TCL;
+   case PCI_CHIP_RADEON_QY:
+   case PCI_CHIP_RADEON_QZ:
+   case PCI_CHIP_RADEON_LY:
+   case PCI_CHIP_RADEON_LZ:
+      break;
+   }
+
+   screen->cpp = dri_priv->bpp / 8;
+   screen->AGPMode = dri_priv->AGPMode;
+
+   screen->frontOffset	= dri_priv->frontOffset;
+   screen->frontPitch	= dri_priv->frontPitch;
+   screen->backOffset	= dri_priv->backOffset;
+   screen->backPitch	= dri_priv->backPitch;
+   screen->depthOffset	= dri_priv->depthOffset;
+   screen->depthPitch	= dri_priv->depthPitch;
+
+   screen->texOffset[RADEON_CARD_HEAP] = dri_priv->textureOffset;
+   screen->texSize[RADEON_CARD_HEAP] = dri_priv->textureSize;
+   screen->logTexGranularity[RADEON_CARD_HEAP] =
+      dri_priv->log2TexGran;
+
+   if ( screen->IsPCI 
+	|| getenv( "RADEON_AGPTEXTURING_FORCE_DISABLE" ) ) {
+      screen->numTexHeaps = RADEON_NR_TEX_HEAPS - 1;
+      screen->texOffset[RADEON_AGP_HEAP] = 0;
+      screen->texSize[RADEON_AGP_HEAP] = 0;
+      screen->logTexGranularity[RADEON_AGP_HEAP] = 0;
+   } else {
+      screen->numTexHeaps = RADEON_NR_TEX_HEAPS;
+      screen->texOffset[RADEON_AGP_HEAP] =
+	 dri_priv->agpTexOffset + RADEON_AGP_TEX_OFFSET;
+      screen->texSize[RADEON_AGP_HEAP] = dri_priv->agpTexMapSize;
+      screen->logTexGranularity[RADEON_AGP_HEAP] =
+	 dri_priv->log2AGPTexGran;
+   }
+
+   screen->driScreen = sPriv;
+   screen->sarea_priv_offset = dri_priv->sarea_priv_offset;
+   return screen;
+}
+
+/* Destroy the device specific screen private data struct.
+ */
+void radeonDestroyScreen( __DRIscreenPrivate *sPriv )
+{
+   radeonScreenPtr screen = (radeonScreenPtr)sPriv->private;
+
+   if (!screen)
+      return;
+
+   if ( !screen->IsPCI ) {
+      drmUnmap( screen->agpTextures.map,
+		screen->agpTextures.size );
+   }
+   drmUnmapBufs( screen->buffers );
+   drmUnmap( screen->status.map, screen->status.size );
+   drmUnmap( screen->mmio.map, screen->mmio.size );
+
+   FREE( screen );
+   sPriv->private = NULL;
+}
+
+
+/* Initialize the driver specific screen private data.
+ */
+static GLboolean
+radeonInitDriver( __DRIscreenPrivate *sPriv )
+{
+   sPriv->private = (void *) radeonCreateScreen( sPriv );
+   if ( !sPriv->private ) {
+      radeonDestroyScreen( sPriv );
+      return GL_FALSE;
+   }
+
+   return GL_TRUE;
+}
+
+
+
+/* Create and initialize the Mesa and driver specific pixmap buffer
+ * data.
+ */
+static GLboolean
+radeonCreateBuffer( __DRIscreenPrivate *driScrnPriv,
+                    __DRIdrawablePrivate *driDrawPriv,
+                    const __GLcontextModes *mesaVis,
+                    GLboolean isPixmap )
+{
+   if (isPixmap) {
+      return GL_FALSE; /* not implemented */
+   }
+   else {
+      const GLboolean swDepth = GL_FALSE;
+      const GLboolean swAlpha = GL_FALSE;
+      const GLboolean swAccum = mesaVis->accumRedBits > 0;
+      const GLboolean swStencil = mesaVis->stencilBits > 0 &&
+         mesaVis->depthBits != 24;
+      driDrawPriv->driverPrivate = (void *)
+         _mesa_create_framebuffer( mesaVis,
+                                   swDepth,
+                                   swStencil,
+                                   swAccum,
+                                   swAlpha );
+      return (driDrawPriv->driverPrivate != NULL);
+   }
+}
+
+
+static void
+radeonDestroyBuffer(__DRIdrawablePrivate *driDrawPriv)
+{
+   _mesa_destroy_framebuffer((GLframebuffer *) (driDrawPriv->driverPrivate));
+}
+
+
+
+
+/* Fullscreen mode isn't used for much -- could be a way to shrink
+ * front/back buffers & get more texture memory if the client has
+ * changed the video resolution.
+ * 
+ * Pageflipping is now done automatically whenever there is a single
+ * 3d client.
+ */
+static GLboolean
+radeonOpenCloseFullScreen( __DRIcontextPrivate *driContextPriv )
+{
+   return GL_TRUE;
+}
+
+static struct __DriverAPIRec radeonAPI = {
+   .InitDriver      = radeonInitDriver,
+   .DestroyScreen   = radeonDestroyScreen,
+   .CreateContext   = radeonCreateContext,
+   .DestroyContext  = radeonDestroyContext,
+   .CreateBuffer    = radeonCreateBuffer,
+   .DestroyBuffer   = radeonDestroyBuffer,
+   .SwapBuffers     = radeonSwapBuffers,
+   .MakeCurrent     = radeonMakeCurrent,
+   .UnbindContext   = radeonUnbindContext,
+   .OpenFullScreen  = radeonOpenCloseFullScreen,
+   .CloseFullScreen = radeonOpenCloseFullScreen,
+   .GetSwapInfo     = getSwapInfo,
+   .GetMSC          = driGetMSC32,
+   .WaitForMSC      = driWaitForMSC32,
+   .WaitForSBC      = NULL,
+   .SwapBuffersMSC  = NULL
+};
+
+
+
+/*
+ * This is the bootstrap function for the driver.
+ * The __driCreateScreen name is the symbol that libGL.so fetches.
+ * Return:  pointer to a __DRIscreenPrivate.
+ */
+#ifndef _SOLO
+void *__driCreateScreen(Display *dpy, int scrn, __DRIscreen *psc,
+                        int numConfigs, __GLXvisualConfig *config)
+{
+   __DRIscreenPrivate *psp;
+   psp = __driUtilCreateScreen(dpy, scrn, psc, numConfigs, config, &radeonAPI);
+   return (void *) psp;
+}
+#else
+void *__driCreateScreen(struct DRIDriverRec *driver,
+                        struct DRIDriverContextRec *driverContext)
+{
+   __DRIscreenPrivate *psp;
+   psp = __driUtilCreateScreen(driver, driverContext, &radeonAPI);
+   return (void *) psp;
+}
+#endif
+
+
+#ifndef _SOLO
+/* This function is called by libGL.so as soon as libGL.so is loaded.
+ * This is where we'd register new extension functions with the dispatcher.
+ */
+void
+__driRegisterExtensions( void )
+{
+   PFNGLXENABLEEXTENSIONPROC glx_enable_extension;
+
+
+   if ( driCompareGLXAPIVersion( 20030317 ) >= 0 ) {
+      glx_enable_extension = (PFNGLXENABLEEXTENSIONPROC)
+	  glXGetProcAddress( "__glXEnableExtension" );
+
+      if ( glx_enable_extension != NULL ) {
+	 glx_enable_extension( "GLX_SGI_swap_control", GL_FALSE );
+	 glx_enable_extension( "GLX_SGI_video_sync", GL_FALSE );
+	 glx_enable_extension( "GLX_MESA_swap_control", GL_FALSE );
+	 glx_enable_extension( "GLX_MESA_swap_frame_usage", GL_FALSE );
+      }
+   }
+}
+#endif
+
+
+/**
+ * Get information about previous buffer swaps.
+ */
+static int
+getSwapInfo( __DRIdrawablePrivate *dPriv, __DRIswapInfo * sInfo )
+{
+   radeonContextPtr  rmesa;
+
+   if ( (dPriv == NULL) || (dPriv->driContextPriv == NULL)
+	|| (dPriv->driContextPriv->driverPrivate == NULL)
+	|| (sInfo == NULL) ) {
+      return -1;
+   }
+
+   rmesa = (radeonContextPtr) dPriv->driContextPriv->driverPrivate;
+   sInfo->swap_count = rmesa->swap_count;
+   sInfo->swap_ust = rmesa->swap_ust;
+   sInfo->swap_missed_count = rmesa->swap_missed_count;
+
+   sInfo->swap_missed_usage = (sInfo->swap_missed_count != 0)
+       ? driCalculateSwapUsage( dPriv, 0, rmesa->swap_missed_ust )
+       : 0.0;
+
+   return 0;
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_screen.h b/src/mesa/drivers/dri/radeon/radeon_screen.h
new file mode 100644
index 00000000000..2c69d8657ae
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_screen.h
@@ -0,0 +1,101 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_screen.h,v 1.5 2002/12/16 16:18:58 dawes Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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 (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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ */
+
+#ifndef __RADEON_SCREEN_H__
+#define __RADEON_SCREEN_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+/*
+ * IMPORTS: these headers contain all the DRI, X and kernel-related
+ * definitions that we need.
+ */
+#include "dri_util.h"
+#include "radeon_common.h"
+#include "radeon_dri.h"
+#include "radeon_reg.h"
+#include "radeon_sarea.h"
+
+
+typedef struct {
+   drmHandle handle;			/* Handle to the DRM region */
+   drmSize size;			/* Size of the DRM region */
+   drmAddress map;			/* Mapping of the DRM region */
+} radeonRegionRec, *radeonRegionPtr;
+
+/* chipset features */
+#define RADEON_CHIPSET_TCL	(1 << 0)
+
+typedef struct {
+
+   int chipset;
+   int cpp;
+   int IsPCI;				/* Current card is a PCI card */
+   int AGPMode;
+   unsigned int irq;			/* IRQ number (0 means none) */
+
+   unsigned int frontOffset;
+   unsigned int frontPitch;
+   unsigned int backOffset;
+   unsigned int backPitch;
+
+   unsigned int depthOffset;
+   unsigned int depthPitch;
+
+    /* Shared texture data */
+   int numTexHeaps;
+   int texOffset[RADEON_NR_TEX_HEAPS];
+   int texSize[RADEON_NR_TEX_HEAPS];
+   int logTexGranularity[RADEON_NR_TEX_HEAPS];
+
+   radeonRegionRec mmio;
+   radeonRegionRec status;
+   radeonRegionRec agpTextures;
+
+   drmBufMapPtr buffers;
+
+   __volatile__ CARD32 *scratch;
+
+   __DRIscreenPrivate *driScreen;
+   unsigned int sarea_priv_offset;
+   unsigned int agp_buffer_offset;	/* offset in card memory space */
+} radeonScreenRec, *radeonScreenPtr;
+
+extern radeonScreenPtr radeonCreateScreen( __DRIscreenPrivate *sPriv );
+extern void radeonDestroyScreen( __DRIscreenPrivate *sPriv );
+
+#endif
+#endif /* __RADEON_SCREEN_H__ */
diff --git a/src/mesa/drivers/dri/radeon/radeon_span.c b/src/mesa/drivers/dri/radeon/radeon_span.c
new file mode 100644
index 00000000000..029d7cd8eef
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_span.c
@@ -0,0 +1,415 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_span.c,v 1.6 2002/10/30 12:51:56 alanh Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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 (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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ *   Keith Whitwell <[email protected]>
+ *
+ */
+
+#include "glheader.h"
+#include "swrast/swrast.h"
+
+#include "radeon_context.h"
+#include "radeon_ioctl.h"
+#include "radeon_state.h"
+#include "radeon_span.h"
+#include "radeon_tex.h"
+
+#define DBG 0
+
+#define LOCAL_VARS							\
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);			\
+   radeonScreenPtr radeonScreen = rmesa->radeonScreen;			\
+   __DRIscreenPrivate *sPriv = rmesa->dri.screen;			\
+   __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;			\
+   GLuint pitch = radeonScreen->frontPitch * radeonScreen->cpp;		\
+   GLuint height = dPriv->h;						\
+   char *buf = (char *)(sPriv->pFB +					\
+			rmesa->state.color.drawOffset +			\
+			(dPriv->x * radeonScreen->cpp) +		\
+			(dPriv->y * pitch));				\
+   char *read_buf = (char *)(sPriv->pFB +				\
+			     rmesa->state.pixel.readOffset +		\
+			     (dPriv->x * radeonScreen->cpp) +		\
+			     (dPriv->y * pitch));			\
+   GLuint p;								\
+   (void) read_buf; (void) buf; (void) p
+
+#define LOCAL_DEPTH_VARS						\
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);			\
+   radeonScreenPtr radeonScreen = rmesa->radeonScreen;			\
+   __DRIscreenPrivate *sPriv = rmesa->dri.screen;			\
+   __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;			\
+   GLuint height = dPriv->h;						\
+   GLuint xo = dPriv->x;						\
+   GLuint yo = dPriv->y;						\
+   char *buf = (char *)(sPriv->pFB + radeonScreen->depthOffset);	\
+   (void) buf
+
+#define LOCAL_STENCIL_VARS	LOCAL_DEPTH_VARS
+
+
+#define CLIPPIXEL( _x, _y )						\
+   ((_x >= minx) && (_x < maxx) && (_y >= miny) && (_y < maxy))
+
+
+#define CLIPSPAN( _x, _y, _n, _x1, _n1, _i )				\
+   if ( _y < miny || _y >= maxy ) {					\
+      _n1 = 0, _x1 = x;							\
+   } else {								\
+      _n1 = _n;								\
+      _x1 = _x;								\
+      if ( _x1 < minx ) _i += (minx-_x1), n1 -= (minx-_x1), _x1 = minx; \
+      if ( _x1 + _n1 >= maxx ) n1 -= (_x1 + n1 - maxx);		        \
+   }
+
+#define Y_FLIP( _y )		(height - _y - 1)
+
+
+#define HW_LOCK() 
+
+#define HW_CLIPLOOP()							\
+   do {									\
+      __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;		\
+      int _nc = dPriv->numClipRects;					\
+									\
+      while ( _nc-- ) {							\
+	 int minx = dPriv->pClipRects[_nc].x1 - dPriv->x;		\
+	 int miny = dPriv->pClipRects[_nc].y1 - dPriv->y;		\
+	 int maxx = dPriv->pClipRects[_nc].x2 - dPriv->x;		\
+	 int maxy = dPriv->pClipRects[_nc].y2 - dPriv->y;
+
+#define HW_ENDCLIPLOOP()						\
+      }									\
+   } while (0)
+
+#define HW_UNLOCK()							
+
+
+
+/* ================================================================
+ * Color buffer
+ */
+
+/* 16 bit, RGB565 color spanline and pixel functions
+ */
+#define INIT_MONO_PIXEL(p, color) \
+  p = PACK_COLOR_565( color[0], color[1], color[2] )
+
+#define WRITE_RGBA( _x, _y, r, g, b, a )				\
+   *(GLushort *)(buf + _x*2 + _y*pitch) = ((((int)r & 0xf8) << 8) |	\
+					   (((int)g & 0xfc) << 3) |	\
+					   (((int)b & 0xf8) >> 3))
+
+#define WRITE_PIXEL( _x, _y, p )					\
+   *(GLushort *)(buf + _x*2 + _y*pitch) = p
+
+#define READ_RGBA( rgba, _x, _y )					\
+   do {									\
+      GLushort p = *(GLushort *)(read_buf + _x*2 + _y*pitch);		\
+      rgba[0] = ((p >> 8) & 0xf8) * 255 / 0xf8;				\
+      rgba[1] = ((p >> 3) & 0xfc) * 255 / 0xfc;				\
+      rgba[2] = ((p << 3) & 0xf8) * 255 / 0xf8;				\
+      rgba[3] = 0xff;							\
+   } while (0)
+
+#define TAG(x) radeon##x##_RGB565
+#include "spantmp.h"
+
+/* 32 bit, ARGB8888 color spanline and pixel functions
+ */
+#undef INIT_MONO_PIXEL
+#define INIT_MONO_PIXEL(p, color) \
+  p = PACK_COLOR_8888( color[3], color[0], color[1], color[2] )
+
+#define WRITE_RGBA( _x, _y, r, g, b, a )			\
+do {								\
+   *(GLuint *)(buf + _x*4 + _y*pitch) = ((b <<  0) |		\
+					 (g <<  8) |		\
+					 (r << 16) |		\
+					 (a << 24) );		\
+} while (0)
+
+#define WRITE_PIXEL( _x, _y, p ) 			\
+do {							\
+   *(GLuint *)(buf + _x*4 + _y*pitch) = p;		\
+} while (0)
+
+#define READ_RGBA( rgba, _x, _y )				\
+do {								\
+   volatile GLuint *ptr = (volatile GLuint *)(read_buf + _x*4 + _y*pitch); \
+   GLuint p = *ptr;					\
+   rgba[0] = (p >> 16) & 0xff;					\
+   rgba[1] = (p >>  8) & 0xff;					\
+   rgba[2] = (p >>  0) & 0xff;					\
+   rgba[3] = (p >> 24) & 0xff;					\
+} while (0)
+
+#define TAG(x) radeon##x##_ARGB8888
+#include "spantmp.h"
+
+
+
+/* ================================================================
+ * Depth buffer
+ */
+
+/* The Radeon family has depth tiling on all the time, so we have to convert
+ * the x,y coordinates into the memory bus address (mba) in the same
+ * manner as the engine.  In each case, the linear block address (ba)
+ * is calculated, and then wired with x and y to produce the final
+ * memory address.
+ */
+
+static GLuint radeon_mba_z32( radeonContextPtr rmesa,
+				       GLint x, GLint y )
+{
+   GLuint pitch = rmesa->radeonScreen->frontPitch;
+   GLuint ba, address = 0;			/* a[0..1] = 0           */
+
+   ba = (y / 16) * (pitch / 16) + (x / 16);
+
+   address |= (x & 0x7) << 2;			/* a[2..4] = x[0..2]     */
+   address |= (y & 0x3) << 5;			/* a[5..6] = y[0..1]     */
+   address |=
+      (((x & 0x10) >> 2) ^ (y & 0x4)) << 5;	/* a[7]    = x[4] ^ y[2] */
+   address |= (ba & 0x3) << 8;			/* a[8..9] = ba[0..1]    */
+
+   address |= (y & 0x8) << 7;			/* a[10]   = y[3]        */
+   address |=
+      (((x & 0x8) << 1) ^ (y & 0x10)) << 7;	/* a[11]   = x[3] ^ y[4] */
+   address |= (ba & ~0x3) << 10;		/* a[12..] = ba[2..]     */
+
+   return address;
+}
+
+static __inline GLuint radeon_mba_z16( radeonContextPtr rmesa, GLint x, GLint y )
+{
+   GLuint pitch = rmesa->radeonScreen->frontPitch;
+   GLuint ba, address = 0;			/* a[0]    = 0           */
+
+   ba = (y / 16) * (pitch / 32) + (x / 32);
+
+   address |= (x & 0x7) << 1;			/* a[1..3] = x[0..2]     */
+   address |= (y & 0x7) << 4;			/* a[4..6] = y[0..2]     */
+   address |= (x & 0x8) << 4;			/* a[7]    = x[3]        */
+   address |= (ba & 0x3) << 8;			/* a[8..9] = ba[0..1]    */
+   address |= (y & 0x8) << 7;			/* a[10]   = y[3]        */
+   address |= ((x & 0x10) ^ (y & 0x10)) << 7;	/* a[11]   = x[4] ^ y[4] */
+   address |= (ba & ~0x3) << 10;		/* a[12..] = ba[2..]     */
+
+   return address;
+}
+
+
+/* 16-bit depth buffer functions
+ */
+#define WRITE_DEPTH( _x, _y, d )					\
+   *(GLushort *)(buf + radeon_mba_z16( rmesa, _x + xo, _y + yo )) = d;
+
+#define READ_DEPTH( d, _x, _y )						\
+   d = *(GLushort *)(buf + radeon_mba_z16( rmesa, _x + xo, _y + yo ));
+
+#define TAG(x) radeon##x##_16
+#include "depthtmp.h"
+
+/* 24 bit depth, 8 bit stencil depthbuffer functions
+ */
+#define WRITE_DEPTH( _x, _y, d )					\
+do {									\
+   GLuint offset = radeon_mba_z32( rmesa, _x + xo, _y + yo );		\
+   GLuint tmp = *(GLuint *)(buf + offset);				\
+   tmp &= 0xff000000;							\
+   tmp |= ((d) & 0x00ffffff);						\
+   *(GLuint *)(buf + offset) = tmp;					\
+} while (0)
+
+#define READ_DEPTH( d, _x, _y )						\
+   d = *(GLuint *)(buf + radeon_mba_z32( rmesa, _x + xo,		\
+					 _y + yo )) & 0x00ffffff;
+
+#define TAG(x) radeon##x##_24_8
+#include "depthtmp.h"
+
+
+/* ================================================================
+ * Stencil buffer
+ */
+
+/* 24 bit depth, 8 bit stencil depthbuffer functions
+ */
+#define WRITE_STENCIL( _x, _y, d )					\
+do {									\
+   GLuint offset = radeon_mba_z32( rmesa, _x + xo, _y + yo );		\
+   GLuint tmp = *(GLuint *)(buf + offset);				\
+   tmp &= 0x00ffffff;							\
+   tmp |= (((d) & 0xff) << 24);						\
+   *(GLuint *)(buf + offset) = tmp;					\
+} while (0)
+
+#define READ_STENCIL( d, _x, _y )					\
+do {									\
+   GLuint offset = radeon_mba_z32( rmesa, _x + xo, _y + yo );		\
+   GLuint tmp = *(GLuint *)(buf + offset);				\
+   tmp &= 0xff000000;							\
+   d = tmp >> 24;							\
+} while (0)
+
+#define TAG(x) radeon##x##_24_8
+#include "stenciltmp.h"
+
+
+/*
+ * This function is called to specify which buffer to read and write
+ * for software rasterization (swrast) fallbacks.  This doesn't necessarily
+ * correspond to glDrawBuffer() or glReadBuffer() calls.
+ */
+static void radeonSetBuffer( GLcontext *ctx,
+                             GLframebuffer *colorBuffer,
+                             GLuint bufferBit )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   switch ( bufferBit ) {
+   case FRONT_LEFT_BIT:
+      if ( rmesa->sarea->pfCurrentPage == 1 ) {
+        rmesa->state.pixel.readOffset = rmesa->radeonScreen->backOffset;
+        rmesa->state.pixel.readPitch  = rmesa->radeonScreen->backPitch;
+        rmesa->state.color.drawOffset = rmesa->radeonScreen->backOffset;
+        rmesa->state.color.drawPitch  = rmesa->radeonScreen->backPitch;
+      } else {
+      	rmesa->state.pixel.readOffset = rmesa->radeonScreen->frontOffset;
+      	rmesa->state.pixel.readPitch  = rmesa->radeonScreen->frontPitch;
+      	rmesa->state.color.drawOffset = rmesa->radeonScreen->frontOffset;
+      	rmesa->state.color.drawPitch  = rmesa->radeonScreen->frontPitch;
+      }
+      break;
+   case BACK_LEFT_BIT:
+      if ( rmesa->sarea->pfCurrentPage == 1 ) {
+      	rmesa->state.pixel.readOffset = rmesa->radeonScreen->frontOffset;
+      	rmesa->state.pixel.readPitch  = rmesa->radeonScreen->frontPitch;
+      	rmesa->state.color.drawOffset = rmesa->radeonScreen->frontOffset;
+      	rmesa->state.color.drawPitch  = rmesa->radeonScreen->frontPitch;
+      } else {
+        rmesa->state.pixel.readOffset = rmesa->radeonScreen->backOffset;
+        rmesa->state.pixel.readPitch  = rmesa->radeonScreen->backPitch;
+        rmesa->state.color.drawOffset = rmesa->radeonScreen->backOffset;
+        rmesa->state.color.drawPitch  = rmesa->radeonScreen->backPitch;
+      }
+      break;
+   default:
+      assert(0);
+      break;
+   }
+}
+
+/* Move locking out to get reasonable span performance (10x better
+ * than doing this in HW_LOCK above).  WaitForIdle() is the main
+ * culprit.
+ */
+
+static void radeonSpanRenderStart( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+
+   RADEON_FIREVERTICES( rmesa );
+   LOCK_HARDWARE( rmesa );
+   radeonWaitForIdleLocked( rmesa );
+}
+
+static void radeonSpanRenderFinish( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+   _swrast_flush( ctx );
+   UNLOCK_HARDWARE( rmesa );
+}
+
+void radeonInitSpanFuncs( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   struct swrast_device_driver *swdd = _swrast_GetDeviceDriverReference(ctx);
+
+   swdd->SetBuffer = radeonSetBuffer;
+
+   switch ( rmesa->radeonScreen->cpp ) {
+   case 2:
+      swdd->WriteRGBASpan	= radeonWriteRGBASpan_RGB565;
+      swdd->WriteRGBSpan	= radeonWriteRGBSpan_RGB565;
+      swdd->WriteMonoRGBASpan	= radeonWriteMonoRGBASpan_RGB565;
+      swdd->WriteRGBAPixels	= radeonWriteRGBAPixels_RGB565;
+      swdd->WriteMonoRGBAPixels	= radeonWriteMonoRGBAPixels_RGB565;
+      swdd->ReadRGBASpan	= radeonReadRGBASpan_RGB565;
+      swdd->ReadRGBAPixels      = radeonReadRGBAPixels_RGB565;
+      break;
+
+   case 4:
+      swdd->WriteRGBASpan	= radeonWriteRGBASpan_ARGB8888;
+      swdd->WriteRGBSpan	= radeonWriteRGBSpan_ARGB8888;
+      swdd->WriteMonoRGBASpan   = radeonWriteMonoRGBASpan_ARGB8888;
+      swdd->WriteRGBAPixels     = radeonWriteRGBAPixels_ARGB8888;
+      swdd->WriteMonoRGBAPixels = radeonWriteMonoRGBAPixels_ARGB8888;
+      swdd->ReadRGBASpan	= radeonReadRGBASpan_ARGB8888;
+      swdd->ReadRGBAPixels      = radeonReadRGBAPixels_ARGB8888;
+      break;
+
+   default:
+      break;
+   }
+
+   switch ( rmesa->glCtx->Visual.depthBits ) {
+   case 16:
+      swdd->ReadDepthSpan	= radeonReadDepthSpan_16;
+      swdd->WriteDepthSpan	= radeonWriteDepthSpan_16;
+      swdd->ReadDepthPixels	= radeonReadDepthPixels_16;
+      swdd->WriteDepthPixels	= radeonWriteDepthPixels_16;
+      break;
+
+   case 24:
+      swdd->ReadDepthSpan	= radeonReadDepthSpan_24_8;
+      swdd->WriteDepthSpan	= radeonWriteDepthSpan_24_8;
+      swdd->ReadDepthPixels	= radeonReadDepthPixels_24_8;
+      swdd->WriteDepthPixels	= radeonWriteDepthPixels_24_8;
+
+      swdd->ReadStencilSpan	= radeonReadStencilSpan_24_8;
+      swdd->WriteStencilSpan	= radeonWriteStencilSpan_24_8;
+      swdd->ReadStencilPixels	= radeonReadStencilPixels_24_8;
+      swdd->WriteStencilPixels	= radeonWriteStencilPixels_24_8;
+      break;
+
+   default:
+      break;
+   }
+
+   swdd->SpanRenderStart          = radeonSpanRenderStart;
+   swdd->SpanRenderFinish         = radeonSpanRenderFinish; 
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_span.h b/src/mesa/drivers/dri/radeon/radeon_span.h
new file mode 100644
index 00000000000..011e8eff57c
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_span.h
@@ -0,0 +1,45 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_span.h,v 1.2 2002/02/22 21:45:01 dawes Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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 (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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ */
+
+#ifndef __RADEON_SPAN_H__
+#define __RADEON_SPAN_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+extern void radeonInitSpanFuncs( GLcontext *ctx );
+
+#endif
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_state.c b/src/mesa/drivers/dri/radeon/radeon_state.c
new file mode 100644
index 00000000000..7b1bbe75fd2
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_state.c
@@ -0,0 +1,2211 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_state.c,v 1.8 2002/12/16 16:18:58 dawes Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 VA Linux Systems Inc., Fremont, California.
+
+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 (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:
+ *   Gareth Hughes <[email protected]>
+ *   Keith Whitwell <[email protected]>
+ */
+
+#include "glheader.h"
+#include "imports.h"
+#include "api_arrayelt.h"
+#include "enums.h"
+#include "colormac.h"
+#include "state.h"
+
+#include "swrast/swrast.h"
+#include "array_cache/acache.h"
+#include "tnl/tnl.h"
+#include "tnl/t_pipeline.h"
+#include "main/light.h"
+#include "swrast_setup/swrast_setup.h"
+
+#include "radeon_context.h"
+#include "radeon_ioctl.h"
+#include "radeon_state.h"
+#include "radeon_tcl.h"
+#include "radeon_tex.h"
+#include "radeon_swtcl.h"
+#include "radeon_vtxfmt.h"
+
+/* =============================================================
+ * Alpha blending
+ */
+
+static void radeonAlphaFunc( GLcontext *ctx, GLenum func, GLfloat ref )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   int pp_misc = rmesa->hw.ctx.cmd[CTX_PP_MISC];
+   GLubyte refByte;
+
+   CLAMPED_FLOAT_TO_UBYTE(refByte, ref);
+
+   RADEON_STATECHANGE( rmesa, ctx );
+
+   pp_misc &= ~(RADEON_ALPHA_TEST_OP_MASK | RADEON_REF_ALPHA_MASK);
+   pp_misc |= (refByte & RADEON_REF_ALPHA_MASK);
+
+   switch ( func ) {
+   case GL_NEVER:
+      pp_misc |= RADEON_ALPHA_TEST_FAIL;
+      break;
+   case GL_LESS:
+      pp_misc |= RADEON_ALPHA_TEST_LESS;
+      break;
+   case GL_EQUAL:
+      pp_misc |= RADEON_ALPHA_TEST_EQUAL;
+      break;
+   case GL_LEQUAL:
+      pp_misc |= RADEON_ALPHA_TEST_LEQUAL;
+      break;
+   case GL_GREATER:
+      pp_misc |= RADEON_ALPHA_TEST_GREATER;
+      break;
+   case GL_NOTEQUAL:
+      pp_misc |= RADEON_ALPHA_TEST_NEQUAL;
+      break;
+   case GL_GEQUAL:
+      pp_misc |= RADEON_ALPHA_TEST_GEQUAL;
+      break;
+   case GL_ALWAYS:
+      pp_misc |= RADEON_ALPHA_TEST_PASS;
+      break;
+   }
+
+   rmesa->hw.ctx.cmd[CTX_PP_MISC] = pp_misc;
+}
+
+static void radeonBlendEquation( GLcontext *ctx, GLenum mode )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint b = rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCNTL] & ~RADEON_COMB_FCN_MASK;
+   GLboolean fallback = GL_FALSE;
+
+   switch ( mode ) {
+   case GL_FUNC_ADD:
+   case GL_LOGIC_OP:
+      b |= RADEON_COMB_FCN_ADD_CLAMP;
+      break;
+
+   case GL_FUNC_SUBTRACT:
+      b |= RADEON_COMB_FCN_SUB_CLAMP;
+      break;
+
+   default:
+      if (ctx->Color.BlendEnabled)
+	 fallback = GL_TRUE;
+      else
+	 b |= RADEON_COMB_FCN_ADD_CLAMP;
+      break;
+   }
+
+   FALLBACK( rmesa, RADEON_FALLBACK_BLEND_EQ, fallback );
+   if ( !fallback ) {
+      RADEON_STATECHANGE( rmesa, ctx );
+      rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCNTL] = b;
+      if ( ctx->Color.ColorLogicOpEnabled ) {
+	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  RADEON_ROP_ENABLE;
+      } else {
+	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~RADEON_ROP_ENABLE;
+      }
+   }
+}
+
+static void radeonBlendFunc( GLcontext *ctx, GLenum sfactor, GLenum dfactor )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint b = rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCNTL] & 
+      ~(RADEON_SRC_BLEND_MASK | RADEON_DST_BLEND_MASK);
+   GLboolean fallback = GL_FALSE;
+
+   switch ( ctx->Color.BlendSrcRGB ) {
+   case GL_ZERO:
+      b |= RADEON_SRC_BLEND_GL_ZERO;
+      break;
+   case GL_ONE:
+      b |= RADEON_SRC_BLEND_GL_ONE;
+      break;
+   case GL_DST_COLOR:
+      b |= RADEON_SRC_BLEND_GL_DST_COLOR;
+      break;
+   case GL_ONE_MINUS_DST_COLOR:
+      b |= RADEON_SRC_BLEND_GL_ONE_MINUS_DST_COLOR;
+      break;
+   case GL_SRC_COLOR:
+      b |= RADEON_SRC_BLEND_GL_SRC_COLOR;
+      break;
+   case GL_ONE_MINUS_SRC_COLOR:
+      b |= RADEON_SRC_BLEND_GL_ONE_MINUS_SRC_COLOR;
+      break;
+   case GL_SRC_ALPHA:
+      b |= RADEON_SRC_BLEND_GL_SRC_ALPHA;
+      break;
+   case GL_ONE_MINUS_SRC_ALPHA:
+      b |= RADEON_SRC_BLEND_GL_ONE_MINUS_SRC_ALPHA;
+      break;
+   case GL_DST_ALPHA:
+      b |= RADEON_SRC_BLEND_GL_DST_ALPHA;
+      break;
+   case GL_ONE_MINUS_DST_ALPHA:
+      b |= RADEON_SRC_BLEND_GL_ONE_MINUS_DST_ALPHA;
+      break;
+   case GL_SRC_ALPHA_SATURATE:
+      b |= RADEON_SRC_BLEND_GL_SRC_ALPHA_SATURATE;
+      break;
+   case GL_CONSTANT_COLOR:
+   case GL_ONE_MINUS_CONSTANT_COLOR:
+   case GL_CONSTANT_ALPHA:
+   case GL_ONE_MINUS_CONSTANT_ALPHA:
+      if (ctx->Color.BlendEnabled)
+	 fallback = GL_TRUE;
+      else
+	 b |= RADEON_SRC_BLEND_GL_ONE;
+      break;
+   default:
+      break;
+   }
+
+   switch ( ctx->Color.BlendDstRGB ) {
+   case GL_ZERO:
+      b |= RADEON_DST_BLEND_GL_ZERO;
+      break;
+   case GL_ONE:
+      b |= RADEON_DST_BLEND_GL_ONE;
+      break;
+   case GL_SRC_COLOR:
+      b |= RADEON_DST_BLEND_GL_SRC_COLOR;
+      break;
+   case GL_ONE_MINUS_SRC_COLOR:
+      b |= RADEON_DST_BLEND_GL_ONE_MINUS_SRC_COLOR;
+      break;
+   case GL_SRC_ALPHA:
+      b |= RADEON_DST_BLEND_GL_SRC_ALPHA;
+      break;
+   case GL_ONE_MINUS_SRC_ALPHA:
+      b |= RADEON_DST_BLEND_GL_ONE_MINUS_SRC_ALPHA;
+      break;
+   case GL_DST_COLOR:
+      b |= RADEON_DST_BLEND_GL_DST_COLOR;
+      break;
+   case GL_ONE_MINUS_DST_COLOR:
+      b |= RADEON_DST_BLEND_GL_ONE_MINUS_DST_COLOR;
+      break;
+   case GL_DST_ALPHA:
+      b |= RADEON_DST_BLEND_GL_DST_ALPHA;
+      break;
+   case GL_ONE_MINUS_DST_ALPHA:
+      b |= RADEON_DST_BLEND_GL_ONE_MINUS_DST_ALPHA;
+      break;
+   case GL_CONSTANT_COLOR:
+   case GL_ONE_MINUS_CONSTANT_COLOR:
+   case GL_CONSTANT_ALPHA:
+   case GL_ONE_MINUS_CONSTANT_ALPHA:
+      if (ctx->Color.BlendEnabled)
+	 fallback = GL_TRUE;
+      else
+	 b |= RADEON_DST_BLEND_GL_ZERO;
+      break;
+   default:
+      break;
+   }
+
+   FALLBACK( rmesa, RADEON_FALLBACK_BLEND_FUNC, fallback );
+   if ( !fallback ) {
+      RADEON_STATECHANGE( rmesa, ctx );
+      rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCNTL] = b;
+   }
+}
+
+static void radeonBlendFuncSeparate( GLcontext *ctx,
+				     GLenum sfactorRGB, GLenum dfactorRGB,
+				     GLenum sfactorA, GLenum dfactorA )
+{
+   radeonBlendFunc( ctx, sfactorRGB, dfactorRGB );
+}
+
+
+/* =============================================================
+ * Depth testing
+ */
+
+static void radeonDepthFunc( GLcontext *ctx, GLenum func )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   RADEON_STATECHANGE( rmesa, ctx );
+   rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] &= ~RADEON_Z_TEST_MASK;
+
+   switch ( ctx->Depth.Func ) {
+   case GL_NEVER:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_Z_TEST_NEVER;
+      break;
+   case GL_LESS:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_Z_TEST_LESS;
+      break;
+   case GL_EQUAL:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_Z_TEST_EQUAL;
+      break;
+   case GL_LEQUAL:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_Z_TEST_LEQUAL;
+      break;
+   case GL_GREATER:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_Z_TEST_GREATER;
+      break;
+   case GL_NOTEQUAL:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_Z_TEST_NEQUAL;
+      break;
+   case GL_GEQUAL:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_Z_TEST_GEQUAL;
+      break;
+   case GL_ALWAYS:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_Z_TEST_ALWAYS;
+      break;
+   }
+}
+
+
+static void radeonDepthMask( GLcontext *ctx, GLboolean flag )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   RADEON_STATECHANGE( rmesa, ctx );
+
+   if ( ctx->Depth.Mask ) {
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |=  RADEON_Z_WRITE_ENABLE;
+   } else {
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] &= ~RADEON_Z_WRITE_ENABLE;
+   }
+}
+
+static void radeonClearDepth( GLcontext *ctx, GLclampd d )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint format = (rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] &
+		    RADEON_DEPTH_FORMAT_MASK);
+
+   switch ( format ) {
+   case RADEON_DEPTH_FORMAT_16BIT_INT_Z:
+      rmesa->state.depth.clear = d * 0x0000ffff;
+      break;
+   case RADEON_DEPTH_FORMAT_24BIT_INT_Z:
+      rmesa->state.depth.clear = d * 0x00ffffff;
+      break;
+   }
+}
+
+
+/* =============================================================
+ * Fog
+ */
+
+
+static void radeonFogfv( GLcontext *ctx, GLenum pname, const GLfloat *param )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   union { int i; float f; } c, d;
+   GLchan col[4];
+
+   c.i = rmesa->hw.fog.cmd[FOG_C];
+   d.i = rmesa->hw.fog.cmd[FOG_D];
+
+   switch (pname) {
+   case GL_FOG_MODE:
+      if (!ctx->Fog.Enabled)
+	 return;
+      RADEON_STATECHANGE(rmesa, tcl);
+      rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] &= ~RADEON_TCL_FOG_MASK;
+      switch (ctx->Fog.Mode) {
+      case GL_LINEAR:
+	 rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] |= RADEON_TCL_FOG_LINEAR;
+	 if (ctx->Fog.Start == ctx->Fog.End) {
+	    c.f = 1.0F;
+	    d.f = 1.0F;
+	 }
+	 else {
+	    c.f = ctx->Fog.End/(ctx->Fog.End-ctx->Fog.Start);
+	    d.f = 1.0/(ctx->Fog.End-ctx->Fog.Start);
+	 }
+	 break;
+      case GL_EXP:
+	 rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] |= RADEON_TCL_FOG_EXP;
+	 c.f = 0.0;
+	 d.f = ctx->Fog.Density;
+	 break;
+      case GL_EXP2:
+	 rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] |= RADEON_TCL_FOG_EXP2;
+	 c.f = 0.0;
+	 d.f = -(ctx->Fog.Density * ctx->Fog.Density);
+	 break;
+      default:
+	 return;
+      }
+      break;
+   case GL_FOG_DENSITY:
+      switch (ctx->Fog.Mode) {
+      case GL_EXP:
+	 c.f = 0.0;
+	 d.f = ctx->Fog.Density;
+	 break;
+      case GL_EXP2:
+	 c.f = 0.0;
+	 d.f = -(ctx->Fog.Density * ctx->Fog.Density);
+	 break;
+      default:
+	 break;
+      }
+      break;
+   case GL_FOG_START:
+   case GL_FOG_END:
+      if (ctx->Fog.Mode == GL_LINEAR) {
+	 if (ctx->Fog.Start == ctx->Fog.End) {
+	    c.f = 1.0F;
+	    d.f = 1.0F;
+	 } else {
+	    c.f = ctx->Fog.End/(ctx->Fog.End-ctx->Fog.Start);
+	    d.f = 1.0/(ctx->Fog.End-ctx->Fog.Start);
+	 }
+      }
+      break;
+   case GL_FOG_COLOR: 
+      RADEON_STATECHANGE( rmesa, ctx );
+      UNCLAMPED_FLOAT_TO_RGB_CHAN( col, ctx->Fog.Color );
+      rmesa->hw.ctx.cmd[CTX_PP_FOG_COLOR] =
+	 radeonPackColor( 4, col[0], col[1], col[2], 0 );
+      break;
+   case GL_FOG_COORDINATE_SOURCE_EXT: 
+      /* What to do?
+       */
+      break;
+   default:
+      return;
+   }
+
+   if (c.i != rmesa->hw.fog.cmd[FOG_C] || d.i != rmesa->hw.fog.cmd[FOG_D]) {
+      RADEON_STATECHANGE( rmesa, fog );
+      rmesa->hw.fog.cmd[FOG_C] = c.i;
+      rmesa->hw.fog.cmd[FOG_D] = d.i;
+   }
+}
+
+
+/* =============================================================
+ * Scissoring
+ */
+
+
+static GLboolean intersect_rect( XF86DRIClipRectPtr out,
+				 XF86DRIClipRectPtr a,
+				 XF86DRIClipRectPtr b )
+{
+   *out = *a;
+   if ( b->x1 > out->x1 ) out->x1 = b->x1;
+   if ( b->y1 > out->y1 ) out->y1 = b->y1;
+   if ( b->x2 < out->x2 ) out->x2 = b->x2;
+   if ( b->y2 < out->y2 ) out->y2 = b->y2;
+   if ( out->x1 >= out->x2 ) return GL_FALSE;
+   if ( out->y1 >= out->y2 ) return GL_FALSE;
+   return GL_TRUE;
+}
+
+
+void radeonRecalcScissorRects( radeonContextPtr rmesa )
+{
+   XF86DRIClipRectPtr out;
+   int i;
+
+   /* Grow cliprect store?
+    */
+   if (rmesa->state.scissor.numAllocedClipRects < rmesa->numClipRects) {
+      while (rmesa->state.scissor.numAllocedClipRects < rmesa->numClipRects) {
+	 rmesa->state.scissor.numAllocedClipRects += 1;	/* zero case */
+	 rmesa->state.scissor.numAllocedClipRects *= 2;
+      }
+
+      if (rmesa->state.scissor.pClipRects)
+	 FREE(rmesa->state.scissor.pClipRects);
+
+      rmesa->state.scissor.pClipRects = 
+	 MALLOC( rmesa->state.scissor.numAllocedClipRects * 
+		 sizeof(XF86DRIClipRectRec) );
+
+      if ( rmesa->state.scissor.pClipRects == NULL ) {
+	 rmesa->state.scissor.numAllocedClipRects = 0;
+	 return;
+      }
+   }
+   
+   out = rmesa->state.scissor.pClipRects;
+   rmesa->state.scissor.numClipRects = 0;
+
+   for ( i = 0 ; i < rmesa->numClipRects ;  i++ ) {
+      if ( intersect_rect( out, 
+			   &rmesa->pClipRects[i], 
+			   &rmesa->state.scissor.rect ) ) {
+	 rmesa->state.scissor.numClipRects++;
+	 out++;
+      }
+   }
+}
+
+
+static void radeonUpdateScissor( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if ( rmesa->dri.drawable ) {
+      __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
+
+      int x = ctx->Scissor.X;
+      int y = dPriv->h - ctx->Scissor.Y - ctx->Scissor.Height;
+      int w = ctx->Scissor.X + ctx->Scissor.Width - 1;
+      int h = dPriv->h - ctx->Scissor.Y - 1;
+
+      rmesa->state.scissor.rect.x1 = x + dPriv->x;
+      rmesa->state.scissor.rect.y1 = y + dPriv->y;
+      rmesa->state.scissor.rect.x2 = w + dPriv->x + 1;
+      rmesa->state.scissor.rect.y2 = h + dPriv->y + 1;
+
+      radeonRecalcScissorRects( rmesa );
+   }
+}
+
+
+static void radeonScissor( GLcontext *ctx,
+			   GLint x, GLint y, GLsizei w, GLsizei h )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if ( ctx->Scissor.Enabled ) {
+      RADEON_FIREVERTICES( rmesa );	/* don't pipeline cliprect changes */
+      radeonUpdateScissor( ctx );
+   }
+
+}
+
+
+/* =============================================================
+ * Culling
+ */
+
+static void radeonCullFace( GLcontext *ctx, GLenum unused )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint s = rmesa->hw.set.cmd[SET_SE_CNTL];
+   GLuint t = rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL];
+
+   s |= RADEON_FFACE_SOLID | RADEON_BFACE_SOLID;
+   t &= ~(RADEON_CULL_FRONT | RADEON_CULL_BACK);
+
+   if ( ctx->Polygon.CullFlag ) {
+      switch ( ctx->Polygon.CullFaceMode ) {
+      case GL_FRONT:
+	 s &= ~RADEON_FFACE_SOLID;
+	 t |= RADEON_CULL_FRONT;
+	 break;
+      case GL_BACK:
+	 s &= ~RADEON_BFACE_SOLID;
+	 t |= RADEON_CULL_BACK;
+	 break;
+      case GL_FRONT_AND_BACK:
+	 s &= ~(RADEON_FFACE_SOLID | RADEON_BFACE_SOLID);
+	 t |= (RADEON_CULL_FRONT | RADEON_CULL_BACK);
+	 break;
+      }
+   }
+
+   if ( rmesa->hw.set.cmd[SET_SE_CNTL] != s ) {
+      RADEON_STATECHANGE(rmesa, set );
+      rmesa->hw.set.cmd[SET_SE_CNTL] = s;
+   }
+
+   if ( rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] != t ) {
+      RADEON_STATECHANGE(rmesa, tcl );
+      rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] = t;
+   }
+}
+
+static void radeonFrontFace( GLcontext *ctx, GLenum mode )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   RADEON_STATECHANGE( rmesa, set );
+   rmesa->hw.set.cmd[SET_SE_CNTL] &= ~RADEON_FFACE_CULL_DIR_MASK;
+
+   RADEON_STATECHANGE( rmesa, tcl );
+   rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] &= ~RADEON_CULL_FRONT_IS_CCW;
+
+   switch ( mode ) {
+   case GL_CW:
+      rmesa->hw.set.cmd[SET_SE_CNTL] |= RADEON_FFACE_CULL_CW;
+      break;
+   case GL_CCW:
+      rmesa->hw.set.cmd[SET_SE_CNTL] |= RADEON_FFACE_CULL_CCW;
+      rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] |= RADEON_CULL_FRONT_IS_CCW;
+      break;
+   }
+}
+
+
+/* =============================================================
+ * Line state
+ */
+static void radeonLineWidth( GLcontext *ctx, GLfloat widthf )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   RADEON_STATECHANGE( rmesa, lin );
+   RADEON_STATECHANGE( rmesa, set );
+
+   /* Line width is stored in U6.4 format.
+    */
+   rmesa->hw.lin.cmd[LIN_SE_LINE_WIDTH] = (GLuint)(widthf * 16.0);
+   if ( widthf > 1.0 ) {
+      rmesa->hw.set.cmd[SET_SE_CNTL] |=  RADEON_WIDELINE_ENABLE;
+   } else {
+      rmesa->hw.set.cmd[SET_SE_CNTL] &= ~RADEON_WIDELINE_ENABLE;
+   }
+}
+
+static void radeonLineStipple( GLcontext *ctx, GLint factor, GLushort pattern )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   RADEON_STATECHANGE( rmesa, lin );
+   rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] = 
+      ((((GLuint)factor & 0xff) << 16) | ((GLuint)pattern));
+}
+
+
+/* =============================================================
+ * Masks
+ */
+static void radeonColorMask( GLcontext *ctx,
+			     GLboolean r, GLboolean g,
+			     GLboolean b, GLboolean a )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint mask = radeonPackColor( rmesa->radeonScreen->cpp,
+				  ctx->Color.ColorMask[RCOMP],
+				  ctx->Color.ColorMask[GCOMP],
+				  ctx->Color.ColorMask[BCOMP],
+				  ctx->Color.ColorMask[ACOMP] );
+
+   if ( rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK] != mask ) {
+      RADEON_STATECHANGE( rmesa, msk );
+      rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK] = mask;
+   }
+}
+
+
+/* =============================================================
+ * Polygon state
+ */
+
+static void radeonPolygonOffset( GLcontext *ctx,
+				 GLfloat factor, GLfloat units )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat constant = units * rmesa->state.depth.scale;
+
+   RADEON_STATECHANGE( rmesa, zbs );
+   rmesa->hw.zbs.cmd[ZBS_SE_ZBIAS_FACTOR]   = *(GLuint *)&factor;
+   rmesa->hw.zbs.cmd[ZBS_SE_ZBIAS_CONSTANT] = *(GLuint *)&constant;
+}
+
+static void radeonPolygonStipple( GLcontext *ctx, const GLubyte *mask )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint i;
+   drmRadeonStipple stipple;
+
+   /* Must flip pattern upside down.
+    */
+   for ( i = 0 ; i < 32 ; i++ ) {
+      rmesa->state.stipple.mask[31 - i] = ((GLuint *) mask)[i];
+   }
+
+   /* TODO: push this into cmd mechanism
+    */
+   RADEON_FIREVERTICES( rmesa );
+   LOCK_HARDWARE( rmesa );
+
+   /* FIXME: Use window x,y offsets into stipple RAM.
+    */
+   stipple.mask = rmesa->state.stipple.mask;
+   drmCommandWrite( rmesa->dri.fd, DRM_RADEON_STIPPLE, 
+                    &stipple, sizeof(drmRadeonStipple) );
+   UNLOCK_HARDWARE( rmesa );
+}
+
+static void radeonPolygonMode( GLcontext *ctx, GLenum face, GLenum mode )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLboolean flag = (ctx->_TriangleCaps & DD_TRI_UNFILLED) != 0;
+
+   /* Can't generally do unfilled via tcl, but some good special
+    * cases work. 
+    */
+   TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_UNFILLED, flag);
+   if (rmesa->TclFallback) {
+      radeonChooseRenderState( ctx );
+      radeonChooseVertexState( ctx );
+   }
+}
+
+
+/* =============================================================
+ * Rendering attributes
+ *
+ * We really don't want to recalculate all this every time we bind a
+ * texture.  These things shouldn't change all that often, so it makes
+ * sense to break them out of the core texture state update routines.
+ */
+
+/* Examine lighting and texture state to determine if separate specular
+ * should be enabled.
+ */
+static void radeonUpdateSpecular( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   CARD32 p = rmesa->hw.ctx.cmd[CTX_PP_CNTL];
+
+   RADEON_STATECHANGE( rmesa, tcl );
+
+   rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] &= ~RADEON_TCL_COMPUTE_SPECULAR;
+   rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] &= ~RADEON_TCL_COMPUTE_DIFFUSE;
+   rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] &= ~RADEON_TCL_VTX_PK_SPEC;
+   rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] &= ~RADEON_TCL_VTX_PK_DIFFUSE;
+   rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &= ~RADEON_LIGHTING_ENABLE;
+
+   p &= ~RADEON_SPECULAR_ENABLE;
+
+   rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] |= RADEON_DIFFUSE_SPECULAR_COMBINE;
+
+
+   if (ctx->Light.Enabled &&
+       ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR) {
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] |= RADEON_TCL_COMPUTE_SPECULAR;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] |= RADEON_TCL_COMPUTE_DIFFUSE;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_SPEC;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_DIFFUSE;
+      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] |= RADEON_LIGHTING_ENABLE;
+      p |=  RADEON_SPECULAR_ENABLE;
+      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &= 
+	 ~RADEON_DIFFUSE_SPECULAR_COMBINE;
+   }
+   else if (ctx->Light.Enabled) {
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] |= RADEON_TCL_COMPUTE_DIFFUSE;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_DIFFUSE;
+      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] |= RADEON_LIGHTING_ENABLE;
+   } else if (ctx->Fog.ColorSumEnabled ) {
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_SPEC;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_DIFFUSE;
+      p |= RADEON_SPECULAR_ENABLE;
+   } else {
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_DIFFUSE;
+   }
+
+   if (ctx->Fog.Enabled) {
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] |= RADEON_TCL_COMPUTE_SPECULAR;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_PK_SPEC;
+
+      /* Bizzare: have to leave lighting enabled to get fog.
+       */
+      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] |= RADEON_LIGHTING_ENABLE;
+   }
+
+   if ( ctx->_TriangleCaps & DD_SEPARATE_SPECULAR ) {
+      assert( (p & RADEON_SPECULAR_ENABLE) != 0 );
+   } else {
+      assert( (p & RADEON_SPECULAR_ENABLE) == 0 );
+   }
+
+   if ( rmesa->hw.ctx.cmd[CTX_PP_CNTL] != p ) {
+      RADEON_STATECHANGE( rmesa, ctx );
+      rmesa->hw.ctx.cmd[CTX_PP_CNTL] = p;
+   }
+
+   /* Update vertex/render formats
+    */
+   if (rmesa->TclFallback) { 
+      radeonChooseRenderState( ctx );
+      radeonChooseVertexState( ctx );
+   }
+}
+
+
+/* =============================================================
+ * Materials
+ */
+
+
+/* Update on colormaterial, material emmissive/ambient, 
+ * lightmodel.globalambient
+ */
+static void update_global_ambient( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   float *fcmd = (float *)RADEON_DB_STATE( glt );
+
+   /* Need to do more if both emmissive & ambient are PREMULT:
+    */
+   if ((rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &
+       ((3 << RADEON_EMISSIVE_SOURCE_SHIFT) |
+	(3 << RADEON_AMBIENT_SOURCE_SHIFT))) == 0) 
+   {
+      COPY_3V( &fcmd[GLT_RED], 
+	       ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_EMISSION]);
+      ACC_SCALE_3V( &fcmd[GLT_RED],
+		   ctx->Light.Model.Ambient,
+		   ctx->Light.Material.Attrib[MAT_ATTRIB_FRONT_AMBIENT]);
+   } 
+   else
+   {
+      COPY_3V( &fcmd[GLT_RED], ctx->Light.Model.Ambient );
+   }
+   
+   RADEON_DB_STATECHANGE(rmesa, &rmesa->hw.glt);
+}
+
+/* Update on change to 
+ *    - light[p].colors
+ *    - light[p].enabled
+ *    - material,
+ *    - colormaterial enabled
+ *    - colormaterial bitmask
+ */
+static void update_light_colors( GLcontext *ctx, GLuint p )
+{
+   struct gl_light *l = &ctx->Light.Light[p];
+
+/*     fprintf(stderr, "%s\n", __FUNCTION__); */
+
+   if (l->Enabled) {
+      radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+      float *fcmd = (float *)RADEON_DB_STATE( lit[p] );
+      GLuint bitmask = ctx->Light.ColorMaterialBitmask;
+      GLfloat (*mat)[4] = ctx->Light.Material.Attrib;
+
+      COPY_4V( &fcmd[LIT_AMBIENT_RED], l->Ambient );	 
+      COPY_4V( &fcmd[LIT_DIFFUSE_RED], l->Diffuse );
+      COPY_4V( &fcmd[LIT_SPECULAR_RED], l->Specular );
+      
+      if (!ctx->Light.ColorMaterialEnabled)
+	 bitmask = 0;
+
+      if ((bitmask & MAT_BIT_FRONT_AMBIENT) == 0) 
+	 SELF_SCALE_3V( &fcmd[LIT_AMBIENT_RED], mat[MAT_ATTRIB_FRONT_AMBIENT] );
+
+      if ((bitmask & MAT_BIT_FRONT_DIFFUSE) == 0) 
+	 SELF_SCALE_3V( &fcmd[LIT_DIFFUSE_RED], mat[MAT_ATTRIB_FRONT_DIFFUSE] );
+      
+      if ((bitmask & MAT_BIT_FRONT_SPECULAR) == 0) 
+	 SELF_SCALE_3V( &fcmd[LIT_SPECULAR_RED], mat[MAT_ATTRIB_FRONT_SPECULAR] );
+
+      RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.lit[p] );
+   }
+}
+
+/* Also fallback for asym colormaterial mode in twoside lighting...
+ */
+static void check_twoside_fallback( GLcontext *ctx )
+{
+   GLboolean fallback = GL_FALSE;
+   GLint i;
+
+   if (ctx->Light.Enabled && ctx->Light.Model.TwoSide) {
+      if (ctx->Light.ColorMaterialEnabled &&
+	  (ctx->Light.ColorMaterialBitmask & BACK_MATERIAL_BITS) != 
+	  ((ctx->Light.ColorMaterialBitmask & FRONT_MATERIAL_BITS)<<1))
+	 fallback = GL_TRUE;
+      else {
+	 for (i = MAT_ATTRIB_FRONT_AMBIENT; i < MAT_ATTRIB_FRONT_INDEXES; i+=2)
+	    if (memcmp( ctx->Light.Material.Attrib[i],
+			ctx->Light.Material.Attrib[i+1],
+			sizeof(GLfloat)*4) != 0) {
+	       fallback = GL_TRUE;  
+	       break;
+	    }
+      }
+   }
+
+   TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_LIGHT_TWOSIDE, fallback );
+}
+
+
+static void radeonColorMaterial( GLcontext *ctx, GLenum face, GLenum mode )
+{
+   if (ctx->Light.ColorMaterialEnabled) {
+      radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+      GLuint light_model_ctl1 = rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL];
+      GLuint mask = ctx->Light.ColorMaterialBitmask;
+
+      /* Default to PREMULT:
+       */
+      light_model_ctl1 &= ~((0xf << RADEON_EMISSIVE_SOURCE_SHIFT) |
+			   (0xf << RADEON_AMBIENT_SOURCE_SHIFT) |
+			   (0xf << RADEON_DIFFUSE_SOURCE_SHIFT) |
+			   (0xf << RADEON_SPECULAR_SOURCE_SHIFT)); 
+   
+      if (mask & MAT_BIT_FRONT_EMISSION) {
+	 light_model_ctl1 |= (3 <<
+			     RADEON_EMISSIVE_SOURCE_SHIFT);
+      }
+
+      if (mask & MAT_BIT_FRONT_AMBIENT) {
+	 light_model_ctl1 |= (3 <<
+			     RADEON_AMBIENT_SOURCE_SHIFT);
+      }
+	 
+      if (mask & MAT_BIT_FRONT_DIFFUSE) {
+	 light_model_ctl1 |= (3 <<
+			     RADEON_DIFFUSE_SOURCE_SHIFT);
+      }
+   
+      if (mask & MAT_BIT_FRONT_SPECULAR) {
+	 light_model_ctl1 |= (3 <<
+			     RADEON_SPECULAR_SOURCE_SHIFT);
+      }
+   
+      if (light_model_ctl1 != rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL]) {
+	 GLuint p;
+
+	 RADEON_STATECHANGE( rmesa, tcl );
+	 rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] = light_model_ctl1;      
+
+	 for (p = 0 ; p < MAX_LIGHTS; p++) 
+	    update_light_colors( ctx, p );
+	 update_global_ambient( ctx );
+      }
+   }
+   
+   check_twoside_fallback( ctx );
+}
+
+void radeonUpdateMaterial( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat (*mat)[4] = ctx->Light.Material.Attrib;
+   GLfloat *fcmd = (GLfloat *)RADEON_DB_STATE( mtl );
+   GLuint p;
+   GLuint mask = ~0;
+   
+   if (ctx->Light.ColorMaterialEnabled)
+      mask &= ~ctx->Light.ColorMaterialBitmask;
+
+   if (RADEON_DEBUG & DEBUG_STATE)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+      
+   if (mask & MAT_BIT_FRONT_EMISSION) {
+      fcmd[MTL_EMMISSIVE_RED]   = mat[MAT_ATTRIB_FRONT_EMISSION][0];
+      fcmd[MTL_EMMISSIVE_GREEN] = mat[MAT_ATTRIB_FRONT_EMISSION][1];
+      fcmd[MTL_EMMISSIVE_BLUE]  = mat[MAT_ATTRIB_FRONT_EMISSION][2];
+      fcmd[MTL_EMMISSIVE_ALPHA] = mat[MAT_ATTRIB_FRONT_EMISSION][3];
+   }
+   if (mask & MAT_BIT_FRONT_AMBIENT) {
+      fcmd[MTL_AMBIENT_RED]     = mat[MAT_ATTRIB_FRONT_AMBIENT][0];
+      fcmd[MTL_AMBIENT_GREEN]   = mat[MAT_ATTRIB_FRONT_AMBIENT][1];
+      fcmd[MTL_AMBIENT_BLUE]    = mat[MAT_ATTRIB_FRONT_AMBIENT][2];
+      fcmd[MTL_AMBIENT_ALPHA]   = mat[MAT_ATTRIB_FRONT_AMBIENT][3];
+   }
+   if (mask & MAT_BIT_FRONT_DIFFUSE) {
+      fcmd[MTL_DIFFUSE_RED]     = mat[MAT_ATTRIB_FRONT_DIFFUSE][0];
+      fcmd[MTL_DIFFUSE_GREEN]   = mat[MAT_ATTRIB_FRONT_DIFFUSE][1];
+      fcmd[MTL_DIFFUSE_BLUE]    = mat[MAT_ATTRIB_FRONT_DIFFUSE][2];
+      fcmd[MTL_DIFFUSE_ALPHA]   = mat[MAT_ATTRIB_FRONT_DIFFUSE][3];
+   }
+   if (mask & MAT_BIT_FRONT_SPECULAR) {
+      fcmd[MTL_SPECULAR_RED]    = mat[MAT_ATTRIB_FRONT_SPECULAR][0];
+      fcmd[MTL_SPECULAR_GREEN]  = mat[MAT_ATTRIB_FRONT_SPECULAR][1];
+      fcmd[MTL_SPECULAR_BLUE]   = mat[MAT_ATTRIB_FRONT_SPECULAR][2];
+      fcmd[MTL_SPECULAR_ALPHA]  = mat[MAT_ATTRIB_FRONT_SPECULAR][3];
+   }
+   if (mask & MAT_BIT_FRONT_SHININESS) {
+      fcmd[MTL_SHININESS]       = mat[MAT_ATTRIB_FRONT_SHININESS][0];
+   }
+
+   if (RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.mtl )) {
+      for (p = 0 ; p < MAX_LIGHTS; p++) 
+	 update_light_colors( ctx, p );
+
+      check_twoside_fallback( ctx );
+      update_global_ambient( ctx );
+   }
+   else if (RADEON_DEBUG & (DEBUG_PRIMS|DEBUG_STATE))
+      fprintf(stderr, "%s: Elided noop material call\n", __FUNCTION__);
+}
+
+/* _NEW_LIGHT
+ * _NEW_MODELVIEW
+ * _MESA_NEW_NEED_EYE_COORDS
+ *
+ * Uses derived state from mesa:
+ *       _VP_inf_norm
+ *       _h_inf_norm
+ *       _Position
+ *       _NormDirection
+ *       _ModelViewInvScale
+ *       _NeedEyeCoords
+ *       _EyeZDir
+ *
+ * which are calculated in light.c and are correct for the current
+ * lighting space (model or eye), hence dependencies on _NEW_MODELVIEW
+ * and _MESA_NEW_NEED_EYE_COORDS.  
+ */
+static void update_light( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   /* Have to check these, or have an automatic shortcircuit mechanism
+    * to remove noop statechanges. (Or just do a better job on the
+    * front end).
+    */
+   {
+      GLuint tmp = rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL];
+
+      if (ctx->_NeedEyeCoords)
+	 tmp &= ~RADEON_LIGHT_IN_MODELSPACE;
+      else
+	 tmp |= RADEON_LIGHT_IN_MODELSPACE;
+      
+
+      /* Leave this test disabled: (unexplained q3 lockup) (even with
+         new packets)
+      */
+      if (tmp != rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL]) 
+      {
+	 RADEON_STATECHANGE( rmesa, tcl );
+	 rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] = tmp;
+      }
+   }
+
+   {
+      GLfloat *fcmd = (GLfloat *)RADEON_DB_STATE( eye );
+      fcmd[EYE_X] = ctx->_EyeZDir[0];
+      fcmd[EYE_Y] = ctx->_EyeZDir[1];
+      fcmd[EYE_Z] = - ctx->_EyeZDir[2];
+      fcmd[EYE_RESCALE_FACTOR] = ctx->_ModelViewInvScale;
+      RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.eye );
+   }
+
+
+
+   if (ctx->Light.Enabled) {
+      GLint p;
+      for (p = 0 ; p < MAX_LIGHTS; p++) {
+	 if (ctx->Light.Light[p].Enabled) {
+	    struct gl_light *l = &ctx->Light.Light[p];
+	    GLfloat *fcmd = (GLfloat *)RADEON_DB_STATE( lit[p] );
+	    
+	    if (l->EyePosition[3] == 0.0) {
+	       COPY_3FV( &fcmd[LIT_POSITION_X], l->_VP_inf_norm ); 
+	       COPY_3FV( &fcmd[LIT_DIRECTION_X], l->_h_inf_norm ); 
+	       fcmd[LIT_POSITION_W] = 0;
+	       fcmd[LIT_DIRECTION_W] = 0;
+	    } else {
+	       COPY_4V( &fcmd[LIT_POSITION_X], l->_Position );
+	       fcmd[LIT_DIRECTION_X] = -l->_NormDirection[0];
+	       fcmd[LIT_DIRECTION_Y] = -l->_NormDirection[1];
+	       fcmd[LIT_DIRECTION_Z] = -l->_NormDirection[2];
+	       fcmd[LIT_DIRECTION_W] = 0;
+	    }
+
+	    RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.lit[p] );
+	 }
+      }
+   }
+}
+
+static void radeonLightfv( GLcontext *ctx, GLenum light,
+			   GLenum pname, const GLfloat *params )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLint p = light - GL_LIGHT0;
+   struct gl_light *l = &ctx->Light.Light[p];
+   GLfloat *fcmd = (GLfloat *)rmesa->hw.lit[p].cmd;
+   
+
+   switch (pname) {
+   case GL_AMBIENT:		
+   case GL_DIFFUSE:
+   case GL_SPECULAR:
+      update_light_colors( ctx, p );
+      break;
+
+   case GL_SPOT_DIRECTION: 
+      /* picked up in update_light */	
+      break;
+
+   case GL_POSITION: {
+      /* positions picked up in update_light, but can do flag here */	
+      GLuint flag;
+      GLuint idx = TCL_PER_LIGHT_CTL_0 + p/2;
+
+      /* FIXME: Set RANGE_ATTEN only when needed */
+      if (p&1) 
+	 flag = RADEON_LIGHT_1_IS_LOCAL;
+      else
+	 flag = RADEON_LIGHT_0_IS_LOCAL;
+
+      RADEON_STATECHANGE(rmesa, tcl);
+      if (l->EyePosition[3] != 0.0F)
+	 rmesa->hw.tcl.cmd[idx] |= flag;
+      else
+	 rmesa->hw.tcl.cmd[idx] &= ~flag;
+      break;
+   }
+
+   case GL_SPOT_EXPONENT:
+      RADEON_STATECHANGE(rmesa, lit[p]);
+      fcmd[LIT_SPOT_EXPONENT] = params[0];
+      break;
+
+   case GL_SPOT_CUTOFF: {
+      GLuint flag = (p&1) ? RADEON_LIGHT_1_IS_SPOT : RADEON_LIGHT_0_IS_SPOT;
+      GLuint idx = TCL_PER_LIGHT_CTL_0 + p/2;
+
+      RADEON_STATECHANGE(rmesa, lit[p]);
+      fcmd[LIT_SPOT_CUTOFF] = l->_CosCutoff;
+
+      RADEON_STATECHANGE(rmesa, tcl);
+      if (l->SpotCutoff != 180.0F)
+	 rmesa->hw.tcl.cmd[idx] |= flag;
+      else
+	 rmesa->hw.tcl.cmd[idx] &= ~flag;
+
+      break;
+   }
+
+   case GL_CONSTANT_ATTENUATION:
+      RADEON_STATECHANGE(rmesa, lit[p]);
+      fcmd[LIT_ATTEN_CONST] = params[0];
+      break;
+   case GL_LINEAR_ATTENUATION:
+      RADEON_STATECHANGE(rmesa, lit[p]);
+      fcmd[LIT_ATTEN_LINEAR] = params[0];
+      break;
+   case GL_QUADRATIC_ATTENUATION:
+      RADEON_STATECHANGE(rmesa, lit[p]);
+      fcmd[LIT_ATTEN_QUADRATIC] = params[0];
+      break;
+   default:
+      return;
+   }
+
+   /* Set RANGE_ATTEN only when needed */
+   switch (pname) {
+   case GL_POSITION:
+   case GL_LINEAR_ATTENUATION:
+   case GL_QUADRATIC_ATTENUATION:
+   {
+      GLuint flag;
+      GLuint idx = TCL_PER_LIGHT_CTL_0 + p/2;
+
+      if (p&1) 
+        flag = RADEON_LIGHT_1_ENABLE_RANGE_ATTEN;
+      else
+        flag = RADEON_LIGHT_0_ENABLE_RANGE_ATTEN;
+
+      RADEON_STATECHANGE(rmesa, tcl);
+      if (l->EyePosition[3] != 0.0F &&
+         (l->LinearAttenuation != 0.0F || l->QuadraticAttenuation != 0.0F))
+        rmesa->hw.tcl.cmd[idx] |= flag;
+      else
+        rmesa->hw.tcl.cmd[idx] &= ~flag;
+      break;
+   }
+   default:
+      break;
+   }
+}
+
+		  
+
+
+static void radeonLightModelfv( GLcontext *ctx, GLenum pname,
+				const GLfloat *param )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   switch (pname) {
+      case GL_LIGHT_MODEL_AMBIENT: 
+	 update_global_ambient( ctx );
+	 break;
+
+      case GL_LIGHT_MODEL_LOCAL_VIEWER:
+	 RADEON_STATECHANGE( rmesa, tcl );
+	 if (ctx->Light.Model.LocalViewer)
+	    rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] |= RADEON_LOCAL_VIEWER;
+	 else
+	    rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &= ~RADEON_LOCAL_VIEWER;
+         break;
+
+      case GL_LIGHT_MODEL_TWO_SIDE:
+	 RADEON_STATECHANGE( rmesa, tcl );
+	 if (ctx->Light.Model.TwoSide)
+	    rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] |= RADEON_LIGHT_TWOSIDE;
+	 else
+	    rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] &= ~RADEON_LIGHT_TWOSIDE;
+
+	 check_twoside_fallback( ctx );
+
+	 if (rmesa->TclFallback) {
+	    radeonChooseRenderState( ctx );
+	    radeonChooseVertexState( ctx );
+	 }
+         break;
+
+      case GL_LIGHT_MODEL_COLOR_CONTROL:
+	 radeonUpdateSpecular(ctx);
+         break;
+
+      default:
+         break;
+   }
+}
+
+static void radeonShadeModel( GLcontext *ctx, GLenum mode )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint s = rmesa->hw.set.cmd[SET_SE_CNTL];
+
+   s &= ~(RADEON_DIFFUSE_SHADE_MASK |
+	  RADEON_ALPHA_SHADE_MASK |
+	  RADEON_SPECULAR_SHADE_MASK |
+	  RADEON_FOG_SHADE_MASK);
+
+   switch ( mode ) {
+   case GL_FLAT:
+      s |= (RADEON_DIFFUSE_SHADE_FLAT |
+	    RADEON_ALPHA_SHADE_FLAT |
+	    RADEON_SPECULAR_SHADE_FLAT |
+	    RADEON_FOG_SHADE_FLAT);
+      break;
+   case GL_SMOOTH:
+      s |= (RADEON_DIFFUSE_SHADE_GOURAUD |
+	    RADEON_ALPHA_SHADE_GOURAUD |
+	    RADEON_SPECULAR_SHADE_GOURAUD |
+	    RADEON_FOG_SHADE_GOURAUD);
+      break;
+   default:
+      return;
+   }
+
+   if ( rmesa->hw.set.cmd[SET_SE_CNTL] != s ) {
+      RADEON_STATECHANGE( rmesa, set );
+      rmesa->hw.set.cmd[SET_SE_CNTL] = s;
+   }
+}
+
+
+/* =============================================================
+ * User clip planes
+ */
+
+static void radeonClipPlane( GLcontext *ctx, GLenum plane, const GLfloat *eq )
+{
+   GLint p = (GLint) plane - (GLint) GL_CLIP_PLANE0;
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLint *ip = (GLint *)ctx->Transform._ClipUserPlane[p];
+
+   RADEON_STATECHANGE( rmesa, ucp[p] );
+   rmesa->hw.ucp[p].cmd[UCP_X] = ip[0];
+   rmesa->hw.ucp[p].cmd[UCP_Y] = ip[1];
+   rmesa->hw.ucp[p].cmd[UCP_Z] = ip[2];
+   rmesa->hw.ucp[p].cmd[UCP_W] = ip[3];
+}
+
+static void radeonUpdateClipPlanes( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint p;
+
+   for (p = 0; p < ctx->Const.MaxClipPlanes; p++) {
+      if (ctx->Transform.ClipPlanesEnabled & (1 << p)) {
+	 GLint *ip = (GLint *)ctx->Transform._ClipUserPlane[p];
+
+	 RADEON_STATECHANGE( rmesa, ucp[p] );
+	 rmesa->hw.ucp[p].cmd[UCP_X] = ip[0];
+	 rmesa->hw.ucp[p].cmd[UCP_Y] = ip[1];
+	 rmesa->hw.ucp[p].cmd[UCP_Z] = ip[2];
+	 rmesa->hw.ucp[p].cmd[UCP_W] = ip[3];
+      }
+   }
+}
+
+
+/* =============================================================
+ * Stencil
+ */
+
+static void radeonStencilFunc( GLcontext *ctx, GLenum func,
+			       GLint ref, GLuint mask )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint refmask = ((ctx->Stencil.Ref[0] << RADEON_STENCIL_REF_SHIFT) |
+		     (ctx->Stencil.ValueMask[0] << RADEON_STENCIL_MASK_SHIFT));
+
+   RADEON_STATECHANGE( rmesa, ctx );
+   RADEON_STATECHANGE( rmesa, msk );
+
+   rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] &= ~RADEON_STENCIL_TEST_MASK;
+   rmesa->hw.msk.cmd[MSK_RB3D_STENCILREFMASK] &= ~(RADEON_STENCIL_REF_MASK|
+						   RADEON_STENCIL_VALUE_MASK);
+
+   switch ( ctx->Stencil.Function[0] ) {
+   case GL_NEVER:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_TEST_NEVER;
+      break;
+   case GL_LESS:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_TEST_LESS;
+      break;
+   case GL_EQUAL:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_TEST_EQUAL;
+      break;
+   case GL_LEQUAL:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_TEST_LEQUAL;
+      break;
+   case GL_GREATER:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_TEST_GREATER;
+      break;
+   case GL_NOTEQUAL:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_TEST_NEQUAL;
+      break;
+   case GL_GEQUAL:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_TEST_GEQUAL;
+      break;
+   case GL_ALWAYS:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_TEST_ALWAYS;
+      break;
+   }
+
+   rmesa->hw.msk.cmd[MSK_RB3D_STENCILREFMASK] |= refmask;
+}
+
+static void radeonStencilMask( GLcontext *ctx, GLuint mask )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   RADEON_STATECHANGE( rmesa, msk );
+   rmesa->hw.msk.cmd[MSK_RB3D_STENCILREFMASK] &= ~RADEON_STENCIL_WRITE_MASK;
+   rmesa->hw.msk.cmd[MSK_RB3D_STENCILREFMASK] |=
+      (ctx->Stencil.WriteMask[0] << RADEON_STENCIL_WRITEMASK_SHIFT);
+}
+
+static void radeonStencilOp( GLcontext *ctx, GLenum fail,
+			     GLenum zfail, GLenum zpass )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   RADEON_STATECHANGE( rmesa, ctx );
+   rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] &= ~(RADEON_STENCIL_FAIL_MASK |
+					       RADEON_STENCIL_ZFAIL_MASK |
+					       RADEON_STENCIL_ZPASS_MASK);
+
+   switch ( ctx->Stencil.FailFunc[0] ) {
+   case GL_KEEP:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_FAIL_KEEP;
+      break;
+   case GL_ZERO:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_FAIL_ZERO;
+      break;
+   case GL_REPLACE:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_FAIL_REPLACE;
+      break;
+   case GL_INCR:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_FAIL_INC;
+      break;
+   case GL_DECR:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_FAIL_DEC;
+      break;
+   case GL_INVERT:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_FAIL_INVERT;
+      break;
+   }
+
+   switch ( ctx->Stencil.ZFailFunc[0] ) {
+   case GL_KEEP:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_ZFAIL_KEEP;
+      break;
+   case GL_ZERO:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_ZFAIL_ZERO;
+      break;
+   case GL_REPLACE:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_ZFAIL_REPLACE;
+      break;
+   case GL_INCR:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_ZFAIL_INC;
+      break;
+   case GL_DECR:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_ZFAIL_DEC;
+      break;
+   case GL_INVERT:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_ZFAIL_INVERT;
+      break;
+   }
+
+   switch ( ctx->Stencil.ZPassFunc[0] ) {
+   case GL_KEEP:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_ZPASS_KEEP;
+      break;
+   case GL_ZERO:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_ZPASS_ZERO;
+      break;
+   case GL_REPLACE:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_ZPASS_REPLACE;
+      break;
+   case GL_INCR:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_ZPASS_INC;
+      break;
+   case GL_DECR:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_ZPASS_DEC;
+      break;
+   case GL_INVERT:
+      rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] |= RADEON_STENCIL_ZPASS_INVERT;
+      break;
+   }
+}
+
+static void radeonClearStencil( GLcontext *ctx, GLint s )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   rmesa->state.stencil.clear = 
+      ((GLuint) ctx->Stencil.Clear |
+       (0xff << RADEON_STENCIL_MASK_SHIFT) |
+       (ctx->Stencil.WriteMask[0] << RADEON_STENCIL_WRITEMASK_SHIFT));
+}
+
+
+/* =============================================================
+ * Window position and viewport transformation
+ */
+
+/*
+ * To correctly position primitives:
+ */
+#define SUBPIXEL_X 0.125
+#define SUBPIXEL_Y 0.125
+
+void radeonUpdateWindow( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
+   GLfloat xoffset = (GLfloat)dPriv->x;
+   GLfloat yoffset = (GLfloat)dPriv->y + dPriv->h;
+   const GLfloat *v = ctx->Viewport._WindowMap.m;
+
+   GLfloat sx = v[MAT_SX];
+   GLfloat tx = v[MAT_TX] + xoffset + SUBPIXEL_X;
+   GLfloat sy = - v[MAT_SY];
+   GLfloat ty = (- v[MAT_TY]) + yoffset + SUBPIXEL_Y;
+   GLfloat sz = v[MAT_SZ] * rmesa->state.depth.scale;
+   GLfloat tz = v[MAT_TZ] * rmesa->state.depth.scale;
+   RADEON_FIREVERTICES( rmesa );
+   RADEON_STATECHANGE( rmesa, vpt );
+
+   rmesa->hw.vpt.cmd[VPT_SE_VPORT_XSCALE]  = *(GLuint *)&sx;
+   rmesa->hw.vpt.cmd[VPT_SE_VPORT_XOFFSET] = *(GLuint *)&tx;
+   rmesa->hw.vpt.cmd[VPT_SE_VPORT_YSCALE]  = *(GLuint *)&sy;
+   rmesa->hw.vpt.cmd[VPT_SE_VPORT_YOFFSET] = *(GLuint *)&ty;
+   rmesa->hw.vpt.cmd[VPT_SE_VPORT_ZSCALE]  = *(GLuint *)&sz;
+   rmesa->hw.vpt.cmd[VPT_SE_VPORT_ZOFFSET] = *(GLuint *)&tz;
+}
+
+
+
+static void radeonViewport( GLcontext *ctx, GLint x, GLint y,
+			    GLsizei width, GLsizei height )
+{
+   /* Don't pipeline viewport changes, conflict with window offset
+    * setting below.  Could apply deltas to rescue pipelined viewport
+    * values, or keep the originals hanging around.
+    */
+   RADEON_FIREVERTICES( RADEON_CONTEXT(ctx) );
+   radeonUpdateWindow( ctx );
+}
+
+static void radeonDepthRange( GLcontext *ctx, GLclampd nearval,
+			      GLclampd farval )
+{
+   radeonUpdateWindow( ctx );
+}
+
+void radeonUpdateViewportOffset( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
+   GLfloat xoffset = (GLfloat)dPriv->x;
+   GLfloat yoffset = (GLfloat)dPriv->y + dPriv->h;
+   const GLfloat *v = ctx->Viewport._WindowMap.m;
+
+   GLfloat tx = v[MAT_TX] + xoffset;
+   GLfloat ty = (- v[MAT_TY]) + yoffset;
+
+   if ( rmesa->hw.vpt.cmd[VPT_SE_VPORT_XOFFSET] != *(GLuint *)&tx ||
+	rmesa->hw.vpt.cmd[VPT_SE_VPORT_YOFFSET] != *(GLuint *)&ty )
+   {
+      /* Note: this should also modify whatever data the context reset
+       * code uses...
+       */
+      rmesa->hw.vpt.cmd[VPT_SE_VPORT_XOFFSET] = *(GLuint *)&tx;
+      rmesa->hw.vpt.cmd[VPT_SE_VPORT_YOFFSET] = *(GLuint *)&ty;
+      
+      /* update polygon stipple x/y screen offset */
+      {
+         GLuint stx, sty;
+         GLuint m = rmesa->hw.msc.cmd[MSC_RE_MISC];
+
+         m &= ~(RADEON_STIPPLE_X_OFFSET_MASK |
+                RADEON_STIPPLE_Y_OFFSET_MASK);
+
+         /* add magic offsets, then invert */
+         stx = 31 - ((rmesa->dri.drawable->x - 1) & RADEON_STIPPLE_COORD_MASK);
+         sty = 31 - ((rmesa->dri.drawable->y + rmesa->dri.drawable->h - 1)
+                     & RADEON_STIPPLE_COORD_MASK);
+
+         m |= ((stx << RADEON_STIPPLE_X_OFFSET_SHIFT) |
+               (sty << RADEON_STIPPLE_Y_OFFSET_SHIFT));
+
+         if ( rmesa->hw.msc.cmd[MSC_RE_MISC] != m ) {
+            RADEON_STATECHANGE( rmesa, msc );
+	    rmesa->hw.msc.cmd[MSC_RE_MISC] = m;
+         }
+      }
+   }
+
+   radeonUpdateScissor( ctx );
+}
+
+
+
+/* =============================================================
+ * Miscellaneous
+ */
+
+static void radeonClearColor( GLcontext *ctx, const GLfloat color[4] )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLubyte c[4];
+   CLAMPED_FLOAT_TO_UBYTE(c[0], color[0]);
+   CLAMPED_FLOAT_TO_UBYTE(c[1], color[1]);
+   CLAMPED_FLOAT_TO_UBYTE(c[2], color[2]);
+   CLAMPED_FLOAT_TO_UBYTE(c[3], color[3]);
+   rmesa->state.color.clear = radeonPackColor( rmesa->radeonScreen->cpp,
+					       c[0], c[1], c[2], c[3] );
+}
+
+
+static void radeonRenderMode( GLcontext *ctx, GLenum mode )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   FALLBACK( rmesa, RADEON_FALLBACK_RENDER_MODE, (mode != GL_RENDER) );
+}
+
+
+static GLuint radeon_rop_tab[] = {
+   RADEON_ROP_CLEAR,
+   RADEON_ROP_AND,
+   RADEON_ROP_AND_REVERSE,
+   RADEON_ROP_COPY,
+   RADEON_ROP_AND_INVERTED,
+   RADEON_ROP_NOOP,
+   RADEON_ROP_XOR,
+   RADEON_ROP_OR,
+   RADEON_ROP_NOR,
+   RADEON_ROP_EQUIV,
+   RADEON_ROP_INVERT,
+   RADEON_ROP_OR_REVERSE,
+   RADEON_ROP_COPY_INVERTED,
+   RADEON_ROP_OR_INVERTED,
+   RADEON_ROP_NAND,
+   RADEON_ROP_SET,
+};
+
+static void radeonLogicOpCode( GLcontext *ctx, GLenum opcode )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint rop = (GLuint)opcode - GL_CLEAR;
+
+   ASSERT( rop < 16 );
+
+   RADEON_STATECHANGE( rmesa, msk );
+   rmesa->hw.msk.cmd[MSK_RB3D_ROPCNTL] = radeon_rop_tab[rop];
+}
+
+
+void radeonSetCliprects( radeonContextPtr rmesa, GLenum mode )
+{
+   __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
+
+   switch ( mode ) {
+   case GL_FRONT_LEFT:
+      rmesa->numClipRects = dPriv->numClipRects;
+      rmesa->pClipRects = (XF86DRIClipRectPtr)dPriv->pClipRects;
+      break;
+   case GL_BACK_LEFT:
+      /* Can't ignore 2d windows if we are page flipping.
+       */
+      if ( dPriv->numBackClipRects == 0 || rmesa->doPageFlip ) {
+	 rmesa->numClipRects = dPriv->numClipRects;
+	 rmesa->pClipRects = (XF86DRIClipRectPtr)dPriv->pClipRects;
+      }
+      else {
+	 rmesa->numClipRects = dPriv->numBackClipRects;
+	 rmesa->pClipRects = (XF86DRIClipRectPtr)dPriv->pBackClipRects;
+      }
+      break;
+   default:
+      fprintf(stderr, "bad mode in radeonSetCliprects\n");
+      return;
+   }
+
+   if (rmesa->state.scissor.enabled)
+      radeonRecalcScissorRects( rmesa );
+}
+
+
+static void radeonDrawBuffer( GLcontext *ctx, GLenum mode )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if (RADEON_DEBUG & DEBUG_DRI)
+      fprintf(stderr, "%s %s\n", __FUNCTION__,
+	      _mesa_lookup_enum_by_nr( mode ));
+
+   RADEON_FIREVERTICES(rmesa);	/* don't pipeline cliprect changes */
+
+   /*
+    * _DrawDestMask is easier to cope with than <mode>.
+    */
+   switch ( ctx->Color._DrawDestMask ) {
+   case FRONT_LEFT_BIT:
+      FALLBACK( rmesa, RADEON_FALLBACK_DRAW_BUFFER, GL_FALSE );
+      radeonSetCliprects( rmesa, GL_FRONT_LEFT );
+      break;
+   case BACK_LEFT_BIT:
+      FALLBACK( rmesa, RADEON_FALLBACK_DRAW_BUFFER, GL_FALSE );
+      radeonSetCliprects( rmesa, GL_BACK_LEFT );
+      break;
+   default:
+      /* GL_NONE or GL_FRONT_AND_BACK or stereo left&right, etc */
+      FALLBACK( rmesa, RADEON_FALLBACK_DRAW_BUFFER, GL_TRUE );
+      return;
+   }
+
+   /* We want to update the s/w rast state too so that r200SetBuffer()
+    * gets called.
+    */
+   _swrast_DrawBuffer(ctx, mode);
+
+   RADEON_STATECHANGE( rmesa, ctx );
+   rmesa->hw.ctx.cmd[CTX_RB3D_COLOROFFSET] = (rmesa->state.color.drawOffset &
+					    RADEON_COLOROFFSET_MASK);
+   rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] = rmesa->state.color.drawPitch;
+}
+
+static void radeonReadBuffer( GLcontext *ctx, GLenum mode )
+{
+   /* nothing, until we implement h/w glRead/CopyPixels or CopyTexImage */
+}
+
+
+/* =============================================================
+ * State enable/disable
+ */
+
+static void radeonEnable( GLcontext *ctx, GLenum cap, GLboolean state )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint p, flag;
+
+   if ( RADEON_DEBUG & DEBUG_STATE )
+      fprintf( stderr, "%s( %s = %s )\n", __FUNCTION__,
+	       _mesa_lookup_enum_by_nr( cap ),
+	       state ? "GL_TRUE" : "GL_FALSE" );
+
+   switch ( cap ) {
+      /* Fast track this one...
+       */
+   case GL_TEXTURE_1D:
+   case GL_TEXTURE_2D:
+   case GL_TEXTURE_3D:
+      break;
+
+   case GL_ALPHA_TEST:
+      RADEON_STATECHANGE( rmesa, ctx );
+      if (state) {
+	 rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= RADEON_ALPHA_TEST_ENABLE;
+      } else {
+	 rmesa->hw.ctx.cmd[CTX_PP_CNTL] &= ~RADEON_ALPHA_TEST_ENABLE;
+      }
+      break;
+
+   case GL_BLEND:
+      RADEON_STATECHANGE( rmesa, ctx );
+      if (state) {
+	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  RADEON_ALPHA_BLEND_ENABLE;
+      } else {
+	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~RADEON_ALPHA_BLEND_ENABLE;
+      }
+      if ( ctx->Color.ColorLogicOpEnabled ) {
+	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  RADEON_ROP_ENABLE;
+      } else {
+	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~RADEON_ROP_ENABLE;
+      }
+
+      /* Catch a possible fallback:
+       */
+      if (state) {
+	 ctx->Driver.BlendEquation( ctx, ctx->Color.BlendEquation );
+	 ctx->Driver.BlendFunc( ctx, ctx->Color.BlendSrcRGB,
+				ctx->Color.BlendDstRGB );
+      }
+      else {
+	 FALLBACK( rmesa, RADEON_FALLBACK_BLEND_FUNC, GL_FALSE );
+	 FALLBACK( rmesa, RADEON_FALLBACK_BLEND_EQ, GL_FALSE );
+      }
+      break;
+
+   case GL_CLIP_PLANE0:
+   case GL_CLIP_PLANE1:
+   case GL_CLIP_PLANE2:
+   case GL_CLIP_PLANE3:
+   case GL_CLIP_PLANE4:
+   case GL_CLIP_PLANE5: 
+      p = cap-GL_CLIP_PLANE0;
+      RADEON_STATECHANGE( rmesa, tcl );
+      if (state) {
+	 rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] |= (RADEON_UCP_ENABLE_0<<p);
+	 radeonClipPlane( ctx, cap, NULL );
+      }
+      else {
+	 rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] &= ~(RADEON_UCP_ENABLE_0<<p);
+      }
+      break;
+
+   case GL_COLOR_MATERIAL:
+      radeonColorMaterial( ctx, 0, 0 );
+      if (!state) 
+	 radeonUpdateMaterial( ctx );
+      break;
+
+   case GL_CULL_FACE:
+      radeonCullFace( ctx, 0 );
+      break;
+
+   case GL_DEPTH_TEST:
+      RADEON_STATECHANGE(rmesa, ctx );
+      if ( state ) {
+	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  RADEON_Z_ENABLE;
+      } else {
+	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~RADEON_Z_ENABLE;
+      }
+      break;
+
+   case GL_DITHER:
+      RADEON_STATECHANGE(rmesa, ctx );
+      if ( state ) {
+	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  RADEON_DITHER_ENABLE;
+      } else {
+	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~RADEON_DITHER_ENABLE;
+      }
+      break;
+
+   case GL_FOG:
+      RADEON_STATECHANGE(rmesa, ctx );
+      if ( state ) {
+	 rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= RADEON_FOG_ENABLE;
+	 radeonFogfv( ctx, GL_FOG_MODE, 0 );
+      } else {
+	 rmesa->hw.ctx.cmd[CTX_PP_CNTL] &= ~RADEON_FOG_ENABLE;
+	 RADEON_STATECHANGE(rmesa, tcl);
+	 rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] &= ~RADEON_TCL_FOG_MASK;
+      }
+      radeonUpdateSpecular( ctx ); /* for PK_SPEC */
+      if (rmesa->TclFallback) 
+	 radeonChooseVertexState( ctx );
+      _mesa_allow_light_in_model( ctx, !state );
+      break;
+
+   case GL_LIGHT0:
+   case GL_LIGHT1:
+   case GL_LIGHT2:
+   case GL_LIGHT3:
+   case GL_LIGHT4:
+   case GL_LIGHT5:
+   case GL_LIGHT6:
+   case GL_LIGHT7:
+      RADEON_STATECHANGE(rmesa, tcl);
+      p = cap - GL_LIGHT0;
+      if (p&1) 
+	 flag = (RADEON_LIGHT_1_ENABLE |
+		 RADEON_LIGHT_1_ENABLE_AMBIENT | 
+		 RADEON_LIGHT_1_ENABLE_SPECULAR);
+      else
+	 flag = (RADEON_LIGHT_0_ENABLE |
+		 RADEON_LIGHT_0_ENABLE_AMBIENT | 
+		 RADEON_LIGHT_0_ENABLE_SPECULAR);
+
+      if (state)
+	 rmesa->hw.tcl.cmd[p/2 + TCL_PER_LIGHT_CTL_0] |= flag;
+      else
+	 rmesa->hw.tcl.cmd[p/2 + TCL_PER_LIGHT_CTL_0] &= ~flag;
+
+      /* 
+       */
+      update_light_colors( ctx, p );
+      break;
+
+   case GL_LIGHTING:
+      RADEON_STATECHANGE(rmesa, tcl);
+      radeonUpdateSpecular(ctx);
+      check_twoside_fallback( ctx );
+      break;
+
+   case GL_LINE_SMOOTH:
+      RADEON_STATECHANGE( rmesa, ctx );
+      if ( state ) {
+	 rmesa->hw.ctx.cmd[CTX_PP_CNTL] |=  RADEON_ANTI_ALIAS_LINE;
+      } else {
+	 rmesa->hw.ctx.cmd[CTX_PP_CNTL] &= ~RADEON_ANTI_ALIAS_LINE;
+      }
+      break;
+
+   case GL_LINE_STIPPLE:
+      RADEON_STATECHANGE( rmesa, ctx );
+      if ( state ) {
+	 rmesa->hw.ctx.cmd[CTX_PP_CNTL] |=  RADEON_PATTERN_ENABLE;
+      } else {
+	 rmesa->hw.ctx.cmd[CTX_PP_CNTL] &= ~RADEON_PATTERN_ENABLE;
+      }
+      break;
+
+   case GL_COLOR_LOGIC_OP:
+      RADEON_STATECHANGE( rmesa, ctx );
+      if ( state ) {
+	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  RADEON_ROP_ENABLE;
+      } else {
+	 rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~RADEON_ROP_ENABLE;
+      }
+      break;
+      
+   case GL_NORMALIZE:
+      RADEON_STATECHANGE( rmesa, tcl );
+      if ( state ) {
+	 rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] |=  RADEON_NORMALIZE_NORMALS;
+      } else {
+	 rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &= ~RADEON_NORMALIZE_NORMALS;
+      }
+      break;
+
+   case GL_POLYGON_OFFSET_POINT:
+      if (rmesa->dri.drmMinor == 1) {
+	 radeonChooseRenderState( ctx );
+      } 
+      else {
+	 RADEON_STATECHANGE( rmesa, set );
+	 if ( state ) {
+	    rmesa->hw.set.cmd[SET_SE_CNTL] |=  RADEON_ZBIAS_ENABLE_POINT;
+	 } else {
+	    rmesa->hw.set.cmd[SET_SE_CNTL] &= ~RADEON_ZBIAS_ENABLE_POINT;
+	 }
+      }
+      break;
+
+   case GL_POLYGON_OFFSET_LINE:
+      if (rmesa->dri.drmMinor == 1) {
+	 radeonChooseRenderState( ctx );
+      } 
+      else {
+	 RADEON_STATECHANGE( rmesa, set );
+	 if ( state ) {
+	    rmesa->hw.set.cmd[SET_SE_CNTL] |=  RADEON_ZBIAS_ENABLE_LINE;
+	 } else {
+	    rmesa->hw.set.cmd[SET_SE_CNTL] &= ~RADEON_ZBIAS_ENABLE_LINE;
+	 }
+      }
+      break;
+
+   case GL_POLYGON_OFFSET_FILL:
+      if (rmesa->dri.drmMinor == 1) {
+	 radeonChooseRenderState( ctx );
+      } 
+      else {
+	 RADEON_STATECHANGE( rmesa, set );
+	 if ( state ) {
+	    rmesa->hw.set.cmd[SET_SE_CNTL] |=  RADEON_ZBIAS_ENABLE_TRI;
+	 } else {
+	    rmesa->hw.set.cmd[SET_SE_CNTL] &= ~RADEON_ZBIAS_ENABLE_TRI;
+	 }
+      }
+      break;
+
+   case GL_POLYGON_SMOOTH:
+      RADEON_STATECHANGE( rmesa, ctx );
+      if ( state ) {
+	 rmesa->hw.ctx.cmd[CTX_PP_CNTL] |=  RADEON_ANTI_ALIAS_POLY;
+      } else {
+	 rmesa->hw.ctx.cmd[CTX_PP_CNTL] &= ~RADEON_ANTI_ALIAS_POLY;
+      }
+      break;
+
+   case GL_POLYGON_STIPPLE:
+      RADEON_STATECHANGE(rmesa, ctx );
+      if ( state ) {
+	 rmesa->hw.ctx.cmd[CTX_PP_CNTL] |=  RADEON_STIPPLE_ENABLE;
+      } else {
+	 rmesa->hw.ctx.cmd[CTX_PP_CNTL] &= ~RADEON_STIPPLE_ENABLE;
+      }
+      break;
+
+   case GL_RESCALE_NORMAL_EXT: {
+      GLboolean tmp = ctx->_NeedEyeCoords ? state : !state;
+      RADEON_STATECHANGE( rmesa, tcl );
+      if ( tmp ) {
+	 rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] |=  RADEON_RESCALE_NORMALS;
+      } else {
+	 rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &= ~RADEON_RESCALE_NORMALS;
+      }
+      break;
+   }
+
+   case GL_SCISSOR_TEST:
+      RADEON_FIREVERTICES( rmesa );
+      rmesa->state.scissor.enabled = state;
+      radeonUpdateScissor( ctx );
+      break;
+
+   case GL_STENCIL_TEST:
+      if ( rmesa->state.stencil.hwBuffer ) {
+	 RADEON_STATECHANGE( rmesa, ctx );
+	 if ( state ) {
+	    rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  RADEON_STENCIL_ENABLE;
+	 } else {
+	    rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &= ~RADEON_STENCIL_ENABLE;
+	 }
+      } else {
+	 FALLBACK( rmesa, RADEON_FALLBACK_STENCIL, state );
+      }
+      break;
+
+   case GL_TEXTURE_GEN_Q:
+   case GL_TEXTURE_GEN_R:
+   case GL_TEXTURE_GEN_S:
+   case GL_TEXTURE_GEN_T:
+      /* Picked up in radeonUpdateTextureState.
+       */
+      rmesa->recheck_texgen[ctx->Texture.CurrentUnit] = GL_TRUE; 
+      break;
+
+   case GL_COLOR_SUM_EXT:
+      radeonUpdateSpecular ( ctx );
+      break;
+
+   default:
+      return;
+   }
+}
+
+
+static void radeonLightingSpaceChange( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLboolean tmp;
+   RADEON_STATECHANGE( rmesa, tcl );
+
+   if (RADEON_DEBUG & DEBUG_STATE)
+      fprintf(stderr, "%s %d BEFORE %x\n", __FUNCTION__, ctx->_NeedEyeCoords,
+	      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL]);
+
+   if (ctx->_NeedEyeCoords)
+      tmp = ctx->Transform.RescaleNormals;
+   else
+      tmp = !ctx->Transform.RescaleNormals;
+
+   if ( tmp ) {
+      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] |=  RADEON_RESCALE_NORMALS;
+   } else {
+      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] &= ~RADEON_RESCALE_NORMALS;
+   }
+
+   if (RADEON_DEBUG & DEBUG_STATE) 
+      fprintf(stderr, "%s %d AFTER %x\n", __FUNCTION__, ctx->_NeedEyeCoords,
+	      rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL]);
+}
+
+/* =============================================================
+ * Deferred state management - matrices, textures, other?
+ */
+
+
+
+
+static void upload_matrix( radeonContextPtr rmesa, GLfloat *src, int idx )
+{
+   float *dest = ((float *)RADEON_DB_STATE( mat[idx] ))+MAT_ELT_0;
+   int i;
+
+
+   for (i = 0 ; i < 4 ; i++) {
+      *dest++ = src[i];
+      *dest++ = src[i+4];
+      *dest++ = src[i+8];
+      *dest++ = src[i+12];
+   }
+
+   RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.mat[idx] );
+}
+
+static void upload_matrix_t( radeonContextPtr rmesa, GLfloat *src, int idx )
+{
+   float *dest = ((float *)RADEON_DB_STATE( mat[idx] ))+MAT_ELT_0;
+   memcpy(dest, src, 16*sizeof(float));
+   RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.mat[idx] );
+}
+
+
+static void update_texturematrix( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+   GLuint tpc = rmesa->hw.tcl.cmd[TCL_TEXTURE_PROC_CTL];
+   GLuint vs = rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL];
+   int unit;
+
+   rmesa->TexMatEnabled = 0;
+
+   for (unit = 0 ; unit < 2; unit++) {
+      if (!ctx->Texture.Unit[unit]._ReallyEnabled) {
+      }
+      else if (ctx->TextureMatrixStack[unit].Top->type != MATRIX_IDENTITY) {
+	 GLuint inputshift = RADEON_TEXGEN_0_INPUT_SHIFT + unit*4;
+	 
+	 rmesa->TexMatEnabled |= (RADEON_TEXGEN_TEXMAT_0_ENABLE|
+				  RADEON_TEXMAT_0_ENABLE) << unit;
+
+	 if (rmesa->TexGenEnabled & (RADEON_TEXMAT_0_ENABLE << unit)) {
+	    /* Need to preconcatenate any active texgen 
+	     * obj/eyeplane matrices:
+	     */
+	    _math_matrix_mul_matrix( &rmesa->tmpmat, 
+				     &rmesa->TexGenMatrix[unit],
+				     ctx->TextureMatrixStack[unit].Top );
+	    upload_matrix( rmesa, rmesa->tmpmat.m, TEXMAT_0+unit );
+	 } 
+	 else {
+	    rmesa->TexMatEnabled |= 
+	       (RADEON_TEXGEN_INPUT_TEXCOORD_0+unit) << inputshift;
+	    upload_matrix( rmesa, ctx->TextureMatrixStack[unit].Top->m, 
+			   TEXMAT_0+unit );
+	 }
+      }
+      else if (rmesa->TexGenEnabled & (RADEON_TEXMAT_0_ENABLE << unit)) {
+	 upload_matrix( rmesa, rmesa->TexGenMatrix[unit].m, 
+			TEXMAT_0+unit );
+      }
+   }
+
+
+   tpc = (rmesa->TexMatEnabled | rmesa->TexGenEnabled);
+
+   vs &= ~((0xf << RADEON_TCL_TEX_0_OUTPUT_SHIFT) |
+	   (0xf << RADEON_TCL_TEX_1_OUTPUT_SHIFT));
+
+   if (tpc & RADEON_TEXGEN_TEXMAT_0_ENABLE)
+      vs |= RADEON_TCL_TEX_COMPUTED_TEX_0 << RADEON_TCL_TEX_0_OUTPUT_SHIFT;
+   else
+      vs |= RADEON_TCL_TEX_INPUT_TEX_0 << RADEON_TCL_TEX_0_OUTPUT_SHIFT;
+
+   if (tpc & RADEON_TEXGEN_TEXMAT_1_ENABLE)
+      vs |= RADEON_TCL_TEX_COMPUTED_TEX_1 << RADEON_TCL_TEX_1_OUTPUT_SHIFT;
+   else
+      vs |= RADEON_TCL_TEX_INPUT_TEX_1 << RADEON_TCL_TEX_1_OUTPUT_SHIFT;
+
+   if (tpc != rmesa->hw.tcl.cmd[TCL_TEXTURE_PROC_CTL] ||
+       vs != rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL]) {
+      
+      RADEON_STATECHANGE(rmesa, tcl);
+      rmesa->hw.tcl.cmd[TCL_TEXTURE_PROC_CTL] = tpc;
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] = vs;
+   }
+}
+
+
+
+void radeonValidateState( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint new_state = rmesa->NewGLState;
+
+   if (new_state & _NEW_TEXTURE) {
+      radeonUpdateTextureState( ctx );
+      new_state |= rmesa->NewGLState; /* may add TEXTURE_MATRIX */
+   }
+
+   /* Need an event driven matrix update?
+    */
+   if (new_state & (_NEW_MODELVIEW|_NEW_PROJECTION)) 
+      upload_matrix( rmesa, ctx->_ModelProjectMatrix.m, MODEL_PROJ );
+
+   /* Need these for lighting (shouldn't upload otherwise)
+    */
+   if (new_state & (_NEW_MODELVIEW)) {
+      upload_matrix( rmesa, ctx->ModelviewMatrixStack.Top->m, MODEL );
+      upload_matrix_t( rmesa, ctx->ModelviewMatrixStack.Top->inv, MODEL_IT );
+   }
+
+   /* Does this need to be triggered on eg. modelview for
+    * texgen-derived objplane/eyeplane matrices?
+    */
+   if (new_state & _NEW_TEXTURE_MATRIX) {
+      update_texturematrix( ctx );
+   }      
+
+   if (new_state & (_NEW_LIGHT|_NEW_MODELVIEW|_MESA_NEW_NEED_EYE_COORDS)) {
+      update_light( ctx );
+   }
+
+   /* emit all active clip planes if projection matrix changes.
+    */
+   if (new_state & (_NEW_PROJECTION)) {
+      if (ctx->Transform.ClipPlanesEnabled) 
+	 radeonUpdateClipPlanes( ctx );
+   }
+
+
+   rmesa->NewGLState = 0;
+}
+
+
+static void radeonInvalidateState( GLcontext *ctx, GLuint new_state )
+{
+   _swrast_InvalidateState( ctx, new_state );
+   _swsetup_InvalidateState( ctx, new_state );
+   _ac_InvalidateState( ctx, new_state );
+   _tnl_InvalidateState( ctx, new_state );
+   _ae_invalidate_state( ctx, new_state );
+   RADEON_CONTEXT(ctx)->NewGLState |= new_state;
+   radeonVtxfmtInvalidate( ctx );
+}
+
+static void radeonWrapRunPipeline( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+
+   if (0)
+      fprintf(stderr, "%s, newstate: %x\n", __FUNCTION__, rmesa->NewGLState);
+
+   /* Validate state:
+    */
+   if (rmesa->NewGLState)
+      radeonValidateState( ctx );
+
+   if (tnl->vb.Material) {
+      TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_MATERIAL, GL_TRUE );
+   }
+
+   /* Run the pipeline.
+    */ 
+   _tnl_run_pipeline( ctx );
+
+   if (tnl->vb.Material) {
+      TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_MATERIAL, GL_FALSE );
+      radeonUpdateMaterial( ctx ); /* not needed any more? */
+   }
+}
+
+
+/* Initialize the driver's state functions.
+ */
+void radeonInitStateFuncs( GLcontext *ctx )
+{
+   ctx->Driver.UpdateState		= radeonInvalidateState;
+   ctx->Driver.LightingSpaceChange      = radeonLightingSpaceChange;
+
+   ctx->Driver.DrawBuffer		= radeonDrawBuffer;
+   ctx->Driver.ReadBuffer		= radeonReadBuffer;
+
+   ctx->Driver.AlphaFunc		= radeonAlphaFunc;
+   ctx->Driver.BlendEquation		= radeonBlendEquation;
+   ctx->Driver.BlendFunc		= radeonBlendFunc;
+   ctx->Driver.BlendFuncSeparate	= radeonBlendFuncSeparate;
+   ctx->Driver.ClearColor		= radeonClearColor;
+   ctx->Driver.ClearDepth		= radeonClearDepth;
+   ctx->Driver.ClearIndex		= NULL;
+   ctx->Driver.ClearStencil		= radeonClearStencil;
+   ctx->Driver.ClipPlane		= radeonClipPlane;
+   ctx->Driver.ColorMask		= radeonColorMask;
+   ctx->Driver.CullFace			= radeonCullFace;
+   ctx->Driver.DepthFunc		= radeonDepthFunc;
+   ctx->Driver.DepthMask		= radeonDepthMask;
+   ctx->Driver.DepthRange		= radeonDepthRange;
+   ctx->Driver.Enable			= radeonEnable;
+   ctx->Driver.Fogfv			= radeonFogfv;
+   ctx->Driver.FrontFace		= radeonFrontFace;
+   ctx->Driver.Hint			= NULL;
+   ctx->Driver.IndexMask		= NULL;
+   ctx->Driver.LightModelfv		= radeonLightModelfv;
+   ctx->Driver.Lightfv			= radeonLightfv;
+   ctx->Driver.LineStipple              = radeonLineStipple;
+   ctx->Driver.LineWidth                = radeonLineWidth;
+   ctx->Driver.LogicOpcode		= radeonLogicOpCode;
+   ctx->Driver.PolygonMode		= radeonPolygonMode;
+
+   if (RADEON_CONTEXT(ctx)->dri.drmMinor > 1)
+      ctx->Driver.PolygonOffset		= radeonPolygonOffset;
+
+   ctx->Driver.PolygonStipple		= radeonPolygonStipple;
+   ctx->Driver.RenderMode		= radeonRenderMode;
+   ctx->Driver.Scissor			= radeonScissor;
+   ctx->Driver.ShadeModel		= radeonShadeModel;
+   ctx->Driver.StencilFunc		= radeonStencilFunc;
+   ctx->Driver.StencilMask		= radeonStencilMask;
+   ctx->Driver.StencilOp		= radeonStencilOp;
+   ctx->Driver.Viewport			= radeonViewport;
+
+   /* Pixel path fallbacks
+    */
+   ctx->Driver.Accum                    = _swrast_Accum;
+   ctx->Driver.Bitmap                   = _swrast_Bitmap;
+   ctx->Driver.CopyPixels               = _swrast_CopyPixels;
+   ctx->Driver.DrawPixels               = _swrast_DrawPixels;
+   ctx->Driver.ReadPixels               = _swrast_ReadPixels;
+
+   /* Swrast hooks for imaging extensions:
+    */
+   ctx->Driver.CopyColorTable		= _swrast_CopyColorTable;
+   ctx->Driver.CopyColorSubTable	= _swrast_CopyColorSubTable;
+   ctx->Driver.CopyConvolutionFilter1D	= _swrast_CopyConvolutionFilter1D;
+   ctx->Driver.CopyConvolutionFilter2D	= _swrast_CopyConvolutionFilter2D;
+
+   TNL_CONTEXT(ctx)->Driver.NotifyMaterialChange = radeonUpdateMaterial;
+   TNL_CONTEXT(ctx)->Driver.RunPipeline = radeonWrapRunPipeline;
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_state.h b/src/mesa/drivers/dri/radeon/radeon_state.h
new file mode 100644
index 00000000000..a181d427e6d
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_state.h
@@ -0,0 +1,77 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_state.h,v 1.5 2002/11/05 17:46:09 tsi Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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 (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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ *
+ */
+
+#ifndef __RADEON_STATE_H__
+#define __RADEON_STATE_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+#include "radeon_context.h"
+
+extern void radeonInitState( radeonContextPtr rmesa );
+extern void radeonInitStateFuncs( GLcontext *ctx );
+
+extern void radeonUpdateMaterial( GLcontext *ctx );
+
+extern void radeonSetCliprects( radeonContextPtr rmesa, GLenum mode );
+extern void radeonRecalcScissorRects( radeonContextPtr rmesa );
+extern void radeonUpdateViewportOffset( GLcontext *ctx );
+extern void radeonUpdateWindow( GLcontext *ctx );
+
+extern void radeonValidateState( GLcontext *ctx );
+
+extern void radeonPrintDirty( radeonContextPtr rmesa,
+			      const char *msg );
+
+
+extern void radeonFallback( GLcontext *ctx, GLuint bit, GLboolean mode );
+#define FALLBACK( rmesa, bit, mode ) do {				\
+   if ( 0 ) fprintf( stderr, "FALLBACK in %s: #%d=%d\n",		\
+		     __FUNCTION__, bit, mode );				\
+   radeonFallback( rmesa->glCtx, bit, mode );				\
+} while (0)
+
+
+#define MODEL_PROJ 0
+#define MODEL      1
+#define MODEL_IT   2
+#define TEXMAT_0   3
+#define TEXMAT_1   4
+#define TEXMAT_2   5
+
+#endif
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_state_init.c b/src/mesa/drivers/dri/radeon/radeon_state_init.c
new file mode 100644
index 00000000000..971ea699ea2
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_state_init.c
@@ -0,0 +1,531 @@
+/* $XFree86$ */
+/*
+ * Copyright 2000, 2001 VA Linux Systems Inc., Fremont, California.
+ *
+ * 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
+ * VA LINUX SYSTEMS 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:
+ *    Gareth Hughes <[email protected]>
+ *    Keith Whitwell <[email protected]>
+ */
+
+#include "glheader.h"
+#include "imports.h"
+#include "api_arrayelt.h"
+
+#include "swrast/swrast.h"
+#include "array_cache/acache.h"
+#include "tnl/tnl.h"
+#include "tnl/t_pipeline.h"
+#include "swrast_setup/swrast_setup.h"
+
+#include "radeon_context.h"
+#include "radeon_ioctl.h"
+#include "radeon_state.h"
+#include "radeon_tcl.h"
+#include "radeon_tex.h"
+#include "radeon_swtcl.h"
+#include "radeon_vtxfmt.h"
+
+/* =============================================================
+ * State initialization
+ */
+
+void radeonPrintDirty( radeonContextPtr rmesa, const char *msg )
+{
+   struct radeon_state_atom *l;
+
+   fprintf(stderr, msg);
+   fprintf(stderr, ": ");
+
+   foreach(l, &(rmesa->hw.dirty)) {
+      fprintf(stderr, "%s, ", l->name);
+   }
+
+   fprintf(stderr, "\n");
+}
+
+static int cmdpkt( int id ) 
+{
+   drmRadeonCmdHeader h;
+   h.i = 0;
+   h.packet.cmd_type = RADEON_CMD_PACKET;
+   h.packet.packet_id = id;
+   return h.i;
+}
+
+static int cmdvec( int offset, int stride, int count ) 
+{
+   drmRadeonCmdHeader h;
+   h.i = 0;
+   h.vectors.cmd_type = RADEON_CMD_VECTORS;
+   h.vectors.offset = offset;
+   h.vectors.stride = stride;
+   h.vectors.count = count;
+   return h.i;
+}
+
+static int cmdscl( int offset, int stride, int count ) 
+{
+   drmRadeonCmdHeader h;
+   h.i = 0;
+   h.scalars.cmd_type = RADEON_CMD_SCALARS;
+   h.scalars.offset = offset;
+   h.scalars.stride = stride;
+   h.scalars.count = count;
+   return h.i;
+}
+
+#define CHECK( NM, FLAG )			\
+static GLboolean check_##NM( GLcontext *ctx )	\
+{						\
+   return FLAG;					\
+}
+
+#define TCL_CHECK( NM, FLAG )				\
+static GLboolean check_##NM( GLcontext *ctx )		\
+{							\
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);	\
+   return !rmesa->TclFallback && (FLAG);		\
+}
+
+
+CHECK( always, GL_TRUE )
+CHECK( tex0, ctx->Texture.Unit[0]._ReallyEnabled )
+CHECK( tex1, ctx->Texture.Unit[1]._ReallyEnabled )
+CHECK( fog, ctx->Fog.Enabled )
+TCL_CHECK( tcl, GL_TRUE )
+TCL_CHECK( tcl_tex0, ctx->Texture.Unit[0]._ReallyEnabled )
+TCL_CHECK( tcl_tex1, ctx->Texture.Unit[1]._ReallyEnabled )
+TCL_CHECK( tcl_lighting, ctx->Light.Enabled )
+TCL_CHECK( tcl_eyespace_or_lighting, ctx->_NeedEyeCoords || ctx->Light.Enabled )
+TCL_CHECK( tcl_lit0, ctx->Light.Enabled && ctx->Light.Light[0].Enabled )
+TCL_CHECK( tcl_lit1, ctx->Light.Enabled && ctx->Light.Light[1].Enabled )
+TCL_CHECK( tcl_lit2, ctx->Light.Enabled && ctx->Light.Light[2].Enabled )
+TCL_CHECK( tcl_lit3, ctx->Light.Enabled && ctx->Light.Light[3].Enabled )
+TCL_CHECK( tcl_lit4, ctx->Light.Enabled && ctx->Light.Light[4].Enabled )
+TCL_CHECK( tcl_lit5, ctx->Light.Enabled && ctx->Light.Light[5].Enabled )
+TCL_CHECK( tcl_lit6, ctx->Light.Enabled && ctx->Light.Light[6].Enabled )
+TCL_CHECK( tcl_lit7, ctx->Light.Enabled && ctx->Light.Light[7].Enabled )
+TCL_CHECK( tcl_ucp0, (ctx->Transform.ClipPlanesEnabled & 0x1) )
+TCL_CHECK( tcl_ucp1, (ctx->Transform.ClipPlanesEnabled & 0x2) )
+TCL_CHECK( tcl_ucp2, (ctx->Transform.ClipPlanesEnabled & 0x4) )
+TCL_CHECK( tcl_ucp3, (ctx->Transform.ClipPlanesEnabled & 0x8) )
+TCL_CHECK( tcl_ucp4, (ctx->Transform.ClipPlanesEnabled & 0x10) )
+TCL_CHECK( tcl_ucp5, (ctx->Transform.ClipPlanesEnabled & 0x20) )
+TCL_CHECK( tcl_eyespace_or_fog, ctx->_NeedEyeCoords || ctx->Fog.Enabled ) 
+
+CHECK( txr0, (ctx->Texture.Unit[0]._ReallyEnabled & TEXTURE_RECT_BIT))
+CHECK( txr1, (ctx->Texture.Unit[1]._ReallyEnabled & TEXTURE_RECT_BIT))
+
+
+
+/* Initialize the context's hardware state.
+ */
+void radeonInitState( radeonContextPtr rmesa )
+{
+   GLcontext *ctx = rmesa->glCtx;
+   GLuint color_fmt, depth_fmt, i;
+
+   switch ( rmesa->radeonScreen->cpp ) {
+   case 2:
+      color_fmt = RADEON_COLOR_FORMAT_RGB565;
+      break;
+   case 4:
+      color_fmt = RADEON_COLOR_FORMAT_ARGB8888;
+      break;
+   default:
+      fprintf( stderr, "Error: Unsupported pixel depth... exiting\n" );
+      exit( -1 );
+   }
+
+   rmesa->state.color.clear = 0x00000000;
+
+   switch ( ctx->Visual.depthBits ) {
+   case 16:
+      rmesa->state.depth.clear = 0x0000ffff;
+      rmesa->state.depth.scale = 1.0 / (GLfloat)0xffff;
+      depth_fmt = RADEON_DEPTH_FORMAT_16BIT_INT_Z;
+      rmesa->state.stencil.clear = 0x00000000;
+      break;
+   case 24:
+      rmesa->state.depth.clear = 0x00ffffff;
+      rmesa->state.depth.scale = 1.0 / (GLfloat)0xffffff;
+      depth_fmt = RADEON_DEPTH_FORMAT_24BIT_INT_Z;
+      rmesa->state.stencil.clear = 0xff000000;
+      break;
+   default:
+      fprintf( stderr, "Error: Unsupported depth %d... exiting\n",
+	       ctx->Visual.depthBits );
+      exit( -1 );
+   }
+
+   /* Only have hw stencil when depth buffer is 24 bits deep */
+   rmesa->state.stencil.hwBuffer = ( ctx->Visual.stencilBits > 0 &&
+				     ctx->Visual.depthBits == 24 );
+
+   rmesa->Fallback = 0;
+
+   if ( ctx->Visual.doubleBufferMode && rmesa->sarea->pfCurrentPage == 0 ) {
+      rmesa->state.color.drawOffset = rmesa->radeonScreen->backOffset;
+      rmesa->state.color.drawPitch  = rmesa->radeonScreen->backPitch;
+   } else {
+      rmesa->state.color.drawOffset = rmesa->radeonScreen->frontOffset;
+      rmesa->state.color.drawPitch  = rmesa->radeonScreen->frontPitch;
+   }
+   rmesa->state.pixel.readOffset = rmesa->state.color.drawOffset;
+   rmesa->state.pixel.readPitch  = rmesa->state.color.drawPitch;
+
+   /* Initialize lists:
+    */
+   make_empty_list(&(rmesa->hw.dirty));
+   make_empty_list(&(rmesa->hw.clean));
+
+
+#define ALLOC_STATE( ATOM, CHK, SZ, NM, FLAG )				\
+   do {								\
+      rmesa->hw.ATOM.cmd_size = SZ;				\
+      rmesa->hw.ATOM.cmd = (int *)CALLOC(SZ * sizeof(int));	\
+      rmesa->hw.ATOM.lastcmd = (int *)CALLOC(SZ * sizeof(int));	\
+      rmesa->hw.ATOM.name = NM;					\
+      rmesa->hw.ATOM.is_tcl = FLAG;					\
+      rmesa->hw.ATOM.check = check_##CHK;				\
+      insert_at_head(&(rmesa->hw.dirty), &(rmesa->hw.ATOM));	\
+   } while (0)
+      
+      
+   /* Allocate state buffers:
+    */
+   ALLOC_STATE( ctx, always, CTX_STATE_SIZE, "CTX/context", 0 );
+   ALLOC_STATE( lin, always, LIN_STATE_SIZE, "LIN/line", 0 );
+   ALLOC_STATE( msk, always, MSK_STATE_SIZE, "MSK/mask", 0 );
+   ALLOC_STATE( vpt, always, VPT_STATE_SIZE, "VPT/viewport", 0 );
+   ALLOC_STATE( set, always, SET_STATE_SIZE, "SET/setup", 0 );
+   ALLOC_STATE( msc, always, MSC_STATE_SIZE, "MSC/misc", 0 );
+   ALLOC_STATE( zbs, always, ZBS_STATE_SIZE, "ZBS/zbias", 0 );
+   ALLOC_STATE( tcl, always, TCL_STATE_SIZE, "TCL/tcl", 1 );
+   ALLOC_STATE( mtl, tcl_lighting, MTL_STATE_SIZE, "MTL/material", 1 );
+   ALLOC_STATE( grd, always, GRD_STATE_SIZE, "GRD/guard-band", 1 );
+   ALLOC_STATE( fog, fog, FOG_STATE_SIZE, "FOG/fog", 1 );
+   ALLOC_STATE( glt, tcl_lighting, GLT_STATE_SIZE, "GLT/light-global", 1 );
+   ALLOC_STATE( eye, tcl_lighting, EYE_STATE_SIZE, "EYE/eye-vector", 1 );
+   ALLOC_STATE( tex[0], tex0, TEX_STATE_SIZE, "TEX/tex-0", 0 );
+   ALLOC_STATE( tex[1], tex1, TEX_STATE_SIZE, "TEX/tex-1", 0 );
+   ALLOC_STATE( mat[0], tcl, MAT_STATE_SIZE, "MAT/modelproject", 1 );
+   ALLOC_STATE( mat[1], tcl_eyespace_or_fog, MAT_STATE_SIZE, "MAT/modelview", 1 );
+   ALLOC_STATE( mat[2], tcl_eyespace_or_lighting, MAT_STATE_SIZE, "MAT/it-modelview", 1 );
+   ALLOC_STATE( mat[3], tcl_tex0, MAT_STATE_SIZE, "MAT/texmat0", 1 );
+   ALLOC_STATE( mat[4], tcl_tex1, MAT_STATE_SIZE, "MAT/texmat1", 1 );
+   ALLOC_STATE( ucp[0], tcl_ucp0, UCP_STATE_SIZE, "UCP/userclip-0", 1 );
+   ALLOC_STATE( ucp[1], tcl_ucp1, UCP_STATE_SIZE, "UCP/userclip-1", 1 );
+   ALLOC_STATE( ucp[2], tcl_ucp2, UCP_STATE_SIZE, "UCP/userclip-2", 1 );
+   ALLOC_STATE( ucp[3], tcl_ucp3, UCP_STATE_SIZE, "UCP/userclip-3", 1 );
+   ALLOC_STATE( ucp[4], tcl_ucp4, UCP_STATE_SIZE, "UCP/userclip-4", 1 );
+   ALLOC_STATE( ucp[5], tcl_ucp5, UCP_STATE_SIZE, "UCP/userclip-5", 1 );
+   ALLOC_STATE( lit[0], tcl_lit0, LIT_STATE_SIZE, "LIT/light-0", 1 );
+   ALLOC_STATE( lit[1], tcl_lit1, LIT_STATE_SIZE, "LIT/light-1", 1 );
+   ALLOC_STATE( lit[2], tcl_lit2, LIT_STATE_SIZE, "LIT/light-2", 1 );
+   ALLOC_STATE( lit[3], tcl_lit3, LIT_STATE_SIZE, "LIT/light-3", 1 );
+   ALLOC_STATE( lit[4], tcl_lit4, LIT_STATE_SIZE, "LIT/light-4", 1 );
+   ALLOC_STATE( lit[5], tcl_lit5, LIT_STATE_SIZE, "LIT/light-5", 1 );
+   ALLOC_STATE( lit[6], tcl_lit6, LIT_STATE_SIZE, "LIT/light-6", 1 );
+   ALLOC_STATE( lit[7], tcl_lit7, LIT_STATE_SIZE, "LIT/light-7", 1 );
+   ALLOC_STATE( txr[0], txr0, TXR_STATE_SIZE, "TXR/txr-0", 0 );
+   ALLOC_STATE( txr[1], txr1, TXR_STATE_SIZE, "TXR/txr-1", 0 );
+
+
+   /* Fill in the packet headers:
+    */
+   rmesa->hw.ctx.cmd[CTX_CMD_0] = cmdpkt(RADEON_EMIT_PP_MISC);
+   rmesa->hw.ctx.cmd[CTX_CMD_1] = cmdpkt(RADEON_EMIT_PP_CNTL);
+   rmesa->hw.ctx.cmd[CTX_CMD_2] = cmdpkt(RADEON_EMIT_RB3D_COLORPITCH);
+   rmesa->hw.lin.cmd[LIN_CMD_0] = cmdpkt(RADEON_EMIT_RE_LINE_PATTERN);
+   rmesa->hw.lin.cmd[LIN_CMD_1] = cmdpkt(RADEON_EMIT_SE_LINE_WIDTH);
+   rmesa->hw.msk.cmd[MSK_CMD_0] = cmdpkt(RADEON_EMIT_RB3D_STENCILREFMASK);
+   rmesa->hw.vpt.cmd[VPT_CMD_0] = cmdpkt(RADEON_EMIT_SE_VPORT_XSCALE);
+   rmesa->hw.set.cmd[SET_CMD_0] = cmdpkt(RADEON_EMIT_SE_CNTL);
+   rmesa->hw.set.cmd[SET_CMD_1] = cmdpkt(RADEON_EMIT_SE_CNTL_STATUS);
+   rmesa->hw.msc.cmd[MSC_CMD_0] = cmdpkt(RADEON_EMIT_RE_MISC);
+   rmesa->hw.tex[0].cmd[TEX_CMD_0] = cmdpkt(RADEON_EMIT_PP_TXFILTER_0);
+   rmesa->hw.tex[0].cmd[TEX_CMD_1] = cmdpkt(RADEON_EMIT_PP_BORDER_COLOR_0);
+   rmesa->hw.tex[1].cmd[TEX_CMD_0] = cmdpkt(RADEON_EMIT_PP_TXFILTER_1);
+   rmesa->hw.tex[1].cmd[TEX_CMD_1] = cmdpkt(RADEON_EMIT_PP_BORDER_COLOR_1);
+   rmesa->hw.zbs.cmd[ZBS_CMD_0] = cmdpkt(RADEON_EMIT_SE_ZBIAS_FACTOR);
+   rmesa->hw.tcl.cmd[TCL_CMD_0] = cmdpkt(RADEON_EMIT_SE_TCL_OUTPUT_VTX_FMT);
+   rmesa->hw.mtl.cmd[MTL_CMD_0] = 
+      cmdpkt(RADEON_EMIT_SE_TCL_MATERIAL_EMMISSIVE_RED);
+   rmesa->hw.txr[0].cmd[TXR_CMD_0] = cmdpkt(RADEON_EMIT_PP_TEX_SIZE_0);
+   rmesa->hw.txr[1].cmd[TXR_CMD_0] = cmdpkt(RADEON_EMIT_PP_TEX_SIZE_1);
+   rmesa->hw.grd.cmd[GRD_CMD_0] = 
+      cmdscl( RADEON_SS_VERT_GUARD_CLIP_ADJ_ADDR, 1, 4 );
+   rmesa->hw.fog.cmd[FOG_CMD_0] = 
+      cmdvec( RADEON_VS_FOG_PARAM_ADDR, 1, 4 );
+   rmesa->hw.glt.cmd[GLT_CMD_0] = 
+      cmdvec( RADEON_VS_GLOBAL_AMBIENT_ADDR, 1, 4 );
+   rmesa->hw.eye.cmd[EYE_CMD_0] = 
+      cmdvec( RADEON_VS_EYE_VECTOR_ADDR, 1, 4 );
+
+   for (i = 0 ; i < 5; i++) {
+      rmesa->hw.mat[i].cmd[MAT_CMD_0] = 
+	 cmdvec( RADEON_VS_MATRIX_0_ADDR + i*4, 1, 16);
+   }
+
+   for (i = 0 ; i < 8; i++) {
+      rmesa->hw.lit[i].cmd[LIT_CMD_0] = 
+	 cmdvec( RADEON_VS_LIGHT_AMBIENT_ADDR + i, 8, 24 );
+      rmesa->hw.lit[i].cmd[LIT_CMD_1] = 
+	 cmdscl( RADEON_SS_LIGHT_DCD_ADDR + i, 8, 6 );
+   }
+
+   for (i = 0 ; i < 6; i++) {
+      rmesa->hw.ucp[i].cmd[UCP_CMD_0] = 
+	 cmdvec( RADEON_VS_UCP_ADDR + i, 1, 4 );
+   }
+
+   rmesa->last_ReallyEnabled = -1;
+
+   /* Initial Harware state:
+    */
+   rmesa->hw.ctx.cmd[CTX_PP_MISC] = (RADEON_ALPHA_TEST_PASS |
+				     RADEON_CHROMA_FUNC_FAIL |
+				     RADEON_CHROMA_KEY_NEAREST |
+				     RADEON_SHADOW_FUNC_EQUAL |
+				     RADEON_SHADOW_PASS_1 |
+				     RADEON_RIGHT_HAND_CUBE_OGL);
+
+   rmesa->hw.ctx.cmd[CTX_PP_FOG_COLOR] = (RADEON_FOG_VERTEX |
+					  RADEON_FOG_USE_DEPTH);
+
+   rmesa->hw.ctx.cmd[CTX_RE_SOLID_COLOR] = 0x00000000;
+
+   rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCNTL] = (RADEON_COMB_FCN_ADD_CLAMP |
+					    RADEON_SRC_BLEND_GL_ONE |
+					    RADEON_DST_BLEND_GL_ZERO );
+
+   rmesa->hw.ctx.cmd[CTX_RB3D_DEPTHOFFSET] =
+      rmesa->radeonScreen->depthOffset;
+
+   rmesa->hw.ctx.cmd[CTX_RB3D_DEPTHPITCH] = 
+      ((rmesa->radeonScreen->depthPitch &
+	RADEON_DEPTHPITCH_MASK) |
+       RADEON_DEPTH_ENDIAN_NO_SWAP);
+
+   rmesa->hw.ctx.cmd[CTX_RB3D_ZSTENCILCNTL] = (depth_fmt |
+					       RADEON_Z_TEST_LESS |
+					       RADEON_STENCIL_TEST_ALWAYS |
+					       RADEON_STENCIL_FAIL_KEEP |
+					       RADEON_STENCIL_ZPASS_KEEP |
+					       RADEON_STENCIL_ZFAIL_KEEP |
+					       RADEON_Z_WRITE_ENABLE);
+
+   rmesa->hw.ctx.cmd[CTX_PP_CNTL] = (RADEON_SCISSOR_ENABLE |
+				     RADEON_ANTI_ALIAS_NONE);
+
+   rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] = (RADEON_PLANE_MASK_ENABLE |
+				       color_fmt |
+				       (1<<15));
+
+   rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] |=  RADEON_DITHER_ENABLE;
+
+   rmesa->hw.ctx.cmd[CTX_RB3D_COLOROFFSET] = (rmesa->state.color.drawOffset &
+					      RADEON_COLOROFFSET_MASK);
+
+   rmesa->hw.ctx.cmd[CTX_RB3D_COLORPITCH] = ((rmesa->state.color.drawPitch &
+					      RADEON_COLORPITCH_MASK) |
+					     RADEON_COLOR_ENDIAN_NO_SWAP);
+
+   rmesa->hw.set.cmd[SET_SE_CNTL] = (RADEON_FFACE_CULL_CCW |
+				     RADEON_BFACE_SOLID |
+				     RADEON_FFACE_SOLID |
+/*  			     RADEON_BADVTX_CULL_DISABLE | */
+				     RADEON_FLAT_SHADE_VTX_LAST |
+				     RADEON_DIFFUSE_SHADE_GOURAUD |
+				     RADEON_ALPHA_SHADE_GOURAUD |
+				     RADEON_SPECULAR_SHADE_GOURAUD |
+				     RADEON_FOG_SHADE_GOURAUD |
+				     RADEON_VPORT_XY_XFORM_ENABLE |
+				     RADEON_VPORT_Z_XFORM_ENABLE |
+				     RADEON_VTX_PIX_CENTER_OGL |
+				     RADEON_ROUND_MODE_TRUNC |
+				     RADEON_ROUND_PREC_8TH_PIX);
+
+   rmesa->hw.set.cmd[SET_SE_CNTL_STATUS] =
+#ifdef MESA_BIG_ENDIAN
+					    RADEON_VC_32BIT_SWAP;
+#else
+  					    RADEON_VC_NO_SWAP;
+#endif
+
+   if (!(rmesa->radeonScreen->chipset & RADEON_CHIPSET_TCL)) {
+     rmesa->hw.set.cmd[SET_SE_CNTL_STATUS] |= RADEON_TCL_BYPASS;
+   }
+
+   rmesa->hw.set.cmd[SET_SE_COORDFMT] = (
+      RADEON_VTX_W0_IS_NOT_1_OVER_W0 |
+      RADEON_TEX1_W_ROUTING_USE_Q1);
+
+
+   rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] = ((1 << 16) | 0xffff);
+
+   rmesa->hw.lin.cmd[LIN_RE_LINE_STATE] = 
+      ((0 << RADEON_LINE_CURRENT_PTR_SHIFT) |
+       (1 << RADEON_LINE_CURRENT_COUNT_SHIFT));
+
+   rmesa->hw.lin.cmd[LIN_SE_LINE_WIDTH] = (1 << 4);
+
+   rmesa->hw.msk.cmd[MSK_RB3D_STENCILREFMASK] = 
+      ((0x00 << RADEON_STENCIL_REF_SHIFT) |
+       (0xff << RADEON_STENCIL_MASK_SHIFT) |
+       (0xff << RADEON_STENCIL_WRITEMASK_SHIFT));
+
+   rmesa->hw.msk.cmd[MSK_RB3D_ROPCNTL] = RADEON_ROP_COPY;
+   rmesa->hw.msk.cmd[MSK_RB3D_PLANEMASK] = 0xffffffff;
+
+   rmesa->hw.msc.cmd[MSC_RE_MISC] = 
+      ((0 << RADEON_STIPPLE_X_OFFSET_SHIFT) |
+       (0 << RADEON_STIPPLE_Y_OFFSET_SHIFT) |
+       RADEON_STIPPLE_BIG_BIT_ORDER);
+
+   rmesa->hw.vpt.cmd[VPT_SE_VPORT_XSCALE]  = 0x00000000;
+   rmesa->hw.vpt.cmd[VPT_SE_VPORT_XOFFSET] = 0x00000000;
+   rmesa->hw.vpt.cmd[VPT_SE_VPORT_YSCALE]  = 0x00000000;
+   rmesa->hw.vpt.cmd[VPT_SE_VPORT_YOFFSET] = 0x00000000;
+   rmesa->hw.vpt.cmd[VPT_SE_VPORT_ZSCALE]  = 0x00000000;
+   rmesa->hw.vpt.cmd[VPT_SE_VPORT_ZOFFSET] = 0x00000000;
+
+   for ( i = 0 ; i < ctx->Const.MaxTextureUnits ; i++ ) {
+      rmesa->hw.tex[i].cmd[TEX_PP_TXFILTER] = RADEON_BORDER_MODE_OGL;
+      rmesa->hw.tex[i].cmd[TEX_PP_TXFORMAT] = 
+	  (RADEON_TXFORMAT_ENDIAN_NO_SWAP |
+	   RADEON_TXFORMAT_PERSPECTIVE_ENABLE |
+	   (i << 24) | /* This is one of RADEON_TXFORMAT_ST_ROUTE_STQ[012] */
+	   (2 << RADEON_TXFORMAT_WIDTH_SHIFT) |
+	   (2 << RADEON_TXFORMAT_HEIGHT_SHIFT));
+
+      /* FIXME: What is this magic value? */
+      rmesa->hw.tex[i].cmd[TEX_PP_TXOFFSET] = 0x2000 << (2 * i);
+
+      rmesa->hw.tex[i].cmd[TEX_PP_BORDER_COLOR] = 0;
+      rmesa->hw.tex[i].cmd[TEX_PP_TXCBLEND] =  
+	  (RADEON_COLOR_ARG_A_ZERO |
+	   RADEON_COLOR_ARG_B_ZERO |
+	   RADEON_COLOR_ARG_C_CURRENT_COLOR |
+	   RADEON_BLEND_CTL_ADD |
+	   RADEON_SCALE_1X |
+	   RADEON_CLAMP_TX);
+      rmesa->hw.tex[i].cmd[TEX_PP_TXABLEND] = 
+	  (RADEON_ALPHA_ARG_A_ZERO |
+	   RADEON_ALPHA_ARG_B_ZERO |
+	   RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+	   RADEON_BLEND_CTL_ADD |
+	   RADEON_SCALE_1X |
+	   RADEON_CLAMP_TX);
+      rmesa->hw.tex[i].cmd[TEX_PP_TFACTOR] = 0;
+   }
+
+   /* Can only add ST1 at the time of doing some multitex but can keep
+    * it after that.  Errors if DIFFUSE is missing.
+    */
+   rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] = 
+      (RADEON_TCL_VTX_Z0 |
+       RADEON_TCL_VTX_W0 |
+       RADEON_TCL_VTX_PK_DIFFUSE
+	 );	/* need to keep this uptodate */
+						   
+   rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXSEL] =
+      ( RADEON_TCL_COMPUTE_XYZW 	|
+	(RADEON_TCL_TEX_INPUT_TEX_0 << RADEON_TCL_TEX_0_OUTPUT_SHIFT) |
+	(RADEON_TCL_TEX_INPUT_TEX_1 << RADEON_TCL_TEX_1_OUTPUT_SHIFT) |
+	(RADEON_TCL_TEX_INPUT_TEX_2 << RADEON_TCL_TEX_2_OUTPUT_SHIFT));
+
+
+   /* XXX */
+   rmesa->hw.tcl.cmd[TCL_MATRIX_SELECT_0] = 
+      ((MODEL << RADEON_MODELVIEW_0_SHIFT) |
+       (MODEL_IT << RADEON_IT_MODELVIEW_0_SHIFT));
+
+   rmesa->hw.tcl.cmd[TCL_MATRIX_SELECT_1] = 
+      ((MODEL_PROJ << RADEON_MODELPROJECT_0_SHIFT) |
+       (TEXMAT_0 << RADEON_TEXMAT_0_SHIFT) |
+       (TEXMAT_1 << RADEON_TEXMAT_1_SHIFT));
+
+   rmesa->hw.tcl.cmd[TCL_UCP_VERT_BLEND_CTL] = 
+      (RADEON_UCP_IN_CLIP_SPACE |
+       RADEON_CULL_FRONT_IS_CCW);
+
+   rmesa->hw.tcl.cmd[TCL_TEXTURE_PROC_CTL] = 0; 
+
+   rmesa->hw.tcl.cmd[TCL_LIGHT_MODEL_CTL] = 
+      (RADEON_SPECULAR_LIGHTS |
+       RADEON_DIFFUSE_SPECULAR_COMBINE |
+       RADEON_LOCAL_LIGHT_VEC_GL |
+       (RADEON_LM_SOURCE_STATE_PREMULT << RADEON_EMISSIVE_SOURCE_SHIFT) |
+       (RADEON_LM_SOURCE_STATE_PREMULT << RADEON_AMBIENT_SOURCE_SHIFT) |
+       (RADEON_LM_SOURCE_STATE_PREMULT << RADEON_DIFFUSE_SOURCE_SHIFT) |
+       (RADEON_LM_SOURCE_STATE_PREMULT << RADEON_SPECULAR_SOURCE_SHIFT)); 
+
+   for (i = 0 ; i < 8; i++) {
+      struct gl_light *l = &ctx->Light.Light[i];
+      GLenum p = GL_LIGHT0 + i;
+      *(float *)&(rmesa->hw.lit[i].cmd[LIT_RANGE_CUTOFF]) = FLT_MAX;
+
+      ctx->Driver.Lightfv( ctx, p, GL_AMBIENT, l->Ambient );
+      ctx->Driver.Lightfv( ctx, p, GL_DIFFUSE, l->Diffuse );
+      ctx->Driver.Lightfv( ctx, p, GL_SPECULAR, l->Specular );
+      ctx->Driver.Lightfv( ctx, p, GL_POSITION, 0 );
+      ctx->Driver.Lightfv( ctx, p, GL_SPOT_DIRECTION, 0 );
+      ctx->Driver.Lightfv( ctx, p, GL_SPOT_EXPONENT, &l->SpotExponent );
+      ctx->Driver.Lightfv( ctx, p, GL_SPOT_CUTOFF, &l->SpotCutoff );
+      ctx->Driver.Lightfv( ctx, p, GL_CONSTANT_ATTENUATION,
+			   &l->ConstantAttenuation );
+      ctx->Driver.Lightfv( ctx, p, GL_LINEAR_ATTENUATION, 
+			   &l->LinearAttenuation );
+      ctx->Driver.Lightfv( ctx, p, GL_QUADRATIC_ATTENUATION, 
+		     &l->QuadraticAttenuation );
+   }
+
+   ctx->Driver.LightModelfv( ctx, GL_LIGHT_MODEL_AMBIENT, 
+			     ctx->Light.Model.Ambient );
+
+   TNL_CONTEXT(ctx)->Driver.NotifyMaterialChange( ctx );
+
+   for (i = 0 ; i < 6; i++) {
+      ctx->Driver.ClipPlane( ctx, GL_CLIP_PLANE0 + i, NULL );
+   }
+
+   ctx->Driver.Fogfv( ctx, GL_FOG_MODE, 0 );
+   ctx->Driver.Fogfv( ctx, GL_FOG_DENSITY, &ctx->Fog.Density );
+   ctx->Driver.Fogfv( ctx, GL_FOG_START, &ctx->Fog.Start );
+   ctx->Driver.Fogfv( ctx, GL_FOG_END, &ctx->Fog.End );
+   ctx->Driver.Fogfv( ctx, GL_FOG_COLOR, ctx->Fog.Color );
+   ctx->Driver.Fogfv( ctx, GL_FOG_COORDINATE_SOURCE_EXT, 0 );
+   
+   rmesa->hw.grd.cmd[GRD_VERT_GUARD_CLIP_ADJ] = IEEE_ONE;
+   rmesa->hw.grd.cmd[GRD_VERT_GUARD_DISCARD_ADJ] = IEEE_ONE;
+   rmesa->hw.grd.cmd[GRD_HORZ_GUARD_CLIP_ADJ] = IEEE_ONE;
+   rmesa->hw.grd.cmd[GRD_HORZ_GUARD_DISCARD_ADJ] = IEEE_ONE;
+
+   rmesa->hw.eye.cmd[EYE_X] = 0;
+   rmesa->hw.eye.cmd[EYE_Y] = 0;
+   rmesa->hw.eye.cmd[EYE_Z] = IEEE_ONE;
+   rmesa->hw.eye.cmd[EYE_RESCALE_FACTOR] = IEEE_ONE;
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_subset.h b/src/mesa/drivers/dri/radeon/radeon_subset.h
new file mode 100644
index 00000000000..1b6e7bd432c
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_subset.h
@@ -0,0 +1,75 @@
+/**
+ * \file radeon_subset.h
+ * \brief Radeon subset driver declarations.
+ *
+ * \author Keith Whitwell <[email protected]>
+ */
+
+/*
+ * Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+ *                      Tungsten Grahpics Inc., Austin, 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
+ * ATI, TUNGSTEN GRAHPICS 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.
+ */
+
+/* $XFree86$ */
+
+#ifndef __RADEON_SUBSET_H__
+#define __RADEON_SUBSET_H__
+
+extern void radeonPointsBitmap( GLsizei width, GLsizei height,
+				GLfloat xorig, GLfloat yorig,
+				GLfloat xmove, GLfloat ymove,
+				const GLubyte *bitmap );
+
+extern void radeonReadPixels( GLint x, GLint y,
+			      GLsizei width, GLsizei height,
+			      GLenum format, GLenum type,
+			      GLvoid *pixels );
+
+extern void radeon_select_Install( GLcontext *ctx );
+
+extern void radeonInitSelect( GLcontext *ctx );
+
+extern void radeonVtxfmtDestroy( GLcontext *ctx );
+
+extern void radeonVtxfmtMakeCurrent( GLcontext *ctx );
+
+extern void radeonVtxfmtUnbindContext( GLcontext *ctx );
+
+extern void radeonVtxfmtInit( GLcontext *ctx );
+
+extern void radeonTclFallback( GLcontext *ctx, GLuint bit, GLboolean mode );
+
+extern void radeonVtxfmtInvalidate( GLcontext *ctx );
+
+extern void radeonSubsetVtxEnableTCL( radeonContextPtr rmesa, GLboolean flag );
+
+extern void radeonUpdateTextureState( GLcontext *ctx );
+
+extern void radeonInitTextureFuncs( GLcontext *ctx );
+
+extern void radeonAgeTextures( radeonContextPtr rmesa, int heap );
+
+extern void radeonDestroyTexObj( radeonContextPtr rmesa, radeonTexObjPtr t );
+
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_subset_bitmap.c b/src/mesa/drivers/dri/radeon/radeon_subset_bitmap.c
new file mode 100644
index 00000000000..cb4a5142219
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_subset_bitmap.c
@@ -0,0 +1,197 @@
+/**
+ * \file radeon_subset_bitmap.c
+ * \brief Bitmap drawing.
+ * 
+ * \author Keith Whitwell <[email protected]>
+ */
+
+/*
+ * Copyright 2003       ATI Technologies Inc., Ontario, Canada, and
+ *                      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
+ * ATI, 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.
+ * 
+ */
+
+/* $XFree86$ */
+
+#include "glheader.h"
+#include "mtypes.h"
+#include "colormac.h"
+#include "context.h"
+#include "enums.h"
+#include "imports.h"
+#include "image.h"
+/*#include "mmath.h"*/
+#include "macros.h"
+#include "state.h"
+
+#include "radeon_context.h"
+#include "radeon_ioctl.h"
+#include "radeon_state.h"
+#include "radeon_subset.h"
+
+/**
+ * \brief Cope with depth operations by drawing individual pixels as points
+ *
+ * \param xorig x coordinate of the bitmap corner.
+ * \param yorig y coordinate of the bitmap corner.
+ * \param xmove increment to the final x coordinate.
+ * \param ymove increment to the final y coordinate.
+ * \param width bitmap width.
+ * \param height bitmap height.
+ * \param bitmap bitmap pointer.
+ * 
+ * Clips the bitmap coordinates and adjusts for windows coordinates. Draws the
+ * bitmap with glPoints(), turning off TCL and hardware viewport transformation
+ * to emit raw pixel coordinates. Finally fires any outstanding vertices and
+ * restores TCL, viewport, texture and color states.
+ */
+void
+radeonPointsBitmap(  GLsizei width, GLsizei height,
+		     GLfloat xorig, GLfloat yorig, 
+		     GLfloat xmove, GLfloat ymove,
+		     const GLubyte *bitmap )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   GLsizei bmwidth = width, bmheight = height;   
+   GLint px, py;
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat saved_color[4], saved_tex0[2];
+   GLint row, col;
+   GLuint orig_se_cntl;
+   GLuint w, h;
+   const struct gl_pixelstore_attrib *unpack = &ctx->Unpack;
+
+   ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx);
+
+   if (width < 0 || height < 0) {
+      _mesa_error( ctx, GL_INVALID_VALUE, "glBitmap(width or height < 0)" );
+      return;
+   }
+
+   if (!ctx->Current.RasterPosValid)
+      return;
+
+   if (ctx->NewState) 
+      _mesa_update_state(ctx);
+
+
+   if (ctx->_RotateMode) {
+      width = bmheight; height = bmwidth;
+
+      px = IFLOOR(ctx->Current.RasterPos[0] + yorig);
+      py = IFLOOR(ctx->Current.RasterPos[1] + xorig);
+
+      ctx->Current.RasterPos[0] += ymove;
+      ctx->Current.RasterPos[1] += xmove;
+   }
+   else {
+      px = IFLOOR(ctx->Current.RasterPos[0] - xorig);
+      py = IFLOOR(ctx->Current.RasterPos[1] - yorig);
+
+      ctx->Current.RasterPos[0] += xmove;
+      ctx->Current.RasterPos[1] += ymove;
+   }
+
+
+
+   /* Turn off tcl and the hw viewport transformation so that we can
+    * emit raw pixel coordinates:
+    */
+   radeonSubsetVtxEnableTCL( rmesa, GL_FALSE );
+   RADEON_STATECHANGE( rmesa, set );
+   orig_se_cntl = rmesa->hw.set.cmd[SET_SE_CNTL];
+   rmesa->hw.set.cmd[SET_SE_CNTL] &= ~(RADEON_VPORT_XY_XFORM_ENABLE |
+				       RADEON_VPORT_Z_XFORM_ENABLE);
+
+
+   /* Adjust for window coordinates, flip y values:
+    */
+   h = rmesa->dri.drawable->h + rmesa->dri.drawable->y - 1;
+   w = rmesa->dri.drawable->w;
+   px += rmesa->dri.drawable->x;
+
+   /* Save current color, texcoord to restore later:
+    */
+   COPY_4V( saved_color, ctx->Current.Attrib[VERT_ATTRIB_COLOR0] );
+   COPY_2V( saved_tex0, ctx->Current.Attrib[VERT_ATTRIB_TEX0] );
+
+   /* Just use the GL entrypoints to talk to radeon_subset_vtx.c:
+    */
+   glBegin( GL_POINTS );
+   glColor4fv( ctx->Current.RasterColor );
+   glTexCoord2fv( ctx->Current.RasterTexCoords[0] );
+
+
+   if (ctx->_RotateMode) {
+      for (col=0; col<width; col++) {
+	 const GLubyte *src = (const GLubyte *) 
+	    _mesa_image_address( unpack, bitmap, height, width, 
+				 GL_COLOR_INDEX, GL_BITMAP, 0, col, 0 );
+	    
+	 /* Msb first */
+	 GLubyte mask = 128U >> (unpack->SkipPixels & 0x7);
+	 for (row=0; row<height; row++) {
+	    if (*src & mask) {
+	       glVertex2f( px-col, h - (py + row) );
+	    }
+	    src += mask & 1;
+	    mask = ((mask << 7) & 0xff) | (mask >> 1);
+	 }
+	 /* get ready for next row */
+	 if (mask != 128)
+	    src++;
+      }
+   }
+   else {
+      for (row=0; row<height; row++) {
+	 const GLubyte *src = (const GLubyte *) 
+	    _mesa_image_address( unpack, bitmap, width, height, 
+				 GL_COLOR_INDEX, GL_BITMAP, 0, row, 0 );
+	    
+	 /* Msb first */
+	 GLubyte mask = 128U >> (unpack->SkipPixels & 0x7);
+	 for (col=0; col<width; col++) {
+	    if (*src & mask) {
+	       glVertex2f( px+col, h - (py + row) );
+	    }
+	    src += mask & 1;
+	    mask = ((mask << 7) & 0xff) | (mask >> 1);
+	 }
+	 /* get ready for next row */
+	 if (mask != 128)
+	    src++;
+      }
+   }
+
+   glEnd();
+   glColor4fv( saved_color );
+   glTexCoord2fv( saved_tex0 );
+
+   /* Fire outstanding vertices, restore state
+    */
+   RADEON_STATECHANGE( rmesa, set );
+   rmesa->hw.set.cmd[SET_SE_CNTL] = orig_se_cntl;
+   radeonSubsetVtxEnableTCL( rmesa, GL_TRUE );
+}
+
diff --git a/src/mesa/drivers/dri/radeon/radeon_subset_readpix.c b/src/mesa/drivers/dri/radeon/radeon_subset_readpix.c
new file mode 100644
index 00000000000..f72d3b84728
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_subset_readpix.c
@@ -0,0 +1,246 @@
+/**
+ * \file radeon_subset_readpix.c
+ * \brief Pixel reading.
+ * 
+ * \author Keith Whitwell <[email protected]>
+ * \author Brian Paul <[email protected]>
+ */
+
+/*
+ * Copyright 2003       ATI Technologies Inc., Ontario, Canada, and
+ *                      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
+ * ATI, 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.
+ */
+
+/* $XFree86$ */
+
+#include "glheader.h"
+#include "mtypes.h"
+#include "colormac.h"
+#include "context.h"
+#include "enums.h"
+#include "imports.h"
+/*#include "mmath.h" */
+#include "macros.h"
+#include "state.h"
+
+#include "radeon_context.h"
+#include "radeon_ioctl.h"
+#include "radeon_state.h"
+#include "radeon_subset.h"
+
+/**
+ * \brief Read pixel in RGBA format on a Radeon 16bpp frame buffer.
+ *
+ * \param rgba destination pointer.
+ * \param ptr pointer to the pixel in the frame buffer.
+ */
+#define READ_RGBA_16( rgba, ptr )		\
+do {						\
+    GLushort p = *(GLushort *)ptr;		\
+    rgba[0] = ((p >> 8) & 0xf8) * 255 / 0xf8;	\
+    rgba[1] = ((p >> 3) & 0xfc) * 255 / 0xfc;	\
+    rgba[2] = ((p << 3) & 0xf8) * 255 / 0xf8;	\
+    rgba[3] = 0xff;				\
+} while (0)
+
+/**
+ * \brief Read pixel in RGBA format on a Radeon 32bpp frame buffer.
+ *
+ * \param rgba destination pointer.
+ * \param ptr pointer to the pixel in the frame buffer.
+ */
+#define READ_RGBA_32( rgba, ptr )		\
+do {						\
+   GLuint p = *(GLuint *)ptr;			\
+   rgba[0] = (p >> 16) & 0xff;			\
+   rgba[1] = (p >>  8) & 0xff;			\
+   rgba[2] = (p >>  0) & 0xff;			\
+   rgba[3] = (p >> 24) & 0xff;			\
+} while (0)
+
+/**
+ * \brief Read a span in RGBA format.
+ * 
+ * \param ctx GL context.
+ * \param n number of pixels in the span.
+ * \param x x position of the span start.
+ * \param y y position of the span.
+ * \param rgba destination buffer.
+ *
+ * Calculates the pointer to the span start in the frame buffer and uses either
+ * #READ_RGBA_16 or #READ_RGBA_32 macros to copy the values.
+ */
+static void ReadRGBASpan( const GLcontext *ctx,
+			       GLuint n, GLint x, GLint y,
+			       GLubyte rgba[][4])
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeonScreenPtr radeonScreen = rmesa->radeonScreen;
+   __DRIdrawablePrivate *dPriv = rmesa->dri.drawable;
+   GLuint cpp = radeonScreen->cpp;
+   GLuint pitch = radeonScreen->frontPitch * cpp;
+   GLint i;
+
+   if (ctx->_RotateMode) {
+      char *ptr = (char *)(rmesa->dri.screen->pFB +
+			   rmesa->state.pixel.readOffset +
+			   ((dPriv->x + (dPriv->w - y - 1)) * cpp) +
+			   ((dPriv->y + (dPriv->h - x - 1)) * pitch));
+
+      if (cpp == 4)
+	 for (i = 0; i < n; i++, ptr -= pitch)
+	    READ_RGBA_32( rgba[i], ptr );
+      else
+	 for (i = 0; i < n; i++, ptr -= pitch)
+	    READ_RGBA_16( rgba[i], ptr );
+   }
+   else {
+      char *ptr = (char *)(rmesa->dri.screen->pFB +
+			   rmesa->state.pixel.readOffset +
+			   ((dPriv->x + x) * cpp) +
+			   ((dPriv->y + (dPriv->h - y - 1)) * pitch));
+
+      if (cpp == 4)
+	 for (i = 0; i < n; i++, ptr += cpp)
+	    READ_RGBA_32( rgba[i], ptr );
+      else
+	 for (i = 0; i < n; i++, ptr += cpp)
+	    READ_RGBA_16( rgba[i], ptr );
+   }
+}
+
+
+/**
+ * \brief Optimized glReadPixels().
+ *
+ * To be used with particular pixel formats GL_UNSIGNED_BYTE and GL_RGBA, when pixel
+ * scaling, biasing and mapping are disabled.
+ *
+ * \param x x start position of the reading rectangle.
+ * \param y y start position of the reading rectangle.
+ * \param width width of the reading rectangle.
+ * \param height height of the reading rectangle.
+ * \param format pixel format. Must be GL_RGBA.
+ * \param type pixel type. Must be GL_UNSIGNED_BYTE.
+ * \param pixels pixel data.
+ * 
+ * After asserting the above conditions, compensates for clipping and calls
+ * ReadRGBASpan() to read each row.
+ */
+void radeonReadPixels( GLint x, GLint y,
+		       GLsizei width, GLsizei height,
+		       GLenum format, GLenum type,
+		       GLvoid *pixels )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   GLint srcX = x;
+   GLint srcY = y;
+   GLint readWidth = width;           /* actual width read */
+   GLint readHeight = height;         /* actual height read */
+   const struct gl_pixelstore_attrib *packing = &ctx->Pack;
+   GLint skipRows = packing->SkipRows;
+   GLint skipPixels = packing->SkipPixels;
+   GLint rowLength;
+   ASSERT_OUTSIDE_BEGIN_END_AND_FLUSH(ctx);
+
+   {
+      GLint tmp, tmps;
+      tmp = x; x = y; y = tmp;
+      tmps = width; width = height; height = tmps;
+   }
+
+   if (width < 0 || height < 0) {
+      _mesa_error( ctx, GL_INVALID_VALUE,
+                   "glReadPixels(width=%d height=%d)", width, height );
+      return;
+   }
+
+   if (!pixels) {
+      _mesa_error( ctx, GL_INVALID_VALUE, "glReadPixels(pixels)" );
+      return;
+   }
+
+   if (ctx->NewState)
+      _mesa_update_state(ctx);
+
+
+   /* can't do scale, bias, mapping, etc */
+   assert(!ctx->_ImageTransferState);
+
+   /* can't do fancy pixel packing */
+   assert (packing->Alignment == 1 &&
+	   !packing->SwapBytes &&
+	   !packing->LsbFirst);
+
+
+   if (packing->RowLength > 0)
+      rowLength = packing->RowLength;
+   else
+      rowLength = width;
+
+   /* horizontal clipping */
+   if (srcX < 0) {
+      skipPixels -= srcX;
+      readWidth += srcX;
+      srcX = 0;
+   }
+   if (srcX + readWidth > (GLint) ctx->ReadBuffer->Width)
+      readWidth -= (srcX + readWidth - (GLint) ctx->ReadBuffer->Width);
+   if (readWidth <= 0)
+      return;
+
+   /* vertical clipping */
+   if (srcY < 0) {
+      skipRows -= srcY;
+      readHeight += srcY;
+      srcY = 0;
+   }
+   if (srcY + readHeight > (GLint) ctx->ReadBuffer->Height)
+      readHeight -= (srcY + readHeight - (GLint) ctx->ReadBuffer->Height);
+   if (readHeight <= 0)
+      return;
+
+   /*
+    * Ready to read!
+    * The window region at (destX, destY) of size (readWidth, readHeight)
+    * will be read back.
+    * We'll write pixel data to buffer pointed to by "pixels" but we'll
+    * skip "skipRows" rows and skip "skipPixels" pixels/row.
+    */
+   if (format == GL_RGBA && type == GL_UNSIGNED_BYTE) {
+      GLchan *dest = (GLchan *) pixels 
+	 + (skipRows * rowLength + skipPixels) * 4;
+      GLint row;
+
+      for (row=0; row<readHeight; row++) {
+	 ReadRGBASpan(ctx, readWidth, srcX, srcY, (GLchan (*)[4]) dest);
+	 dest += rowLength * 4;
+	 srcY++;
+      }
+   }
+   else {
+      /* can't do this format/type combination */
+      assert(0);
+   }
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_subset_select.c b/src/mesa/drivers/dri/radeon/radeon_subset_select.c
new file mode 100644
index 00000000000..bd5003ffe51
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_subset_select.c
@@ -0,0 +1,998 @@
+/**
+ * \file radeon_subset_select.c
+ * \brief Selection.
+ */
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  4.1
+ *
+ * Copyright (C) 1999-2002  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.
+ */
+
+/* $Id: radeon_subset_select.c,v 1.2 2003/08/22 20:11:45 brianp Exp $ */
+
+
+#include "glheader.h"
+#include "imports.h"
+#include "context.h"
+/*#include "mmath.h"*/
+#include "mtypes.h"
+#include "enums.h"
+#include "glapi.h"
+#include "feedback.h"
+
+#include "radeon_context.h"
+#include "radeon_subset.h"
+
+/**
+ * \brief Vertex.
+ */
+typedef struct {
+   struct { GLfloat x, y, z, w; } pos;      /**< \brief position */
+   struct { GLfloat x, y, z, w; } eyePos;   /**< \brief position, eye coordinates */
+   struct { GLfloat x, y, z, w; } clipPos;  /**< \brief clipped coordinates */
+   struct { GLfloat x, y, z, w; } winPos;   /**< \brief position, windows coordinates */
+   struct { GLfloat s, t; } texCoord;       /**< \brief texture coordinates */
+   struct { GLfloat r, g, b, a; } color;    /**< \brief color */
+} vertex;
+
+
+/**
+ * \brief Vertex buffer.
+ */
+static struct select_vb_t {
+   GLuint    vCount;		/**< \brief vertex count */
+   vertex  vBuffer[4];		/**< \brief vertex buffer */
+   GLboolean lineReset;
+   GLboolean partialLineLoop;	/**< \brief whether we are in a middle of a line loop */
+} vb;
+
+
+
+
+/**********************************************************************/
+/** \name        Vertex Transformation and Clipping                   */
+/**********************************************************************/
+/*@{*/
+
+/**
+ * \brief Transform a point (column vector) by a matrix:  Q = M * P.
+ *
+ * \param Q destination point.
+ * \param P source point.
+ * \param M transformation matrix.
+ */
+#define TRANSFORM_POINT( Q, M, P )				\
+   Q.x = M[0] * P.x + M[4] * P.y + M[8] *  P.z + M[12] * P.w;	\
+   Q.y = M[1] * P.x + M[5] * P.y + M[9] *  P.z + M[13] * P.w;	\
+   Q.z = M[2] * P.x + M[6] * P.y + M[10] * P.z + M[14] * P.w;	\
+   Q.w = M[3] * P.x + M[7] * P.y + M[11] * P.z + M[15] * P.w;
+
+/**
+ * \brief Clip coord to window coord mapping.
+ *
+ * \param Q destination point.
+ * \param P source point.
+ * \param VP view port.
+ */
+#define MAP_POINT( Q, P, VP )                                      \
+   Q.x = (GLfloat) (((P.x / P.w) + 1.0) * VP.Width / 2.0 + VP.X);  \
+   Q.y = (GLfloat) (((P.y / P.w) + 1.0) * VP.Height / 2.0 + VP.Y);  \
+   Q.z = (GLfloat) (((P.z / P.w) + 1.0) * (VP.Far - VP.Near) / 2.0 + VP.Near);\
+   Q.w = (GLfloat) P.w;
+
+
+/**
+ * \brief Linear interpolation:  (1 - T) * A + T * B.
+ *
+ * \param T interpolation factor.
+ * \param A first value.
+ * \param B second value.
+ * \result interpolated value.
+ */
+#define INTERPOLATE(T, A, B)   ((A) + ((B) - (A)) * (T))
+
+
+
+/**
+ * \brief Interpolate vertex position, color, texcoords, etc.
+ *
+ * \param t interpolation factor.
+ * \param v0 first vertex.
+ * \param v1 second vertex.
+ * \param vOut output vertex.
+ * 
+ * Uses the #INTERPOLATE macro for all the interpolation of all elements.
+ */
+static void
+interpolate_vertex(GLfloat t, const vertex *v0, const vertex *v1, 
+		   vertex *vOut)
+{
+   vOut->eyePos.x = INTERPOLATE(t, v0->eyePos.x, v1->eyePos.x);
+   vOut->eyePos.y = INTERPOLATE(t, v0->eyePos.y, v1->eyePos.y);
+   vOut->eyePos.z = INTERPOLATE(t, v0->eyePos.z, v1->eyePos.z);
+   vOut->eyePos.w = INTERPOLATE(t, v0->eyePos.w, v1->eyePos.w);
+
+   vOut->clipPos.x = INTERPOLATE(t, v0->clipPos.x, v1->clipPos.x);
+   vOut->clipPos.y = INTERPOLATE(t, v0->clipPos.y, v1->clipPos.y);
+   vOut->clipPos.z = INTERPOLATE(t, v0->clipPos.z, v1->clipPos.z);
+   vOut->clipPos.w = INTERPOLATE(t, v0->clipPos.w, v1->clipPos.w);
+
+   vOut->color.r = INTERPOLATE(t, v0->color.r, v1->color.r);
+   vOut->color.g = INTERPOLATE(t, v0->color.g, v1->color.g);
+   vOut->color.b = INTERPOLATE(t, v0->color.b, v1->color.b);
+   vOut->color.a = INTERPOLATE(t, v0->color.a, v1->color.a);
+
+   vOut->texCoord.s = INTERPOLATE(t, v0->texCoord.s, v1->texCoord.s);
+   vOut->texCoord.t = INTERPOLATE(t, v0->texCoord.t, v1->texCoord.t);
+}
+
+
+
+
+/*
+ * Clip bit codes 
+ */
+#define CLIP_LEFT    1
+#define CLIP_RIGHT   2
+#define CLIP_BOTTOM  4
+#define CLIP_TOP     8
+#define CLIP_NEAR   16
+#define CLIP_FAR    32
+
+
+/**
+ * \brief Apply view volume clip testing to a point.
+ *
+ * \param v point to test.
+ * \return zero if visible, or the clip code mask, i.e., binary OR of a
+ * combination of the #CLIP_LEFT, #CLIP_RIGHT, #CLIP_BOTTOM, #CLIP_TOP, #CLIP_NEAR,
+ * #CLIP_FAR clip bit codes.
+ */
+static GLuint
+clip_point(const vertex *v)
+{
+   GLuint mask = 0;
+   if (v->clipPos.x > v->clipPos.w) mask |= CLIP_RIGHT;
+   if (v->clipPos.x < -v->clipPos.w) mask |= CLIP_LEFT;
+   if (v->clipPos.y > v->clipPos.w) mask |= CLIP_TOP;
+   if (v->clipPos.y < -v->clipPos.w) mask |= CLIP_BOTTOM;
+   if (v->clipPos.z > v->clipPos.w) mask |= CLIP_FAR;
+   if (v->clipPos.z < -v->clipPos.w) mask |= CLIP_NEAR;
+   return mask;
+}
+
+
+/**
+ * \def GENERAL_CLIP
+ * \brief Clipping utility macro.
+ * 
+ * We use 6 instances of this code in each of the clip_line() and
+ * clip_polygon() to clip against the 6 planes.  For each plane, we define the
+ * #OUTSIDE and #COMPUTE_INTERSECTION macros appropriately.
+ */
+
+
+/**
+ * \brief Apply clipping to a line segment.
+ *
+ * \param v0in input start vertex
+ * \param v1in input end vertesx
+ * \param v0new output start vertex
+ * \param v1new output end vertex
+ *
+ * \return GL_TRUE if the line segment is visible, or GL_FALSE if it is totally
+ * clipped.
+ *
+ * \sa #GENERAL_CLIP.
+ */
+static GLboolean
+clip_line(const vertex *v0in, const vertex *v1in,
+	  vertex *v0new, vertex *v1new)
+{
+   vertex v0, v1, vNew;
+   GLfloat dx, dy, dz, dw, t;
+   GLuint code0, code1;
+
+   code0 = clip_point(v0in);
+   code1 = clip_point(v1in);
+   if (code0 & code1)
+      return GL_FALSE;  /* totally clipped */
+   
+   *v0new = *v0in;
+   *v1new = *v1in;
+   if (code0 == 0 && code1 == 0)
+      return GL_TRUE;   /* no clipping needed */
+   
+   v0 = *v0in;
+   v1 = *v1in;
+
+
+#define GENERAL_CLIP                                                    \
+   if (OUTSIDE(v0)) {                                                   \
+      if (OUTSIDE(v1)) {                                                \
+         /* both verts are outside ==> return 0 */                      \
+         return 0;                                                      \
+      }                                                                 \
+      else {                                                            \
+         /* v0 is outside, v1 is inside ==> clip */                     \
+         COMPUTE_INTERSECTION( v1, v0, vNew )                           \
+         interpolate_vertex(t, &v1, &v0, &vNew);                        \
+         v0 = vNew;                                                     \
+      }                                                                 \
+   }                                                                    \
+   else {                                                               \
+      if (OUTSIDE(v1)) {                                                \
+         /* v0 is inside, v1 is outside ==> clip */                     \
+         COMPUTE_INTERSECTION( v0, v1, vNew )                           \
+         interpolate_vertex(t, &v0, &v1, &vNew);                        \
+         v1 = vNew;                                                     \
+      }                                                                 \
+      /* else both verts are inside ==> do nothing */                   \
+   }
+
+   /* Clip against +X side */
+#define OUTSIDE(V)      (V.clipPos.x > V.clipPos.w)
+#define COMPUTE_INTERSECTION( IN, OUT, NEW )                         \
+        dx = OUT.clipPos.x - IN.clipPos.x;                           \
+        dw = OUT.clipPos.w - IN.clipPos.w;                           \
+        t = (IN.clipPos.x - IN.clipPos.w) / (dw-dx);
+   GENERAL_CLIP
+#undef OUTSIDE
+#undef COMPUTE_INTERSECTION
+
+   /* Clip against -X side */
+#define OUTSIDE(V)      (V.clipPos.x < -(V.clipPos.w))
+#define COMPUTE_INTERSECTION( IN, OUT, NEW )                         \
+        dx = OUT.clipPos.x - IN.clipPos.x;                           \
+        dw = OUT.clipPos.w - IN.clipPos.w;                           \
+        t = -(IN.clipPos.x + IN.clipPos.w) / (dw+dx);
+   GENERAL_CLIP
+#undef OUTSIDE
+#undef COMPUTE_INTERSECTION
+
+   /* Clip against +Y side */
+#define OUTSIDE(V)      (V.clipPos.y > V.clipPos.w)
+#define COMPUTE_INTERSECTION( IN, OUT, NEW )                         \
+        dy = OUT.clipPos.y - IN.clipPos.y;                           \
+        dw = OUT.clipPos.w - IN.clipPos.w;                           \
+        t = (IN.clipPos.y - IN.clipPos.w) / (dw-dy);
+   GENERAL_CLIP
+#undef OUTSIDE
+#undef COMPUTE_INTERSECTION
+
+   /* Clip against -Y side */
+#define OUTSIDE(V)      (V.clipPos.y < -(V.clipPos.w))
+#define COMPUTE_INTERSECTION( IN, OUT, NEW )                         \
+        dy = OUT.clipPos.y - IN.clipPos.y;                           \
+        dw = OUT.clipPos.w - IN.clipPos.w;                           \
+        t = -(IN.clipPos.y + IN.clipPos.w) / (dw+dy);
+   GENERAL_CLIP
+#undef OUTSIDE
+#undef COMPUTE_INTERSECTION
+
+   /* Clip against +Z side */
+#define OUTSIDE(V)      (V.clipPos.z > V.clipPos.w)
+#define COMPUTE_INTERSECTION( IN, OUT, NEW )                         \
+        dz = OUT.clipPos.z - IN.clipPos.z;                           \
+        dw = OUT.clipPos.w - IN.clipPos.w;                           \
+        t = (IN.clipPos.z - IN.clipPos.w) / (dw-dz);
+   GENERAL_CLIP
+#undef OUTSIDE
+#undef COMPUTE_INTERSECTION
+
+   /* Clip against -Z side */
+#define OUTSIDE(V)      (V.clipPos.z < -(V.clipPos.w))
+#define COMPUTE_INTERSECTION( IN, OUT, NEW )                         \
+        dz = OUT.clipPos.z - IN.clipPos.z;                           \
+        dw = OUT.clipPos.w - IN.clipPos.w;                           \
+        t = -(IN.clipPos.z + IN.clipPos.w) / (dw+dz);
+   GENERAL_CLIP
+#undef OUTSIDE
+#undef COMPUTE_INTERSECTION
+
+#undef GENERAL_CLIP
+
+   *v0new = v0;
+   *v1new = v1;
+   return GL_TRUE;
+}
+
+
+
+/**
+ * \brief Apply clipping to a polygon.
+ *
+ * \param vIn array of input vertices.
+ * \param inCount number of input vertices
+ * \param vOut array of output vertices.
+ *
+ * \return number of vertices in \p vOut.
+ *
+ * \sa #GENERAL_CLIP.
+ */
+static GLuint
+clip_polygon(const vertex *vIn, unsigned int inCount, vertex *vOut)
+{
+   vertex inlist[20], outlist[20];
+   GLfloat dx, dy, dz, dw, t;
+   GLuint incount, outcount, previ, curri, result;
+   const vertex *currVert, *prevVert;
+   vertex *newVert;
+
+
+#define GENERAL_CLIP(INCOUNT, INLIST, OUTCOUNT, OUTLIST)                \
+   if (INCOUNT < 3)                                                     \
+      return GL_FALSE;                                                  \
+   previ = INCOUNT - 1;         /* let previous = last vertex */        \
+   prevVert = INLIST + previ;                                           \
+   OUTCOUNT = 0;                                                        \
+   for (curri = 0; curri < INCOUNT; curri++) {                          \
+      currVert = INLIST + curri;                                        \
+      if (INSIDE(currVert)) {                                           \
+         if (INSIDE(prevVert)) {                                        \
+            /* both verts are inside ==> copy current to outlist */     \
+            OUTLIST[OUTCOUNT] = *currVert;                              \
+            OUTCOUNT++;                                                 \
+         }                                                              \
+         else {                                                         \
+            newVert = OUTLIST + OUTCOUNT;                               \
+            /* current is inside and previous is outside ==> clip */    \
+            COMPUTE_INTERSECTION( currVert, prevVert, newVert )         \
+            OUTCOUNT++;                                                 \
+            /* Output current */                                        \
+            OUTLIST[OUTCOUNT] = *currVert;                              \
+            OUTCOUNT++;                                                 \
+         }                                                              \
+      }                                                                 \
+      else {                                                            \
+         if (INSIDE(prevVert)) {                                        \
+            newVert = OUTLIST + OUTCOUNT;                               \
+            /* current is outside and previous is inside ==> clip */    \
+            COMPUTE_INTERSECTION( prevVert, currVert, newVert );        \
+            OUTLIST[OUTCOUNT] = *newVert;                               \
+            OUTCOUNT++;                                                 \
+         }                                                              \
+         /* else both verts are outside ==> do nothing */               \
+      }                                                                 \
+      /* let previous = current */                                      \
+      previ = curri;                                                    \
+      prevVert = currVert;                                              \
+   }
+
+/*
+ * Clip against +X
+ */
+#define INSIDE(V)       (V->clipPos.x <= V->clipPos.w)
+#define COMPUTE_INTERSECTION( IN, OUT, NEW )              \
+        dx = OUT->clipPos.x - IN->clipPos.x;              \
+        dw = OUT->clipPos.w - IN->clipPos.w;              \
+        t = (IN->clipPos.x - IN->clipPos.w) / (dw - dx);  \
+        interpolate_vertex(t, IN, OUT, NEW );
+
+   GENERAL_CLIP(inCount, vIn, outcount, outlist)
+
+#undef INSIDE
+#undef COMPUTE_INTERSECTION
+
+/*
+ * Clip against -X
+ */
+#define INSIDE(V)       (V->clipPos.x >= -V->clipPos.w)
+#define COMPUTE_INTERSECTION( IN, OUT, NEW )               \
+        dx = OUT->clipPos.x - IN->clipPos.x;               \
+        dw = OUT->clipPos.w - IN->clipPos.w;               \
+        t = -(IN->clipPos.x + IN->clipPos.w) / (dw + dx);  \
+        interpolate_vertex(t, IN, OUT, NEW );
+
+   GENERAL_CLIP(outcount, outlist, incount, inlist)
+
+#undef INSIDE
+#undef COMPUTE_INTERSECTION
+
+/*
+ * Clip against +Y
+ */
+#define INSIDE(V)       (V->clipPos.y <= V->clipPos.w)
+#define COMPUTE_INTERSECTION( IN, OUT, NEW )              \
+        dy = OUT->clipPos.y - IN->clipPos.y;              \
+        dw = OUT->clipPos.w - IN->clipPos.w;              \
+        t = (IN->clipPos.y - IN->clipPos.w) / (dw - dy);  \
+        interpolate_vertex(t, IN, OUT, NEW );
+
+   GENERAL_CLIP(incount, inlist, outcount, outlist)
+
+#undef INSIDE
+#undef COMPUTE_INTERSECTION
+
+/*
+ * Clip against -Y
+ */
+#define INSIDE(V)       (V->clipPos.y >= -V->clipPos.w)
+#define COMPUTE_INTERSECTION( IN, OUT, NEW )               \
+        dy = OUT->clipPos.y - IN->clipPos.y;               \
+        dw = OUT->clipPos.w - IN->clipPos.w;               \
+        t = -(IN->clipPos.y + IN->clipPos.w) / (dw + dy);  \
+        interpolate_vertex(t, IN, OUT, NEW );
+
+   GENERAL_CLIP(outcount, outlist, incount, inlist)
+
+#undef INSIDE
+#undef COMPUTE_INTERSECTION
+
+/*
+ * Clip against +Z
+ */
+#define INSIDE(V)       (V->clipPos.z <= V->clipPos.w)
+#define COMPUTE_INTERSECTION( IN, OUT, NEW )              \
+        dz = OUT->clipPos.z - IN->clipPos.z;              \
+        dw = OUT->clipPos.w - IN->clipPos.w;              \
+        t = (IN->clipPos.z - IN->clipPos.w) / (dw - dz);  \
+        interpolate_vertex(t, IN, OUT, NEW );
+
+   GENERAL_CLIP(incount, inlist, outcount, outlist)
+
+#undef INSIDE
+#undef COMPUTE_INTERSECTION
+
+/*
+ * Clip against -Z
+ */
+#define INSIDE(V)       (V->clipPos.z >= -V->clipPos.w)
+#define COMPUTE_INTERSECTION( IN, OUT, NEW )               \
+        dz = OUT->clipPos.z - IN->clipPos.z;               \
+        dw = OUT->clipPos.w - IN->clipPos.w;               \
+        t = -(IN->clipPos.z + IN->clipPos.w) / (dw + dz);  \
+        interpolate_vertex(t, IN, OUT, NEW );
+
+   GENERAL_CLIP(outcount, outlist, result, vOut)
+
+#undef INSIDE
+#undef COMPUTE_INTERSECTION
+
+#undef GENERAL_CLIP
+
+	return result;
+}
+
+/*@}*/
+
+
+
+/**********************************************************************/
+/** \name                    Selection                                */
+/**********************************************************************/
+/*@{*/
+
+/**
+ * \brief Select point.
+ * 
+ * \param v vertex.
+ *
+ * If the clipped point is visible then maps the vertex into window coordinates
+ * and calls _mesa_update_hitflag().
+ */
+static void
+select_point(const vertex *v)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   if (clip_point(v) == 0)
+   {
+      vertex c = *v;
+      MAP_POINT(c.winPos, c.clipPos, ctx->Viewport);
+      _mesa_update_hitflag(ctx, c.winPos.z);
+   }
+}
+
+/**
+ * \brief Select line.
+ * 
+ * \param v0 first vertex.
+ * \param v1 second vertex.
+ *
+ * If the clipped line is visible then maps the vertices into window coordinates
+ * and calls _mesa_update_hitflag().
+ */
+static void
+select_line(const vertex *v0, const vertex *v1)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   vertex c0, c1;
+   if (clip_line(v0, v1, &c0, &c1))
+   {
+      MAP_POINT(c0.winPos, c0.clipPos, ctx->Viewport);
+      MAP_POINT(c1.winPos, c1.clipPos, ctx->Viewport);
+      _mesa_update_hitflag(ctx, c0.winPos.z);
+      _mesa_update_hitflag(ctx, c1.winPos.z);
+   }
+}
+
+/**
+ * \brief Select line.
+ * 
+ * \param v0 first vertex.
+ * \param v1 second vertex.
+ * \param v2 third vertex.
+ *
+ * If the clipped polygon is visible then maps the vertices into window
+ * coordinates and calls _mesa_update_hitflag().
+ */
+static void
+select_triangle(const vertex *v0,
+		const vertex *v1,
+		const vertex *v2)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   vertex vlist[3], vclipped[8];
+   GLuint i, n;
+
+   vlist[0] = *v0;
+   vlist[1] = *v1;
+   vlist[2] = *v2;
+   n = clip_polygon(vlist, 3, vclipped);
+   for (i = 0; i < n; i++) {
+      MAP_POINT(vclipped[i].winPos, vclipped[i].clipPos, ctx->Viewport);
+      _mesa_update_hitflag(ctx, vclipped[i].winPos.z);
+   }
+}
+
+/**
+ * \brief Set current vertex coordinates.
+ *
+ * \param x x vertex coordinate.
+ * \param y y vertex coordinate.
+ * \param z z vertex coordinate.
+ * \param w homogeneous coordinate.
+ * 
+ * Stores the vertex and current attributes in ::vb, transforms it into eye space and then clip space.
+ * 
+ * If a sufficient number of vertices is stored calls one of select_point(),
+ * select_line() or select_triangle(), according to the current primitive.
+ */
+static void
+radeon_select_Vertex4f(GLfloat x, GLfloat y, GLfloat z, GLfloat w)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   struct gl_polygon_attrib *p = &(ctx->Polygon);
+   vertex *v = vb.vBuffer + vb.vCount;
+
+   /* store the vertex */
+   v->pos.x = x;
+   v->pos.y = y;
+   v->pos.z = z;
+   v->pos.w = w;
+   v->color.r = ctx->Current.Attrib[VERT_ATTRIB_COLOR0][0];  
+   v->color.g = ctx->Current.Attrib[VERT_ATTRIB_COLOR0][1];  
+   v->color.b = ctx->Current.Attrib[VERT_ATTRIB_COLOR0][2];  
+   v->color.a = ctx->Current.Attrib[VERT_ATTRIB_COLOR0][3];  
+   v->texCoord.s = ctx->Current.Attrib[VERT_ATTRIB_TEX0][0]; 
+   v->texCoord.t = ctx->Current.Attrib[VERT_ATTRIB_TEX0][1]; 
+
+   /* transform to eye space, then clip space */
+   TRANSFORM_POINT(v->eyePos, ctx->ModelviewMatrixStack.Top->m, v->pos);
+   TRANSFORM_POINT(v->clipPos, ctx->ProjectionMatrixStack.Top->m, v->eyePos);
+
+   switch (ctx->Driver.CurrentExecPrimitive) {
+   case GL_POINTS:
+      assert(vb.vCount == 0);
+      select_point(v);
+      break;
+   case GL_LINES:
+      if (vb.vCount == 0)
+      {
+	 vb.vCount = 1;
+      }
+      else
+      {
+	 assert(vb.vCount == 1);
+	 select_line(vb.vBuffer + 0, vb.vBuffer + 1);
+	 vb.vCount = 0;
+      }
+      break;
+   case GL_LINE_STRIP:
+      if (vb.vCount == 0)
+      {
+	 vb.vCount = 1;
+      }
+      else
+      {
+	 assert(vb.vCount == 1);
+	 select_line(vb.vBuffer + 0, vb.vBuffer + 1);
+	 vb.vBuffer[0] = vb.vBuffer[1];
+	 /* leave vb.vCount at 1 */
+      }
+      break;
+   case GL_LINE_LOOP:
+      if (vb.vCount == 0)
+      {
+	 vb.vCount = 1;
+	 vb.partialLineLoop = GL_FALSE;
+      }
+      else if (vb.vCount == 1)
+      {
+	 select_line(vb.vBuffer + 0, vb.vBuffer + 1);
+	 vb.partialLineLoop = GL_TRUE;
+	 vb.vCount = 2;
+      }
+      else
+      {
+	 assert(vb.vCount == 2);
+	 vb.partialLineLoop = GL_FALSE;
+	 select_line(vb.vBuffer + 1, vb.vBuffer + 2);
+	 vb.vBuffer[1] = vb.vBuffer[2];
+	 /* leave vb.vCount at 2 */
+      }
+      break;
+   case GL_TRIANGLES:
+      if (vb.vCount == 0 || vb.vCount == 1)
+      {
+	 vb.vCount++;
+      }
+      else
+      {
+	 assert(vb.vCount == 2);
+	 select_triangle(vb.vBuffer + 0, vb.vBuffer + 1, vb.vBuffer + 2);
+	 vb.vCount = 0;
+      }
+      break;
+   case GL_TRIANGLE_STRIP:
+      if (vb.vCount == 0 || vb.vCount == 1)
+      {
+	 vb.vCount++;
+      }
+      else if (vb.vCount == 2)
+      {
+	 select_triangle(vb.vBuffer + 0, vb.vBuffer + 1, vb.vBuffer + 2);
+	 vb.vCount = 3;
+      }
+      else
+      {
+	 assert(vb.vCount == 3);
+	 select_triangle(vb.vBuffer + 1, vb.vBuffer + 3, vb.vBuffer + 2);
+	 vb.vBuffer[0] = vb.vBuffer[2];
+	 vb.vBuffer[1] = vb.vBuffer[3];
+	 vb.vCount = 2;
+      }
+      break;		
+   case GL_TRIANGLE_FAN:
+      if (vb.vCount == 0 || vb.vCount == 1)
+      {
+	 vb.vCount++;
+      }
+      else
+      {
+	 assert(vb.vCount == 2);
+	 select_triangle(vb.vBuffer + 0, vb.vBuffer + 1, vb.vBuffer + 2);
+	 vb.vBuffer[1] = vb.vBuffer[2];
+	 /* leave vb.vCount = 2 */
+      }
+      break;
+   case GL_QUADS:
+      if (vb.vCount < 3)
+      {
+	 vb.vCount++;
+      }
+      else
+      {
+	 assert(vb.vCount == 3);
+	 select_triangle(vb.vBuffer + 0, vb.vBuffer + 1, vb.vBuffer + 2);
+	 select_triangle(vb.vBuffer + 0, vb.vBuffer + 2, vb.vBuffer + 3);
+	 vb.vCount = 0;
+      }
+      break;		
+   case GL_QUAD_STRIP:
+      if (vb.vCount < 3)
+      {
+	 vb.vCount++;
+      }
+      else
+      {
+	 assert(vb.vCount == 3);
+	 select_triangle(vb.vBuffer + 0, vb.vBuffer + 1, vb.vBuffer + 2);
+	 select_triangle(vb.vBuffer + 1, vb.vBuffer + 3, vb.vBuffer + 2);
+	 vb.vBuffer[0] = vb.vBuffer[2];
+	 vb.vBuffer[1] = vb.vBuffer[3];
+	 vb.vCount = 2;
+      }
+      break;		
+   case GL_POLYGON:
+      switch (p->FrontMode) {
+      case GL_POINT:
+	 assert(vb.vCount == 0);
+	 select_point(v);
+	 break;
+      case GL_LINE:
+	 if (vb.vCount == 0)
+	 {
+	    vb.vCount = 1;
+	    vb.partialLineLoop = GL_FALSE;
+	 }
+	 else if (vb.vCount == 1)
+	 {
+	    select_line(vb.vBuffer + 0, vb.vBuffer + 1);
+	    vb.partialLineLoop = GL_TRUE;
+	    vb.vCount = 2;
+	 }
+	 else
+	 {
+	    assert(vb.vCount == 2);
+	    vb.partialLineLoop = GL_FALSE;
+	    select_line(vb.vBuffer + 1, vb.vBuffer + 2);
+	    vb.vBuffer[1] = vb.vBuffer[2];
+	    /* leave vb.vCount at 2 */
+	 }
+	 break;
+      case GL_FILL:
+	 /* draw as a tri-fan */
+	 if (vb.vCount == 0 || vb.vCount == 1)
+	 {
+	    vb.vCount++;
+	 }
+	 else
+	 {
+	    assert(vb.vCount == 2);
+	    select_triangle(vb.vBuffer + 0, vb.vBuffer + 1, vb.vBuffer + 2);
+	    vb.vBuffer[1] = vb.vBuffer[2];
+	    /* leave vb.vCount = 2 */
+	 }
+	 break;
+      default:
+	 ; /* impossible */
+      }
+      break;
+   default:
+      ; /* outside begin/end -- no action required */
+   }
+}
+
+/**
+ * \brief Calls radeon_select_Vertex4f().
+ */
+static void radeon_select_Vertex2f(GLfloat x, GLfloat y)
+{
+   radeon_select_Vertex4f(x, y, 0.0, 1.0);
+}
+
+/**
+ * \brief Calls radeon_select_Vertex4f().
+ */
+static void radeon_select_Vertex2fv(const GLfloat * v)
+{
+   radeon_select_Vertex4f(v[0], v[1], 0.0, 1.0);
+}
+
+/**
+ * \brief Calls radeon_select_Vertex4f().
+ */
+static void radeon_select_Vertex3f(GLfloat x, GLfloat y, GLfloat z)
+{
+   radeon_select_Vertex4f(x, y, z, 1.0);
+}
+
+/**
+ * \brief Calls radeon_select_Vertex4f().
+ */
+static void radeon_select_Vertex3fv(const GLfloat * v)
+{
+   radeon_select_Vertex4f(v[0], v[1], v[2], 1.0);
+}
+
+
+/**
+ * \brief Set current vertex color.
+ *
+ * \param r red color component.
+ * \param g gree color component.
+ * \param b blue color component.
+ * \param a alpha color component.
+ *
+ * Updates the GL context's current vertex color.
+ */
+static void radeon_select_Color4f( GLfloat r, GLfloat g,
+				   GLfloat b, GLfloat a )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   GLfloat *dest = ctx->Current.Attrib[VERT_ATTRIB_COLOR0];
+   dest[0] = r;
+   dest[1] = g;
+   dest[2] = b;
+   dest[3] = a;
+}
+
+/**
+ * \brief Calls radeon_select_Color4f().
+ */
+static void radeon_select_Color4fv( const GLfloat *v )
+{
+   radeon_select_Color4f( v[0], v[1], v[2], v[3] );
+}
+
+/**
+ * \brief Calls radeon_select_Color4f().
+ */
+static void radeon_select_Color3f( GLfloat r, GLfloat g, GLfloat b )
+{
+   radeon_select_Color4f( r, g, b, 1.0 );
+}
+
+/**
+ * \brief Calls radeon_select_Color4f().
+ */
+static void radeon_select_Color3fv( const GLfloat *v )
+{
+   radeon_select_Color4f( v[0], v[1], v[2], 1.0 );
+}
+
+/**
+ * \brief Set current vertex texture coordinates.
+ *
+ * \param s texture coordinate.
+ * \param t texture coordinate.
+ *
+ * Updates the GL context's current vertex texture coordinates.
+ */
+static __inline__ void radeon_select_TexCoord2f( GLfloat s, GLfloat t )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   GLfloat *dest = ctx->Current.Attrib[VERT_ATTRIB_TEX0];
+   dest[0] = s;
+   dest[1] = t;
+}
+
+/**
+ * \brief Calls radeon_select_TexCoord2f().
+ */
+static void radeon_select_TexCoord2fv( const GLfloat *v )
+{
+   radeon_select_TexCoord2f( v[0], v[1] );
+}
+
+
+/**
+ * \brief Process glBegin().
+ *
+ * \param mode primitive.
+ */ 
+static void radeon_select_Begin(GLenum mode)
+{
+   GET_CURRENT_CONTEXT(ctx);
+
+   if (mode > GL_POLYGON) {
+      _mesa_error( ctx, GL_INVALID_ENUM, "glBegin" );
+      return;
+   }
+
+   if (ctx->Driver.CurrentExecPrimitive != GL_POLYGON+1) {
+      _mesa_error( ctx, GL_INVALID_OPERATION, "glBegin" );
+      return;
+   }
+
+   ctx->Driver.CurrentExecPrimitive = mode;
+
+   vb.vCount = 0;
+   vb.lineReset = GL_TRUE;
+   vb.partialLineLoop = GL_FALSE;
+}
+
+/**
+ * \brief Process glEnd().
+ */
+static void radeon_select_End(void)
+{
+   GET_CURRENT_CONTEXT(ctx);
+
+   if ( (ctx->Driver.CurrentExecPrimitive == GL_LINE_LOOP ||
+	 (ctx->Driver.CurrentExecPrimitive == GL_POLYGON && 
+	  ctx->Polygon.FrontMode == GL_LINE))
+	&& vb.vCount == 2 )
+   {
+      /* draw the last line segment */
+      if (vb.partialLineLoop)
+	 select_line(vb.vBuffer + 1, vb.vBuffer + 0);
+      else
+	 select_line(vb.vBuffer + 2, vb.vBuffer + 0);
+   }
+
+   ctx->Driver.CurrentExecPrimitive = GL_POLYGON+1;
+}
+
+
+/**
+ * \brief Flush vertices.
+ * 
+ * \param ctx GL context.
+ * \param flags not used.
+ *
+ * Nothing much to do here, besides marking the vertices as flushed, as we
+ * don't buffer anything.
+ */
+static void radeonSelectFlushVertices( GLcontext *ctx, GLuint flags )
+{
+   ctx->Driver.NeedFlush = 0;
+}
+
+/**
+ * \brief Install the select callbacks.
+ *
+ * \param ctx GL context.
+ *
+ * Installs the glBegin()/glEnd() associated select callbacks into the glapi
+ * table.
+ */
+void radeon_select_Install( GLcontext *ctx )
+{
+   struct _glapi_table *exec = ctx->Exec;
+
+   exec->Color3f = radeon_select_Color3f;
+   exec->Color3fv = radeon_select_Color3fv;
+   exec->Color4f = radeon_select_Color4f;
+   exec->Color4fv = radeon_select_Color4fv;
+   exec->TexCoord2f = radeon_select_TexCoord2f;
+   exec->TexCoord2fv = radeon_select_TexCoord2fv;
+   exec->Vertex2f = radeon_select_Vertex2f;
+   exec->Vertex2fv = radeon_select_Vertex2fv;
+   exec->Vertex3f = radeon_select_Vertex3f;
+   exec->Vertex3fv = radeon_select_Vertex3fv;
+   exec->Begin = radeon_select_Begin;
+   exec->End = radeon_select_End;
+
+   ctx->Driver.FlushVertices = radeonSelectFlushVertices;
+}
+/*@}*/
+
+
+
+/**
+ * \brief Set rasterization mode.
+ *
+ * \param ctx GL context.
+ * \param mode rasterization mode. Supports GL_RENDER or
+ *
+ * If mode is GL_RENDER, calls either radeonVtxfmtInit() or
+ * radeon_noop_Install depending on whether the application has focus
+ * (i.e., a fullscreen-cliprect) or not.  If mode is GL_SELECT, calls
+ * radeon_select_Install().
+ */
+static void radeonRenderMode( GLcontext *ctx, GLenum mode )
+{
+   switch (mode) {
+   case GL_RENDER:
+      radeonVtxfmtInit( ctx );
+      break;
+   case GL_SELECT:
+      radeon_select_Install( ctx );
+      break;
+   default:
+      break;
+   }
+}
+
+/**
+ * \brief Setup the GL context driver callbacks.
+ *
+ * \param ctx GL context.
+ *
+ * \sa Called by radeonCreateContext().
+ */
+void radeonInitSelect( GLcontext *ctx )
+{
+   ctx->Driver.RenderMode = radeonRenderMode;
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_subset_tex.c b/src/mesa/drivers/dri/radeon/radeon_subset_tex.c
new file mode 100644
index 00000000000..e401779513f
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_subset_tex.c
@@ -0,0 +1,1018 @@
+/**
+ * \file radeon_subset_tex.c
+ * \brief Texturing.
+ *
+ * \author Gareth Hughes <[email protected]>
+ * \author Brian Paul <[email protected]>
+ */
+
+/*
+ * Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+ *                      VA Linux Systems Inc., Fremont, California.
+ *
+ * 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
+ * ATI, VA LINUX SYSTEMS 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.
+ */
+
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_tex.c,v 1.6 2002/09/16 18:05:20 eich Exp $ */
+
+#include "glheader.h"
+#include "imports.h"
+#include "colormac.h"
+#include "context.h"
+#include "enums.h"
+#include "image.h"
+#include "simple_list.h"
+#include "texformat.h"
+#include "texstore.h"
+
+#include "radeon_context.h"
+#include "radeon_state.h"
+#include "radeon_ioctl.h"
+#include "radeon_subset.h"
+
+#include <errno.h>
+#include <stdio.h>
+
+
+/**
+ * \brief Destroy hardware state associated with a texture.
+ *
+ * \param rmesa Radeon context.
+ * \param t Radeon texture object to be destroyed.
+ *
+ * Frees the memory associated with the texture and if the texture is bound to
+ * a texture unit cleans the associated hardware state.
+ */
+void radeonDestroyTexObj( radeonContextPtr rmesa, radeonTexObjPtr t )
+{
+   if ( t->memBlock ) {
+      mmFreeMem( t->memBlock );
+      t->memBlock = NULL;
+   }
+
+   if ( t->tObj )
+      t->tObj->DriverData = NULL;
+
+   if ( rmesa ) {
+      if ( t == rmesa->state.texture.unit[0].texobj ) {
+         rmesa->state.texture.unit[0].texobj = NULL;
+	 remove_from_list( &rmesa->hw.tex[0] );
+	 make_empty_list( &rmesa->hw.tex[0] );
+      }
+   }
+
+   remove_from_list( t );
+   FREE( t );
+}
+
+
+/**
+ * \brief Keep track of swapped out texture objects.
+ * 
+ * \param rmesa Radeon context.
+ * \param t Radeon texture object.
+ *
+ * Frees the memory associated with the texture, marks all mipmap images in
+ * the texture as dirty and add it to the radeon_texture::swapped list.
+ */
+static void radeonSwapOutTexObj( radeonContextPtr rmesa, radeonTexObjPtr t )
+{
+   if ( t->memBlock ) {
+      mmFreeMem( t->memBlock );
+      t->memBlock = NULL;
+   }
+
+   t->dirty_images = ~0;
+   move_to_tail( &rmesa->texture.swapped, t );
+}
+
+
+/**
+ * Texture space has been invalidated.
+ *
+ * \param rmesa Radeon context.
+ * \param heap texture heap number.
+ * 
+ * Swaps out every texture in the specified heap.
+ */
+void radeonAgeTextures( radeonContextPtr rmesa, int heap )
+{
+   radeonTexObjPtr t, tmp;
+
+   foreach_s ( t, tmp, &rmesa->texture.objects[heap] ) 
+      radeonSwapOutTexObj( rmesa, t );
+}
+
+
+/***************************************************************/
+/** \name Texture image conversions
+ */
+/*@{*/
+
+/**
+ * \brief Upload texture image.
+ *
+ * \param rmesa Radeon context.
+ * \param t Radeon texture object.
+ * \param level level of the image to take the sub-image.
+ * \param x sub-image abscissa.
+ * \param y sub-image ordinate.
+ * \param width sub-image width.
+ * \param height sub-image height.
+ *
+ * Fills in a drmRadeonTexture and drmRadeonTexImage structures and uploads the
+ * texture via the DRM_RADEON_TEXTURE ioctl, aborting in case of failure.
+ */
+static void radeonUploadSubImage( radeonContextPtr rmesa,
+				  radeonTexObjPtr t, GLint level,
+				  GLint x, GLint y, GLint width, GLint height )
+{
+   struct gl_texture_image *texImage;
+   GLint ret;
+   drmRadeonTexture tex;
+   drmRadeonTexImage tmp;
+
+   level += t->firstLevel;
+   texImage = t->tObj->Image[level];
+
+   if ( !texImage || !texImage->Data ) 
+      return;
+
+   t->image[level].data = texImage->Data;
+
+   tex.offset = t->bufAddr;
+   tex.pitch = (t->image[0].width * texImage->TexFormat->TexelBytes) / 64;
+   tex.format = t->pp_txformat & RADEON_TXFORMAT_FORMAT_MASK;
+   tex.width = texImage->Width;
+   tex.height = texImage->Height;
+   tex.image = &tmp;
+
+   memcpy( &tmp, &t->image[level], sizeof(drmRadeonTexImage) );
+
+   do {
+      ret = drmCommandWriteRead( rmesa->dri.fd, DRM_RADEON_TEXTURE,
+                                 &tex, sizeof(drmRadeonTexture) );
+   } while ( ret && errno == EAGAIN );
+
+   if ( ret ) {
+      UNLOCK_HARDWARE( rmesa );
+      fprintf( stderr, "DRM_RADEON_TEXTURE: return = %d\n", ret );
+      exit( 1 );
+   }
+}
+
+/**
+ * \brief Upload texture images.
+ *
+ * This might require removing our own and/or other client's texture objects to
+ * make room for these images.
+ *
+ * \param rmesa Radeon context.
+ * \param tObj texture object to upload.
+ *
+ * Sets the matching hardware texture format. Calculates which mipmap levels to
+ * send, depending of the base image size, GL_TEXTURE_MIN_LOD,
+ * GL_TEXTURE_MAX_LOD, GL_TEXTURE_BASE_LEVEL, and GL_TEXTURE_MAX_LEVEL and the
+ * Radeon offset rules. Kicks out textures until the requested texture fits,
+ * sets the texture hardware state and, while holding the hardware lock,
+ * uploads any images that are new.
+ */
+static void radeonSetTexImages( radeonContextPtr rmesa,
+				struct gl_texture_object *tObj )
+{
+   radeonTexObjPtr t = (radeonTexObjPtr)tObj->DriverData;
+   const struct gl_texture_image *baseImage = tObj->Image[tObj->BaseLevel];
+   GLint totalSize;
+   GLint texelsPerDword = 0, blitWidth = 0, blitPitch = 0;
+   GLint x, y, width, height;
+   GLint i;
+   GLint firstLevel, lastLevel, numLevels;
+   GLint log2Width, log2Height;
+   GLuint txformat = 0;
+
+   /* This code cannot be reached once we have lost focus
+    */
+   assert(rmesa->radeonScreen->buffers);
+
+   /* Set the hardware texture format
+    */
+   switch (baseImage->TexFormat->MesaFormat) {
+   case MESA_FORMAT_I8:
+      txformat = RADEON_TXFORMAT_I8;
+      texelsPerDword = 4;
+      blitPitch = 64;
+      break;
+   case MESA_FORMAT_RGBA8888:
+      txformat = RADEON_TXFORMAT_RGBA8888 | RADEON_TXFORMAT_ALPHA_IN_MAP;
+      texelsPerDword = 1;
+      blitPitch = 16;
+      break;
+   case MESA_FORMAT_RGB565:
+      txformat = RADEON_TXFORMAT_RGB565;
+      texelsPerDword = 2;
+      blitPitch = 32;
+      break;
+   default:
+      _mesa_problem(NULL, "unexpected texture format in radeonTexImage2D");
+      return;
+   }
+
+   t->pp_txformat &= ~(RADEON_TXFORMAT_FORMAT_MASK |
+		       RADEON_TXFORMAT_ALPHA_IN_MAP);
+   t->pp_txformat |= txformat;
+
+
+   /* Select the larger of the two widths for our global texture image
+    * coordinate space.  As the Radeon has very strict offset rules, we
+    * can't upload mipmaps directly and have to reference their location
+    * from the aligned start of the whole image.
+    */
+   blitWidth = MAX2( baseImage->Width, blitPitch );
+
+   /* Calculate mipmap offsets and dimensions.
+    */
+   totalSize = 0;
+   x = 0;
+   y = 0;
+
+   /* Compute which mipmap levels we really want to send to the hardware.
+    * This depends on the base image size, GL_TEXTURE_MIN_LOD,
+    * GL_TEXTURE_MAX_LOD, GL_TEXTURE_BASE_LEVEL, and GL_TEXTURE_MAX_LEVEL.
+    * Yes, this looks overly complicated, but it's all needed.
+    */
+   firstLevel = tObj->BaseLevel + (GLint) (tObj->MinLod + 0.5);
+   firstLevel = MAX2(firstLevel, tObj->BaseLevel);
+   lastLevel = tObj->BaseLevel + (GLint) (tObj->MaxLod + 0.5);
+   lastLevel = MAX2(lastLevel, tObj->BaseLevel);
+   lastLevel = MIN2(lastLevel, tObj->BaseLevel + baseImage->MaxLog2);
+   lastLevel = MIN2(lastLevel, tObj->MaxLevel);
+   lastLevel = MAX2(firstLevel, lastLevel); /* need at least one level */
+
+   /* save these values */
+   t->firstLevel = firstLevel;
+   t->lastLevel = lastLevel;
+
+   numLevels = lastLevel - firstLevel + 1;
+
+   log2Width = tObj->Image[firstLevel]->WidthLog2;
+   log2Height = tObj->Image[firstLevel]->HeightLog2;
+
+   for ( i = 0 ; i < numLevels ; i++ ) {
+      const struct gl_texture_image *texImage = tObj->Image[i + firstLevel];
+      if ( !texImage )
+	 break;
+
+      width = texImage->Width;
+      height = texImage->Height;
+
+      /* Texture images have a minimum pitch of 32 bytes (half of the
+       * 64-byte minimum pitch for blits).  For images that have a
+       * width smaller than this, we must pad each texture image
+       * scanline out to this amount.
+       */
+      if ( width < blitPitch / 2 ) {
+	 width = blitPitch / 2;
+      }
+
+      totalSize += width * height * baseImage->TexFormat->TexelBytes;
+      ASSERT( (totalSize & 31) == 0 );
+
+      while ( width < blitWidth && height > 1 ) {
+	 width *= 2;
+	 height /= 2;
+      }
+
+      ASSERT(i < RADEON_MAX_TEXTURE_LEVELS);
+      t->image[i].x = x;
+      t->image[i].y = y;
+      t->image[i].width  = width;
+      t->image[i].height = height;
+
+      /* While blits must have a pitch of at least 64 bytes, mipmaps
+       * must be aligned on a 32-byte boundary (just like each texture
+       * image scanline).
+       */
+      if ( width >= blitWidth ) {
+	 y += height;
+      } else {
+	 x += width;
+	 if ( x >= blitWidth ) {
+	    x = 0;
+	    y++;
+	 }
+      }
+   }
+
+   /* Align the total size of texture memory block.
+    */
+   t->totalSize = (totalSize + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK;
+
+   /* Hardware state:
+    */
+   t->pp_txfilter &= ~RADEON_MAX_MIP_LEVEL_MASK;
+   t->pp_txfilter |= (numLevels - 1) << RADEON_MAX_MIP_LEVEL_SHIFT;
+
+   t->pp_txformat &= ~(RADEON_TXFORMAT_WIDTH_MASK |
+		       RADEON_TXFORMAT_HEIGHT_MASK);
+   t->pp_txformat |= ((log2Width << RADEON_TXFORMAT_WIDTH_SHIFT) |
+		      (log2Height << RADEON_TXFORMAT_HEIGHT_SHIFT));
+   t->dirty_state = TEX_ALL;
+
+   /* Update the local texture LRU.
+    */
+   move_to_head( &rmesa->texture.objects[0], t );
+
+   LOCK_HARDWARE( rmesa );
+
+   /* Kick out textures until the requested texture fits */
+   while ( !t->memBlock ) {
+      t->memBlock = mmAllocMem( rmesa->texture.heap[0], t->totalSize, 12, 0);
+	 
+      if (!t->memBlock)
+	 radeonSwapOutTexObj( rmesa, rmesa->texture.objects[0].prev );
+	 
+   }
+
+   /* Set the base offset of the texture image */
+   t->bufAddr = rmesa->radeonScreen->texOffset[0] + t->memBlock->ofs;
+   t->pp_txoffset = t->bufAddr;
+
+   /* Upload any images that are new 
+    */
+   for ( i = 0 ; i < numLevels ; i++ ) {
+      if ( t->dirty_images & (1 << i) ) {
+	 radeonUploadSubImage( rmesa, t, i, 0, 0,
+			       t->image[i].width, t->image[i].height );
+      }
+   }
+
+   rmesa->texture.age[0] = ++rmesa->sarea->texAge[0]; 
+   UNLOCK_HARDWARE( rmesa );
+   t->dirty_images = 0;
+}
+
+/*@}*/
+
+
+/******************************************************************/
+/** \name Texture combine functions
+ */
+/*@{*/
+
+enum {
+   RADEON_DISABLE	= 0, /**< \brief disabled */
+   RADEON_REPLACE	= 1, /**< \brief replace function */
+   RADEON_MODULATE	= 2, /**< \brief modulate function */
+   RADEON_DECAL		= 3, /**< \brief decal function */
+   RADEON_BLEND		= 4, /**< \brief blend function */
+   RADEON_MAX_COMBFUNC	= 5  /**< \brief max number of combine functions */
+} ;
+
+
+/**
+ * \brief Color combine function hardware state table.
+ */
+static GLuint radeon_color_combine[][RADEON_MAX_COMBFUNC] =
+{
+   /* Unit 0:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_CURRENT_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00802800
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T0_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x00800142
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T0_COLOR |
+       RADEON_COLOR_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x008c2d42
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T0_COLOR |
+       RADEON_COLOR_ARG_C_T0_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x008c2902
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_TFACTOR_COLOR |
+       RADEON_COLOR_ARG_C_T0_COLOR |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   },
+
+};
+
+/**
+ * \brief Alpha combine function hardware state table.
+ */
+static GLuint radeon_alpha_combine[][RADEON_MAX_COMBFUNC] =
+{
+   /* Unit 0:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00800500
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T0_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x00800051
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_T0_ALPHA |
+       RADEON_ALPHA_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x00800100
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x00800051
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_TFACTOR_ALPHA |
+       RADEON_ALPHA_ARG_C_T0_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+   },
+
+};
+
+/*@}*/
+
+
+/******************************************************************/
+/** \name Texture unit state management
+ */
+/*@{*/
+
+/**
+ * \brief Update the texture environment.
+ *
+ * \param ctx GL context
+ * \param unit texture unit to update.
+ *
+ * Sets the state of the RADEON_TEX_PP_TXCBLEND and RADEON_TEX_PP_TXABLEND
+ * registers using the ::radeon_color_combine and ::radeon_alpha_combine tables,
+ * and informs of the state change.
+ */
+static void radeonUpdateTextureEnv( GLcontext *ctx, int unit )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
+   const struct gl_texture_object *tObj = texUnit->_Current;
+   const GLenum format = tObj->Image[tObj->BaseLevel]->Format;
+   GLuint color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+   GLuint alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
+
+
+   /* Set the texture environment state.  Isn't this nice and clean?
+    * The Radeon will automagically set the texture alpha to 0xff when
+    * the texture format does not include an alpha component.  This
+    * reduces the amount of special-casing we have to do, alpha-only
+    * textures being a notable exception.
+    */
+   switch ( texUnit->EnvMode ) {
+   case GL_REPLACE:
+      switch ( format ) {
+      case GL_RGBA:
+      case GL_INTENSITY:
+	 color_combine = radeon_color_combine[unit][RADEON_REPLACE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_REPLACE];
+	 break;
+      case GL_RGB:
+	 color_combine = radeon_color_combine[unit][RADEON_REPLACE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
+         break;
+      default:
+	 break;
+      }
+      break;
+
+   case GL_MODULATE:
+      switch ( format ) {
+      case GL_RGBA:
+      case GL_INTENSITY:
+	 color_combine = radeon_color_combine[unit][RADEON_MODULATE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	 break;
+      case GL_RGB:
+	 color_combine = radeon_color_combine[unit][RADEON_MODULATE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
+	 break;
+      default:
+	 break;
+      }
+      break;
+
+   case GL_DECAL:
+      switch ( format ) {
+      case GL_RGBA:
+      case GL_RGB:
+	 color_combine = radeon_color_combine[unit][RADEON_DECAL];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
+	 break;
+      case GL_INTENSITY:
+	 color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
+	 break;
+      default:
+	 break;
+      }
+      break;
+
+   case GL_BLEND:
+      switch ( format ) {
+      case GL_RGBA:
+      case GL_RGB:
+	 color_combine = radeon_color_combine[unit][RADEON_BLEND];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	 break;
+      case GL_INTENSITY:
+	 color_combine = radeon_color_combine[unit][RADEON_BLEND];
+	 alpha_combine = radeon_alpha_combine[unit][RADEON_BLEND];
+	 break;
+      default:
+	 break;
+      }
+      break;
+
+   default:
+      break;
+   }
+
+   if ( rmesa->hw.tex[unit].cmd[TEX_PP_TXCBLEND] != color_combine ||
+	rmesa->hw.tex[unit].cmd[TEX_PP_TXABLEND] != alpha_combine ) {
+      RADEON_STATECHANGE( rmesa, tex[unit] );
+      rmesa->hw.tex[unit].cmd[TEX_PP_TXCBLEND] = color_combine;
+      rmesa->hw.tex[unit].cmd[TEX_PP_TXABLEND] = alpha_combine;
+   }
+}
+
+
+#define TEXOBJ_TXFILTER_MASK (RADEON_MAX_MIP_LEVEL_MASK |	\
+			      RADEON_MIN_FILTER_MASK | 		\
+			      RADEON_MAG_FILTER_MASK |		\
+			      RADEON_MAX_ANISO_MASK |		\
+			      RADEON_CLAMP_S_MASK | 		\
+			      RADEON_CLAMP_T_MASK)
+
+#define TEXOBJ_TXFORMAT_MASK (RADEON_TXFORMAT_WIDTH_MASK |	\
+			      RADEON_TXFORMAT_HEIGHT_MASK |	\
+			      RADEON_TXFORMAT_FORMAT_MASK |	\
+			      RADEON_TXFORMAT_ALPHA_IN_MAP)
+
+
+
+void radeonUpdateTextureState( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[0];
+
+   if ( texUnit->_ReallyEnabled & (TEXTURE_1D_BIT | TEXTURE_2D_BIT) ) {
+      struct gl_texture_object *tObj = texUnit->_Current;
+      radeonTexObjPtr t = (radeonTexObjPtr) tObj->DriverData;
+
+      /* Upload teximages (not pipelined)
+       */
+      if ( t->dirty_images ) {
+	 RADEON_FIREVERTICES( rmesa );
+	 radeonSetTexImages( rmesa, tObj );
+      }
+
+      /* Update state if this is a different texture object to last
+       * time.
+       */
+      if ( rmesa->state.texture.unit[0].texobj != t ) {
+	 rmesa->state.texture.unit[0].texobj = t;
+	 t->dirty_state |= 1<<0;
+	 move_to_head( &rmesa->texture.objects[0], t );
+      }
+
+      if (t->dirty_state) {
+	 GLuint *cmd = RADEON_DB_STATE( tex[0] );
+
+	 cmd[TEX_PP_TXFILTER] &= ~TEXOBJ_TXFILTER_MASK;
+	 cmd[TEX_PP_TXFORMAT] &= ~TEXOBJ_TXFORMAT_MASK;
+	 cmd[TEX_PP_TXFILTER] |= t->pp_txfilter & TEXOBJ_TXFILTER_MASK;
+	 cmd[TEX_PP_TXFORMAT] |= t->pp_txformat & TEXOBJ_TXFORMAT_MASK;
+	 cmd[TEX_PP_TXOFFSET] = t->pp_txoffset;
+	 cmd[TEX_PP_BORDER_COLOR] = t->pp_border_color;
+	 
+	 RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.tex[0] );
+	 t->dirty_state = 0;
+      }
+
+      /* Newly enabled?
+       */
+      if (!(rmesa->hw.ctx.cmd[CTX_PP_CNTL] & RADEON_TEX_0_ENABLE)) {
+	 RADEON_STATECHANGE( rmesa, ctx );
+	 rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= (RADEON_TEX_0_ENABLE | 
+					    RADEON_TEX_BLEND_0_ENABLE);
+
+	 RADEON_STATECHANGE( rmesa, tcl );
+	 rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_ST0;
+      }
+
+      radeonUpdateTextureEnv( ctx, 0 );
+   }
+   else if (rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (RADEON_TEX_0_ENABLE<<0)) {
+      /* Texture unit disabled */
+      rmesa->state.texture.unit[0].texobj = 0;
+      RADEON_STATECHANGE( rmesa, ctx );
+      rmesa->hw.ctx.cmd[CTX_PP_CNTL] &= 
+	 ~((RADEON_TEX_0_ENABLE | RADEON_TEX_BLEND_0_ENABLE) << 0);
+
+      RADEON_STATECHANGE( rmesa, tcl );
+      rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] &= ~(RADEON_TCL_VTX_ST0 |
+						RADEON_TCL_VTX_Q0);
+   }
+}
+
+
+
+/**
+ * \brief Choose texture format.
+ *
+ * \param ctx GL context.
+ * \param internalFormat texture internal format.
+ * \param format pixel format. Not used.
+ * \param type pixel data type. Not used.
+ *
+ * \return pointer to chosen texture format.
+ *
+ * Returns a pointer to one of the Mesa texture formats which is supported by
+ * Radeon and matches the internal format.
+ */
+static const struct gl_texture_format *
+radeonChooseTextureFormat( GLcontext *ctx, GLint internalFormat,
+                           GLenum format, GLenum type )
+{
+   switch ( internalFormat ) {
+   case GL_RGBA:
+   case GL_RGBA8:
+      return &_mesa_texformat_rgba8888;
+
+   case GL_RGB:
+   case GL_RGB5:
+      return &_mesa_texformat_rgb565;
+
+   case GL_INTENSITY:
+   case GL_INTENSITY8:
+      return &_mesa_texformat_i8;
+
+   default:
+      _mesa_problem(ctx, "unexpected texture format in radeonChoosTexFormat");
+      return NULL;
+   }
+}
+
+/**
+ * \brief Allocate a Radeon texture object.
+ *
+ * \param texObj texture object.
+ *
+ * \return pointer to the device specific texture object on success, or NULL on failure.
+ *
+ * Allocates and initializes a radeon_tex_obj structure to connect it to the
+ * driver private data pointer in \p texObj.
+ */
+static radeonTexObjPtr radeonAllocTexObj( struct gl_texture_object *texObj )
+{
+   radeonTexObjPtr t;
+
+   t = CALLOC_STRUCT( radeon_tex_obj );
+   if (!t)
+      return NULL;
+
+   t->tObj = texObj;
+   texObj->DriverData = t;
+   make_empty_list( t );
+   t->dirty_images = ~0;
+   return t;
+}
+
+
+/**
+ * \brief Load a texture image.
+ *
+ * \param ctx GL context.
+ * \param texObj texture object
+ * \param target target texture.
+ * \param level level of detail number.
+ * \param internalFormat internal format.
+ * \param width texture image width.
+ * \param height texture image height.
+ * \param border border width.
+ * \param format pixel format.
+ * \param type pixel data type.
+ * \param pixels image data.
+ * \param packing passed to _mesa_store_teximage2d() unchanged.
+ * \param texImage passed to _mesa_store_teximage2d() unchanged.
+ * 
+ * If there is a device specific texture object associated with the given
+ * texture object then swaps that texture out. Calls _mesa_store_teximage2d()
+ * with all other parameters unchanged.
+ */
+static void radeonTexImage2D( GLcontext *ctx, GLenum target, GLint level,
+                              GLint internalFormat,
+                              GLint width, GLint height, GLint border,
+                              GLenum format, GLenum type, const GLvoid *pixels,
+                              const struct gl_pixelstore_attrib *packing,
+                              struct gl_texture_object *texObj,
+                              struct gl_texture_image *texImage )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeonTexObjPtr t = (radeonTexObjPtr)texObj->DriverData;
+
+   if ( t ) 
+      radeonSwapOutTexObj( rmesa, t );
+
+   /* Note, this will call radeonChooseTextureFormat */
+   _mesa_store_teximage2d(ctx, target, level, internalFormat,
+                          width, height, border, format, type, pixels,
+                          &ctx->Unpack, texObj, texImage);
+}
+
+/**
+ * \brief Set texture environment parameters.
+ *
+ * \param ctx GL context.
+ * \param target texture environment.
+ * \param pname texture parameter. Accepted value is GL_TEXTURE_ENV_COLOR.
+ * \param param parameter value.
+ *
+ * Updates the current unit's RADEON_TEX_PP_TFACTOR register and informs of the
+ * state change.
+ */
+static void radeonTexEnv( GLcontext *ctx, GLenum target,
+			  GLenum pname, const GLfloat *param )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint unit = ctx->Texture.CurrentUnit;
+   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
+
+   switch ( pname ) {
+   case GL_TEXTURE_ENV_COLOR: {
+      GLubyte c[4];
+      GLuint envColor;
+      UNCLAMPED_FLOAT_TO_RGBA_CHAN( c, texUnit->EnvColor );
+      envColor = radeonPackColor( 4, c[0], c[1], c[2], c[3] );
+      if ( rmesa->hw.tex[unit].cmd[TEX_PP_TFACTOR] != envColor ) {
+	 RADEON_STATECHANGE( rmesa, tex[unit] );
+	 rmesa->hw.tex[unit].cmd[TEX_PP_TFACTOR] = envColor;
+      }
+      break;
+   }
+
+   default:
+      return;
+   }
+}
+
+/**
+ * \brief Set texture parameter.
+ *
+ * \param ctx GL context.
+ * \param target target texture.
+ * \param texObj texture object.
+ * \param pname texture parameter.
+ * \param params parameter value.
+ * 
+ * Allocates the device specific texture object data if it doesn't exist
+ * already.
+ * 
+ * Updates the texture object radeon_tex_obj::pp_txfilter register and marks
+ * the texture state (radeon_tex_obj::dirty_state) as dirty.
+ */
+static void radeonTexParameter( GLcontext *ctx, GLenum target,
+				struct gl_texture_object *texObj,
+				GLenum pname, const GLfloat *params )
+{
+   radeonTexObjPtr t = (radeonTexObjPtr) texObj->DriverData;
+   
+   if (!t)
+      t = radeonAllocTexObj( texObj );
+
+   switch ( pname ) {
+   case GL_TEXTURE_MIN_FILTER:
+      t->pp_txfilter &= ~RADEON_MIN_FILTER_MASK;
+      switch ( texObj->MinFilter ) {
+      case GL_NEAREST:
+	 t->pp_txfilter |= RADEON_MIN_FILTER_NEAREST;
+	 break;
+      case GL_LINEAR:
+	 t->pp_txfilter |= RADEON_MIN_FILTER_LINEAR;
+	 break;
+      case GL_NEAREST_MIPMAP_NEAREST:
+	 t->pp_txfilter |= RADEON_MIN_FILTER_NEAREST_MIP_NEAREST;
+	 break;
+      case GL_NEAREST_MIPMAP_LINEAR:
+	 t->pp_txfilter |= RADEON_MIN_FILTER_LINEAR_MIP_NEAREST;
+	 break;
+      case GL_LINEAR_MIPMAP_NEAREST:
+	 t->pp_txfilter |= RADEON_MIN_FILTER_NEAREST_MIP_LINEAR;
+	 break;
+      case GL_LINEAR_MIPMAP_LINEAR:
+	 t->pp_txfilter |= RADEON_MIN_FILTER_LINEAR_MIP_LINEAR;
+	 break;
+      }
+      break;
+
+   case GL_TEXTURE_MAG_FILTER:
+      t->pp_txfilter &= ~RADEON_MAG_FILTER_MASK;
+      switch ( texObj->MagFilter ) {
+      case GL_NEAREST:
+	 t->pp_txfilter |= RADEON_MAG_FILTER_NEAREST;
+	 break;
+      case GL_LINEAR:
+	 t->pp_txfilter |= RADEON_MAG_FILTER_LINEAR;
+	 break;
+      }
+      break;
+
+   case GL_TEXTURE_WRAP_S:
+      t->pp_txfilter &= ~RADEON_CLAMP_S_MASK;
+      switch ( texObj->WrapS ) {
+      case GL_REPEAT:
+	 t->pp_txfilter |= RADEON_CLAMP_S_WRAP;
+	 break;
+      case GL_CLAMP_TO_EDGE:
+	 t->pp_txfilter |= RADEON_CLAMP_S_CLAMP_LAST;
+	 break;
+      }
+      break;
+
+   case GL_TEXTURE_WRAP_T:
+      t->pp_txfilter &= ~RADEON_CLAMP_T_MASK;
+      switch ( texObj->WrapT ) {
+      case GL_REPEAT:
+	 t->pp_txfilter |= RADEON_CLAMP_T_WRAP;
+	 break;
+      case GL_CLAMP_TO_EDGE:
+	 t->pp_txfilter |= RADEON_CLAMP_T_CLAMP_LAST;
+	 break;
+      }
+      break;
+
+   default:
+      return;
+   }
+
+   /* Mark this texobj as dirty (one bit per tex unit)
+    */
+   t->dirty_state = TEX_ALL;
+}
+
+/**
+ * \brief Bind texture.
+ *
+ * \param ctx GL context.
+ * \param target not used.
+ * \param texObj texture object.
+ * 
+ * Allocates the device specific texture data if it doesn't exist already.
+ */
+static void radeonBindTexture( GLcontext *ctx, GLenum target,
+			       struct gl_texture_object *texObj )
+{
+   if ( !texObj->DriverData ) 
+      radeonAllocTexObj( texObj );
+}
+
+/**
+ * \brief Delete texture.
+ *
+ * \param ctx GL context.
+ * \param texObj texture object.
+ *
+ * Fires any outstanding vertices and destroy the device specific texture
+ * object.
+ */ 
+static void radeonDeleteTexture( GLcontext *ctx,
+				 struct gl_texture_object *texObj )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeonTexObjPtr t = (radeonTexObjPtr) texObj->DriverData;
+
+   if ( t ) {
+      if ( rmesa ) 
+         RADEON_FIREVERTICES( rmesa );
+      radeonDestroyTexObj( rmesa, t );
+   }
+}
+
+/**
+ * \brief Initialize context texture object data.
+ * 
+ * \param ctx GL context.
+ *
+ * Called by radeonInitTextureFuncs() to setup the context initial texture
+ * objects.
+ */
+static void radeonInitTextureObjects( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   struct gl_texture_object *texObj;
+   GLuint tmp = ctx->Texture.CurrentUnit;
+
+   ctx->Texture.CurrentUnit = 0;
+
+   texObj = ctx->Texture.Unit[0].Current2D;
+   radeonBindTexture( ctx, GL_TEXTURE_2D, texObj );
+   move_to_tail( &rmesa->texture.swapped,
+		 (radeonTexObjPtr)texObj->DriverData );
+
+
+   ctx->Texture.CurrentUnit = tmp;
+}
+
+/**
+ * \brief Setup the GL context driver callbacks.
+ *
+ * \param ctx GL context.
+ *
+ * \sa Called by radeonCreateContext().
+ */
+void radeonInitTextureFuncs( GLcontext *ctx )
+{
+   ctx->Driver.ChooseTextureFormat	= radeonChooseTextureFormat;
+   ctx->Driver.TexImage2D		= radeonTexImage2D;
+
+   ctx->Driver.BindTexture		= radeonBindTexture;
+   ctx->Driver.CreateTexture		= NULL; /* FIXME: Is this used??? */
+   ctx->Driver.DeleteTexture		= radeonDeleteTexture;
+   ctx->Driver.PrioritizeTexture	= NULL;
+   ctx->Driver.ActiveTexture		= NULL;
+   ctx->Driver.UpdateTexturePalette	= NULL;
+
+   ctx->Driver.TexEnv			= radeonTexEnv;
+   ctx->Driver.TexParameter		= radeonTexParameter;
+
+   radeonInitTextureObjects( ctx );
+}
+
+/*@}*/
diff --git a/src/mesa/drivers/dri/radeon/radeon_subset_vtx.c b/src/mesa/drivers/dri/radeon/radeon_subset_vtx.c
new file mode 100644
index 00000000000..aa6ec73d5b2
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_subset_vtx.c
@@ -0,0 +1,989 @@
+/**
+ * \file radeon_subset_vtx.c
+ * \brief Vertex buffering.
+ *
+ * \author Keith Whitwell <[email protected]>
+ */
+
+/*
+ * Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+ *                      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
+ * ATI, 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.
+ * 
+ */
+
+/* $XFree86$ */
+
+#include "glheader.h"
+#include "imports.h"
+#include "api_noop.h"
+#include "context.h"
+/*#include "mmath.h" */
+#include "mtypes.h"
+#include "enums.h"
+#include "glapi.h"
+#include "colormac.h"
+#include "state.h"
+
+#include "radeon_context.h"
+#include "radeon_state.h"
+#include "radeon_ioctl.h"
+#include "radeon_subset.h"
+
+/**
+ * \brief Union for vertex data.
+ */
+union vertex_dword { 
+	float f; /**< \brief floating point value */
+	int i;   /**< \brief integer point value */
+};
+
+
+/**
+ * \brief Maximum number of dwords per vertex.
+ *
+ * Defined as 10 to hold: \code xyzw rgba st \endcode
+ */
+#define MAX_VERTEX_DWORDS 10
+
+
+/**
+ * \brief Global vertex buffer data.
+ */
+static struct vb_t {
+   /**
+    * \brief Notification mechanism.  
+    *
+    * These are treated as a stack to allow us to do things like build quads in
+    * temporary storage and then emit them as triangles.
+    */
+   struct {
+      GLint vertspace;         /**< \brief free vertices count */
+      GLint initial_vertspace; /**< \brief total vertices count */
+      GLint *dmaptr;           /**< \brief */
+      void (*notify)( void );  /**< \brief notification callback */
+   } stack[2];
+
+   /**
+    * \brief Storage for current vertex.
+    */
+   union vertex_dword vertex[MAX_VERTEX_DWORDS];
+
+   /**
+    * \brief Temporary storage for quads, etc.
+    */
+   union vertex_dword vertex_store[MAX_VERTEX_DWORDS * 4];
+
+   /**
+    * \name Color/texture
+    *
+    * Pointers to either vertex or ctx->Current.Attrib, depending on whether
+    * color/texture participates in the current vertex.
+    */
+   /*@{*/
+   GLfloat *floatcolorptr; /**< \brief color */
+   GLfloat *texcoordptr;   /**< \brief texture */
+   /*@}*/
+
+   /**
+    * \brief Pointer to the GL context.
+    */
+   GLcontext *context;
+
+   /**
+    * \brief Active primitive.
+    *
+    * \note May differ from ctx->Driver.CurrentExecPrimitive.
+    */
+   /*@{*/
+   GLenum prim;          /**< \brief primitive */
+   GLuint vertex_format; /**< \brief vertex format */
+   GLint vertex_size;    /**< \brief vertex size */
+   GLboolean recheck;    /**< \brief set if it's needed to validate this information */
+   /*@}*/
+} vb;
+
+
+static void radeonFlushVertices( GLcontext *, GLuint );
+
+
+/**
+ * \brief Primitive information table.
+ */
+static struct prims_t { 
+   int start,  /**< \brief vertex count for the starting primitive */
+       incr,   /**< \brief vertex increment for a further primitive */
+       hwprim; /**< \brief hardware primitive */
+} prims[10] = {
+   { 1, 1, RADEON_CP_VC_CNTL_PRIM_TYPE_POINT },
+   { 2, 2, RADEON_CP_VC_CNTL_PRIM_TYPE_LINE }, 
+   { 2, 1, RADEON_CP_VC_CNTL_PRIM_TYPE_LINE_STRIP },
+   { 2, 1, RADEON_CP_VC_CNTL_PRIM_TYPE_LINE_STRIP },
+   { 3, 3, RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST },
+   { 3, 1, RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_STRIP },
+   { 3, 1, RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_FAN }, 
+   { 4, 4, RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST },
+   { 4, 2, RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_STRIP }, 
+   { 3, 1, RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_FAN }, 
+};
+
+
+/**
+ * \brief Finish the primitive in the vertex buffer.
+ *
+ * \param rmesa Radeon context.
+ *
+ * Truncates any redundant vertices off the end of the buffer, emit the
+ * remaining vertices and advances the current DMA region.
+ */
+static void finish_prim( radeonContextPtr rmesa )
+{
+   GLuint prim_end = vb.stack[0].initial_vertspace - vb.stack[0].vertspace;
+   
+   /* Too few vertices? (eg: 2 vertices for a triangles prim?)
+    */
+   if (prim_end < prims[vb.prim].start) 
+      return;
+
+   /* Drop redundant vertices off end of primitive.  (eg: 5 vertices
+    * for triangles prim?)
+    */
+   prim_end -= (prim_end - prims[vb.prim].start) % prims[vb.prim].incr;
+
+   radeonEmitVertexAOS( rmesa, vb.vertex_size, GET_START(&rmesa->dma.current) );
+
+   radeonEmitVbufPrim( rmesa, vb.vertex_format,
+		       prims[vb.prim].hwprim | rmesa->tcl.tcl_flag, 
+		       prim_end );
+
+   rmesa->dma.current.ptr = 
+      rmesa->dma.current.start += prim_end * vb.vertex_size * 4; 
+}
+
+
+/**
+ * \brief Copy a vertex from the current DMA region
+ *
+ * \param rmesa Radeon context.
+ * \param n vertex index relative to the current DMA region.
+ * \param dst destination pointer.
+ *
+ * Used internally by copy_dma_verts().
+ */
+static void copy_vertex( radeonContextPtr rmesa, GLuint n, GLfloat *dst )
+{
+   GLuint i;
+   GLfloat *src = (GLfloat *)(rmesa->dma.current.address + 
+			      rmesa->dma.current.ptr + 
+			      n * vb.vertex_size * 4);
+
+   for (i = 0 ; i < vb.vertex_size; i++) 
+      dst[i] = src[i];
+}
+
+
+/**
+ * \brief Copy last vertices from the current DMA buffer to resume in a new buffer.
+ *
+ * \param rmesa Radeon context.
+ * \param tmp destination buffer.
+ *
+ * Takes from the current DMA buffer the last vertices necessary to resume in a
+ * new buffer, according to the current primitive.  Uses internally
+ * copy_vertex() for the vertex copying.
+ * 
+ */
+static GLuint copy_dma_verts( radeonContextPtr rmesa, 
+			      GLfloat (*tmp)[MAX_VERTEX_DWORDS] )
+{
+   GLuint ovf, i;
+   GLuint nr = vb.stack[0].initial_vertspace - vb.stack[0].vertspace;
+
+   switch( vb.prim )
+   {
+   case GL_POINTS:
+      return 0;
+   case GL_LINES:
+      ovf = nr&1;
+      for (i = 0 ; i < ovf ; i++)
+	 copy_vertex( rmesa, nr-ovf+i, tmp[i] );
+      return i;
+   case GL_LINE_STRIP:
+      if (nr == 0) 
+	 return 0;
+      copy_vertex( rmesa, nr-1, tmp[0] );
+      return 1;
+   case GL_LINE_LOOP:
+   case GL_TRIANGLE_FAN:
+   case GL_POLYGON:
+      if (nr == 0) 
+	 return 0;
+      else if (nr == 1) {
+	 copy_vertex( rmesa, 0, tmp[0] );
+	 return 1;
+      } else {
+	 copy_vertex( rmesa, 0, tmp[0] );
+	 copy_vertex( rmesa, nr-1, tmp[1] );
+	 return 2;
+      }
+   case GL_TRIANGLES:
+      ovf = nr % 3;
+      for (i = 0 ; i < ovf ; i++)
+	 copy_vertex( rmesa, nr-ovf+i, tmp[i] );
+      return i;
+   case GL_QUADS:
+      ovf = nr % 4;
+      for (i = 0 ; i < ovf ; i++)
+	 copy_vertex( rmesa, nr-ovf+i, tmp[i] );
+      return i;
+   case GL_TRIANGLE_STRIP:
+   case GL_QUAD_STRIP:
+      ovf = MIN2(nr, 2);
+      for (i = 0 ; i < ovf ; i++)
+	 copy_vertex( rmesa, nr-ovf+i, tmp[i] );
+      return i;
+   default:
+      return 0;
+   }
+}
+
+static void notify_wrap_buffer( void );
+
+/**
+ * \brief Resets the vertex buffer notification mechanism.
+ *
+ * Fills in vb_t::stack with the values from the current DMA region in
+ * radeon_dma::current and sets the notification callback to
+ * notify_wrap_buffer().
+ */
+static void reset_notify( void )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( vb.context );
+
+   vb.stack[0].dmaptr = (int *)(rmesa->dma.current.address +
+				rmesa->dma.current.ptr);
+   vb.stack[0].vertspace = ((rmesa->dma.current.end - rmesa->dma.current.ptr) / 
+			    (vb.vertex_size * 4));
+   vb.stack[0].vertspace &= ~1;	/* even numbers only -- avoid tristrip parity */
+   vb.stack[0].initial_vertspace = vb.stack[0].vertspace;
+   vb.stack[0].notify = notify_wrap_buffer;
+}      
+
+/**
+ * \brief Full buffer notification callback.
+ *
+ * Makes a copy of the necessary vertices of the current buffer via
+ * copy_dma_verts(), gets and resets new buffer via radeon and re-emits the
+ * saved vertices.
+ */
+static void notify_wrap_buffer( void )
+{
+   GLcontext *ctx = vb.context;
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat tmp[3][MAX_VERTEX_DWORDS];
+   GLuint i, nrverts = 0;
+
+   /* Copy vertices out of dma:
+    */
+   nrverts = copy_dma_verts( rmesa, tmp );
+   finish_prim( rmesa );
+
+   /* Get new buffer
+    */
+   radeonRefillCurrentDmaRegion( rmesa );
+
+   /* Reset vertspace[0], dmaptr
+    */
+   reset_notify();
+
+   /* Reemit saved vertices
+    */
+   for (i = 0 ; i < nrverts; i++) {
+      memcpy( vb.stack[0].dmaptr, tmp[i], vb.vertex_size * 4 );
+      vb.stack[0].dmaptr += vb.vertex_size;
+      vb.stack[0].vertspace--;
+   }
+}
+
+
+static void notify_noop( void )
+{
+   vb.stack[0].dmaptr = (int *)vb.vertex;
+   vb.stack[0].notify = notify_noop;
+   vb.stack[0].vertspace = 1;
+}
+
+/**
+ * \brief Pop the notification mechanism stack.
+ *
+ * Simply copy the second stack array element into the first.
+ *
+ * \sa vb_t::stack and push_notify().
+ */
+static void pop_notify( void )
+{
+   vb.stack[0] = vb.stack[1];
+}
+
+/**
+ * \brief Push the notification mechanism stack.
+ *
+ * \param notify new notify callback for the stack head.
+ * \param space space available for vertices in \p store.
+ * \param store buffer where to store the vertices.
+ * 
+ * Copy the second stack array element into the first and makes the stack head
+ * use the given resources.
+ * 
+ * \sa vb_t::stack and pop_notify().
+ */
+static void push_notify( void (*notify)( void ), int space, 
+			 union vertex_dword *store )
+{
+   vb.stack[1] = vb.stack[0];
+   vb.stack[0].notify = notify;
+   vb.stack[0].initial_vertspace = space;
+   vb.stack[0].vertspace = space;
+   vb.stack[0].dmaptr = (int *)store;
+}
+
+
+/**
+ * \brief Emit a stored vertex (in vb_t::vertex_store) to DMA.
+ *
+ * \param v vertex index.
+ *
+ * Adds the vertex into the current vertex buffer and calls the notification
+ * callback vb_t::notify().
+ */
+static void emit_vertex( int v )
+{
+   int i, *tmp = (int *)vb.vertex_store + v * vb.vertex_size;
+   
+   for (i = 0 ; i < vb.vertex_size ; i++) 
+      *vb.stack[0].dmaptr++ = *tmp++;
+
+   if (--vb.stack[0].vertspace == 0)
+      vb.stack[0].notify();
+}
+
+
+/**
+ * \brief Emit a quad (in vb_t::vertex_store) to DMA as two triangles.
+ *
+ * \param v0 first vertex index.
+ * \param v1 second vertex index.
+ * \param v2 third vertex index.
+ * \param v3 fourth vertex index.
+ *
+ * Calls emit_vertex() to emit the triangles' vertices.
+ */
+static void emit_quad( int v0, int v1, int v2, int v3 )
+{
+   emit_vertex( v0 ); emit_vertex( v1 ); emit_vertex( v3 );
+   emit_vertex( v1 ); emit_vertex( v2 ); emit_vertex( v3 );
+}
+
+/**
+ * \brief Every fourth vertex in a quad primitive, this is called to emit it.
+ *
+ * Pops the notification stack, calls emit_quad() and pushes the notification
+ * stack again, with itself and the vb_t::vertex_store to process another four
+ * vertices.
+ */
+static void notify_quad( void )
+{
+   pop_notify();
+   emit_quad( 0, 1, 2, 3 ); 
+   push_notify( notify_quad, 4, vb.vertex_store );
+}
+
+static void notify_qstrip1( void );
+
+/**
+ * \brief After the 4th vertex, emit either a quad or a flipped quad each two
+ * vertices.
+ *
+ * Pops the notification stack, calls emit_quad() with the flipped vertices and
+ * pushes the notification stack again, with notify_qstrip1() and the
+ * vb_t::vertex_store to process another two vertices.
+ *
+ * \sa notify_qstrip1().
+ */
+static void notify_qstrip0( void )
+{
+   pop_notify();
+   emit_quad( 0, 1, 3, 2 );
+   push_notify( notify_qstrip1, 2, vb.vertex_store );
+}
+
+/**
+ * \brief After the 4th vertex, emit either a quad or a flipped quad each two
+ * vertices.
+ *
+ * Pops the notification stack, calls emit_quad() with the straight vertices
+ * and pushes the notification stack again, with notify_qstrip0() and the
+ * vb_t::vertex_store to process another two vertices.
+ *
+ * \sa notify_qstrip0().
+ */
+static void notify_qstrip1( void )
+{
+   pop_notify();
+   emit_quad( 2, 3, 1, 0 ); 
+   push_notify( notify_qstrip0, 2, vb.vertex_store + 2*vb.vertex_size );
+}
+
+/**
+ * \brief Emit the saved vertex (but hang on to it for later).
+ *
+ * Continue processing this primitive as a linestrip.
+ *
+ * Pops the notification stack and calls emit_quad with the first vertex.
+ */
+static void notify_lineloop0( void )
+{
+   pop_notify();
+   emit_vertex(0);
+}
+
+/**
+ * \brief Invalidate the current vertex format.
+ *
+ * \param ctx GL context.
+ *
+ * Sets the vb_t::recheck flag.
+ */
+void radeonVtxfmtInvalidate( GLcontext *ctx )
+{
+   vb.recheck = GL_TRUE;
+}
+
+
+/**
+ * \brief Validate the vertex format from the context.
+ *
+ * \param ctx GL context.
+ *
+ * Signals a new primitive and determines the appropriate vertex format and
+ * size. Points vb_t::floatcolorptr and vb_t::texcoordptr to the current vertex
+ * and sets them to the current color and texture attributes.
+ *
+ * Clears the vb_t::recheck flag on exit.
+ */
+static void radeonVtxfmtValidate( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+   GLuint ind = (RADEON_CP_VC_FRMT_Z |
+		 RADEON_CP_VC_FRMT_FPCOLOR | 
+		 RADEON_CP_VC_FRMT_FPALPHA);
+
+   if (ctx->Driver.NeedFlush)
+      ctx->Driver.FlushVertices( ctx, ctx->Driver.NeedFlush );
+
+   if (ctx->Texture.Unit[0]._ReallyEnabled) 
+      ind |= RADEON_CP_VC_FRMT_ST0;
+
+   RADEON_NEWPRIM(rmesa);
+   vb.vertex_format = ind;
+   vb.vertex_size = 3;
+
+   /* Would prefer to use ubyte floats in the vertex:
+    */
+   vb.floatcolorptr = &vb.vertex[vb.vertex_size].f;
+   vb.vertex_size += 4;
+   vb.floatcolorptr[0] = ctx->Current.Attrib[VERT_ATTRIB_COLOR0][0];
+   vb.floatcolorptr[1] = ctx->Current.Attrib[VERT_ATTRIB_COLOR0][1];
+   vb.floatcolorptr[2] = ctx->Current.Attrib[VERT_ATTRIB_COLOR0][2];
+   vb.floatcolorptr[3] = ctx->Current.Attrib[VERT_ATTRIB_COLOR0][3];
+   
+   if (ind & RADEON_CP_VC_FRMT_ST0) {
+      vb.texcoordptr = &vb.vertex[vb.vertex_size].f;
+      vb.vertex_size += 2;
+      vb.texcoordptr[0] = ctx->Current.Attrib[VERT_ATTRIB_TEX0][0];
+      vb.texcoordptr[1] = ctx->Current.Attrib[VERT_ATTRIB_TEX0][1];   
+   } 
+   else
+      vb.texcoordptr = ctx->Current.Attrib[VERT_ATTRIB_TEX0];
+
+   vb.recheck = GL_FALSE;
+   ctx->Driver.NeedFlush = FLUSH_UPDATE_CURRENT;
+}
+
+
+#define RESET_STIPPLE() do {			\
+   RADEON_STATECHANGE( rmesa, lin );		\
+   radeonEmitState( rmesa );			\
+} while (0)
+
+#define AUTO_STIPPLE( mode )  do {		\
+   RADEON_STATECHANGE( rmesa, lin );		\
+   if (mode)					\
+      rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] |=	\
+	 RADEON_LINE_PATTERN_AUTO_RESET;	\
+   else						\
+      rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] &=	\
+	 ~RADEON_LINE_PATTERN_AUTO_RESET;	\
+   radeonEmitState( rmesa );			\
+} while (0)
+
+
+/**
+ * \brief Process glBegin().
+ *
+ * \param mode primitive.
+ */
+static void radeon_Begin( GLenum mode )
+{
+   GLcontext *ctx = vb.context;
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint se_cntl;
+   
+   if (mode > GL_POLYGON) {
+      _mesa_error( ctx, GL_INVALID_ENUM, "glBegin" );
+      return;
+   }
+
+   if (ctx->Driver.CurrentExecPrimitive != GL_POLYGON+1) {
+      _mesa_error( ctx, GL_INVALID_OPERATION, "glBegin" );
+      return;
+   }
+   
+   if (ctx->NewState) 
+      _mesa_update_state( ctx );
+
+   if (rmesa->NewGLState)
+      radeonValidateState( ctx );
+
+   if (vb.recheck) 
+      radeonVtxfmtValidate( ctx );
+
+   /* Do we need to grab a new DMA region for the vertices?
+    */
+   if (rmesa->dma.current.ptr + 12*vb.vertex_size*4 > rmesa->dma.current.end) {
+      RADEON_NEWPRIM( rmesa );
+      radeonRefillCurrentDmaRegion( rmesa );
+   }
+
+   reset_notify();
+   vb.prim = ctx->Driver.CurrentExecPrimitive = mode;
+   se_cntl = rmesa->hw.set.cmd[SET_SE_CNTL] | RADEON_FLAT_SHADE_VTX_LAST;
+
+   if (ctx->Line.StippleFlag && 
+       (mode == GL_LINES || 
+	mode == GL_LINE_LOOP ||
+	mode == GL_LINE_STRIP))
+      RESET_STIPPLE();
+
+   switch( mode ) {
+   case GL_LINES:
+      if (ctx->Line.StippleFlag) 
+	 AUTO_STIPPLE( GL_TRUE );
+      break;
+   case GL_LINE_LOOP:
+      vb.prim = GL_LINE_STRIP;
+      push_notify( notify_lineloop0, 1, vb.vertex_store );
+      break;
+   case GL_QUADS:
+      vb.prim = GL_TRIANGLES;
+      push_notify( notify_quad, 4, vb.vertex_store );
+      break;
+   case GL_QUAD_STRIP:
+      if (ctx->_TriangleCaps & DD_FLATSHADE) {
+	 vb.prim = GL_TRIANGLES;
+	 push_notify( notify_qstrip0, 4, vb.vertex_store );
+      }
+      break;
+   case GL_POLYGON:
+      if (ctx->_TriangleCaps & DD_FLATSHADE)
+	 se_cntl &= ~RADEON_FLAT_SHADE_VTX_LAST;
+      break;
+   default:
+      break;
+   }
+
+   if (se_cntl != rmesa->hw.set.cmd[SET_SE_CNTL]) {
+      RADEON_STATECHANGE( rmesa, set );
+      rmesa->hw.set.cmd[SET_SE_CNTL] = se_cntl;
+   }
+}
+
+
+/**
+ * \brief Process glEnd().
+ *
+ */
+static void radeon_End( void )
+{
+   GLcontext *ctx = vb.context;
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if (ctx->Driver.CurrentExecPrimitive == GL_POLYGON+1) {
+      _mesa_error( ctx, GL_INVALID_OPERATION, "glEnd" );
+      return;
+   }
+
+   /* Need to finish a line loop?
+    */
+   if (ctx->Driver.CurrentExecPrimitive == GL_LINE_LOOP) 
+      emit_vertex( 0 );
+
+   /* Need to pop off quads/quadstrip/etc notification?
+    */
+   if (vb.stack[0].notify != notify_wrap_buffer)
+      pop_notify();
+
+   finish_prim( rmesa );
+
+   if (ctx->Driver.CurrentExecPrimitive == GL_LINES && ctx->Line.StippleFlag) 
+      AUTO_STIPPLE( GL_FALSE );
+	  
+   ctx->Driver.CurrentExecPrimitive = GL_POLYGON+1;
+   notify_noop();
+}
+
+
+
+/**
+ * \brief Flush vertices.
+ *
+ * \param ctx GL context.
+ * \param flags flags.
+ *
+ * If FLUSH_UPDATE_CURRENT is et in \p flags then the current vertex attributes
+ * in the GL context is updated from vb_t::floatcolorptr and vb_t::texcoordptr.
+ */
+static void radeonFlushVertices( GLcontext *ctx, GLuint flags )
+{
+   if (flags & FLUSH_UPDATE_CURRENT) {
+      ctx->Current.Attrib[VERT_ATTRIB_COLOR0][0] = vb.floatcolorptr[0];
+      ctx->Current.Attrib[VERT_ATTRIB_COLOR0][1] = vb.floatcolorptr[1];
+      ctx->Current.Attrib[VERT_ATTRIB_COLOR0][2] = vb.floatcolorptr[2];
+      ctx->Current.Attrib[VERT_ATTRIB_COLOR0][3] = vb.floatcolorptr[3];
+
+      if (vb.vertex_format & RADEON_CP_VC_FRMT_ST0) {
+	 ctx->Current.Attrib[VERT_ATTRIB_TEX0][0] = vb.texcoordptr[0];
+	 ctx->Current.Attrib[VERT_ATTRIB_TEX0][1] = vb.texcoordptr[1];
+	 ctx->Current.Attrib[VERT_ATTRIB_TEX0][2] = 0.0F;
+	 ctx->Current.Attrib[VERT_ATTRIB_TEX0][3] = 1.0F;
+      }
+   }
+
+   ctx->Driver.NeedFlush &= ~FLUSH_STORED_VERTICES;
+}
+
+
+/**
+ * \brief Set current vertex coordinates.
+ *
+ * \param x x vertex coordinate.
+ * \param y y vertex coordinate.
+ * \param z z vertex coordinate.
+ * 
+ * Set the current vertex coordinates. If run out of space in this buffer call
+ * the notification callback.
+ */
+static __inline__ void radeon_Vertex3f( GLfloat x, GLfloat y, GLfloat z )
+{
+   int i;
+
+   *vb.stack[0].dmaptr++ = *(int *)&x;
+   *vb.stack[0].dmaptr++ = *(int *)&y;
+   *vb.stack[0].dmaptr++ = *(int *)&z;
+
+   for (i = 3; i < vb.vertex_size; i++) 
+      *vb.stack[0].dmaptr++ = vb.vertex[i].i;
+
+   if (--vb.stack[0].vertspace == 0)
+      vb.stack[0].notify();
+}
+
+/**
+ * \brief Set current vertex color.
+ *
+ * \param r red color component.
+ * \param g gree color component.
+ * \param b blue color component.
+ * \param a alpha color component.
+ *
+ * Sets the current vertex color via vb_t::floatcolorptr.
+ */
+static __inline__  void radeon_Color4f( GLfloat r, GLfloat g,
+					GLfloat b, GLfloat a )
+{
+   GLfloat *dest = vb.floatcolorptr;
+   dest[0] = r;
+   dest[1] = g;
+   dest[2] = b;
+   dest[3] = a;
+}
+
+/**
+ * \brief Set current vertex texture coordinates.
+ *
+ * \param s texture coordinate.
+ * \param t texture coordinate.
+ *
+ * Sets the current vertex color via vb_t::texcoordptr.
+ */
+static __inline__ void radeon_TexCoord2f( GLfloat s, GLfloat t )
+{
+   GLfloat *dest = vb.texcoordptr;
+   dest[0] = s;
+   dest[1] = t;
+}
+
+/**
+ * Calls radeon_Vertex3f(), which is expanded inline by the compiler to be
+ * efficient.
+ */
+static void radeon_Vertex3fv( const GLfloat *v )
+{
+   radeon_Vertex3f( v[0], v[1], v[2] );
+}
+
+/**
+ * Calls radeon_Vertex3f(), which is expanded inline by the compiler to be
+ * efficient.
+ */
+static void radeon_Vertex2f( GLfloat x, GLfloat y )
+{
+   radeon_Vertex3f( x, y, 0 );
+}
+
+/**
+ * Calls radeon_Vertex3f(), which is expanded inline by the compiler to be
+ * efficient.
+ */
+static void radeon_Vertex2fv( const GLfloat *v )
+{
+   radeon_Vertex3f( v[0], v[1], 0 );
+}
+
+/**
+ * Calls radeon_Vertex3f(), which is expanded inline by the compiler to be
+ * efficient.
+ */
+static void radeon_Color4fv( const GLfloat *v )
+{
+   radeon_Color4f( v[0], v[1], v[2], v[3] );
+}
+
+/**
+ * Calls radeon_Color4f(), which is expanded inline by the compiler to be
+ * efficient.
+ */
+static void radeon_Color3f( GLfloat r, GLfloat g, GLfloat b )
+{
+   radeon_Color4f( r, g, b, 1.0 );
+}
+
+/**
+ * Calls radeon_Color4f(), which is expanded inline by the compiler to be
+ * efficient.
+ */
+static void radeon_Color3fv( const GLfloat *v )
+{
+   radeon_Color4f( v[0], v[1], v[2], 1.0 );
+}
+
+/**
+ * Calls radeon_TexCoord2f(), which is expanded inline by the compiler to be
+ * efficient.
+ */
+static void radeon_TexCoord2fv( const GLfloat *v )
+{
+   radeon_TexCoord2f( v[0], v[1] );
+}
+
+
+/**
+ * No-op.
+ */
+void radeonVtxfmtUnbindContext( GLcontext *ctx )
+{
+}
+
+/**
+ * No-op.
+ */
+void radeonVtxfmtMakeCurrent( GLcontext *ctx )
+{
+}
+
+/**
+ * No-op.
+ */
+void radeonVtxfmtDestroy( GLcontext *ctx )
+{
+}
+
+/**
+ * \brief Software rendering fallback.
+ *
+ * \param ctx GL context.
+ * \param bit fallback bitmask.
+ * \param mode enable or disable.
+ * 
+ * Does nothing except display a warning message if \p mode is set.
+ */
+void radeonFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
+{
+   if (mode)
+      fprintf(stderr, "Warning: hit nonexistant fallback path!\n");
+}
+
+/**
+ * \brief Software TCL fallback.
+ *
+ * \param ctx GL context.
+ * \param bit fallback bitmask.
+ * \param mode enable or disable.
+ * 
+ * Does nothing except display a warning message if \p mode is set.
+ */
+void radeonTclFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
+{
+   if (mode)
+      fprintf(stderr, "Warning: hit nonexistant fallback path!\n");
+}
+
+/**
+ * \brief Called by radeonPointsBitmap() to disable TCL.
+ *
+ * \param rmesa Radeon context.
+ * \param flag whether to enable or disable TCL.
+ * 
+ * Updates radeon_tcl_info::tcl_flag.
+ */
+void radeonSubsetVtxEnableTCL( radeonContextPtr rmesa, GLboolean flag )
+{
+   rmesa->tcl.tcl_flag = flag ? RADEON_CP_VC_CNTL_TCL_ENABLE : 0;
+}
+
+
+
+/**********************************************************************/
+/** \name        Noop mode for operation without focus                */
+/**********************************************************************/
+/*@{*/
+
+
+/**
+ * \brief Process glBegin().
+ *
+ * \param mode primitive.
+ */ 
+static void radeon_noop_Begin(GLenum mode)
+{
+   GET_CURRENT_CONTEXT(ctx);
+
+   if (mode > GL_POLYGON) {
+      _mesa_error( ctx, GL_INVALID_ENUM, "glBegin" );
+      return;
+   }
+
+   if (ctx->Driver.CurrentExecPrimitive != GL_POLYGON+1) {
+      _mesa_error( ctx, GL_INVALID_OPERATION, "glBegin" );
+      return;
+   }
+
+   ctx->Driver.CurrentExecPrimitive = mode;
+}
+
+/**
+ * \brief Process glEnd().
+ */
+static void radeon_noop_End(void)
+{
+   GET_CURRENT_CONTEXT(ctx);
+   ctx->Driver.CurrentExecPrimitive = GL_POLYGON+1;
+}
+
+
+/**
+ * \brief Install the noop callbacks.
+ *
+ * \param ctx GL context.
+ *
+ * Installs the noop callbacks into the glapi table.  These functions
+ * will not attempt to emit any DMA vertices, but will keep internal
+ * GL state updated.  Borrows heavily from the select code.
+ */
+static void radeon_noop_Install( GLcontext *ctx )
+{
+   ctx->Exec->Begin = radeon_noop_Begin;
+   ctx->Exec->End = radeon_noop_End;
+
+   vb.texcoordptr = ctx->Current.Attrib[VERT_ATTRIB_TEX0];
+   vb.floatcolorptr = ctx->Current.Attrib[VERT_ATTRIB_COLOR0];
+
+   notify_noop();
+}
+
+
+/**
+ * \brief Setup the GL context callbacks.
+ * 
+ * \param ctx GL context.
+ * 
+ * Setups the GL context callbacks and links _glapi_table entries related to
+ * the glBegin()/glEnd() pairs to the functions in this module.
+ * 
+ * Called by radeonCreateContext() and radeonRenderMode().
+ */
+void radeonVtxfmtInit( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   struct _glapi_table *exec = ctx->Exec;
+
+   exec->Color3f = radeon_Color3f;
+   exec->Color3fv = radeon_Color3fv;
+   exec->Color4f = radeon_Color4f;
+   exec->Color4fv = radeon_Color4fv;
+   exec->TexCoord2f = radeon_TexCoord2f;
+   exec->TexCoord2fv = radeon_TexCoord2fv;
+   exec->Vertex2f = radeon_Vertex2f;
+   exec->Vertex2fv = radeon_Vertex2fv;
+   exec->Vertex3f = radeon_Vertex3f;
+   exec->Vertex3fv = radeon_Vertex3fv;
+   exec->Begin = radeon_Begin;
+   exec->End = radeon_End;
+
+   vb.context = ctx;
+   
+   ctx->Driver.FlushVertices = radeonFlushVertices;
+   ctx->Driver.CurrentExecPrimitive = GL_POLYGON+1;
+
+   if (rmesa->radeonScreen->buffers) {
+      radeonVtxfmtValidate( ctx );
+      notify_noop();
+   }
+   else 
+      radeon_noop_Install( ctx );
+}
+
+
+/*@}*/
+
+
diff --git a/src/mesa/drivers/dri/radeon/radeon_swtcl.c b/src/mesa/drivers/dri/radeon/radeon_swtcl.c
new file mode 100644
index 00000000000..926b1523d62
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_swtcl.c
@@ -0,0 +1,1332 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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 (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 <[email protected]>
+ */
+
+#include "glheader.h"
+#include "mtypes.h"
+#include "colormac.h"
+#include "enums.h"
+#include "imports.h"
+#include "macros.h"
+
+#include "swrast_setup/swrast_setup.h"
+#include "math/m_translate.h"
+#include "tnl/tnl.h"
+#include "tnl/t_context.h"
+#include "tnl/t_imm_exec.h"
+#include "tnl/t_pipeline.h"
+
+#include "radeon_context.h"
+#include "radeon_ioctl.h"
+#include "radeon_state.h"
+#include "radeon_swtcl.h"
+#include "radeon_tcl.h"
+
+/***********************************************************************
+ *              Build render functions from dd templates               *
+ ***********************************************************************/
+
+
+#define RADEON_XYZW_BIT		0x01
+#define RADEON_RGBA_BIT		0x02
+#define RADEON_SPEC_BIT		0x04
+#define RADEON_TEX0_BIT		0x08
+#define RADEON_TEX1_BIT		0x10
+#define RADEON_PTEX_BIT		0x20
+#define RADEON_MAX_SETUP	0x40
+
+static void flush_last_swtcl_prim( radeonContextPtr rmesa  );
+static void flush_last_swtcl_prim_compat( radeonContextPtr rmesa );
+
+static struct {
+   void                (*emit)( GLcontext *, GLuint, GLuint, void *, GLuint );
+   interp_func		interp;
+   copy_pv_func	        copy_pv;
+   GLboolean           (*check_tex_sizes)( GLcontext *ctx );
+   GLuint               vertex_size;
+   GLuint               vertex_stride_shift;
+   GLuint               vertex_format;
+} setup_tab[RADEON_MAX_SETUP];
+
+
+#define TINY_VERTEX_FORMAT	        (RADEON_CP_VC_FRMT_XY |		\
+					 RADEON_CP_VC_FRMT_Z |		\
+					 RADEON_CP_VC_FRMT_PKCOLOR)
+
+#define NOTEX_VERTEX_FORMAT	        (RADEON_CP_VC_FRMT_XY |		\
+					 RADEON_CP_VC_FRMT_Z |		\
+					 RADEON_CP_VC_FRMT_W0 |		\
+					 RADEON_CP_VC_FRMT_PKCOLOR |	\
+					 RADEON_CP_VC_FRMT_PKSPEC)
+
+#define TEX0_VERTEX_FORMAT	        (RADEON_CP_VC_FRMT_XY |		\
+					 RADEON_CP_VC_FRMT_Z |		\
+					 RADEON_CP_VC_FRMT_W0 |		\
+					 RADEON_CP_VC_FRMT_PKCOLOR |	\
+					 RADEON_CP_VC_FRMT_PKSPEC |	\
+					 RADEON_CP_VC_FRMT_ST0)
+
+#define TEX1_VERTEX_FORMAT	        (RADEON_CP_VC_FRMT_XY |		\
+					 RADEON_CP_VC_FRMT_Z |		\
+					 RADEON_CP_VC_FRMT_W0 |		\
+					 RADEON_CP_VC_FRMT_PKCOLOR |	\
+					 RADEON_CP_VC_FRMT_PKSPEC |	\
+					 RADEON_CP_VC_FRMT_ST0 |	\
+					 RADEON_CP_VC_FRMT_ST1)
+
+#define PROJ_TEX1_VERTEX_FORMAT	        (RADEON_CP_VC_FRMT_XY |		\
+					 RADEON_CP_VC_FRMT_Z |		\
+					 RADEON_CP_VC_FRMT_W0 |		\
+					 RADEON_CP_VC_FRMT_PKCOLOR |	\
+					 RADEON_CP_VC_FRMT_PKSPEC |	\
+					 RADEON_CP_VC_FRMT_ST0 |	\
+					 RADEON_CP_VC_FRMT_Q0 |         \
+					 RADEON_CP_VC_FRMT_ST1 |	\
+					 RADEON_CP_VC_FRMT_Q1)
+
+#define TEX2_VERTEX_FORMAT 0
+#define TEX3_VERTEX_FORMAT 0
+#define PROJ_TEX3_VERTEX_FORMAT 0
+
+#define DO_XYZW (IND & RADEON_XYZW_BIT)
+#define DO_RGBA (IND & RADEON_RGBA_BIT)
+#define DO_SPEC (IND & RADEON_SPEC_BIT)
+#define DO_FOG  (IND & RADEON_SPEC_BIT)
+#define DO_TEX0 (IND & RADEON_TEX0_BIT)
+#define DO_TEX1 (IND & RADEON_TEX1_BIT)
+#define DO_TEX2 0
+#define DO_TEX3 0
+#define DO_PTEX (IND & RADEON_PTEX_BIT)
+
+#define VERTEX radeonVertex
+#define VERTEX_COLOR radeon_color_t
+#define GET_VIEWPORT_MAT() 0
+#define GET_TEXSOURCE(n)  n
+#define GET_VERTEX_FORMAT() RADEON_CONTEXT(ctx)->swtcl.vertex_format
+#define GET_VERTEX_STORE() RADEON_CONTEXT(ctx)->swtcl.verts
+#define GET_VERTEX_STRIDE_SHIFT() RADEON_CONTEXT(ctx)->swtcl.vertex_stride_shift
+#define GET_UBYTE_COLOR_STORE() &RADEON_CONTEXT(ctx)->UbyteColor
+#define GET_UBYTE_SPEC_COLOR_STORE() &RADEON_CONTEXT(ctx)->UbyteSecondaryColor
+
+#define HAVE_HW_VIEWPORT    1
+/* Tiny vertices don't seem to work atm - haven't looked into why.
+ */
+#define HAVE_HW_DIVIDE      (IND & ~(RADEON_XYZW_BIT|RADEON_RGBA_BIT))
+#define HAVE_TINY_VERTICES  1
+#define HAVE_RGBA_COLOR     1
+#define HAVE_NOTEX_VERTICES 1
+#define HAVE_TEX0_VERTICES  1
+#define HAVE_TEX1_VERTICES  1
+#define HAVE_TEX2_VERTICES  0
+#define HAVE_TEX3_VERTICES  0
+#define HAVE_PTEX_VERTICES  1
+
+#define CHECK_HW_DIVIDE    (!(ctx->_TriangleCaps & (DD_TRI_LIGHT_TWOSIDE| \
+                                                    DD_TRI_UNFILLED)))
+
+#define IMPORT_QUALIFIER
+#define IMPORT_FLOAT_COLORS radeon_import_float_colors
+#define IMPORT_FLOAT_SPEC_COLORS radeon_import_float_spec_colors
+
+#define INTERP_VERTEX setup_tab[RADEON_CONTEXT(ctx)->swtcl.SetupIndex].interp
+#define COPY_PV_VERTEX setup_tab[RADEON_CONTEXT(ctx)->swtcl.SetupIndex].copy_pv
+
+
+/***********************************************************************
+ *         Generate  pv-copying and translation functions              *
+ ***********************************************************************/
+
+#define TAG(x) radeon_##x
+#define IND ~0
+#include "tnl_dd/t_dd_vb.c"
+#undef IND
+
+
+/***********************************************************************
+ *             Generate vertex emit and interp functions               *
+ ***********************************************************************/
+
+#define IND (RADEON_XYZW_BIT|RADEON_RGBA_BIT)
+#define TAG(x) x##_wg
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (RADEON_XYZW_BIT|RADEON_RGBA_BIT|RADEON_TEX0_BIT)
+#define TAG(x) x##_wgt0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (RADEON_XYZW_BIT|RADEON_RGBA_BIT|RADEON_TEX0_BIT|RADEON_PTEX_BIT)
+#define TAG(x) x##_wgpt0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (RADEON_XYZW_BIT|RADEON_RGBA_BIT|RADEON_TEX0_BIT|RADEON_TEX1_BIT)
+#define TAG(x) x##_wgt0t1
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (RADEON_XYZW_BIT|RADEON_RGBA_BIT|RADEON_TEX0_BIT|RADEON_TEX1_BIT|\
+             RADEON_PTEX_BIT)
+#define TAG(x) x##_wgpt0t1
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (RADEON_XYZW_BIT|RADEON_RGBA_BIT|RADEON_SPEC_BIT)
+#define TAG(x) x##_wgfs
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (RADEON_XYZW_BIT|RADEON_RGBA_BIT|RADEON_SPEC_BIT|\
+	     RADEON_TEX0_BIT)
+#define TAG(x) x##_wgfst0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (RADEON_XYZW_BIT|RADEON_RGBA_BIT|RADEON_SPEC_BIT|\
+	     RADEON_TEX0_BIT|RADEON_PTEX_BIT)
+#define TAG(x) x##_wgfspt0
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (RADEON_XYZW_BIT|RADEON_RGBA_BIT|RADEON_SPEC_BIT|\
+	     RADEON_TEX0_BIT|RADEON_TEX1_BIT)
+#define TAG(x) x##_wgfst0t1
+#include "tnl_dd/t_dd_vbtmp.h"
+
+#define IND (RADEON_XYZW_BIT|RADEON_RGBA_BIT|RADEON_SPEC_BIT|\
+	     RADEON_TEX0_BIT|RADEON_TEX1_BIT|RADEON_PTEX_BIT)
+#define TAG(x) x##_wgfspt0t1
+#include "tnl_dd/t_dd_vbtmp.h"
+
+
+/***********************************************************************
+ *                         Initialization 
+ ***********************************************************************/
+
+static void init_setup_tab( void )
+{
+   init_wg();
+   init_wgt0();
+   init_wgpt0();
+   init_wgt0t1();
+   init_wgpt0t1();
+   init_wgfs();
+   init_wgfst0();
+   init_wgfspt0();
+   init_wgfst0t1();
+   init_wgfspt0t1();
+}
+
+
+
+void radeonPrintSetupFlags(char *msg, GLuint flags )
+{
+   fprintf(stderr, "%s(%x): %s%s%s%s%s%s\n",
+	   msg,
+	   (int)flags,
+	   (flags & RADEON_XYZW_BIT)      ? " xyzw," : "",
+	   (flags & RADEON_RGBA_BIT)     ? " rgba," : "",
+	   (flags & RADEON_SPEC_BIT)     ? " spec/fog," : "",
+	   (flags & RADEON_TEX0_BIT)     ? " tex-0," : "",
+	   (flags & RADEON_TEX1_BIT)     ? " tex-1," : "",
+	   (flags & RADEON_PTEX_BIT)     ? " proj-tex," : "");
+}
+
+
+static void radeonRenderStart( GLcontext *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+
+   if (!setup_tab[rmesa->swtcl.SetupIndex].check_tex_sizes(ctx)) {
+      GLuint ind = rmesa->swtcl.SetupIndex |= (RADEON_PTEX_BIT|RADEON_RGBA_BIT);
+
+      /* Projective textures are handled nicely; just have to change
+       * up to the new vertex format.
+       */
+      if (setup_tab[ind].vertex_format != rmesa->swtcl.vertex_format) {
+	 RADEON_NEWPRIM(rmesa);
+	 rmesa->swtcl.vertex_format = setup_tab[ind].vertex_format;
+	 rmesa->swtcl.vertex_size = setup_tab[ind].vertex_size;
+	 rmesa->swtcl.vertex_stride_shift = setup_tab[ind].vertex_stride_shift;
+      }
+
+      if (!(ctx->_TriangleCaps & (DD_TRI_LIGHT_TWOSIDE|DD_TRI_UNFILLED))) {
+	 tnl->Driver.Render.Interp = setup_tab[rmesa->swtcl.SetupIndex].interp;
+	 tnl->Driver.Render.CopyPV = setup_tab[rmesa->swtcl.SetupIndex].copy_pv;
+      }
+   }
+   
+   if (rmesa->dma.flush != 0 && 
+       rmesa->dma.flush != flush_last_swtcl_prim_compat &&
+       rmesa->dma.flush != flush_last_swtcl_prim)
+      rmesa->dma.flush( rmesa );
+}
+
+
+void radeonBuildVertices( GLcontext *ctx, GLuint start, GLuint count,
+			   GLuint newinputs )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+   GLubyte *v = ((GLubyte *)rmesa->swtcl.verts + 
+		 (start << rmesa->swtcl.vertex_stride_shift));
+   GLuint stride = 1 << rmesa->swtcl.vertex_stride_shift;
+
+   newinputs |= rmesa->swtcl.SetupNewInputs;
+   rmesa->swtcl.SetupNewInputs = 0;
+
+   if (!newinputs)
+      return;
+
+   setup_tab[rmesa->swtcl.SetupIndex].emit( ctx, start, count, v, stride );
+}
+
+void radeonChooseVertexState( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   GLuint ind = (RADEON_XYZW_BIT | RADEON_RGBA_BIT);
+
+   if (!rmesa->TclFallback || rmesa->Fallback)
+      return;
+
+   if (ctx->Fog.Enabled || (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR))
+      ind |= RADEON_SPEC_BIT;
+
+   if (ctx->Texture._EnabledUnits & 0x2)
+      /* unit 1 enabled */
+      ind |= RADEON_TEX0_BIT|RADEON_TEX1_BIT;
+   else if (ctx->Texture._EnabledUnits & 0x1)
+      /* unit 0 enabled */
+      ind |= RADEON_TEX0_BIT;
+
+   rmesa->swtcl.SetupIndex = ind;
+
+   if (ctx->_TriangleCaps & (DD_TRI_LIGHT_TWOSIDE|DD_TRI_UNFILLED)) {
+      tnl->Driver.Render.Interp = radeon_interp_extras;
+      tnl->Driver.Render.CopyPV = radeon_copy_pv_extras;
+   }
+   else {
+      tnl->Driver.Render.Interp = setup_tab[ind].interp;
+      tnl->Driver.Render.CopyPV = setup_tab[ind].copy_pv;
+   }
+
+   if (setup_tab[ind].vertex_format != rmesa->swtcl.vertex_format) {
+      RADEON_NEWPRIM(rmesa);
+      rmesa->swtcl.vertex_format = setup_tab[ind].vertex_format;
+      rmesa->swtcl.vertex_size = setup_tab[ind].vertex_size;
+      rmesa->swtcl.vertex_stride_shift = setup_tab[ind].vertex_stride_shift;
+   }
+
+   {
+      GLuint se_coord_fmt, needproj;
+
+      /* HW perspective divide is a win, but tiny vertex formats are a
+       * bigger one.
+       */
+      if (setup_tab[ind].vertex_format == TINY_VERTEX_FORMAT ||
+	  (ctx->_TriangleCaps & (DD_TRI_LIGHT_TWOSIDE|DD_TRI_UNFILLED))) {
+	 needproj = GL_TRUE;
+	 se_coord_fmt = (RADEON_VTX_XY_PRE_MULT_1_OVER_W0 |
+			 RADEON_VTX_Z_PRE_MULT_1_OVER_W0 |
+			 RADEON_TEX1_W_ROUTING_USE_Q1);
+      }
+      else {
+	 needproj = GL_FALSE;
+	 se_coord_fmt = (RADEON_VTX_W0_IS_NOT_1_OVER_W0 |
+			 RADEON_TEX1_W_ROUTING_USE_Q1);
+      }
+
+      if ( se_coord_fmt != rmesa->hw.set.cmd[SET_SE_COORDFMT] ) {
+	 RADEON_STATECHANGE( rmesa, set );
+	 rmesa->hw.set.cmd[SET_SE_COORDFMT] = se_coord_fmt;
+      }
+      _tnl_need_projected_coords( ctx, needproj );
+   }
+}
+
+
+/* Flush vertices in the current dma region.
+ */
+static void flush_last_swtcl_prim( radeonContextPtr rmesa  )
+{
+   if (RADEON_DEBUG & DEBUG_IOCTL)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   rmesa->dma.flush = 0;
+
+   if (rmesa->dma.current.buf) {
+      struct radeon_dma_region *current = &rmesa->dma.current;
+      GLuint current_offset = (rmesa->radeonScreen->agp_buffer_offset +
+			       current->buf->buf->idx * RADEON_BUFFER_SIZE + 
+			       current->start);
+
+      assert (!(rmesa->swtcl.hw_primitive & RADEON_CP_VC_CNTL_PRIM_WALK_IND));
+
+      assert (current->start + 
+	      rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 ==
+	      current->ptr);
+
+      if (rmesa->dma.current.start != rmesa->dma.current.ptr) {
+	 radeonEmitVertexAOS( rmesa,
+			      rmesa->swtcl.vertex_size,
+			      current_offset);
+
+	 radeonEmitVbufPrim( rmesa,
+			     rmesa->swtcl.vertex_format,
+			     rmesa->swtcl.hw_primitive,
+			     rmesa->swtcl.numverts);
+      }
+
+      rmesa->swtcl.numverts = 0;
+      current->start = current->ptr;
+   }
+}
+
+
+static void flush_last_swtcl_prim_compat( radeonContextPtr rmesa )
+{
+   struct radeon_dma_region *current = &rmesa->dma.current;
+
+   if (RADEON_DEBUG & DEBUG_IOCTL)
+      fprintf(stderr, "%s buf %p start %d ptr %d\n", 
+	      __FUNCTION__,
+	      current->buf,
+	      current->start,
+	      current->ptr);
+
+   assert (!(rmesa->swtcl.hw_primitive & RADEON_CP_VC_CNTL_PRIM_WALK_IND));
+   assert (current->start + 
+	   rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 ==
+	   current->ptr);
+   assert (current->start == 0);
+
+   rmesa->dma.flush = 0;
+
+   if (current->ptr && current->buf) {
+      assert (current->buf->refcount == 1);
+
+      radeonCompatEmitPrimitive( rmesa,
+				 rmesa->swtcl.vertex_format,
+				 rmesa->swtcl.hw_primitive,
+				 rmesa->swtcl.numverts);
+      
+      /* The buffer has been released:
+       */
+      FREE(current->buf);
+      current->buf = 0;
+      current->start = 0;
+      current->ptr = current->end;
+
+   }
+
+   rmesa->swtcl.numverts = 0;
+}
+
+
+/* Alloc space in the current dma region.
+ */
+static __inline void *radeonAllocDmaLowVerts( radeonContextPtr rmesa,
+					      int nverts, int vsize )
+{
+   GLuint bytes = vsize * nverts;
+
+   if ( rmesa->dma.current.ptr + bytes > rmesa->dma.current.end ) 
+      radeonRefillCurrentDmaRegion( rmesa );
+
+   if (!rmesa->dma.flush) {
+      rmesa->glCtx->Driver.NeedFlush |= FLUSH_STORED_VERTICES;
+      if (rmesa->dri.drmMinor == 1)
+	 rmesa->dma.flush = flush_last_swtcl_prim_compat;
+      else
+	 rmesa->dma.flush = flush_last_swtcl_prim;
+   }
+
+   assert( vsize == rmesa->swtcl.vertex_size * 4 );
+   assert( rmesa->dma.flush == flush_last_swtcl_prim ||
+	   rmesa->dma.flush == flush_last_swtcl_prim_compat);
+   assert (rmesa->dma.current.start + 
+	   rmesa->swtcl.numverts * rmesa->swtcl.vertex_size * 4 ==
+	   rmesa->dma.current.ptr);
+
+
+   {
+      char *head = rmesa->dma.current.address + rmesa->dma.current.ptr;
+      rmesa->dma.current.ptr += bytes;
+      rmesa->swtcl.numverts += nverts;
+      return head;
+   }
+
+}
+
+
+
+
+void radeon_emit_contiguous_verts( GLcontext *ctx, GLuint start, GLuint count )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint vertex_size = rmesa->swtcl.vertex_size * 4;
+   CARD32 *dest = radeonAllocDmaLowVerts( rmesa, count-start, vertex_size );
+   setup_tab[rmesa->swtcl.SetupIndex].emit( ctx, start, count, dest, 
+					    vertex_size );
+}
+
+
+
+void radeon_emit_indexed_verts( GLcontext *ctx, GLuint start, GLuint count )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   radeonAllocDmaRegionVerts( rmesa, 
+			      &rmesa->swtcl.indexed_verts, 
+			      count - start,
+			      rmesa->swtcl.vertex_size * 4, 
+			      64);
+
+   setup_tab[rmesa->swtcl.SetupIndex].emit( 
+      ctx, start, count, 
+      rmesa->swtcl.indexed_verts.address + rmesa->swtcl.indexed_verts.start, 
+      rmesa->swtcl.vertex_size * 4 );
+}
+
+
+/*
+ * Render unclipped vertex buffers by emitting vertices directly to
+ * dma buffers.  Use strip/fan hardware primitives where possible.
+ * Try to simulate missing primitives with indexed vertices.
+ */
+#define HAVE_POINTS      1
+#define HAVE_LINES       1
+#define HAVE_LINE_STRIPS 1
+#define HAVE_TRIANGLES   1
+#define HAVE_TRI_STRIPS  1
+#define HAVE_TRI_STRIP_1 0
+#define HAVE_TRI_FANS    1
+#define HAVE_QUADS       0
+#define HAVE_QUAD_STRIPS 0
+#define HAVE_POLYGONS    0
+#define HAVE_ELTS        1
+
+static const GLuint hw_prim[GL_POLYGON+1] = {
+   RADEON_CP_VC_CNTL_PRIM_TYPE_POINT,
+   RADEON_CP_VC_CNTL_PRIM_TYPE_LINE,
+   0,
+   RADEON_CP_VC_CNTL_PRIM_TYPE_LINE_STRIP,
+   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST,
+   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_STRIP,
+   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_FAN,
+   0,
+   0,
+   0
+};
+
+static __inline void radeonDmaPrimitive( radeonContextPtr rmesa, GLenum prim )
+{
+   RADEON_NEWPRIM( rmesa );
+   rmesa->swtcl.hw_primitive = hw_prim[prim];
+   assert(rmesa->dma.current.ptr == rmesa->dma.current.start);
+}
+
+static __inline void radeonEltPrimitive( radeonContextPtr rmesa, GLenum prim )
+{
+   RADEON_NEWPRIM( rmesa );
+   rmesa->swtcl.hw_primitive = hw_prim[prim] | RADEON_CP_VC_CNTL_PRIM_WALK_IND;
+}
+
+
+static void VERT_FALLBACK( GLcontext *ctx,
+			   GLuint start,
+			   GLuint count,
+			   GLuint flags )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   tnl->Driver.Render.PrimitiveNotify( ctx, flags & PRIM_MODE_MASK );
+   tnl->Driver.Render.BuildVertices( ctx, start, count, ~0 );
+   tnl->Driver.Render.PrimTabVerts[flags&PRIM_MODE_MASK]( ctx, start, count, flags );
+   RADEON_CONTEXT(ctx)->swtcl.SetupNewInputs = VERT_BIT_CLIP;
+}
+
+static void ELT_FALLBACK( GLcontext *ctx,
+			  GLuint start,
+			  GLuint count,
+			  GLuint flags )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   tnl->Driver.Render.PrimitiveNotify( ctx, flags & PRIM_MODE_MASK );
+   tnl->Driver.Render.BuildVertices( ctx, start, count, ~0 );
+   tnl->Driver.Render.PrimTabElts[flags&PRIM_MODE_MASK]( ctx, start, count, flags );
+   RADEON_CONTEXT(ctx)->swtcl.SetupNewInputs = VERT_BIT_CLIP;
+}
+
+
+#define LOCAL_VARS radeonContextPtr rmesa = RADEON_CONTEXT(ctx)
+#define ELTS_VARS  GLushort *dest
+#define INIT( prim ) radeonDmaPrimitive( rmesa, prim )
+#define ELT_INIT(prim) radeonEltPrimitive( rmesa, prim )
+#define NEW_PRIMITIVE()  RADEON_NEWPRIM( rmesa )
+#define NEW_BUFFER()  radeonRefillCurrentDmaRegion( rmesa )
+#define GET_CURRENT_VB_MAX_VERTS() \
+  (((int)rmesa->dma.current.end - (int)rmesa->dma.current.ptr) / (rmesa->swtcl.vertex_size*4))
+#define GET_SUBSEQUENT_VB_MAX_VERTS() \
+  ((RADEON_BUFFER_SIZE) / (rmesa->swtcl.vertex_size*4))
+
+#if RADEON_OLD_PACKETS
+# define GET_CURRENT_VB_MAX_ELTS() \
+  ((RADEON_CMD_BUF_SZ - (rmesa->store.cmd_used + 24)) / 2)
+#else
+# define GET_CURRENT_VB_MAX_ELTS() \
+  ((RADEON_CMD_BUF_SZ - (rmesa->store.cmd_used + 16)) / 2)
+#endif
+#define GET_SUBSEQUENT_VB_MAX_ELTS() \
+  ((RADEON_CMD_BUF_SZ - 1024) / 2)
+
+
+
+/* How do you extend an existing primitive?
+ */
+#define ALLOC_ELTS(nr)							\
+do {									\
+   if (rmesa->dma.flush == radeonFlushElts &&				\
+       rmesa->store.cmd_used + nr*2 < RADEON_CMD_BUF_SZ) {		\
+									\
+      dest = (GLushort *)(rmesa->store.cmd_buf +			\
+			  rmesa->store.cmd_used);			\
+      rmesa->store.cmd_used += nr*2;					\
+   }									\
+   else {								\
+      if (rmesa->dma.flush) {						\
+	 rmesa->dma.flush( rmesa );					\
+      }									\
+									\
+      radeonEmitVertexAOS( rmesa,					\
+			   rmesa->swtcl.vertex_size,			\
+			   (rmesa->radeonScreen->agp_buffer_offset +		\
+			    rmesa->swtcl.indexed_verts.buf->buf->idx * 	\
+			    RADEON_BUFFER_SIZE +			\
+			    rmesa->swtcl.indexed_verts.start));		\
+									\
+      dest = radeonAllocEltsOpenEnded( rmesa,				\
+				       rmesa->swtcl.vertex_format,	\
+				       rmesa->swtcl.hw_primitive,	\
+				       nr );				\
+   }									\
+} while (0)
+
+#define ALLOC_ELTS_NEW_PRIMITIVE(nr) ALLOC_ELTS( nr )
+
+#ifdef MESA_BIG_ENDIAN
+/* We could do without (most of) this ugliness if dest was always 32 bit word aligned... */
+#define EMIT_ELT(offset, x) do {				\
+	int off = offset + ( ( (GLuint)dest & 0x2 ) >> 1 );	\
+	GLushort *des = (GLushort *)( (GLuint)dest & ~0x2 );	\
+	(des)[ off + 1 - 2 * ( off & 1 ) ] = (GLushort)(x); } while (0)
+#else
+#define EMIT_ELT(offset, x) (dest)[offset] = (GLushort) (x)
+#endif
+#define EMIT_TWO_ELTS(offset, x, y)  *(GLuint *)(dest+offset) = ((y)<<16)|(x);
+#define INCR_ELTS( nr ) dest += nr
+#define RELEASE_ELT_VERTS() \
+  radeonReleaseDmaRegion( rmesa, &rmesa->swtcl.indexed_verts, __FUNCTION__ )
+#define EMIT_VERTS( ctx, j, nr ) \
+  radeon_emit_contiguous_verts(ctx, j, (j)+(nr))
+#define EMIT_INDEXED_VERTS( ctx, start, count ) \
+  radeon_emit_indexed_verts( ctx, start, count )
+
+
+#define TAG(x) radeon_dma_##x
+#include "tnl_dd/t_dd_dmatmp.h"
+
+
+/**********************************************************************/
+/*                          Render pipeline stage                     */
+/**********************************************************************/
+
+
+static GLboolean radeon_run_render( GLcontext *ctx,
+				    struct gl_pipeline_stage *stage )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   struct vertex_buffer *VB = &tnl->vb;
+   GLuint i, length, flags = 0;
+   render_func *tab = TAG(render_tab_verts);
+
+   if (rmesa->swtcl.indexed_verts.buf && (!VB->Elts || stage->changed_inputs)) 
+      RELEASE_ELT_VERTS();
+   	
+   if (VB->ClipOrMask ||	     /* No clipping */
+       rmesa->swtcl.RenderIndex != 0 ||    /* No per-vertex manipulations */
+       ctx->Line.StippleFlag)        /* GH: THIS IS A HACK!!! */
+      return GL_TRUE;		
+
+   if (rmesa->dri.drmMinor < 3) {
+      /* drm 1.1 doesn't support vertex primitives starting in the
+       * middle of a buffer.  It doesn't support sane indexed vertices
+       * either.  drm 1.2 fixes both of these problems, but we don't have a
+       * compatibility layer to that version yet.  
+       */
+      return GL_TRUE;
+   }
+
+   tnl->Driver.Render.Start( ctx );
+
+   if (VB->Elts) {
+      tab = TAG(render_tab_elts);
+      if (!rmesa->swtcl.indexed_verts.buf)
+	 if (!TAG(emit_elt_verts)(ctx, 0, VB->Count))
+	    return GL_TRUE;	/* too many vertices */
+   }
+
+   for (i = 0 ; !(flags & PRIM_LAST) ; i += length)
+   {
+      flags = VB->Primitive[i];
+      length = VB->PrimitiveLength[i];
+
+      if (RADEON_DEBUG & DEBUG_PRIMS)
+	 fprintf(stderr, "radeon_render.c: prim %s %d..%d\n", 
+		 _mesa_lookup_enum_by_nr(flags & PRIM_MODE_MASK), 
+		 i, i+length);
+
+      if (length)
+	 tab[flags & PRIM_MODE_MASK]( ctx, i, i + length, flags );
+   }
+
+   tnl->Driver.Render.Finish( ctx );
+
+   return GL_FALSE;		/* finished the pipe */
+}
+
+
+
+static void radeon_check_render( GLcontext *ctx,
+				 struct gl_pipeline_stage *stage )
+{
+   GLuint inputs = VERT_BIT_POS | VERT_BIT_CLIP | VERT_BIT_COLOR0;
+
+   if (ctx->RenderMode == GL_RENDER) {
+      if (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR)
+	 inputs |= VERT_BIT_COLOR1;
+
+      if (ctx->Texture.Unit[0]._ReallyEnabled)
+	 inputs |= VERT_BIT_TEX0;
+
+      if (ctx->Texture.Unit[1]._ReallyEnabled)
+	 inputs |= VERT_BIT_TEX1;
+
+      if (ctx->Fog.Enabled)
+	 inputs |= VERT_BIT_FOG;
+   }
+
+   stage->inputs = inputs;
+}
+
+
+static void dtr( struct gl_pipeline_stage *stage )
+{
+   (void)stage;
+}
+
+
+const struct gl_pipeline_stage _radeon_render_stage =
+{
+   "radeon render",
+   (_DD_NEW_SEPARATE_SPECULAR |
+    _NEW_TEXTURE|
+    _NEW_FOG|
+    _NEW_RENDERMODE),		/* re-check (new inputs) */
+   0,				/* re-run (always runs) */
+   GL_TRUE,			/* active */
+   0, 0,			/* inputs (set in check_render), outputs */
+   0, 0,			/* changed_inputs, private */
+   dtr,				/* destructor */
+   radeon_check_render,		/* check - initially set to alloc data */
+   radeon_run_render		/* run */
+};
+
+
+/**************************************************************************/
+
+/* Radeon texture rectangle expects coords in 0..1 range, not 0..dimension
+ * as in the extension spec.  Need to translate here.
+ *
+ * Note that swrast expects 0..dimension, so if a fallback is active,
+ * don't do anything.  (Maybe need to configure swrast to match hw)
+ */
+struct texrect_stage_data {
+   GLvector4f texcoord[MAX_TEXTURE_UNITS];
+};
+
+#define TEXRECT_STAGE_DATA(stage) ((struct texrect_stage_data *)stage->privatePtr)
+
+
+static GLboolean run_texrect_stage( GLcontext *ctx,
+				    struct gl_pipeline_stage *stage )
+{
+   struct texrect_stage_data *store = TEXRECT_STAGE_DATA(stage);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   struct vertex_buffer *VB = &tnl->vb;
+   GLuint i;
+
+   if (rmesa->Fallback)
+      return GL_TRUE;
+
+   for (i = 0 ; i < ctx->Const.MaxTextureUnits ; i++) {
+      if (!(ctx->Texture.Unit[i]._ReallyEnabled & TEXTURE_RECT_BIT))
+	 continue;
+   
+      if (stage->changed_inputs & VERT_BIT_TEX(i)) {
+	 struct gl_texture_object *texObj = ctx->Texture.Unit[i].CurrentRect;
+	 struct gl_texture_image *texImage = texObj->Image[texObj->BaseLevel];
+	 const GLfloat iw = 1.0/texImage->Width;
+	 const GLfloat ih = 1.0/texImage->Height;
+	 GLfloat *in = (GLfloat *)VB->TexCoordPtr[i]->data;
+	 GLint instride = VB->TexCoordPtr[i]->stride;
+	 GLfloat (*out)[4] = store->texcoord[i].data;
+	 GLint j;
+	 
+	 for (j = 0 ; j < VB->Count ; j++) {
+	    out[j][0] = in[0] * iw;
+	    out[j][1] = in[1] * ih;
+	    in = (GLfloat *)((GLubyte *)in + instride);
+	 }
+      }
+
+      VB->TexCoordPtr[i] = &store->texcoord[i];
+   }
+
+   return GL_TRUE;
+}
+
+
+/* Called the first time stage->run() is invoked.
+ */
+static GLboolean alloc_texrect_data( GLcontext *ctx,
+				     struct gl_pipeline_stage *stage )
+{
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+   struct texrect_stage_data *store;
+   GLuint i;
+
+   stage->privatePtr = CALLOC(sizeof(*store));
+   store = TEXRECT_STAGE_DATA(stage);
+   if (!store)
+      return GL_FALSE;
+
+   for (i = 0 ; i < ctx->Const.MaxTextureUnits ; i++)
+      _mesa_vector4f_alloc( &store->texcoord[i], 0, VB->Size, 32 );
+
+   /* Now run the stage.
+    */
+   stage->run = run_texrect_stage;
+   return stage->run( ctx, stage );
+}
+
+
+static void check_texrect( GLcontext *ctx,
+			   struct gl_pipeline_stage *stage )
+{
+   GLuint flags = 0;
+
+   if (ctx->Texture.Unit[0]._ReallyEnabled & TEXTURE_RECT_BIT)
+      flags |= VERT_BIT_TEX0;
+
+   if (ctx->Texture.Unit[1]._ReallyEnabled & TEXTURE_RECT_BIT)
+      flags |= VERT_BIT_TEX1;
+
+   stage->inputs = flags;
+   stage->outputs = flags;
+   stage->active = (flags != 0);
+}
+
+
+static void free_texrect_data( struct gl_pipeline_stage *stage )
+{
+   struct texrect_stage_data *store = TEXRECT_STAGE_DATA(stage);
+   GLuint i;
+
+   if (store) {
+      for (i = 0 ; i < MAX_TEXTURE_UNITS ; i++)
+	 if (store->texcoord[i].data)
+	    _mesa_vector4f_free( &store->texcoord[i] );
+      FREE( store );
+      stage->privatePtr = 0;
+   }
+}
+
+
+const struct gl_pipeline_stage _radeon_texrect_stage =
+{
+   "radeon texrect stage",			/* name */
+   _NEW_TEXTURE,	/* check_state */
+   _NEW_TEXTURE,	/* run_state */
+   GL_TRUE,				/* active? */
+   0,					/* inputs */
+   0,					/* outputs */
+   0,					/* changed_inputs */
+   NULL,				/* private data */
+   free_texrect_data,			/* destructor */
+   check_texrect,			/* check */
+   alloc_texrect_data,			/* run -- initially set to init */
+};
+
+
+/**************************************************************************/
+
+
+static const GLuint reduced_hw_prim[GL_POLYGON+1] = {
+   RADEON_CP_VC_CNTL_PRIM_TYPE_POINT,
+   RADEON_CP_VC_CNTL_PRIM_TYPE_LINE,
+   RADEON_CP_VC_CNTL_PRIM_TYPE_LINE,
+   RADEON_CP_VC_CNTL_PRIM_TYPE_LINE,
+   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST,
+   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST,
+   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST,
+   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST,
+   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST,
+   RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST
+};
+
+static void radeonRasterPrimitive( GLcontext *ctx, GLuint hwprim );
+static void radeonRenderPrimitive( GLcontext *ctx, GLenum prim );
+static void radeonResetLineStipple( GLcontext *ctx );
+
+
+/***********************************************************************
+ *                    Emit primitives as inline vertices               *
+ ***********************************************************************/
+
+#undef LOCAL_VARS
+#define CTX_ARG radeonContextPtr rmesa
+#define CTX_ARG2 rmesa
+#define GET_VERTEX_DWORDS() rmesa->swtcl.vertex_size
+#define ALLOC_VERTS( n, size ) radeonAllocDmaLowVerts( rmesa, n, size * 4 )
+#undef LOCAL_VARS
+#define LOCAL_VARS						\
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);		\
+   const GLuint shift = rmesa->swtcl.vertex_stride_shift;	\
+   const char *radeonverts = (char *)rmesa->swtcl.verts;
+#define VERT(x) (radeonVertex *)(radeonverts + (x << shift))
+#define VERTEX radeonVertex 
+#undef TAG
+#define TAG(x) radeon_##x
+#include "tnl_dd/t_dd_triemit.h"
+
+
+/***********************************************************************
+ *          Macros for t_dd_tritmp.h to draw basic primitives          *
+ ***********************************************************************/
+
+#define QUAD( a, b, c, d ) radeon_quad( rmesa, a, b, c, d )
+#define TRI( a, b, c )     radeon_triangle( rmesa, a, b, c )
+#define LINE( a, b )       radeon_line( rmesa, a, b )
+#define POINT( a )         radeon_point( rmesa, a )
+
+/***********************************************************************
+ *              Build render functions from dd templates               *
+ ***********************************************************************/
+
+#define RADEON_TWOSIDE_BIT	0x01
+#define RADEON_UNFILLED_BIT	0x02
+#define RADEON_OFFSET_BIT	0x04 /* drmMinor == 1 */
+#define RADEON_MAX_TRIFUNC	0x08
+
+
+static struct {
+   points_func	        points;
+   line_func		line;
+   triangle_func	triangle;
+   quad_func		quad;
+} rast_tab[RADEON_MAX_TRIFUNC];
+
+
+#define DO_FALLBACK  0
+#define DO_OFFSET   (IND & RADEON_OFFSET_BIT)
+#define DO_UNFILLED (IND & RADEON_UNFILLED_BIT)
+#define DO_TWOSIDE  (IND & RADEON_TWOSIDE_BIT)
+#define DO_FLAT      0
+#define DO_TRI       1
+#define DO_QUAD      1
+#define DO_LINE      1
+#define DO_POINTS    1
+#define DO_FULL_QUAD 1
+
+#define HAVE_RGBA   1
+#define HAVE_SPEC   1
+#define HAVE_INDEX  0
+#define HAVE_BACK_COLORS  0
+#define HAVE_HW_FLATSHADE 1
+#define TAB rast_tab
+
+#define DEPTH_SCALE 1.0
+#define UNFILLED_TRI unfilled_tri
+#define UNFILLED_QUAD unfilled_quad
+#define VERT_X(_v) _v->v.x
+#define VERT_Y(_v) _v->v.y
+#define VERT_Z(_v) _v->v.z
+#define AREA_IS_CCW( a ) (a < 0)
+#define GET_VERTEX(e) (rmesa->swtcl.verts + (e<<rmesa->swtcl.vertex_stride_shift))
+
+#define VERT_SET_RGBA( v, c )    v->ui[coloroffset] = LE32_TO_CPU(*(GLuint *)c)
+#define VERT_COPY_RGBA( v0, v1 ) v0->ui[coloroffset] = v1->ui[coloroffset]
+#define VERT_SAVE_RGBA( idx )    color[idx] = CPU_TO_LE32(v[idx]->ui[coloroffset])
+#define VERT_RESTORE_RGBA( idx ) v[idx]->ui[coloroffset] = LE32_TO_CPU(color[idx])
+
+#define VERT_SET_SPEC( v0, c )   if (havespec) {			\
+					v0->v.specular.red   = (c)[0];	\
+					v0->v.specular.green = (c)[1];	\
+					v0->v.specular.blue  = (c)[2]; }
+#define VERT_COPY_SPEC( v0, v1 ) if (havespec) {					\
+					v0->v.specular.red   = v1->v.specular.red;	\
+					v0->v.specular.green = v1->v.specular.green;	\
+					v0->v.specular.blue  = v1->v.specular.blue; }
+#define VERT_SAVE_SPEC( idx )    if (havespec) spec[idx] = CPU_TO_LE32(v[idx]->ui[5])
+#define VERT_RESTORE_SPEC( idx ) if (havespec) v[idx]->ui[5] = LE32_TO_CPU(spec[idx])
+
+#undef LOCAL_VARS
+#undef TAG
+#undef INIT
+
+#define LOCAL_VARS(n)							\
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);			\
+   GLuint color[n], spec[n];						\
+   GLuint coloroffset = (rmesa->swtcl.vertex_size == 4 ? 3 : 4);	\
+   GLboolean havespec = (rmesa->swtcl.vertex_size > 4);			\
+   (void) color; (void) spec; (void) coloroffset; (void) havespec;
+
+/***********************************************************************
+ *                Helpers for rendering unfilled primitives            *
+ ***********************************************************************/
+
+#define RASTERIZE(x) radeonRasterPrimitive( ctx, reduced_hw_prim[x] )
+#define RENDER_PRIMITIVE rmesa->swtcl.render_primitive
+#undef TAG
+#define TAG(x) x
+#include "tnl_dd/t_dd_unfilled.h"
+#undef IND
+
+
+/***********************************************************************
+ *                      Generate GL render functions                   *
+ ***********************************************************************/
+
+
+#define IND (0)
+#define TAG(x) x
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (RADEON_TWOSIDE_BIT)
+#define TAG(x) x##_twoside
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (RADEON_UNFILLED_BIT)
+#define TAG(x) x##_unfilled
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (RADEON_TWOSIDE_BIT|RADEON_UNFILLED_BIT)
+#define TAG(x) x##_twoside_unfilled
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (RADEON_OFFSET_BIT)
+#define TAG(x) x##_offset
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (RADEON_TWOSIDE_BIT|RADEON_OFFSET_BIT)
+#define TAG(x) x##_twoside_offset
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (RADEON_UNFILLED_BIT|RADEON_OFFSET_BIT)
+#define TAG(x) x##_unfilled_offset
+#include "tnl_dd/t_dd_tritmp.h"
+
+#define IND (RADEON_TWOSIDE_BIT|RADEON_UNFILLED_BIT|RADEON_OFFSET_BIT)
+#define TAG(x) x##_twoside_unfilled_offset
+#include "tnl_dd/t_dd_tritmp.h"
+
+
+static void init_rast_tab( void )
+{
+   init();
+   init_twoside();
+   init_unfilled();
+   init_twoside_unfilled();
+   init_offset();
+   init_twoside_offset();
+   init_unfilled_offset();
+   init_twoside_unfilled_offset();
+}
+
+/**********************************************************************/
+/*               Render unclipped begin/end objects                   */
+/**********************************************************************/
+
+#define VERT(x) (radeonVertex *)(radeonverts + (x << shift))
+#define RENDER_POINTS( start, count )		\
+   for ( ; start < count ; start++)		\
+      radeon_point( rmesa, VERT(start) )
+#define RENDER_LINE( v0, v1 ) \
+   radeon_line( rmesa, VERT(v0), VERT(v1) )
+#define RENDER_TRI( v0, v1, v2 )  \
+   radeon_triangle( rmesa, VERT(v0), VERT(v1), VERT(v2) )
+#define RENDER_QUAD( v0, v1, v2, v3 ) \
+   radeon_quad( rmesa, VERT(v0), VERT(v1), VERT(v2), VERT(v3) )
+#undef INIT
+#define INIT(x) do {					\
+   radeonRenderPrimitive( ctx, x );			\
+} while (0)
+#undef LOCAL_VARS
+#define LOCAL_VARS						\
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);		\
+   const GLuint shift = rmesa->swtcl.vertex_stride_shift;		\
+   const char *radeonverts = (char *)rmesa->swtcl.verts;		\
+   const GLuint * const elt = TNL_CONTEXT(ctx)->vb.Elts;	\
+   const GLboolean stipple = ctx->Line.StippleFlag;		\
+   (void) elt; (void) stipple;
+#define RESET_STIPPLE	if ( stipple ) radeonResetLineStipple( ctx );
+#define RESET_OCCLUSION
+#define PRESERVE_VB_DEFS
+#define ELT(x) (x)
+#define TAG(x) radeon_##x##_verts
+#include "tnl/t_vb_rendertmp.h"
+#undef ELT
+#undef TAG
+#define TAG(x) radeon_##x##_elts
+#define ELT(x) elt[x]
+#include "tnl/t_vb_rendertmp.h"
+
+
+
+/**********************************************************************/
+/*                    Choose render functions                         */
+/**********************************************************************/
+
+void radeonChooseRenderState( GLcontext *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint index = 0;
+   GLuint flags = ctx->_TriangleCaps;
+
+   if (!rmesa->TclFallback || rmesa->Fallback) 
+      return;
+
+   if (flags & DD_TRI_LIGHT_TWOSIDE) index |= RADEON_TWOSIDE_BIT;
+   if (flags & DD_TRI_UNFILLED)      index |= RADEON_UNFILLED_BIT;
+   if ((flags & DD_TRI_OFFSET) &&
+       rmesa->dri.drmMinor == 1)  index |= RADEON_OFFSET_BIT;
+
+   if (index != rmesa->swtcl.RenderIndex) {
+      tnl->Driver.Render.Points = rast_tab[index].points;
+      tnl->Driver.Render.Line = rast_tab[index].line;
+      tnl->Driver.Render.ClippedLine = rast_tab[index].line;
+      tnl->Driver.Render.Triangle = rast_tab[index].triangle;
+      tnl->Driver.Render.Quad = rast_tab[index].quad;
+
+      if (index == 0) {
+	 tnl->Driver.Render.PrimTabVerts = radeon_render_tab_verts;
+	 tnl->Driver.Render.PrimTabElts = radeon_render_tab_elts;
+	 tnl->Driver.Render.ClippedPolygon = radeon_fast_clipped_poly;
+      } else {
+	 tnl->Driver.Render.PrimTabVerts = _tnl_render_tab_verts;
+	 tnl->Driver.Render.PrimTabElts = _tnl_render_tab_elts;
+	 tnl->Driver.Render.ClippedPolygon = _tnl_RenderClippedPolygon;
+      }
+
+      rmesa->swtcl.RenderIndex = index;
+   }
+}
+
+
+/**********************************************************************/
+/*                 High level hooks for t_vb_render.c                 */
+/**********************************************************************/
+
+
+static void radeonRasterPrimitive( GLcontext *ctx, GLuint hwprim )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if (rmesa->swtcl.hw_primitive != hwprim) {
+      RADEON_NEWPRIM( rmesa );
+      rmesa->swtcl.hw_primitive = hwprim;
+   }
+}
+
+static void radeonRenderPrimitive( GLcontext *ctx, GLenum prim )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   rmesa->swtcl.render_primitive = prim;
+   if (prim < GL_TRIANGLES || !(ctx->_TriangleCaps & DD_TRI_UNFILLED)) 
+      radeonRasterPrimitive( ctx, reduced_hw_prim[prim] );
+}
+
+static void radeonRenderFinish( GLcontext *ctx )
+{
+}
+
+static void radeonResetLineStipple( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   RADEON_STATECHANGE( rmesa, lin );
+}
+
+
+/**********************************************************************/
+/*           Transition to/from hardware rasterization.               */
+/**********************************************************************/
+
+static const char * const fallbackStrings[] = {
+   "Texture mode",
+   "glDrawBuffer(GL_FRONT_AND_BACK)",
+   "glEnable(GL_STENCIL) without hw stencil buffer",
+   "glRenderMode(selection or feedback)",
+   "glBlendEquation",
+   "glBlendFunc",
+   "RADEON_NO_RAST",
+   "Mixing GL_CLAMP_TO_BORDER and GL_CLAMP (or GL_MIRROR_CLAMP_ATI)"
+};
+
+
+static const char *getFallbackString(GLuint bit)
+{
+   int i = 0;
+   while (bit > 1) {
+      i++;
+      bit >>= 1;
+   }
+   return fallbackStrings[i];
+}
+
+
+void radeonFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   GLuint oldfallback = rmesa->Fallback;
+
+   if (mode) {
+      rmesa->Fallback |= bit;
+      if (oldfallback == 0) {
+	 RADEON_FIREVERTICES( rmesa );
+	 TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_RASTER, GL_TRUE );
+	 _swsetup_Wakeup( ctx );
+	 _tnl_need_projected_coords( ctx, GL_TRUE );
+	 rmesa->swtcl.RenderIndex = ~0;
+         if (RADEON_DEBUG & DEBUG_FALLBACKS) {
+            fprintf(stderr, "Radeon begin rasterization fallback: 0x%x %s\n",
+                    bit, getFallbackString(bit));
+         }
+      }
+   }
+   else {
+      rmesa->Fallback &= ~bit;
+      if (oldfallback == bit) {
+	 _swrast_flush( ctx );
+	 tnl->Driver.Render.Start = radeonRenderStart;
+	 tnl->Driver.Render.PrimitiveNotify = radeonRenderPrimitive;
+	 tnl->Driver.Render.Finish = radeonRenderFinish;
+	 tnl->Driver.Render.BuildVertices = radeonBuildVertices;
+	 tnl->Driver.Render.ResetLineStipple = radeonResetLineStipple;
+	 TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_RASTER, GL_FALSE );
+	 if (rmesa->TclFallback) {
+	    /* These are already done if rmesa->TclFallback goes to
+	     * zero above. But not if it doesn't (RADEON_NO_TCL for
+	     * example?)
+	     */
+	    radeonChooseVertexState( ctx );
+	    radeonChooseRenderState( ctx );
+	 }
+         if (RADEON_DEBUG & DEBUG_FALLBACKS) {
+            fprintf(stderr, "Radeon end rasterization fallback: 0x%x %s\n",
+                    bit, getFallbackString(bit));
+         }
+      }
+   }
+}
+
+
+void radeonFlushVertices( GLcontext *ctx, GLuint flags )
+{
+   _tnl_flush_vertices( ctx, flags );
+
+   if (flags & FLUSH_STORED_VERTICES)
+      RADEON_NEWPRIM( RADEON_CONTEXT( ctx ) );
+}
+
+/**********************************************************************/
+/*                            Initialization.                         */
+/**********************************************************************/
+
+void radeonInitSwtcl( GLcontext *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint size = TNL_CONTEXT(ctx)->vb.Size;
+   static int firsttime = 1;
+
+   if (firsttime) {
+      init_rast_tab();
+      init_setup_tab();
+      firsttime = 0;
+   }
+
+   tnl->Driver.Render.Start = radeonRenderStart;
+   tnl->Driver.Render.Finish = radeonRenderFinish;
+   tnl->Driver.Render.PrimitiveNotify = radeonRenderPrimitive;
+   tnl->Driver.Render.ResetLineStipple = radeonResetLineStipple;
+   tnl->Driver.Render.BuildVertices = radeonBuildVertices;
+
+   rmesa->swtcl.verts = ALIGN_MALLOC( size * 16 * 4, 32 );
+   rmesa->swtcl.RenderIndex = ~0;
+   rmesa->swtcl.render_primitive = GL_TRIANGLES;
+   rmesa->swtcl.hw_primitive = 0;
+}
+
+
+void radeonDestroySwtcl( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if (rmesa->swtcl.indexed_verts.buf) 
+      radeonReleaseDmaRegion( rmesa, &rmesa->swtcl.indexed_verts, 
+			      __FUNCTION__ );
+
+   if (rmesa->swtcl.verts) {
+      ALIGN_FREE(rmesa->swtcl.verts);
+      rmesa->swtcl.verts = 0;
+   }
+
+   if (rmesa->UbyteSecondaryColor.Ptr) {
+      ALIGN_FREE(rmesa->UbyteSecondaryColor.Ptr);
+      rmesa->UbyteSecondaryColor.Ptr = 0;
+   }
+
+   if (rmesa->UbyteColor.Ptr) {
+      ALIGN_FREE(rmesa->UbyteColor.Ptr);
+      rmesa->UbyteColor.Ptr = 0;
+   }
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_swtcl.h b/src/mesa/drivers/dri/radeon/radeon_swtcl.h
new file mode 100644
index 00000000000..c45e711c0b1
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_swtcl.h
@@ -0,0 +1,77 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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
+ATI, VA LINUX SYSTEMS AND/OR THEIR SUPPLIERS BE LIABLE FOR ANY CLAIM,
+DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR
+OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE
+USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell <[email protected]>
+ *
+ */
+
+#ifndef __RADEON_TRIS_H__
+#define __RADEON_TRIS_H__
+
+#include "mtypes.h"
+#include "swrast/swrast.h"
+#include "radeon_context.h"
+
+extern void radeonInitSwtcl( GLcontext *ctx );
+extern void radeonDestroySwtcl( GLcontext *ctx );
+
+extern void radeonFlushVertices( GLcontext *ctx, GLuint flags );
+extern void radeonChooseRenderState( GLcontext *ctx );
+extern void radeonChooseVertexState( GLcontext *ctx );
+
+extern void radeonCheckTexSizes( GLcontext *ctx );
+
+extern void radeonBuildVertices( GLcontext *ctx, GLuint start, GLuint count,
+				 GLuint newinputs );
+
+extern void radeonPrintSetupFlags(char *msg, GLuint flags );
+
+
+extern void radeon_emit_contiguous_verts( GLcontext *ctx,
+					  GLuint start,
+					  GLuint count );
+
+extern void radeon_emit_indexed_verts( GLcontext *ctx,
+				       GLuint start,
+				       GLuint count );
+
+extern void radeon_translate_vertex( GLcontext *ctx, 
+				     const radeonVertex *src, 
+				     SWvertex *dst );
+
+extern void radeon_print_vertex( GLcontext *ctx, const radeonVertex *v );
+
+extern void radeon_import_float_colors( GLcontext *ctx );
+extern void radeon_import_float_spec_colors( GLcontext *ctx );
+
+
+
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_tcl.c b/src/mesa/drivers/dri/radeon/radeon_tcl.c
new file mode 100644
index 00000000000..651194a8047
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_tcl.c
@@ -0,0 +1,527 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     Tungsten Graphics Inc., Austin, 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, 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 <[email protected]>
+ */
+
+#include "glheader.h"
+#include "imports.h"
+#include "light.h"
+#include "mtypes.h"
+#include "enums.h"
+
+#include "array_cache/acache.h"
+#include "tnl/tnl.h"
+#include "tnl/t_pipeline.h"
+
+#include "radeon_context.h"
+#include "radeon_state.h"
+#include "radeon_ioctl.h"
+#include "radeon_tex.h"
+#include "radeon_tcl.h"
+#include "radeon_swtcl.h"
+#include "radeon_maos.h"
+
+
+
+/*
+ * Render unclipped vertex buffers by emitting vertices directly to
+ * dma buffers.  Use strip/fan hardware primitives where possible.
+ * Try to simulate missing primitives with indexed vertices.
+ */
+#define HAVE_POINTS      1
+#define HAVE_LINES       1
+#define HAVE_LINE_LOOP   0
+#define HAVE_LINE_STRIPS 1
+#define HAVE_TRIANGLES   1
+#define HAVE_TRI_STRIPS  1
+#define HAVE_TRI_STRIP_1 0
+#define HAVE_TRI_FANS    1
+#define HAVE_QUADS       0
+#define HAVE_QUAD_STRIPS 0
+#define HAVE_POLYGONS    1
+#define HAVE_ELTS        1
+
+
+#define HW_POINTS           RADEON_CP_VC_CNTL_PRIM_TYPE_POINT
+#define HW_LINES            RADEON_CP_VC_CNTL_PRIM_TYPE_LINE
+#define HW_LINE_LOOP        0
+#define HW_LINE_STRIP       RADEON_CP_VC_CNTL_PRIM_TYPE_LINE_STRIP
+#define HW_TRIANGLES        RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_LIST
+#define HW_TRIANGLE_STRIP_0 RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_STRIP
+#define HW_TRIANGLE_STRIP_1 0
+#define HW_TRIANGLE_FAN     RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_FAN
+#define HW_QUADS            0
+#define HW_QUAD_STRIP       0
+#define HW_POLYGON          RADEON_CP_VC_CNTL_PRIM_TYPE_TRI_FAN
+
+
+static GLboolean discrete_prim[0x10] = {
+   0,				/* 0 none */
+   1,				/* 1 points */
+   1,				/* 2 lines */
+   0,				/* 3 line_strip */
+   1,				/* 4 tri_list */
+   0,				/* 5 tri_fan */
+   0,				/* 6 tri_type2 */
+   1,				/* 7 rect list (unused) */
+   1,				/* 8 3vert point */
+   1,				/* 9 3vert line */
+   0,
+   0,
+   0,
+   0,
+   0,
+   0,
+};
+   
+
+#define LOCAL_VARS radeonContextPtr rmesa = RADEON_CONTEXT(ctx)
+#define ELT_TYPE  GLushort
+
+#define ELT_INIT(prim, hw_prim) \
+   radeonTclPrimitive( ctx, prim, hw_prim | RADEON_CP_VC_CNTL_PRIM_WALK_IND )
+
+#define GET_MESA_ELTS() rmesa->tcl.Elts
+
+
+/* Don't really know how many elts will fit in what's left of cmdbuf,
+ * as there is state to emit, etc:
+ */
+
+/* Testing on isosurf shows a maximum around here.  Don't know if it's
+ * the card or driver or kernel module that is causing the behaviour.
+ */
+#define GET_MAX_HW_ELTS() 300
+
+
+#define RESET_STIPPLE() do {			\
+   RADEON_STATECHANGE( rmesa, lin );		\
+   radeonEmitState( rmesa );			\
+} while (0)
+
+#define AUTO_STIPPLE( mode )  do {		\
+   RADEON_STATECHANGE( rmesa, lin );		\
+   if (mode)					\
+      rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] |=	\
+	 RADEON_LINE_PATTERN_AUTO_RESET;	\
+   else						\
+      rmesa->hw.lin.cmd[LIN_RE_LINE_PATTERN] &=	\
+	 ~RADEON_LINE_PATTERN_AUTO_RESET;	\
+   radeonEmitState( rmesa );			\
+} while (0)
+
+
+
+#define ALLOC_ELTS(nr)	radeonAllocElts( rmesa, nr )
+
+static GLushort *radeonAllocElts( radeonContextPtr rmesa, GLuint nr ) 
+{
+   if (rmesa->dma.flush)
+      rmesa->dma.flush( rmesa );
+
+   radeonEmitAOS( rmesa,
+		rmesa->tcl.aos_components,
+		rmesa->tcl.nr_aos_components, 0 );
+
+   return radeonAllocEltsOpenEnded( rmesa,
+				    rmesa->tcl.vertex_format, 
+				    rmesa->tcl.hw_primitive, nr );
+}
+
+#define CLOSE_ELTS()  RADEON_NEWPRIM( rmesa )
+
+
+
+/* TODO: Try to extend existing primitive if both are identical,
+ * discrete and there are no intervening state changes.  (Somewhat
+ * duplicates changes to DrawArrays code)
+ */
+static void EMIT_PRIM( GLcontext *ctx, 
+		       GLenum prim, 
+		       GLuint hwprim, 
+		       GLuint start, 
+		       GLuint count)	
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+   radeonTclPrimitive( ctx, prim, hwprim );
+   
+   radeonEmitAOS( rmesa,
+		  rmesa->tcl.aos_components,
+		  rmesa->tcl.nr_aos_components,
+		  start );
+   
+   /* Why couldn't this packet have taken an offset param?
+    */
+   radeonEmitVbufPrim( rmesa,
+		       rmesa->tcl.vertex_format,
+		       rmesa->tcl.hw_primitive,
+		       count - start );
+}
+
+
+
+/* Try & join small primitives
+ */
+#if 0
+#define PREFER_DISCRETE_ELT_PRIM( NR, PRIM ) 0
+#else
+#define PREFER_DISCRETE_ELT_PRIM( NR, PRIM )			\
+  ((NR) < 20 ||							\
+   ((NR) < 40 &&						\
+    rmesa->tcl.hw_primitive == (PRIM|				\
+			    RADEON_CP_VC_CNTL_PRIM_WALK_IND|	\
+			    RADEON_CP_VC_CNTL_TCL_ENABLE)))
+#endif
+
+#ifdef MESA_BIG_ENDIAN
+/* We could do without (most of) this ugliness if dest was always 32 bit word aligned... */
+#define EMIT_ELT(dest, offset, x) do {				\
+	int off = offset + ( ( (GLuint)dest & 0x2 ) >> 1 );	\
+	GLushort *des = (GLushort *)( (GLuint)dest & ~0x2 );	\
+	(des)[ off + 1 - 2 * ( off & 1 ) ] = (GLushort)(x); } while (0)
+#else
+#define EMIT_ELT(dest, offset, x) (dest)[offset] = (GLushort) (x)
+#endif
+
+#define EMIT_TWO_ELTS(dest, offset, x, y)  *(GLuint *)(dest+offset) = ((y)<<16)|(x);
+
+
+
+#define TAG(x) tcl_##x
+#include "tnl_dd/t_dd_dmatmp2.h"
+
+/**********************************************************************/
+/*                          External entrypoints                     */
+/**********************************************************************/
+
+void radeonEmitPrimitive( GLcontext *ctx, 
+			  GLuint first,
+			  GLuint last,
+			  GLuint flags )
+{
+   tcl_render_tab_verts[flags&PRIM_MODE_MASK]( ctx, first, last, flags );
+}
+
+void radeonEmitEltPrimitive( GLcontext *ctx, 
+			     GLuint first,
+			     GLuint last,
+			     GLuint flags )
+{
+   tcl_render_tab_elts[flags&PRIM_MODE_MASK]( ctx, first, last, flags );
+}
+
+void radeonTclPrimitive( GLcontext *ctx, 
+			 GLenum prim,
+			 int hw_prim )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint se_cntl;
+   GLuint newprim = hw_prim | RADEON_CP_VC_CNTL_TCL_ENABLE;
+
+   if (newprim != rmesa->tcl.hw_primitive ||
+       !discrete_prim[hw_prim&0xf]) {
+      RADEON_NEWPRIM( rmesa );
+      rmesa->tcl.hw_primitive = newprim;
+   }
+
+   se_cntl = rmesa->hw.set.cmd[SET_SE_CNTL];
+   se_cntl &= ~RADEON_FLAT_SHADE_VTX_LAST;
+
+   if (prim == GL_POLYGON && (ctx->_TriangleCaps & DD_FLATSHADE)) 
+      se_cntl |= RADEON_FLAT_SHADE_VTX_0;
+   else
+      se_cntl |= RADEON_FLAT_SHADE_VTX_LAST;
+
+   if (se_cntl != rmesa->hw.set.cmd[SET_SE_CNTL]) {
+      RADEON_STATECHANGE( rmesa, set );
+      rmesa->hw.set.cmd[SET_SE_CNTL] = se_cntl;
+   }
+}
+
+
+/**********************************************************************/
+/*                          Render pipeline stage                     */
+/**********************************************************************/
+
+
+/* TCL render.
+ */
+static GLboolean radeon_run_tcl_render( GLcontext *ctx,
+					struct gl_pipeline_stage *stage )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   struct vertex_buffer *VB = &tnl->vb;
+   GLuint i,flags = 0,length;
+
+   /* TODO: separate this from the swtnl pipeline 
+    */
+   if (rmesa->TclFallback)
+      return GL_TRUE;	/* fallback to software t&l */
+
+   if (VB->Count == 0)
+      return GL_FALSE;
+
+   radeonReleaseArrays( ctx, stage->changed_inputs );
+   radeonEmitArrays( ctx, stage->inputs );
+
+   rmesa->tcl.Elts = VB->Elts;
+
+   for (i = VB->FirstPrimitive ; !(flags & PRIM_LAST) ; i += length)
+   {
+      flags = VB->Primitive[i];
+      length = VB->PrimitiveLength[i];
+
+      if (RADEON_DEBUG & DEBUG_PRIMS)
+	 fprintf(stderr, "%s: prim %s %d..%d\n", 
+		 __FUNCTION__,
+		 _mesa_lookup_enum_by_nr(flags & PRIM_MODE_MASK), 
+		 i, i+length);
+
+      if (!length)
+	 continue;
+
+      if (rmesa->tcl.Elts)
+	 radeonEmitEltPrimitive( ctx, i, i+length, flags );
+      else
+	 radeonEmitPrimitive( ctx, i, i+length, flags );
+   }
+
+   return GL_FALSE;		/* finished the pipe */
+}
+
+
+
+static void radeon_check_tcl_render( GLcontext *ctx,
+				     struct gl_pipeline_stage *stage )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint inputs = VERT_BIT_POS;
+
+   if (ctx->RenderMode == GL_RENDER) {
+      /* Make all this event-driven:
+       */
+      if (ctx->Light.Enabled) {
+	 inputs |= VERT_BIT_NORMAL;
+
+	 if (1 || ctx->Light.ColorMaterialEnabled) {
+	    inputs |= VERT_BIT_COLOR0;
+	 }
+      }
+      else {
+	 inputs |= VERT_BIT_COLOR0;
+	 
+	 if (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR) {
+	    inputs |= VERT_BIT_COLOR1;
+	 }
+      }
+
+      if (ctx->Texture.Unit[0]._ReallyEnabled) {
+	 if (ctx->Texture.Unit[0].TexGenEnabled) {
+	    if (rmesa->TexGenNeedNormals[0]) {
+	       inputs |= VERT_BIT_NORMAL;
+	    }
+	 } else {
+	    inputs |= VERT_BIT_TEX0;
+	 }
+      }
+
+      if (ctx->Texture.Unit[1]._ReallyEnabled) {
+	 if (ctx->Texture.Unit[1].TexGenEnabled) {
+	    if (rmesa->TexGenNeedNormals[1]) {
+	       inputs |= VERT_BIT_NORMAL;
+	    }
+	 } else {
+	    inputs |= VERT_BIT_TEX1;
+	 }
+      }
+
+      stage->inputs = inputs;
+      stage->active = 1;
+   }
+   else
+      stage->active = 0;
+}
+
+static void radeon_init_tcl_render( GLcontext *ctx,
+				    struct gl_pipeline_stage *stage )
+{
+   stage->check = radeon_check_tcl_render;
+   stage->check( ctx, stage );
+}
+
+static void dtr( struct gl_pipeline_stage *stage )
+{
+   (void)stage;
+}
+
+
+/* Initial state for tcl stage.  
+ */
+const struct gl_pipeline_stage _radeon_tcl_stage =
+{
+   "radeon render",
+   (_DD_NEW_SEPARATE_SPECULAR |
+    _NEW_LIGHT|
+    _NEW_TEXTURE|
+    _NEW_FOG|
+    _NEW_RENDERMODE),		/* re-check (new inputs) */
+   0,				/* re-run (always runs) */
+   GL_TRUE,			/* active */
+   0, 0,			/* inputs (set in check_render), outputs */
+   0, 0,			/* changed_inputs, private */
+   dtr,				/* destructor */
+   radeon_init_tcl_render,	/* check - initially set to alloc data */
+   radeon_run_tcl_render	/* run */
+};
+
+
+
+/**********************************************************************/
+/*                 Validate state at pipeline start                   */
+/**********************************************************************/
+
+
+/*-----------------------------------------------------------------------
+ * Manage TCL fallbacks
+ */
+
+
+static void transition_to_swtnl( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   GLuint se_cntl;
+
+   RADEON_NEWPRIM( rmesa );
+   rmesa->swtcl.vertex_format = 0;
+
+   radeonChooseVertexState( ctx );
+   radeonChooseRenderState( ctx );
+
+   _mesa_validate_all_lighting_tables( ctx ); 
+
+   tnl->Driver.NotifyMaterialChange = 
+      _mesa_validate_all_lighting_tables;
+
+   radeonReleaseArrays( ctx, ~0 );
+
+   se_cntl = rmesa->hw.set.cmd[SET_SE_CNTL];
+   se_cntl |= RADEON_FLAT_SHADE_VTX_LAST;
+	 
+   if (se_cntl != rmesa->hw.set.cmd[SET_SE_CNTL]) {
+      RADEON_STATECHANGE( rmesa, set );
+      rmesa->hw.set.cmd[SET_SE_CNTL] = se_cntl;
+   }
+}
+
+
+static void transition_to_hwtnl( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   GLuint se_coord_fmt = (RADEON_VTX_W0_IS_NOT_1_OVER_W0 |
+			  RADEON_TEX1_W_ROUTING_USE_Q1);
+
+   if ( se_coord_fmt != rmesa->hw.set.cmd[SET_SE_COORDFMT] ) {
+      RADEON_STATECHANGE( rmesa, set );
+      rmesa->hw.set.cmd[SET_SE_COORDFMT] = se_coord_fmt;
+      _tnl_need_projected_coords( ctx, GL_FALSE );
+   }
+
+   radeonUpdateMaterial( ctx );
+
+   tnl->Driver.NotifyMaterialChange = radeonUpdateMaterial;
+
+   if ( rmesa->dma.flush )			
+      rmesa->dma.flush( rmesa );	
+
+   rmesa->dma.flush = 0;
+   rmesa->swtcl.vertex_format = 0;
+   
+   if (rmesa->swtcl.indexed_verts.buf) 
+      radeonReleaseDmaRegion( rmesa, &rmesa->swtcl.indexed_verts, 
+			      __FUNCTION__ );
+
+   if (RADEON_DEBUG & DEBUG_FALLBACKS) 
+      fprintf(stderr, "Radeon end tcl fallback\n");
+}
+
+static char *fallbackStrings[] = {
+   "Rasterization fallback",
+   "Unfilled triangles",
+   "Twosided lighting, differing materials",
+   "Materials in VB (maybe between begin/end)",
+   "Texgen unit 0",
+   "Texgen unit 1",
+   "Texgen unit 2",
+   "User disable"
+};
+
+
+static char *getFallbackString(GLuint bit)
+{
+   int i = 0;
+   while (bit > 1) {
+      i++;
+      bit >>= 1;
+   }
+   return fallbackStrings[i];
+}
+
+
+
+void radeonTclFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint oldfallback = rmesa->TclFallback;
+
+   if (mode) {
+      rmesa->TclFallback |= bit;
+      if (oldfallback == 0) {
+	 if (RADEON_DEBUG & DEBUG_FALLBACKS) 
+	    fprintf(stderr, "Radeon begin tcl fallback %s\n",
+		    getFallbackString( bit ));
+	 transition_to_swtnl( ctx );
+      }
+   }
+   else {
+      rmesa->TclFallback &= ~bit;
+      if (oldfallback == bit) {
+	 if (RADEON_DEBUG & DEBUG_FALLBACKS) 
+	    fprintf(stderr, "Radeon end tcl fallback %s\n",
+		    getFallbackString( bit ));
+	 transition_to_hwtnl( ctx );
+      }
+   }
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_tcl.h b/src/mesa/drivers/dri/radeon/radeon_tcl.h
new file mode 100644
index 00000000000..1e97d32148a
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_tcl.h
@@ -0,0 +1,70 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     Tungsten Grahpics Inc., Austin, 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, 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 <[email protected]>
+ *
+ */
+
+#ifndef __RADEON_TCL_H__
+#define __RADEON_TCL_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+#include "radeon_context.h"
+
+extern void radeonTclPrimitive( GLcontext *ctx, GLenum prim, int hw_prim );
+extern void radeonEmitEltPrimitive( GLcontext *ctx, GLuint first, GLuint last,
+				    GLuint flags );
+extern void radeonEmitPrimitive( GLcontext *ctx, GLuint first, GLuint last,
+				 GLuint flags );
+
+extern void radeonTclFallback( GLcontext *ctx, GLuint bit, GLboolean mode );
+					      
+#define RADEON_TCL_FALLBACK_RASTER            0x1 /* rasterization */
+#define RADEON_TCL_FALLBACK_UNFILLED          0x2 /* unfilled tris */
+#define RADEON_TCL_FALLBACK_LIGHT_TWOSIDE     0x4 /* twoside tris */
+#define RADEON_TCL_FALLBACK_MATERIAL          0x8 /* material in vb */
+#define RADEON_TCL_FALLBACK_TEXGEN_0          0x10 /* texgen, unit 0 */
+#define RADEON_TCL_FALLBACK_TEXGEN_1          0x20 /* texgen, unit 1 */
+#define RADEON_TCL_FALLBACK_TEXGEN_2          0x40 /* texgen, unit 2 */
+#define RADEON_TCL_FALLBACK_TCL_DISABLE       0x80 /* user disable */
+#define RADEON_TCL_FALLBACK_TEXRECT_0         0x100 /* texture rectangle */
+#define RADEON_TCL_FALLBACK_TEXRECT_1         0x200 /* texture rectangle */
+#define RADEON_TCL_FALLBACK_TEXRECT_2         0x400 /* texture rectangle */
+
+#define RADEON_MAX_TCL_VERTSIZE (15*4)
+
+#define TCL_FALLBACK( ctx, bit, mode )	radeonTclFallback( ctx, bit, mode )
+
+
+#endif
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_tex.c b/src/mesa/drivers/dri/radeon/radeon_tex.c
new file mode 100644
index 00000000000..e068202c5ee
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_tex.c
@@ -0,0 +1,733 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_tex.c,v 1.6 2002/09/16 18:05:20 eich Exp $ */
+/*
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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 (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:
+ *    Gareth Hughes <[email protected]>
+ *    Brian Paul <[email protected]>
+ */
+
+#include "glheader.h"
+#include "imports.h"
+#include "colormac.h"
+#include "context.h"
+#include "enums.h"
+#include "image.h"
+#include "simple_list.h"
+#include "texformat.h"
+#include "texstore.h"
+#include "teximage.h"
+
+
+#include "radeon_context.h"
+#include "radeon_state.h"
+#include "radeon_ioctl.h"
+#include "radeon_swtcl.h"
+#include "radeon_tex.h"
+
+
+
+/**
+ * Set the texture wrap modes.
+ * 
+ * \param t Texture object whose wrap modes are to be set
+ * \param swrap Wrap mode for the \a s texture coordinate
+ * \param twrap Wrap mode for the \a t texture coordinate
+ */
+
+static void radeonSetTexWrap( radeonTexObjPtr t, GLenum swrap, GLenum twrap )
+{
+   GLboolean  is_clamp = GL_FALSE;
+   GLboolean  is_clamp_to_border = GL_FALSE;
+
+   t->pp_txfilter &= ~(RADEON_CLAMP_S_MASK | RADEON_CLAMP_T_MASK | RADEON_BORDER_MODE_D3D);
+
+   switch ( swrap ) {
+   case GL_REPEAT:
+      t->pp_txfilter |= RADEON_CLAMP_S_WRAP;
+      break;
+   case GL_CLAMP:
+      t->pp_txfilter |= RADEON_CLAMP_S_CLAMP_GL;
+      is_clamp = GL_TRUE;
+      break;
+   case GL_CLAMP_TO_EDGE:
+      t->pp_txfilter |= RADEON_CLAMP_S_CLAMP_LAST;
+      break;
+   case GL_CLAMP_TO_BORDER:
+      t->pp_txfilter |= RADEON_CLAMP_S_CLAMP_GL;
+      is_clamp_to_border = GL_TRUE;
+      break;
+   case GL_MIRRORED_REPEAT:
+      t->pp_txfilter |= RADEON_CLAMP_S_MIRROR;
+      break;
+   case GL_MIRROR_CLAMP_ATI:
+      t->pp_txfilter |= RADEON_CLAMP_S_MIRROR_CLAMP_GL;
+      is_clamp = GL_TRUE;
+      break;
+   case GL_MIRROR_CLAMP_TO_EDGE_ATI:
+      t->pp_txfilter |= RADEON_CLAMP_S_MIRROR_CLAMP_LAST;
+      break;
+   default:
+      _mesa_problem(NULL, "bad S wrap mode in %s", __FUNCTION__);
+   }
+
+   switch ( twrap ) {
+   case GL_REPEAT:
+      t->pp_txfilter |= RADEON_CLAMP_T_WRAP;
+      break;
+   case GL_CLAMP:
+      t->pp_txfilter |= RADEON_CLAMP_T_CLAMP_GL;
+      is_clamp = GL_TRUE;
+      break;
+   case GL_CLAMP_TO_EDGE:
+      t->pp_txfilter |= RADEON_CLAMP_T_CLAMP_LAST;
+      break;
+   case GL_CLAMP_TO_BORDER:
+      t->pp_txfilter |= RADEON_CLAMP_T_CLAMP_GL;
+      is_clamp_to_border = GL_TRUE;
+      break;
+   case GL_MIRRORED_REPEAT:
+      t->pp_txfilter |= RADEON_CLAMP_T_MIRROR;
+      break;
+   case GL_MIRROR_CLAMP_ATI:
+      t->pp_txfilter |= RADEON_CLAMP_T_MIRROR_CLAMP_GL;
+      is_clamp = GL_TRUE;
+      break;
+   case GL_MIRROR_CLAMP_TO_EDGE_ATI:
+      t->pp_txfilter |= RADEON_CLAMP_T_MIRROR_CLAMP_LAST;
+      break;
+   default:
+      _mesa_problem(NULL, "bad T wrap mode in %s", __FUNCTION__);
+   }
+
+   if ( is_clamp_to_border ) {
+      t->pp_txfilter |= RADEON_BORDER_MODE_D3D;
+   }
+
+   t->border_fallback = (is_clamp && is_clamp_to_border);
+}
+
+static void radeonSetTexMaxAnisotropy( radeonTexObjPtr t, GLfloat max )
+{
+   t->pp_txfilter &= ~RADEON_MAX_ANISO_MASK;
+
+   if ( max == 1.0 ) {
+      t->pp_txfilter |= RADEON_MAX_ANISO_1_TO_1;
+   } else if ( max <= 2.0 ) {
+      t->pp_txfilter |= RADEON_MAX_ANISO_2_TO_1;
+   } else if ( max <= 4.0 ) {
+      t->pp_txfilter |= RADEON_MAX_ANISO_4_TO_1;
+   } else if ( max <= 8.0 ) {
+      t->pp_txfilter |= RADEON_MAX_ANISO_8_TO_1;
+   } else {
+      t->pp_txfilter |= RADEON_MAX_ANISO_16_TO_1;
+   }
+}
+
+/**
+ * Set the texture magnification and minification modes.
+ * 
+ * \param t Texture whose filter modes are to be set
+ * \param minf Texture minification mode
+ * \param magf Texture magnification mode
+ */
+
+static void radeonSetTexFilter( radeonTexObjPtr t, GLenum minf, GLenum magf )
+{
+   GLuint anisotropy = (t->pp_txfilter & RADEON_MAX_ANISO_MASK);
+
+   t->pp_txfilter &= ~(RADEON_MIN_FILTER_MASK | RADEON_MAG_FILTER_MASK);
+
+   if ( anisotropy == RADEON_MAX_ANISO_1_TO_1 ) {
+      switch ( minf ) {
+      case GL_NEAREST:
+	 t->pp_txfilter |= RADEON_MIN_FILTER_NEAREST;
+	 break;
+      case GL_LINEAR:
+	 t->pp_txfilter |= RADEON_MIN_FILTER_LINEAR;
+	 break;
+      case GL_NEAREST_MIPMAP_NEAREST:
+	 t->pp_txfilter |= RADEON_MIN_FILTER_NEAREST_MIP_NEAREST;
+	 break;
+      case GL_NEAREST_MIPMAP_LINEAR:
+	 t->pp_txfilter |= RADEON_MIN_FILTER_LINEAR_MIP_NEAREST;
+	 break;
+      case GL_LINEAR_MIPMAP_NEAREST:
+	 t->pp_txfilter |= RADEON_MIN_FILTER_NEAREST_MIP_LINEAR;
+	 break;
+      case GL_LINEAR_MIPMAP_LINEAR:
+	 t->pp_txfilter |= RADEON_MIN_FILTER_LINEAR_MIP_LINEAR;
+	 break;
+      }
+   } else {
+      switch ( minf ) {
+      case GL_NEAREST:
+	 t->pp_txfilter |= RADEON_MIN_FILTER_ANISO_NEAREST;
+	 break;
+      case GL_LINEAR:
+	 t->pp_txfilter |= RADEON_MIN_FILTER_ANISO_LINEAR;
+	 break;
+      case GL_NEAREST_MIPMAP_NEAREST:
+      case GL_LINEAR_MIPMAP_NEAREST:
+	 t->pp_txfilter |= RADEON_MIN_FILTER_ANISO_NEAREST_MIP_NEAREST;
+	 break;
+      case GL_NEAREST_MIPMAP_LINEAR:
+      case GL_LINEAR_MIPMAP_LINEAR:
+	 t->pp_txfilter |= RADEON_MIN_FILTER_ANISO_NEAREST_MIP_LINEAR;
+	 break;
+      }
+   }
+
+   switch ( magf ) {
+   case GL_NEAREST:
+      t->pp_txfilter |= RADEON_MAG_FILTER_NEAREST;
+      break;
+   case GL_LINEAR:
+      t->pp_txfilter |= RADEON_MAG_FILTER_LINEAR;
+      break;
+   }
+}
+
+static void radeonSetTexBorderColor( radeonTexObjPtr t, GLubyte c[4] )
+{
+   t->pp_border_color = radeonPackColor( 4, c[0], c[1], c[2], c[3] );
+}
+
+
+/**
+ * Allocate space for and load the mesa images into the texture memory block.
+ * This will happen before drawing with a new texture, or drawing with a
+ * texture after it was swapped out or teximaged again.
+ */
+
+static radeonTexObjPtr radeonAllocTexObj( struct gl_texture_object *texObj )
+{
+   radeonTexObjPtr t;
+
+   t = CALLOC_STRUCT( radeon_tex_obj );
+   texObj->DriverData = t;
+   if ( t != NULL ) {
+      if ( RADEON_DEBUG & DEBUG_TEXTURE ) {
+	 fprintf( stderr, "%s( %p, %p )\n", __FUNCTION__, texObj, t );
+      }
+
+      /* Initialize non-image-dependent parts of the state:
+       */
+      t->base.tObj = texObj;
+      t->border_fallback = GL_FALSE;
+
+      t->pp_txfilter = RADEON_BORDER_MODE_OGL;
+      t->pp_txformat = (RADEON_TXFORMAT_ENDIAN_NO_SWAP |
+			RADEON_TXFORMAT_PERSPECTIVE_ENABLE);
+
+      make_empty_list( & t->base );
+
+      radeonSetTexWrap( t, texObj->WrapS, texObj->WrapT );
+      radeonSetTexMaxAnisotropy( t, texObj->MaxAnisotropy );
+      radeonSetTexFilter( t, texObj->MinFilter, texObj->MagFilter );
+      radeonSetTexBorderColor( t, texObj->_BorderChan );
+   }
+
+   return t;
+}
+
+
+static const struct gl_texture_format *
+radeonChooseTextureFormat( GLcontext *ctx, GLint internalFormat,
+                           GLenum format, GLenum type )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   const GLboolean do32bpt = ( rmesa->radeonScreen->cpp == 4 );
+
+   switch ( internalFormat ) {
+   case 4:
+   case GL_RGBA:
+   case GL_COMPRESSED_RGBA:
+      if ( format == GL_BGRA ) {
+	 if ( type == GL_UNSIGNED_INT_8_8_8_8_REV ) {
+	    return &_mesa_texformat_argb8888;
+	 }
+         else if ( type == GL_UNSIGNED_SHORT_4_4_4_4_REV ) {
+            return &_mesa_texformat_argb4444;
+	 }
+         else if ( type == GL_UNSIGNED_SHORT_1_5_5_5_REV ) {
+	    return &_mesa_texformat_argb1555;
+	 }
+      }
+      return do32bpt ? &_mesa_texformat_rgba8888 : &_mesa_texformat_argb4444;
+
+   case 3:
+   case GL_RGB:
+   case GL_COMPRESSED_RGB:
+      if ( format == GL_RGB && type == GL_UNSIGNED_SHORT_5_6_5 ) {
+	 return &_mesa_texformat_rgb565;
+      }
+      return do32bpt ? &_mesa_texformat_rgba8888 : &_mesa_texformat_rgb565;
+
+   case GL_RGBA8:
+   case GL_RGB10_A2:
+   case GL_RGBA12:
+   case GL_RGBA16:
+      return do32bpt ? &_mesa_texformat_rgba8888 : &_mesa_texformat_argb4444;
+
+   case GL_RGBA4:
+   case GL_RGBA2:
+      return &_mesa_texformat_argb4444;
+
+   case GL_RGB5_A1:
+      return &_mesa_texformat_argb1555;
+
+   case GL_RGB8:
+   case GL_RGB10:
+   case GL_RGB12:
+   case GL_RGB16:
+      return do32bpt ? &_mesa_texformat_rgba8888 : &_mesa_texformat_rgb565;
+
+   case GL_RGB5:
+   case GL_RGB4:
+   case GL_R3_G3_B2:
+      return &_mesa_texformat_rgb565;
+
+   case GL_ALPHA:
+   case GL_ALPHA4:
+   case GL_ALPHA8:
+   case GL_ALPHA12:
+   case GL_ALPHA16:
+   case GL_COMPRESSED_ALPHA:
+      return &_mesa_texformat_al88;
+
+   case 1:
+   case GL_LUMINANCE:
+   case GL_LUMINANCE4:
+   case GL_LUMINANCE8:
+   case GL_LUMINANCE12:
+   case GL_LUMINANCE16:
+   case GL_COMPRESSED_LUMINANCE:
+      return &_mesa_texformat_al88;
+
+   case 2:
+   case GL_LUMINANCE_ALPHA:
+   case GL_LUMINANCE4_ALPHA4:
+   case GL_LUMINANCE6_ALPHA2:
+   case GL_LUMINANCE8_ALPHA8:
+   case GL_LUMINANCE12_ALPHA4:
+   case GL_LUMINANCE12_ALPHA12:
+   case GL_LUMINANCE16_ALPHA16:
+   case GL_COMPRESSED_LUMINANCE_ALPHA:
+      return &_mesa_texformat_al88;
+
+   case GL_INTENSITY:
+   case GL_INTENSITY4:
+   case GL_INTENSITY8:
+   case GL_INTENSITY12:
+   case GL_INTENSITY16:
+   case GL_COMPRESSED_INTENSITY:
+      return &_mesa_texformat_i8;
+
+   case GL_YCBCR_MESA:
+      if (type == GL_UNSIGNED_SHORT_8_8_APPLE ||
+          type == GL_UNSIGNED_BYTE)
+         return &_mesa_texformat_ycbcr;
+      else
+         return &_mesa_texformat_ycbcr_rev;
+
+   default:
+      _mesa_problem(ctx, "unexpected texture format in %s", __FUNCTION__);
+      return NULL;
+   }
+
+   return NULL; /* never get here */
+}
+
+
+static void radeonTexImage1D( GLcontext *ctx, GLenum target, GLint level,
+                              GLint internalFormat,
+                              GLint width, GLint border,
+                              GLenum format, GLenum type, const GLvoid *pixels,
+                              const struct gl_pixelstore_attrib *packing,
+                              struct gl_texture_object *texObj,
+                              struct gl_texture_image *texImage )
+{
+   driTextureObject * t = (driTextureObject *) texObj->DriverData;
+
+   if ( t ) {
+      driSwapOutTextureObject( t );
+   }
+   else {
+      t = (driTextureObject *) radeonAllocTexObj( texObj );
+      if (!t) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage1D");
+         return;
+      }
+   }
+
+   /* Note, this will call ChooseTextureFormat */
+   _mesa_store_teximage1d(ctx, target, level, internalFormat,
+                          width, border, format, type, pixels,
+                          &ctx->Unpack, texObj, texImage);
+
+   t->dirty_images[0] |= (1 << level);
+}
+
+
+static void radeonTexSubImage1D( GLcontext *ctx, GLenum target, GLint level,
+                                 GLint xoffset,
+                                 GLsizei width,
+                                 GLenum format, GLenum type,
+                                 const GLvoid *pixels,
+                                 const struct gl_pixelstore_attrib *packing,
+                                 struct gl_texture_object *texObj,
+                                 struct gl_texture_image *texImage )
+{
+   driTextureObject * t = (driTextureObject *) texObj->DriverData;
+
+   assert( t ); /* this _should_ be true */
+   if ( t ) {
+      driSwapOutTextureObject( t );
+   }
+   else {
+      t = (driTextureObject *) radeonAllocTexObj( texObj );
+      if (!t) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage1D");
+         return;
+      }
+   }
+
+   _mesa_store_texsubimage1d(ctx, target, level, xoffset, width,
+			     format, type, pixels, packing, texObj,
+			     texImage);
+
+   t->dirty_images[0] |= (1 << level);
+}
+
+
+static void radeonTexImage2D( GLcontext *ctx, GLenum target, GLint level,
+                              GLint internalFormat,
+                              GLint width, GLint height, GLint border,
+                              GLenum format, GLenum type, const GLvoid *pixels,
+                              const struct gl_pixelstore_attrib *packing,
+                              struct gl_texture_object *texObj,
+                              struct gl_texture_image *texImage )
+{
+   driTextureObject * t = (driTextureObject *) texObj->DriverData;
+   GLuint face;
+
+   /* which cube face or ordinary 2D image */
+   switch (target) {
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
+      face = (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X;
+      ASSERT(face < 6);
+      break;
+   default:
+      face = 0;
+   }
+
+   if ( t != NULL ) {
+      driSwapOutTextureObject( t );
+   }
+   else {
+      t = (driTextureObject *) radeonAllocTexObj( texObj );
+      if (!t) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexImage2D");
+         return;
+      }
+   }
+
+   /* Note, this will call ChooseTextureFormat */
+   _mesa_store_teximage2d(ctx, target, level, internalFormat,
+                          width, height, border, format, type, pixels,
+                          &ctx->Unpack, texObj, texImage);
+
+   t->dirty_images[face] |= (1 << level);
+}
+
+
+static void radeonTexSubImage2D( GLcontext *ctx, GLenum target, GLint level,
+                                 GLint xoffset, GLint yoffset,
+                                 GLsizei width, GLsizei height,
+                                 GLenum format, GLenum type,
+                                 const GLvoid *pixels,
+                                 const struct gl_pixelstore_attrib *packing,
+                                 struct gl_texture_object *texObj,
+                                 struct gl_texture_image *texImage )
+{
+   driTextureObject * t = (driTextureObject *) texObj->DriverData;
+   GLuint face;
+
+
+   /* which cube face or ordinary 2D image */
+   switch (target) {
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_X:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_X:
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_Y:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Y:
+   case GL_TEXTURE_CUBE_MAP_POSITIVE_Z:
+   case GL_TEXTURE_CUBE_MAP_NEGATIVE_Z:
+      face = (GLuint) target - (GLuint) GL_TEXTURE_CUBE_MAP_POSITIVE_X;
+      ASSERT(face < 6);
+      break;
+   default:
+      face = 0;
+   }
+
+   assert( t ); /* this _should_ be true */
+   if ( t ) {
+      driSwapOutTextureObject( t );
+   }
+   else {
+      t = (driTextureObject *) radeonAllocTexObj( texObj );
+      if (!t) {
+         _mesa_error(ctx, GL_OUT_OF_MEMORY, "glTexSubImage2D");
+         return;
+      }
+   }
+
+   _mesa_store_texsubimage2d(ctx, target, level, xoffset, yoffset, width,
+			     height, format, type, pixels, packing, texObj,
+			     texImage);
+
+   t->dirty_images[face] |= (1 << level);
+}
+
+
+
+#define SCALED_FLOAT_TO_BYTE( x, scale ) \
+		(((GLuint)((255.0F / scale) * (x))) / 2)
+
+static void radeonTexEnv( GLcontext *ctx, GLenum target,
+			  GLenum pname, const GLfloat *param )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint unit = ctx->Texture.CurrentUnit;
+   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
+
+   if ( RADEON_DEBUG & DEBUG_STATE ) {
+      fprintf( stderr, "%s( %s )\n",
+	       __FUNCTION__, _mesa_lookup_enum_by_nr( pname ) );
+   }
+
+   switch ( pname ) {
+   case GL_TEXTURE_ENV_COLOR: {
+      GLubyte c[4];
+      GLuint envColor;
+      UNCLAMPED_FLOAT_TO_RGBA_CHAN( c, texUnit->EnvColor );
+      envColor = radeonPackColor( 4, c[0], c[1], c[2], c[3] );
+      if ( rmesa->hw.tex[unit].cmd[TEX_PP_TFACTOR] != envColor ) {
+	 RADEON_STATECHANGE( rmesa, tex[unit] );
+	 rmesa->hw.tex[unit].cmd[TEX_PP_TFACTOR] = envColor;
+      }
+      break;
+   }
+
+   case GL_TEXTURE_LOD_BIAS_EXT: {
+      GLfloat bias;
+      GLuint b;
+
+      /* The Radeon's LOD bias is a signed 2's complement value with a
+       * range of -1.0 <= bias < 4.0.  We break this into two linear
+       * functions, one mapping [-1.0,0.0] to [-128,0] and one mapping
+       * [0.0,4.0] to [0,127].
+       */
+      bias = CLAMP( *param, -1.0, 4.0 );
+      if ( bias == 0 ) {
+	 b = 0;
+      } else if ( bias > 0 ) {
+	 b = ((GLuint)SCALED_FLOAT_TO_BYTE( bias, 4.0 )) << RADEON_LOD_BIAS_SHIFT;
+      } else {
+	 b = ((GLuint)SCALED_FLOAT_TO_BYTE( bias, 1.0 )) << RADEON_LOD_BIAS_SHIFT;
+      }
+      if ( (rmesa->hw.tex[unit].cmd[TEX_PP_TXFILTER] & RADEON_LOD_BIAS_MASK) != b ) {
+	 RADEON_STATECHANGE( rmesa, tex[unit] );
+	 rmesa->hw.tex[unit].cmd[TEX_PP_TXFILTER] &= ~RADEON_LOD_BIAS_MASK;
+	 rmesa->hw.tex[unit].cmd[TEX_PP_TXFILTER] |= (b & RADEON_LOD_BIAS_MASK);
+      }
+      break;
+   }
+
+   default:
+      return;
+   }
+}
+
+
+/**
+ * Changes variables and flags for a state update, which will happen at the
+ * next UpdateTextureState
+ */
+
+static void radeonTexParameter( GLcontext *ctx, GLenum target,
+				struct gl_texture_object *texObj,
+				GLenum pname, const GLfloat *params )
+{
+   radeonTexObjPtr t = (radeonTexObjPtr) texObj->DriverData;
+
+   if ( RADEON_DEBUG & (DEBUG_STATE|DEBUG_TEXTURE) ) {
+      fprintf( stderr, "%s( %s )\n", __FUNCTION__,
+	       _mesa_lookup_enum_by_nr( pname ) );
+   }
+
+   if ( ( target != GL_TEXTURE_2D ) &&
+	( target != GL_TEXTURE_1D ) )
+      return;
+
+   switch ( pname ) {
+   case GL_TEXTURE_MIN_FILTER:
+   case GL_TEXTURE_MAG_FILTER:
+   case GL_TEXTURE_MAX_ANISOTROPY_EXT:
+      radeonSetTexMaxAnisotropy( t, texObj->MaxAnisotropy );
+      radeonSetTexFilter( t, texObj->MinFilter, texObj->MagFilter );
+      break;
+
+   case GL_TEXTURE_WRAP_S:
+   case GL_TEXTURE_WRAP_T:
+      radeonSetTexWrap( t, texObj->WrapS, texObj->WrapT );
+      break;
+
+   case GL_TEXTURE_BORDER_COLOR:
+      radeonSetTexBorderColor( t, texObj->_BorderChan );
+      break;
+
+   case GL_TEXTURE_BASE_LEVEL:
+   case GL_TEXTURE_MAX_LEVEL:
+   case GL_TEXTURE_MIN_LOD:
+   case GL_TEXTURE_MAX_LOD:
+      /* This isn't the most efficient solution but there doesn't appear to
+       * be a nice alternative.  Since there's no LOD clamping,
+       * we just have to rely on loading the right subset of mipmap levels
+       * to simulate a clamped LOD.
+       */
+      driSwapOutTextureObject( (driTextureObject *) t );
+      break;
+
+   default:
+      return;
+   }
+
+   /* Mark this texobj as dirty (one bit per tex unit)
+    */
+   t->dirty_state = TEX_ALL;
+}
+
+
+
+static void radeonBindTexture( GLcontext *ctx, GLenum target,
+			       struct gl_texture_object *texObj )
+{
+   if ( RADEON_DEBUG & (DEBUG_STATE|DEBUG_TEXTURE) ) {
+      fprintf( stderr, "%s( %p ) unit=%d\n", __FUNCTION__, texObj,
+	       ctx->Texture.CurrentUnit );
+   }
+
+   if ( target == GL_TEXTURE_2D || target == GL_TEXTURE_1D ) {
+      if ( texObj->DriverData == NULL ) {
+	 radeonAllocTexObj( texObj );
+      }
+   }
+}
+
+static void radeonDeleteTexture( GLcontext *ctx,
+				 struct gl_texture_object *texObj )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   driTextureObject * t = (driTextureObject *) texObj->DriverData;
+
+   if ( RADEON_DEBUG & (DEBUG_STATE|DEBUG_TEXTURE) ) {
+      fprintf( stderr, "%s( %p (target = %s) )\n", __FUNCTION__, texObj,
+	       _mesa_lookup_enum_by_nr( texObj->Target ) );
+   }
+
+   if ( t != NULL ) {
+      if ( rmesa ) {
+         RADEON_FIREVERTICES( rmesa );
+      }
+
+      driDestroyTextureObject( t );
+   }
+}
+
+/* Need:  
+ *  - Same GEN_MODE for all active bits
+ *  - Same EyePlane/ObjPlane for all active bits when using Eye/Obj
+ *  - STRQ presumably all supported (matrix means incoming R values
+ *    can end up in STQ, this has implications for vertex support,
+ *    presumably ok if maos is used, though?)
+ *  
+ * Basically impossible to do this on the fly - just collect some
+ * basic info & do the checks from ValidateState().
+ */
+static void radeonTexGen( GLcontext *ctx,
+			  GLenum coord,
+			  GLenum pname,
+			  const GLfloat *params )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint unit = ctx->Texture.CurrentUnit;
+   rmesa->recheck_texgen[unit] = GL_TRUE;
+}
+
+
+void radeonInitTextureFuncs( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+
+   ctx->Driver.ChooseTextureFormat	= radeonChooseTextureFormat;
+   ctx->Driver.TexImage1D		= radeonTexImage1D;
+   ctx->Driver.TexImage2D		= radeonTexImage2D;
+   ctx->Driver.TexImage3D		= _mesa_store_teximage3d;
+   ctx->Driver.TexSubImage1D		= radeonTexSubImage1D;
+   ctx->Driver.TexSubImage2D		= radeonTexSubImage2D;
+   ctx->Driver.TexSubImage3D		= _mesa_store_texsubimage3d;
+   ctx->Driver.CopyTexImage1D		= _swrast_copy_teximage1d;
+   ctx->Driver.CopyTexImage2D		= _swrast_copy_teximage2d;
+   ctx->Driver.CopyTexSubImage1D	= _swrast_copy_texsubimage1d;
+   ctx->Driver.CopyTexSubImage2D	= _swrast_copy_texsubimage2d;
+   ctx->Driver.CopyTexSubImage3D 	= _swrast_copy_texsubimage3d;
+   ctx->Driver.TestProxyTexImage	= _mesa_test_proxy_teximage;
+
+   ctx->Driver.BindTexture		= radeonBindTexture;
+   ctx->Driver.CreateTexture		= NULL; /* FIXME: Is this used??? */
+   ctx->Driver.DeleteTexture		= radeonDeleteTexture;
+   ctx->Driver.IsTextureResident	= driIsTextureResident;
+   ctx->Driver.PrioritizeTexture	= NULL;
+   ctx->Driver.ActiveTexture		= NULL;
+   ctx->Driver.UpdateTexturePalette	= NULL;
+
+   ctx->Driver.TexEnv			= radeonTexEnv;
+   ctx->Driver.TexParameter		= radeonTexParameter;
+   ctx->Driver.TexGen                   = radeonTexGen;
+
+   driInitTextureObjects( ctx, & rmesa->swapped,
+			  DRI_TEXMGR_DO_TEXTURE_1D
+			  | DRI_TEXMGR_DO_TEXTURE_2D );
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_tex.h b/src/mesa/drivers/dri/radeon/radeon_tex.h
new file mode 100644
index 00000000000..ce079baec2e
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_tex.h
@@ -0,0 +1,53 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_tex.h,v 1.3 2002/02/22 21:45:01 dawes Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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 (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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ *
+ */
+
+#ifndef __RADEON_TEX_H__
+#define __RADEON_TEX_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+extern void radeonUpdateTextureState( GLcontext *ctx );
+
+extern int radeonUploadTexImages( radeonContextPtr rmesa, radeonTexObjPtr t,
+				  GLuint face );
+
+extern void radeonDestroyTexObj( radeonContextPtr rmesa, radeonTexObjPtr t );
+
+extern void radeonInitTextureFuncs( GLcontext *ctx );
+
+#endif
+#endif /* __RADEON_TEX_H__ */
diff --git a/src/mesa/drivers/dri/radeon/radeon_texmem.c b/src/mesa/drivers/dri/radeon/radeon_texmem.c
new file mode 100644
index 00000000000..3adc2a951c6
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_texmem.c
@@ -0,0 +1,378 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_texmem.c,v 1.7 2002/12/16 16:18:59 dawes Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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 ATI, VA LINUX SYSTEMS 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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ *
+ */
+#include <errno.h>
+
+#include "glheader.h"
+#include "imports.h"
+#include "context.h"
+#include "macros.h"
+#include "simple_list.h"
+
+#include "radeon_context.h"
+#include "radeon_ioctl.h"
+#include "radeon_tex.h"
+
+
+/**
+ * Destroy any device-dependent state associated with the texture.  This may
+ * include NULLing out hardware state that points to the texture.
+ */
+void
+radeonDestroyTexObj( radeonContextPtr rmesa, radeonTexObjPtr t )
+{
+   if ( RADEON_DEBUG & DEBUG_TEXTURE ) {
+      fprintf( stderr, "%s( %p, %p )\n", __FUNCTION__, t, t->base.tObj );
+   }
+
+   if ( rmesa != NULL ) {
+      unsigned   i;
+
+
+      for ( i = 0 ; i < rmesa->glCtx->Const.MaxTextureUnits ; i++ ) {
+	 if ( t == rmesa->state.texture.unit[i].texobj ) {
+	    rmesa->state.texture.unit[i].texobj = NULL;
+	    remove_from_list( &rmesa->hw.tex[i] );
+	    make_empty_list( &rmesa->hw.tex[i] );
+	 }
+      }
+   }
+}
+
+
+/* ------------------------------------------------------------
+ * Texture image conversions
+ */
+
+
+static void radeonUploadRectSubImage( radeonContextPtr rmesa,
+				      radeonTexObjPtr t, 
+				      struct gl_texture_image *texImage,
+				      GLint x, GLint y, 
+				      GLint width, GLint height )
+{
+   const struct gl_texture_format *texFormat = texImage->TexFormat;
+   int blit_format, dstPitch, done;
+
+   switch ( texFormat->TexelBytes ) {
+   case 1:
+      blit_format = RADEON_GMC_DST_8BPP_CI;
+      break;
+   case 2:
+      blit_format = RADEON_GMC_DST_16BPP;
+      break;
+   case 4:
+      blit_format = RADEON_GMC_DST_32BPP;
+      break;
+   default:
+      fprintf( stderr, "radeonUploadRectSubImage: unknown blit_format (texelbytes=%d)\n", 
+      	       texFormat->TexelBytes);
+      return;
+   }
+
+   t->image[0][0].data = texImage->Data;
+
+   /* Currently don't need to cope with small pitches.
+    */
+   width = texImage->Width;
+   height = texImage->Height;
+   dstPitch = t->pp_txpitch + 32;
+
+   {	/* FIXME: prefer AGP-texturing if possible */
+      /* Data not in agp memory, or bad pitch.
+       */
+      for (done = 0; done < height ; ) {
+	 struct radeon_dma_region region;
+	 int lines = MIN2( height - done, RADEON_BUFFER_SIZE / dstPitch );
+	 int src_pitch;
+	 char *tex;
+
+         src_pitch = texImage->RowStride * texFormat->TexelBytes;
+
+	 tex = (char *)texImage->Data + done * src_pitch;
+
+	 memset(&region, 0, sizeof(region));
+	 radeonAllocDmaRegion( rmesa, &region, lines * dstPitch, 1024 );
+
+	 /* Copy texdata to dma:
+	  */
+	 if (0)
+	    fprintf(stderr, "%s: src_pitch %d dst_pitch %d\n",
+		    __FUNCTION__, src_pitch, dstPitch);
+
+	 if (src_pitch == dstPitch) {
+	    memcpy( region.address, tex, lines * src_pitch );
+	 } 
+	 else {
+	    char *buf = region.address;
+	    int i;
+	    for (i = 0 ; i < lines ; i++) {
+	       memcpy( buf, tex, src_pitch );
+	       buf += dstPitch;
+	       tex += src_pitch;
+	    }
+	 }
+
+	 radeonEmitWait( rmesa, RADEON_WAIT_3D );
+
+	 
+
+	 /* Blit to framebuffer
+	  */
+	 radeonEmitBlit( rmesa, 
+		       blit_format, 
+ 		       dstPitch, GET_START( &region ),    
+ 		       dstPitch, t->bufAddr, 
+		       0, 0, 
+		       0, done, 
+		       width, lines );
+	 
+	 radeonEmitWait( rmesa, RADEON_WAIT_2D );
+
+	 radeonReleaseDmaRegion( rmesa, &region, __FUNCTION__ );
+	 done += lines;
+      }
+   }
+}
+
+
+/**
+ * Upload the texture image associated with texture \a t at the specified
+ * level at the address relative to \a start.
+ */
+static void uploadSubImage( radeonContextPtr rmesa, radeonTexObjPtr t, 
+			    GLint hwlevel,
+			    GLint x, GLint y, GLint width, GLint height,
+			    GLuint face )
+{
+   struct gl_texture_image *texImage = NULL;
+   GLuint offset;
+   GLint imageWidth, imageHeight;
+   GLint ret;
+   drmRadeonTexture tex;
+   drmRadeonTexImage tmp;
+   const int level = hwlevel + t->base.firstLevel;
+
+   if ( RADEON_DEBUG & DEBUG_TEXTURE ) {
+      fprintf( stderr, "%s( %p, %p ) level/width/height/face = %d/%d/%d/%u\n", 
+	       __FUNCTION__, t, t->base.tObj, level, width, height, face );
+   }
+
+   ASSERT(face < 6);
+
+   /* Ensure we have a valid texture to upload */
+   if ( ( hwlevel < 0 ) || ( hwlevel >= RADEON_MAX_TEXTURE_LEVELS ) ) {
+      _mesa_problem(NULL, "bad texture level in %s", __FUNCTION__);
+      return;
+   }
+
+   switch (face) {
+   case 0:
+      texImage = t->base.tObj->Image[level];
+      break;
+   case 1:
+      texImage = t->base.tObj->NegX[level];
+      break;
+   case 2:
+      texImage = t->base.tObj->PosY[level];
+      break;
+   case 3:
+      texImage = t->base.tObj->NegY[level];
+      break;
+   case 4:
+      texImage = t->base.tObj->PosZ[level];
+      break;
+   case 5:
+      texImage = t->base.tObj->NegZ[level];
+      break;
+   }
+
+   if ( !texImage ) {
+      if ( RADEON_DEBUG & DEBUG_TEXTURE )
+	 fprintf( stderr, "%s: texImage %d is NULL!\n", __FUNCTION__, level );
+      return;
+   }
+   if ( !texImage->Data ) {
+      if ( RADEON_DEBUG & DEBUG_TEXTURE )
+	 fprintf( stderr, "%s: image data is NULL!\n", __FUNCTION__ );
+      return;
+   }
+
+
+   if (t->base.tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
+      assert(level == 0);
+      assert(hwlevel == 0);
+      if ( RADEON_DEBUG & DEBUG_TEXTURE )
+	 fprintf( stderr, "%s: image data is rectangular\n", __FUNCTION__);
+      radeonUploadRectSubImage( rmesa, t, texImage, x, y, width, height );
+      return;
+   }
+
+   imageWidth = texImage->Width;
+   imageHeight = texImage->Height;
+
+   offset = t->bufAddr;
+
+   if ( RADEON_DEBUG & (DEBUG_TEXTURE|DEBUG_IOCTL) ) {
+      GLint imageX = 0;
+      GLint imageY = 0;
+      GLint blitX = t->image[face][hwlevel].x;
+      GLint blitY = t->image[face][hwlevel].y;
+      GLint blitWidth = t->image[face][hwlevel].width;
+      GLint blitHeight = t->image[face][hwlevel].height;
+      fprintf( stderr, "   upload image: %d,%d at %d,%d\n",
+	       imageWidth, imageHeight, imageX, imageY );
+      fprintf( stderr, "   upload  blit: %d,%d at %d,%d\n",
+	       blitWidth, blitHeight, blitX, blitY );
+      fprintf( stderr, "       blit ofs: 0x%07x level: %d/%d\n",
+	       (GLuint)offset, hwlevel, level );
+   }
+
+   t->image[face][hwlevel].data = texImage->Data;
+
+   /* Init the DRM_RADEON_TEXTURE command / drmRadeonTexture struct.
+    * NOTE: we're always use a 1KB-wide blit and I8 texture format.
+    * We used to use 1, 2 and 4-byte texels and used to use the texture
+    * width to dictate the blit width - but that won't work for compressed
+    * textures. (Brian)
+    */
+   tex.offset = offset;
+   tex.pitch = BLIT_WIDTH_BYTES / 64;
+   tex.format = RADEON_TXFORMAT_I8; /* any 1-byte texel format */
+   if (texImage->TexFormat->TexelBytes) {
+      tex.width = imageWidth * texImage->TexFormat->TexelBytes; /* in bytes */
+      tex.height = imageHeight;
+   }
+   else {
+      tex.width = imageWidth; /* compressed */
+      tex.height = imageHeight;
+      if (tex.height < 4)
+         tex.height = 4;
+   }
+   tex.image = &tmp;
+
+   /* copy (x,y,width,height,data) */
+   memcpy( &tmp, &t->image[face][hwlevel], sizeof(drmRadeonTexImage) );
+
+   LOCK_HARDWARE( rmesa );
+   do {
+      ret = drmCommandWriteRead( rmesa->dri.fd, DRM_RADEON_TEXTURE,
+                                 &tex, sizeof(drmRadeonTexture) );
+   } while ( ret && errno == EAGAIN );
+
+   UNLOCK_HARDWARE( rmesa );
+
+   if ( ret ) {
+      fprintf( stderr, "DRM_RADEON_TEXTURE: return = %d\n", ret );
+      fprintf( stderr, "   offset=0x%08x\n",
+	       offset );
+      fprintf( stderr, "   image width=%d height=%d\n",
+	       imageWidth, imageHeight );
+      fprintf( stderr, "    blit width=%d height=%d data=%p\n",
+	       t->image[face][hwlevel].width, t->image[face][hwlevel].height,
+	       t->image[face][hwlevel].data );
+      exit( 1 );
+   }
+}
+
+
+/**
+ * Upload the texture images associated with texture \a t.  This might
+ * require the allocation of texture memory.
+ * 
+ * \param rmesa Context pointer
+ * \param t Texture to be uploaded
+ * \param face Cube map face to be uploaded.  Zero for non-cube maps.
+ */
+
+int radeonUploadTexImages( radeonContextPtr rmesa, radeonTexObjPtr t, GLuint face )
+{
+   const int numLevels = t->base.lastLevel - t->base.firstLevel + 1;
+
+   if ( RADEON_DEBUG & (DEBUG_TEXTURE|DEBUG_IOCTL) ) {
+      fprintf( stderr, "%s( %p, %p ) sz=%d lvls=%d-%d\n", __FUNCTION__,
+	       rmesa->glCtx, t->base.tObj, t->base.totalSize,
+	       t->base.firstLevel, t->base.lastLevel );
+   }
+
+   if ( !t || t->base.totalSize == 0 )
+      return 0;
+
+   LOCK_HARDWARE( rmesa );
+
+   if ( t->base.memBlock == NULL ) {
+      int heap;
+
+      heap = driAllocateTexture( rmesa->texture_heaps, rmesa->nr_heaps,
+				 (driTextureObject *) t );
+      if ( heap == -1 ) {
+	 UNLOCK_HARDWARE( rmesa );
+	 return -1;
+      }
+
+      /* Set the base offset of the texture image */
+      t->bufAddr = rmesa->radeonScreen->texOffset[heap] 
+	   + t->base.memBlock->ofs;
+      t->pp_txoffset = t->bufAddr;
+
+
+      /* Mark this texobj as dirty on all units:
+       */
+      t->dirty_state = TEX_ALL;
+   }
+
+
+   /* Let the world know we've used this memory recently.
+    */
+   driUpdateTextureLRU( (driTextureObject *) t );
+   UNLOCK_HARDWARE( rmesa );
+
+
+   /* Upload any images that are new */
+   if (t->base.dirty_images[face]) {
+      int i;
+      for ( i = 0 ; i < numLevels ; i++ ) {
+         if ( (t->base.dirty_images[face] & (1 << (i+t->base.firstLevel))) != 0 ) {
+            uploadSubImage( rmesa, t, i, 0, 0, t->image[face][i].width,
+			    t->image[face][i].height, face );
+         }
+      }
+      t->base.dirty_images[face] = 0;
+   }
+
+   return 0;
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_texstate.c b/src/mesa/drivers/dri/radeon/radeon_texstate.c
new file mode 100644
index 00000000000..6dccd311800
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_texstate.c
@@ -0,0 +1,1628 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_texstate.c,v 1.6 2002/12/16 16:18:59 dawes Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     VA Linux Systems Inc., Fremont, California.
+
+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 (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:
+ *   Kevin E. Martin <[email protected]>
+ *   Gareth Hughes <[email protected]>
+ */
+
+#include "glheader.h"
+#include "imports.h"
+#include "colormac.h"
+#include "context.h"
+#include "macros.h"
+#include "texformat.h"
+#include "enums.h"
+
+#include "radeon_context.h"
+#include "radeon_state.h"
+#include "radeon_ioctl.h"
+#include "radeon_swtcl.h"
+#include "radeon_tex.h"
+#include "radeon_tcl.h"
+
+
+#define RADEON_TXFORMAT_AL88      RADEON_TXFORMAT_AI88
+#define RADEON_TXFORMAT_YCBCR     RADEON_TXFORMAT_YVYU422
+#define RADEON_TXFORMAT_YCBCR_REV RADEON_TXFORMAT_VYUY422
+
+#define _COLOR(f) \
+    [ MESA_FORMAT_ ## f ] = { RADEON_TXFORMAT_ ## f, 0 }
+#define _ALPHA(f) \
+    [ MESA_FORMAT_ ## f ] = { RADEON_TXFORMAT_ ## f | RADEON_TXFORMAT_ALPHA_IN_MAP, 0 }
+#define _YUV(f) \
+   [ MESA_FORMAT_ ## f ] = { RADEON_TXFORMAT_ ## f, RADEON_YUV_TO_RGB }
+#define _INVALID(f) \
+    [ MESA_FORMAT_ ## f ] = { 0xffffffff, 0 }
+#define VALID_FORMAT(f) ( ((f) <= MESA_FORMAT_YCBCR_REV) \
+			     && (tx_table[f].format != 0xffffffff) )
+
+static const struct {
+   GLuint format, filter;
+}
+tx_table[] =
+{
+   _ALPHA(RGBA8888),
+   _ALPHA(ARGB8888),
+   _INVALID(RGB888),
+   _COLOR(RGB565),
+   _ALPHA(ARGB4444),
+   _ALPHA(ARGB1555),
+   _ALPHA(AL88),
+   _INVALID(A8),
+   _INVALID(L8),
+   _COLOR(I8),
+   _INVALID(CI8),
+   _YUV(YCBCR),
+   _YUV(YCBCR_REV),
+};
+
+#undef _COLOR
+#undef _ALPHA
+#undef _INVALID
+
+/**
+ * This function computes the number of bytes of storage needed for
+ * the given texture object (all mipmap levels, all cube faces).
+ * The \c image[face][level].x/y/width/height parameters for upload/blitting
+ * are computed here.  \c pp_txfilter, \c pp_txformat, etc. will be set here
+ * too.
+ * 
+ * \param rmesa Context pointer
+ * \param tObj GL texture object whose images are to be posted to
+ *                 hardware state.
+ */
+static void radeonSetTexImages( radeonContextPtr rmesa,
+				struct gl_texture_object *tObj )
+{
+   radeonTexObjPtr t = (radeonTexObjPtr)tObj->DriverData;
+   const struct gl_texture_image *baseImage = tObj->Image[tObj->BaseLevel];
+   GLint curOffset;
+   GLint i;
+   GLint firstLevel=0, lastLevel=0, numLevels;
+   GLint log2Width, log2Height, log2Depth;
+
+   /* Set the hardware texture format
+    */
+
+   t->pp_txformat &= ~(RADEON_TXFORMAT_FORMAT_MASK |
+		       RADEON_TXFORMAT_ALPHA_IN_MAP);
+   t->pp_txfilter &= ~RADEON_YUV_TO_RGB;
+
+   if ( VALID_FORMAT( baseImage->TexFormat->MesaFormat ) ) {
+      t->pp_txformat |= tx_table[ baseImage->TexFormat->MesaFormat ].format;
+      t->pp_txfilter |= tx_table[ baseImage->TexFormat->MesaFormat ].filter;
+   }
+   else {
+      _mesa_problem(NULL, "unexpected texture format in %s", __FUNCTION__);
+      return;
+   }
+
+
+
+   /* Compute which mipmap levels we really want to send to the hardware.
+    * This depends on the base image size, GL_TEXTURE_MIN_LOD,
+    * GL_TEXTURE_MAX_LOD, GL_TEXTURE_BASE_LEVEL, and GL_TEXTURE_MAX_LEVEL.
+    * Yes, this looks overly complicated, but it's all needed.
+    */
+   switch (tObj->Target) {
+   case GL_TEXTURE_1D:
+   case GL_TEXTURE_2D:
+      firstLevel = tObj->BaseLevel + (GLint)(tObj->MinLod + 0.5);
+      firstLevel = MAX2(firstLevel, tObj->BaseLevel);
+      lastLevel = tObj->BaseLevel + (GLint)(tObj->MaxLod + 0.5);
+      lastLevel = MAX2(lastLevel, tObj->BaseLevel);
+      lastLevel = MIN2(lastLevel, tObj->BaseLevel + baseImage->MaxLog2);
+      lastLevel = MIN2(lastLevel, tObj->MaxLevel);
+      lastLevel = MAX2(firstLevel, lastLevel); /* need at least one level */
+      log2Width = tObj->Image[firstLevel]->WidthLog2;
+      log2Height = tObj->Image[firstLevel]->HeightLog2;
+      log2Depth = 0;
+      break;
+   case GL_TEXTURE_RECTANGLE_NV:
+      firstLevel = lastLevel = 0;
+      log2Width = log2Height = 1; /* ? */
+      log2Depth = 0;
+      break;
+   default:
+      return;
+   }
+
+   /* save these values */
+   t->base.firstLevel = firstLevel;
+   t->base.lastLevel = lastLevel;
+
+   numLevels = lastLevel - firstLevel + 1;
+
+   assert(numLevels <= RADEON_MAX_TEXTURE_LEVELS);
+
+   /* Calculate mipmap offsets and dimensions for blitting (uploading)
+    * The idea is that we lay out the mipmap levels within a block of
+    * memory organized as a rectangle of width BLIT_WIDTH_BYTES.
+    */
+   curOffset = 0;
+
+   for (i = 0; i < numLevels; i++) {
+      const struct gl_texture_image *texImage;
+      GLuint size;
+
+      texImage = tObj->Image[i + firstLevel];
+      if ( !texImage )
+	 break;
+
+      /* find image size in bytes */
+      if (texImage->IsCompressed) {
+         size = texImage->CompressedSize;
+      }
+      else if (tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
+      	 size = ((texImage->Width * texImage->TexFormat->TexelBytes + 63)
+      	         & ~63) * texImage->Height;
+      }
+      else {
+         int w = texImage->Width * texImage->TexFormat->TexelBytes;
+         if (w < 32)
+            w = 32;
+         size = w * texImage->Height * texImage->Depth;
+      }
+      assert(size > 0);
+
+      if (curOffset & 0x1f) {
+         /* align to 32-byte offset */
+         curOffset = (curOffset + 0x1f) & ~0x1f;
+      }
+
+      t->image[0][i].x = curOffset % BLIT_WIDTH_BYTES;
+      t->image[0][i].y = curOffset / BLIT_WIDTH_BYTES;
+      t->image[0][i].width  = MIN2(size, BLIT_WIDTH_BYTES);
+      t->image[0][i].height = size / t->image[0][i].width;
+
+#if 0
+      /* for debugging only and only  applicable to non-rectangle targets */
+      assert(size % t->image[0][i].width == 0);
+      assert(t->image[0][i].x == 0
+             || (size < BLIT_WIDTH_BYTES && t->image[0][i].height == 1));
+#endif
+
+      if (0)
+         fprintf(stderr,
+                 "level %d: %dx%d x=%d y=%d w=%d h=%d size=%d at %d\n",
+                 i, texImage->Width, texImage->Height,
+                 t->image[0][i].x, t->image[0][i].y,
+                 t->image[0][i].width, t->image[0][i].height, size, curOffset);
+
+      curOffset += size;
+
+   }
+
+   /* Align the total size of texture memory block.
+    */
+   t->base.totalSize = (curOffset + RADEON_OFFSET_MASK) & ~RADEON_OFFSET_MASK;
+
+   /* Hardware state:
+    */
+   t->pp_txfilter &= ~RADEON_MAX_MIP_LEVEL_MASK;
+   t->pp_txfilter |= (numLevels - 1) << RADEON_MAX_MIP_LEVEL_SHIFT;
+
+   t->pp_txformat &= ~(RADEON_TXFORMAT_WIDTH_MASK |
+		       RADEON_TXFORMAT_HEIGHT_MASK |
+                       RADEON_TXFORMAT_CUBIC_MAP_ENABLE);
+   t->pp_txformat |= ((log2Width << RADEON_TXFORMAT_WIDTH_SHIFT) |
+		      (log2Height << RADEON_TXFORMAT_HEIGHT_SHIFT));
+
+   t->pp_txsize = (((tObj->Image[firstLevel]->Width - 1) << 0) |
+                   ((tObj->Image[firstLevel]->Height - 1) << 16));
+
+   /* Only need to round to nearest 32 for textures, but the blitter
+    * requires 64-byte aligned pitches, and we may/may not need the
+    * blitter.   NPOT only!
+    */
+   if (baseImage->IsCompressed)
+      t->pp_txpitch = (tObj->Image[firstLevel]->Width + 63) & ~(63);
+   else
+      t->pp_txpitch = ((tObj->Image[firstLevel]->Width * baseImage->TexFormat->TexelBytes) + 63) & ~(63);
+   t->pp_txpitch -= 32;
+
+   t->dirty_state = TEX_ALL;
+
+   /* FYI: radeonUploadTexImages( rmesa, t ); used to be called here */
+}
+
+
+
+/* ================================================================
+ * Texture combine functions
+ */
+
+#define RADEON_DISABLE		0
+#define RADEON_REPLACE		1
+#define RADEON_MODULATE		2
+#define RADEON_DECAL		3
+#define RADEON_BLEND		4
+#define RADEON_ADD		5
+#define RADEON_MAX_COMBFUNC	6
+
+static GLuint radeon_color_combine[][RADEON_MAX_COMBFUNC] =
+{
+   /* Unit 0:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_CURRENT_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00802800
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T0_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x00800142
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T0_COLOR |
+       RADEON_COLOR_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x008c2d42
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T0_COLOR |
+       RADEON_COLOR_ARG_C_T0_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x008c2902
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_TFACTOR_COLOR |
+       RADEON_COLOR_ARG_C_T0_COLOR |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00812802
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T0_COLOR |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   },
+
+   /* Unit 1:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_CURRENT_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00803000
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T1_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x00800182
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T1_COLOR |
+       RADEON_COLOR_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x008c3582
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T1_COLOR |
+       RADEON_COLOR_ARG_C_T1_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x008c3102
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_TFACTOR_COLOR |
+       RADEON_COLOR_ARG_C_T1_COLOR |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00813002
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T1_COLOR |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   },
+
+   /* Unit 2:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_CURRENT_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00803800
+       */
+      (RADEON_COLOR_ARG_A_ZERO |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T2_COLOR |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x008001c2
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T2_COLOR |
+       RADEON_COLOR_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x008c3dc2
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_T2_COLOR |
+       RADEON_COLOR_ARG_C_T2_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x008c3902
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_TFACTOR_COLOR |
+       RADEON_COLOR_ARG_C_T2_COLOR |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00813802
+       */
+      (RADEON_COLOR_ARG_A_CURRENT_COLOR |
+       RADEON_COLOR_ARG_B_ZERO |
+       RADEON_COLOR_ARG_C_T2_COLOR |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   }
+};
+
+static GLuint radeon_alpha_combine[][RADEON_MAX_COMBFUNC] =
+{
+   /* Unit 0:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00800500
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T0_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x00800051
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_T0_ALPHA |
+       RADEON_ALPHA_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x00800100
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x00800051
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_TFACTOR_ALPHA |
+       RADEON_ALPHA_ARG_C_T0_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00800051
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T0_ALPHA |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   },
+
+   /* Unit 1:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00800600
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T1_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x00800061
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_T1_ALPHA |
+       RADEON_ALPHA_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x00800100
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x00800061
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_TFACTOR_ALPHA |
+       RADEON_ALPHA_ARG_C_T1_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00800061
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T1_ALPHA |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   },
+
+   /* Unit 2:
+    */
+   {
+      /* Disable combiner stage
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_REPLACE = 0x00800700
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T2_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_MODULATE = 0x00800071
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_T2_ALPHA |
+       RADEON_ALPHA_ARG_C_ZERO |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_DECAL = 0x00800100
+       */
+      (RADEON_ALPHA_ARG_A_ZERO |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_CURRENT_ALPHA |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_BLEND = 0x00800071
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_TFACTOR_ALPHA |
+       RADEON_ALPHA_ARG_C_T2_ALPHA |
+       RADEON_BLEND_CTL_BLEND |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+
+      /* GL_ADD = 0x00800021
+       */
+      (RADEON_ALPHA_ARG_A_CURRENT_ALPHA |
+       RADEON_ALPHA_ARG_B_ZERO |
+       RADEON_ALPHA_ARG_C_T2_ALPHA |
+       RADEON_COMP_ARG_B |
+       RADEON_BLEND_CTL_ADD |
+       RADEON_SCALE_1X |
+       RADEON_CLAMP_TX),
+   }
+};
+
+
+/* GL_ARB_texture_env_combine support
+ */
+
+/* The color tables have combine functions for GL_SRC_COLOR,
+ * GL_ONE_MINUS_SRC_COLOR, GL_SRC_ALPHA and GL_ONE_MINUS_SRC_ALPHA.
+ */
+static GLuint radeon_texture_color[][RADEON_MAX_TEXTURE_UNITS] =
+{
+   {
+      RADEON_COLOR_ARG_A_T0_COLOR,
+      RADEON_COLOR_ARG_A_T1_COLOR,
+      RADEON_COLOR_ARG_A_T2_COLOR
+   },
+   {
+      RADEON_COLOR_ARG_A_T0_COLOR | RADEON_COMP_ARG_A,
+      RADEON_COLOR_ARG_A_T1_COLOR | RADEON_COMP_ARG_A,
+      RADEON_COLOR_ARG_A_T2_COLOR | RADEON_COMP_ARG_A
+   },
+   {
+      RADEON_COLOR_ARG_A_T0_ALPHA,
+      RADEON_COLOR_ARG_A_T1_ALPHA,
+      RADEON_COLOR_ARG_A_T2_ALPHA
+   },
+   {
+      RADEON_COLOR_ARG_A_T0_ALPHA | RADEON_COMP_ARG_A,
+      RADEON_COLOR_ARG_A_T1_ALPHA | RADEON_COMP_ARG_A,
+      RADEON_COLOR_ARG_A_T2_ALPHA | RADEON_COMP_ARG_A
+   },
+};
+
+static GLuint radeon_tfactor_color[] =
+{
+   RADEON_COLOR_ARG_A_TFACTOR_COLOR,
+   RADEON_COLOR_ARG_A_TFACTOR_COLOR | RADEON_COMP_ARG_A,
+   RADEON_COLOR_ARG_A_TFACTOR_ALPHA,
+   RADEON_COLOR_ARG_A_TFACTOR_ALPHA | RADEON_COMP_ARG_A
+};
+
+static GLuint radeon_primary_color[] =
+{
+   RADEON_COLOR_ARG_A_DIFFUSE_COLOR,
+   RADEON_COLOR_ARG_A_DIFFUSE_COLOR | RADEON_COMP_ARG_A,
+   RADEON_COLOR_ARG_A_DIFFUSE_ALPHA,
+   RADEON_COLOR_ARG_A_DIFFUSE_ALPHA | RADEON_COMP_ARG_A
+};
+
+static GLuint radeon_previous_color[] =
+{
+   RADEON_COLOR_ARG_A_CURRENT_COLOR,
+   RADEON_COLOR_ARG_A_CURRENT_COLOR | RADEON_COMP_ARG_A,
+   RADEON_COLOR_ARG_A_CURRENT_ALPHA,
+   RADEON_COLOR_ARG_A_CURRENT_ALPHA | RADEON_COMP_ARG_A
+};
+
+/* GL_ZERO table - indices 0-3
+ * GL_ONE  table - indices 1-4
+ */
+static GLuint radeon_zero_color[] =
+{
+   RADEON_COLOR_ARG_A_ZERO,
+   RADEON_COLOR_ARG_A_ZERO | RADEON_COMP_ARG_A,
+   RADEON_COLOR_ARG_A_ZERO,
+   RADEON_COLOR_ARG_A_ZERO | RADEON_COMP_ARG_A,
+   RADEON_COLOR_ARG_A_ZERO
+};
+
+
+/* The alpha tables only have GL_SRC_ALPHA and GL_ONE_MINUS_SRC_ALPHA.
+ */
+static GLuint radeon_texture_alpha[][RADEON_MAX_TEXTURE_UNITS] =
+{
+   {
+      RADEON_ALPHA_ARG_A_T0_ALPHA,
+      RADEON_ALPHA_ARG_A_T1_ALPHA,
+      RADEON_ALPHA_ARG_A_T2_ALPHA
+   },
+   {
+      RADEON_ALPHA_ARG_A_T0_ALPHA | RADEON_COMP_ARG_A,
+      RADEON_ALPHA_ARG_A_T1_ALPHA | RADEON_COMP_ARG_A,
+      RADEON_ALPHA_ARG_A_T2_ALPHA | RADEON_COMP_ARG_A
+   },
+};
+
+static GLuint radeon_tfactor_alpha[] =
+{
+   RADEON_ALPHA_ARG_A_TFACTOR_ALPHA,
+   RADEON_ALPHA_ARG_A_TFACTOR_ALPHA | RADEON_COMP_ARG_A
+};
+
+static GLuint radeon_primary_alpha[] =
+{
+   RADEON_ALPHA_ARG_A_DIFFUSE_ALPHA,
+   RADEON_ALPHA_ARG_A_DIFFUSE_ALPHA | RADEON_COMP_ARG_A
+};
+
+static GLuint radeon_previous_alpha[] =
+{
+   RADEON_ALPHA_ARG_A_CURRENT_ALPHA,
+   RADEON_ALPHA_ARG_A_CURRENT_ALPHA | RADEON_COMP_ARG_A
+};
+
+/* GL_ZERO table - indices 0-1
+ * GL_ONE  table - indices 1-2
+ */
+static GLuint radeon_zero_alpha[] =
+{
+   RADEON_ALPHA_ARG_A_ZERO,
+   RADEON_ALPHA_ARG_A_ZERO | RADEON_COMP_ARG_A,
+   RADEON_ALPHA_ARG_A_ZERO
+};
+
+
+/* Extract the arg from slot A, shift it into the correct argument slot
+ * and set the corresponding complement bit.
+ */
+#define RADEON_COLOR_ARG( n, arg )			\
+do {							\
+   color_combine |=					\
+      ((color_arg[n] & RADEON_COLOR_ARG_MASK)		\
+       << RADEON_COLOR_ARG_##arg##_SHIFT);		\
+   color_combine |=					\
+      ((color_arg[n] >> RADEON_COMP_ARG_SHIFT)		\
+       << RADEON_COMP_ARG_##arg##_SHIFT);		\
+} while (0)
+
+#define RADEON_ALPHA_ARG( n, arg )			\
+do {							\
+   alpha_combine |=					\
+      ((alpha_arg[n] & RADEON_ALPHA_ARG_MASK)		\
+       << RADEON_ALPHA_ARG_##arg##_SHIFT);		\
+   alpha_combine |=					\
+      ((alpha_arg[n] >> RADEON_COMP_ARG_SHIFT)		\
+       << RADEON_COMP_ARG_##arg##_SHIFT);		\
+} while (0)
+
+
+/* ================================================================
+ * Texture unit state management
+ */
+
+static GLboolean radeonUpdateTextureEnv( GLcontext *ctx, int unit )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   const struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
+   GLuint color_combine, alpha_combine;
+
+   /* texUnit->_Current can be NULL if and only if the texture unit is
+    * not actually enabled.
+    */
+   assert( (texUnit->_ReallyEnabled == 0)
+	   || (texUnit->_Current != NULL) );
+
+   if ( RADEON_DEBUG & DEBUG_TEXTURE ) {
+      fprintf( stderr, "%s( %p, %d )\n", __FUNCTION__, ctx, unit );
+   }
+
+   /* Set the texture environment state.  Isn't this nice and clean?
+    * The chip will automagically set the texture alpha to 0xff when
+    * the texture format does not include an alpha component.  This
+    * reduces the amount of special-casing we have to do, alpha-only
+    * textures being a notable exception.
+    */
+   if ( !texUnit->_ReallyEnabled ) {
+      /* Don't cache these results.
+       */
+      rmesa->state.texture.unit[unit].format = 0;
+      rmesa->state.texture.unit[unit].envMode = 0;
+      color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+      alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
+   }
+   else {
+      const struct gl_texture_object *tObj = texUnit->_Current;
+      const GLenum format = tObj->Image[tObj->BaseLevel]->Format;
+      GLuint color_arg[3], alpha_arg[3];
+      GLuint i, numColorArgs = 0, numAlphaArgs = 0;
+      GLuint RGBshift = texUnit->CombineScaleShiftRGB;
+      GLuint Ashift = texUnit->CombineScaleShiftA;
+
+      switch ( texUnit->EnvMode ) {
+      case GL_REPLACE:
+         switch ( format ) {
+	 case GL_RGBA:
+         case GL_LUMINANCE_ALPHA:
+         case GL_INTENSITY:
+	    color_combine = radeon_color_combine[unit][RADEON_REPLACE];
+	    alpha_combine = radeon_alpha_combine[unit][RADEON_REPLACE];
+	    break;
+	 case GL_ALPHA:
+	    color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+	    alpha_combine = radeon_alpha_combine[unit][RADEON_REPLACE];
+	    break;
+	 case GL_LUMINANCE:
+	 case GL_RGB:
+	 case GL_YCBCR_MESA:
+	    color_combine = radeon_color_combine[unit][RADEON_REPLACE];
+	    alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
+	    break;
+	 case GL_COLOR_INDEX:
+	 default:
+	    return GL_FALSE;
+	 }
+	 break;
+
+      case GL_MODULATE:
+	 switch ( format ) {
+	 case GL_RGBA:
+	 case GL_LUMINANCE_ALPHA:
+	 case GL_INTENSITY:
+	    color_combine = radeon_color_combine[unit][RADEON_MODULATE];
+	    alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	    break;
+	 case GL_ALPHA:
+	    color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+	    alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	    break;
+	 case GL_RGB:
+	 case GL_LUMINANCE:
+	 case GL_YCBCR_MESA:
+	    color_combine = radeon_color_combine[unit][RADEON_MODULATE];
+	    alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
+	    break;
+	 case GL_COLOR_INDEX:
+	 default:
+	    return GL_FALSE;
+	 }
+	 break;
+
+      case GL_DECAL:
+	 switch ( format ) {
+	 case GL_RGBA:
+	 case GL_RGB:
+	 case GL_YCBCR_MESA:
+	    color_combine = radeon_color_combine[unit][RADEON_DECAL];
+	    alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
+	    break;
+	 case GL_ALPHA:
+	 case GL_LUMINANCE:
+	 case GL_LUMINANCE_ALPHA:
+	 case GL_INTENSITY:
+	    color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+	    alpha_combine = radeon_alpha_combine[unit][RADEON_DISABLE];
+	    break;
+	 case GL_COLOR_INDEX:
+	 default:
+	    return GL_FALSE;
+	 }
+	 break;
+
+      case GL_BLEND:
+	 switch ( format ) {
+	 case GL_RGBA:
+	 case GL_RGB:
+	 case GL_LUMINANCE:
+	 case GL_LUMINANCE_ALPHA:
+	 case GL_YCBCR_MESA:
+	    color_combine = radeon_color_combine[unit][RADEON_BLEND];
+	    alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	    break;
+	 case GL_ALPHA:
+	    color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+	    alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	    break;
+	 case GL_INTENSITY:
+	    color_combine = radeon_color_combine[unit][RADEON_BLEND];
+	    alpha_combine = radeon_alpha_combine[unit][RADEON_BLEND];
+	    break;
+	 case GL_COLOR_INDEX:
+	 default:
+	    return GL_FALSE;
+	 }
+	 break;
+
+      case GL_ADD:
+	 switch ( format ) {
+	 case GL_RGBA:
+	 case GL_RGB:
+	 case GL_LUMINANCE:
+	 case GL_LUMINANCE_ALPHA:
+	 case GL_YCBCR_MESA:
+	    color_combine = radeon_color_combine[unit][RADEON_ADD];
+	    alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	    break;
+	 case GL_ALPHA:
+	    color_combine = radeon_color_combine[unit][RADEON_DISABLE];
+	    alpha_combine = radeon_alpha_combine[unit][RADEON_MODULATE];
+	    break;
+	 case GL_INTENSITY:
+	    color_combine = radeon_color_combine[unit][RADEON_ADD];
+	    alpha_combine = radeon_alpha_combine[unit][RADEON_ADD];
+	    break;
+	 case GL_COLOR_INDEX:
+	 default:
+	    return GL_FALSE;
+	 }
+	 break;
+
+      case GL_COMBINE:
+	 /* Don't cache these results.
+	  */
+	 rmesa->state.texture.unit[unit].format = 0;
+	 rmesa->state.texture.unit[unit].envMode = 0;
+
+	 /* Step 0:
+	  * Calculate how many arguments we need to process.
+	  */
+	 switch ( texUnit->CombineModeRGB ) {
+	 case GL_REPLACE:
+	    numColorArgs = 1;
+	    break;
+	 case GL_MODULATE:
+	 case GL_ADD:
+	 case GL_ADD_SIGNED:
+	 case GL_SUBTRACT:
+	 case GL_DOT3_RGB:
+	 case GL_DOT3_RGBA:
+	 case GL_DOT3_RGB_EXT:
+	 case GL_DOT3_RGBA_EXT:
+	    numColorArgs = 2;
+	    break;
+	 case GL_INTERPOLATE:
+	 case GL_MODULATE_ADD_ATI:
+	 case GL_MODULATE_SIGNED_ADD_ATI:
+	 case GL_MODULATE_SUBTRACT_ATI:
+	    numColorArgs = 3;
+	    break;
+	 default:
+	    return GL_FALSE;
+	 }
+
+	 switch ( texUnit->CombineModeA ) {
+	 case GL_REPLACE:
+	    numAlphaArgs = 1;
+	    break;
+	 case GL_MODULATE:
+	 case GL_ADD:
+	 case GL_ADD_SIGNED:
+	 case GL_SUBTRACT:
+	    numAlphaArgs = 2;
+	    break;
+	 case GL_INTERPOLATE:
+	 case GL_MODULATE_ADD_ATI:
+	 case GL_MODULATE_SIGNED_ADD_ATI:
+	 case GL_MODULATE_SUBTRACT_ATI:
+	    numAlphaArgs = 3;
+	    break;
+	 default:
+	    return GL_FALSE;
+	 }
+
+	 /* Step 1:
+	  * Extract the color and alpha combine function arguments.
+	  */
+	 for ( i = 0 ; i < numColorArgs ; i++ ) {
+	    const GLuint op = texUnit->CombineOperandRGB[i] - GL_SRC_COLOR;
+	    assert(op >= 0);
+	    assert(op <= 3);
+	    switch ( texUnit->CombineSourceRGB[i] ) {
+	    case GL_TEXTURE:
+	       color_arg[i] = radeon_texture_color[op][unit];
+	       break;
+	    case GL_CONSTANT:
+	       color_arg[i] = radeon_tfactor_color[op];
+	       break;
+	    case GL_PRIMARY_COLOR:
+	       color_arg[i] = radeon_primary_color[op];
+	       break;
+	    case GL_PREVIOUS:
+	       color_arg[i] = radeon_previous_color[op];
+	       break;
+	    case GL_ZERO:
+	       color_arg[i] = radeon_zero_color[op];
+	       break;
+	    case GL_ONE:
+	       color_arg[i] = radeon_zero_color[op+1];
+	       break;
+	    default:
+	       return GL_FALSE;
+	    }
+	 }
+
+	 for ( i = 0 ; i < numAlphaArgs ; i++ ) {
+	    const GLuint op = texUnit->CombineOperandA[i] - GL_SRC_ALPHA;
+	    assert(op >= 0);
+	    assert(op <= 1);
+	    switch ( texUnit->CombineSourceA[i] ) {
+	    case GL_TEXTURE:
+	       alpha_arg[i] = radeon_texture_alpha[op][unit];
+	       break;
+	    case GL_CONSTANT:
+	       alpha_arg[i] = radeon_tfactor_alpha[op];
+	       break;
+	    case GL_PRIMARY_COLOR:
+	       alpha_arg[i] = radeon_primary_alpha[op];
+	       break;
+	    case GL_PREVIOUS:
+	       alpha_arg[i] = radeon_previous_alpha[op];
+	       break;
+	    case GL_ZERO:
+	       alpha_arg[i] = radeon_zero_alpha[op];
+	       break;
+	    case GL_ONE:
+	       alpha_arg[i] = radeon_zero_alpha[op+1];
+	       break;
+	    default:
+	       return GL_FALSE;
+	    }
+	 }
+
+	 /* Step 2:
+	  * Build up the color and alpha combine functions.
+	  */
+	 switch ( texUnit->CombineModeRGB ) {
+	 case GL_REPLACE:
+	    color_combine = (RADEON_COLOR_ARG_A_ZERO |
+			     RADEON_COLOR_ARG_B_ZERO |
+			     RADEON_BLEND_CTL_ADD |
+			     RADEON_CLAMP_TX);
+	    RADEON_COLOR_ARG( 0, C );
+	    break;
+	 case GL_MODULATE:
+	    color_combine = (RADEON_COLOR_ARG_C_ZERO |
+			     RADEON_BLEND_CTL_ADD |
+			     RADEON_CLAMP_TX);
+	    RADEON_COLOR_ARG( 0, A );
+	    RADEON_COLOR_ARG( 1, B );
+	    break;
+	 case GL_ADD:
+	    color_combine = (RADEON_COLOR_ARG_B_ZERO |
+			     RADEON_COMP_ARG_B |
+			     RADEON_BLEND_CTL_ADD |
+			     RADEON_CLAMP_TX);
+	    RADEON_COLOR_ARG( 0, A );
+	    RADEON_COLOR_ARG( 1, C );
+	    break;
+	 case GL_ADD_SIGNED:
+	    color_combine = (RADEON_COLOR_ARG_B_ZERO |
+			     RADEON_COMP_ARG_B |
+			     RADEON_BLEND_CTL_ADDSIGNED |
+			     RADEON_CLAMP_TX);
+	    RADEON_COLOR_ARG( 0, A );
+	    RADEON_COLOR_ARG( 1, C );
+	    break;
+	 case GL_SUBTRACT:
+	    color_combine = (RADEON_COLOR_ARG_B_ZERO |
+			     RADEON_COMP_ARG_B |
+			     RADEON_BLEND_CTL_SUBTRACT |
+			     RADEON_CLAMP_TX);
+	    RADEON_COLOR_ARG( 0, A );
+	    RADEON_COLOR_ARG( 1, C );
+	    break;
+	 case GL_INTERPOLATE:
+	    color_combine = (RADEON_BLEND_CTL_BLEND |
+			     RADEON_CLAMP_TX);
+	    RADEON_COLOR_ARG( 0, B );
+	    RADEON_COLOR_ARG( 1, A );
+	    RADEON_COLOR_ARG( 2, C );
+	    break;
+
+	 case GL_DOT3_RGB_EXT:
+	 case GL_DOT3_RGBA_EXT:
+	    /* The EXT version of the DOT3 extension does not support the
+	     * scale factor, but the ARB version (and the version in OpenGL
+	     * 1.3) does.
+	     */
+	    RGBshift = 0;
+	    Ashift = 0;
+	    /* FALLTHROUGH */
+
+	 case GL_DOT3_RGB:
+	 case GL_DOT3_RGBA:
+	    /* The R100 / RV200 only support a 1X multiplier in hardware
+	     * w/the ARB version.
+	     */
+	    if ( RGBshift != (RADEON_SCALE_1X >> RADEON_SCALE_SHIFT) ) {
+	       return GL_FALSE;
+	    }
+
+	    RGBshift += 2;
+	    Ashift = RGBshift;
+
+	    color_combine = (RADEON_COLOR_ARG_C_ZERO |
+			     RADEON_BLEND_CTL_DOT3 |
+			     RADEON_CLAMP_TX);
+	    RADEON_COLOR_ARG( 0, A );
+	    RADEON_COLOR_ARG( 1, B );
+	    break;
+
+	 case GL_MODULATE_ADD_ATI:
+	    color_combine = (RADEON_BLEND_CTL_ADD |
+			     RADEON_CLAMP_TX);
+	    RADEON_COLOR_ARG( 0, A );
+	    RADEON_COLOR_ARG( 1, C );
+	    RADEON_COLOR_ARG( 2, B );
+	    break;
+	 case GL_MODULATE_SIGNED_ADD_ATI:
+	    color_combine = (RADEON_BLEND_CTL_ADDSIGNED |
+			     RADEON_CLAMP_TX);
+	    RADEON_COLOR_ARG( 0, A );
+	    RADEON_COLOR_ARG( 1, C );
+	    RADEON_COLOR_ARG( 2, B );
+	    break;
+	 case GL_MODULATE_SUBTRACT_ATI:
+	    color_combine = (RADEON_BLEND_CTL_SUBTRACT |
+			     RADEON_CLAMP_TX);
+	    RADEON_COLOR_ARG( 0, A );
+	    RADEON_COLOR_ARG( 1, C );
+	    RADEON_COLOR_ARG( 2, B );
+	    break;
+	 default:
+	    return GL_FALSE;
+	 }
+
+	 switch ( texUnit->CombineModeA ) {
+	 case GL_REPLACE:
+	    alpha_combine = (RADEON_ALPHA_ARG_A_ZERO |
+			     RADEON_ALPHA_ARG_B_ZERO |
+			     RADEON_BLEND_CTL_ADD |
+			     RADEON_CLAMP_TX);
+	    RADEON_ALPHA_ARG( 0, C );
+	    break;
+	 case GL_MODULATE:
+	    alpha_combine = (RADEON_ALPHA_ARG_C_ZERO |
+			     RADEON_BLEND_CTL_ADD |
+			     RADEON_CLAMP_TX);
+	    RADEON_ALPHA_ARG( 0, A );
+	    RADEON_ALPHA_ARG( 1, B );
+	    break;
+	 case GL_ADD:
+	    alpha_combine = (RADEON_ALPHA_ARG_B_ZERO |
+			     RADEON_COMP_ARG_B |
+			     RADEON_BLEND_CTL_ADD |
+			     RADEON_CLAMP_TX);
+	    RADEON_ALPHA_ARG( 0, A );
+	    RADEON_ALPHA_ARG( 1, C );
+	    break;
+	 case GL_ADD_SIGNED:
+	    alpha_combine = (RADEON_ALPHA_ARG_B_ZERO |
+			     RADEON_COMP_ARG_B |
+			     RADEON_BLEND_CTL_ADDSIGNED |
+			     RADEON_CLAMP_TX);
+	    RADEON_ALPHA_ARG( 0, A );
+	    RADEON_ALPHA_ARG( 1, C );
+	    break;
+	 case GL_SUBTRACT:
+	    alpha_combine = (RADEON_COLOR_ARG_B_ZERO |
+			     RADEON_COMP_ARG_B |
+			     RADEON_BLEND_CTL_SUBTRACT |
+			     RADEON_CLAMP_TX);
+	    RADEON_ALPHA_ARG( 0, A );
+	    RADEON_ALPHA_ARG( 1, C );
+	    break;
+	 case GL_INTERPOLATE:
+	    alpha_combine = (RADEON_BLEND_CTL_BLEND |
+			     RADEON_CLAMP_TX);
+	    RADEON_ALPHA_ARG( 0, B );
+	    RADEON_ALPHA_ARG( 1, A );
+	    RADEON_ALPHA_ARG( 2, C );
+	    break;
+
+	 case GL_MODULATE_ADD_ATI:
+	    alpha_combine = (RADEON_BLEND_CTL_ADD |
+			     RADEON_CLAMP_TX);
+	    RADEON_ALPHA_ARG( 0, A );
+	    RADEON_ALPHA_ARG( 1, C );
+	    RADEON_ALPHA_ARG( 2, B );
+	    break;
+	 case GL_MODULATE_SIGNED_ADD_ATI:
+	    alpha_combine = (RADEON_BLEND_CTL_ADDSIGNED |
+			     RADEON_CLAMP_TX);
+	    RADEON_ALPHA_ARG( 0, A );
+	    RADEON_ALPHA_ARG( 1, C );
+	    RADEON_ALPHA_ARG( 2, B );
+	    break;
+	 case GL_MODULATE_SUBTRACT_ATI:
+	    alpha_combine = (RADEON_BLEND_CTL_SUBTRACT |
+			     RADEON_CLAMP_TX);
+	    RADEON_ALPHA_ARG( 0, A );
+	    RADEON_ALPHA_ARG( 1, C );
+	    RADEON_ALPHA_ARG( 2, B );
+	    break;
+	 default:
+	    return GL_FALSE;
+	 }
+
+	 if ( (texUnit->CombineModeRGB == GL_DOT3_RGB_EXT)
+	      || (texUnit->CombineModeRGB == GL_DOT3_RGB) ) {
+	    alpha_combine |= RADEON_DOT_ALPHA_DONT_REPLICATE;
+	 }
+
+	 /* Step 3:
+	  * Apply the scale factor.
+	  */
+	 color_combine |= (RGBshift << RADEON_SCALE_SHIFT);
+	 alpha_combine |= (Ashift   << RADEON_SCALE_SHIFT);
+
+	 /* All done!
+	  */
+	 break;
+
+      default:
+	 return GL_FALSE;
+      }
+   }
+
+   if ( rmesa->hw.tex[unit].cmd[TEX_PP_TXCBLEND] != color_combine ||
+	rmesa->hw.tex[unit].cmd[TEX_PP_TXABLEND] != alpha_combine ) {
+      RADEON_STATECHANGE( rmesa, tex[unit] );
+      rmesa->hw.tex[unit].cmd[TEX_PP_TXCBLEND] = color_combine;
+      rmesa->hw.tex[unit].cmd[TEX_PP_TXABLEND] = alpha_combine;
+   }
+
+   return GL_TRUE;
+}
+
+#define TEXOBJ_TXFILTER_MASK (RADEON_MAX_MIP_LEVEL_MASK |	\
+			      RADEON_MIN_FILTER_MASK | 		\
+			      RADEON_MAG_FILTER_MASK |		\
+			      RADEON_MAX_ANISO_MASK |		\
+			      RADEON_YUV_TO_RGB |		\
+			      RADEON_YUV_TEMPERATURE_MASK |	\
+			      RADEON_CLAMP_S_MASK | 		\
+			      RADEON_CLAMP_T_MASK | 		\
+			      RADEON_BORDER_MODE_D3D )
+
+#define TEXOBJ_TXFORMAT_MASK (RADEON_TXFORMAT_WIDTH_MASK |	\
+			      RADEON_TXFORMAT_HEIGHT_MASK |	\
+			      RADEON_TXFORMAT_FORMAT_MASK |	\
+                              RADEON_TXFORMAT_F5_WIDTH_MASK |	\
+                              RADEON_TXFORMAT_F5_HEIGHT_MASK |	\
+			      RADEON_TXFORMAT_ALPHA_IN_MAP |	\
+			      RADEON_TXFORMAT_CUBIC_MAP_ENABLE |	\
+                              RADEON_TXFORMAT_NON_POWER2)
+
+
+static void import_tex_obj_state( radeonContextPtr rmesa,
+				  int unit,
+				  radeonTexObjPtr texobj )
+{
+   GLuint *cmd = RADEON_DB_STATE( tex[unit] );
+
+   cmd[TEX_PP_TXFILTER] &= ~TEXOBJ_TXFILTER_MASK;
+   cmd[TEX_PP_TXFILTER] |= texobj->pp_txfilter & TEXOBJ_TXFILTER_MASK;
+   cmd[TEX_PP_TXFORMAT] &= ~TEXOBJ_TXFORMAT_MASK;
+   cmd[TEX_PP_TXFORMAT] |= texobj->pp_txformat & TEXOBJ_TXFORMAT_MASK;
+   cmd[TEX_PP_TXOFFSET] = texobj->pp_txoffset;
+   cmd[TEX_PP_BORDER_COLOR] = texobj->pp_border_color;
+   RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.tex[unit] );
+
+   if (texobj->base.tObj->Target == GL_TEXTURE_RECTANGLE_NV) {
+      GLuint *txr_cmd = RADEON_DB_STATE( txr[unit] );
+      txr_cmd[TXR_PP_TEX_SIZE] = texobj->pp_txsize; /* NPOT only! */
+      txr_cmd[TXR_PP_TEX_PITCH] = texobj->pp_txpitch; /* NPOT only! */
+      RADEON_DB_STATECHANGE( rmesa, &rmesa->hw.txr[unit] );
+   }
+
+   texobj->dirty_state &= ~(1<<unit);
+}
+
+
+
+
+static void set_texgen_matrix( radeonContextPtr rmesa, 
+			       GLuint unit,
+			       const GLfloat *s_plane,
+			       const GLfloat *t_plane )
+{
+   static const GLfloat scale_identity[4] = { 1,1,1,1 };
+
+   if (!TEST_EQ_4V( s_plane, scale_identity) ||
+       !TEST_EQ_4V( t_plane, scale_identity)) {
+      rmesa->TexGenEnabled |= RADEON_TEXMAT_0_ENABLE<<unit;
+      rmesa->TexGenMatrix[unit].m[0]  = s_plane[0];
+      rmesa->TexGenMatrix[unit].m[4]  = s_plane[1];
+      rmesa->TexGenMatrix[unit].m[8]  = s_plane[2];
+      rmesa->TexGenMatrix[unit].m[12] = s_plane[3];
+
+      rmesa->TexGenMatrix[unit].m[1]  = t_plane[0];
+      rmesa->TexGenMatrix[unit].m[5]  = t_plane[1];
+      rmesa->TexGenMatrix[unit].m[9]  = t_plane[2];
+      rmesa->TexGenMatrix[unit].m[13] = t_plane[3];
+      rmesa->NewGLState |= _NEW_TEXTURE_MATRIX;
+   }
+}
+
+/* Ignoring the Q texcoord for now.
+ *
+ * Returns GL_FALSE if fallback required.  
+ */
+static GLboolean radeon_validate_texgen( GLcontext *ctx, GLuint unit )
+{  
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
+   GLuint inputshift = RADEON_TEXGEN_0_INPUT_SHIFT + unit*4;
+   GLuint tmp = rmesa->TexGenEnabled;
+
+   rmesa->TexGenEnabled &= ~(RADEON_TEXGEN_TEXMAT_0_ENABLE<<unit);
+   rmesa->TexGenEnabled &= ~(RADEON_TEXMAT_0_ENABLE<<unit);
+   rmesa->TexGenEnabled &= ~(RADEON_TEXGEN_INPUT_MASK<<inputshift);
+   rmesa->TexGenNeedNormals[unit] = 0;
+
+   if ((texUnit->TexGenEnabled & (S_BIT|T_BIT)) == 0) {
+      /* Disabled, no fallback:
+       */
+      rmesa->TexGenEnabled |= 
+	 (RADEON_TEXGEN_INPUT_TEXCOORD_0+unit) << inputshift;
+      return GL_TRUE;
+   }
+   else if (texUnit->TexGenEnabled & Q_BIT) {
+      /* Very easy to do this, in fact would remove a fallback case
+       * elsewhere, but I haven't done it yet...  Fallback: 
+       */
+      fprintf(stderr, "fallback Q_BIT\n");
+      return GL_FALSE;
+   }
+   else if ((texUnit->TexGenEnabled & (S_BIT|T_BIT)) != (S_BIT|T_BIT) ||
+	    texUnit->GenModeS != texUnit->GenModeT) {
+      /* Mixed modes, fallback:
+       */
+      /* fprintf(stderr, "fallback mixed texgen\n"); */
+      return GL_FALSE;
+   }
+   else
+      rmesa->TexGenEnabled |= RADEON_TEXGEN_TEXMAT_0_ENABLE << unit;
+
+   switch (texUnit->GenModeS) {
+   case GL_OBJECT_LINEAR:
+      rmesa->TexGenEnabled |= RADEON_TEXGEN_INPUT_OBJ << inputshift;
+      set_texgen_matrix( rmesa, unit, 
+			 texUnit->ObjectPlaneS,
+			 texUnit->ObjectPlaneT);
+      break;
+
+   case GL_EYE_LINEAR:
+      rmesa->TexGenEnabled |= RADEON_TEXGEN_INPUT_EYE << inputshift;
+      set_texgen_matrix( rmesa, unit, 
+			 texUnit->EyePlaneS,
+			 texUnit->EyePlaneT);
+      break;
+
+   case GL_REFLECTION_MAP_NV:
+      rmesa->TexGenNeedNormals[unit] = GL_TRUE;
+      rmesa->TexGenEnabled |= RADEON_TEXGEN_INPUT_EYE_REFLECT<<inputshift;
+      break;
+
+   case GL_NORMAL_MAP_NV:
+      rmesa->TexGenNeedNormals[unit] = GL_TRUE;
+      rmesa->TexGenEnabled |= RADEON_TEXGEN_INPUT_EYE_NORMAL<<inputshift;
+      break;
+
+   case GL_SPHERE_MAP:
+   default:
+      /* Unsupported mode, fallback:
+       */
+      /*  fprintf(stderr, "fallback unsupported texgen\n"); */
+      return GL_FALSE;
+   }
+
+   if (tmp != rmesa->TexGenEnabled) {
+      rmesa->NewGLState |= _NEW_TEXTURE_MATRIX;
+   }
+
+   return GL_TRUE;
+}
+
+
+static void disable_tex( GLcontext *ctx, int unit )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if (rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (RADEON_TEX_0_ENABLE<<unit)) {
+      /* Texture unit disabled */
+      if ( rmesa->state.texture.unit[unit].texobj != NULL ) {
+	 /* The old texture is no longer bound to this texture unit.
+	  * Mark it as such.
+	  */
+
+	 rmesa->state.texture.unit[unit].texobj->base.bound &= ~(1UL << unit);
+	 rmesa->state.texture.unit[unit].texobj = NULL;
+      }
+
+      RADEON_STATECHANGE( rmesa, ctx );
+      rmesa->hw.ctx.cmd[CTX_PP_CNTL] &= 
+	  ~((RADEON_TEX_0_ENABLE | RADEON_TEX_BLEND_0_ENABLE) << unit);
+
+      RADEON_STATECHANGE( rmesa, tcl );
+      switch (unit) {
+      case 0:
+	 rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] &= ~(RADEON_TCL_VTX_ST0 |
+						   RADEON_TCL_VTX_Q0);
+	    break;
+      case 1:
+	 rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] &= ~(RADEON_TCL_VTX_ST1 |
+						   RADEON_TCL_VTX_Q1);
+	 break;
+      default:
+	 break;
+      }
+
+
+      if (rmesa->TclFallback & (RADEON_TCL_FALLBACK_TEXGEN_0<<unit)) {
+	 TCL_FALLBACK( ctx, (RADEON_TCL_FALLBACK_TEXGEN_0<<unit), GL_FALSE);
+	 rmesa->recheck_texgen[unit] = GL_TRUE;
+      }
+
+
+
+      {
+	 GLuint inputshift = RADEON_TEXGEN_0_INPUT_SHIFT + unit*4;
+	 GLuint tmp = rmesa->TexGenEnabled;
+
+	 rmesa->TexGenEnabled &= ~(RADEON_TEXGEN_TEXMAT_0_ENABLE<<unit);
+	 rmesa->TexGenEnabled &= ~(RADEON_TEXMAT_0_ENABLE<<unit);
+	 rmesa->TexGenEnabled &= ~(RADEON_TEXGEN_INPUT_MASK<<inputshift);
+	 rmesa->TexGenNeedNormals[unit] = 0;
+	 rmesa->TexGenEnabled |= 
+	     (RADEON_TEXGEN_INPUT_TEXCOORD_0+unit) << inputshift;
+
+	 if (tmp != rmesa->TexGenEnabled) {
+	    rmesa->recheck_texgen[unit] = GL_TRUE;
+	    rmesa->NewGLState |= _NEW_TEXTURE_MATRIX;
+	 }
+      }
+   }
+}
+
+static GLboolean enable_tex_2d( GLcontext *ctx, int unit )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
+   struct gl_texture_object *tObj = texUnit->_Current;
+   radeonTexObjPtr t = (radeonTexObjPtr) tObj->DriverData;
+
+   /* Need to load the 2d images associated with this unit.
+    */
+   if (t->pp_txformat & RADEON_TXFORMAT_NON_POWER2) {
+      t->pp_txformat &= ~RADEON_TXFORMAT_NON_POWER2;
+      t->base.dirty_images[0] = ~0;
+   }
+
+   ASSERT(tObj->Target == GL_TEXTURE_2D || tObj->Target == GL_TEXTURE_1D);
+
+   if ( t->base.dirty_images[0] ) {
+      RADEON_FIREVERTICES( rmesa );
+      radeonSetTexImages( rmesa, tObj );
+      radeonUploadTexImages( rmesa, (radeonTexObjPtr) tObj->DriverData, 0 );
+      if ( !t->base.memBlock ) 
+	return GL_FALSE;
+   }
+
+   return GL_TRUE;
+}
+
+static GLboolean enable_tex_rect( GLcontext *ctx, int unit )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
+   struct gl_texture_object *tObj = texUnit->_Current;
+   radeonTexObjPtr t = (radeonTexObjPtr) tObj->DriverData;
+
+   if (!(t->pp_txformat & RADEON_TXFORMAT_NON_POWER2)) {
+      t->pp_txformat |= RADEON_TXFORMAT_NON_POWER2;
+      t->base.dirty_images[0] = ~0;
+   }
+
+   ASSERT(tObj->Target == GL_TEXTURE_RECTANGLE_NV);
+
+   if ( t->base.dirty_images[0] ) {
+      RADEON_FIREVERTICES( rmesa );
+      radeonSetTexImages( rmesa, tObj );
+      radeonUploadTexImages( rmesa, (radeonTexObjPtr) tObj->DriverData, 0 );
+      if ( !t->base.memBlock /* && !rmesa->prefer_agp_client_texturing  FIXME */ ) {
+	 fprintf(stderr, "%s: upload failed\n", __FUNCTION__);
+	 return GL_FALSE;
+      }
+   }
+
+   return GL_TRUE;
+}
+
+
+static GLboolean update_tex_common( GLcontext *ctx, int unit )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
+   struct gl_texture_object *tObj = texUnit->_Current;
+   radeonTexObjPtr t = (radeonTexObjPtr) tObj->DriverData;
+   GLenum format;
+
+   /* Fallback if there's a texture border */
+   if ( tObj->Image[tObj->BaseLevel]->Border > 0 ) {
+      fprintf(stderr, "%s: border\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+   /* Update state if this is a different texture object to last
+    * time.
+    */
+   if ( rmesa->state.texture.unit[unit].texobj != t ) {
+      if ( rmesa->state.texture.unit[unit].texobj != NULL ) {
+	 /* The old texture is no longer bound to this texture unit.
+	  * Mark it as such.
+	  */
+
+	 rmesa->state.texture.unit[unit].texobj->base.bound &= 
+	     ~(1UL << unit);
+      }
+
+      rmesa->state.texture.unit[unit].texobj = t;
+      t->base.bound |= (1UL << unit);
+      t->dirty_state |= 1<<unit;
+      driUpdateTextureLRU( (driTextureObject *) t ); /* XXX: should be locked! */
+   }
+
+
+   /* Newly enabled?
+    */
+   if ( !(rmesa->hw.ctx.cmd[CTX_PP_CNTL] & (RADEON_TEX_0_ENABLE<<unit))) {
+      RADEON_STATECHANGE( rmesa, ctx );
+      rmesa->hw.ctx.cmd[CTX_PP_CNTL] |= 
+	  (RADEON_TEX_0_ENABLE | RADEON_TEX_BLEND_0_ENABLE) << unit;
+
+      RADEON_STATECHANGE( rmesa, tcl );
+
+      if (unit == 0)
+	  rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_ST0;
+      else 
+	  rmesa->hw.tcl.cmd[TCL_OUTPUT_VTXFMT] |= RADEON_TCL_VTX_ST1;
+
+      rmesa->recheck_texgen[unit] = GL_TRUE;
+   }
+
+   if (t->dirty_state & (1<<unit)) {
+      import_tex_obj_state( rmesa, unit, t );
+   }
+
+   if (rmesa->recheck_texgen[unit]) {
+      GLboolean fallback = !radeon_validate_texgen( ctx, unit );
+      TCL_FALLBACK( ctx, (RADEON_TCL_FALLBACK_TEXGEN_0<<unit), fallback);
+      rmesa->recheck_texgen[unit] = 0;
+      rmesa->NewGLState |= _NEW_TEXTURE_MATRIX;
+   }
+
+   format = tObj->Image[tObj->BaseLevel]->Format;
+   if ( rmesa->state.texture.unit[unit].format != format ||
+	rmesa->state.texture.unit[unit].envMode != texUnit->EnvMode ) {
+      rmesa->state.texture.unit[unit].format = format;
+      rmesa->state.texture.unit[unit].envMode = texUnit->EnvMode;
+      if ( ! radeonUpdateTextureEnv( ctx, unit ) ) {
+	 return GL_FALSE;
+      }
+   }
+
+   FALLBACK( rmesa, RADEON_FALLBACK_BORDER_MODE, t->border_fallback );
+   return !t->border_fallback;
+}
+
+
+
+static GLboolean radeonUpdateTextureUnit( GLcontext *ctx, int unit )
+{
+   struct gl_texture_unit *texUnit = &ctx->Texture.Unit[unit];
+
+   TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_TEXRECT_0 << unit, 0 );
+
+   if ( texUnit->_ReallyEnabled & (TEXTURE_RECT_BIT) ) {
+      TCL_FALLBACK( ctx, RADEON_TCL_FALLBACK_TEXRECT_0 << unit, 1 );
+
+      return (enable_tex_rect( ctx, unit ) &&
+	      update_tex_common( ctx, unit ));
+   }
+   else if ( texUnit->_ReallyEnabled & (TEXTURE_1D_BIT | TEXTURE_2D_BIT) ) {
+      return (enable_tex_2d( ctx, unit ) &&
+	      update_tex_common( ctx, unit ));
+   }
+   else if ( texUnit->_ReallyEnabled ) {
+      return GL_FALSE;
+   }
+   else {
+      disable_tex( ctx, unit );
+      return GL_TRUE;
+   }
+}
+
+void radeonUpdateTextureState( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLboolean ok;
+
+   ok = (radeonUpdateTextureUnit( ctx, 0 ) &&
+	 radeonUpdateTextureUnit( ctx, 1 ));
+
+   FALLBACK( rmesa, RADEON_FALLBACK_TEXTURE, !ok );
+
+   if (rmesa->TclFallback)
+      radeonChooseVertexState( ctx );
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_vtxfmt.c b/src/mesa/drivers/dri/radeon/radeon_vtxfmt.c
new file mode 100644
index 00000000000..b613e9eb434
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_vtxfmt.c
@@ -0,0 +1,1089 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_vtxfmt.c,v 1.5 2002/12/16 16:18:59 dawes Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     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, 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 <[email protected]>
+ */
+#include "glheader.h"
+#include "imports.h"
+#include "api_noop.h"
+#include "api_arrayelt.h"
+#include "context.h"
+#include "mtypes.h"
+#include "enums.h"
+#include "glapi.h"
+#include "colormac.h"
+#include "light.h"
+#include "state.h"
+#include "vtxfmt.h"
+
+#include "tnl/tnl.h"
+#include "tnl/t_context.h"
+#include "tnl/t_array_api.h"
+
+#include "radeon_context.h"
+#include "radeon_state.h"
+#include "radeon_ioctl.h"
+#include "radeon_tex.h"
+#include "radeon_tcl.h"
+#include "radeon_swtcl.h"
+#include "radeon_vtxfmt.h"
+
+static void radeonVtxfmtFlushVertices( GLcontext *, GLuint );
+
+static void count_func( const char *name,  struct dynfn *l )
+{
+   int i = 0;
+   struct dynfn *f;
+   foreach (f, l) i++;
+   if (i) fprintf(stderr, "%s: %d\n", name, i );
+}
+
+static void count_funcs( radeonContextPtr rmesa )
+{
+   count_func( "Vertex2f", &rmesa->vb.dfn_cache.Vertex2f );
+   count_func( "Vertex2fv", &rmesa->vb.dfn_cache.Vertex2fv );
+   count_func( "Vertex3f", &rmesa->vb.dfn_cache.Vertex3f );
+   count_func( "Vertex3fv", &rmesa->vb.dfn_cache.Vertex3fv );
+   count_func( "Color4ub", &rmesa->vb.dfn_cache.Color4ub );
+   count_func( "Color4ubv", &rmesa->vb.dfn_cache.Color4ubv );
+   count_func( "Color3ub", &rmesa->vb.dfn_cache.Color3ub );
+   count_func( "Color3ubv", &rmesa->vb.dfn_cache.Color3ubv );
+   count_func( "Color4f", &rmesa->vb.dfn_cache.Color4f );
+   count_func( "Color4fv", &rmesa->vb.dfn_cache.Color4fv );
+   count_func( "Color3f", &rmesa->vb.dfn_cache.Color3f );
+   count_func( "Color3fv", &rmesa->vb.dfn_cache.Color3fv );
+   count_func( "SecondaryColor3f", &rmesa->vb.dfn_cache.SecondaryColor3fEXT );
+   count_func( "SecondaryColor3fv", &rmesa->vb.dfn_cache.SecondaryColor3fvEXT );
+   count_func( "SecondaryColor3ub", &rmesa->vb.dfn_cache.SecondaryColor3ubEXT );
+   count_func( "SecondaryColor3ubv", &rmesa->vb.dfn_cache.SecondaryColor3ubvEXT );
+   count_func( "Normal3f", &rmesa->vb.dfn_cache.Normal3f );
+   count_func( "Normal3fv", &rmesa->vb.dfn_cache.Normal3fv );
+   count_func( "TexCoord2f", &rmesa->vb.dfn_cache.TexCoord2f );
+   count_func( "TexCoord2fv", &rmesa->vb.dfn_cache.TexCoord2fv );
+   count_func( "TexCoord1f", &rmesa->vb.dfn_cache.TexCoord1f );
+   count_func( "TexCoord1fv", &rmesa->vb.dfn_cache.TexCoord1fv );
+   count_func( "MultiTexCoord2fARB", &rmesa->vb.dfn_cache.MultiTexCoord2fARB );
+   count_func( "MultiTexCoord2fvARB", &rmesa->vb.dfn_cache.MultiTexCoord2fvARB );
+   count_func( "MultiTexCoord1fARB", &rmesa->vb.dfn_cache.MultiTexCoord1fARB );
+   count_func( "MultiTexCoord1fvARB", &rmesa->vb.dfn_cache.MultiTexCoord1fvARB );
+}
+
+
+void radeon_copy_to_current( GLcontext *ctx ) 
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   assert(ctx->Driver.NeedFlush & FLUSH_UPDATE_CURRENT);
+
+   if (rmesa->vb.vertex_format & RADEON_CP_VC_FRMT_N0) {
+      ctx->Current.Attrib[VERT_ATTRIB_NORMAL][0] = rmesa->vb.normalptr[0];
+      ctx->Current.Attrib[VERT_ATTRIB_NORMAL][1] = rmesa->vb.normalptr[1];
+      ctx->Current.Attrib[VERT_ATTRIB_NORMAL][2] = rmesa->vb.normalptr[2];
+   }
+
+   if (rmesa->vb.vertex_format & RADEON_CP_VC_FRMT_PKCOLOR) {
+      ctx->Current.Attrib[VERT_ATTRIB_COLOR0][0] = UBYTE_TO_FLOAT( rmesa->vb.colorptr->red );
+      ctx->Current.Attrib[VERT_ATTRIB_COLOR0][1] = UBYTE_TO_FLOAT( rmesa->vb.colorptr->green );
+      ctx->Current.Attrib[VERT_ATTRIB_COLOR0][2] = UBYTE_TO_FLOAT( rmesa->vb.colorptr->blue );
+      ctx->Current.Attrib[VERT_ATTRIB_COLOR0][3] = UBYTE_TO_FLOAT( rmesa->vb.colorptr->alpha );
+   } 
+   
+   if (rmesa->vb.vertex_format & RADEON_CP_VC_FRMT_FPCOLOR) {
+      ctx->Current.Attrib[VERT_ATTRIB_COLOR0][0] = rmesa->vb.floatcolorptr[0];
+      ctx->Current.Attrib[VERT_ATTRIB_COLOR0][1] = rmesa->vb.floatcolorptr[1];
+      ctx->Current.Attrib[VERT_ATTRIB_COLOR0][2] = rmesa->vb.floatcolorptr[2];
+   }
+
+   if (rmesa->vb.vertex_format & RADEON_CP_VC_FRMT_FPALPHA)
+      ctx->Current.Attrib[VERT_ATTRIB_COLOR0][3] = rmesa->vb.floatcolorptr[3];
+      
+   if (rmesa->vb.vertex_format & RADEON_CP_VC_FRMT_PKSPEC) {
+      ctx->Current.Attrib[VERT_ATTRIB_COLOR1][0] = UBYTE_TO_FLOAT( rmesa->vb.specptr->red );
+      ctx->Current.Attrib[VERT_ATTRIB_COLOR1][1] = UBYTE_TO_FLOAT( rmesa->vb.specptr->green );
+      ctx->Current.Attrib[VERT_ATTRIB_COLOR1][2] = UBYTE_TO_FLOAT( rmesa->vb.specptr->blue );
+   } 
+
+   if (rmesa->vb.vertex_format & RADEON_CP_VC_FRMT_ST0) {
+      ctx->Current.Attrib[VERT_ATTRIB_TEX0][0] = rmesa->vb.texcoordptr[0][0];
+      ctx->Current.Attrib[VERT_ATTRIB_TEX0][1] = rmesa->vb.texcoordptr[0][1];
+      ctx->Current.Attrib[VERT_ATTRIB_TEX0][2] = 0.0F;
+      ctx->Current.Attrib[VERT_ATTRIB_TEX0][3] = 1.0F;
+   }
+
+   if (rmesa->vb.vertex_format & RADEON_CP_VC_FRMT_ST1) {
+      ctx->Current.Attrib[VERT_ATTRIB_TEX1][0] = rmesa->vb.texcoordptr[1][0];
+      ctx->Current.Attrib[VERT_ATTRIB_TEX1][1] = rmesa->vb.texcoordptr[1][1];
+      ctx->Current.Attrib[VERT_ATTRIB_TEX1][2] = 0.0F;
+      ctx->Current.Attrib[VERT_ATTRIB_TEX1][3] = 1.0F;
+   }
+
+   ctx->Driver.NeedFlush &= ~FLUSH_UPDATE_CURRENT;
+}
+
+static GLboolean discreet_gl_prim[GL_POLYGON+1] = {
+   1,				/* 0 points */
+   1,				/* 1 lines */
+   0,				/* 2 line_strip */
+   0,				/* 3 line_loop */
+   1,				/* 4 tris */
+   0,				/* 5 tri_fan */
+   0,				/* 6 tri_strip */
+   1,				/* 7 quads */
+   0,				/* 8 quadstrip */
+   0,				/* 9 poly */
+};
+
+static void flush_prims( radeonContextPtr rmesa )
+{
+   int i,j;
+   struct radeon_dma_region tmp = rmesa->dma.current;
+   
+   tmp.buf->refcount++;
+   tmp.aos_size = rmesa->vb.vertex_size;
+   tmp.aos_stride = rmesa->vb.vertex_size;
+   tmp.aos_start = GET_START(&tmp);
+
+   rmesa->dma.current.ptr = rmesa->dma.current.start += 
+      (rmesa->vb.initial_counter - rmesa->vb.counter) * rmesa->vb.vertex_size * 4; 
+
+   rmesa->tcl.vertex_format = rmesa->vb.vertex_format;
+   rmesa->tcl.aos_components[0] = &tmp;
+   rmesa->tcl.nr_aos_components = 1;
+   rmesa->dma.flush = 0;
+
+   /* Optimize the primitive list:
+    */
+   if (rmesa->vb.nrprims > 1) {
+      for (j = 0, i = 1 ; i < rmesa->vb.nrprims; i++) {
+	 int pj = rmesa->vb.primlist[j].prim & 0xf;
+	 int pi = rmesa->vb.primlist[i].prim & 0xf;
+      
+	 if (pj == pi && discreet_gl_prim[pj] &&
+	     rmesa->vb.primlist[i].start == rmesa->vb.primlist[j].end) {
+	    rmesa->vb.primlist[j].end = rmesa->vb.primlist[i].end;
+	 }
+	 else {
+	    j++;
+	    if (j != i) rmesa->vb.primlist[j] = rmesa->vb.primlist[i];
+	 }
+      }
+      rmesa->vb.nrprims = j+1;
+   }
+
+   for (i = 0 ; i < rmesa->vb.nrprims; i++) {
+      if (RADEON_DEBUG & DEBUG_PRIMS)
+	 fprintf(stderr, "vtxfmt prim %d: %s %d..%d\n", i,
+		 _mesa_lookup_enum_by_nr( rmesa->vb.primlist[i].prim & 
+					  PRIM_MODE_MASK ),
+		 rmesa->vb.primlist[i].start,
+		 rmesa->vb.primlist[i].end);
+
+      radeonEmitPrimitive( rmesa->glCtx,
+			   rmesa->vb.primlist[i].start,
+			   rmesa->vb.primlist[i].end,
+			   rmesa->vb.primlist[i].prim );
+   }
+
+   rmesa->vb.nrprims = 0;
+   radeonReleaseDmaRegion( rmesa, &tmp, __FUNCTION__ );
+}
+
+
+static void start_prim( radeonContextPtr rmesa, GLuint mode )
+{
+   if (RADEON_DEBUG & DEBUG_VFMT)
+      fprintf(stderr, "%s %d\n", __FUNCTION__, rmesa->vb.initial_counter - rmesa->vb.counter);
+
+   rmesa->vb.primlist[rmesa->vb.nrprims].start = rmesa->vb.initial_counter - rmesa->vb.counter;
+   rmesa->vb.primlist[rmesa->vb.nrprims].prim = mode;
+}
+
+static void note_last_prim( radeonContextPtr rmesa, GLuint flags )
+{
+   if (RADEON_DEBUG & DEBUG_VFMT)
+      fprintf(stderr, "%s %d\n", __FUNCTION__, rmesa->vb.initial_counter - rmesa->vb.counter);
+
+   if (rmesa->vb.prim[0] != GL_POLYGON+1) {
+      rmesa->vb.primlist[rmesa->vb.nrprims].prim |= flags;
+      rmesa->vb.primlist[rmesa->vb.nrprims].end = rmesa->vb.initial_counter - rmesa->vb.counter;
+
+      if (++(rmesa->vb.nrprims) == RADEON_MAX_PRIMS)
+	 flush_prims( rmesa );
+   }
+}
+
+
+static void copy_vertex( radeonContextPtr rmesa, GLuint n, GLfloat *dst )
+{
+   GLuint i;
+   GLfloat *src = (GLfloat *)(rmesa->dma.current.address + 
+			      rmesa->dma.current.ptr + 
+			      (rmesa->vb.primlist[rmesa->vb.nrprims].start + n) * 
+			      rmesa->vb.vertex_size * 4);
+
+   if (RADEON_DEBUG & DEBUG_VFMT) 
+      fprintf(stderr, "copy_vertex %d\n", rmesa->vb.primlist[rmesa->vb.nrprims].start + n);
+
+   for (i = 0 ; i < rmesa->vb.vertex_size; i++) {
+      dst[i] = src[i];
+   }
+}
+
+/* NOTE: This actually reads the copied vertices back from uncached
+ * memory.  Could also use the counter/notify mechanism to populate
+ * tmp on the fly as vertices are generated.  
+ */
+static GLuint copy_dma_verts( radeonContextPtr rmesa, GLfloat (*tmp)[15] )
+{
+   GLuint ovf, i;
+   GLuint nr = (rmesa->vb.initial_counter - rmesa->vb.counter) - rmesa->vb.primlist[rmesa->vb.nrprims].start;
+
+   if (RADEON_DEBUG & DEBUG_VFMT)
+      fprintf(stderr, "%s %d verts\n", __FUNCTION__, nr);
+
+   switch( rmesa->vb.prim[0] )
+   {
+   case GL_POINTS:
+      return 0;
+   case GL_LINES:
+      ovf = nr&1;
+      for (i = 0 ; i < ovf ; i++)
+	 copy_vertex( rmesa, nr-ovf+i, tmp[i] );
+      return i;
+   case GL_TRIANGLES:
+      ovf = nr%3;
+      for (i = 0 ; i < ovf ; i++)
+	 copy_vertex( rmesa, nr-ovf+i, tmp[i] );
+      return i;
+   case GL_QUADS:
+      ovf = nr&3;
+      for (i = 0 ; i < ovf ; i++)
+	 copy_vertex( rmesa, nr-ovf+i, tmp[i] );
+      return i;
+   case GL_LINE_STRIP:
+      if (nr == 0) 
+	 return 0;
+      copy_vertex( rmesa, nr-1, tmp[0] );
+      return 1;
+   case GL_LINE_LOOP:
+   case GL_TRIANGLE_FAN:
+   case GL_POLYGON:
+      if (nr == 0) 
+	 return 0;
+      else if (nr == 1) {
+	 copy_vertex( rmesa, 0, tmp[0] );
+	 return 1;
+      } else {
+	 copy_vertex( rmesa, 0, tmp[0] );
+	 copy_vertex( rmesa, nr-1, tmp[1] );
+	 return 2;
+      }
+   case GL_TRIANGLE_STRIP:
+      ovf = MIN2(nr, 2);
+      for (i = 0 ; i < ovf ; i++)
+	 copy_vertex( rmesa, nr-ovf+i, tmp[i] );
+      return i;
+   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++)
+	 copy_vertex( rmesa, nr-ovf+i, tmp[i] );
+      return i;
+   default:
+      assert(0);
+      return 0;
+   }
+}
+
+static void VFMT_FALLBACK_OUTSIDE_BEGIN_END( const char *caller )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if (RADEON_DEBUG & (DEBUG_VFMT|DEBUG_FALLBACKS))
+      fprintf(stderr, "%s from %s\n", __FUNCTION__, caller);
+
+   if (ctx->Driver.NeedFlush) 
+      radeonVtxfmtFlushVertices( ctx, ctx->Driver.NeedFlush );
+
+   if (ctx->NewState)
+      _mesa_update_state( ctx ); /* clear state so fell_back sticks */
+
+   _tnl_wakeup_exec( ctx );
+   ctx->Driver.FlushVertices = radeonFlushVertices;
+
+   assert( rmesa->dma.flush == 0 );
+   rmesa->vb.fell_back = GL_TRUE;
+   rmesa->vb.installed = GL_FALSE;
+}
+
+
+static void VFMT_FALLBACK( const char *caller )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat tmp[3][15];
+   GLuint i, prim;
+   GLuint ind = rmesa->vb.vertex_format;
+   GLuint nrverts;
+   GLfloat alpha = 1.0;
+
+   if (RADEON_DEBUG & (DEBUG_FALLBACKS|DEBUG_VFMT))
+      fprintf(stderr, "%s from %s\n", __FUNCTION__, caller);
+
+   if (rmesa->vb.prim[0] == GL_POLYGON+1) {
+      VFMT_FALLBACK_OUTSIDE_BEGIN_END( __FUNCTION__ );
+      return;
+   }
+
+   /* Copy vertices out of dma:
+    */
+   nrverts = copy_dma_verts( rmesa, tmp );
+
+   /* Finish the prim at this point:
+    */
+   note_last_prim( rmesa, 0 );
+   flush_prims( rmesa );
+
+   /* Update ctx->Driver.CurrentExecPrimitive and swap in swtnl. 
+    */
+   prim = rmesa->vb.prim[0];
+   ctx->Driver.CurrentExecPrimitive = GL_POLYGON+1;
+   _tnl_wakeup_exec( ctx );
+   ctx->Driver.FlushVertices = radeonFlushVertices;
+
+   assert(rmesa->dma.flush == 0);
+   rmesa->vb.fell_back = GL_TRUE;
+   rmesa->vb.installed = GL_FALSE;
+   glBegin( prim );
+   
+   if (rmesa->vb.installed_color_3f_sz == 4)
+      alpha = ctx->Current.Attrib[VERT_ATTRIB_COLOR0][3];
+
+   /* Replay saved vertices
+    */
+   for (i = 0 ; i < nrverts; i++) {
+      GLuint offset = 3;
+      if (ind & RADEON_CP_VC_FRMT_N0) {
+	 glNormal3fv( &tmp[i][offset] ); 
+	 offset += 3;
+      }
+
+      if (ind & RADEON_CP_VC_FRMT_PKCOLOR) {
+	 radeon_color_t *col = (radeon_color_t *)&tmp[i][offset];
+	 glColor4ub( col->red, col->green, col->blue, col->alpha );
+	 offset++;
+      }
+      else if (ind & RADEON_CP_VC_FRMT_FPALPHA) {
+	 glColor4fv( &tmp[i][offset] ); 
+	 offset+=4;
+      } 
+      else if (ind & RADEON_CP_VC_FRMT_FPCOLOR) {
+	 glColor3fv( &tmp[i][offset] ); 
+	 offset+=3;
+      }
+
+      if (ind & RADEON_CP_VC_FRMT_PKSPEC) {
+	 radeon_color_t *spec = (radeon_color_t *)&tmp[i][offset];
+	 _glapi_Dispatch->SecondaryColor3ubEXT( spec->red, spec->green, spec->blue );
+	 offset++;
+      }
+
+      if (ind & RADEON_CP_VC_FRMT_ST0) {
+	 glTexCoord2fv( &tmp[i][offset] ); 
+	 offset += 2;
+      }
+
+      if (ind & RADEON_CP_VC_FRMT_ST1) {
+	 glMultiTexCoord2fvARB( GL_TEXTURE1_ARB, &tmp[i][offset] );
+	 offset += 2;
+      }
+      glVertex3fv( &tmp[i][0] );
+   }
+
+   /* Replay current vertex
+    */
+   if (ind & RADEON_CP_VC_FRMT_N0) 
+      glNormal3fv( rmesa->vb.normalptr );
+
+   if (ind & RADEON_CP_VC_FRMT_PKCOLOR)
+      glColor4ub( rmesa->vb.colorptr->red, rmesa->vb.colorptr->green, rmesa->vb.colorptr->blue, rmesa->vb.colorptr->alpha );
+   else if (ind & RADEON_CP_VC_FRMT_FPALPHA)
+      glColor4fv( rmesa->vb.floatcolorptr );
+   else if (ind & RADEON_CP_VC_FRMT_FPCOLOR) {
+      if (rmesa->vb.installed_color_3f_sz == 4 && alpha != 1.0)
+	 glColor4f( rmesa->vb.floatcolorptr[0],
+		    rmesa->vb.floatcolorptr[1],
+		    rmesa->vb.floatcolorptr[2],
+		    alpha );
+      else
+	 glColor3fv( rmesa->vb.floatcolorptr );
+   }
+
+   if (ind & RADEON_CP_VC_FRMT_PKSPEC) 
+      _glapi_Dispatch->SecondaryColor3ubEXT( rmesa->vb.specptr->red, rmesa->vb.specptr->green, rmesa->vb.specptr->blue ); 
+
+   if (ind & RADEON_CP_VC_FRMT_ST0) 
+      glTexCoord2fv( rmesa->vb.texcoordptr[0] );
+
+   if (ind & RADEON_CP_VC_FRMT_ST1) 
+      glMultiTexCoord2fvARB( GL_TEXTURE1_ARB, rmesa->vb.texcoordptr[1] );
+}
+
+
+
+static void wrap_buffer( void )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat tmp[3][15];
+   GLuint i, nrverts;
+
+   if (RADEON_DEBUG & (DEBUG_VFMT|DEBUG_PRIMS))
+      fprintf(stderr, "%s %d\n", __FUNCTION__, rmesa->vb.initial_counter - rmesa->vb.counter);
+
+   /* Don't deal with parity.
+    */
+   if ((((rmesa->vb.initial_counter - rmesa->vb.counter) -  
+	 rmesa->vb.primlist[rmesa->vb.nrprims].start) & 1)) {
+      rmesa->vb.counter++;
+      rmesa->vb.initial_counter++;
+      return;
+   }
+
+   /* Copy vertices out of dma:
+    */
+   if (rmesa->vb.prim[0] == GL_POLYGON+1) 
+      nrverts = 0;
+   else {
+      nrverts = copy_dma_verts( rmesa, tmp );
+
+      if (RADEON_DEBUG & DEBUG_VFMT)
+	 fprintf(stderr, "%d vertices to copy\n", nrverts);
+   
+      /* Finish the prim at this point:
+       */
+      note_last_prim( rmesa, 0 );
+   }
+
+   /* Fire any buffered primitives
+    */
+   flush_prims( rmesa );
+
+   /* Get new buffer
+    */
+   radeonRefillCurrentDmaRegion( rmesa );
+
+   /* Reset counter, dmaptr
+    */
+   rmesa->vb.dmaptr = (int *)(rmesa->dma.current.ptr + rmesa->dma.current.address);
+   rmesa->vb.counter = (rmesa->dma.current.end - rmesa->dma.current.ptr) / 
+      (rmesa->vb.vertex_size * 4);
+   rmesa->vb.counter--;
+   rmesa->vb.initial_counter = rmesa->vb.counter;
+   rmesa->vb.notify = wrap_buffer;
+
+   rmesa->dma.flush = flush_prims;
+
+   /* Restart wrapped primitive:
+    */
+   if (rmesa->vb.prim[0] != GL_POLYGON+1)
+      start_prim( rmesa, rmesa->vb.prim[0] );
+
+   /* Reemit saved vertices
+    */
+   for (i = 0 ; i < nrverts; i++) {
+      if (RADEON_DEBUG & DEBUG_VERTS) {
+	 int j;
+	 fprintf(stderr, "re-emit vertex %d to %p\n", i, rmesa->vb.dmaptr);
+	 if (RADEON_DEBUG & DEBUG_VERBOSE)
+	    for (j = 0 ; j < rmesa->vb.vertex_size; j++) 
+	       fprintf(stderr, "\t%08x/%f\n", *(int*)&tmp[i][j], tmp[i][j]);
+      }
+
+      memcpy( rmesa->vb.dmaptr, tmp[i], rmesa->vb.vertex_size * 4 );
+      rmesa->vb.dmaptr += rmesa->vb.vertex_size;
+      rmesa->vb.counter--;
+   }
+}
+
+
+
+static GLboolean check_vtx_fmt( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLuint ind = RADEON_CP_VC_FRMT_Z;
+
+   if (rmesa->TclFallback || rmesa->vb.fell_back || ctx->CompileFlag)
+      return GL_FALSE;
+
+   if (ctx->Driver.NeedFlush & FLUSH_UPDATE_CURRENT) 
+      ctx->Driver.FlushVertices( ctx, FLUSH_UPDATE_CURRENT );
+   
+   /* Make all this event-driven:
+    */
+   if (ctx->Light.Enabled) {
+      ind |= RADEON_CP_VC_FRMT_N0;
+
+      /* TODO: make this data driven: If we receive only ubytes, send
+       * color as ubytes.  Also check if converting (with free
+       * checking for overflow) is cheaper than sending floats
+       * directly.
+       */
+      if (ctx->Light.ColorMaterialEnabled) {
+	 ind |= (RADEON_CP_VC_FRMT_FPCOLOR |
+		 RADEON_CP_VC_FRMT_FPALPHA);
+      }
+      else
+	 ind |= RADEON_CP_VC_FRMT_PKCOLOR; /* for alpha? */
+   }
+   else {
+      /* TODO: make this data driven?
+       */
+      ind |= RADEON_CP_VC_FRMT_PKCOLOR;
+	 
+      if (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR) {
+	 ind |= RADEON_CP_VC_FRMT_PKSPEC;
+      }
+   }
+
+   if (ctx->Texture.Unit[0]._ReallyEnabled) {
+      if (ctx->Texture.Unit[0].TexGenEnabled) {
+	 if (rmesa->TexGenNeedNormals[0]) {
+	    ind |= RADEON_CP_VC_FRMT_N0;
+	 }
+      } else {
+	 if (ctx->Current.Attrib[VERT_ATTRIB_TEX0][2] != 0.0F ||
+	     ctx->Current.Attrib[VERT_ATTRIB_TEX0][3] != 1.0) {
+	    if (RADEON_DEBUG & (DEBUG_VFMT|DEBUG_FALLBACKS))
+	       fprintf(stderr, "%s: rq0\n", __FUNCTION__);
+	    return GL_FALSE;
+	 }
+	 ind |= RADEON_CP_VC_FRMT_ST0;
+      }
+   }
+
+   if (ctx->Texture.Unit[1]._ReallyEnabled) {
+      if (ctx->Texture.Unit[1].TexGenEnabled) {
+	 if (rmesa->TexGenNeedNormals[1]) {
+	    ind |= RADEON_CP_VC_FRMT_N0;
+	 }
+      } else {
+	 if (ctx->Current.Attrib[VERT_ATTRIB_TEX1][2] != 0.0F ||
+	     ctx->Current.Attrib[VERT_ATTRIB_TEX1][3] != 1.0) {
+	    if (RADEON_DEBUG & (DEBUG_VFMT|DEBUG_FALLBACKS))
+	       fprintf(stderr, "%s: rq1\n", __FUNCTION__);
+	    return GL_FALSE;
+	 }
+	 ind |= RADEON_CP_VC_FRMT_ST1;
+      }
+   }
+
+   if (RADEON_DEBUG & (DEBUG_VFMT|DEBUG_STATE))
+      fprintf(stderr, "%s: format: 0x%x\n", __FUNCTION__, ind );
+
+   RADEON_NEWPRIM(rmesa);
+   rmesa->vb.vertex_format = ind;
+   rmesa->vb.vertex_size = 3;
+   rmesa->vb.prim = &ctx->Driver.CurrentExecPrimitive;
+
+   rmesa->vb.normalptr = ctx->Current.Attrib[VERT_ATTRIB_NORMAL];
+   rmesa->vb.colorptr = NULL;
+   rmesa->vb.floatcolorptr = ctx->Current.Attrib[VERT_ATTRIB_COLOR0];
+   rmesa->vb.specptr = NULL;
+   rmesa->vb.floatspecptr = ctx->Current.Attrib[VERT_ATTRIB_COLOR1];
+   rmesa->vb.texcoordptr[0] = ctx->Current.Attrib[VERT_ATTRIB_TEX0];
+   rmesa->vb.texcoordptr[1] = ctx->Current.Attrib[VERT_ATTRIB_TEX1];
+
+   /* Run through and initialize the vertex components in the order
+    * the hardware understands:
+    */
+   if (ind & RADEON_CP_VC_FRMT_N0) {
+      rmesa->vb.normalptr = &rmesa->vb.vertex[rmesa->vb.vertex_size].f;
+      rmesa->vb.vertex_size += 3;
+      rmesa->vb.normalptr[0] = ctx->Current.Attrib[VERT_ATTRIB_NORMAL][0];
+      rmesa->vb.normalptr[1] = ctx->Current.Attrib[VERT_ATTRIB_NORMAL][1];
+      rmesa->vb.normalptr[2] = ctx->Current.Attrib[VERT_ATTRIB_NORMAL][2];
+   }
+
+   if (ind & RADEON_CP_VC_FRMT_PKCOLOR) {
+      rmesa->vb.colorptr = &rmesa->vb.vertex[rmesa->vb.vertex_size].color;
+      rmesa->vb.vertex_size += 1;
+      UNCLAMPED_FLOAT_TO_CHAN( rmesa->vb.colorptr->red,   ctx->Current.Attrib[VERT_ATTRIB_COLOR0][0] );
+      UNCLAMPED_FLOAT_TO_CHAN( rmesa->vb.colorptr->green, ctx->Current.Attrib[VERT_ATTRIB_COLOR0][1] );
+      UNCLAMPED_FLOAT_TO_CHAN( rmesa->vb.colorptr->blue,  ctx->Current.Attrib[VERT_ATTRIB_COLOR0][2] );
+      UNCLAMPED_FLOAT_TO_CHAN( rmesa->vb.colorptr->alpha, ctx->Current.Attrib[VERT_ATTRIB_COLOR0][3] );
+   }
+
+   if (ind & RADEON_CP_VC_FRMT_FPCOLOR) {
+      assert(!(ind & RADEON_CP_VC_FRMT_PKCOLOR));
+      rmesa->vb.floatcolorptr = &rmesa->vb.vertex[rmesa->vb.vertex_size].f;
+      rmesa->vb.vertex_size += 3;
+      rmesa->vb.floatcolorptr[0] = ctx->Current.Attrib[VERT_ATTRIB_COLOR0][0];
+      rmesa->vb.floatcolorptr[1] = ctx->Current.Attrib[VERT_ATTRIB_COLOR0][1];
+      rmesa->vb.floatcolorptr[2] = ctx->Current.Attrib[VERT_ATTRIB_COLOR0][2];
+
+      if (ind & RADEON_CP_VC_FRMT_FPALPHA) {
+	 rmesa->vb.vertex_size += 1;
+	 rmesa->vb.floatcolorptr[3] = ctx->Current.Attrib[VERT_ATTRIB_COLOR0][3];
+      }
+   }
+   
+   if (ind & RADEON_CP_VC_FRMT_PKSPEC) {
+      rmesa->vb.specptr = &rmesa->vb.vertex[rmesa->vb.vertex_size].color;
+      rmesa->vb.vertex_size += 1;
+      UNCLAMPED_FLOAT_TO_CHAN( rmesa->vb.specptr->red,   ctx->Current.Attrib[VERT_ATTRIB_COLOR1][0] );
+      UNCLAMPED_FLOAT_TO_CHAN( rmesa->vb.specptr->green, ctx->Current.Attrib[VERT_ATTRIB_COLOR1][1] );
+      UNCLAMPED_FLOAT_TO_CHAN( rmesa->vb.specptr->blue,  ctx->Current.Attrib[VERT_ATTRIB_COLOR1][2] );
+   }
+
+   if (ind & RADEON_CP_VC_FRMT_ST0) {
+      rmesa->vb.texcoordptr[0] = &rmesa->vb.vertex[rmesa->vb.vertex_size].f;
+      rmesa->vb.vertex_size += 2;
+      rmesa->vb.texcoordptr[0][0] = ctx->Current.Attrib[VERT_ATTRIB_TEX0][0];
+      rmesa->vb.texcoordptr[0][1] = ctx->Current.Attrib[VERT_ATTRIB_TEX0][1];   
+   } 
+
+   if (ind & RADEON_CP_VC_FRMT_ST1) {
+      rmesa->vb.texcoordptr[1] = &rmesa->vb.vertex[rmesa->vb.vertex_size].f;
+      rmesa->vb.vertex_size += 2;
+      rmesa->vb.texcoordptr[1][0] = ctx->Current.Attrib[VERT_ATTRIB_TEX1][0];
+      rmesa->vb.texcoordptr[1][1] = ctx->Current.Attrib[VERT_ATTRIB_TEX1][1];
+   } 
+
+   if (rmesa->vb.installed_vertex_format != rmesa->vb.vertex_format) {
+      if (RADEON_DEBUG & DEBUG_VFMT)
+	 fprintf(stderr, "reinstall on vertex_format change\n");
+      _mesa_install_exec_vtxfmt( ctx, &rmesa->vb.vtxfmt );
+      rmesa->vb.installed_vertex_format = rmesa->vb.vertex_format;
+   }
+
+   if (RADEON_DEBUG & DEBUG_VFMT)
+      fprintf(stderr, "%s -- success\n", __FUNCTION__);
+   
+   return GL_TRUE;
+}
+
+void radeonVtxfmtInvalidate( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+
+   rmesa->vb.recheck = GL_TRUE;
+   rmesa->vb.fell_back = GL_FALSE;
+}
+
+
+static void radeonNewList( GLcontext *ctx, GLuint list, GLenum mode )
+{
+   VFMT_FALLBACK_OUTSIDE_BEGIN_END( __FUNCTION__ );
+}
+
+
+static void radeonVtxfmtValidate( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+
+   if (RADEON_DEBUG & DEBUG_VFMT)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   if (ctx->Driver.NeedFlush)
+      ctx->Driver.FlushVertices( ctx, ctx->Driver.NeedFlush );
+
+   rmesa->vb.recheck = GL_FALSE;
+
+   if (check_vtx_fmt( ctx )) {
+      if (!rmesa->vb.installed) {
+	 if (RADEON_DEBUG & DEBUG_VFMT)
+	    fprintf(stderr, "reinstall (new install)\n");
+
+	 _mesa_install_exec_vtxfmt( ctx, &rmesa->vb.vtxfmt );
+	 ctx->Driver.FlushVertices = radeonVtxfmtFlushVertices;
+	 ctx->Driver.NewList = radeonNewList;
+	 rmesa->vb.installed = GL_TRUE;
+      }
+      else if (RADEON_DEBUG & DEBUG_VFMT)
+	 fprintf(stderr, "%s: already installed", __FUNCTION__);
+   } 
+   else {
+      if (RADEON_DEBUG & DEBUG_VFMT)
+	 fprintf(stderr, "%s: failed\n", __FUNCTION__);
+
+      if (rmesa->vb.installed) {
+	 if (rmesa->dma.flush)
+	    rmesa->dma.flush( rmesa );
+	 _tnl_wakeup_exec( ctx );
+	 ctx->Driver.FlushVertices = radeonFlushVertices;
+	 rmesa->vb.installed = GL_FALSE;
+      }
+   }      
+}
+
+
+
+/* Materials:
+ */
+static void radeon_Materialfv( GLenum face, GLenum pname, 
+			       const GLfloat *params )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+
+   if (RADEON_DEBUG & DEBUG_VFMT)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   if (rmesa->vb.prim[0] != GL_POLYGON+1) {
+      VFMT_FALLBACK( __FUNCTION__ );
+      glMaterialfv( face, pname, params );
+      return;
+   }
+   _mesa_noop_Materialfv( face, pname, params );
+   radeonUpdateMaterial( ctx );
+}
+
+
+/* Begin/End
+ */
+static void radeon_Begin( GLenum mode )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if (RADEON_DEBUG & DEBUG_VFMT)
+      fprintf(stderr, "%s( %s )\n", __FUNCTION__,
+	      _mesa_lookup_enum_by_nr( mode ));
+
+   if (mode > GL_POLYGON) {
+      _mesa_error( ctx, GL_INVALID_ENUM, "glBegin" );
+      return;
+   }
+
+   if (rmesa->vb.prim[0] != GL_POLYGON+1) {
+      _mesa_error( ctx, GL_INVALID_OPERATION, "glBegin" );
+      return;
+   }
+   
+   if (ctx->NewState) 
+      _mesa_update_state( ctx );
+
+   if (rmesa->NewGLState)
+      radeonValidateState( ctx );
+
+   if (rmesa->vb.recheck) 
+      radeonVtxfmtValidate( ctx );
+
+   if (!rmesa->vb.installed) {
+      glBegin( mode );
+      return;
+   }
+
+
+   if (rmesa->dma.flush && rmesa->vb.counter < 12) {
+      if (RADEON_DEBUG & DEBUG_VFMT)
+	 fprintf(stderr, "%s: flush almost-empty buffers\n", __FUNCTION__);
+      flush_prims( rmesa );
+   }
+
+   /* Need to arrange to save vertices here?  Or always copy from dma (yuk)?
+    */
+   if (!rmesa->dma.flush) {
+      if (rmesa->dma.current.ptr + 12*rmesa->vb.vertex_size*4 > 
+	  rmesa->dma.current.end) {
+	 RADEON_NEWPRIM( rmesa );
+	 radeonRefillCurrentDmaRegion( rmesa );
+      }
+
+      rmesa->vb.dmaptr = (int *)(rmesa->dma.current.address + rmesa->dma.current.ptr);
+      rmesa->vb.counter = (rmesa->dma.current.end - rmesa->dma.current.ptr) / 
+	 (rmesa->vb.vertex_size * 4);
+      rmesa->vb.counter--;
+      rmesa->vb.initial_counter = rmesa->vb.counter;
+      rmesa->vb.notify = wrap_buffer;
+      rmesa->dma.flush = flush_prims;
+      ctx->Driver.NeedFlush |= FLUSH_STORED_VERTICES;
+   }
+   
+   
+   rmesa->vb.prim[0] = mode;
+   start_prim( rmesa, mode | PRIM_BEGIN );
+}
+
+
+
+static void radeon_End( void )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if (RADEON_DEBUG & DEBUG_VFMT)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   if (rmesa->vb.prim[0] == GL_POLYGON+1) {
+      _mesa_error( ctx, GL_INVALID_OPERATION, "glEnd" );
+      return;
+   }
+	  
+   note_last_prim( rmesa, PRIM_END );
+   rmesa->vb.prim[0] = GL_POLYGON+1;
+}
+
+
+/* Fallback on difficult entrypoints:
+ */
+#define PRE_LOOPBACK( FUNC )			\
+do {						\
+   if (RADEON_DEBUG & DEBUG_VFMT) 		\
+      fprintf(stderr, "%s\n", __FUNCTION__);	\
+   VFMT_FALLBACK( __FUNCTION__ );		\
+} while (0)
+#define TAG(x) radeon_fallback_##x
+#include "vtxfmt_tmp.h"
+
+
+
+static GLboolean radeonNotifyBegin( GLcontext *ctx, GLenum p )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+   
+   if (RADEON_DEBUG & DEBUG_VFMT)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   assert(!rmesa->vb.installed);
+
+   if (ctx->NewState) 
+      _mesa_update_state( ctx );
+
+   if (rmesa->NewGLState)
+      radeonValidateState( ctx );
+
+   if (ctx->Driver.NeedFlush)
+      ctx->Driver.FlushVertices( ctx, ctx->Driver.NeedFlush );
+
+   if (rmesa->vb.recheck) 
+      radeonVtxfmtValidate( ctx );
+
+   if (!rmesa->vb.installed) {
+      if (RADEON_DEBUG & DEBUG_VFMT)
+	 fprintf(stderr, "%s -- failed\n", __FUNCTION__);
+      return GL_FALSE;
+   }
+
+   radeon_Begin( p );
+   return GL_TRUE;
+}
+
+static void radeonVtxfmtFlushVertices( GLcontext *ctx, GLuint flags )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+
+   if (RADEON_DEBUG & DEBUG_VFMT)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
+   assert(rmesa->vb.installed);
+
+   if (flags & FLUSH_UPDATE_CURRENT) {
+      radeon_copy_to_current( ctx );
+      if (RADEON_DEBUG & DEBUG_VFMT)
+	 fprintf(stderr, "reinstall on update_current\n");
+      _mesa_install_exec_vtxfmt( ctx, &rmesa->vb.vtxfmt );
+      ctx->Driver.NeedFlush &= ~FLUSH_UPDATE_CURRENT;
+   }
+
+   if (flags & FLUSH_STORED_VERTICES) {
+      radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+      assert (rmesa->dma.flush == 0 ||
+	      rmesa->dma.flush == flush_prims);
+      if (rmesa->dma.flush == flush_prims)
+	 flush_prims( RADEON_CONTEXT( ctx ) );
+      ctx->Driver.NeedFlush &= ~FLUSH_STORED_VERTICES;
+   }
+}
+
+
+
+/* At this point, don't expect very many versions of each function to
+ * be generated, so not concerned about freeing them?
+ */
+
+
+void radeonVtxfmtInit( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+   GLvertexformat *vfmt = &(rmesa->vb.vtxfmt);
+
+   MEMSET( vfmt, 0, sizeof(GLvertexformat) );
+
+   /* Hook in chooser functions for codegen, etc:
+    */
+   radeonVtxfmtInitChoosers( vfmt );
+
+   /* Handled fully in supported states, but no codegen:
+    */
+   vfmt->Materialfv = radeon_Materialfv;
+   vfmt->ArrayElement = _ae_loopback_array_elt;	        /* generic helper */
+   vfmt->Rectf = _mesa_noop_Rectf;			/* generic helper */
+   vfmt->Begin = radeon_Begin;
+   vfmt->End = radeon_End;
+
+   /* Fallback for performance reasons:  (Fix with cva/elt path here and
+    * dmatmp2.h style primitive-merging)
+    *
+    * These should call NotifyBegin(), as should _tnl_EvalMesh, to allow
+    * a driver-hook.
+    */
+   vfmt->DrawArrays = radeon_fallback_DrawArrays;
+   vfmt->DrawElements = radeon_fallback_DrawElements;
+   vfmt->DrawRangeElements = radeon_fallback_DrawRangeElements; 
+
+
+   /* Not active in supported states; just keep ctx->Current uptodate:
+    */
+   vfmt->FogCoordfvEXT = _mesa_noop_FogCoordfvEXT;
+   vfmt->FogCoordfEXT = _mesa_noop_FogCoordfEXT;
+   vfmt->EdgeFlag = _mesa_noop_EdgeFlag;
+   vfmt->EdgeFlagv = _mesa_noop_EdgeFlagv;
+   vfmt->Indexi = _mesa_noop_Indexi;
+   vfmt->Indexiv = _mesa_noop_Indexiv;
+
+
+   /* Active but unsupported -- fallback if we receive these:
+    */
+   vfmt->CallList = radeon_fallback_CallList;
+   vfmt->EvalCoord1f = radeon_fallback_EvalCoord1f;
+   vfmt->EvalCoord1fv = radeon_fallback_EvalCoord1fv;
+   vfmt->EvalCoord2f = radeon_fallback_EvalCoord2f;
+   vfmt->EvalCoord2fv = radeon_fallback_EvalCoord2fv;
+   vfmt->EvalMesh1 = radeon_fallback_EvalMesh1;
+   vfmt->EvalMesh2 = radeon_fallback_EvalMesh2;
+   vfmt->EvalPoint1 = radeon_fallback_EvalPoint1;
+   vfmt->EvalPoint2 = radeon_fallback_EvalPoint2;
+   vfmt->TexCoord3f = radeon_fallback_TexCoord3f;
+   vfmt->TexCoord3fv = radeon_fallback_TexCoord3fv;
+   vfmt->TexCoord4f = radeon_fallback_TexCoord4f;
+   vfmt->TexCoord4fv = radeon_fallback_TexCoord4fv;
+   vfmt->MultiTexCoord3fARB = radeon_fallback_MultiTexCoord3fARB;
+   vfmt->MultiTexCoord3fvARB = radeon_fallback_MultiTexCoord3fvARB;
+   vfmt->MultiTexCoord4fARB = radeon_fallback_MultiTexCoord4fARB;
+   vfmt->MultiTexCoord4fvARB = radeon_fallback_MultiTexCoord4fvARB;
+   vfmt->Vertex4f = radeon_fallback_Vertex4f;
+   vfmt->Vertex4fv = radeon_fallback_Vertex4fv;
+
+   (void)radeon_fallback_vtxfmt;
+
+   TNL_CONTEXT(ctx)->Driver.NotifyBegin = radeonNotifyBegin;
+
+   rmesa->vb.enabled = 1;
+   rmesa->vb.prim = &ctx->Driver.CurrentExecPrimitive;
+   rmesa->vb.primflags = 0;
+
+   make_empty_list( &rmesa->vb.dfn_cache.Vertex2f );
+   make_empty_list( &rmesa->vb.dfn_cache.Vertex2fv );
+   make_empty_list( &rmesa->vb.dfn_cache.Vertex3f );
+   make_empty_list( &rmesa->vb.dfn_cache.Vertex3fv );
+   make_empty_list( &rmesa->vb.dfn_cache.Color4ub );
+   make_empty_list( &rmesa->vb.dfn_cache.Color4ubv );
+   make_empty_list( &rmesa->vb.dfn_cache.Color3ub );
+   make_empty_list( &rmesa->vb.dfn_cache.Color3ubv );
+   make_empty_list( &rmesa->vb.dfn_cache.Color4f );
+   make_empty_list( &rmesa->vb.dfn_cache.Color4fv );
+   make_empty_list( &rmesa->vb.dfn_cache.Color3f );
+   make_empty_list( &rmesa->vb.dfn_cache.Color3fv );
+   make_empty_list( &rmesa->vb.dfn_cache.SecondaryColor3fEXT );
+   make_empty_list( &rmesa->vb.dfn_cache.SecondaryColor3fvEXT );
+   make_empty_list( &rmesa->vb.dfn_cache.SecondaryColor3ubEXT );
+   make_empty_list( &rmesa->vb.dfn_cache.SecondaryColor3ubvEXT );
+   make_empty_list( &rmesa->vb.dfn_cache.Normal3f );
+   make_empty_list( &rmesa->vb.dfn_cache.Normal3fv );
+   make_empty_list( &rmesa->vb.dfn_cache.TexCoord2f );
+   make_empty_list( &rmesa->vb.dfn_cache.TexCoord2fv );
+   make_empty_list( &rmesa->vb.dfn_cache.TexCoord1f );
+   make_empty_list( &rmesa->vb.dfn_cache.TexCoord1fv );
+   make_empty_list( &rmesa->vb.dfn_cache.MultiTexCoord2fARB );
+   make_empty_list( &rmesa->vb.dfn_cache.MultiTexCoord2fvARB );
+   make_empty_list( &rmesa->vb.dfn_cache.MultiTexCoord1fARB );
+   make_empty_list( &rmesa->vb.dfn_cache.MultiTexCoord1fvARB );
+
+   radeonInitCodegen( &rmesa->vb.codegen );
+}
+
+static void free_funcs( struct dynfn *l )
+{
+   struct dynfn *f, *tmp;
+   foreach_s (f, tmp, l) {
+      remove_from_list( f );
+      ALIGN_FREE( f->code );
+      FREE( f );
+   }
+}
+
+
+
+void radeonVtxfmtMakeCurrent( GLcontext *ctx )
+{
+}
+
+
+void radeonVtxfmtDestroy( GLcontext *ctx )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT( ctx );
+
+   count_funcs( rmesa );
+   free_funcs( &rmesa->vb.dfn_cache.Vertex2f );
+   free_funcs( &rmesa->vb.dfn_cache.Vertex2fv );
+   free_funcs( &rmesa->vb.dfn_cache.Vertex3f );
+   free_funcs( &rmesa->vb.dfn_cache.Vertex3fv );
+   free_funcs( &rmesa->vb.dfn_cache.Color4ub );
+   free_funcs( &rmesa->vb.dfn_cache.Color4ubv );
+   free_funcs( &rmesa->vb.dfn_cache.Color3ub );
+   free_funcs( &rmesa->vb.dfn_cache.Color3ubv );
+   free_funcs( &rmesa->vb.dfn_cache.Color4f );
+   free_funcs( &rmesa->vb.dfn_cache.Color4fv );
+   free_funcs( &rmesa->vb.dfn_cache.Color3f );
+   free_funcs( &rmesa->vb.dfn_cache.Color3fv );
+   free_funcs( &rmesa->vb.dfn_cache.SecondaryColor3ubEXT );
+   free_funcs( &rmesa->vb.dfn_cache.SecondaryColor3ubvEXT );
+   free_funcs( &rmesa->vb.dfn_cache.SecondaryColor3fEXT );
+   free_funcs( &rmesa->vb.dfn_cache.SecondaryColor3fvEXT );
+   free_funcs( &rmesa->vb.dfn_cache.Normal3f );
+   free_funcs( &rmesa->vb.dfn_cache.Normal3fv );
+   free_funcs( &rmesa->vb.dfn_cache.TexCoord2f );
+   free_funcs( &rmesa->vb.dfn_cache.TexCoord2fv );
+   free_funcs( &rmesa->vb.dfn_cache.TexCoord1f );
+   free_funcs( &rmesa->vb.dfn_cache.TexCoord1fv );
+   free_funcs( &rmesa->vb.dfn_cache.MultiTexCoord2fARB );
+   free_funcs( &rmesa->vb.dfn_cache.MultiTexCoord2fvARB );
+   free_funcs( &rmesa->vb.dfn_cache.MultiTexCoord1fARB );
+   free_funcs( &rmesa->vb.dfn_cache.MultiTexCoord1fvARB );
+}
+
diff --git a/src/mesa/drivers/dri/radeon/radeon_vtxfmt.h b/src/mesa/drivers/dri/radeon/radeon_vtxfmt.h
new file mode 100644
index 00000000000..9792fcbb788
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_vtxfmt.h
@@ -0,0 +1,124 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/radeon/radeon_vtxfmt.h,v 1.3 2002/12/21 17:02:16 dawes Exp $ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     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, 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 <[email protected]>
+ */
+
+#ifndef __RADEON_VTXFMT_H__
+#define __RADEON_VTXFMT_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+#include "radeon_context.h"
+
+
+extern void radeonVtxfmtUpdate( GLcontext *ctx );
+extern void radeonVtxfmtInit( GLcontext *ctx );
+extern void radeonVtxfmtInvalidate( GLcontext *ctx );
+extern void radeonVtxfmtDestroy( GLcontext *ctx );
+extern void radeonVtxfmtInitChoosers( GLvertexformat *vfmt );
+
+extern void radeonVtxfmtMakeCurrent( GLcontext *ctx );
+extern void radeonVtxfmtUnbindContext( GLcontext *ctx );
+
+extern void radeon_copy_to_current( GLcontext *ctx );
+
+#define DFN( FUNC, CACHE)				\
+do {							\
+   char *start = (char *)&FUNC;				\
+   char *end = (char *)&FUNC##_end;			\
+   insert_at_head( &CACHE, dfn );			\
+   dfn->key = key;					\
+   dfn->code = ALIGN_MALLOC( end - start, 16 );		\
+   memcpy (dfn->code, start, end - start);		\
+}							\
+while ( 0 )
+
+#define FIXUP( CODE, OFFSET, CHECKVAL, NEWVAL )	\
+do {						\
+   int *icode = (int *)(CODE+OFFSET);		\
+   assert (*icode == CHECKVAL);			\
+   *icode = (int)NEWVAL;			\
+} while (0)
+
+
+/* Useful for figuring out the offsets:
+ */
+#define FIXUP2( CODE, OFFSET, CHECKVAL, NEWVAL )		\
+do {								\
+   while (*(int *)(CODE+OFFSET) != CHECKVAL) OFFSET++;		\
+   fprintf(stderr, "%s/%d CVAL %x OFFSET %d VAL %x\n", __FUNCTION__,	\
+	   __LINE__, CHECKVAL, OFFSET, (int)(NEWVAL));			\
+   *(int *)(CODE+OFFSET) = (int)(NEWVAL);				\
+   OFFSET += 4;							\
+} while (0)
+
+/* 
+ */
+void radeonInitCodegen( struct dfn_generators *gen );
+void radeonInitX86Codegen( struct dfn_generators *gen );
+void radeonInitSSECodegen( struct dfn_generators *gen );
+
+
+
+/* Defined in radeon_vtxfmt_x86.c
+ */
+struct dynfn *radeon_makeX86Vertex2f( GLcontext *, int );
+struct dynfn *radeon_makeX86Vertex2fv( GLcontext *, int );
+struct dynfn *radeon_makeX86Vertex3f( GLcontext *, int );
+struct dynfn *radeon_makeX86Vertex3fv( GLcontext *, int );
+struct dynfn *radeon_makeX86Color4ub( GLcontext *, int );
+struct dynfn *radeon_makeX86Color4ubv( GLcontext *, int );
+struct dynfn *radeon_makeX86Color3ub( GLcontext *, int );
+struct dynfn *radeon_makeX86Color3ubv( GLcontext *, int );
+struct dynfn *radeon_makeX86Color4f( GLcontext *, int );
+struct dynfn *radeon_makeX86Color4fv( GLcontext *, int );
+struct dynfn *radeon_makeX86Color3f( GLcontext *, int );
+struct dynfn *radeon_makeX86Color3fv( GLcontext *, int );
+struct dynfn *radeon_makeX86SecondaryColor3ubEXT( GLcontext *, int );
+struct dynfn *radeon_makeX86SecondaryColor3ubvEXT( GLcontext *, int );
+struct dynfn *radeon_makeX86SecondaryColor3fEXT( GLcontext *, int );
+struct dynfn *radeon_makeX86SecondaryColor3fvEXT( GLcontext *, int );
+struct dynfn *radeon_makeX86Normal3f( GLcontext *, int );
+struct dynfn *radeon_makeX86Normal3fv( GLcontext *, int );
+struct dynfn *radeon_makeX86TexCoord2f( GLcontext *, int );
+struct dynfn *radeon_makeX86TexCoord2fv( GLcontext *, int );
+struct dynfn *radeon_makeX86TexCoord1f( GLcontext *, int );
+struct dynfn *radeon_makeX86TexCoord1fv( GLcontext *, int );
+struct dynfn *radeon_makeX86MultiTexCoord2fARB( GLcontext *, int );
+struct dynfn *radeon_makeX86MultiTexCoord2fvARB( GLcontext *, int );
+struct dynfn *radeon_makeX86MultiTexCoord1fARB( GLcontext *, int );
+struct dynfn *radeon_makeX86MultiTexCoord1fvARB( GLcontext *, int );
+
+
+#endif
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_vtxfmt_c.c b/src/mesa/drivers/dri/radeon/radeon_vtxfmt_c.c
new file mode 100644
index 00000000000..188e34a4208
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_vtxfmt_c.c
@@ -0,0 +1,905 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 2002 ATI Technologies Inc., Ontario, Canada, and
+                     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, 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 <[email protected]>
+ */
+#include "glheader.h"
+#include "mtypes.h"
+#include "colormac.h"
+#include "simple_list.h"
+#include "api_noop.h"
+#include "vtxfmt.h"
+
+#include "radeon_vtxfmt.h"
+
+/* Fallback versions of all the entrypoints for situations where
+ * codegen isn't available.  This is still a lot faster than the
+ * vb/pipeline implementation in Mesa.
+ */
+static void radeon_Vertex3f( GLfloat x, GLfloat y, GLfloat z )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   int i;
+
+   *rmesa->vb.dmaptr++ = *(int *)&x;
+   *rmesa->vb.dmaptr++ = *(int *)&y;
+   *rmesa->vb.dmaptr++ = *(int *)&z;
+
+   for (i = 3; i < rmesa->vb.vertex_size; i++)
+      *rmesa->vb.dmaptr++ = rmesa->vb.vertex[i].i;
+   
+   if (--rmesa->vb.counter == 0)
+      rmesa->vb.notify();
+}
+
+
+static void radeon_Vertex3fv( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   int i;
+
+   *rmesa->vb.dmaptr++ = *(int *)&v[0];
+   *rmesa->vb.dmaptr++ = *(int *)&v[1];
+   *rmesa->vb.dmaptr++ = *(int *)&v[2];
+
+   for (i = 3; i < rmesa->vb.vertex_size; i++)
+      *rmesa->vb.dmaptr++ = rmesa->vb.vertex[i].i;
+   
+   if (--rmesa->vb.counter == 0)
+      rmesa->vb.notify();
+}
+
+
+static void radeon_Vertex2f( GLfloat x, GLfloat y )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   int i;
+
+   *rmesa->vb.dmaptr++ = *(int *)&x;
+   *rmesa->vb.dmaptr++ = *(int *)&y;
+   *rmesa->vb.dmaptr++ = 0;
+
+   for (i = 3; i < rmesa->vb.vertex_size; i++)
+      *rmesa->vb.dmaptr++ = *(int *)&rmesa->vb.vertex[i];
+   
+   if (--rmesa->vb.counter == 0)
+      rmesa->vb.notify();
+}
+
+
+static void radeon_Vertex2fv( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   int i;
+
+   *rmesa->vb.dmaptr++ = *(int *)&v[0];
+   *rmesa->vb.dmaptr++ = *(int *)&v[1];
+   *rmesa->vb.dmaptr++ = 0;
+
+   for (i = 3; i < rmesa->vb.vertex_size; i++)
+      *rmesa->vb.dmaptr++ = rmesa->vb.vertex[i].i;
+   
+   if (--rmesa->vb.counter == 0)
+      rmesa->vb.notify();
+}
+
+
+
+/* Color for ubyte (packed) color formats:
+ */
+static void radeon_Color3ub_ub( GLubyte r, GLubyte g, GLubyte b )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeon_color_t *dest = rmesa->vb.colorptr;
+   dest->red	= r;
+   dest->green	= g;
+   dest->blue	= b;
+   dest->alpha	= 0xff;
+}
+
+static void radeon_Color3ubv_ub( const GLubyte *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeon_color_t *dest = rmesa->vb.colorptr;
+   dest->red	= v[0];
+   dest->green	= v[1];
+   dest->blue	= v[2];
+   dest->alpha	= 0xff;
+}
+
+static void radeon_Color4ub_ub( GLubyte r, GLubyte g, GLubyte b, GLubyte a )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeon_color_t *dest = rmesa->vb.colorptr;
+   dest->red	= r;
+   dest->green	= g;
+   dest->blue	= b;
+   dest->alpha	= a;
+}
+
+static void radeon_Color4ubv_ub( const GLubyte *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   *(GLuint *)rmesa->vb.colorptr = LE32_TO_CPU(*(GLuint *)v);
+}
+
+
+static void radeon_Color3f_ub( GLfloat r, GLfloat g, GLfloat b )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeon_color_t *dest = rmesa->vb.colorptr;
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->red,   r );
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->green, g );
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->blue,  b );
+   dest->alpha = 255;
+}
+
+static void radeon_Color3fv_ub( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeon_color_t *dest = rmesa->vb.colorptr;
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->red,   v[0] );
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->green, v[1] );
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->blue,  v[2] );
+   dest->alpha = 255;
+}
+
+static void radeon_Color4f_ub( GLfloat r, GLfloat g, GLfloat b, GLfloat a )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeon_color_t *dest = rmesa->vb.colorptr;
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->red,   r );
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->green, g );
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->blue,  b );
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->alpha, a );
+}
+
+static void radeon_Color4fv_ub( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeon_color_t *dest = rmesa->vb.colorptr;
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->red,	  v[0] );
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->green, v[1] );
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->blue,  v[2] );
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->alpha, v[3] );
+}
+
+
+/* Color for float color+alpha formats:
+ */
+static void radeon_Color3ub_4f( GLubyte r, GLubyte g, GLubyte b )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatcolorptr;
+   dest[0] = UBYTE_TO_FLOAT(r);
+   dest[1] = UBYTE_TO_FLOAT(g);
+   dest[2] = UBYTE_TO_FLOAT(b);
+   dest[3] = 1.0;
+}
+
+static void radeon_Color3ubv_4f( const GLubyte *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatcolorptr;
+   dest[0] = UBYTE_TO_FLOAT(v[0]);
+   dest[1] = UBYTE_TO_FLOAT(v[1]);
+   dest[2] = UBYTE_TO_FLOAT(v[2]);
+   dest[3] = 1.0;
+}
+
+static void radeon_Color4ub_4f( GLubyte r, GLubyte g, GLubyte b, GLubyte a )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatcolorptr;
+   dest[0] = UBYTE_TO_FLOAT(r);
+   dest[1] = UBYTE_TO_FLOAT(g);
+   dest[2] = UBYTE_TO_FLOAT(b);
+   dest[3] = UBYTE_TO_FLOAT(a);
+}
+
+static void radeon_Color4ubv_4f( const GLubyte *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatcolorptr;
+   dest[0] = UBYTE_TO_FLOAT(v[0]);
+   dest[1] = UBYTE_TO_FLOAT(v[1]);
+   dest[2] = UBYTE_TO_FLOAT(v[2]);
+   dest[3] = UBYTE_TO_FLOAT(v[3]);
+}
+
+
+static void radeon_Color3f_4f( GLfloat r, GLfloat g, GLfloat b )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatcolorptr;
+   dest[0] = r;
+   dest[1] = g;
+   dest[2] = b;
+   dest[3] = 1.0;		
+}
+
+static void radeon_Color3fv_4f( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatcolorptr;
+   dest[0] = v[0];
+   dest[1] = v[1];
+   dest[2] = v[2];
+   dest[3] = 1.0;
+}
+
+static void radeon_Color4f_4f( GLfloat r, GLfloat g, GLfloat b, GLfloat a )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatcolorptr;
+   dest[0] = r;
+   dest[1] = g;
+   dest[2] = b;
+   dest[3] = a;
+}
+
+static void radeon_Color4fv_4f( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatcolorptr;
+   dest[0] = v[0];
+   dest[1] = v[1];
+   dest[2] = v[2];
+   dest[3] = v[3];
+}
+
+
+/* Color for float color formats:
+ */
+static void radeon_Color3ub_3f( GLubyte r, GLubyte g, GLubyte b )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatcolorptr;
+   dest[0] = UBYTE_TO_FLOAT(r);
+   dest[1] = UBYTE_TO_FLOAT(g);
+   dest[2] = UBYTE_TO_FLOAT(b);
+}
+
+static void radeon_Color3ubv_3f( const GLubyte *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatcolorptr;
+   dest[0] = UBYTE_TO_FLOAT(v[0]);
+   dest[1] = UBYTE_TO_FLOAT(v[1]);
+   dest[2] = UBYTE_TO_FLOAT(v[2]);
+}
+
+static void radeon_Color4ub_3f( GLubyte r, GLubyte g, GLubyte b, GLubyte a )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatcolorptr;
+   dest[0] = UBYTE_TO_FLOAT(r);
+   dest[1] = UBYTE_TO_FLOAT(g);
+   dest[2] = UBYTE_TO_FLOAT(b);
+   ctx->Current.Attrib[VERT_ATTRIB_COLOR0][3] = UBYTE_TO_FLOAT(a);
+}
+
+static void radeon_Color4ubv_3f( const GLubyte *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatcolorptr;
+   dest[0] = UBYTE_TO_FLOAT(v[0]);
+   dest[1] = UBYTE_TO_FLOAT(v[1]);
+   dest[2] = UBYTE_TO_FLOAT(v[2]);
+   ctx->Current.Attrib[VERT_ATTRIB_COLOR0][3] = UBYTE_TO_FLOAT(v[3]);
+}
+
+
+static void radeon_Color3f_3f( GLfloat r, GLfloat g, GLfloat b )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatcolorptr;
+   dest[0] = r;
+   dest[1] = g;
+   dest[2] = b;
+}
+
+static void radeon_Color3fv_3f( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatcolorptr;
+   dest[0] = v[0];
+   dest[1] = v[1];
+   dest[2] = v[2];
+}
+
+static void radeon_Color4f_3f( GLfloat r, GLfloat g, GLfloat b, GLfloat a )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatcolorptr;
+   dest[0] = r;
+   dest[1] = g;
+   dest[2] = b;
+   ctx->Current.Attrib[VERT_ATTRIB_COLOR0][3] = a;
+}
+
+static void radeon_Color4fv_3f( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatcolorptr;
+   dest[0] = v[0];
+   dest[1] = v[1];
+   dest[2] = v[2];
+   ctx->Current.Attrib[VERT_ATTRIB_COLOR0][3] = v[3]; 
+}
+
+
+/* Secondary Color:
+ */
+static void radeon_SecondaryColor3ubEXT_ub( GLubyte r, GLubyte g, GLubyte b )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeon_color_t *dest = rmesa->vb.specptr;
+   dest->red	= r;
+   dest->green	= g;
+   dest->blue	= b;
+   dest->alpha	= 0xff;
+}
+
+static void radeon_SecondaryColor3ubvEXT_ub( const GLubyte *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeon_color_t *dest = rmesa->vb.specptr;
+   dest->red	= v[0];
+   dest->green	= v[1];
+   dest->blue	= v[2];
+   dest->alpha	= 0xff;
+}
+
+static void radeon_SecondaryColor3fEXT_ub( GLfloat r, GLfloat g, GLfloat b )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeon_color_t *dest = rmesa->vb.specptr;
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->red,	  r );
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->green, g );
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->blue,  b );
+   dest->alpha = 255;
+}
+
+static void radeon_SecondaryColor3fvEXT_ub( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   radeon_color_t *dest = rmesa->vb.specptr;
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->red,	  v[0] );
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->green, v[1] );
+   UNCLAMPED_FLOAT_TO_UBYTE( dest->blue,  v[2] );
+   dest->alpha = 255;
+}
+
+static void radeon_SecondaryColor3ubEXT_3f( GLubyte r, GLubyte g, GLubyte b )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatspecptr;
+   dest[0] = UBYTE_TO_FLOAT(r);
+   dest[1] = UBYTE_TO_FLOAT(g);
+   dest[2] = UBYTE_TO_FLOAT(b);
+   dest[3] = 1.0;
+}
+
+static void radeon_SecondaryColor3ubvEXT_3f( const GLubyte *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatspecptr;
+   dest[0] = UBYTE_TO_FLOAT(v[0]);
+   dest[1] = UBYTE_TO_FLOAT(v[1]);
+   dest[2] = UBYTE_TO_FLOAT(v[2]);
+   dest[3] = 1.0;
+}
+
+static void radeon_SecondaryColor3fEXT_3f( GLfloat r, GLfloat g, GLfloat b )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatspecptr;
+   dest[0] = r;
+   dest[1] = g;
+   dest[2] = b;
+   dest[3] = 1.0;
+}
+
+static void radeon_SecondaryColor3fvEXT_3f( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.floatspecptr;
+   dest[0] = v[0];
+   dest[1] = v[1];
+   dest[2] = v[2];
+   dest[3] = 1.0;
+}
+
+
+/* Normal
+ */
+static void radeon_Normal3f( GLfloat n0, GLfloat n1, GLfloat n2 )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.normalptr;
+   dest[0] = n0;
+   dest[1] = n1;
+   dest[2] = n2;
+}
+
+static void radeon_Normal3fv( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.normalptr;
+   dest[0] = v[0];
+   dest[1] = v[1];
+   dest[2] = v[2];
+}
+
+
+/* TexCoord
+ */
+static void radeon_TexCoord1f( GLfloat s )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.texcoordptr[0];
+   dest[0] = s;
+   dest[1] = 0;
+}
+
+static void radeon_TexCoord1fv( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.texcoordptr[0];
+   dest[0] = v[0];
+   dest[1] = 0;
+}
+
+static void radeon_TexCoord2f( GLfloat s, GLfloat t )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.texcoordptr[0];
+   dest[0] = s;
+   dest[1] = t;
+}
+
+static void radeon_TexCoord2fv( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.texcoordptr[0];
+   dest[0] = v[0];
+   dest[1] = v[1];
+}
+
+
+/* MultiTexcoord
+ * 
+ * Technically speaking, these functions should subtract GL_TEXTURE0 from
+ * \c target before masking and using it.  The value of GL_TEXTURE0 is 0x84C0,
+ * which has the low-order 5 bits 0.  For all possible valid values of 
+ * \c target.  Subtracting GL_TEXTURE0 has the net effect of masking \c target
+ * with 0x1F.  Masking with 0x1F and then masking with 0x01 is redundant, so
+ * the subtraction has been omitted.
+ */
+
+static void radeon_MultiTexCoord1fARB( GLenum target, GLfloat s  )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.texcoordptr[target & 1];
+   dest[0] = s;
+   dest[1] = 0;
+}
+
+static void radeon_MultiTexCoord1fvARB( GLenum target, const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.texcoordptr[target & 1];
+   dest[0] = v[0];
+   dest[1] = 0;
+}
+
+static void radeon_MultiTexCoord2fARB( GLenum target, GLfloat s, GLfloat t )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.texcoordptr[target & 1];
+   dest[0] = s;
+   dest[1] = t;
+}
+
+static void radeon_MultiTexCoord2fvARB( GLenum target, const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   GLfloat *dest = rmesa->vb.texcoordptr[target & 1];
+   dest[0] = v[0];
+   dest[1] = v[1];
+}
+
+static struct dynfn *lookup( struct dynfn *l, int key )
+{
+   struct dynfn *f;
+
+   foreach( f, l ) {
+      if (f->key == key) 
+	 return f;
+   }
+
+   return 0;
+}
+
+/* Can't use the loopback template for this:
+ */
+
+#define CHOOSE(FN, FNTYPE, MASK, ACTIVE, ARGS1, ARGS2 )			\
+static void choose_##FN ARGS1						\
+{									\
+   GET_CURRENT_CONTEXT(ctx);						\
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);			\
+   int key = rmesa->vb.vertex_format & (MASK|ACTIVE);			\
+   struct dynfn *dfn;							\
+									\
+   dfn = lookup( &rmesa->vb.dfn_cache.FN, key );			\
+   if (dfn == 0)							\
+      dfn = rmesa->vb.codegen.FN( ctx, key );				\
+   else if (RADEON_DEBUG & DEBUG_CODEGEN)				\
+      fprintf(stderr, "%s -- cached codegen\n", __FUNCTION__ );		\
+									\
+   if (dfn)								\
+      ctx->Exec->FN = (FNTYPE)(dfn->code);				\
+   else {								\
+      if (RADEON_DEBUG & DEBUG_CODEGEN)					\
+	 fprintf(stderr, "%s -- generic version\n", __FUNCTION__ );	\
+      ctx->Exec->FN = radeon_##FN;					\
+   }									\
+									\
+   ctx->Driver.NeedFlush |= FLUSH_UPDATE_CURRENT;			\
+   ctx->Exec->FN ARGS2;							\
+}
+
+
+
+/* For the _3f case, only allow one color function to be hooked in at
+ * a time.  Eventually, use a similar mechanism to allow selecting the
+ * color component of the vertex format based on client behaviour.  
+ *
+ * Note:  Perform these actions even if there is a codegen or cached 
+ * codegen version of the chosen function.
+ */
+#define CHOOSE_COLOR(FN, FNTYPE, NR, MASK, ACTIVE, ARGS1, ARGS2 )	\
+static void choose_##FN ARGS1						\
+{									\
+   GET_CURRENT_CONTEXT(ctx); \
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);			\
+   int key = rmesa->vb.vertex_format & (MASK|ACTIVE);			\
+   struct dynfn *dfn;							\
+									\
+   if (rmesa->vb.vertex_format & ACTIVE_PKCOLOR) {			\
+      ctx->Exec->FN = radeon_##FN##_ub;					\
+   }									\
+   else if ((rmesa->vb.vertex_format &					\
+            (ACTIVE_FPCOLOR|ACTIVE_FPALPHA)) == ACTIVE_FPCOLOR) {	\
+									\
+      if (rmesa->vb.installed_color_3f_sz != NR) {			\
+         rmesa->vb.installed_color_3f_sz = NR;				\
+         if (NR == 3) ctx->Current.Attrib[VERT_ATTRIB_COLOR0][3] = 1.0;	\
+         if (ctx->Driver.NeedFlush & FLUSH_UPDATE_CURRENT) {		\
+            radeon_copy_to_current( ctx );				\
+            _mesa_install_exec_vtxfmt( ctx, &rmesa->vb.vtxfmt );	\
+            ctx->Exec->FN ARGS2;					\
+            return;							\
+         }								\
+      }									\
+									\
+      ctx->Exec->FN = radeon_##FN##_3f;					\
+   }									\
+   else {								\
+      ctx->Exec->FN = radeon_##FN##_4f;					\
+   }									\
+									\
+									\
+   dfn = lookup( &rmesa->vb.dfn_cache.FN, key );			\
+   if (!dfn) dfn = rmesa->vb.codegen.FN( ctx, key );			\
+									\
+   if (dfn) {								\
+      if (RADEON_DEBUG & DEBUG_CODEGEN)					\
+         fprintf(stderr, "%s -- codegen version\n", __FUNCTION__ );	\
+      ctx->Exec->FN = (FNTYPE)dfn->code;				\
+   }									\
+   else if (RADEON_DEBUG & DEBUG_CODEGEN)				\
+         fprintf(stderr, "%s -- 'c' version\n", __FUNCTION__ );		\
+									\
+   ctx->Driver.NeedFlush |= FLUSH_UPDATE_CURRENT;			\
+   ctx->Exec->FN ARGS2;							\
+}
+
+
+
+/* Right now there are both _ub and _3f versions of the secondary color
+ * functions.  Currently, we only set-up the hardware to use the _ub versions.
+ * The _3f versions are needed for the cases where secondary color isn't used
+ * in the vertex format, but it still needs to be stored in the context
+ * state vector.
+ */
+#define CHOOSE_SECONDARY_COLOR(FN, FNTYPE, MASK, ACTIVE, ARGS1, ARGS2 )	\
+static void choose_##FN ARGS1						\
+{									\
+   GET_CURRENT_CONTEXT(ctx);						\
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);			\
+   int key = rmesa->vb.vertex_format & (MASK|ACTIVE);			\
+   struct dynfn *dfn = lookup( &rmesa->vb.dfn_cache.FN, key );		\
+									\
+   if (dfn == 0)							\
+      dfn = rmesa->vb.codegen.FN( ctx, key );				\
+   else  if (RADEON_DEBUG & DEBUG_CODEGEN)				\
+      fprintf(stderr, "%s -- cached version\n", __FUNCTION__ );		\
+									\
+   if (dfn)								\
+      ctx->Exec->FN = (FNTYPE)(dfn->code);				\
+   else {								\
+      if (RADEON_DEBUG & DEBUG_CODEGEN)					\
+         fprintf(stderr, "%s -- generic version\n", __FUNCTION__ );	\
+      ctx->Exec->FN = ((rmesa->vb.vertex_format & ACTIVE_PKSPEC) != 0)	\
+	  ? radeon_##FN##_ub : radeon_##FN##_3f;			\
+   }									\
+									\
+   ctx->Driver.NeedFlush |= FLUSH_UPDATE_CURRENT;			\
+   ctx->Exec->FN ARGS2;							\
+}
+
+
+
+
+
+/* Shorthands
+ */
+#define ACTIVE_XYZW (RADEON_CP_VC_FRMT_W0|RADEON_CP_VC_FRMT_Z)
+#define ACTIVE_NORM RADEON_CP_VC_FRMT_N0
+
+#define ACTIVE_PKCOLOR RADEON_CP_VC_FRMT_PKCOLOR
+#define ACTIVE_FPCOLOR RADEON_CP_VC_FRMT_FPCOLOR
+#define ACTIVE_FPALPHA RADEON_CP_VC_FRMT_FPALPHA
+#define ACTIVE_COLOR (ACTIVE_FPCOLOR|ACTIVE_PKCOLOR)
+
+#define ACTIVE_PKSPEC RADEON_CP_VC_FRMT_PKSPEC
+#define ACTIVE_FPSPEC RADEON_CP_VC_FRMT_FPSPEC
+#define ACTIVE_SPEC   (ACTIVE_FPSPEC|ACTIVE_PKSPEC)
+
+#define ACTIVE_ST0 RADEON_CP_VC_FRMT_ST0
+#define ACTIVE_ST1 RADEON_CP_VC_FRMT_ST1
+#define ACTIVE_ST_ALL (RADEON_CP_VC_FRMT_ST1|RADEON_CP_VC_FRMT_ST0)
+
+/* Each codegen function should be able to be fully specified by a
+ * subsetted version of rmesa->vb.vertex_format.
+ */
+#define MASK_NORM    (ACTIVE_XYZW)
+#define MASK_COLOR   (MASK_NORM|ACTIVE_NORM)
+#define MASK_SPEC    (MASK_COLOR|ACTIVE_COLOR)
+#define MASK_ST0     (MASK_SPEC|ACTIVE_SPEC)
+#define MASK_ST1     (MASK_ST0|ACTIVE_ST0)
+#define MASK_ST_ALL  (MASK_ST1|ACTIVE_ST1)
+#define MASK_VERTEX  (MASK_ST_ALL|ACTIVE_FPALPHA) 
+
+
+typedef void (*p4f)( GLfloat, GLfloat, GLfloat, GLfloat );
+typedef void (*p3f)( GLfloat, GLfloat, GLfloat );
+typedef void (*p2f)( GLfloat, GLfloat );
+typedef void (*p1f)( GLfloat );
+typedef void (*pe2f)( GLenum, GLfloat, GLfloat );
+typedef void (*pe1f)( GLenum, GLfloat );
+typedef void (*p4ub)( GLubyte, GLubyte, GLubyte, GLubyte );
+typedef void (*p3ub)( GLubyte, GLubyte, GLubyte );
+typedef void (*pfv)( const GLfloat * );
+typedef void (*pefv)( GLenum, const GLfloat * );
+typedef void (*pubv)( const GLubyte * );
+
+
+CHOOSE(Normal3f, p3f, MASK_NORM, ACTIVE_NORM, 
+       (GLfloat a,GLfloat b,GLfloat c), (a,b,c))
+CHOOSE(Normal3fv, pfv, MASK_NORM, ACTIVE_NORM, 
+       (const GLfloat *v), (v))
+
+CHOOSE_COLOR(Color4ub, p4ub, 4, MASK_COLOR, ACTIVE_COLOR,
+	(GLubyte a,GLubyte b, GLubyte c, GLubyte d), (a,b,c,d))
+CHOOSE_COLOR(Color4ubv, pubv, 4, MASK_COLOR, ACTIVE_COLOR, 
+	(const GLubyte *v), (v))
+CHOOSE_COLOR(Color3ub, p3ub, 3, MASK_COLOR, ACTIVE_COLOR, 
+	(GLubyte a,GLubyte b, GLubyte c), (a,b,c))
+CHOOSE_COLOR(Color3ubv, pubv, 3, MASK_COLOR, ACTIVE_COLOR, 
+	(const GLubyte *v), (v))
+
+CHOOSE_COLOR(Color4f, p4f, 4, MASK_COLOR, ACTIVE_COLOR, 
+	(GLfloat a,GLfloat b, GLfloat c, GLfloat d), (a,b,c,d))
+CHOOSE_COLOR(Color4fv, pfv, 4, MASK_COLOR, ACTIVE_COLOR, 
+	(const GLfloat *v), (v))
+CHOOSE_COLOR(Color3f, p3f, 3, MASK_COLOR, ACTIVE_COLOR,
+	(GLfloat a,GLfloat b, GLfloat c), (a,b,c))
+CHOOSE_COLOR(Color3fv, pfv, 3, MASK_COLOR, ACTIVE_COLOR,
+	(const GLfloat *v), (v))
+
+
+CHOOSE_SECONDARY_COLOR(SecondaryColor3ubEXT, p3ub, MASK_SPEC, ACTIVE_SPEC,
+	(GLubyte a,GLubyte b, GLubyte c), (a,b,c))
+CHOOSE_SECONDARY_COLOR(SecondaryColor3ubvEXT, pubv, MASK_SPEC, ACTIVE_SPEC,
+	(const GLubyte *v), (v))
+CHOOSE_SECONDARY_COLOR(SecondaryColor3fEXT, p3f, MASK_SPEC, ACTIVE_SPEC,
+	(GLfloat a,GLfloat b, GLfloat c), (a,b,c))
+CHOOSE_SECONDARY_COLOR(SecondaryColor3fvEXT, pfv, MASK_SPEC, ACTIVE_SPEC,
+	(const GLfloat *v), (v))
+
+CHOOSE(TexCoord2f, p2f, MASK_ST0, ACTIVE_ST0, 
+       (GLfloat a,GLfloat b), (a,b))
+CHOOSE(TexCoord2fv, pfv, MASK_ST0, ACTIVE_ST0, 
+       (const GLfloat *v), (v))
+CHOOSE(TexCoord1f, p1f, MASK_ST0, ACTIVE_ST0, 
+       (GLfloat a), (a))
+CHOOSE(TexCoord1fv, pfv, MASK_ST0, ACTIVE_ST0, 
+       (const GLfloat *v), (v))
+
+CHOOSE(MultiTexCoord2fARB, pe2f, MASK_ST_ALL, ACTIVE_ST_ALL,
+	 (GLenum u,GLfloat a,GLfloat b), (u,a,b))
+CHOOSE(MultiTexCoord2fvARB, pefv, MASK_ST_ALL, ACTIVE_ST_ALL,
+	(GLenum u,const GLfloat *v), (u,v))
+CHOOSE(MultiTexCoord1fARB, pe1f, MASK_ST_ALL, ACTIVE_ST_ALL,
+	 (GLenum u,GLfloat a), (u,a))
+CHOOSE(MultiTexCoord1fvARB, pefv, MASK_ST_ALL, ACTIVE_ST_ALL,
+	(GLenum u,const GLfloat *v), (u,v))
+
+CHOOSE(Vertex3f, p3f, MASK_VERTEX, MASK_VERTEX, 
+       (GLfloat a,GLfloat b,GLfloat c), (a,b,c))
+CHOOSE(Vertex3fv, pfv, MASK_VERTEX, MASK_VERTEX, 
+       (const GLfloat *v), (v))
+CHOOSE(Vertex2f, p2f, MASK_VERTEX, MASK_VERTEX, 
+       (GLfloat a,GLfloat b), (a,b))
+CHOOSE(Vertex2fv, pfv, MASK_VERTEX, MASK_VERTEX, 
+       (const GLfloat *v), (v))
+
+
+
+
+
+void radeonVtxfmtInitChoosers( GLvertexformat *vfmt )
+{
+   vfmt->Color3f = choose_Color3f;
+   vfmt->Color3fv = choose_Color3fv;
+   vfmt->Color3ub = choose_Color3ub;
+   vfmt->Color3ubv = choose_Color3ubv;
+   vfmt->Color4f = choose_Color4f;
+   vfmt->Color4fv = choose_Color4fv;
+   vfmt->Color4ub = choose_Color4ub;
+   vfmt->Color4ubv = choose_Color4ubv;
+   vfmt->SecondaryColor3fEXT = choose_SecondaryColor3fEXT;
+   vfmt->SecondaryColor3fvEXT = choose_SecondaryColor3fvEXT;
+   vfmt->SecondaryColor3ubEXT = choose_SecondaryColor3ubEXT;
+   vfmt->SecondaryColor3ubvEXT = choose_SecondaryColor3ubvEXT;
+   vfmt->MultiTexCoord1fARB = choose_MultiTexCoord1fARB;
+   vfmt->MultiTexCoord1fvARB = choose_MultiTexCoord1fvARB;
+   vfmt->MultiTexCoord2fARB = choose_MultiTexCoord2fARB;
+   vfmt->MultiTexCoord2fvARB = choose_MultiTexCoord2fvARB;
+   vfmt->Normal3f = choose_Normal3f;
+   vfmt->Normal3fv = choose_Normal3fv;
+   vfmt->TexCoord1f = choose_TexCoord1f;
+   vfmt->TexCoord1fv = choose_TexCoord1fv;
+   vfmt->TexCoord2f = choose_TexCoord2f;
+   vfmt->TexCoord2fv = choose_TexCoord2fv;
+   vfmt->Vertex2f = choose_Vertex2f;
+   vfmt->Vertex2fv = choose_Vertex2fv;
+   vfmt->Vertex3f = choose_Vertex3f;
+   vfmt->Vertex3fv = choose_Vertex3fv;
+}
+
+
+static struct dynfn *codegen_noop( GLcontext *ctx, int key )
+{
+   (void) ctx; (void) key;
+   return 0;
+}
+
+void radeonInitCodegen( struct dfn_generators *gen )
+{
+   gen->Vertex3f = codegen_noop;
+   gen->Vertex3fv = codegen_noop;
+   gen->Color4ub = codegen_noop;
+   gen->Color4ubv = codegen_noop;
+   gen->Normal3f = codegen_noop;
+   gen->Normal3fv = codegen_noop;
+   gen->TexCoord2f = codegen_noop;
+   gen->TexCoord2fv = codegen_noop;
+   gen->MultiTexCoord2fARB = codegen_noop;
+   gen->MultiTexCoord2fvARB = codegen_noop;
+   gen->Vertex2f = codegen_noop;
+   gen->Vertex2fv = codegen_noop;
+   gen->Color3ub = codegen_noop;
+   gen->Color3ubv = codegen_noop;
+   gen->Color4f = codegen_noop;
+   gen->Color4fv = codegen_noop;
+   gen->Color3f = codegen_noop;
+   gen->Color3fv = codegen_noop;
+   gen->SecondaryColor3fEXT = codegen_noop;
+   gen->SecondaryColor3fvEXT = codegen_noop;
+   gen->SecondaryColor3ubEXT = codegen_noop;
+   gen->SecondaryColor3ubvEXT = codegen_noop;
+   gen->TexCoord1f = codegen_noop;
+   gen->TexCoord1fv = codegen_noop;
+   gen->MultiTexCoord1fARB = codegen_noop;
+   gen->MultiTexCoord1fvARB = codegen_noop;
+
+   if (!getenv("RADEON_NO_CODEGEN")) {
+#if defined(USE_X86_ASM)
+      radeonInitX86Codegen( gen );
+#endif
+
+#if defined(USE_SSE_ASM)
+      radeonInitSSECodegen( gen );
+#endif
+   }
+}
diff --git a/src/mesa/drivers/dri/radeon/radeon_vtxfmt_sse.c b/src/mesa/drivers/dri/radeon/radeon_vtxfmt_sse.c
new file mode 100644
index 00000000000..0f2c82bd878
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_vtxfmt_sse.c
@@ -0,0 +1,232 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     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, 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 <[email protected]>
+ */
+
+#include "glheader.h"
+#include "imports.h"
+#include "simple_list.h" 
+#include "radeon_vtxfmt.h"
+
+#if defined(USE_SSE_ASM)
+#include "X86/common_x86_asm.h"
+
+#define EXTERN( FUNC )		\
+extern const char *FUNC;	\
+extern const char *FUNC##_end
+
+EXTERN( _sse_Attribute2fv );
+EXTERN( _sse_Attribute2f );
+EXTERN( _sse_Attribute3fv );
+EXTERN( _sse_Attribute3f );
+EXTERN( _sse_MultiTexCoord2fv );
+EXTERN( _sse_MultiTexCoord2f );
+EXTERN( _sse_MultiTexCoord2fv_2 );
+EXTERN( _sse_MultiTexCoord2f_2 );
+
+/* Build specialized versions of the immediate calls on the fly for
+ * the current state.
+ */
+
+static struct dynfn *radeon_makeSSEAttribute2fv( struct dynfn * cache, int key,
+					       const char * name, void * dest)
+{
+   struct dynfn *dfn = MALLOC_STRUCT( dynfn );
+
+   if (RADEON_DEBUG & DEBUG_CODEGEN)
+      fprintf(stderr, "%s 0x%08x\n", name, key );
+
+   DFN ( _sse_Attribute2fv, (*cache) );
+   FIXUP(dfn->code, 10, 0x0, (int)dest);
+   return dfn;
+}
+
+static struct dynfn *radeon_makeSSEAttribute2f( struct dynfn * cache, int key,
+					      const char * name, void * dest )
+{
+   struct dynfn *dfn = MALLOC_STRUCT( dynfn );
+
+   if (RADEON_DEBUG & DEBUG_CODEGEN)
+      fprintf(stderr, "%s 0x%08x\n", name, key );
+
+   DFN ( _sse_Attribute2f, (*cache) );
+   FIXUP(dfn->code, 8, 0x0, (int)dest); 
+   return dfn;
+}
+
+static struct dynfn *radeon_makeSSEAttribute3fv( struct dynfn * cache, int key,
+					       const char * name, void * dest)
+{
+   struct dynfn *dfn = MALLOC_STRUCT( dynfn );
+
+   if (RADEON_DEBUG & DEBUG_CODEGEN)
+      fprintf(stderr, "%s 0x%08x\n", name, key );
+
+   DFN ( _sse_Attribute3fv, (*cache) );
+   FIXUP(dfn->code, 13, 0x0, (int)dest);
+   FIXUP(dfn->code, 18, 0x8, 8+(int)dest);
+   return dfn;
+}
+
+static struct dynfn *radeon_makeSSEAttribute3f( struct dynfn * cache, int key,
+					      const char * name, void * dest )
+{
+   struct dynfn *dfn = MALLOC_STRUCT( dynfn );
+
+   if (RADEON_DEBUG & DEBUG_CODEGEN)
+      fprintf(stderr, "%s 0x%08x\n", name, key );
+
+   DFN ( _sse_Attribute3f, (*cache) );
+   FIXUP(dfn->code, 12, 0x0, (int)dest); 
+   FIXUP(dfn->code, 17, 0x8, 8+(int)dest); 
+   return dfn;
+}
+
+static struct dynfn * radeon_makeSSENormal3fv( GLcontext *ctx, int key )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   return radeon_makeSSEAttribute3fv( & rmesa->vb.dfn_cache.Normal3fv, key,
+				      __FUNCTION__, rmesa->vb.normalptr );
+}
+
+static struct dynfn *radeon_makeSSENormal3f( GLcontext *ctx, int key )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   return radeon_makeSSEAttribute3f( & rmesa->vb.dfn_cache.Normal3f, key,
+				     __FUNCTION__, rmesa->vb.normalptr );
+}
+
+static struct dynfn *radeon_makeSSEColor3fv( GLcontext *ctx, int key )
+{
+   if (key & (RADEON_CP_VC_FRMT_PKCOLOR|RADEON_CP_VC_FRMT_FPALPHA))
+      return 0;
+   else
+   {
+      radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+      return radeon_makeSSEAttribute3fv( & rmesa->vb.dfn_cache.Color3fv, key,
+					 __FUNCTION__, rmesa->vb.floatcolorptr );
+   }
+}
+
+static struct dynfn *radeon_makeSSEColor3f( GLcontext *ctx, int key )
+{
+   if (key & (RADEON_CP_VC_FRMT_PKCOLOR|RADEON_CP_VC_FRMT_FPALPHA))
+      return 0;
+   else
+   {
+      radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+      return radeon_makeSSEAttribute3f( & rmesa->vb.dfn_cache.Color3f, key,
+					__FUNCTION__, rmesa->vb.floatcolorptr );
+   }
+}
+
+static struct dynfn *radeon_makeSSETexCoord2fv( GLcontext *ctx, int key )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   return radeon_makeSSEAttribute2fv( & rmesa->vb.dfn_cache.TexCoord2fv, key,
+				       __FUNCTION__, rmesa->vb.texcoordptr[0] );
+}
+
+static struct dynfn *radeon_makeSSETexCoord2f( GLcontext *ctx, int key )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   return radeon_makeSSEAttribute2f( & rmesa->vb.dfn_cache.TexCoord2f, key,
+				     __FUNCTION__, rmesa->vb.texcoordptr[0] );
+}
+
+static struct dynfn *radeon_makeSSEMultiTexCoord2fv( GLcontext *ctx, int key )
+{
+   struct dynfn *dfn = MALLOC_STRUCT( dynfn );
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if (RADEON_DEBUG & DEBUG_CODEGEN)
+      fprintf(stderr, "%s 0x%08x\n", __FUNCTION__, key );
+
+   if ((key & (RADEON_CP_VC_FRMT_ST0|RADEON_CP_VC_FRMT_ST1)) ==
+      (RADEON_CP_VC_FRMT_ST0|RADEON_CP_VC_FRMT_ST1)) {
+      DFN ( _sse_MultiTexCoord2fv, rmesa->vb.dfn_cache.MultiTexCoord2fvARB );
+      FIXUP(dfn->code, 18, 0xdeadbeef, (int)rmesa->vb.texcoordptr[0]);	
+   } else {
+      DFN ( _sse_MultiTexCoord2fv_2, rmesa->vb.dfn_cache.MultiTexCoord2fvARB );
+      FIXUP(dfn->code, 14, 0x0, (int)rmesa->vb.texcoordptr);
+   }
+   return dfn;
+}
+
+static struct dynfn *radeon_makeSSEMultiTexCoord2f( GLcontext *ctx, int key )
+{
+   struct dynfn *dfn = MALLOC_STRUCT( dynfn );
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if (RADEON_DEBUG & DEBUG_CODEGEN)
+      fprintf(stderr, "%s 0x%08x\n", __FUNCTION__, key );
+
+   if ((key & (RADEON_CP_VC_FRMT_ST0|RADEON_CP_VC_FRMT_ST1)) ==
+      (RADEON_CP_VC_FRMT_ST0|RADEON_CP_VC_FRMT_ST1)) {
+      DFN ( _sse_MultiTexCoord2f, rmesa->vb.dfn_cache.MultiTexCoord2fARB );
+      FIXUP(dfn->code, 16, 0xdeadbeef, (int)rmesa->vb.texcoordptr[0]);	
+   } else {
+      DFN ( _sse_MultiTexCoord2f_2, rmesa->vb.dfn_cache.MultiTexCoord2fARB );
+      FIXUP(dfn->code, 15, 0x0, (int)rmesa->vb.texcoordptr);
+   }
+   return dfn;
+}
+
+void radeonInitSSECodegen( struct dfn_generators *gen )
+{
+   if ( cpu_has_xmm ) {
+      gen->Normal3fv = (void *) radeon_makeSSENormal3fv;
+      gen->Normal3f = (void *) radeon_makeSSENormal3f;
+      gen->Color3fv = (void *) radeon_makeSSEColor3fv;
+      gen->Color3f = (void *) radeon_makeSSEColor3f;
+      gen->TexCoord2fv = (void *) radeon_makeSSETexCoord2fv;
+      gen->TexCoord2f = (void *) radeon_makeSSETexCoord2f;
+      gen->MultiTexCoord2fvARB = (void *) radeon_makeSSEMultiTexCoord2fv;
+      gen->MultiTexCoord2fARB = (void *) radeon_makeSSEMultiTexCoord2f;
+   }
+}
+
+#else 
+
+void radeonInitSSECodegen( struct dfn_generators *gen )
+{
+   (void) gen;
+}
+
+#endif
diff --git a/src/mesa/drivers/dri/radeon/radeon_vtxfmt_x86.c b/src/mesa/drivers/dri/radeon/radeon_vtxfmt_x86.c
new file mode 100644
index 00000000000..92941ca5f80
--- /dev/null
+++ b/src/mesa/drivers/dri/radeon/radeon_vtxfmt_x86.c
@@ -0,0 +1,437 @@
+/* $XFree86$ */
+/**************************************************************************
+
+Copyright 2000, 2001 ATI Technologies Inc., Ontario, Canada, and
+                     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, 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 <[email protected]>
+ */
+
+#include "glheader.h"
+#include "imports.h"
+#include "simple_list.h" 
+#include "radeon_vtxfmt.h"
+
+#if defined(USE_X86_ASM)
+
+#define EXTERN( FUNC )		\
+extern const char *FUNC;	\
+extern const char *FUNC##_end
+
+EXTERN ( _x86_Attribute2fv );
+EXTERN ( _x86_Attribute2f );
+EXTERN ( _x86_Attribute3fv );
+EXTERN ( _x86_Attribute3f );
+EXTERN ( _x86_Vertex3fv_6 );
+EXTERN ( _x86_Vertex3fv_8 );
+EXTERN ( _x86_Vertex3fv );
+EXTERN ( _x86_Vertex3f_4 );
+EXTERN ( _x86_Vertex3f_6 );
+EXTERN ( _x86_Vertex3f );
+EXTERN ( _x86_Color4ubv_ub );
+EXTERN ( _x86_Color4ubv_4f );
+EXTERN ( _x86_Color4ub_ub );
+EXTERN ( _x86_MultiTexCoord2fv );
+EXTERN ( _x86_MultiTexCoord2fv_2 );
+EXTERN ( _x86_MultiTexCoord2f );
+EXTERN ( _x86_MultiTexCoord2f_2 );
+
+
+/* Build specialized versions of the immediate calls on the fly for
+ * the current state.  Generic x86 versions.
+ */
+
+struct dynfn *radeon_makeX86Vertex3f( GLcontext *ctx, int key )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   struct dynfn *dfn = MALLOC_STRUCT( dynfn );
+
+   if (RADEON_DEBUG & DEBUG_CODEGEN)
+      fprintf(stderr, "%s 0x%08x %d\n", __FUNCTION__, key, rmesa->vb.vertex_size );
+
+   switch (rmesa->vb.vertex_size) {
+   case 4: {
+
+      DFN ( _x86_Vertex3f_4, rmesa->vb.dfn_cache.Vertex3f );
+      FIXUP(dfn->code, 2, 0x0, (int)&rmesa->vb.dmaptr);
+      FIXUP(dfn->code, 25, 0x0, (int)&rmesa->vb.vertex[3]);
+      FIXUP(dfn->code, 36, 0x0, (int)&rmesa->vb.counter);
+      FIXUP(dfn->code, 46, 0x0, (int)&rmesa->vb.dmaptr);
+      FIXUP(dfn->code, 51, 0x0, (int)&rmesa->vb.counter);
+      FIXUP(dfn->code, 60, 0x0, (int)&rmesa->vb.notify);
+      break;
+   }
+   case 6: {
+
+      DFN ( _x86_Vertex3f_6, rmesa->vb.dfn_cache.Vertex3f );
+      FIXUP(dfn->code, 3, 0x0, (int)&rmesa->vb.dmaptr);
+      FIXUP(dfn->code, 28, 0x0, (int)&rmesa->vb.vertex[3]);
+      FIXUP(dfn->code, 34, 0x0, (int)&rmesa->vb.vertex[4]);
+      FIXUP(dfn->code, 40, 0x0, (int)&rmesa->vb.vertex[5]);
+      FIXUP(dfn->code, 57, 0x0, (int)&rmesa->vb.counter);
+      FIXUP(dfn->code, 63, 0x0, (int)&rmesa->vb.dmaptr);
+      FIXUP(dfn->code, 70, 0x0, (int)&rmesa->vb.counter);
+      FIXUP(dfn->code, 79, 0x0, (int)&rmesa->vb.notify);
+      break;
+   }
+   default: {
+
+      DFN ( _x86_Vertex3f, rmesa->vb.dfn_cache.Vertex3f );
+      FIXUP(dfn->code, 3, 0x0, (int)&rmesa->vb.vertex[3]);
+      FIXUP(dfn->code, 9, 0x0, (int)&rmesa->vb.dmaptr);
+      FIXUP(dfn->code, 37, 0x0, rmesa->vb.vertex_size-3);
+      FIXUP(dfn->code, 44, 0x0, (int)&rmesa->vb.counter);
+      FIXUP(dfn->code, 50, 0x0, (int)&rmesa->vb.dmaptr);
+      FIXUP(dfn->code, 56, 0x0, (int)&rmesa->vb.counter);
+      FIXUP(dfn->code, 67, 0x0, (int)&rmesa->vb.notify);
+   break;
+   }
+   }
+
+   return dfn;
+}
+
+
+
+struct dynfn *radeon_makeX86Vertex3fv( GLcontext *ctx, int key )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+   struct dynfn *dfn = MALLOC_STRUCT( dynfn );
+
+   if (RADEON_DEBUG & DEBUG_CODEGEN)
+      fprintf(stderr, "%s 0x%08x %d\n", __FUNCTION__, key, rmesa->vb.vertex_size );
+
+   switch (rmesa->vb.vertex_size) {
+   case 6: {
+
+      DFN ( _x86_Vertex3fv_6, rmesa->vb.dfn_cache.Vertex3fv );
+      FIXUP(dfn->code, 1, 0x00000000, (int)&rmesa->vb.dmaptr);
+      FIXUP(dfn->code, 27, 0x0000001c, (int)&rmesa->vb.vertex[3]);
+      FIXUP(dfn->code, 33, 0x00000020, (int)&rmesa->vb.vertex[4]);
+      FIXUP(dfn->code, 45, 0x00000024, (int)&rmesa->vb.vertex[5]);
+      FIXUP(dfn->code, 56, 0x00000000, (int)&rmesa->vb.dmaptr);
+      FIXUP(dfn->code, 61, 0x00000004, (int)&rmesa->vb.counter);
+      FIXUP(dfn->code, 67, 0x00000004, (int)&rmesa->vb.counter);
+      FIXUP(dfn->code, 76, 0x00000008, (int)&rmesa->vb.notify);
+      break;
+   }
+   
+
+   case 8: {
+
+      DFN ( _x86_Vertex3fv_8, rmesa->vb.dfn_cache.Vertex3fv );
+      FIXUP(dfn->code, 1, 0x00000000, (int)&rmesa->vb.dmaptr);
+      FIXUP(dfn->code, 27, 0x0000001c, (int)&rmesa->vb.vertex[3]);
+      FIXUP(dfn->code, 33, 0x00000020, (int)&rmesa->vb.vertex[4]);
+      FIXUP(dfn->code, 45, 0x0000001c, (int)&rmesa->vb.vertex[5]);
+      FIXUP(dfn->code, 51, 0x00000020, (int)&rmesa->vb.vertex[6]);
+      FIXUP(dfn->code, 63, 0x00000024, (int)&rmesa->vb.vertex[7]);
+      FIXUP(dfn->code, 74, 0x00000000, (int)&rmesa->vb.dmaptr);
+      FIXUP(dfn->code, 79, 0x00000004, (int)&rmesa->vb.counter);
+      FIXUP(dfn->code, 85, 0x00000004, (int)&rmesa->vb.counter);
+      FIXUP(dfn->code, 94, 0x00000008, (int)&rmesa->vb.notify);
+      break;
+   }
+   
+
+
+   default: {
+
+      DFN ( _x86_Vertex3fv, rmesa->vb.dfn_cache.Vertex3fv );
+      FIXUP(dfn->code, 8, 0x01010101, (int)&rmesa->vb.dmaptr);
+      FIXUP(dfn->code, 32, 0x00000006, rmesa->vb.vertex_size-3);
+      FIXUP(dfn->code, 37, 0x00000058, (int)&rmesa->vb.vertex[3]);
+      FIXUP(dfn->code, 45, 0x01010101, (int)&rmesa->vb.dmaptr);
+      FIXUP(dfn->code, 50, 0x02020202, (int)&rmesa->vb.counter);
+      FIXUP(dfn->code, 58, 0x02020202, (int)&rmesa->vb.counter);
+      FIXUP(dfn->code, 67, 0x0, (int)&rmesa->vb.notify);
+   break;
+   }
+   }
+
+   return dfn;
+}
+
+static struct dynfn *
+radeon_makeX86Attribute2fv( struct dynfn * cache, int key,
+			    const char * name, void * dest )
+{
+   struct dynfn *dfn = MALLOC_STRUCT( dynfn );
+
+   if (RADEON_DEBUG & DEBUG_CODEGEN)
+      fprintf(stderr, "%s 0x%08x\n", name, key );
+
+   DFN ( _x86_Attribute2fv, (*cache) );
+   FIXUP(dfn->code, 11, 0x0, (int)dest); 
+   FIXUP(dfn->code, 16, 0x4, 4+(int)dest); 
+
+   return dfn;
+}
+
+static struct dynfn *
+radeon_makeX86Attribute2f( struct dynfn * cache, int key,
+			   const char * name, void * dest )
+{
+   struct dynfn *dfn = MALLOC_STRUCT( dynfn );
+
+   if (RADEON_DEBUG & DEBUG_CODEGEN)
+      fprintf(stderr, "%s 0x%08x\n", name, key );
+
+   DFN ( _x86_Attribute2f, (*cache) );
+   FIXUP(dfn->code, 1, 0x0, (int)dest); 
+
+   return dfn;
+}
+
+
+static struct dynfn *
+radeon_makeX86Attribute3fv( struct dynfn * cache, int key,
+			    const char * name, void * dest )
+{
+   struct dynfn *dfn = MALLOC_STRUCT( dynfn );
+
+   if (RADEON_DEBUG & DEBUG_CODEGEN)
+      fprintf(stderr, "%s 0x%08x\n", name, key );
+
+   DFN ( _x86_Attribute3fv, (*cache) );
+   FIXUP(dfn->code, 14, 0x0, (int)dest); 
+   FIXUP(dfn->code, 20, 0x4, 4+(int)dest); 
+   FIXUP(dfn->code, 25, 0x8, 8+(int)dest);
+
+   return dfn;
+}
+
+static struct dynfn *
+radeon_makeX86Attribute3f( struct dynfn * cache, int key,
+			 const char * name, void * dest )
+{
+   struct dynfn *dfn = MALLOC_STRUCT( dynfn );
+
+   if (RADEON_DEBUG & DEBUG_CODEGEN)
+      fprintf(stderr, "%s 0x%08x\n", name, key );
+
+   DFN ( _x86_Attribute3f, (*cache) );
+   FIXUP(dfn->code, 14, 0x0, (int)dest); 
+   FIXUP(dfn->code, 20, 0x4, 4+(int)dest); 
+   FIXUP(dfn->code, 25, 0x8, 8+(int)dest);
+
+   return dfn;
+}
+
+struct dynfn *radeon_makeX86Normal3fv( GLcontext *ctx, int key )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   return radeon_makeX86Attribute3fv( & rmesa->vb.dfn_cache.Normal3fv, key,
+				      __FUNCTION__, rmesa->vb.normalptr );
+}
+
+struct dynfn *radeon_makeX86Normal3f( GLcontext *ctx, int key )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   return radeon_makeX86Attribute3f( & rmesa->vb.dfn_cache.Normal3f, key,
+				     __FUNCTION__, rmesa->vb.normalptr );
+}
+
+struct dynfn *radeon_makeX86Color4ubv( GLcontext *ctx, int key )
+{
+   struct dynfn *dfn = MALLOC_STRUCT( dynfn );
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+
+   if (RADEON_DEBUG & DEBUG_CODEGEN)
+      fprintf(stderr, "%s 0x%08x\n", __FUNCTION__, key );
+
+   if (key & RADEON_CP_VC_FRMT_PKCOLOR) {
+      DFN ( _x86_Color4ubv_ub, rmesa->vb.dfn_cache.Color4ubv);
+      FIXUP(dfn->code, 5, 0x12345678, (int)rmesa->vb.colorptr); 
+      return dfn;
+   } 
+   else {
+
+      DFN ( _x86_Color4ubv_4f, rmesa->vb.dfn_cache.Color4ubv);
+      FIXUP(dfn->code, 2, 0x00000000, (int)_mesa_ubyte_to_float_color_tab); 
+      FIXUP(dfn->code, 27, 0xdeadbeaf, (int)rmesa->vb.floatcolorptr); 
+      FIXUP(dfn->code, 33, 0xdeadbeaf, (int)rmesa->vb.floatcolorptr+4); 
+      FIXUP(dfn->code, 55, 0xdeadbeaf, (int)rmesa->vb.floatcolorptr+8); 
+      FIXUP(dfn->code, 61, 0xdeadbeaf, (int)rmesa->vb.floatcolorptr+12); 
+      return dfn;
+   }
+}
+
+struct dynfn *radeon_makeX86Color4ub( GLcontext *ctx, int key )
+{
+   if (RADEON_DEBUG & DEBUG_CODEGEN)
+      fprintf(stderr, "%s 0x%08x\n", __FUNCTION__, key );
+
+   if (key & RADEON_CP_VC_FRMT_PKCOLOR) {
+      struct dynfn *dfn = MALLOC_STRUCT( dynfn );
+      radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+      DFN ( _x86_Color4ub_ub, rmesa->vb.dfn_cache.Color4ub );
+      FIXUP(dfn->code, 18, 0x0, (int)rmesa->vb.colorptr); 
+      FIXUP(dfn->code, 24, 0x0, (int)rmesa->vb.colorptr+1); 
+      FIXUP(dfn->code, 30, 0x0, (int)rmesa->vb.colorptr+2); 
+      FIXUP(dfn->code, 36, 0x0, (int)rmesa->vb.colorptr+3); 
+      return dfn;
+   }
+   else
+      return 0;
+}
+
+
+struct dynfn *radeon_makeX86Color3fv( GLcontext *ctx, int key )
+{
+   if (key & (RADEON_CP_VC_FRMT_PKCOLOR|RADEON_CP_VC_FRMT_FPALPHA))
+      return 0;
+   else
+   {
+      radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+      return radeon_makeX86Attribute3fv( & rmesa->vb.dfn_cache.Color3fv, key,
+					 __FUNCTION__, rmesa->vb.floatcolorptr );
+   }
+}
+
+struct dynfn *radeon_makeX86Color3f( GLcontext *ctx, int key )
+{
+   if (key & (RADEON_CP_VC_FRMT_PKCOLOR|RADEON_CP_VC_FRMT_FPALPHA))
+      return 0;
+   else
+   {
+      radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+      return radeon_makeX86Attribute3f( & rmesa->vb.dfn_cache.Color3f, key,
+					__FUNCTION__, rmesa->vb.floatcolorptr );
+   }
+}
+
+
+
+struct dynfn *radeon_makeX86TexCoord2fv( GLcontext *ctx, int key )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   return radeon_makeX86Attribute2fv( & rmesa->vb.dfn_cache.TexCoord2fv, key,
+				      __FUNCTION__, rmesa->vb.texcoordptr[0] );
+}
+
+struct dynfn *radeon_makeX86TexCoord2f( GLcontext *ctx, int key )
+{
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   return radeon_makeX86Attribute2f( & rmesa->vb.dfn_cache.TexCoord2f, key,
+				     __FUNCTION__, rmesa->vb.texcoordptr[0] );
+}
+
+struct dynfn *radeon_makeX86MultiTexCoord2fvARB( GLcontext *ctx, int key )
+{
+   struct dynfn *dfn = MALLOC_STRUCT( dynfn );
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if (RADEON_DEBUG & DEBUG_CODEGEN)
+      fprintf(stderr, "%s 0x%08x\n", __FUNCTION__, key );
+
+   if ((key & (RADEON_CP_VC_FRMT_ST0|RADEON_CP_VC_FRMT_ST1)) ==
+      (RADEON_CP_VC_FRMT_ST0|RADEON_CP_VC_FRMT_ST1)) {
+      DFN ( _x86_MultiTexCoord2fv, rmesa->vb.dfn_cache.MultiTexCoord2fvARB );
+      FIXUP(dfn->code, 21, 0xdeadbeef, (int)rmesa->vb.texcoordptr[0]);
+      FIXUP(dfn->code, 27, 0xdeadbeef, (int)rmesa->vb.texcoordptr[0]+4);
+   } else {
+      DFN ( _x86_MultiTexCoord2fv_2, rmesa->vb.dfn_cache.MultiTexCoord2fvARB );
+      FIXUP(dfn->code, 14, 0x0, (int)rmesa->vb.texcoordptr);
+   }
+   return dfn;
+}
+
+struct dynfn *radeon_makeX86MultiTexCoord2fARB( GLcontext *ctx, 
+						int key )
+{
+   struct dynfn *dfn = MALLOC_STRUCT( dynfn );
+   radeonContextPtr rmesa = RADEON_CONTEXT(ctx);
+
+   if (RADEON_DEBUG & DEBUG_CODEGEN)
+      fprintf(stderr, "%s 0x%08x\n", __FUNCTION__, key );
+
+   if ((key & (RADEON_CP_VC_FRMT_ST0|RADEON_CP_VC_FRMT_ST1)) ==
+       (RADEON_CP_VC_FRMT_ST0|RADEON_CP_VC_FRMT_ST1)) {
+      DFN ( _x86_MultiTexCoord2f, rmesa->vb.dfn_cache.MultiTexCoord2fARB );
+      FIXUP(dfn->code, 20, 0xdeadbeef, (int)rmesa->vb.texcoordptr[0]);
+      FIXUP(dfn->code, 26, 0xdeadbeef, (int)rmesa->vb.texcoordptr[0]+4); 
+   }
+   else {
+      /* Note: this might get generated multiple times, even though the
+       * actual emitted code is the same.
+       */
+      DFN ( _x86_MultiTexCoord2f_2, rmesa->vb.dfn_cache.MultiTexCoord2fARB );
+      FIXUP(dfn->code, 18, 0x0, (int)rmesa->vb.texcoordptr); 
+   }      
+   return dfn;
+}
+
+
+void radeonInitX86Codegen( struct dfn_generators *gen )
+{
+   gen->Vertex3f = radeon_makeX86Vertex3f;
+   gen->Vertex3fv = radeon_makeX86Vertex3fv;
+   gen->Color4ub = radeon_makeX86Color4ub; /* PKCOLOR only */
+   gen->Color4ubv = radeon_makeX86Color4ubv; /* PKCOLOR only */
+   gen->Normal3f = radeon_makeX86Normal3f;
+   gen->Normal3fv = radeon_makeX86Normal3fv;
+   gen->TexCoord2f = radeon_makeX86TexCoord2f;
+   gen->TexCoord2fv = radeon_makeX86TexCoord2fv;
+   gen->MultiTexCoord2fARB = radeon_makeX86MultiTexCoord2fARB;
+   gen->MultiTexCoord2fvARB = radeon_makeX86MultiTexCoord2fvARB;
+   gen->Color3f = radeon_makeX86Color3f;
+   gen->Color3fv = radeon_makeX86Color3fv;
+
+   /* Not done:
+    */
+/*     gen->Vertex2f = radeon_makeX86Vertex2f; */
+/*     gen->Vertex2fv = radeon_makeX86Vertex2fv; */
+/*     gen->Color3ub = radeon_makeX86Color3ub; */
+/*     gen->Color3ubv = radeon_makeX86Color3ubv; */
+/*     gen->Color4f = radeon_makeX86Color4f; */
+/*     gen->Color4fv = radeon_makeX86Color4fv; */
+/*     gen->TexCoord1f = radeon_makeX86TexCoord1f; */
+/*     gen->TexCoord1fv = radeon_makeX86TexCoord1fv; */
+/*     gen->MultiTexCoord1fARB = radeon_makeX86MultiTexCoord1fARB; */
+/*     gen->MultiTexCoord1fvARB = radeon_makeX86MultiTexCoord1fvARB; */
+}
+
+
+#else 
+
+void radeonInitX86Codegen( struct dfn_generators *gen )
+{
+   (void) gen;
+}
+
+#endif
diff --git a/src/mesa/drivers/dri/radeon/server/radeon_common.h b/src/mesa/drivers/dri/radeon/server/radeon_common.h
index c26ccd3cc2d..0792b5c2e0e 100644
--- a/src/mesa/drivers/dri/radeon/server/radeon_common.h
+++ b/src/mesa/drivers/dri/radeon/server/radeon_common.h
@@ -1,19 +1,5 @@
-/**
- * \file server/radeon_common.h 
- * \brief Common header definitions for Radeon 2D/3D/DRM driver suite.
+/* radeon_common.h -- common header definitions for Radeon 2D/3D/DRM suite
  *
- * \note Some of these structures are meant for backward compatibility and
- * aren't used by the subset driver.
- *
- * \author Gareth Hughes <[email protected]>
- * \author Kevin E. Martin <[email protected]>
- * \author Keith Whitwell <[email protected]>
- * 
- * \author Converted to common header format by
- * Jens Owen <[email protected]>
- */
-
-/*
  * Copyright 2000 VA Linux Systems, Inc., Fremont, California.
  * Copyright 2002 Tungsten Graphics, Inc., Cedar Park, Texas.
  * All Rights Reserved.
@@ -36,10 +22,19 @@
  * 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:
+ *   Gareth Hughes <[email protected]>
+ *   Kevin E. Martin <[email protected]>
+ *   Keith Whitwell <[email protected]>
+ *
+ * Converted to common header format:
+ *   Jens Owen <[email protected]>
+ *
+ * $XFree86: xc/programs/Xserver/hw/xfree86/os-support/xf86drmRadeon.h,v 1.6 2001/04/16 15:02:13 tsi Exp $
+ *
  */
 
-/* $XFree86: xc/programs/Xserver/hw/xfree86/os-support/xf86drmRadeon.h,v 1.6 2001/04/16 15:02:13 tsi Exp $ */
-
 #ifndef _RADEON_COMMON_H_
 #define _RADEON_COMMON_H_
 
@@ -90,42 +85,33 @@
 #define RADEON_CLEAR_DEPTH     4
 
 
-/**
- * \brief DRM_RADEON_CP_INIT ioctl argument type.
- */
 typedef struct {
    enum {
-      DRM_RADEON_INIT_CP    = 0x01,   /**< \brief initialize CP */
-      DRM_RADEON_CLEANUP_CP = 0x02,   /**< \brief clean up CP */
-      DRM_RADEON_INIT_R200_CP = 0x03  /**< \brief initialize R200 CP */
-   } func;                            /**< \brief request */
-   unsigned long sarea_priv_offset;   /**< \brief SAREA private offset */
-   int is_pci;                        /**< \brief is current card a PCI card? */
-   int cp_mode;                       /**< \brief CP mode */
-   int agp_size;                      /**< \brief AGP space size */
-   int ring_size;                     /**< \brief CP ring buffer size */
-   int usec_timeout;                  /**< \brief timeout for DRM operations in usecs */
-
-   unsigned int fb_bpp;               
-   unsigned int front_offset;         /**< \brief front color buffer offset */
-   unsigned int front_pitch;          /**< \brief front color buffer pitch */
-   unsigned int back_offset;          /**< \brief back color buffer offset */
-   unsigned int back_pitch;           /**< \brief back color buffer pitch*/
-   unsigned int depth_bpp;            /**< \brief depth buffer bits-per-pixel */
-   unsigned int depth_offset;         /**< \brief depth buffer offset */
-   unsigned int depth_pitch;          /**< \brief depth buffer pitch */
-
-   unsigned long fb_offset;           /**< \brief framebuffer offset */
-   unsigned long mmio_offset;         /**< \brief MMIO register offset */
-   unsigned long ring_offset;         /**< \brief CP ring buffer offset */
-   unsigned long ring_rptr_offset;    /**< \brief CP ring buffer read pointer offset */
-   unsigned long buffers_offset;      /**< \brief vertex buffers offset */
-   unsigned long agp_textures_offset; /**< \brief AGP textures offset */
+      DRM_RADEON_INIT_CP    = 0x01,
+      DRM_RADEON_CLEANUP_CP = 0x02,
+      DRM_RADEON_INIT_R200_CP = 0x03
+   } func;
+   unsigned long sarea_priv_offset;
+   int is_pci;
+   int cp_mode;
+   int agp_size;
+   int ring_size;
+   int usec_timeout;
+
+   unsigned int fb_bpp;
+   unsigned int front_offset, front_pitch;
+   unsigned int back_offset, back_pitch;
+   unsigned int depth_bpp;
+   unsigned int depth_offset, depth_pitch;
+
+   unsigned long fb_offset;
+   unsigned long mmio_offset;
+   unsigned long ring_offset;
+   unsigned long ring_rptr_offset;
+   unsigned long buffers_offset;
+   unsigned long agp_textures_offset;
 } drmRadeonInit;
 
-/**
- * \brief DRM_RADEON_CP_STOP ioctl argument type.
- */
 typedef struct {
    int flush;
    int idle;
@@ -143,17 +129,13 @@ typedef union drmRadeonClearR {
         unsigned int ui[5];
 } drmRadeonClearRect;
 
-/**
- * \brief DRM_RADEON_CLEAR ioctl argument type.
- */
 typedef struct drmRadeonClearT {
-        unsigned int flags;              /**< \brief bitmask of the planes to clear */
-        unsigned int clear_color;        /**< \brief color buffer clear value */
-        unsigned int clear_depth;        /**< \brief depth buffer clear value */
-        unsigned int color_mask;         /**< \brief color buffer clear mask */
-        unsigned int depth_mask;         /**< \brief stencil buffer clear value
-					   *  \todo Misnamed field. */
-        drmRadeonClearRect *depth_boxes; /**< \brief depth buffer cliprects */
+        unsigned int flags;
+        unsigned int clear_color;
+        unsigned int clear_depth;
+        unsigned int color_mask;
+        unsigned int depth_mask;   /* misnamed field:  should be stencil */
+        drmRadeonClearRect *depth_boxes;
 } drmRadeonClearType;
 
 typedef struct drmRadeonFullscreenT {
@@ -163,16 +145,10 @@ typedef struct drmRadeonFullscreenT {
         } func;
 } drmRadeonFullscreenType;
 
-/**
- * \brief DRM_RADEON_STIPPLE ioctl argument type.
- */
 typedef struct {
         unsigned int *mask;
 } drmRadeonStipple;
 
-/**
- * \brief Texture image for drmRadeonTexture.
- */
 typedef struct {
         unsigned int x;
         unsigned int y;
@@ -181,22 +157,18 @@ typedef struct {
         const void *data;
 } drmRadeonTexImage;
 
-/**
- * \brief DRM_RADEON_TEXTURE ioctl argument type.
- */
 typedef struct {
-        int offset;               /**< \brief texture offset */
-        int pitch;                /**< \brief texture pitch */
-        int format;               /**< \brief pixel format */
-        int width;                /**< \brief texture width */
-        int height;               /**< \brief texture height */
-	drmRadeonTexImage *image; /**< \brief image */
+        int offset;
+        int pitch;
+        int format;
+        int width;                      /* Texture image coordinates */
+        int height;
+        drmRadeonTexImage *image;
 } drmRadeonTexture;
 
 
 #define RADEON_MAX_TEXTURE_UNITS 3
 
-
 /* Layout matches drm_radeon_state_t in linux drm_radeon.h.  
  */
 typedef struct {
@@ -266,16 +238,13 @@ typedef struct {
 	unsigned int dirty;
 } drmRadeonState;
 
-/**
- * \brief DRM 1.1 vertex ioctl.
- *
- * Used in compatibility modes.
+/* 1.1 vertex ioctl.  Used in compatibility modes.
  */
 typedef struct {
-	int prim;			/**< \brief Primitive number */
-	int idx;			/**< \brief Index of vertex buffer */
-	int count;			/**< \brief Number of vertices in buffer */
-	int discard;			/**< \brief Client finished with buffer? */
+	int prim;
+	int idx;			/* Index of vertex buffer */
+	int count;			/* Number of vertices in buffer */
+	int discard;			/* Client finished with buffer? */
 } drmRadeonVertex;
 
 typedef struct {
@@ -283,13 +252,13 @@ typedef struct {
 	unsigned int finish;
 	unsigned int prim:8;
 	unsigned int stateidx:8;
-	unsigned int numverts:16;	/**< overloaded as offset/64 for elt prims */
+	unsigned int numverts:16; /* overloaded as offset/64 for elt prims */
         unsigned int vc_format;
 } drmRadeonPrim;
 
 typedef struct {
-        int idx;                        /**< \brief Index of vertex buffer */
-        int discard;                    /**< \brief Client finished with buffer? */
+        int idx;                        /* Index of vertex buffer */
+        int discard;                    /* Client finished with buffer? */
         int nr_states;
         drmRadeonState *state;
         int nr_prims;
@@ -299,156 +268,127 @@ typedef struct {
 #define RADEON_MAX_STATES 16
 #define RADEON_MAX_PRIMS  64
 
-
-/**
- * \brief Command buffer.  
- *
- * \todo Replace with true DMA stream?
+/* Command buffer.  Replace with true dma stream?
  */
 typedef struct {
-	int bufsz;          /**< \brief buffer size */
-	char *buf;          /**< \brief buffer */
-	int nbox;           /**< \brief number of cliprects */
-        drmClipRect *boxes; /**< \brief cliprects */
+	int bufsz;
+	char *buf;
+	int nbox;
+        drmClipRect *boxes;
 } drmRadeonCmdBuffer;
 
-
-/**
- * \brief Per-packet identifiers for use with the ::RADEON_CMD_PACKET command
- * in the DRM_RADEON_CMDBUF ioctl.  
- *
- * \note Comments relate new packets to old state bits and the packet size.
+/* New style per-packet identifiers for use in cmd_buffer ioctl with
+ * the RADEON_EMIT_PACKET command.  Comments relate new packets to old
+ * state bits and the packet size:
  */
-enum drmRadeonCmdPkt {
-   RADEON_EMIT_PP_MISC                       = 0, /* context/7 */
-   RADEON_EMIT_PP_CNTL                       = 1, /* context/3 */
-   RADEON_EMIT_RB3D_COLORPITCH               = 2, /* context/1 */
-   RADEON_EMIT_RE_LINE_PATTERN               = 3, /* line/2 */
-   RADEON_EMIT_SE_LINE_WIDTH                 = 4, /* line/1 */
-   RADEON_EMIT_PP_LUM_MATRIX                 = 5, /* bumpmap/1 */
-   RADEON_EMIT_PP_ROT_MATRIX_0               = 6, /* bumpmap/2 */
-   RADEON_EMIT_RB3D_STENCILREFMASK           = 7, /* masks/3 */
-   RADEON_EMIT_SE_VPORT_XSCALE               = 8, /* viewport/6 */
-   RADEON_EMIT_SE_CNTL                       = 9, /* setup/2 */
-   RADEON_EMIT_SE_CNTL_STATUS                = 10, /* setup/1 */
-   RADEON_EMIT_RE_MISC                       = 11, /* misc/1 */
-   RADEON_EMIT_PP_TXFILTER_0                 = 12, /* tex0/6 */
-   RADEON_EMIT_PP_BORDER_COLOR_0             = 13, /* tex0/1 */
-   RADEON_EMIT_PP_TXFILTER_1                 = 14, /* tex1/6 */
-   RADEON_EMIT_PP_BORDER_COLOR_1             = 15, /* tex1/1 */
-   RADEON_EMIT_PP_TXFILTER_2                 = 16, /* tex2/6 */
-   RADEON_EMIT_PP_BORDER_COLOR_2             = 17, /* tex2/1 */
-   RADEON_EMIT_SE_ZBIAS_FACTOR               = 18, /* zbias/2 */
-   RADEON_EMIT_SE_TCL_OUTPUT_VTX_FMT         = 19, /* tcl/11 */
-   RADEON_EMIT_SE_TCL_MATERIAL_EMMISSIVE_RED = 20, /* material/17 */
-   R200_EMIT_PP_TXCBLEND_0                   = 21, /* tex0/4 */
-   R200_EMIT_PP_TXCBLEND_1                   = 22, /* tex1/4 */
-   R200_EMIT_PP_TXCBLEND_2                   = 23, /* tex2/4 */
-   R200_EMIT_PP_TXCBLEND_3                   = 24, /* tex3/4 */
-   R200_EMIT_PP_TXCBLEND_4                   = 25, /* tex4/4 */
-   R200_EMIT_PP_TXCBLEND_5                   = 26, /* tex5/4 */
-   R200_EMIT_PP_TXCBLEND_6                   = 27, /* /4 */
-   R200_EMIT_PP_TXCBLEND_7                   = 28, /* /4 */
-   R200_EMIT_TCL_LIGHT_MODEL_CTL_0           = 29, /* tcl/6 */
-   R200_EMIT_TFACTOR_0                       = 30, /* tf/6 */
-   R200_EMIT_VTX_FMT_0                       = 31, /* vtx/4 */
-   R200_EMIT_VAP_CTL                         = 32, /* vap/1 */
-   R200_EMIT_MATRIX_SELECT_0                 = 33, /* msl/5 */
-   R200_EMIT_TEX_PROC_CTL_2                  = 34, /* tcg/5 */
-   R200_EMIT_TCL_UCP_VERT_BLEND_CTL          = 35, /* tcl/1 */
-   R200_EMIT_PP_TXFILTER_0                   = 36, /* tex0/6 */
-   R200_EMIT_PP_TXFILTER_1                   = 37, /* tex1/6 */
-   R200_EMIT_PP_TXFILTER_2                   = 38, /* tex2/6 */
-   R200_EMIT_PP_TXFILTER_3                   = 39, /* tex3/6 */
-   R200_EMIT_PP_TXFILTER_4                   = 40, /* tex4/6 */
-   R200_EMIT_PP_TXFILTER_5                   = 41, /* tex5/6 */
-   R200_EMIT_PP_TXOFFSET_0                   = 42, /* tex0/1 */
-   R200_EMIT_PP_TXOFFSET_1                   = 43, /* tex1/1 */
-   R200_EMIT_PP_TXOFFSET_2                   = 44, /* tex2/1 */
-   R200_EMIT_PP_TXOFFSET_3                   = 45, /* tex3/1 */
-   R200_EMIT_PP_TXOFFSET_4                   = 46, /* tex4/1 */
-   R200_EMIT_PP_TXOFFSET_5                   = 47, /* tex5/1 */
-   R200_EMIT_VTE_CNTL                        = 48, /* vte/1 */
-   R200_EMIT_OUTPUT_VTX_COMP_SEL             = 49, /* vtx/1 */
-   R200_EMIT_PP_TAM_DEBUG3                   = 50, /* tam/1 */
-   R200_EMIT_PP_CNTL_X                       = 51, /* cst/1 */
-   R200_EMIT_RB3D_DEPTHXY_OFFSET             = 52, /* cst/1 */
-   R200_EMIT_RE_AUX_SCISSOR_CNTL             = 53, /* cst/1 */
-   R200_EMIT_RE_SCISSOR_TL_0                 = 54, /* cst/2 */
-   R200_EMIT_RE_SCISSOR_TL_1                 = 55, /* cst/2 */
-   R200_EMIT_RE_SCISSOR_TL_2                 = 56, /* cst/2 */
-   R200_EMIT_SE_VAP_CNTL_STATUS              = 57, /* cst/1 */
-   R200_EMIT_SE_VTX_STATE_CNTL               = 58, /* cst/1 */
-   R200_EMIT_RE_POINTSIZE                    = 59, /* cst/1 */
-   R200_EMIT_TCL_INPUT_VTX_VECTOR_ADDR_0     = 60, /* cst/4 */
-   R200_EMIT_PP_CUBIC_FACES_0                = 61,
-   R200_EMIT_PP_CUBIC_OFFSETS_0              = 62,
-   R200_EMIT_PP_CUBIC_FACES_1                = 63,
-   R200_EMIT_PP_CUBIC_OFFSETS_1              = 64,
-   R200_EMIT_PP_CUBIC_FACES_2                = 65,
-   R200_EMIT_PP_CUBIC_OFFSETS_2              = 66,
-   R200_EMIT_PP_CUBIC_FACES_3                = 67,
-   R200_EMIT_PP_CUBIC_OFFSETS_3              = 68,
-   R200_EMIT_PP_CUBIC_FACES_4                = 69,
-   R200_EMIT_PP_CUBIC_OFFSETS_4              = 70,
-   R200_EMIT_PP_CUBIC_FACES_5                = 71,
-   R200_EMIT_PP_CUBIC_OFFSETS_5              = 72,
-   RADEON_MAX_STATE_PACKETS                  = 73
-} ;
-
-
-/**
- * \brief Command types understood by the DRM_RADEON_CMDBUF ioctl.  
- * 
- * More can be added but obviously these can't be removed or changed.
- *
- * \sa drmRadeonCmdHeader.
- */
-enum drmRadeonCmdType {
-   RADEON_CMD_PACKET       = 1, /**< \brief emit one of the ::drmRadeonCmdPkt register packets */
-   RADEON_CMD_SCALARS      = 2, /**< \brief emit scalar data */
-   RADEON_CMD_VECTORS      = 3, /**< \brief emit vector data */
-   RADEON_CMD_DMA_DISCARD  = 4, /**< \brief discard current DMA buffer */
-   RADEON_CMD_PACKET3      = 5, /**< \brief emit hardware packet */
-   RADEON_CMD_PACKET3_CLIP = 6, /**< \brief emit hardware packet wrapped in cliprects */
-   RADEON_CMD_SCALARS2     = 7, /**< \brief R200 stopgap */
-   RADEON_CMD_WAIT         = 8  /**< \brief synchronization */
-} ;
-
-/**
- * \brief Command packet headers understood by the DRM_RADEON_CMDBUF ioctl.
- *
- * \sa drmRadeonCmdType.
+#define RADEON_EMIT_PP_MISC                         0 /* context/7 */
+#define RADEON_EMIT_PP_CNTL                         1 /* context/3 */
+#define RADEON_EMIT_RB3D_COLORPITCH                 2 /* context/1 */
+#define RADEON_EMIT_RE_LINE_PATTERN                 3 /* line/2 */
+#define RADEON_EMIT_SE_LINE_WIDTH                   4 /* line/1 */
+#define RADEON_EMIT_PP_LUM_MATRIX                   5 /* bumpmap/1 */
+#define RADEON_EMIT_PP_ROT_MATRIX_0                 6 /* bumpmap/2 */
+#define RADEON_EMIT_RB3D_STENCILREFMASK             7 /* masks/3 */
+#define RADEON_EMIT_SE_VPORT_XSCALE                 8 /* viewport/6 */
+#define RADEON_EMIT_SE_CNTL                         9 /* setup/2 */
+#define RADEON_EMIT_SE_CNTL_STATUS                  10 /* setup/1 */
+#define RADEON_EMIT_RE_MISC                         11 /* misc/1 */
+#define RADEON_EMIT_PP_TXFILTER_0                   12 /* tex0/6 */
+#define RADEON_EMIT_PP_BORDER_COLOR_0               13 /* tex0/1 */
+#define RADEON_EMIT_PP_TXFILTER_1                   14 /* tex1/6 */
+#define RADEON_EMIT_PP_BORDER_COLOR_1               15 /* tex1/1 */
+#define RADEON_EMIT_PP_TXFILTER_2                   16 /* tex2/6 */
+#define RADEON_EMIT_PP_BORDER_COLOR_2               17 /* tex2/1 */
+#define RADEON_EMIT_SE_ZBIAS_FACTOR                 18 /* zbias/2 */
+#define RADEON_EMIT_SE_TCL_OUTPUT_VTX_FMT           19 /* tcl/11 */
+#define RADEON_EMIT_SE_TCL_MATERIAL_EMMISSIVE_RED   20 /* material/17 */
+#define R200_EMIT_PP_TXCBLEND_0                     21 /* tex0/4 */
+#define R200_EMIT_PP_TXCBLEND_1                     22 /* tex1/4 */
+#define R200_EMIT_PP_TXCBLEND_2                     23 /* tex2/4 */
+#define R200_EMIT_PP_TXCBLEND_3                     24 /* tex3/4 */
+#define R200_EMIT_PP_TXCBLEND_4                     25 /* tex4/4 */
+#define R200_EMIT_PP_TXCBLEND_5                     26 /* tex5/4 */
+#define R200_EMIT_PP_TXCBLEND_6                     27 /* /4 */
+#define R200_EMIT_PP_TXCBLEND_7                     28 /* /4 */
+#define R200_EMIT_TCL_LIGHT_MODEL_CTL_0             29 /* tcl/6 */
+#define R200_EMIT_TFACTOR_0                         30 /* tf/6 */
+#define R200_EMIT_VTX_FMT_0                         31 /* vtx/4 */
+#define R200_EMIT_VAP_CTL                           32 /* vap/1 */
+#define R200_EMIT_MATRIX_SELECT_0                   33 /* msl/5 */
+#define R200_EMIT_TEX_PROC_CTL_2                    34 /* tcg/5 */
+#define R200_EMIT_TCL_UCP_VERT_BLEND_CTL            35 /* tcl/1 */
+#define R200_EMIT_PP_TXFILTER_0                     36 /* tex0/6 */
+#define R200_EMIT_PP_TXFILTER_1                     37 /* tex1/6 */
+#define R200_EMIT_PP_TXFILTER_2                     38 /* tex2/6 */
+#define R200_EMIT_PP_TXFILTER_3                     39 /* tex3/6 */
+#define R200_EMIT_PP_TXFILTER_4                     40 /* tex4/6 */
+#define R200_EMIT_PP_TXFILTER_5                     41 /* tex5/6 */
+#define R200_EMIT_PP_TXOFFSET_0                     42 /* tex0/1 */
+#define R200_EMIT_PP_TXOFFSET_1                     43 /* tex1/1 */
+#define R200_EMIT_PP_TXOFFSET_2                     44 /* tex2/1 */
+#define R200_EMIT_PP_TXOFFSET_3                     45 /* tex3/1 */
+#define R200_EMIT_PP_TXOFFSET_4                     46 /* tex4/1 */
+#define R200_EMIT_PP_TXOFFSET_5                     47 /* tex5/1 */
+#define R200_EMIT_VTE_CNTL                          48 /* vte/1 */
+#define R200_EMIT_OUTPUT_VTX_COMP_SEL               49 /* vtx/1 */
+#define R200_EMIT_PP_TAM_DEBUG3                     50 /* tam/1 */
+#define R200_EMIT_PP_CNTL_X                         51 /* cst/1 */
+#define R200_EMIT_RB3D_DEPTHXY_OFFSET               52 /* cst/1 */
+#define R200_EMIT_RE_AUX_SCISSOR_CNTL               53 /* cst/1 */
+#define R200_EMIT_RE_SCISSOR_TL_0                   54 /* cst/2 */
+#define R200_EMIT_RE_SCISSOR_TL_1                   55 /* cst/2 */
+#define R200_EMIT_RE_SCISSOR_TL_2                   56 /* cst/2 */
+#define R200_EMIT_SE_VAP_CNTL_STATUS                57 /* cst/1 */
+#define R200_EMIT_SE_VTX_STATE_CNTL                 58 /* cst/1 */
+#define R200_EMIT_RE_POINTSIZE                      59 /* cst/1 */
+#define R200_EMIT_TCL_INPUT_VTX_VECTOR_ADDR_0       60 /* cst/4 */
+#define R200_EMIT_PP_CUBIC_FACES_0                  61
+#define R200_EMIT_PP_CUBIC_OFFSETS_0                62
+#define R200_EMIT_PP_CUBIC_FACES_1                  63
+#define R200_EMIT_PP_CUBIC_OFFSETS_1                64
+#define R200_EMIT_PP_CUBIC_FACES_2                  65
+#define R200_EMIT_PP_CUBIC_OFFSETS_2                66
+#define R200_EMIT_PP_CUBIC_FACES_3                  67
+#define R200_EMIT_PP_CUBIC_OFFSETS_3                68
+#define R200_EMIT_PP_CUBIC_FACES_4                  69
+#define R200_EMIT_PP_CUBIC_OFFSETS_4                70
+#define R200_EMIT_PP_CUBIC_FACES_5                  71
+#define R200_EMIT_PP_CUBIC_OFFSETS_5                72
+#define RADEON_EMIT_PP_TEX_SIZE_0                   73
+#define RADEON_EMIT_PP_TEX_SIZE_1                   74
+#define RADEON_EMIT_PP_TEX_SIZE_2                   75
+#define RADEON_MAX_STATE_PACKETS                    76
+
+
+/* Commands understood by cmd_buffer ioctl.  More can be added but
+ * obviously these can't be removed or changed:
  */
+#define RADEON_CMD_PACKET      1 /* emit one of the register packets above */
+#define RADEON_CMD_SCALARS     2 /* emit scalar data */
+#define RADEON_CMD_VECTORS     3 /* emit vector data */
+#define RADEON_CMD_DMA_DISCARD 4 /* discard current dma buf */
+#define RADEON_CMD_PACKET3     5 /* emit hw packet */
+#define RADEON_CMD_PACKET3_CLIP 6 /* emit hw packet wrapped in cliprects */
+#define RADEON_CMD_SCALARS2     7 /* R200 stopgap */
+#define RADEON_CMD_WAIT         8 /* synchronization */
+
 typedef union {
-	/** \brief integer equivalent */
 	int i;
-
 	struct { 
 	   unsigned char cmd_type, pad0, pad1, pad2;
 	} header;
-
-	/** \brief emit a register packet */
 	struct { 
 	   unsigned char cmd_type, packet_id, pad0, pad1;
 	} packet;
-	
-	/** \brief scalar data */
 	struct { 
 	   unsigned char cmd_type, offset, stride, count; 
 	} scalars;
-	
-	/** \brief vector data */
 	struct { 
 	   unsigned char cmd_type, offset, stride, count; 
 	} vectors;
-	
-	/** \brief discard current DMA buffer */
 	struct { 
 	   unsigned char cmd_type, buf_idx, pad0, pad1; 
 	} dma;
-	
-	/** \brief synchronization */
 	struct { 
 	   unsigned char cmd_type, flags, pad0, pad1; 
 	} wait;
@@ -458,12 +398,10 @@ typedef union {
 #define RADEON_WAIT_2D  0x1
 #define RADEON_WAIT_3D  0x2
 
-/**
- * \brief DRM_RADEON_GETPARAM ioctl argument type.
- */
+
 typedef struct drm_radeon_getparam {
-	int param;  /**< \brief parameter number */
-	int *value; /**< \brief parameter value */
+	int param;
+	void *value;
 } drmRadeonGetParam;
 
 #define RADEON_PARAM_AGP_BUFFER_OFFSET 1
@@ -472,10 +410,6 @@ typedef struct drm_radeon_getparam {
 #define RADEON_PARAM_LAST_CLEAR        4
 #define RADEON_PARAM_IRQ_NR            5
 #define RADEON_PARAM_AGP_BASE          6
-#define RADEON_PARAM_REGISTER_HANDLE   7 
-#define RADEON_PARAM_STATUS_HANDLE     8
-#define RADEON_PARAM_SAREA_HANDLE      9
-#define RADEON_PARAM_AGP_TEX_HANDLE    10
 
 
 #define RADEON_MEM_REGION_AGP 1
@@ -493,29 +427,18 @@ typedef struct drm_radeon_mem_free {
 	int region_offset;
 } drmRadeonMemFree;
 
-/**
- * \brief DRM_RADEON_INIT_HEAP argument type.
- */
 typedef struct drm_radeon_mem_init_heap {
-	int region; /**< \brief region type */
-	int size;   /**< \brief region size */
-	int start;  /**< \brief region start offset */
+	int region;
+	int size;
+	int start;	
 } drmRadeonMemInitHeap;
 
-/**
- * \brief DRM_RADEON_IRQ_EMIT ioctl argument type.
- *
- * New in DRM 1.6: userspace can request and wait on IRQ's.
+/* 1.6: Userspace can request & wait on irq's:
  */
 typedef struct drm_radeon_irq_emit {
 	int *irq_seq;
 } drmRadeonIrqEmit;
 
-/**
- * \brief DRM_RADEON_IRQ_WAIT ioctl argument type.
- *
- * New in DRM 1.6: userspace can request and wait on IRQ's.
- */
 typedef struct drm_radeon_irq_wait {
 	int irq_seq;
 } drmRadeonIrqWait;
diff --git a/src/mesa/drivers/dri/radeon/server/radeon_dri.c b/src/mesa/drivers/dri/radeon/server/radeon_dri.c
index f14bd13a952..4271aa7da23 100644
--- a/src/mesa/drivers/dri/radeon/server/radeon_dri.c
+++ b/src/mesa/drivers/dri/radeon/server/radeon_dri.c
@@ -12,6 +12,7 @@
 #include <stdlib.h>
 #include <string.h>
 #include <errno.h>
+#include <unistd.h>
 
 #include "driver.h"
 #include "drm.h"
@@ -23,9 +24,6 @@
 #include "radeon_sarea.h"
 #include "sarea.h"
 
-#include <unistd.h>
-
-
 
 /* HACK - for now, put this here... */
 /* Alpha - this may need to be a variable to handle UP1x00 vs TITAN */
@@ -735,7 +733,7 @@ static int RADEONMemoryInit( const DRIDriverContext *ctx, RADEONInfoPtr info )
  * Setups a RADEONDRIRec structure to be passed to radeon_dri.so for its
  * initialization.
  */
-static int RADEONScreenInit( DRIDriverContext *ctx, RADEONInfoPtr info )
+static int RADEONScreenInit( const DRIDriverContext *ctx, RADEONInfoPtr info )
 {
    RADEONDRIPtr   pRADEONDRI;
    int err;
diff --git a/src/mesa/drivers/dri/radeon/server/radeon_reg.h b/src/mesa/drivers/dri/radeon/server/radeon_reg.h
index 2cd9dbe094f..5570a439458 100644
--- a/src/mesa/drivers/dri/radeon/server/radeon_reg.h
+++ b/src/mesa/drivers/dri/radeon/server/radeon_reg.h
@@ -1,24 +1,4 @@
-/**
- * \file server/radeon_reg.h
- * \brief Registers and register definitions for the Radeon.
- * 
- * \authors Kevin E. Martin <[email protected]>
- * \authors Rickard E. Faith <[email protected]>
- * \authors Alan Hourihane <[email protected]>
- *
- * \par References
- *
- * - RAGE 128 VR/ RAGE 128 GL Register Reference Manual (Technical
- *   Reference Manual P/N RRG-G04100-C Rev. 0.04), ATI Technologies: April
- *   1999.
- * - RAGE 128 Software Development Manual (Technical Reference Manual P/N
- *   SDK-G04000 Rev. 0.01), ATI Technologies: June 1999.
- *
- * \note !!!! FIXME !!!! THIS FILE HAS BEEN CONVERTED FROM r128_reg.h
- * AND CONTAINS REGISTERS AND REGISTER DEFINITIONS THAT ARE NOT CORRECT
- * ON THE RADEON.  A FULL AUDIT OF THIS CODE IS NEEDED!
- */
-
+/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_reg.h,v 1.25 2003/02/07 18:08:59 martin Exp $ */
 /*
  * Copyright 2000 ATI Technologies Inc., Markham, Ontario, and
  *                VA Linux Systems Inc., Fremont, California.
@@ -47,7 +27,28 @@
  * DEALINGS IN THE SOFTWARE.
  */
 
-/* $XFree86: xc/programs/Xserver/hw/xfree86/drivers/ati/radeon_reg.h,v 1.20 2002/10/12 01:38:07 martin Exp $ */
+/*
+ * Authors:
+ *   Kevin E. Martin <[email protected]>
+ *   Rickard E. Faith <[email protected]>
+ *   Alan Hourihane <[email protected]>
+ *
+ * References:
+ *
+ * !!!! FIXME !!!!
+ *   RAGE 128 VR/ RAGE 128 GL Register Reference Manual (Technical
+ *   Reference Manual P/N RRG-G04100-C Rev. 0.04), ATI Technologies: April
+ *   1999.
+ *
+ * !!!! FIXME !!!!
+ *   RAGE 128 Software Development Manual (Technical Reference Manual P/N
+ *   SDK-G04000 Rev. 0.01), ATI Technologies: June 1999.
+ *
+ */
+
+/* !!!! FIXME !!!!  NOTE: THIS FILE HAS BEEN CONVERTED FROM r128_reg.h
+ * AND CONTAINS REGISTERS AND REGISTER DEFINITIONS THAT ARE NOT CORRECT
+ * ON THE RADEON.  A FULL AUDIT OF THIS CODE IS NEEDED!  */
 
 #ifndef _RADEON_REG_H_
 #define _RADEON_REG_H_
@@ -216,6 +217,10 @@
 #define RADEON_CONFIG_APER_SIZE             0x0108
 #define RADEON_CONFIG_BONDS                 0x00e8
 #define RADEON_CONFIG_CNTL                  0x00e0
+#       define RADEON_CFG_ATI_REV_A11       (0   << 16)
+#       define RADEON_CFG_ATI_REV_A12       (1   << 16)
+#       define RADEON_CFG_ATI_REV_A13       (2   << 16)
+#       define RADEON_CFG_ATI_REV_ID_MASK   (0xf << 16)
 #define RADEON_CONFIG_MEMSIZE               0x00f8
 #define RADEON_CONFIG_MEMSIZE_EMBEDDED      0x0114
 #define RADEON_CONFIG_REG_1_BASE            0x010c
@@ -302,6 +307,10 @@
 #define RADEON_CRTC2_PITCH                  0x032c
 #define RADEON_CRTC_STATUS                  0x005c
 #       define RADEON_CRTC_VBLANK_SAVE      (1 <<  1)
+#       define RADEON_CRTC_VBLANK_SAVE_CLEAR  (1 <<  1)
+#define RADEON_CRTC2_STATUS                  0x03fc
+#       define RADEON_CRTC2_VBLANK_SAVE      (1 <<  1)
+#       define RADEON_CRTC2_VBLANK_SAVE_CLEAR  (1 <<  1)
 #define RADEON_CRTC_V_SYNC_STRT_WID         0x020c
 #       define RADEON_CRTC_V_SYNC_STRT        (0x7ff <<  0)
 #       define RADEON_CRTC_V_SYNC_STRT_SHIFT  0
@@ -492,6 +501,7 @@
 #define RADEON_DST_LINE_START               0x1600
 #define RADEON_DST_LINE_END                 0x1604
 #define RADEON_DST_LINE_PATCOUNT            0x1608
+#       define RADEON_BRES_CNTL_SHIFT       8
 #define RADEON_DST_OFFSET                   0x1404
 #define RADEON_DST_PITCH                    0x1408
 #define RADEON_DST_PITCH_OFFSET             0x142c
@@ -554,6 +564,7 @@
 #define RADEON_FP_GEN_CNTL                  0x0284
 #       define RADEON_FP_FPON                  (1 <<  0)
 #       define RADEON_FP_TMDS_EN               (1 <<  2)
+#       define RADEON_FP_PANEL_FORMAT          (1 <<  3)
 #       define RADEON_FP_EN_TMDS               (1 <<  7)
 #       define RADEON_FP_DETECT_SENSE          (1 <<  8)
 #       define RADEON_FP_SEL_CRTC2             (1 << 13)
@@ -612,6 +623,8 @@
 #define RADEON_GEN_INT_STATUS               0x0044
 #       define RADEON_VSYNC_INT_AK          (1 <<  2)
 #       define RADEON_VSYNC_INT             (1 <<  2)
+#       define RADEON_VSYNC2_INT_AK         (1 <<  6)
+#       define RADEON_VSYNC2_INT            (1 <<  6)
 #define RADEON_GENENB                       0x03c3 /* VGA */
 #define RADEON_GENFC_RD                     0x03ca /* VGA */
 #define RADEON_GENFC_WT                     0x03da /* VGA, 0x03ba */
@@ -708,6 +721,9 @@
 #define RADEON_MM_DATA                      0x0004
 #define RADEON_MM_INDEX                     0x0000
 #define RADEON_MPLL_CNTL                    0x000e /* PLL */
+#define RADEON_MPP_TB_CONFIG                0x01c0 /* ? */
+#define RADEON_MPP_GP_CONFIG                0x01c8 /* ? */
+
 
 #define RADEON_N_VIF_COUNT                  0x0248
 
@@ -863,6 +879,8 @@
 #       define RADEON_P2PLL_REF_DIV_MASK    0x03ff
 #       define RADEON_P2PLL_ATOMIC_UPDATE_R (1 << 15) /* same as _W */
 #       define RADEON_P2PLL_ATOMIC_UPDATE_W (1 << 15) /* same as _R */
+#       define R300_PPLL_REF_DIV_ACC_MASK   (0x3ff < 18)
+#       define R300_PPLL_REF_DIV_ACC_SHIFT  18
 #define RADEON_PALETTE_DATA                 0x00b4
 #define RADEON_PALETTE_30_DATA              0x00b8
 #define RADEON_PALETTE_INDEX                0x00b0
@@ -1122,6 +1140,10 @@
 #       define RADEON_LOD_BIAS_SHIFT                       8
 #       define RADEON_MAX_MIP_LEVEL_MASK                   (0x0f << 16)
 #       define RADEON_MAX_MIP_LEVEL_SHIFT                  16
+#       define RADEON_YUV_TO_RGB                           (1  << 20)
+#       define RADEON_YUV_TEMPERATURE_COOL                 (0  << 21)
+#       define RADEON_YUV_TEMPERATURE_HOT                  (1  << 21)
+#       define RADEON_YUV_TEMPERATURE_MASK                 (1  << 21)
 #       define RADEON_WRAPEN_S                             (1  << 22)
 #       define RADEON_CLAMP_S_WRAP                         (0  << 23)
 #       define RADEON_CLAMP_S_MIRROR                       (1  << 23)
@@ -1129,6 +1151,8 @@
 #       define RADEON_CLAMP_S_MIRROR_CLAMP_LAST            (3  << 23)
 #       define RADEON_CLAMP_S_CLAMP_BORDER                 (4  << 23)
 #       define RADEON_CLAMP_S_MIRROR_CLAMP_BORDER          (5  << 23)
+#       define RADEON_CLAMP_S_CLAMP_GL                     (6  << 23)
+#       define RADEON_CLAMP_S_MIRROR_CLAMP_GL              (7  << 23)
 #       define RADEON_CLAMP_S_MASK                         (7  << 23)
 #       define RADEON_WRAPEN_T                             (1  << 26)
 #       define RADEON_CLAMP_T_WRAP                         (0  << 27)
@@ -1137,6 +1161,8 @@
 #       define RADEON_CLAMP_T_MIRROR_CLAMP_LAST            (3  << 27)
 #       define RADEON_CLAMP_T_CLAMP_BORDER                 (4  << 27)
 #       define RADEON_CLAMP_T_MIRROR_CLAMP_BORDER          (5  << 27)
+#       define RADEON_CLAMP_T_CLAMP_GL                     (6  << 27)
+#       define RADEON_CLAMP_T_MIRROR_CLAMP_GL              (7  << 27)
 #       define RADEON_CLAMP_T_MASK                         (7  << 27)
 #       define RADEON_BORDER_MODE_OGL                      (0  << 31)
 #       define RADEON_BORDER_MODE_D3D                      (1  << 31)
@@ -1152,6 +1178,11 @@
 #       define RADEON_TXFORMAT_ARGB8888           (6  <<  0)
 #       define RADEON_TXFORMAT_RGBA8888           (7  <<  0)
 #       define RADEON_TXFORMAT_Y8                 (8  <<  0)
+#       define RADEON_TXFORMAT_VYUY422            (10 <<  0)
+#       define RADEON_TXFORMAT_YVYU422            (11 <<  0)
+#       define RADEON_TXFORMAT_DXT1               (12 <<  0)
+#       define RADEON_TXFORMAT_DXT23              (14 <<  0)
+#       define RADEON_TXFORMAT_DXT45              (15 <<  0)
 #       define RADEON_TXFORMAT_FORMAT_MASK        (31 <<  0)
 #       define RADEON_TXFORMAT_FORMAT_SHIFT       0
 #       define RADEON_TXFORMAT_APPLE_YUV_MODE     (1  <<  5)
@@ -1161,6 +1192,10 @@
 #       define RADEON_TXFORMAT_WIDTH_SHIFT        8
 #       define RADEON_TXFORMAT_HEIGHT_MASK        (15 << 12)
 #       define RADEON_TXFORMAT_HEIGHT_SHIFT       12
+#       define RADEON_TXFORMAT_F5_WIDTH_MASK      (15 << 16)
+#       define RADEON_TXFORMAT_F5_WIDTH_SHIFT     16
+#       define RADEON_TXFORMAT_F5_HEIGHT_MASK     (15 << 20)
+#       define RADEON_TXFORMAT_F5_HEIGHT_SHIFT    20
 #       define RADEON_TXFORMAT_ST_ROUTE_STQ0      (0  << 24)
 #       define RADEON_TXFORMAT_ST_ROUTE_MASK      (3  << 24)
 #       define RADEON_TXFORMAT_ST_ROUTE_STQ1      (1  << 24)
@@ -1173,6 +1208,26 @@
 #       define RADEON_TXFORMAT_CHROMA_KEY_ENABLE  (1  << 29)
 #       define RADEON_TXFORMAT_CUBIC_MAP_ENABLE   (1  << 30)
 #       define RADEON_TXFORMAT_PERSPECTIVE_ENABLE (1  << 31)
+#define RADEON_PP_CUBIC_FACES_0             0x1d24
+#define RADEON_PP_CUBIC_FACES_1             0x1d28
+#define RADEON_PP_CUBIC_FACES_2             0x1d2c
+#       define RADEON_FACE_WIDTH_1_SHIFT          0
+#       define RADEON_FACE_HEIGHT_1_SHIFT         4
+#       define RADEON_FACE_WIDTH_1_MASK           (0xf << 0)
+#       define RADEON_FACE_HEIGHT_1_MASK          (0xf << 4)
+#       define RADEON_FACE_WIDTH_2_SHIFT          8
+#       define RADEON_FACE_HEIGHT_2_SHIFT         12
+#       define RADEON_FACE_WIDTH_2_MASK           (0xf << 8)
+#       define RADEON_FACE_HEIGHT_2_MASK          (0xf << 12)
+#       define RADEON_FACE_WIDTH_3_SHIFT          16
+#       define RADEON_FACE_HEIGHT_3_SHIFT         20
+#       define RADEON_FACE_WIDTH_3_MASK           (0xf << 16)
+#       define RADEON_FACE_HEIGHT_3_MASK          (0xf << 20)
+#       define RADEON_FACE_WIDTH_4_SHIFT          24
+#       define RADEON_FACE_HEIGHT_4_SHIFT         28
+#       define RADEON_FACE_WIDTH_4_MASK           (0xf << 24)
+#       define RADEON_FACE_HEIGHT_4_MASK          (0xf << 28)
+
 #define RADEON_PP_TXOFFSET_0                0x1c5c
 #define RADEON_PP_TXOFFSET_1                0x1c74
 #define RADEON_PP_TXOFFSET_2                0x1c8c
@@ -1187,6 +1242,39 @@
 #       define RADEON_TXO_MICRO_TILE_OPT     (2 << 3)
 #       define RADEON_TXO_OFFSET_MASK        0xffffffe0
 #       define RADEON_TXO_OFFSET_SHIFT       5
+
+#define RADEON_PP_CUBIC_OFFSET_T0_0         0x1dd0  /* bits [31:5] */
+#define RADEON_PP_CUBIC_OFFSET_T0_1         0x1dd4
+#define RADEON_PP_CUBIC_OFFSET_T0_2         0x1dd8
+#define RADEON_PP_CUBIC_OFFSET_T0_3         0x1ddc
+#define RADEON_PP_CUBIC_OFFSET_T0_4         0x1de0
+#define RADEON_PP_CUBIC_OFFSET_T1_0         0x1e00
+#define RADEON_PP_CUBIC_OFFSET_T1_1         0x1e04
+#define RADEON_PP_CUBIC_OFFSET_T1_2         0x1e08
+#define RADEON_PP_CUBIC_OFFSET_T1_3         0x1e0c
+#define RADEON_PP_CUBIC_OFFSET_T1_4         0x1e10
+#define RADEON_PP_CUBIC_OFFSET_T2_0         0x1e14
+#define RADEON_PP_CUBIC_OFFSET_T2_1         0x1e18
+#define RADEON_PP_CUBIC_OFFSET_T2_2         0x1e1c
+#define RADEON_PP_CUBIC_OFFSET_T2_3         0x1e20
+#define RADEON_PP_CUBIC_OFFSET_T2_4         0x1e24
+
+#define RADEON_PP_TEX_SIZE_0                0x1d04  /* NPOT */
+#define RADEON_PP_TEX_SIZE_1                0x1d0c
+#define RADEON_PP_TEX_SIZE_2                0x1d14
+#       define RADEON_TEX_USIZE_MASK        (0x7ff << 0)
+#       define RADEON_TEX_USIZE_SHIFT       0
+#       define RADEON_TEX_VSIZE_MASK        (0x7ff << 16)
+#       define RADEON_TEX_VSIZE_SHIFT       16
+#       define RADEON_SIGNED_RGB_MASK       (1 << 30)
+#       define RADEON_SIGNED_RGB_SHIFT      30
+#       define RADEON_SIGNED_ALPHA_MASK     (1 << 31)
+#       define RADEON_SIGNED_ALPHA_SHIFT    31
+#define RADEON_PP_TEX_PITCH_0               0x1d08  /* NPOT */
+#define RADEON_PP_TEX_PITCH_1               0x1d10  /* NPOT */
+#define RADEON_PP_TEX_PITCH_2               0x1d18  /* NPOT */
+/* note: bits 13-5: 32 byte aligned stride of texture map */
+
 #define RADEON_PP_TXCBLEND_0                0x1c60
 #define RADEON_PP_TXCBLEND_1                0x1c78
 #define RADEON_PP_TXCBLEND_2                0x1c90
diff --git a/src/mesa/drivers/dri/radeon/server/radeon_sarea.h b/src/mesa/drivers/dri/radeon/server/radeon_sarea.h
index f682bb6b6a6..81e4325d7a1 100644
--- a/src/mesa/drivers/dri/radeon/server/radeon_sarea.h
+++ b/src/mesa/drivers/dri/radeon/server/radeon_sarea.h
@@ -296,9 +296,9 @@ typedef struct {
     /** \brief Texture regions. 
      * Last element is sentinal
      */
-    radeon_tex_region_t texList[RADEON_NR_TEX_HEAPS][RADEON_NR_TEX_REGIONS+1];
+    drmTextureRegion texList[RADEON_NR_TEX_HEAPS][RADEON_NR_TEX_REGIONS+1];
     /** \brief last time texture was uploaded */
-    int texAge[RADEON_NR_TEX_HEAPS];
+    unsigned int texAge[RADEON_NR_TEX_HEAPS];
     /*@}*/
 
     int ctxOwner;		/**< \brief last context to upload state */
diff --git a/src/mesa/tnl_dd/imm/NOTES.imm b/src/mesa/tnl_dd/imm/NOTES.imm
new file mode 100644
index 00000000000..9b2bd65e4bf
--- /dev/null
+++ b/src/mesa/tnl_dd/imm/NOTES.imm
@@ -0,0 +1,112 @@
+
+NOTE:
+
+These files are incomplete.  They do not yet form a working
+implementation of hte concepts discused below.
+
+
+OVERVIEW
+
+The t_dd_imm_* files form a set of templates to produce driver -
+specific software tnl modules for a small subset of transformation and
+lighting states.
+
+The approach is quite different to the large vertex buffers of the
+src/tnl module, and is based around a cache of four recent vertices
+and a 'current' vertex which is updated directly from the Color,
+Normal, Texcoord, SecondaryColor and Fog entrypoints.
+
+The current vertex is actually a composite of the ctx->Current values
+and a partial hardware vertex maintained where the hardware values
+differ from those in ctx->Current.  For example, clamped color values
+are kept in the hardware vertex, while texcoords remain in
+ctx->Current.
+
+A crude diagram:
+
+       		+--------------+	+-------------------+
+		| ctx->Current |	| Current-HW-vertex |
+		+--------------+	+-------------------+
+		       \       	       	       	  /
+		       	\      			 /
+			 \     			/
+			  \    		       /
+			   ---------   --------
+			       	   |   |      
+			       	   v   v     
+	+--------+   +--------+  +--------+  +--------+ 
+	| vert-0 |   | vert-1 |  | vert-2 |  | vert-3 |		       	 
+	+--------+   +--------+  +--------+  +--------+		       	 
+				     |
+				     |
+				     v
+				     
+				    DMA
+
+
+Here values from ctx->Current and current-HW-vertex are merged to
+build vert-2, which is then dumped to hardware (DMA).  A state machine
+determines which vertex is built in turn, and how the vertices are
+used to present primitives to hardware.  These actions all occur
+during a call to Vertex{234}f{v}.
+
+Each vert-n includes clip coordinates and a clipmask in addition to
+the hardware (window) coordinates.  This information allows clipping
+to take place directly on these vertices, if need be.
+
+t_dd_imm_capi.h
+	
+	Color{34}{fub}{v}() implementations.  These update both
+	ctx->Current (unclamped float colors) and current-HW-vertex
+	with hardware-specific color values (typically unsigned
+	bytes).
+
+	When lighting is enabled, the functions from src/api_noop.c
+	should be used, which just update ctx->Current.  (The
+	current-hw-vertex colors are produced from lighting, which is
+	keyed to Normal3f).
+
+t_dd_imm_vb.c
+
+	Support functions for clipping and fallback.  See
+	t_dd_imm_primtmp.h.
+
+t_dd_imm_napi.c
+t_dd_imm_napi.h
+
+	Versions of Normal3f{v} to perform lighting with one or more
+	infinite lights.  Updates ctx->Current.Normal and the current
+	HW colors.
+
+	When lighting is disabled, use the functions from api_noop.c
+	instead.
+
+
+t_dd_imm_primtmp.h
+
+	State machine to control emission of vertices and primitives
+	to hardware.  Called indirectly from Vertex{234}f{v}.  Capable
+	of supporting hardware strip and fan primitives, and of
+	decomposing to discreet primitives for clipping or fallback,
+	or where the native primitive is unavailable.
+
+t_dd_imm_tapi.h
+
+	Implementations of TexCoord{v} and MultiTexCoord4f{v}ARB to
+	fire a callback when transitioning to projective texture.
+	Most drivers will need to change vertex format at this point,
+	some may need to enable a software rasterization fallback.
+
+t_dd_imm_vapi.h
+
+	Implementations of Vertex{234}f{v}.  These perform
+	transformation and cliptesting on their arguments, then jump
+	into the state machine implemented in primtmp.h.
+
+t_dd_imm_vertex.h
+
+	Support functions for building and clip-interpolating hardware
+	vertices.  Called from primtmp.h.
+
+
+Keith Whitwell, June 2001.
\ No newline at end of file
diff --git a/src/mesa/tnl_dd/imm/t_dd_imm_capi.h b/src/mesa/tnl_dd/imm/t_dd_imm_capi.h
new file mode 100644
index 00000000000..77f3a4ab497
--- /dev/null
+++ b/src/mesa/tnl_dd/imm/t_dd_imm_capi.h
@@ -0,0 +1,419 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999-2001  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:
+ *    Gareth Hughes <[email protected]>
+ */
+
+/* Template for immediate mode color functions.
+ *
+ * FIXME: Floating-point color versions of these...
+ */
+
+
+static void TAG(Color3f)( GLfloat r, GLfloat g, GLfloat b )
+{
+   GET_CURRENT;
+#ifdef COLOR_IS_FLOAT
+   CURRENT_COLOR( RCOMP ) = CLAMP(r, 0.0f, 1.0f);
+   CURRENT_COLOR( GCOMP ) = CLAMP(g, 0.0f, 1.0f);
+   CURRENT_COLOR( BCOMP ) = CLAMP(b, 0.0f, 1.0f);
+   CURRENT_COLOR( ACOMP ) = 1.0f;
+#else
+   UNCLAMPED_FLOAT_TO_UBYTE( CURRENT_COLOR( RCOMP ), r );
+   UNCLAMPED_FLOAT_TO_UBYTE( CURRENT_COLOR( GCOMP ), g );
+   UNCLAMPED_FLOAT_TO_UBYTE( CURRENT_COLOR( BCOMP ), b );
+   CURRENT_COLOR( ACOMP ) = 255;
+#endif
+}
+
+static void TAG(Color3fv)( const GLfloat *v )
+{
+   GET_CURRENT;
+#ifdef COLOR_IS_FLOAT
+   CURRENT_COLOR( RCOMP ) = CLAMP(v[0], 0.0f, 1.0f);
+   CURRENT_COLOR( GCOMP ) = CLAMP(v[1], 0.0f, 1.0f);
+   CURRENT_COLOR( BCOMP ) = CLAMP(v[2], 0.0f, 1.0f);
+   CURRENT_COLOR( ACOMP ) = 1.0f;
+#else
+   UNCLAMPED_FLOAT_TO_UBYTE( CURRENT_COLOR( RCOMP ), v[0] );
+   UNCLAMPED_FLOAT_TO_UBYTE( CURRENT_COLOR( GCOMP ), v[1] );
+   UNCLAMPED_FLOAT_TO_UBYTE( CURRENT_COLOR( BCOMP ), v[2] );
+   CURRENT_COLOR( ACOMP ) = 255;
+#endif
+}
+
+static void TAG(Color3ub)( GLubyte r, GLubyte g, GLubyte b )
+{
+   GET_CURRENT;
+#ifdef COLOR_IS_FLOAT
+   CURRENT_COLOR( RCOMP ) = UBYTE_TO_FLOAT( r );
+   CURRENT_COLOR( GCOMP ) = UBYTE_TO_FLOAT( g );
+   CURRENT_COLOR( BCOMP ) = UBYTE_TO_FLOAT( b );
+   CURRENT_COLOR( ACOMP ) = 1.0f;
+#else
+   CURRENT_COLOR( RCOMP ) = r;
+   CURRENT_COLOR( GCOMP ) = g;
+   CURRENT_COLOR( BCOMP ) = b;
+   CURRENT_COLOR( ACOMP ) = 255;
+#endif
+}
+
+static void TAG(Color3ubv)( const GLubyte *v )
+{
+   GET_CURRENT;
+#ifdef COLOR_IS_FLOAT
+   CURRENT_COLOR( RCOMP ) = UBYTE_TO_FLOAT( v[0] );
+   CURRENT_COLOR( GCOMP ) = UBYTE_TO_FLOAT( v[1] );
+   CURRENT_COLOR( BCOMP ) = UBYTE_TO_FLOAT( v[2] );
+   CURRENT_COLOR( ACOMP ) = 1.0f;
+#else
+   CURRENT_COLOR( RCOMP ) = v[0];
+   CURRENT_COLOR( GCOMP ) = v[1];
+   CURRENT_COLOR( BCOMP ) = v[2];
+   CURRENT_COLOR( ACOMP ) = 255;
+#endif
+}
+
+static void TAG(Color4f)( GLfloat r, GLfloat g, GLfloat b, GLfloat a )
+{
+   GET_CURRENT;
+#ifdef COLOR_IS_FLOAT
+   CURRENT_COLOR( RCOMP ) = CLAMP(r, 0.0f, 1.0f);
+   CURRENT_COLOR( GCOMP ) = CLAMP(g, 0.0f, 1.0f);
+   CURRENT_COLOR( BCOMP ) = CLAMP(b, 0.0f, 1.0f);
+   CURRENT_COLOR( ACOMP ) = CLAMP(a, 0.0f, 1.0f);
+#else
+   UNCLAMPED_FLOAT_TO_UBYTE( CURRENT_COLOR( RCOMP ), r );
+   UNCLAMPED_FLOAT_TO_UBYTE( CURRENT_COLOR( GCOMP ), g );
+   UNCLAMPED_FLOAT_TO_UBYTE( CURRENT_COLOR( BCOMP ), b );
+   UNCLAMPED_FLOAT_TO_UBYTE( CURRENT_COLOR( ACOMP ), a );
+#endif
+}
+
+static void TAG(Color4fv)( const GLfloat *v )
+{
+   GET_CURRENT;
+#ifdef COLOR_IS_FLOAT
+   CURRENT_COLOR( RCOMP ) = CLAMP(v[0], 0.0f, 1.0f);
+   CURRENT_COLOR( GCOMP ) = CLAMP(v[1], 0.0f, 1.0f);
+   CURRENT_COLOR( BCOMP ) = CLAMP(v[2], 0.0f, 1.0f);
+   CURRENT_COLOR( ACOMP ) = CLAMP(v[3], 0.0f, 1.0f);
+#else
+   UNCLAMPED_FLOAT_TO_UBYTE( CURRENT_COLOR( RCOMP ), v[0] );
+   UNCLAMPED_FLOAT_TO_UBYTE( CURRENT_COLOR( GCOMP ), v[1] );
+   UNCLAMPED_FLOAT_TO_UBYTE( CURRENT_COLOR( BCOMP ), v[2] );
+   UNCLAMPED_FLOAT_TO_UBYTE( CURRENT_COLOR( ACOMP ), v[3] );
+#endif
+}
+
+static void TAG(Color4ub)( GLubyte r, GLubyte g, GLubyte b, GLubyte a )
+{
+   GET_CURRENT;
+#ifdef COLOR_IS_FLOAT
+   CURRENT_COLOR( RCOMP ) = UBYTE_TO_FLOAT( r );
+   CURRENT_COLOR( GCOMP ) = UBYTE_TO_FLOAT( g );
+   CURRENT_COLOR( BCOMP ) = UBYTE_TO_FLOAT( b );
+   CURRENT_COLOR( ACOMP ) = UBYTE_TO_FLOAT( a );
+#else
+   CURRENT_COLOR( RCOMP ) = r;
+   CURRENT_COLOR( GCOMP ) = g;
+   CURRENT_COLOR( BCOMP ) = b;
+   CURRENT_COLOR( ACOMP ) = a;
+#endif
+}
+
+static void TAG(Color4ubv)( const GLubyte *v )
+{
+   GET_CURRENT;
+#ifdef COLOR_IS_FLOAT
+   CURRENT_COLOR( RCOMP ) = UBYTE_TO_FLOAT( v[0] );
+   CURRENT_COLOR( GCOMP ) = UBYTE_TO_FLOAT( v[1] );
+   CURRENT_COLOR( BCOMP ) = UBYTE_TO_FLOAT( v[2] );
+   CURRENT_COLOR( ACOMP ) = UBYTE_TO_FLOAT( v[3] );
+#else
+   CURRENT_COLOR( RCOMP ) = v[0];
+   CURRENT_COLOR( GCOMP ) = v[1];
+   CURRENT_COLOR( BCOMP ) = v[2];
+   CURRENT_COLOR( ACOMP ) = v[3];
+#endif
+}
+
+
+static void TAG(ColorMaterial3f)( GLfloat r, GLfloat g, GLfloat b )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   GLfloat *color = ctx->Current.Color;
+
+   color[0] = r;
+   color[1] = g;
+   color[2] = b;
+   color[3] = 1.0;
+
+   _mesa_update_color_material( ctx, color );
+   RECALC_BASE_COLOR( ctx );
+}
+
+static void TAG(ColorMaterial3fv)( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   GLfloat *color = ctx->Current.Color;
+
+   color[0] = v[0];
+   color[1] = v[1];
+   color[2] = v[2];
+   color[3] = 1.0;
+
+   _mesa_update_color_material( ctx, color );
+   RECALC_BASE_COLOR( ctx );
+}
+
+static void TAG(ColorMaterial3ub)( GLubyte r, GLubyte g, GLubyte b )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   GLfloat *color = ctx->Current.Color;
+
+   color[0] = UBYTE_TO_FLOAT( r );
+   color[1] = UBYTE_TO_FLOAT( g );
+   color[2] = UBYTE_TO_FLOAT( b );
+   color[3] = 1.0;
+
+   _mesa_update_color_material( ctx, color );
+   RECALC_BASE_COLOR( ctx );
+}
+
+static void TAG(ColorMaterial3ubv)( const GLubyte *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   GLfloat *color = ctx->Current.Color;
+
+   color[0] = UBYTE_TO_FLOAT( v[0] );
+   color[1] = UBYTE_TO_FLOAT( v[1] );
+   color[2] = UBYTE_TO_FLOAT( v[2] );
+   color[3] = 1.0;
+
+   _mesa_update_color_material( ctx, color );
+   RECALC_BASE_COLOR( ctx );
+}
+
+static void TAG(ColorMaterial4f)( GLfloat r, GLfloat g, GLfloat b, GLfloat a )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   GLfloat *color = ctx->Current.Color;
+
+   color[0] = r;
+   color[1] = g;
+   color[2] = b;
+   color[3] = a;
+
+   _mesa_update_color_material( ctx, color );
+   RECALC_BASE_COLOR( ctx );
+}
+
+static void TAG(ColorMaterial4fv)( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   GLfloat *color = ctx->Current.Color;
+
+   color[0] = v[0];
+   color[1] = v[1];
+   color[2] = v[2];
+   color[3] = v[3];
+
+   _mesa_update_color_material( ctx, color );
+   RECALC_BASE_COLOR( ctx );
+}
+
+static void TAG(ColorMaterial4ub)( GLubyte r, GLubyte g, GLubyte b, GLubyte a )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   GLfloat *color = ctx->Current.Color;
+
+   color[0] = UBYTE_TO_FLOAT( r );
+   color[1] = UBYTE_TO_FLOAT( g );
+   color[2] = UBYTE_TO_FLOAT( b );
+   color[3] = UBYTE_TO_FLOAT( a );
+
+   _mesa_update_color_material( ctx, color );
+   RECALC_BASE_COLOR( ctx );
+}
+
+static void TAG(ColorMaterial4ubv)( const GLubyte *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+   GLfloat *color = ctx->Current.Color;
+
+   color[0] = UBYTE_TO_FLOAT( v[0] );
+   color[1] = UBYTE_TO_FLOAT( v[1] );
+   color[2] = UBYTE_TO_FLOAT( v[2] );
+   color[3] = UBYTE_TO_FLOAT( v[3] );
+
+   _mesa_update_color_material( ctx, color );
+   RECALC_BASE_COLOR( ctx );
+}
+
+
+
+
+
+/* =============================================================
+ * Color chooser functions:
+ */
+
+static void TAG(choose_Color3f)( GLfloat r, GLfloat g, GLfloat b )
+{
+   GET_CURRENT_CONTEXT(ctx);
+
+   if ( ctx->Light.Enabled ) {
+      if ( ctx->Light.ColorMaterialEnabled ) {
+	 ctx->Exec->Color3f = TAG(ColorMaterial3f);
+      } else {
+	 ctx->Exec->Color3f = _mesa_noop_Color3f;
+      }
+   } else {
+      ctx->Exec->Color3f = TAG(Color3f);
+   }
+   glColor3f( r, g, b );
+}
+
+static void TAG(choose_Color3fv)( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+
+   if ( ctx->Light.Enabled ) {
+      if ( ctx->Light.ColorMaterialEnabled ) {
+	 ctx->Exec->Color3fv = TAG(ColorMaterial3fv);
+      } else {
+	 ctx->Exec->Color3fv = _mesa_noop_Color3fv;
+      }
+   } else {
+      ctx->Exec->Color3fv = TAG(Color3fv);
+   }
+   glColor3fv( v );
+}
+
+static void TAG(choose_Color3ub)( GLubyte r, GLubyte g, GLubyte b )
+{
+   GET_CURRENT_CONTEXT(ctx);
+
+   if ( ctx->Light.Enabled ) {
+      if ( ctx->Light.ColorMaterialEnabled ) {
+	 ctx->Exec->Color3ub = TAG(ColorMaterial3ub);
+      } else {
+	 ctx->Exec->Color3ub = _mesa_noop_Color3ub;
+      }
+   } else {
+      ctx->Exec->Color3ub = TAG(Color3ub);
+   }
+   glColor3ub( r, g, b );
+}
+
+static void TAG(choose_Color3ubv)( const GLubyte *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+
+   if ( ctx->Light.Enabled ) {
+      if ( ctx->Light.ColorMaterialEnabled ) {
+	 ctx->Exec->Color3ubv = TAG(ColorMaterial3ubv);
+      } else {
+	 ctx->Exec->Color3ubv = _mesa_noop_Color3ubv;
+      }
+   } else {
+      ctx->Exec->Color3ubv = TAG(Color3ubv);
+   }
+   glColor3ubv( v );
+}
+
+static void TAG(choose_Color4f)( GLfloat r, GLfloat g, GLfloat b, GLfloat a )
+{
+   GET_CURRENT_CONTEXT(ctx);
+
+   if ( ctx->Light.Enabled ) {
+      if ( ctx->Light.ColorMaterialEnabled ) {
+	 ctx->Exec->Color4f = TAG(ColorMaterial4f);
+      } else {
+	 ctx->Exec->Color4f = _mesa_noop_Color4f;
+      }
+   } else {
+      ctx->Exec->Color4f = TAG(Color4f);
+   }
+   glColor4f( r, g, b, a );
+}
+
+static void TAG(choose_Color4fv)( const GLfloat *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+
+   if ( ctx->Light.Enabled ) {
+      if ( ctx->Light.ColorMaterialEnabled ) {
+	 ctx->Exec->Color4fv = TAG(ColorMaterial4fv);
+      } else {
+	 ctx->Exec->Color4fv = _mesa_noop_Color4fv;
+      }
+   } else {
+      ctx->Exec->Color4fv = TAG(Color4fv);
+   }
+   glColor4fv( v );
+}
+
+static void TAG(choose_Color4ub)( GLubyte r, GLubyte g, GLubyte b, GLubyte a )
+{
+   GET_CURRENT_CONTEXT(ctx);
+
+   if ( ctx->Light.Enabled ) {
+      if ( ctx->Light.ColorMaterialEnabled ) {
+	 ctx->Exec->Color4ub = TAG(ColorMaterial4ub);
+      } else {
+	 ctx->Exec->Color4ub = _mesa_noop_Color4ub;
+      }
+   } else {
+      ctx->Exec->Color4ub = TAG(Color4ub);
+   }
+   glColor4ub( r, g, b, a );
+}
+
+static void TAG(choose_Color4ubv)( const GLubyte *v )
+{
+   GET_CURRENT_CONTEXT(ctx);
+
+   if ( ctx->Light.Enabled ) {
+      if ( ctx->Light.ColorMaterialEnabled ) {
+	 ctx->Exec->Color4ubv = TAG(ColorMaterial4ubv);
+      } else {
+	 ctx->Exec->Color4ubv = _mesa_noop_Color4ubv;
+      }
+   } else {
+      ctx->Exec->Color4ubv = TAG(Color4ubv);
+   }
+   glColor4ubv( v );
+}
+
+
+
+#undef GET_CURRENT
+#undef CURRENT_COLOR
+#undef CURRENT_SPECULAR
+#undef COLOR_IS_FLOAT
+#undef RECALC_BASE_COLOR
+#undef TAG
diff --git a/src/mesa/tnl_dd/imm/t_dd_imm_napi.h b/src/mesa/tnl_dd/imm/t_dd_imm_napi.h
new file mode 100644
index 00000000000..9844f615ffe
--- /dev/null
+++ b/src/mesa/tnl_dd/imm/t_dd_imm_napi.h
@@ -0,0 +1,226 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999-2001  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:
+ *    Gareth Hughes <[email protected]>
+ *    Keith Whitwell <[email protected]>
+ */
+
+/* Template for immediate mode normal functions.  Optimize for infinite
+ * lights when doing software lighting.
+ */
+
+static void TAG(Normal3f_single)( GLfloat x, GLfloat y, GLfloat z )
+{
+   GET_CURRENT_VERTEX;
+   const struct gl_light *light = ctx->Light.EnabledList.prev;
+   GLfloat n_dot_h, n_dot_VP, spec, sum[3];
+   GLfloat *normal = ctx->Current.Normal;
+   GLfloat scale = 1.0;
+
+   ASSIGN_3V( normal, x, y, z );
+   COPY_3V( sum, BASE_COLOR );
+
+   if ( IND & NORM_RESCALE ) {
+      scale = ctx->_ModelViewInvScale;
+   } else if ( IND & NORM_NORMALIZE ) {
+      scale = LEN_3FV( normal );
+      if ( scale != 0.0 ) scale = 1.0 / scale;
+   }
+
+   n_dot_VP = DOT3( normal, light->_VP_inf_norm ) * scale;
+   if ( n_dot_VP > 0.0F ) {
+      ACC_SCALE_SCALAR_3V( sum, n_dot_VP, light->_MatDiffuse[0] );
+      n_dot_h = DOT3( normal, light->_h_inf_norm ) * scale;
+      if ( n_dot_h > 0.0F ) {
+	 GET_SHINE_TAB_ENTRY( ctx->_ShineTable[0], n_dot_h, spec );
+	 ACC_SCALE_SCALAR_3V( sum, spec, light->_MatSpecular[0] );
+      }
+   }
+
+#ifdef LIT_COLOR_IS_FLOAT
+   LIT_COLOR ( RCOMP ) = CLAMP(sum[0], 0.0f, 0.1f);
+   LIT_COLOR ( GCOMP ) = CLAMP(sum[1], 0.0f, 0.1f);
+   LIT_COLOR ( BCOMP ) = CLAMP(sum[2], 0.0f, 0.1f);
+#else
+   UNCLAMPED_FLOAT_TO_UBYTE( LIT_COLOR( RCOMP ), sum[0] );
+   UNCLAMPED_FLOAT_TO_UBYTE( LIT_COLOR( GCOMP ), sum[1] );
+   UNCLAMPED_FLOAT_TO_UBYTE( LIT_COLOR( BCOMP ), sum[2] );
+#endif
+   LIT_COLOR( ACOMP ) = LIT_ALPHA;
+}
+
+static void TAG(Normal3fv_single)( const GLfloat *normal )
+{
+   GET_CURRENT_VERTEX;
+   const struct gl_light *light = ctx->Light.EnabledList.prev;
+   GLfloat n_dot_h, n_dot_VP, spec, sum[3];
+   GLfloat scale = 1.0;
+
+   COPY_3V( ctx->Current.Normal, normal );
+   COPY_3V( sum, BASE_COLOR );
+
+   if ( IND & NORM_RESCALE ) {
+      scale = ctx->_ModelViewInvScale;
+   } else if ( IND & NORM_NORMALIZE ) {
+      scale = LEN_3FV( normal );
+      if ( scale != 0.0 ) scale = 1.0 / scale;
+   }
+
+   n_dot_VP = DOT3( normal, light->_VP_inf_norm ) * scale;
+   if ( n_dot_VP > 0.0F ) {
+      ACC_SCALE_SCALAR_3V( sum, n_dot_VP, light->_MatDiffuse[0] );
+      n_dot_h = DOT3( normal, light->_h_inf_norm ) * scale;
+      if ( n_dot_h > 0.0F ) {
+	 GET_SHINE_TAB_ENTRY( ctx->_ShineTable[0], n_dot_h, spec );
+	 ACC_SCALE_SCALAR_3V( sum, spec, light->_MatSpecular[0] );
+      }
+   }
+
+#ifdef LIT_COLOR_IS_FLOAT
+   LIT_COLOR ( RCOMP ) = CLAMP(sum[0], 0.0f, 0.1f);
+   LIT_COLOR ( GCOMP ) = CLAMP(sum[1], 0.0f, 0.1f);
+   LIT_COLOR ( BCOMP ) = CLAMP(sum[2], 0.0f, 0.1f);
+#else
+   UNCLAMPED_FLOAT_TO_UBYTE( LIT_COLOR( RCOMP ), sum[0] );
+   UNCLAMPED_FLOAT_TO_UBYTE( LIT_COLOR( GCOMP ), sum[1] );
+   UNCLAMPED_FLOAT_TO_UBYTE( LIT_COLOR( BCOMP ), sum[2] );
+#endif
+   LIT_COLOR( ACOMP ) = LIT_ALPHA;
+}
+
+
+static void TAG(Normal3f_multi)( GLfloat x, GLfloat y, GLfloat z )
+{
+   GET_CURRENT_VERTEX;
+   struct gl_light *light;
+   GLfloat n_dot_h, n_dot_VP, spec, sum[3], tmp[3];
+   GLfloat *normal;
+
+   ASSIGN_3V( ctx->Current.Normal, x, y, z );
+   COPY_3V( sum, BASE_COLOR );
+
+   if ( IND & NORM_RESCALE ) {
+      normal = tmp;
+      ASSIGN_3V( normal, x, y, z );
+      SELF_SCALE_SCALAR_3V( normal, ctx->_ModelViewInvScale );
+   } else if ( IND & NORM_NORMALIZE ) {
+      normal = tmp;
+      ASSIGN_3V( normal, x, y, z );
+      NORMALIZE_3FV( normal );
+   } else {
+      normal = ctx->Current.Normal;
+   }
+
+   foreach ( light, &ctx->Light.EnabledList ) {
+      n_dot_VP = DOT3( normal, light->_VP_inf_norm );
+      if ( n_dot_VP > 0.0F ) {
+	 ACC_SCALE_SCALAR_3V( sum, n_dot_VP, light->_MatDiffuse[0] );
+	 n_dot_h = DOT3( normal, light->_h_inf_norm );
+	 if ( n_dot_h > 0.0F ) {
+	    GET_SHINE_TAB_ENTRY( ctx->_ShineTable[0], n_dot_h, spec );
+	    ACC_SCALE_SCALAR_3V( sum, spec, light->_MatSpecular[0] );
+	 }
+      }
+   }
+
+#ifdef LIT_COLOR_IS_FLOAT
+   LIT_COLOR ( RCOMP ) = CLAMP(sum[0], 0.0f, 0.1f);
+   LIT_COLOR ( GCOMP ) = CLAMP(sum[1], 0.0f, 0.1f);
+   LIT_COLOR ( BCOMP ) = CLAMP(sum[2], 0.0f, 0.1f);
+#else
+   UNCLAMPED_FLOAT_TO_UBYTE( LIT_COLOR( RCOMP ), sum[0] );
+   UNCLAMPED_FLOAT_TO_UBYTE( LIT_COLOR( GCOMP ), sum[1] );
+   UNCLAMPED_FLOAT_TO_UBYTE( LIT_COLOR( BCOMP ), sum[2] );
+#endif
+   LIT_COLOR( ACOMP ) = LIT_ALPHA;
+}
+
+static void TAG(Normal3fv_multi)( const GLfloat *n )
+{
+   GET_CURRENT_VERTEX;
+   struct gl_light *light;
+   GLfloat n_dot_h, n_dot_VP, spec, sum[3], tmp[3];
+   GLfloat *normal;
+
+   COPY_3V( ctx->Current.Normal, n );
+   COPY_3V( sum, BASE_COLOR );
+
+   if ( IND & NORM_RESCALE ) {
+      normal = tmp;
+      COPY_3V( normal, n );
+      SELF_SCALE_SCALAR_3V( normal, ctx->_ModelViewInvScale );
+   } else if ( IND & NORM_NORMALIZE ) {
+      normal = tmp;
+      COPY_3V( normal, n );
+      NORMALIZE_3FV( normal );
+   } else {
+      normal = ctx->Current.Normal;
+   }
+
+   foreach ( light, &ctx->Light.EnabledList ) {
+      n_dot_VP = DOT3( normal, light->_VP_inf_norm );
+      if ( n_dot_VP > 0.0F ) {
+	 ACC_SCALE_SCALAR_3V( sum, n_dot_VP, light->_MatDiffuse[0] );
+	 n_dot_h = DOT3( normal, light->_h_inf_norm );
+	 if ( n_dot_h > 0.0F ) {
+	    GET_SHINE_TAB_ENTRY( ctx->_ShineTable[0], n_dot_h, spec );
+	    ACC_SCALE_SCALAR_3V( sum, spec, light->_MatSpecular[0] );
+	 }
+      }
+   }
+
+#ifdef LIT_COLOR_IS_FLOAT
+   LIT_COLOR ( RCOMP ) = CLAMP(sum[0], 0.0f, 0.1f);
+   LIT_COLOR ( GCOMP ) = CLAMP(sum[1], 0.0f, 0.1f);
+   LIT_COLOR ( BCOMP ) = CLAMP(sum[2], 0.0f, 0.1f);
+#else
+   UNCLAMPED_FLOAT_TO_UBYTE( LIT_COLOR( RCOMP ), sum[0] );
+   UNCLAMPED_FLOAT_TO_UBYTE( LIT_COLOR( GCOMP ), sum[1] );
+   UNCLAMPED_FLOAT_TO_UBYTE( LIT_COLOR( BCOMP ), sum[2] );
+#endif
+   LIT_COLOR( ACOMP ) = LIT_ALPHA;
+}
+
+
+
+static void TAG(init_norm)( void )
+{
+   norm_tab[IND].normal3f_single = TAG(Normal3f_single);
+   norm_tab[IND].normal3fv_single = TAG(Normal3fv_single);
+   norm_tab[IND].normal3f_multi = TAG(Normal3f_multi);
+   norm_tab[IND].normal3fv_multi = TAG(Normal3fv_multi);
+}
+
+
+
+#ifndef PRESERVE_NORMAL_DEFS
+#undef GET_CURRENT
+#undef GET_CURRENT_VERTEX
+#undef LIT_COLOR
+#undef LIT_COLOR_IS_FLOAT
+#endif
+#undef PRESERVE_NORMAL_DEFS
+#undef IND
+#undef TAG
diff --git a/src/mesa/tnl_dd/imm/t_dd_imm_primtmp.h b/src/mesa/tnl_dd/imm/t_dd_imm_primtmp.h
new file mode 100644
index 00000000000..97dca3fd421
--- /dev/null
+++ b/src/mesa/tnl_dd/imm/t_dd_imm_primtmp.h
@@ -0,0 +1,570 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999-2001  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 <[email protected]>
+ *    Gareth Hughes <[email protected]>
+ */
+
+/* Template for immediate mode vertices. 
+ *
+ * Probably instantiate once for each vertex format used:
+ *   - TINY_VERTICES
+ *   - TEX0_VERTICES
+ *   - TEX1_VERTICES
+ *   - PTEX_VERTICES
+ *
+ * Have to handle TEX->PTEX transition somehow.
+ */
+
+#define DBG 0
+
+
+
+/* =============================================================
+ * GL_POINTS
+ */
+
+static void TAG(flush_point_0)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   if ( !v0->mask ) {
+      LOCAL_VARS;
+      DRAW_POINT( v0 );
+   }
+}
+
+
+/* =============================================================
+ * GL_LINES
+ */
+
+static void TAG(flush_line_1)( GLcontext *ctx, TNL_VERTEX *v0 );
+
+static void TAG(flush_line_0)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   FLUSH_VERTEX = TAG(flush_line_1);
+   ACTIVE_VERTEX = IMM_VERTICES( 1 );
+}
+
+static void TAG(flush_line_1)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   TNL_VERTEX *v1 = v0 - 1;
+   ACTIVE_VERTEX = IMM_VERTICES( 0 );
+   FLUSH_VERTEX = TAG(flush_line_0);
+   if (FALLBACK_OR_CLIPPING) 
+      CLIP_OR_DRAW_LINE( ctx, v1, v0 ); 
+   else
+      DRAW_LINE( ctx, v1, v0 ); 
+}
+
+
+/* =============================================================
+ * GL_LINE_LOOP
+ */
+
+static void TAG(flush_line_loop_2)( GLcontext *ctx, TNL_VERTEX *v0 );
+static void TAG(flush_line_loop_1)( GLcontext *ctx, TNL_VERTEX *v0 );
+
+static void TAG(flush_line_loop_0)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+
+   ACTIVE_VERTEX = v0 + 1;
+   FLUSH_VERTEX = TAG(flush_line_loop_1);
+}
+
+#define DRAW_LINELOOP_LINE( a, b )			\
+   if (!HAVE_LINE_STRIP || FALLBACK_OR_CLIPPING) {	\
+      CLIP_OR_DRAW_LINE( ctx, a, b );			\
+   } else if (EXTEND_PRIM( 1 )) {			\
+      EMIT_VERTEX( b );					\
+   } else {						\
+      BEGIN_PRIM( GL_LINE_STRIP, 2 ); 			\
+      EMIT_VERTEX( a );					\
+      EMIT_VERTEX( b );      				\
+   }
+
+static void TAG(flush_line_loop_1)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   TNL_VERTEX *v1 = v0 - 1;
+   ACTIVE_VERTEX = v1;
+   FLUSH_VERTEX = TAG(flush_line_loop_2);
+   DRAW_LINELOOP_LINE( v1, v0 );
+}
+
+static void TAG(flush_line_loop_2)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   TNL_VERTEX *v1 = v0 + 1;
+   ACTIVE_VERTEX = v1;
+   FLUSH_VERTEX = TAG(flush_line_loop_1);
+   DRAW_LINELOOP_LINE( v1, v0 );
+}
+
+static void TAG(end_line_loop)( GLcontext *ctx )
+{
+   LOCAL_VARS;
+
+   if ( FLUSH_VERTEX != TAG(flush_line_loop_0) ) {
+      TNL_VERTEX *v1 = ACTIVE_VERTEX;
+      TNL_VERTEX *v0 = IMM_VERTICES( 0 );
+      DRAW_LINELOOP_LINE( v1, v0 );
+   }
+}
+
+
+
+/* =============================================================
+ * GL_LINE_STRIP
+ */
+
+static void TAG(flush_line_strip_2)( GLcontext *ctx, TNL_VERTEX *v0 );
+static void TAG(flush_line_strip_1)( GLcontext *ctx, TNL_VERTEX *v0 );
+
+static void TAG(flush_line_strip_0)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+
+   ACTIVE_VERTEX = v0 + 1;
+   FLUSH_VERTEX = TAG(flush_line_strip_0b);
+}
+
+
+static void TAG(flush_line_strip_1)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   TNL_VERTEX *v1 = v0 - 1;
+
+   ACTIVE_VERTEX = v1;
+   FLUSH_VERTEX = TAG(flush_line_strip_2);
+
+   if (!HAVE_LINE_STRIP || FALLBACK_OR_CLIPPING)
+      CLIP_OR_DRAW_LINE( ctx, v1, v0 );
+   else if (EXTEND_PRIM( 1 )) {
+      EMIT_VERTEX( v0 );
+   } else {
+      BEGIN_PRIM( GL_LINE_STRIP, 2 ); 
+      EMIT_VERTEX( v1 );
+      EMIT_VERTEX( v0 );      
+   }      
+}
+
+static void TAG(flush_line_strip_2)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   TNL_VERTEX *v1 = v0 + 1;
+
+   ACTIVE_VERTEX = v1;
+   FLUSH_VERTEX = TAG(flush_line_strip_1);
+
+   if (!HAVE_LINE_STRIP || FALLBACK_OR_CLIPPING)
+      CLIP_OR_DRAW_LINE( ctx, v1, v0 );
+   else if (EXTEND_PRIM( 1 )) {
+      EMIT_VERTEX( v0 );
+   } else {
+      BEGIN_PRIM( GL_LINE_STRIP, 2 ); 
+      EMIT_VERTEX( v1 );
+      EMIT_VERTEX( v0 );      
+   }      
+}
+
+
+
+/* =============================================================
+ * GL_TRIANGLES
+ */
+
+static void TAG(flush_triangle_2)( GLcontext *ctx, TNL_VERTEX *v0 );
+static void TAG(flush_triangle_1)( GLcontext *ctx, TNL_VERTEX *v0 );
+
+static void TAG(flush_triangle_0)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+
+   if ( DBG ) fprintf( stderr, __FUNCTION__ "\n" );
+
+   ACTIVE_VERTEX = v0 + 1;
+   FLUSH_VERTEX = TAG(flush_triangle_1);
+   BEGIN_PRIM( GL_TRIANGLES, 0 );
+}
+
+static void TAG(flush_triangle_1)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+
+   if ( DBG ) fprintf( stderr, __FUNCTION__ "\n" );
+
+   ACTIVE_VERTEX = v0 + 1;
+   FLUSH_VERTEX = TAG(flush_triangle_2);
+}
+
+static void TAG(flush_triangle_2)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   TNL_VERTEX *v2 = v0 - 2;
+   TNL_VERTEX *v1 = v0 - 1;
+
+   if ( DBG ) fprintf( stderr, __FUNCTION__ "\n" );
+
+   ACTIVE_VERTEX = v2;
+   FLUSH_VERTEX = TAG(flush_triangle_0);
+
+   /* nothing gained by trying to emit as hw primitives -- that
+    * happens normally in this case.  
+    */
+   if (FALLBACK_OR_CLIPPING)
+      CLIP_OR_DRAW_TRI( ctx, v2, v1, v0 );
+   else
+      DRAW_TRI( ctx, v2, v1, v0 );
+}
+
+
+
+
+/* =============================================================
+ * GL_TRIANGLE_STRIP
+ */
+
+static void TAG(flush_tri_strip_3)( GLcontext *ctx, TNL_VERTEX *v0 );
+static void TAG(flush_tri_strip_2)( GLcontext *ctx, TNL_VERTEX *v0 );
+static void TAG(flush_tri_strip_1)( GLcontext *ctx, TNL_VERTEX *v0 );
+
+static void TAG(flush_tri_strip_0)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   ACTIVE_VERTEX = IMM_VERTICES( 1 );
+   FLUSH_VERTEX = TAG(flush_tri_strip_1);
+}
+
+static void TAG(flush_tri_strip_1)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   ACTIVE_VERTEX = IMM_VERTICES( 2 );
+   FLUSH_VERTEX = TAG(flush_tri_strip_2);
+}
+
+#define DO_TRISTRIP_TRI( vert0, vert1 )				\
+   if (!HAVE_TRI_STRIP || FALLBACK_OR_CLIPPING) {	\
+      TNL_VERTEX *v2 = IMM_VERTICES( vert0 );			\
+      TNL_VERTEX *v1 = IMM_VERTICES( vert1 );			\
+      TAG(draw_tri)( ctx, v2, v1, v0 );				\
+   } else if (EXTEND_PRIM( 1 )) {				\
+      EMIT_VERTEX( v0 );					\
+   } else {							\
+      TNL_VERTEX *v2 = IMM_VERTICES( vert0 );			\
+      TNL_VERTEX *v1 = IMM_VERTICES( vert1 );			\
+      BEGIN_PRIM( GL_TRIANGLE_STRIP, 3 ); 			\
+      EMIT_VERTEX( v2 );					\
+      EMIT_VERTEX( v1 );      					\
+      EMIT_VERTEX( v0 );      					\
+   }      
+
+static void TAG(flush_tri_strip_2)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   FLUSH_VERTEX = TAG(flush_tri_strip_3);
+   ACTIVE_VERTEX = IMM_VERTICES( 3 );
+   DO_TRISTRIP_TRI( 0, 1 );
+}
+
+static void TAG(flush_tri_strip_3)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   FLUSH_VERTEX = TAG(flush_tri_strip_4);
+   ACTIVE_VERTEX = IMM_VERTICES( 0 );
+   DO_TRISTRIP_TRI( 1, 2 );
+}
+
+static void TAG(flush_tri_strip_4)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   FLUSH_VERTEX = TAG(flush_tri_strip_5);
+   ACTIVE_VERTEX = IMM_VERTICES( 1 );
+   DO_TRISTRIP_TRI( 2, 3 );
+}
+
+static void TAG(flush_tri_strip_5)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   FLUSH_VERTEX = TAG(flush_tri_strip_2);
+   ACTIVE_VERTEX = IMM_VERTICES( 2 );
+   DO_TRISTRIP_TRI( 0, 3 );
+}
+
+
+
+/* =============================================================
+ * GL_TRIANGLE_FAN
+ */
+
+static void TAG(flush_tri_fan_2)( GLcontext *ctx, TNL_VERTEX *v0 );
+static void TAG(flush_tri_fan_1)( GLcontext *ctx, TNL_VERTEX *v0 );
+
+static void TAG(flush_tri_fan_0)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+
+   ACTIVE_VERTEX = v0 + 1;
+   FLUSH_VERTEX = TAG(flush_tri_fan_1);
+}
+
+static void TAG(flush_tri_fan_1)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+
+   ACTIVE_VERTEX = v0 + 1;
+   FLUSH_VERTEX = TAG(flush_tri_fan_2);
+}
+
+#define DO_TRIFAN_TRI( vert0, vert1 )				\
+   if (!HAVE_TRI_FAN || FALLBACK_OR_CLIPPING) {	\
+      TNL_VERTEX *v2 = IMM_VERTICES( vert0 );			\
+      TNL_VERTEX *v1 = IMM_VERTICES( vert1 );			\
+      TAG(draw_tri)( ctx, v2, v1, v0 );				\
+   } else if (EXTEND_PRIM( 1 )) {				\
+      EMIT_VERTEX( v0 );					\
+   } else {							\
+      TNL_VERTEX *v2 = IMM_VERTICES( vert0 );			\
+      TNL_VERTEX *v1 = IMM_VERTICES( vert1 );			\
+      BEGIN_PRIM( GL_TRIANGLE_FAN, 3 ); 			\
+      EMIT_VERTEX( v2 );					\
+      EMIT_VERTEX( v1 );      					\
+      EMIT_VERTEX( v0 );      					\
+   }      
+
+static void TAG(flush_tri_fan_2)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   ACTIVE_VERTEX = IMM_VERTICES( 1 );
+   FLUSH_VERTEX = TAG(flush_tri_fan_3 );
+   DO_TRIFAN_TRI( 0, 1 );
+}
+
+static void TAG(flush_tri_fan_3)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   ACTIVE_VERTEX = IMM_VERTICES( 2 );
+   FLUSH_VERTEX = TAG(flush_tri_fan_2 );
+   DO_TRIFAN_TRI( 0, 2 );
+}
+
+
+
+/* =============================================================
+ * GL_QUADS
+ */
+
+static void TAG(flush_quad_3)( GLcontext *ctx, TNL_VERTEX *v0 );
+static void TAG(flush_quad_2)( GLcontext *ctx, TNL_VERTEX *v0 );
+static void TAG(flush_quad_1)( GLcontext *ctx, TNL_VERTEX *v0 );
+
+static void TAG(flush_quad_0)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   IMM_VERTEX( v0 ) = v0 + 1;
+   FLUSH_VERTEX = TAG(flush_quad_1);
+}
+
+static void TAG(flush_quad_1)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   IMM_VERTEX( v0 ) = v0 + 1;
+   FLUSH_VERTEX = TAG(flush_quad_2);
+}
+
+static void TAG(flush_quad_2)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   IMM_VERTEX( v0 ) = v0 + 1;
+   FLUSH_VERTEX = TAG(flush_quad_3);
+}
+
+static void TAG(flush_quad_3)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   TNL_VERTEX *v3 = v0 - 3;
+   TNL_VERTEX *v2 = v0 - 2;
+   TNL_VERTEX *v1 = v0 - 1;
+
+   IMM_VERTEX( v0 ) = v3;
+   FLUSH_VERTEX = TAG(flush_quad_0);
+
+   if (!HAVE_HW_QUADS || FALLBACK_OR_CLIPPING) {
+      CLIP_OR_DRAW_TRI( ctx, v3, v2, v0 );
+      CLIP_OR_DRAW_TRI( ctx, v2, v1, v0 );
+   } else {
+      EXTEND_PRIM_NF( GL_QUADS, 4 );
+      EMIT_VERTEX( v3 );
+      EMIT_VERTEX( v2 );
+      EMIT_VERTEX( v1 );      
+      EMIT_VERTEX( v0 );      
+   }
+}
+
+
+
+/* =============================================================
+ * GL_QUAD_STRIP
+ */
+
+static void TAG(flush_quad_strip_3)( GLcontext *ctx, TNL_VERTEX *v0 );
+static void TAG(flush_quad_strip_2)( GLcontext *ctx, TNL_VERTEX *v0 );
+static void TAG(flush_quad_strip_1)( GLcontext *ctx, TNL_VERTEX *v0 );
+
+static void TAG(flush_quad_strip_0)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+
+   IMM_VERTEX( v3 ) = v0;
+   IMM_VERTEX( v0 ) = v0 + 1;
+   FLUSH_VERTEX = TAG(flush_quad_strip_1);
+}
+
+static void TAG(flush_quad_strip_1)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+
+   IMM_VERTEX( v2 ) = v0;
+   IMM_VERTEX( v0 ) = v0 + 1;
+   FLUSH_VERTEX = TAG(flush_quad_strip_2);
+}
+
+static void TAG(flush_quad_strip_2)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+
+   IMM_VERTEX( v1 ) = v0;
+   IMM_VERTEX( v0 ) = v0 + 1;
+   FLUSH_VERTEX = TAG(flush_quad_strip_3);
+}
+
+static void TAG(flush_quad_strip_3)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   TNL_VERTEX *v3 = IMM_VERTEX( v3 );
+   TNL_VERTEX *v2 = IMM_VERTEX( v2 );
+   TNL_VERTEX *v1 = IMM_VERTEX( v1 );
+
+   IMM_VERTEX( v0 ) = v3;
+   IMM_VERTEX( v2 ) = v0;
+   IMM_VERTEX( v3 ) = v1;
+   FLUSH_VERTEX = TAG(flush_quad_strip_2);
+
+   if (FALLBACK_OR_CLIPPING) {
+      CLIP_OR_DRAW_TRI( ctx, v3, v2, v0 );
+      CLIP_OR_DRAW_TRI( ctx, v2, v1, v0 );
+   } else {
+      DRAW_TRI( ctx, v3, v2, v0 );
+      DRAW_TRI( ctx, v2, v1, v0 );
+   }
+}
+
+
+
+/* =============================================================
+ * GL_POLYGON
+ */
+
+static void TAG(flush_poly_2)( GLcontext *ctx, TNL_VERTEX *v0 );
+static void TAG(flush_poly_1)( GLcontext *ctx, TNL_VERTEX *v0 );
+
+static void TAG(flush_poly_0)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   ACTIVE_VERTEX = IMM_VERTICES( 1 );
+   FLUSH_VERTEX = TAG(flush_poly_1);
+}
+
+static void TAG(flush_poly_1)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   ACTIVE_VERTEX = IMM_VERTICES( 2 );
+   FLUSH_VERTEX = TAG(flush_poly_2);
+}
+
+#define DO_POLY_TRI( vert0, vert1 )				\
+   if (!HAVE_POLYGONS || FALLBACK_OR_CLIPPING) {	\
+      TNL_VERTEX *v2 = IMM_VERTICES( vert0 );			\
+      TNL_VERTEX *v1 = IMM_VERTICES( vert1 );			\
+      TAG(draw_tri)( ctx, v1, v0, v2 );				\
+   } else if (EXTEND_PRIM( 1 )) {				\
+      EMIT_VERTEX( v0 );					\
+   } else {							\
+      TNL_VERTEX *v2 = IMM_VERTICES( vert0 );			\
+      TNL_VERTEX *v1 = IMM_VERTICES( vert1 );			\
+      BEGIN_PRIM( GL_POLYGON, 3 );				\
+      EMIT_VERTEX( v2 );					\
+      EMIT_VERTEX( v1 );					\
+      EMIT_VERTEX( v0 );					\
+   }      
+
+static void TAG(flush_poly_2)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   ACTIVE_VERTEX = IMM_VERTICES( 1 );
+   FLUSH_VERTEX = TAG(flush_poly_3);
+   DO_POLY_TRI( 0, 1 );
+}
+
+static void TAG(flush_poly_3)( GLcontext *ctx, TNL_VERTEX *v0 )
+{
+   LOCAL_VARS;
+   ACTIVE_VERTEX = IMM_VERTICES( 2 );
+   FLUSH_VERTEX = TAG(flush_poly_2);
+   DO_POLY_TRI( 0, 2 );
+}
+
+
+void (*TAG(flush_tab)[GL_POLYGON+1])( GLcontext *, TNL_VERTEX * ) =
+{
+   TAG(flush_point),
+   TAG(flush_line_0),
+   TAG(flush_line_loop_0),
+   TAG(flush_line_strip_0),
+   TAG(flush_triangle_0),
+   TAG(flush_tri_strip_0),
+   TAG(flush_tri_fan_0),
+   TAG(flush_quad_0),
+   TAG(flush_quad_strip_0),
+   TAG(flush_poly_0),
+};
+
+
+#ifndef PRESERVE_PRIM_DEFS
+#undef LOCAL_VARS
+#undef GET_INTERP_FUNC
+#undef IMM_VERTEX
+#undef IMM_VERTICES
+#undef FLUSH_VERTEX
+#endif
+#undef PRESERVE_PRIM_DEFS
+#undef EXTEND_PRIM
+#undef EMIT_VERTEX
+#undef EMIT_VERTEX_TRI
+#undef EMIT_VERTEX_LINE
+#undef EMIT_VERTEX_POINT
+#undef TAG
diff --git a/src/mesa/tnl_dd/imm/t_dd_imm_tapi.h b/src/mesa/tnl_dd/imm/t_dd_imm_tapi.h
new file mode 100644
index 00000000000..b82a9abebff
--- /dev/null
+++ b/src/mesa/tnl_dd/imm/t_dd_imm_tapi.h
@@ -0,0 +1,75 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999-2001  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:
+ *    Gareth Hughes <[email protected]>
+ *    Keith Whitwell <[email protected]>
+ */
+
+/* Template for immediate mode texture coordinate functions.
+ */
+
+#ifndef DO_PROJ_TEX
+#error "Need to define DO_PROJ_TEX"
+#endif
+
+
+/* =============================================================
+ * Notify on calls to texture4f, to allow switch to projected texture
+ * vertex format:
+ */
+
+static void TAG(TexCoord4f)( GLfloat s, GLfloat t, GLfloat r, GLfloat q )
+{
+   GET_CURRENT;
+   DO_PROJ_TEX;
+   TEXCOORD4( s, t, r, q );
+}
+
+static void TAG(TexCoord4fv)( const GLfloat *v )
+{
+   GET_CURRENT;
+   DO_PROJ_TEX;
+   TEXCOORD4( v[0], v[1], v[2], v[3] );
+}
+
+static void TAG(MultiTexCoord4fARB)( GLenum target, GLfloat s,
+				     GLfloat t, GLfloat r, GLfloat q )
+{
+   GET_CURRENT;
+   DO_PROJ_TEX;
+   MULTI_TEXCOORD4( unit, s, t, r, q );
+}
+
+static void TAG(MultiTexCoord4fvARB)( GLenum target, const GLfloat *v )
+{
+   GET_CURRENT;
+   DO_PROJ_TEX;
+   MULTI_TEXCOORD4( unit, v[0], v[1], v[2], v[3] );
+}
+
+
+
+#undef DO_PROJ_TEX
+#undef TAG
diff --git a/src/mesa/tnl_dd/imm/t_dd_imm_vapi.h b/src/mesa/tnl_dd/imm/t_dd_imm_vapi.h
new file mode 100644
index 00000000000..95e93b1c839
--- /dev/null
+++ b/src/mesa/tnl_dd/imm/t_dd_imm_vapi.h
@@ -0,0 +1,159 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999-2001  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:
+ *    Gareth Hughes <[email protected]>
+ *    Keith Whitwell <[email protected]>
+ */
+
+/* Template for immediate mode vertex functions.
+ */
+
+#define DBG 0
+
+#define VERTEX( ox, oy, oz, ow )
+do {
+   GET_CURRENT_VERTEX;
+   GLfloat w;
+   GLuint mask;
+   const GLfloat * const m = ctx->_ModelProjectMatrix.m;
+
+   if (DO_FULL_MATRIX) {
+      VERTEX_CLIP(0) = m[0] * ox + m[4] * oy + m[8]  * oz + m[12] * ow;
+      VERTEX_CLIP(1) = m[1] * ox + m[5] * oy + m[9]  * oz + m[13] * ow;
+      VERTEX_CLIP(2) = m[2] * ox + m[6] * oy + m[10] * oz + m[14] * ow;
+      VERTEX_CLIP(3) = m[3] * ox + m[7] * oy + m[11] * oz + m[15] * ow;
+      w = VERTEX_CLIP(3);
+   }
+   else if (DO_NOROT_MATRIX) {
+      VERTEX_CLIP(0) = m[0] * ox                          + m[12] * ow;
+      VERTEX_CLIP(1) =             m[5] * oy              + m[13] * ow;
+      VERTEX_CLIP(2) =                         m[10] * oz + m[14] * ow;
+      VERTEX_CLIP(3) =                                              ow;
+      w = ow;
+   } 
+   else {
+      ASSERT (DO_IDENTITY_MATRIX);
+      VERTEX_CLIP(0) = ox;
+      VERTEX_CLIP(1) = oy;
+      VERTEX_CLIP(2) = oz;
+      VERTEX_CLIP(3) = ow;
+      w = ow;
+   }
+
+   mask = 0;
+   if (DO_CLIP_TEST) {
+      if ( VERTEX_CLIP(0) >  w ) mask |= CLIP_RIGHT_BIT;
+      if ( VERTEX_CLIP(0) < -w ) mask |= CLIP_LEFT_BIT;
+      if ( VERTEX_CLIP(1) >  w ) mask |= CLIP_TOP_BIT;
+      if ( VERTEX_CLIP(1) < -w ) mask |= CLIP_BOTTOM_BIT;
+      if ( VERTEX_CLIP(2) >  w ) mask |= CLIP_FAR_BIT;
+      if ( VERTEX_CLIP(2) < -w ) mask |= CLIP_NEAR_BIT;
+      VERTEX_MASK(v) = mask;
+   }
+
+   if (!mask) {
+      if (HAVE_VERTEX_WIN) {
+	 if (!HAVE_HW_VIEWPORT) {
+	    const GLfloat *s = GET_VIEWPORT_MATRIX();
+	    if (HAVE_W && HAVE_HW_DIVIDE) {
+	       VERTEX_WIN( 0 ) = s[0]  * VERTEX_CLIP( 0 ) + s[12];
+	       VERTEX_WIN( 1 ) = s[5]  * VERTEX_CLIP( 1 ) + s[13];
+	       VERTEX_WIN( 2 ) = s[10] * VERTEX_CLIP( 2 ) + s[14];
+	       VERTEX_WIN( 3 ) = w;
+	    }
+	    else {
+	       const GLfloat oow = 1.0/w; /* possibly opt away */
+	       VERTEX_WIN( 0 ) = s[0]  * VERTEX_CLIP( 0 ) * oow + s[12];
+	       VERTEX_WIN( 1 ) = s[5]  * VERTEX_CLIP( 1 ) * oow + s[13];
+	       VERTEX_WIN( 2 ) = s[10] * VERTEX_CLIP( 2 ) * oow + s[14];
+	       if (HAVE_W)
+		  VERTEX_WIN( 3 ) = oow;
+	    }
+	 }
+	 else if (HAVE_W && HAVE_HW_DIVIDE) {
+	    if (!VERTEX_WIN_IS_VERTEX_CLIP) {
+	       VERTEX_WIN( 0 ) = VERTEX_CLIP( 0 );
+	       VERTEX_WIN( 1 ) = VERTEX_CLIP( 1 );
+	       VERTEX_WIN( 2 ) = VERTEX_CLIP( 2 );
+	       VERTEX_WIN( 3 ) = w;
+	    }
+	 }
+	 else {
+	    const GLfloat oow = 1.0/w; /* possibly opt away */
+	    VERTEX_WIN( 0 ) = VERTEX_CLIP( 0 ) * oow;
+	    VERTEX_WIN( 1 ) = VERTEX_CLIP( 1 ) * oow;
+	    VERTEX_WIN( 2 ) = VERTEX_CLIP( 2 ) * oow;
+	    if (HAVE_W)
+	       VERTEX_WIN( 3 ) = oow;
+	 }
+      }
+   } else if (!FALLBACK_OR_CLIPPING) {
+      SET_CLIPPING();		/* transition to clipping */
+   }
+
+   COPY_VERTEX_FROM_CURRENT;
+   BUILD_PRIM_FROM_VERTEX;
+}
+
+/* Let the compiler optimize away the constant operations:
+ */
+static void VTAG(Vertex2f)( GLfloat ox, GLfloat oy )
+{
+   /* Cliptest on clip[2] could also be eliminated...
+    */
+   VERTEX( ox, oy, 0, 1 );
+}
+
+static void VTAG(Vertex2fv)( const GLfloat *obj )
+{
+   /* Cliptest on clip[2] could also be eliminated...
+    */
+   VERTEX( obj[0], obj[1], 0, 1 );
+}
+
+static void VTAG(Vertex3f)( GLfloat ox, GLfloat oy, GLfloat oz )
+{
+   VERTEX( ox, oy, oz, 1 );
+}
+
+static void VTAG(Vertex3fv)( const GLfloat *obj )
+{
+   VERTEX( obj[0], obj[1], obj[2], 1 );
+}
+
+static void VTAG(Vertex4f)( GLfloat ox, GLfloat oy, GLfloat oz, GLfloat ow )
+{
+   VERTEX( ox, oy, oz, ow );
+}
+
+static void VTAG(Vertex4fv)( const GLfloat *obj )
+{
+   VERTEX( obj[0], obj[1], obj[2], obj[3] );
+}
+
+
+#undef DO_FULL_MATRIX
+#undef VTAG
+#undef VERTEX
diff --git a/src/mesa/tnl_dd/imm/t_dd_imm_vb.c b/src/mesa/tnl_dd/imm/t_dd_imm_vb.c
new file mode 100644
index 00000000000..0c4462f556f
--- /dev/null
+++ b/src/mesa/tnl_dd/imm/t_dd_imm_vb.c
@@ -0,0 +1,204 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999-2001  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 <[email protected]>
+ */
+
+/* Template to build clipping routines to support t_dd_imm_primtmp.h.
+ *
+ * The TAG(draw_line) and TAG(draw_triangle) routines are called in
+ * clipping and fallback scenarios, and when the native hardware
+ * primitive (eg polygons) is unavailable.
+ */
+
+
+#define CLIP_DOTPROD(K, A, B, C, D)		\
+   (CLIP_X(K)*A + CLIP_Y(K)*B + 	\
+    CLIP_Z(K)*C + CLIP_W(K)*D)
+
+#define POLY_CLIP( PLANE, A, B, C, D )					\
+do {									\
+   if (mask & PLANE) {							\
+      TNL_VERTEX **indata = inlist[in];					\
+      TNL_VERTEX **outdata = inlist[in ^= 1];				\
+      TNL_VERTEX *J = indata[0];					\
+      GLfloat dpJ = CLIP_DOTPROD(J, A, B, C, D );			\
+      GLuint outcount = 0;						\
+      GLuint i;								\
+									\
+      indata[n] = indata[0]; /* prevent rotation of vertices */		\
+      for (i = 1; i <= n; i++) {					\
+	 TNL_VERTEX *I = indata[i];					\
+	 GLfloat dpI = CLIP_DOTPROD(idx, A, B, C, D );			\
+									\
+	 if (!NEGATIVE(dpPrev)) {					\
+	    outdata[outcount++] = J;					\
+	 }								\
+									\
+	 if (DIFFERENT_SIGNS(dpI, dpJ)) {				\
+            TNL_VERTEX *O = verts++;					\
+            outdata[outcount++] = O;					\
+	    if (NEGATIVE(dpI)) {					\
+	       /* Going out of bounds.  Avoid division by zero as we	\
+		* know dp != dpPrev from DIFFERENT_SIGNS, above.	\
+		*/							\
+	       GLfloat t = dpI / (dpI - dpJ);				\
+      	       INTERP( ctx, t, O, I, J );				\
+	    } else {							\
+	       /* Coming back in.					\
+		*/							\
+	       GLfloat t = dpJ / (dpJ - dpI);				\
+	       INTERP( ctx, t, O, J, I );				\
+	    }								\
+	 }								\
+									\
+	 J = I;								\
+	 dpJ = dpI;							\
+      }									\
+									\
+      if (outcount < 3)							\
+	 return;							\
+									\
+      nr = outcount;							\
+   }									\
+} while (0)
+
+
+#define LINE_CLIP(PLANE, A, B, C, D )			\
+do {							\
+   if (mask & PLANE) {					\
+      GLfloat dpI = CLIP_DOTPROD( I, A, B, C, D );	\
+      GLfloat dpJ = CLIP_DOTPROD( J, A, B, C, D );	\
+							\
+      if (DIFFERENT_SIGNS(dpI, dpJ)) {			\
+         TNL_VERTEX *O = verts++;			\
+	 if (NEGATIVE(dpJ)) {				\
+	    GLfloat t = dpI / (dpI - dpJ);		\
+	    INTERP( ctx, t, O, I, J );	\
+            J = O;					\
+	 } else {					\
+  	    GLfloat t = dpJ / (dpJ - dpI);		\
+	    INTERP( ctx, t, O, J, I );	\
+            I = O;					\
+	 }						\
+      }							\
+      else if (NEGATIVE(dpI))				\
+	 return;					\
+  }							\
+} while (0)
+
+
+
+/* Clip a line against the viewport and user clip planes.
+ */
+static void TAG(clip_draw_line)( GLcontext *ctx,
+				 TNL_VERTEX *I,
+				 TNL_VERTEX *J,
+				 GLuint mask )
+{
+   LOCAL_VARS;
+   GET_INTERP_FUNC;
+   TNL_VERTEX tmp[MAX_CLIPPED_VERTICES];
+   TNL_VERTEX *verts = tmp;
+   TNL_VERTEX *pv = J;
+
+   LINE_CLIP( CLIP_RIGHT_BIT,  -1,  0,  0, 1 );
+   LINE_CLIP( CLIP_LEFT_BIT,    1,  0,  0, 1 );
+   LINE_CLIP( CLIP_TOP_BIT,     0, -1,  0, 1 );
+   LINE_CLIP( CLIP_BOTTOM_BIT,  0,  1,  0, 1 );
+   LINE_CLIP( CLIP_FAR_BIT,     0,  0, -1, 1 );
+   LINE_CLIP( CLIP_NEAR_BIT,    0,  0,  1, 1 );
+
+   if ((ctx->_TriangleCaps & DD_FLATSHADE) && J != pv)
+      COPY_PV( ctx, J, pv );
+
+   DRAW_LINE( I, J );
+}
+
+
+/* Clip a triangle against the viewport and user clip planes.
+ */
+static void TAG(clip_draw_triangle)( GLcontext *ctx,
+				     TNL_VERTEX *v0,
+				     TNL_VERTEX *v1,
+				     TNL_VERTEX *v2,
+				     GLuint mask )
+{
+   LOCAL_VARS;
+   GET_INTERP_FUNC;
+   TNL_VERTEX tmp[MAX_CLIPPED_VERTICES];
+   TNL_VERTEX *verts = tmp;
+   TNL_VERTEX *(inlist[2][MAX_CLIPPED_VERTICES]);
+   TNL_VERTEX **out;
+   GLuint in = 0;
+   GLuint n = 3;
+   GLuint i;
+
+   ASSIGN_3V(inlist, v2, v0, v1 ); /* pv rotated to slot zero */
+
+   POLY_CLIP( CLIP_RIGHT_BIT,  -1,  0,  0, 1 );
+   POLY_CLIP( CLIP_LEFT_BIT,    1,  0,  0, 1 );
+   POLY_CLIP( CLIP_TOP_BIT,     0, -1,  0, 1 );
+   POLY_CLIP( CLIP_BOTTOM_BIT,  0,  1,  0, 1 );
+   POLY_CLIP( CLIP_FAR_BIT,     0,  0, -1, 1 );
+   POLY_CLIP( CLIP_NEAR_BIT,    0,  0,  1, 1 );
+
+   if ((ctx->_TriangleCaps & DD_FLATSHADE) && v2 != inlist[0]) 
+      COPY_PV( ctx, inlist[0], v2 );
+
+   out = inlist[in];
+   DRAW_POLYGON( out, n );
+}
+
+
+static __inline void TAG(draw_triangle)( GLcontext *ctx,
+					 TNL_VERTEX *v0,
+					 TNL_VERTEX *v1,
+					 TNL_VERTEX *v2 )
+{
+   LOCAL_VARS;
+   GLubyte ormask = (v0->mask | v1->mask | v2->mask);
+
+   if ( !ormask ) {
+      DRAW_TRI( v0, v1, v2 );
+   } else if ( !(v0->mask & v1->mask & v2->mask) ) {
+      TAG(clip_draw_triangle)( ctx, v0, v1, v2, ormask );
+   }
+}
+
+static __inline void TAG(draw_line)( GLcontext *ctx,
+				     TNL_VERTEX *v0,
+				     TNL_VERTEX *v1 )
+{
+   LOCAL_VARS;
+   GLubyte ormask = (v0->mask | v1->mask);
+
+   if ( !ormask ) {
+      DRAW_LINE( v0, v1 );
+   } else if ( !(v0->mask & v1->mask) ) {
+      TAG(clip_draw_line)( ctx, v0, v1, ormask );
+   }
+}
+
diff --git a/src/mesa/tnl_dd/imm/t_dd_imm_vbtmp.h b/src/mesa/tnl_dd/imm/t_dd_imm_vbtmp.h
new file mode 100644
index 00000000000..2f76553cff9
--- /dev/null
+++ b/src/mesa/tnl_dd/imm/t_dd_imm_vbtmp.h
@@ -0,0 +1,268 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999-2001  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 <[email protected]>
+ */
+
+/* Template to build support for t_dd_imm_* tnl module using vertices
+ * as defined in t_dd_vertex.h.
+ *
+ * See t_dd_vbtmp.h for definitions of arguments to this file.
+ * Unfortunately it seems necessary to duplicate a lot of that code.
+ */
+
+#ifndef LOCALVARS
+#define LOCALVARS
+#endif
+
+
+
+/* COPY_VERTEX_FROM_CURRENT in t_dd_imm_vapi.c
+ */
+static void TAG(emit_vfmt)( GLcontext *ctx, VERTEX *v )
+{
+   LOCALVARS
+      ;
+
+   /* This template assumes (like t_dd_vbtmp.h) that color is ubyte.
+    */
+   if (DO_TEX0 || DO_TEX1 || !HAVE_TINY_VERTICES)
+   {
+      const GLubyte *col = GET_HARDWARE_COLOR();
+      if (HAVE_RGBA_COLOR) {
+	 v->v.ui[4] = *(GLuint *)&col;
+      } else {
+	 v->v.color.blue  = col[2];
+	 v->v.color.green = col[1];
+	 v->v.color.red   = col[0];
+	 v->v.color.alpha = col[3];
+      }
+   }
+   else {
+      if (HAVE_RGBA_COLOR) {
+	 v->v.ui[3] = *(GLuint *)col;
+      }
+      else {
+	 v->tv.color.blue  = col[2];
+	 v->tv.color.green = col[1];
+	 v->tv.color.red   = col[0];
+	 v->tv.color.alpha = col[3];
+      }
+   }
+
+   if (DO_TEX0) {
+      GLfloat *tc = ctx->Current.Texture[0];
+      v->v.u0 = tc[0];
+      v->v.v0 = tc[1];
+      if (DO_PTEX) {
+	 if (HAVE_PTEX_VERTICES) {
+	    v->pv.q0 = tc[3];
+	 } 
+	 else {
+	    float rhw = 1.0 / tc[3];
+	    v->v.w *= tc[3];
+	    v->v.u0 *= rhw;
+	    v->v.v0 *= rhw;
+	 } 
+      } 
+   }
+   if (DO_TEX1) {
+      GLfloat *tc = ctx->Current.Texture[1];
+      if (DO_PTEX) {
+	 v->pv.u1 = tc[0];
+	 v->pv.v1 = tc[1];
+	 v->pv.q1 = tc[3];
+      } 
+      else {
+	 v->v.u1 = tc[0];
+	 v->v.v1 = tc[1];
+      }
+   } 
+   else if (DO_PTEX) {
+      *(GLuint *)&v->pv.q1 = 0;	/* avoid culling on radeon */
+   }
+   if (DO_TEX2) {
+      GLfloat *tc = ctx->Current.Texture[2];
+      if (DO_PTEX) {
+	 v->pv.u2 = tc[0];
+	 v->pv.v2 = tc[1];
+	 v->pv.q2 = tc[3];
+      } 
+      else {
+	 v->v.u2 = tc[0];
+	 v->v.v2 = tc[1];
+      }
+   } 
+   if (DO_TEX3) {
+      GLfloat *tc = ctx->Current.Texture[3];
+      if (DO_PTEX) {
+	 v->pv.u3 = tc[0];
+	 v->pv.v3 = tc[1];
+	 v->pv.q3 = tc[3];
+      } 
+      else {
+	 v->v.u3 = tc[0];
+	 v->v.v3 = tc[1];
+      }
+   } 
+}
+
+
+
+
+static void TAG(interp)( GLcontext *ctx,
+			 GLfloat t,
+			 TNL_VERTEX *dst,
+			 TNL_VERTEX *in,
+			 TNL_VERTEX *out )
+{
+   LOCALVARS
+   const GLfloat *s = GET_VIEWPORT_MAT();
+   GLfloat w;
+
+   (void)s;
+
+   if (HAVE_HW_DIVIDE) {
+      VIEWPORT_X( dst->v.v.x, dst->clip[0] );
+      VIEWPORT_Y( dst->v.v.y, dst->clip[1] );
+      VIEWPORT_Z( dst->v.v.z, dst->clip[2] );
+      w = dstclip[3];
+   }
+   else {
+      w = 1.0 / dst->clip[3];
+      VIEWPORT_X( dst->v.v.x, dst->clip[0] * w );
+      VIEWPORT_Y( dst->v.v.y, dst->clip[1] * w );
+      VIEWPORT_Z( dst->v.v.z, dst->clip[2] * w );
+   }
+
+   if (HAVE_HW_DIVIDE || DO_TEX0) {
+
+      dst->v.v.w = w;
+
+      INTERP_UB( t, dst->v.ub4[4][0], out->v.ub4[4][0], in->v.ub4[4][0] );
+      INTERP_UB( t, dst->v.ub4[4][1], out->v.ub4[4][1], in->v.ub4[4][1] );
+      INTERP_UB( t, dst->v.ub4[4][2], out->v.ub4[4][2], in->v.ub4[4][2] );
+      INTERP_UB( t, dst->v.ub4[4][3], out->v.ub4[4][3], in->v.ub4[4][3] );
+
+      if (DO_TEX0) {
+	 if (DO_PTEX) {
+	    if (HAVE_PTEX_VERTICES) {
+	       INTERP_F( t, dst->v.pv.u0, out->v.pv.u0, in->v.pv.u0 );
+	       INTERP_F( t, dst->v.pv.v0, out->v.pv.v0, in->v.pv.v0 );
+	       INTERP_F( t, dst->v.pv.q0, out->v.pv.q0, in->v.pv.q0 );
+	    } else {
+	       GLfloat wout = out->clip[3]; /* projected clip */
+	       GLfloat win = in->clip[3]; /* projected clip */
+	       GLfloat qout = out->v.pv.w / wout;
+	       GLfloat qin = in->v.pv.w / win;
+	       GLfloat qdst, rqdst;
+
+	       ASSERT( !HAVE_HW_DIVIDE ); /* assert win, wout projected clip */
+
+	       INTERP_F( t, dst->v.v.u0, out->v.v.u0 * qout, in->v.v.u0 * qin );
+	       INTERP_F( t, dst->v.v.v0, out->v.v.v0 * qout, in->v.v.v0 * qin );
+	       INTERP_F( t, qdst, qout, qin );
+
+	       rqdst = 1.0 / qdst;
+	       dst->v.v.u0 *= rqdst;
+	       dst->v.v.v0 *= rqdst;
+	       dst->v.v.w *= rqdst;
+	    }
+	 }
+	 else {
+	    INTERP_F( t, dst->v.v.u0, out->v.v.u0, in->v.v.u0 );
+	    INTERP_F( t, dst->v.v.v0, out->v.v.v0, in->v.v.v0 );
+	 }
+      }
+      if (DO_TEX1) {
+	 if (DO_PTEX) {
+	    INTERP_F( t, dst->v.pv.u1, out->v.pv.u1, in->v.pv.u1 );
+	    INTERP_F( t, dst->v.pv.v1, out->v.pv.v1, in->v.pv.v1 );
+	    INTERP_F( t, dst->v.pv.q1, out->v.pv.q1, in->v.pv.q1 );
+	 } else {
+	    INTERP_F( t, dst->v.v.u1, out->v.v.u1, in->v.v.u1 );
+	    INTERP_F( t, dst->v.v.v1, out->v.v.v1, in->v.v.v1 );
+	 }
+      }
+      else if (DO_PTEX) {
+	 dst->v.pv.q0 = 0.0;	/* must be a valid float on radeon */
+      }
+      if (DO_TEX2) {
+	 if (DO_PTEX) {
+	    INTERP_F( t, dst->v.pv.u2, out->v.pv.u2, in->v.pv.u2 );
+	    INTERP_F( t, dst->v.pv.v2, out->v.pv.v2, in->v.pv.v2 );
+	    INTERP_F( t, dst->v.pv.q2, out->v.pv.q2, in->v.pv.q2 );
+	 } else {
+	    INTERP_F( t, dst->v.v.u2, out->v.v.u2, in->v.v.u2 );
+	    INTERP_F( t, dst->v.v.v2, out->v.v.v2, in->v.v.v2 );
+	 }
+      }
+      if (DO_TEX3) {
+	 if (DO_PTEX) {
+	    INTERP_F( t, dst->v.pv.u3, out->v.pv.u3, in->v.pv.u3 );
+	    INTERP_F( t, dst->v.pv.v3, out->v.pv.v3, in->v.pv.v3 );
+	    INTERP_F( t, dst->v.pv.q3, out->v.pv.q3, in->v.pv.q3 );
+	 } else {
+	    INTERP_F( t, dst->v.v.u3, out->v.v.u3, in->v.v.u3 );
+	    INTERP_F( t, dst->v.v.v3, out->v.v.v3, in->v.v.v3 );
+	 }
+      }
+   } else {
+      /* 4-dword vertex.  Color is in v[3] and there is no oow coordinate.
+       */
+      INTERP_UB( t, dst->v.ub4[3][0], out->v.ub4[3][0], in->v.ub4[3][0] );
+      INTERP_UB( t, dst->v.ub4[3][1], out->v.ub4[3][1], in->v.ub4[3][1] );
+      INTERP_UB( t, dst->v.ub4[3][2], out->v.ub4[3][2], in->v.ub4[3][2] );
+      INTERP_UB( t, dst->v.ub4[3][3], out->v.ub4[3][3], in->v.ub4[3][3] );
+   }
+}
+
+
+static __inline void TAG(copy_pv)( GLcontext *ctx,
+				   TNL_VERTEX *dst, 
+				   TNL_VERTEX *src )
+{
+   if (DO_TEX0 || DO_TEX1 || !HAVE_TINY_VERTICES) {
+      dst->v.v.ui[4] = src->v.v.ui[4];
+   }
+   else {
+      dst->v.v.ui[3] = src->v.v.ui[3];
+   }
+}
+
+
+
+static void TAG(init)( void )
+{
+   setup_tab[IND].emit = TAG(emit_vfmt);
+   setup_tab[IND].interp = TAG(interp_vfmt);
+}
+
+
+#undef IND
+#undef TAG
+
+
+
diff --git a/src/mesa/tnl_dd/t_dd.c b/src/mesa/tnl_dd/t_dd.c
new file mode 100644
index 00000000000..731da5c320d
--- /dev/null
+++ b/src/mesa/tnl_dd/t_dd.c
@@ -0,0 +1,58 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999-2001  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 <[email protected]>
+ */
+
+static void copy_pv_rgba4_spec5( GLcontext *ctx, GLuint edst, GLuint esrc )
+{
+   i810ContextPtr imesa = I810_CONTEXT( ctx );
+   GLubyte *i810verts = (GLubyte *)imesa->verts;
+   GLuint shift = imesa->vertex_stride_shift;
+   i810Vertex *dst = (i810Vertex *)(i810verts + (edst << shift));
+   i810Vertex *src = (i810Vertex *)(i810verts + (esrc << shift));
+   dst->ui[4] = src->ui[4];
+   dst->ui[5] = src->ui[5];
+}
+
+static void copy_pv_rgba4( GLcontext *ctx, GLuint edst, GLuint esrc )
+{
+   i810ContextPtr imesa = I810_CONTEXT( ctx );
+   GLubyte *i810verts = (GLubyte *)imesa->verts;
+   GLuint shift = imesa->vertex_stride_shift;
+   i810Vertex *dst = (i810Vertex *)(i810verts + (edst << shift));
+   i810Vertex *src = (i810Vertex *)(i810verts + (esrc << shift));
+   dst->ui[4] = src->ui[4];
+}
+
+static void copy_pv_rgba3( GLcontext *ctx, GLuint edst, GLuint esrc )
+{
+   i810ContextPtr imesa = I810_CONTEXT( ctx );
+   GLubyte *i810verts = (GLubyte *)imesa->verts;
+   GLuint shift = imesa->vertex_stride_shift;
+   i810Vertex *dst = (i810Vertex *)(i810verts + (edst << shift));
+   i810Vertex *src = (i810Vertex *)(i810verts + (esrc << shift));
+   dst->ui[3] = src->ui[3];
+}
diff --git a/src/mesa/tnl_dd/t_dd_dmatmp.h b/src/mesa/tnl_dd/t_dd_dmatmp.h
new file mode 100644
index 00000000000..d9f709389d1
--- /dev/null
+++ b/src/mesa/tnl_dd/t_dd_dmatmp.h
@@ -0,0 +1,1105 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999-2001  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 <[email protected]>
+ */
+
+
+/* Template for render stages which build and emit vertices directly
+ * to fixed-size dma buffers.  Useful for rendering strips and other
+ * native primitives where clipping and per-vertex tweaks such as
+ * those in t_dd_tritmp.h are not required.
+ *
+ * Produces code for both inline triangles and indexed triangles.
+ * Where various primitive types are unaccelerated by hardware, the
+ * code attempts to fallback to other primitive types (quadstrips to
+ * tristrips, lineloops to linestrips), or to indexed vertices.
+ * Ultimately, a FALLBACK() macro is invoked if there is no way to
+ * render the primitive natively.
+ */
+
+#if !defined(HAVE_TRIANGLES)
+#error "must have at least triangles to use render template"
+#endif
+
+#if !HAVE_ELTS
+#define ELTS_VARS
+#define ALLOC_ELTS( nr )
+#define EMIT_ELT( offset, elt )
+#define EMIT_TWO_ELTS( offset, elt0, elt1 )
+#define INCR_ELTS( nr )
+#define ELT_INIT(prim)
+#define GET_CURRENT_VB_MAX_ELTS() 0
+#define GET_SUBSEQUENT_VB_MAX_ELTS() 0
+#define ALLOC_ELTS_NEW_PRIMITIVE(nr)
+#define RELEASE_ELT_VERTS()
+#define EMIT_INDEXED_VERTS( ctx, start, count )
+#endif
+
+#ifndef EMIT_TWO_ELTS
+#define EMIT_TWO_ELTS( offset, elt0, elt1 )	\
+do { 						\
+   EMIT_ELT( offset, elt0 ); 			\
+   EMIT_ELT( offset+1, elt1 ); 			\
+} while (0)
+#endif
+
+#ifndef FINISH
+#define FINISH
+#endif
+
+/**********************************************************************/
+/*                  Render whole begin/end objects                    */
+/**********************************************************************/
+
+
+
+static GLboolean TAG(emit_elt_verts)( GLcontext *ctx,
+				      GLuint start, GLuint count )
+{
+   if (HAVE_ELTS) {
+      LOCAL_VARS;
+      GLuint nr = count - start;
+
+      if ( nr >= GET_SUBSEQUENT_VB_MAX_VERTS() ) /* assumes same packing for
+						  * indexed and regualar verts
+						  */
+	 return GL_FALSE;
+
+      NEW_PRIMITIVE(); /* finish last prim */
+      EMIT_INDEXED_VERTS( ctx, start, count );
+      return GL_TRUE;
+   } else {
+      return GL_FALSE;
+   }
+}
+
+#if (HAVE_ELTS)
+static void TAG(emit_elts)( GLcontext *ctx, GLuint *elts, GLuint nr )
+{
+   GLint i;
+   LOCAL_VARS;
+   ELTS_VARS;
+
+   ALLOC_ELTS( nr );
+
+   for ( i = 0 ; i < nr ; i+=2, elts += 2 ) {
+      EMIT_TWO_ELTS( 0, elts[0], elts[1] );
+      INCR_ELTS( 2 );
+   }
+}
+#endif
+
+
+/***********************************************************************
+ *                    Render non-indexed primitives.
+ ***********************************************************************/
+
+
+
+static void TAG(render_points_verts)( GLcontext *ctx,
+				      GLuint start,
+				      GLuint count,
+				      GLuint flags )
+{
+   if (HAVE_POINTS) {
+      LOCAL_VARS;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_VERTS();
+      int currentsz = GET_CURRENT_VB_MAX_VERTS();
+      GLuint j, nr;
+
+      INIT( GL_POINTS );
+
+      if (currentsz < 8)
+	 currentsz = dmasz;
+
+      for (j = start; j < count; j += nr ) {
+	 nr = MIN2( currentsz, count - j );
+	 EMIT_VERTS( ctx, j, nr );
+	 currentsz = dmasz;
+      }
+
+      FINISH;
+
+   } else {
+      VERT_FALLBACK( ctx, start, count, flags );
+   }
+}
+
+static void TAG(render_lines_verts)( GLcontext *ctx,
+				     GLuint start,
+				     GLuint count,
+				     GLuint flags )
+{
+   if (HAVE_LINES) {
+      LOCAL_VARS;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_VERTS();
+      int currentsz = GET_CURRENT_VB_MAX_VERTS();
+      GLuint j, nr;
+
+      INIT( GL_LINES );
+
+      /* Emit whole number of lines in total and in each buffer:
+       */
+      count -= (count-start) & 1;
+      currentsz -= currentsz & 1;
+      dmasz -= dmasz & 1;
+
+      if (currentsz < 8)
+	 currentsz = dmasz;
+
+      for (j = start; j < count; j += nr ) {
+	 nr = MIN2( currentsz, count - j );
+	 EMIT_VERTS( ctx, j, nr );
+	 currentsz = dmasz;
+      }
+
+      FINISH;
+
+   } else {
+      VERT_FALLBACK( ctx, start, count, flags );
+   }
+}
+
+
+static void TAG(render_line_strip_verts)( GLcontext *ctx,
+					  GLuint start,
+					  GLuint count,
+					  GLuint flags )
+{
+   if (HAVE_LINE_STRIPS) {
+      LOCAL_VARS;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_VERTS();
+      int currentsz = GET_CURRENT_VB_MAX_VERTS();
+      GLuint j, nr;
+
+      NEW_PRIMITIVE(); /* always a new primitive */
+      INIT( GL_LINE_STRIP );
+
+      if (currentsz < 8)
+	 currentsz = dmasz;
+
+      for (j = start; j + 1 < count; j += nr - 1 ) {
+	 nr = MIN2( currentsz, count - j );
+	 EMIT_VERTS( ctx, j, nr );
+	 currentsz = dmasz;
+      }
+ 
+      FINISH;
+
+   } else {
+      VERT_FALLBACK( ctx, start, count, flags );
+   }
+}
+
+
+static void TAG(render_line_loop_verts)( GLcontext *ctx,
+					 GLuint start,
+					 GLuint count,
+					 GLuint flags )
+{
+   if (HAVE_LINE_STRIPS) {
+      LOCAL_VARS;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_VERTS();
+      int currentsz = GET_CURRENT_VB_MAX_VERTS();
+      GLuint j, nr;
+
+      NEW_PRIMITIVE();
+      INIT( GL_LINE_STRIP );
+
+      if (flags & PRIM_BEGIN)
+	 j = start;
+      else
+	 j = start + 1;
+
+      /* Ensure last vertex won't wrap buffers:
+       */
+      currentsz--;
+      dmasz--;
+
+      if (currentsz < 8) {
+	 NEW_BUFFER();
+	 currentsz = dmasz;
+      }
+
+      if (j + 1 < count) {
+	 for ( ; j + 1 < count; j += nr - 1 ) {
+	    nr = MIN2( currentsz, count - j );
+	    EMIT_VERTS( ctx, j, nr );
+	    currentsz = dmasz;
+	 }
+
+	 if (start < count - 1 && (flags & PRIM_END))
+	    EMIT_VERTS( ctx, start, 1 );
+      }
+      else if (start + 1 < count && (flags & PRIM_END)) {
+	 EMIT_VERTS( ctx, start+1, 1 );
+	 EMIT_VERTS( ctx, start, 1 );
+      }
+
+      FINISH;
+
+   } else {
+      VERT_FALLBACK( ctx, start, count, flags );
+   }
+}
+
+
+static void TAG(render_triangles_verts)( GLcontext *ctx,
+					 GLuint start,
+					 GLuint count,
+					 GLuint flags )
+{
+   LOCAL_VARS;
+   int dmasz = (GET_SUBSEQUENT_VB_MAX_VERTS()/3) * 3;
+   int currentsz = (GET_CURRENT_VB_MAX_VERTS()/3) * 3;
+   GLuint j, nr;
+
+   INIT(GL_TRIANGLES);
+
+   /* Emit whole number of tris in total.  dmasz is already a multiple
+    * of 3.
+    */
+   count -= (count-start)%3;
+
+   if (currentsz < 8)
+      currentsz = dmasz;
+
+   for (j = start; j < count; j += nr) {
+      nr = MIN2( currentsz, count - j );
+      EMIT_VERTS( ctx, j, nr );
+      currentsz = dmasz;
+   }
+   FINISH;
+}
+
+
+
+static void TAG(render_tri_strip_verts)( GLcontext *ctx,
+					 GLuint start,
+					 GLuint count,
+					 GLuint flags )
+{
+   if (HAVE_TRI_STRIPS) {
+      LOCAL_VARS;
+      GLuint j, nr;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_VERTS();
+      int currentsz;
+
+      INIT(GL_TRIANGLE_STRIP);
+      NEW_PRIMITIVE();
+
+      currentsz = GET_CURRENT_VB_MAX_VERTS();
+
+      if (currentsz < 8) {
+	 NEW_BUFFER();
+	 currentsz = dmasz;
+      }
+
+      if ((flags & PRIM_PARITY) && count - start > 2) {
+	 if (HAVE_TRI_STRIP_1 && 0) {
+	 } else {
+	    EMIT_VERTS( ctx, start, 1 );
+	    currentsz--;
+	 }
+      }
+
+      /* From here on emit even numbers of tris when wrapping over buffers:
+       */
+      dmasz -= (dmasz & 1);
+      currentsz -= (currentsz & 1);
+
+      for (j = start ; j + 2 < count; j += nr - 2 ) {
+	 nr = MIN2( currentsz, count - j );
+	 EMIT_VERTS( ctx, j, nr );
+	 currentsz = dmasz;
+      }
+
+      FINISH;
+
+   } else {
+      VERT_FALLBACK( ctx, start, count, flags );
+   }
+}
+
+static void TAG(render_tri_fan_verts)( GLcontext *ctx,
+				       GLuint start,
+				       GLuint count,
+				       GLuint flags )
+{
+   if (HAVE_TRI_FANS) {
+      LOCAL_VARS;
+      GLuint j, nr;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_VERTS();
+      int currentsz = GET_CURRENT_VB_MAX_VERTS();
+
+      NEW_PRIMITIVE();
+      INIT(GL_TRIANGLE_FAN);
+
+      if (currentsz < 8) {
+	 NEW_BUFFER();
+	 currentsz = dmasz;
+      }
+
+      for (j = start + 1 ; j + 1 < count; j += nr - 1 ) {
+	 nr = MIN2( currentsz, count - j + 1 );
+	 EMIT_VERTS( ctx, start, 1 );
+	 EMIT_VERTS( ctx, j, nr - 1 );
+	 currentsz = dmasz;
+      }
+
+      FINISH;
+
+   }
+   else {
+      /* Could write code to emit these as indexed vertices (for the
+       * g400, for instance).
+       */
+      VERT_FALLBACK( ctx, start, count, flags );
+   }
+}
+
+
+static void TAG(render_poly_verts)( GLcontext *ctx,
+				    GLuint start,
+				    GLuint count,
+				    GLuint flags )
+{
+   if (HAVE_POLYGONS) {
+      LOCAL_VARS;
+      GLuint j, nr;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_VERTS();
+      int currentsz = GET_CURRENT_VB_MAX_VERTS();
+
+      NEW_PRIMITIVE();
+      INIT(GL_POLYGON);
+
+      if (currentsz < 8) {
+	 NEW_BUFFER();
+	 currentsz = dmasz;
+      }
+
+      for (j = start + 1 ; j + 1 < count ; j += nr - 1 ) {
+	 nr = MIN2( currentsz, count - j + 1 );
+	 EMIT_VERTS( ctx, start, 1 );
+	 EMIT_VERTS( ctx, j, nr - 1 );
+	 currentsz = dmasz;
+      }
+
+      FINISH;
+
+   }
+   else if (HAVE_TRI_FANS && !(ctx->_TriangleCaps & DD_FLATSHADE)) {
+      TAG(render_tri_fan_verts)( ctx, start, count, flags );
+   } else {
+      VERT_FALLBACK( ctx, start, count, flags );
+   }
+}
+
+static void TAG(render_quad_strip_verts)( GLcontext *ctx,
+					  GLuint start,
+					  GLuint count,
+					  GLuint flags )
+{
+   GLuint j, nr;
+
+   if (HAVE_QUAD_STRIPS) {
+      LOCAL_VARS;
+      GLuint j, nr;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_VERTS();
+      int currentsz;
+
+      INIT(GL_QUAD_STRIP);
+      NEW_PRIMITIVE();
+
+      currentsz = GET_CURRENT_VB_MAX_VERTS();
+
+      if (currentsz < 8) {
+	 NEW_BUFFER();
+	 currentsz = dmasz;
+      }
+
+      dmasz -= (dmasz & 2);
+      currentsz -= (currentsz & 2);
+
+      for (j = start ; j + 3 < count; j += nr - 2 ) {
+	 nr = MIN2( currentsz, count - j );
+	 EMIT_VERTS( ctx, j, nr );
+	 currentsz = dmasz;
+      }
+
+      FINISH;
+
+   } else if (HAVE_TRI_STRIPS && (ctx->_TriangleCaps & DD_FLATSHADE)) {
+      if (TAG(emit_elt_verts)( ctx, start, count )) {
+	 LOCAL_VARS;
+	 int dmasz = GET_SUBSEQUENT_VB_MAX_ELTS();
+	 int currentsz;
+	 GLuint j, nr;
+
+	 /* Simulate flat-shaded quadstrips using indexed vertices:
+	  */
+	 NEW_PRIMITIVE();
+	 ELT_INIT( GL_TRIANGLES );
+
+	 currentsz = GET_CURRENT_VB_MAX_ELTS();
+
+	 /* Emit whole number of quads in total, and in each buffer.
+	  */
+	 dmasz -= dmasz & 1;
+	 count -= (count-start) & 1;
+	 currentsz -= currentsz & 1;
+
+	 if (currentsz < 12)
+	    currentsz = dmasz;
+
+	 currentsz = currentsz/6*2;
+	 dmasz = dmasz/6*2;
+
+	 for (j = start; j + 3 < count; j += nr - 2 ) {
+	    nr = MIN2( currentsz, count - j );
+	    if (nr >= 4) {
+	       GLint quads = (nr/2)-1;
+	       GLint i;
+	       ELTS_VARS;
+
+	       NEW_PRIMITIVE();
+	       ALLOC_ELTS_NEW_PRIMITIVE( quads*6 );
+
+	       for ( i = j-start ; i < j-start+quads*2 ; i+=2 ) {
+		  EMIT_TWO_ELTS( 0, (i+0), (i+1) );
+		  EMIT_TWO_ELTS( 2, (i+2), (i+1) );
+		  EMIT_TWO_ELTS( 4, (i+3), (i+2) );
+		  INCR_ELTS( 6 );
+	       }
+
+	       NEW_PRIMITIVE();
+	    }
+	    currentsz = dmasz;
+	 }
+
+	 RELEASE_ELT_VERTS();
+      }
+      else {
+	 /* Vertices won't fit in a single buffer or elts not available,
+	  * VERT_FALLBACK.
+	  */
+	 VERT_FALLBACK( ctx, start, count, flags );
+      }
+   }
+   else if (HAVE_TRI_STRIPS) {
+      LOCAL_VARS;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_VERTS();
+      int currentsz = GET_CURRENT_VB_MAX_VERTS();
+
+      /* Emit smooth-shaded quadstrips as tristrips:
+       */
+      NEW_PRIMITIVE();
+      INIT( GL_TRIANGLE_STRIP );
+
+      /* Emit whole number of quads in total, and in each buffer.
+       */
+      dmasz -= dmasz & 1;
+      currentsz -= currentsz & 1;
+      count -= (count-start) & 1;
+
+      if (currentsz < 8) {
+	 NEW_BUFFER();
+	 currentsz = dmasz;
+      }
+
+      for (j = start; j + 3 < count; j += nr - 2 ) {
+	 nr = MIN2( currentsz, count - j );
+	 EMIT_VERTS( ctx, j, nr );
+	 currentsz = dmasz;
+      }
+
+      FINISH;
+
+   } else {
+      VERT_FALLBACK( ctx, start, count, flags );
+   }
+}
+
+
+static void TAG(render_quads_verts)( GLcontext *ctx,
+				     GLuint start,
+				     GLuint count,
+				     GLuint flags )
+{
+   if (HAVE_QUADS) {
+      LOCAL_VARS;
+      int dmasz = (GET_SUBSEQUENT_VB_MAX_VERTS()/4) * 4;
+      int currentsz = (GET_CURRENT_VB_MAX_VERTS()/4) * 4;
+      GLuint j, nr;
+
+      INIT(GL_QUADS);
+
+      /* Emit whole number of quads in total.  dmasz is already a multiple
+       * of 4.
+       */
+      count -= (count-start)%4;
+
+      if (currentsz < 8)
+         currentsz = dmasz;
+
+      for (j = start; j < count; j += nr) {
+         nr = MIN2( currentsz, count - j );
+         EMIT_VERTS( ctx, j, nr );
+         currentsz = dmasz;
+      }
+      FINISH;
+   } else if (TAG(emit_elt_verts)( ctx, start, count )) {
+      /* Hardware doesn't have a quad primitive type -- try to
+       * simulate it using indexed vertices and the triangle
+       * primitive:
+       */
+      LOCAL_VARS;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_ELTS();
+      int currentsz;
+      GLuint j, nr;
+
+      NEW_PRIMITIVE();
+      ELT_INIT( GL_TRIANGLES );
+      currentsz = GET_CURRENT_VB_MAX_ELTS();
+
+      /* Emit whole number of quads in total, and in each buffer.
+       */
+      dmasz -= dmasz & 3;
+      count -= (count-start) & 3;
+      currentsz -= currentsz & 3;
+
+      /* Adjust for rendering as triangles:
+       */
+      currentsz = currentsz/6*4;
+      dmasz = dmasz/6*4;
+
+      if (currentsz < 8)
+	 currentsz = dmasz;
+
+      for (j = start; j < count; j += nr ) {
+	 nr = MIN2( currentsz, count - j );
+	 if (nr >= 4) {
+	    GLint quads = nr/4;
+	    GLint i;
+	    ELTS_VARS;
+
+	    NEW_PRIMITIVE();
+	    ALLOC_ELTS_NEW_PRIMITIVE( quads*6 );
+
+	    for ( i = j-start ; i < j-start+quads*4 ; i+=4 ) {
+	       EMIT_TWO_ELTS( 0, (i+0), (i+1) );
+	       EMIT_TWO_ELTS( 2, (i+3), (i+1) );
+	       EMIT_TWO_ELTS( 4, (i+2), (i+3) );
+	       INCR_ELTS( 6 );
+	    }
+
+	    NEW_PRIMITIVE();
+	 }
+	 currentsz = dmasz;
+      }
+
+      RELEASE_ELT_VERTS();
+   }
+   else {
+      /* Vertices won't fit in a single buffer, fallback.
+       */
+      VERT_FALLBACK( ctx, start, count, flags );
+   }
+}
+
+static void TAG(render_noop)( GLcontext *ctx,
+			      GLuint start,
+			      GLuint count,
+			      GLuint flags )
+{
+}
+
+
+
+
+static render_func TAG(render_tab_verts)[GL_POLYGON+2] =
+{
+   TAG(render_points_verts),
+   TAG(render_lines_verts),
+   TAG(render_line_loop_verts),
+   TAG(render_line_strip_verts),
+   TAG(render_triangles_verts),
+   TAG(render_tri_strip_verts),
+   TAG(render_tri_fan_verts),
+   TAG(render_quads_verts),
+   TAG(render_quad_strip_verts),
+   TAG(render_poly_verts),
+   TAG(render_noop),
+};
+
+
+/****************************************************************************
+ *                 Render elts using hardware indexed verts                 *
+ ****************************************************************************/
+
+#if (HAVE_ELTS)
+static void TAG(render_points_elts)( GLcontext *ctx,
+				     GLuint start,
+				     GLuint count,
+				     GLuint flags )
+{
+   if (HAVE_POINTS) {
+      LOCAL_VARS;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_ELTS();
+      int currentsz;
+      GLuint *elts = TNL_CONTEXT(ctx)->vb.Elts;
+      GLuint j, nr;
+
+      ELT_INIT( GL_POINTS );
+
+      currentsz = GET_CURRENT_VB_MAX_ELTS();
+      if (currentsz < 8)
+	 currentsz = dmasz;
+
+      for (j = start; j < count; j += nr ) {
+	 nr = MIN2( currentsz, count - j );
+	 TAG(emit_elts)( ctx, elts+j, nr );
+	 NEW_PRIMITIVE();
+	 currentsz = dmasz;
+      }
+   } else {
+      ELT_FALLBACK( ctx, start, count, flags );
+   }
+}
+
+
+
+static void TAG(render_lines_elts)( GLcontext *ctx,
+				    GLuint start,
+				    GLuint count,
+				    GLuint flags )
+{
+   if (HAVE_LINES) {
+      LOCAL_VARS;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_ELTS();
+      int currentsz;
+      GLuint *elts = TNL_CONTEXT(ctx)->vb.Elts;
+      GLuint j, nr;
+
+      ELT_INIT( GL_LINES );
+
+      /* Emit whole number of lines in total and in each buffer:
+       */
+      count -= (count-start) & 1;
+      currentsz -= currentsz & 1;
+      dmasz -= dmasz & 1;
+
+      currentsz = GET_CURRENT_VB_MAX_ELTS();
+      if (currentsz < 8)
+	 currentsz = dmasz;
+
+      for (j = start; j < count; j += nr ) {
+	 nr = MIN2( currentsz, count - j );
+	 TAG(emit_elts)( ctx, elts+j, nr );
+	 NEW_PRIMITIVE();
+	 currentsz = dmasz;
+      }
+   } else {
+      ELT_FALLBACK( ctx, start, count, flags );
+   }
+}
+
+
+static void TAG(render_line_strip_elts)( GLcontext *ctx,
+					 GLuint start,
+					 GLuint count,
+					 GLuint flags )
+{
+   if (HAVE_LINE_STRIPS) {
+      LOCAL_VARS;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_ELTS();
+      int currentsz;
+      GLuint *elts = TNL_CONTEXT(ctx)->vb.Elts;
+      GLuint j, nr;
+
+      NEW_PRIMITIVE(); /* always a new primitive */
+      ELT_INIT( GL_LINE_STRIP );
+
+      currentsz = GET_CURRENT_VB_MAX_ELTS();
+      if (currentsz < 8)
+	 currentsz = dmasz;
+
+      for (j = start; j + 1 < count; j += nr - 1 ) {
+	 nr = MIN2( currentsz, count - j );
+	 TAG(emit_elts)( ctx, elts+j, nr );
+	 NEW_PRIMITIVE();
+	 currentsz = dmasz;
+      }
+   } else {
+      /* TODO: Try to emit as indexed lines.
+       */
+      ELT_FALLBACK( ctx, start, count, flags );
+   }
+}
+
+
+static void TAG(render_line_loop_elts)( GLcontext *ctx,
+					GLuint start,
+					GLuint count,
+					GLuint flags )
+{
+   if (HAVE_LINE_STRIPS) {
+      LOCAL_VARS;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_ELTS();
+      int currentsz;
+      GLuint *elts = TNL_CONTEXT(ctx)->vb.Elts;
+      GLuint j, nr;
+
+      NEW_PRIMITIVE();
+      ELT_INIT( GL_LINE_STRIP );
+
+      if (flags & PRIM_BEGIN)
+	 j = start;
+      else
+	 j = start + 1;
+
+      currentsz = GET_CURRENT_VB_MAX_ELTS();
+      if (currentsz < 8) {
+	 NEW_BUFFER();
+	 currentsz = dmasz;
+      }
+
+      /* Ensure last vertex doesn't wrap:
+       */
+      currentsz--;
+      dmasz--;
+
+      for ( ; j + 1 < count; j += nr - 1 ) {
+	 nr = MIN2( currentsz, count - j );
+/*  	 NEW_PRIMITIVE(); */
+	 TAG(emit_elts)( ctx, elts+j, nr );
+	 currentsz = dmasz;
+      }
+
+      if (flags & PRIM_END)
+	 TAG(emit_elts)( ctx, elts+start, 1 );
+
+      NEW_PRIMITIVE();
+   } else {
+      /* TODO: Try to emit as indexed lines */
+      ELT_FALLBACK( ctx, start, count, flags );
+   }
+}
+
+
+/* For verts, we still eliminate the copy from main memory to dma
+ * buffers.  For elts, this is probably no better (worse?) than the
+ * standard path.
+ */
+static void TAG(render_triangles_elts)( GLcontext *ctx,
+					GLuint start,
+					GLuint count,
+					GLuint flags )
+{
+   LOCAL_VARS;
+   GLuint *elts = TNL_CONTEXT(ctx)->vb.Elts;
+   int dmasz = GET_SUBSEQUENT_VB_MAX_ELTS()/3*3;
+   int currentsz;
+   GLuint j, nr;
+
+   NEW_PRIMITIVE();
+   ELT_INIT( GL_TRIANGLES );
+
+   currentsz = GET_CURRENT_VB_MAX_ELTS();
+
+   /* Emit whole number of tris in total.  dmasz is already a multiple
+    * of 3.
+    */
+   count -= (count-start)%3;
+   currentsz -= currentsz%3;
+   if (currentsz < 8)
+      currentsz = dmasz;
+
+   for (j = start; j < count; j += nr) {
+      nr = MIN2( currentsz, count - j );
+      TAG(emit_elts)( ctx, elts+j, nr );
+      NEW_PRIMITIVE();
+      currentsz = dmasz;
+   }
+}
+
+
+
+static void TAG(render_tri_strip_elts)( GLcontext *ctx,
+					GLuint start,
+					GLuint count,
+					GLuint flags )
+{
+   if (HAVE_TRI_STRIPS) {
+      LOCAL_VARS;
+      GLuint j, nr;
+      GLuint *elts = TNL_CONTEXT(ctx)->vb.Elts;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_ELTS();
+      int currentsz;
+
+      NEW_PRIMITIVE();
+      ELT_INIT( GL_TRIANGLE_STRIP );
+
+      currentsz = GET_CURRENT_VB_MAX_ELTS();
+      if (currentsz < 8) {
+	 NEW_BUFFER();
+	 currentsz = dmasz;
+      }
+
+      if ((flags & PRIM_PARITY) && count - start > 2) {
+	 TAG(emit_elts)( ctx, elts+start, 1 );
+      }
+
+      /* Keep the same winding over multiple buffers:
+       */
+      dmasz -= (dmasz & 1);
+      currentsz -= (currentsz & 1);
+
+      for (j = start ; j + 2 < count; j += nr - 2 ) {
+	 nr = MIN2( currentsz, count - j );
+	 TAG(emit_elts)( ctx, elts+j, nr );
+	 NEW_PRIMITIVE();
+	 currentsz = dmasz;
+      }
+   } else {
+      /* TODO: try to emit as indexed triangles */
+      ELT_FALLBACK( ctx, start, count, flags );
+   }
+}
+
+static void TAG(render_tri_fan_elts)( GLcontext *ctx,
+				      GLuint start,
+				      GLuint count,
+				      GLuint flags )
+{
+   if (HAVE_TRI_FANS) {
+      LOCAL_VARS;
+      GLuint *elts = TNL_CONTEXT(ctx)->vb.Elts;
+      GLuint j, nr;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_ELTS();
+      int currentsz;
+
+      NEW_PRIMITIVE();
+      ELT_INIT( GL_TRIANGLE_FAN );
+
+      currentsz = GET_CURRENT_VB_MAX_ELTS();
+      if (currentsz < 8) {
+	 NEW_BUFFER();
+	 currentsz = dmasz;
+      }
+
+      for (j = start + 1 ; j + 1 < count; j += nr - 1 ) {
+	 nr = MIN2( currentsz, count - j + 1 );
+	 TAG(emit_elts)( ctx, elts+start, 1 );
+	 TAG(emit_elts)( ctx, elts+j, nr - 1 );
+	 NEW_PRIMITIVE();
+	 currentsz = dmasz;
+      }
+   } else {
+      /* TODO: try to emit as indexed triangles */
+      ELT_FALLBACK( ctx, start, count, flags );
+   }
+}
+
+
+static void TAG(render_poly_elts)( GLcontext *ctx,
+				   GLuint start,
+				   GLuint count,
+				   GLuint flags )
+{
+   if (HAVE_POLYGONS && 0) {
+   } else if (HAVE_TRI_FANS && !(ctx->_TriangleCaps & DD_FLATSHADE)) {
+      LOCAL_VARS;
+      GLuint *elts = TNL_CONTEXT(ctx)->vb.Elts;
+      GLuint j, nr;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_ELTS();
+      int currentsz;
+
+      NEW_PRIMITIVE();
+      ELT_INIT( GL_TRIANGLE_FAN );
+
+      currentsz = GET_CURRENT_VB_MAX_ELTS();
+      if (currentsz < 8) {
+	 NEW_BUFFER();
+	 currentsz = dmasz;
+      }
+
+      for (j = start + 1 ; j + 1 < count ; j += nr - 1 ) {
+	 nr = MIN2( currentsz, count - j + 1 );
+	 TAG(emit_elts)( ctx, elts+start, 1 );
+	 TAG(emit_elts)( ctx, elts+j, nr - 1 );
+	 NEW_PRIMITIVE();
+	 currentsz = dmasz;
+      }
+   } else {
+      ELT_FALLBACK( ctx, start, count, flags );
+   }
+}
+
+static void TAG(render_quad_strip_elts)( GLcontext *ctx,
+					 GLuint start,
+					 GLuint count,
+					 GLuint flags )
+{
+   if (HAVE_QUAD_STRIPS && 0) {
+   }
+   else if (HAVE_TRI_STRIPS) {
+      LOCAL_VARS;
+      GLuint *elts = TNL_CONTEXT(ctx)->vb.Elts;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_ELTS();
+      int currentsz;
+      GLuint j, nr;
+
+      NEW_PRIMITIVE();
+      currentsz = GET_CURRENT_VB_MAX_ELTS();
+
+      /* Emit whole number of quads in total, and in each buffer.
+       */
+      dmasz -= dmasz & 1;
+      count -= (count-start) & 1;
+      currentsz -= currentsz & 1;
+
+      if (currentsz < 12)
+	 currentsz = dmasz;
+
+      if (ctx->_TriangleCaps & DD_FLATSHADE) {
+	 ELT_INIT( GL_TRIANGLES );
+
+	 currentsz = currentsz/6*2;
+	 dmasz = dmasz/6*2;
+
+	 for (j = start; j + 3 < count; j += nr - 2 ) {
+	    nr = MIN2( currentsz, count - j );
+
+	    if (nr >= 4)
+	    {
+	       GLint i;
+	       GLint quads = (nr/2)-1;
+	       ELTS_VARS;
+
+
+	       NEW_PRIMITIVE();
+	       ALLOC_ELTS_NEW_PRIMITIVE( quads*6 );
+
+	       for ( i = j-start ; i < j-start+quads ; i++, elts += 2 ) {
+		  EMIT_TWO_ELTS( 0, elts[0], elts[1] );
+		  EMIT_TWO_ELTS( 2, elts[2], elts[1] );
+		  EMIT_TWO_ELTS( 4, elts[3], elts[2] );
+		  INCR_ELTS( 6 );
+	       }
+
+	       NEW_PRIMITIVE();
+	    }
+
+	    currentsz = dmasz;
+	 }
+      }
+      else {
+	 ELT_INIT( GL_TRIANGLE_STRIP );
+
+	 for (j = start; j + 3 < count; j += nr - 2 ) {
+	    nr = MIN2( currentsz, count - j );
+	    TAG(emit_elts)( ctx, elts+j, nr );
+	    NEW_PRIMITIVE();
+	    currentsz = dmasz;
+	 }
+      }
+   }
+}
+
+
+static void TAG(render_quads_elts)( GLcontext *ctx,
+				    GLuint start,
+				    GLuint count,
+				    GLuint flags )
+{
+   if (HAVE_QUADS && 0) {
+   } else {
+      LOCAL_VARS;
+      GLuint *elts = TNL_CONTEXT(ctx)->vb.Elts;
+      int dmasz = GET_SUBSEQUENT_VB_MAX_ELTS();
+      int currentsz;
+      GLuint j, nr;
+
+      ELT_INIT( GL_TRIANGLES );
+      currentsz = GET_CURRENT_VB_MAX_ELTS();
+
+      /* Emit whole number of quads in total, and in each buffer.
+       */
+      dmasz -= dmasz & 3;
+      count -= (count-start) & 3;
+      currentsz -= currentsz & 3;
+
+      /* Adjust for rendering as triangles:
+       */
+      currentsz = currentsz/6*4;
+      dmasz = dmasz/6*4;
+
+      if (currentsz < 8)
+	 currentsz = dmasz;
+
+      for (j = start; j + 3 < count; j += nr - 2 ) {
+	 nr = MIN2( currentsz, count - j );
+
+	 if (nr >= 4)
+	 {
+	    GLint quads = nr/4;
+	    GLint i;
+	    ELTS_VARS;
+	    NEW_PRIMITIVE();
+	    ALLOC_ELTS_NEW_PRIMITIVE( quads * 6 );
+
+	    for ( i = j-start ; i < j-start+quads ; i++, elts += 4 ) {
+	       EMIT_TWO_ELTS( 0, elts[0], elts[1] );
+	       EMIT_TWO_ELTS( 2, elts[3], elts[1] );
+	       EMIT_TWO_ELTS( 4, elts[2], elts[3] );
+	       INCR_ELTS( 6 );
+	    }
+	 }
+
+	 NEW_PRIMITIVE();
+	 currentsz = dmasz;
+      }
+   }
+}
+
+
+
+static render_func TAG(render_tab_elts)[GL_POLYGON+2] =
+{
+   TAG(render_points_elts),
+   TAG(render_lines_elts),
+   TAG(render_line_loop_elts),
+   TAG(render_line_strip_elts),
+   TAG(render_triangles_elts),
+   TAG(render_tri_strip_elts),
+   TAG(render_tri_fan_elts),
+   TAG(render_quads_elts),
+   TAG(render_quad_strip_elts),
+   TAG(render_poly_elts),
+   TAG(render_noop),
+};
+#endif
diff --git a/src/mesa/tnl_dd/t_dd_dmatmp2.h b/src/mesa/tnl_dd/t_dd_dmatmp2.h
new file mode 100644
index 00000000000..3f58d24c952
--- /dev/null
+++ b/src/mesa/tnl_dd/t_dd_dmatmp2.h
@@ -0,0 +1,949 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  4.0.3
+ *
+ * Copyright (C) 1999-2002  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 <[email protected]>
+ */
+
+
+/* Template for render stages which build and emit vertices directly
+ * to fixed-size dma buffers.  Useful for rendering strips and other
+ * native primitives where clipping and per-vertex tweaks such as
+ * those in t_dd_tritmp.h are not required.
+ *
+ */
+
+#if !HAVE_TRIANGLES || !HAVE_POINTS || !HAVE_LINES
+#error "must have points, lines & triangles to use render template"
+#endif
+
+#if !HAVE_TRI_STRIPS || !HAVE_TRI_FANS
+#error "must have tri strip and fans to use render template"
+#endif
+
+#if !HAVE_LINE_STRIPS
+#error "must have line strips to use render template"
+#endif
+
+#if !HAVE_POLYGONS
+#error "must have polygons to use render template"
+#endif
+
+#if !HAVE_ELTS
+#error "must have elts to use render template"
+#endif
+
+
+#ifndef EMIT_TWO_ELTS
+#define EMIT_TWO_ELTS( dest, offset, elt0, elt1 )	\
+do { 						\
+   (dest)[offset] = (elt0); 			\
+   (dest)[offset+1] = (elt1); 			\
+} while (0)
+#endif
+
+
+/**********************************************************************/
+/*                  Render whole begin/end objects                    */
+/**********************************************************************/
+
+
+static ELT_TYPE *TAG(emit_elts)( GLcontext *ctx, 
+			    ELT_TYPE *dest,
+			    GLuint *elts, GLuint nr )
+{
+   GLint i;
+   LOCAL_VARS;
+
+   for ( i = 0 ; i+1 < nr ; i+=2, elts += 2 ) {
+      EMIT_TWO_ELTS( dest, 0, elts[0], elts[1] );
+      dest += 2;
+   }
+   if (i < nr) {
+      EMIT_ELT( dest, 0, elts[0] );
+      dest += 1;
+   }
+   
+   return dest;
+}
+
+static ELT_TYPE *TAG(emit_consecutive_elts)( GLcontext *ctx, 
+					ELT_TYPE *dest,
+					GLuint start, GLuint nr )
+{
+   GLint i;
+   LOCAL_VARS;
+
+   for ( i = 0 ; i+1 < nr ; i+=2, start += 2 ) {
+      EMIT_TWO_ELTS( dest, 0, start, start+1 );
+      dest += 2;
+   }
+   if (i < nr) {
+      EMIT_ELT( dest, 0, start );
+      dest += 1;
+   }
+
+   return dest;
+}
+
+/***********************************************************************
+ *                    Render non-indexed primitives.
+ ***********************************************************************/
+
+
+
+static void TAG(render_points_verts)( GLcontext *ctx,
+				      GLuint start,
+				      GLuint count,
+				      GLuint flags )
+{
+   if (start < count) {
+      LOCAL_VARS;
+      if (0) fprintf(stderr, "%s\n", __FUNCTION__);
+      EMIT_PRIM( ctx, GL_POINTS, HW_POINTS, start, count );
+   }
+}
+
+static void TAG(render_lines_verts)( GLcontext *ctx,
+				     GLuint start,
+				     GLuint count,
+				     GLuint flags )
+{
+   LOCAL_VARS;
+   if (0) fprintf(stderr, "%s\n", __FUNCTION__);
+   count -= (count-start) & 1;
+
+   if (start+1 >= count)
+      return;
+
+   if ((flags & PRIM_BEGIN) && ctx->Line.StippleFlag) {
+      RESET_STIPPLE();
+      AUTO_STIPPLE( GL_TRUE );
+   }
+      
+   EMIT_PRIM( ctx, GL_LINES, HW_LINES, start, count );
+
+   if ((flags & PRIM_END) && ctx->Line.StippleFlag)
+      AUTO_STIPPLE( GL_FALSE );
+}
+
+
+static void TAG(render_line_strip_verts)( GLcontext *ctx,
+					  GLuint start,
+					  GLuint count,
+					  GLuint flags )
+{
+   LOCAL_VARS;
+   if (0) fprintf(stderr, "%s\n", __FUNCTION__);
+
+   if (start+1 >= count)
+      return;
+
+   if ((flags & PRIM_BEGIN) && ctx->Line.StippleFlag)
+      RESET_STIPPLE();
+
+
+   if (PREFER_DISCRETE_ELT_PRIM( count-start, HW_LINES ))
+   {   
+      int dmasz = GET_MAX_HW_ELTS();
+      GLuint j, nr;
+
+      ELT_INIT( GL_LINES, HW_LINES );
+
+      /* Emit whole number of lines in each full buffer.
+       */
+      dmasz = dmasz/2;
+
+
+      for (j = start; j + 1 < count; j += nr - 1 ) {
+	 ELT_TYPE *dest;
+	 GLint i;
+
+	 nr = MIN2( dmasz, count - j );
+	 dest = ALLOC_ELTS( (nr-1)*2 );
+	    
+	 for ( i = j ; i+1 < j+nr ; i+=1 ) {
+	    EMIT_TWO_ELTS( dest, 0, (i+0), (i+1) );
+	    dest += 2;
+	 }
+
+	 CLOSE_ELTS();
+      }
+   }
+   else
+      EMIT_PRIM( ctx, GL_LINE_STRIP, HW_LINE_STRIP, start, count );
+}
+
+
+static void TAG(render_line_loop_verts)( GLcontext *ctx,
+					 GLuint start,
+					 GLuint count,
+					 GLuint flags )
+{
+   LOCAL_VARS;
+   GLuint j, nr;
+   if (0) fprintf(stderr, "%s\n", __FUNCTION__);
+
+   if (flags & PRIM_BEGIN) {
+      j = start;
+      if (ctx->Line.StippleFlag)
+	 RESET_STIPPLE( );
+   }
+   else
+      j = start + 1;
+
+   if (flags & PRIM_END) {
+
+      if (start+1 >= count)
+	 return;
+
+      if (PREFER_DISCRETE_ELT_PRIM( count-start, HW_LINES )) {
+	 int dmasz = GET_MAX_HW_ELTS();
+
+	 ELT_INIT( GL_LINES, HW_LINES );
+
+	 /* Emit whole number of lines in each full buffer.
+	  */
+	 dmasz = dmasz/2;
+
+	 /* Ensure last vertex doesn't wrap:
+	  */
+	 dmasz--;
+
+	 for (; j + 1 < count;  ) {
+	    GLint i;
+	    ELT_TYPE *dest;
+
+	    nr = MIN2( dmasz, count - j );
+	    dest = ALLOC_ELTS( nr*2 );	/* allocs room for 1 more line */
+
+	    for ( i = 0 ; i < nr - 1 ; i+=1 ) {
+	       EMIT_TWO_ELTS( dest, 0, (j+i), (j+i+1) );
+	       dest += 2;
+	    }
+
+	    j += nr - 1;
+
+	    /* Emit 1 more line into space alloced above */
+	    if (j + 1 >= count) {
+ 	       EMIT_TWO_ELTS( dest, 0, (j), (start) ); 
+ 	       dest += 2; 
+ 	    }
+ 
+	    CLOSE_ELTS();
+	 }
+      }
+      else
+      {
+	 int dmasz = GET_MAX_HW_ELTS() - 1;
+
+	 ELT_INIT( GL_LINE_STRIP, HW_LINE_STRIP );
+
+	 for ( ; j + 1 < count;  ) {
+	    nr = MIN2( dmasz, count - j );
+	    if (j + nr < count) {
+	       ELT_TYPE *dest = ALLOC_ELTS( nr );
+	       dest = TAG(emit_consecutive_elts)( ctx, dest, j, nr );
+	       j += nr - 1;
+	       CLOSE_ELTS();
+	    }
+	    else if (nr) {
+	       ELT_TYPE *dest = ALLOC_ELTS( nr + 1 );
+	       dest = TAG(emit_consecutive_elts)( ctx, dest, j, nr );
+	       dest = TAG(emit_consecutive_elts)( ctx, dest, start, 1 );
+	       j += nr;
+	       CLOSE_ELTS();
+	    }
+	 }   
+      }
+   } else {
+      TAG(render_line_strip_verts)( ctx, j, count, flags );
+   }
+}
+
+
+static void TAG(render_triangles_verts)( GLcontext *ctx,
+					 GLuint start,
+					 GLuint count,
+					 GLuint flags )
+{
+   LOCAL_VARS;
+   if (0) fprintf(stderr, "%s\n", __FUNCTION__);
+
+   count -= (count-start)%3;
+
+   if (start+2 >= count) {
+      return;
+   }
+
+   /* need a PREFER_DISCRETE_ELT_PRIM here too..
+    */
+   EMIT_PRIM( ctx, GL_TRIANGLES, HW_TRIANGLES, start, count );
+}
+
+
+
+static void TAG(render_tri_strip_verts)( GLcontext *ctx,
+					 GLuint start,
+					 GLuint count,
+					 GLuint flags )
+{
+   LOCAL_VARS;
+   if (0) fprintf(stderr, "%s\n", __FUNCTION__);
+
+   if (start + 2 >= count)
+      return;
+
+   if (PREFER_DISCRETE_ELT_PRIM( count-start, HW_TRIANGLES ))
+   {   
+      int dmasz = GET_MAX_HW_ELTS();
+      int parity = 0;
+      GLuint j, nr;
+
+      ELT_INIT( GL_TRIANGLES, HW_TRIANGLES );
+
+      if (flags & PRIM_PARITY)
+	 parity = 1;
+
+      /* Emit even number of tris in each full buffer.
+       */
+      dmasz = dmasz/3;
+      dmasz -= dmasz & 1;
+
+      for (j = start; j + 2 < count; j += nr - 2 ) {
+	 ELT_TYPE *dest;
+	 GLint i;
+
+	 nr = MIN2( dmasz, count - j );
+	 dest = ALLOC_ELTS( (nr-2)*3 );
+	    
+	 for ( i = j ; i+2 < j+nr ; i++, parity^=1 ) {
+	    EMIT_ELT( dest, 0, (i+0+parity) );
+	    EMIT_ELT( dest, 1, (i+1-parity) );
+	    EMIT_ELT( dest, 2, (i+2) );
+	    dest += 3;
+	 }
+
+	 CLOSE_ELTS();
+      }
+   }
+   else if ((flags & PRIM_PARITY) == 0)  
+      EMIT_PRIM( ctx, GL_TRIANGLE_STRIP, HW_TRIANGLE_STRIP_0, start, count );
+   else if (HAVE_TRI_STRIP_1)
+      EMIT_PRIM( ctx, GL_TRIANGLE_STRIP, HW_TRIANGLE_STRIP_1, start, count );
+   else {
+      /* Emit the first triangle with elts, then the rest as a regular strip.
+       * TODO:  Make this unlikely in t_imm_api.c
+       */
+      ELT_TYPE *dest;
+
+      ELT_INIT( GL_TRIANGLES, HW_TRIANGLES );
+      dest = ALLOC_ELTS( 3 );
+      EMIT_ELT( dest, 0, (start+1) );
+      EMIT_ELT( dest, 1, (start+0) );
+      EMIT_ELT( dest, 2, (start+2) );
+      dest += 3;
+      CLOSE_ELTS();
+
+      start++;
+      if (start + 2 >= count)
+	 return;
+
+      EMIT_PRIM( ctx, GL_TRIANGLE_STRIP, HW_TRIANGLE_STRIP_0, start, 
+		 count );
+   }
+}
+
+static void TAG(render_tri_fan_verts)( GLcontext *ctx,
+				       GLuint start,
+				       GLuint count,
+				       GLuint flags )
+{
+   LOCAL_VARS;
+   if (0) fprintf(stderr, "%s\n", __FUNCTION__);
+
+   if (start+2 >= count) 
+      return;
+
+   if (PREFER_DISCRETE_ELT_PRIM( count-start, HW_TRIANGLES ))
+   {   
+      int dmasz = GET_MAX_HW_ELTS();
+      GLuint j, nr;
+
+      ELT_INIT( GL_TRIANGLES, HW_TRIANGLES );
+
+      dmasz = dmasz/3;
+
+      for (j = start + 1; j + 1 < count; j += nr - 1 ) {
+	 ELT_TYPE *dest;
+	 GLint i;
+
+	 nr = MIN2( dmasz, count - j );
+	 dest = ALLOC_ELTS( (nr-1)*3 );
+	    
+	 for ( i = j ; i+1 < j+nr ; i++ ) {
+	    EMIT_ELT( dest, 0, (start) );
+	    EMIT_ELT( dest, 1, (i) );
+	    EMIT_ELT( dest, 2, (i+1) );
+	    dest += 3;
+	 }
+	 
+	 CLOSE_ELTS();
+      }
+   }
+   else {
+      EMIT_PRIM( ctx, GL_TRIANGLE_FAN, HW_TRIANGLE_FAN, start, count );
+   }
+}
+
+
+static void TAG(render_poly_verts)( GLcontext *ctx,
+				    GLuint start,
+				    GLuint count,
+				    GLuint flags )
+{
+   LOCAL_VARS;
+   if (0) fprintf(stderr, "%s\n", __FUNCTION__);
+
+   if (start+2 >= count) 
+      return;
+
+   EMIT_PRIM( ctx, GL_POLYGON, HW_POLYGON, start, count );
+}
+
+static void TAG(render_quad_strip_verts)( GLcontext *ctx,
+					  GLuint start,
+					  GLuint count,
+					  GLuint flags )
+{
+   LOCAL_VARS;
+   if (0) fprintf(stderr, "%s\n", __FUNCTION__);
+
+   count -= (count-start) & 1;
+
+   if (start+3 >= count) 
+      return;
+
+   if (HAVE_QUAD_STRIPS) {
+      EMIT_PRIM( ctx, GL_QUAD_STRIP, HW_QUAD_STRIP, start, count );
+   } 
+   else if (ctx->_TriangleCaps & DD_FLATSHADE) {
+      LOCAL_VARS;
+      int dmasz = GET_MAX_HW_ELTS();
+      GLuint j, nr;
+
+      ELT_INIT( GL_TRIANGLES, HW_TRIANGLES );
+
+      /* Emit whole number of quads in total, and in each buffer.
+       */
+      dmasz = (dmasz/6)*2;
+
+      for (j = start; j + 3 < count; j += nr - 2 ) {
+	 ELT_TYPE *dest;
+	 GLint quads, i;
+
+	 nr = MIN2( dmasz, count - j );
+	 quads = (nr/2)-1;
+	 dest = ALLOC_ELTS( quads*6 );
+	    
+	 for ( i = j ; i < j+quads*2 ; i+=2 ) {
+	    EMIT_TWO_ELTS( dest, 0, (i+0), (i+1) );
+	    EMIT_TWO_ELTS( dest, 2, (i+2), (i+1) );
+	    EMIT_TWO_ELTS( dest, 4, (i+3), (i+2) );
+	    dest += 6;
+	 }
+
+	 CLOSE_ELTS();
+      }
+   }
+   else {
+      EMIT_PRIM( ctx, GL_TRIANGLE_STRIP, HW_TRIANGLE_STRIP_0, start, count );
+   }
+}
+
+
+static void TAG(render_quads_verts)( GLcontext *ctx,
+				     GLuint start,
+				     GLuint count,
+				     GLuint flags )
+{
+   LOCAL_VARS;
+   if (0) fprintf(stderr, "%s\n", __FUNCTION__);
+   count -= (count-start)%4;
+
+   if (start+3 >= count) 
+      return;
+
+   if (HAVE_QUADS) {
+      EMIT_PRIM( ctx, HW_QUADS, GL_QUADS, start, count );
+   } 
+   else {
+      /* Hardware doesn't have a quad primitive type -- simulate it
+       * using indexed vertices and the triangle primitive: 
+       */
+      LOCAL_VARS;
+      int dmasz = GET_MAX_HW_ELTS();
+      GLuint j, nr;
+
+      ELT_INIT( GL_TRIANGLES, HW_TRIANGLES );
+
+      /* Adjust for rendering as triangles:
+       */
+      dmasz = (dmasz/6)*4;
+
+      for (j = start; j < count; j += nr ) {
+	 ELT_TYPE *dest;
+	 GLint quads, i;
+
+	 nr = MIN2( dmasz, count - j );
+	 quads = nr/4;
+	 dest = ALLOC_ELTS( quads*6 );
+
+	 for ( i = j ; i < j+quads*4 ; i+=4 ) {
+	    EMIT_TWO_ELTS( dest, 0, (i+0), (i+1) );
+	    EMIT_TWO_ELTS( dest, 2, (i+3), (i+1) );
+	    EMIT_TWO_ELTS( dest, 4, (i+2), (i+3) );
+	    dest += 6;
+	 }
+
+	 CLOSE_ELTS();
+      }
+   }
+}
+
+static void TAG(render_noop)( GLcontext *ctx,
+			      GLuint start,
+			      GLuint count,
+			      GLuint flags )
+{
+}
+
+
+
+
+static render_func TAG(render_tab_verts)[GL_POLYGON+2] =
+{
+   TAG(render_points_verts),
+   TAG(render_lines_verts),
+   TAG(render_line_loop_verts),
+   TAG(render_line_strip_verts),
+   TAG(render_triangles_verts),
+   TAG(render_tri_strip_verts),
+   TAG(render_tri_fan_verts),
+   TAG(render_quads_verts),
+   TAG(render_quad_strip_verts),
+   TAG(render_poly_verts),
+   TAG(render_noop),
+};
+
+
+/****************************************************************************
+ *                 Render elts using hardware indexed verts                 *
+ ****************************************************************************/
+
+static void TAG(render_points_elts)( GLcontext *ctx,
+				     GLuint start,
+				     GLuint count,
+				     GLuint flags )
+{
+   LOCAL_VARS;
+   int dmasz = GET_MAX_HW_ELTS();
+   GLuint *elts = GET_MESA_ELTS();
+   GLuint j, nr;
+   ELT_TYPE *dest;
+
+   ELT_INIT( GL_POINTS, HW_POINTS );
+
+   for (j = start; j < count; j += nr ) {
+      nr = MIN2( dmasz, count - j );
+      dest = ALLOC_ELTS( nr );
+      dest = TAG(emit_elts)( ctx, dest, elts+j, nr );
+      CLOSE_ELTS();
+   }
+}
+
+
+
+static void TAG(render_lines_elts)( GLcontext *ctx,
+				    GLuint start,
+				    GLuint count,
+				    GLuint flags )
+{
+   LOCAL_VARS;
+   int dmasz = GET_MAX_HW_ELTS();
+   GLuint *elts = GET_MESA_ELTS();
+   GLuint j, nr;
+   ELT_TYPE *dest;
+
+   if (start+1 >= count)
+      return;
+
+   if ((flags & PRIM_BEGIN) && ctx->Line.StippleFlag) {
+      RESET_STIPPLE();
+      AUTO_STIPPLE( GL_TRUE );
+   }
+
+   ELT_INIT( GL_LINES, HW_LINES );
+
+   /* Emit whole number of lines in total and in each buffer:
+    */
+   count -= (count-start) & 1;
+   dmasz -= dmasz & 1;
+
+   for (j = start; j < count; j += nr ) {
+      nr = MIN2( dmasz, count - j );
+      dest = ALLOC_ELTS( nr );
+      dest = TAG(emit_elts)( ctx, dest, elts+j, nr );
+      CLOSE_ELTS();
+   }
+
+   if ((flags & PRIM_END) && ctx->Line.StippleFlag)
+      AUTO_STIPPLE( GL_FALSE );
+}
+
+
+static void TAG(render_line_strip_elts)( GLcontext *ctx,
+					 GLuint start,
+					 GLuint count,
+					 GLuint flags )
+{
+   LOCAL_VARS;
+   int dmasz = GET_MAX_HW_ELTS();
+   GLuint *elts = GET_MESA_ELTS();
+   GLuint j, nr;
+   ELT_TYPE *dest;
+
+   if (start+1 >= count)
+      return;
+
+   ELT_INIT( GL_LINE_STRIP, HW_LINE_STRIP );
+
+   if ((flags & PRIM_BEGIN) && ctx->Line.StippleFlag)
+      RESET_STIPPLE();
+
+   for (j = start; j + 1 < count; j += nr - 1 ) {
+      nr = MIN2( dmasz, count - j );
+      dest = ALLOC_ELTS( nr );
+      dest = TAG(emit_elts)( ctx, dest, elts+j, nr );
+      CLOSE_ELTS();
+   }
+}
+
+
+static void TAG(render_line_loop_elts)( GLcontext *ctx,
+					GLuint start,
+					GLuint count,
+					GLuint flags )
+{
+   LOCAL_VARS;
+   int dmasz = GET_MAX_HW_ELTS();
+   GLuint *elts = GET_MESA_ELTS();
+   GLuint j, nr;
+   ELT_TYPE *dest;
+
+   if (0) fprintf(stderr, "%s\n", __FUNCTION__);
+
+   if (flags & PRIM_BEGIN)
+      j = start;
+   else
+      j = start + 1;
+
+   
+   if (flags & PRIM_END) {
+      if (start+1 >= count)
+	 return;
+   } 
+   else {
+      if (j+1 >= count)
+	 return;
+   }
+
+   ELT_INIT( GL_LINE_STRIP, HW_LINE_STRIP );
+
+   if ((flags & PRIM_BEGIN) && ctx->Line.StippleFlag)
+      RESET_STIPPLE();
+
+   
+   /* Ensure last vertex doesn't wrap:
+    */
+   dmasz--;
+
+   for ( ; j + 1 < count; ) {
+      nr = MIN2( dmasz, count - j );
+      dest = ALLOC_ELTS( nr+1 );	/* Reserve possible space for last elt */
+      dest = TAG(emit_elts)( ctx, dest, elts+j, nr );
+      j += nr - 1;
+      if (j + 1 >= count && (flags & PRIM_END)) {
+	 dest = TAG(emit_elts)( ctx, dest, elts+start, 1 );
+      }
+      CLOSE_ELTS();
+   }
+}
+
+
+static void TAG(render_triangles_elts)( GLcontext *ctx,
+					GLuint start,
+					GLuint count,
+					GLuint flags )
+{
+   LOCAL_VARS;
+   GLuint *elts = GET_MESA_ELTS();
+   int dmasz = GET_MAX_HW_ELTS()/3*3;
+   GLuint j, nr;
+   ELT_TYPE *dest;
+
+   if (start+2 >= count)
+      return;
+
+   ELT_INIT( GL_TRIANGLES, HW_TRIANGLES );
+
+
+   /* Emit whole number of tris in total.  dmasz is already a multiple
+    * of 3.
+    */
+   count -= (count-start)%3;
+
+   for (j = start; j < count; j += nr) {
+      nr = MIN2( dmasz, count - j );
+      dest = ALLOC_ELTS( nr );
+      dest = TAG(emit_elts)( ctx, dest, elts+j, nr );
+      CLOSE_ELTS();
+   }
+}
+
+
+
+static void TAG(render_tri_strip_elts)( GLcontext *ctx,
+					GLuint start,
+					GLuint count,
+					GLuint flags )
+{
+   LOCAL_VARS;
+   GLuint j, nr;
+   GLuint *elts = GET_MESA_ELTS();
+   int dmasz = GET_MAX_HW_ELTS();
+   ELT_TYPE *dest;
+
+   if (start+2 >= count)
+      return;
+
+   ELT_INIT( GL_TRIANGLE_STRIP, HW_TRIANGLE_STRIP_0 );
+
+   /* Keep the same winding over multiple buffers:
+    */
+   dmasz -= (dmasz & 1);
+
+   for (j = start ; j + 2 < count; j += nr - 2 ) {
+      nr = MIN2( dmasz, count - j );
+
+      if (flags & PRIM_PARITY) {
+	 dest = ALLOC_ELTS( nr );	
+	 dest = TAG(emit_elts)( ctx, dest, elts+j, 1 );
+	 dest = TAG(emit_elts)( ctx, dest, elts+j, nr-1 );
+	 nr--; flags &= ~PRIM_PARITY;
+	 CLOSE_ELTS();
+      }
+      else {
+	 dest = ALLOC_ELTS( nr );
+	 dest = TAG(emit_elts)( ctx, dest, elts+j, nr );
+	 CLOSE_ELTS();
+      }
+   }
+}
+
+static void TAG(render_tri_fan_elts)( GLcontext *ctx,
+				      GLuint start,
+				      GLuint count,
+				      GLuint flags )
+{
+   LOCAL_VARS;
+   GLuint *elts = GET_MESA_ELTS();
+   GLuint j, nr;
+   int dmasz = GET_MAX_HW_ELTS();
+   ELT_TYPE *dest;
+
+   if (start+2 >= count)
+      return;
+
+   ELT_INIT( GL_TRIANGLE_FAN, HW_TRIANGLE_FAN );
+
+   for (j = start + 1 ; j + 1 < count; j += nr - 1 ) {
+      nr = MIN2( dmasz, count - j + 1 );
+      dest = ALLOC_ELTS( nr );
+      dest = TAG(emit_elts)( ctx, dest, elts+start, 1 );
+      dest = TAG(emit_elts)( ctx, dest, elts+j, nr - 1 );
+      CLOSE_ELTS();
+   }
+}
+
+
+static void TAG(render_poly_elts)( GLcontext *ctx,
+				   GLuint start,
+				   GLuint count,
+				   GLuint flags )
+{
+   LOCAL_VARS;
+   GLuint *elts = GET_MESA_ELTS();
+   GLuint j, nr;
+   int dmasz = GET_MAX_HW_ELTS();
+   ELT_TYPE *dest;
+
+   if (start+2 >= count)
+      return;
+
+   ELT_INIT( GL_POLYGON, HW_POLYGON );
+
+   for (j = start + 1 ; j + 1 < count ; j += nr - 1 ) {
+      nr = MIN2( dmasz, count - j + 1 );
+      dest = ALLOC_ELTS( nr );
+      dest = TAG(emit_elts)( ctx, dest, elts+start, 1 );
+      dest = TAG(emit_elts)( ctx, dest, elts+j, nr - 1 );
+      CLOSE_ELTS();
+   }
+}
+
+static void TAG(render_quad_strip_elts)( GLcontext *ctx,
+					 GLuint start,
+					 GLuint count,
+					 GLuint flags )
+{
+   if (start+3 >= count)
+      return;
+
+   if (HAVE_QUAD_STRIPS && 0) {
+   }
+   else {
+      LOCAL_VARS;
+      GLuint *elts = GET_MESA_ELTS();
+      int dmasz = GET_MAX_HW_ELTS();
+      GLuint j, nr;
+      ELT_TYPE *dest;
+
+      /* Emit whole number of quads in total, and in each buffer.
+       */
+      dmasz -= dmasz & 1;
+      count -= (count-start) & 1;
+
+      if (ctx->_TriangleCaps & DD_FLATSHADE) {
+	 ELT_INIT( GL_TRIANGLES, HW_TRIANGLES );
+
+	 dmasz = dmasz/6*2;
+
+	 for (j = start; j + 3 < count; j += nr - 2 ) {
+	    nr = MIN2( dmasz, count - j );
+
+	    if (nr >= 4)
+	    {
+	       GLint quads = (nr/2)-1;
+	       ELT_TYPE *dest = ALLOC_ELTS( quads*6 );
+	       GLint i;
+
+	       for ( i = j-start ; i < j-start+quads ; i++, elts += 2 ) {
+		  EMIT_TWO_ELTS( dest, 0, elts[0], elts[1] );
+		  EMIT_TWO_ELTS( dest, 2, elts[2], elts[1] );
+		  EMIT_TWO_ELTS( dest, 4, elts[3], elts[2] );
+		  dest += 6;
+	       }
+
+	       CLOSE_ELTS();
+	    }
+	 }
+      }
+      else {
+	 ELT_INIT( GL_TRIANGLE_STRIP, HW_TRIANGLE_STRIP_0 );
+
+	 for (j = start; j + 3 < count; j += nr - 2 ) {
+	    nr = MIN2( dmasz, count - j );
+	    dest = ALLOC_ELTS( nr );
+	    dest = TAG(emit_elts)( ctx, dest, elts+j, nr );
+	    CLOSE_ELTS();
+	 }
+      }
+   }
+}
+
+
+static void TAG(render_quads_elts)( GLcontext *ctx,
+				    GLuint start,
+				    GLuint count,
+				    GLuint flags )
+{
+   if (start+3 >= count)
+      return;
+
+   if (HAVE_QUADS && 0) {
+   } else {
+      LOCAL_VARS;
+      GLuint *elts = GET_MESA_ELTS();
+      int dmasz = GET_MAX_HW_ELTS();
+      GLuint j, nr;
+
+      ELT_INIT( GL_TRIANGLES, HW_TRIANGLES );
+
+      /* Emit whole number of quads in total, and in each buffer.
+       */
+      dmasz -= dmasz & 3;
+      count -= (count-start) & 3;
+
+      /* Adjust for rendering as triangles:
+       */
+      dmasz = dmasz/6*4;
+
+      for (j = start; j + 3 < count; j += nr ) {
+	 nr = MIN2( dmasz, count - j );
+
+	 {
+	    GLint quads = nr/4;
+	    ELT_TYPE *dest = ALLOC_ELTS( quads * 6 );
+	    GLint i;
+
+	    for ( i = j-start ; i < j-start+quads ; i++, elts += 4 ) {
+	       EMIT_TWO_ELTS( dest, 0, elts[0], elts[1] );
+	       EMIT_TWO_ELTS( dest, 2, elts[3], elts[1] );
+	       EMIT_TWO_ELTS( dest, 4, elts[2], elts[3] );
+	       dest += 6;
+	    }
+
+	    CLOSE_ELTS();
+	 }
+      }
+   }
+}
+
+
+
+static render_func TAG(render_tab_elts)[GL_POLYGON+2] =
+{
+   TAG(render_points_elts),
+   TAG(render_lines_elts),
+   TAG(render_line_loop_elts),
+   TAG(render_line_strip_elts),
+   TAG(render_triangles_elts),
+   TAG(render_tri_strip_elts),
+   TAG(render_tri_fan_elts),
+   TAG(render_quads_elts),
+   TAG(render_quad_strip_elts),
+   TAG(render_poly_elts),
+   TAG(render_noop),
+};
diff --git a/src/mesa/tnl_dd/t_dd_rendertmp.h b/src/mesa/tnl_dd/t_dd_rendertmp.h
new file mode 100644
index 00000000000..fc00e76934b
--- /dev/null
+++ b/src/mesa/tnl_dd/t_dd_rendertmp.h
@@ -0,0 +1,437 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999-2001  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 <[email protected]>
+ */
+
+
+#ifndef POSTFIX
+#define POSTFIX
+#endif
+
+#ifndef INIT
+#define INIT(x)
+#endif
+
+#ifndef NEED_EDGEFLAG_SETUP
+#define NEED_EDGEFLAG_SETUP 0
+#define EDGEFLAG_GET(a) 0
+#define EDGEFLAG_SET(a,b) (void)b
+#endif
+
+#ifndef RESET_STIPPLE
+#define RESET_STIPPLE
+#endif
+
+#ifndef RESET_OCCLUSION
+#define RESET_OCCLUSION
+#endif
+
+#ifndef TEST_PRIM_END
+#define TEST_PRIM_END(flags) (flags & PRIM_END)
+#define TEST_PRIM_BEGIN(flags) (flags & PRIM_BEGIN)
+#define TEST_PRIM_PARITY(flags) (flags & PRIM_PARITY)
+#endif
+
+#ifndef ELT
+#define ELT(x) x
+#endif
+
+#ifndef RENDER_TAB_QUALIFIER
+#define RENDER_TAB_QUALIFIER static
+#endif
+
+static void TAG(render_points)( GLcontext *ctx,
+				GLuint start,
+				GLuint count,
+				GLuint flags )
+{
+   LOCAL_VARS;
+   (void) flags;
+
+   RESET_OCCLUSION;
+   INIT(GL_POINTS);
+   RENDER_POINTS( start, count );
+   POSTFIX;
+}
+
+static void TAG(render_lines)( GLcontext *ctx,
+			       GLuint start,
+			       GLuint count,
+			       GLuint flags )
+{
+   GLuint j;
+   LOCAL_VARS;
+   (void) flags;
+
+   RESET_OCCLUSION;
+   INIT(GL_LINES);
+   for (j=start+1; j<count; j+=2 ) {
+      RENDER_LINE( ELT(j-1), ELT(j) );
+      RESET_STIPPLE;
+   }
+   POSTFIX;
+}
+
+
+static void TAG(render_line_strip)( GLcontext *ctx,
+				    GLuint start,
+				    GLuint count,
+				    GLuint flags )
+{
+   GLuint j;
+   LOCAL_VARS;
+   (void) flags;
+
+   RESET_OCCLUSION;
+   INIT(GL_LINE_STRIP);
+
+   for (j=start+1; j<count; j++ )
+      RENDER_LINE( ELT(j-1), ELT(j) );
+
+   if (TEST_PRIM_END(flags))
+      RESET_STIPPLE;
+
+   POSTFIX;
+}
+
+
+static void TAG(render_line_loop)( GLcontext *ctx,
+				   GLuint start,
+				   GLuint count,
+				   GLuint flags )
+{
+   GLuint i;
+   LOCAL_VARS;
+
+   (void) flags;
+
+   RESET_OCCLUSION;
+   INIT(GL_LINE_LOOP);
+
+   if (start+1 < count) {
+      if (TEST_PRIM_BEGIN(flags)) {
+	 RENDER_LINE( ELT(start), ELT(start+1) );
+      }
+
+      for ( i = start+2 ; i < count ; i++) {
+	 RENDER_LINE( ELT(i-1), ELT(i) );
+      }
+
+      if ( TEST_PRIM_END(flags)) {
+	 RENDER_LINE( ELT(count-1), ELT(start) );
+	 RESET_STIPPLE;
+      }
+   }
+
+   POSTFIX;
+}
+
+
+static void TAG(render_triangles)( GLcontext *ctx,
+				   GLuint start,
+				   GLuint count,
+				   GLuint flags )
+{
+   GLuint j;
+   LOCAL_VARS;
+   (void) flags;
+
+   INIT(GL_TRIANGLES);
+   if (NEED_EDGEFLAG_SETUP) {
+      for (j=start+2; j<count; j+=3) {
+	 /* Leave the edgeflags as supplied by the user.
+	  */
+	 RENDER_TRI( ELT(j-2), ELT(j-1), ELT(j) );
+	 RESET_STIPPLE;
+      }
+   } else {
+      for (j=start+2; j<count; j+=3) {
+	 RENDER_TRI( ELT(j-2), ELT(j-1), ELT(j) );
+      }
+   }
+   POSTFIX;
+}
+
+
+
+static void TAG(render_tri_strip)( GLcontext *ctx,
+				   GLuint start,
+				   GLuint count,
+				   GLuint flags )
+{
+   GLuint j;
+   GLuint parity = 0;
+   LOCAL_VARS;
+
+   if (TEST_PRIM_PARITY(flags))
+      parity = 1;
+
+   INIT(GL_TRIANGLE_STRIP);
+   if (NEED_EDGEFLAG_SETUP) {
+      for (j=start+2;j<count;j++,parity^=1) {
+	 GLuint ej2 = ELT(j-2+parity);
+	 GLuint ej1 = ELT(j-1-parity);
+	 GLuint ej = ELT(j);
+	 GLboolean ef2 = EDGEFLAG_GET( ej2 );
+	 GLboolean ef1 = EDGEFLAG_GET( ej1 );
+	 GLboolean ef = EDGEFLAG_GET( ej );
+	 EDGEFLAG_SET( ej2, GL_TRUE );
+	 EDGEFLAG_SET( ej1, GL_TRUE );
+	 EDGEFLAG_SET( ej, GL_TRUE );
+	 RENDER_TRI( ej2, ej1, ej );
+	 EDGEFLAG_SET( ej2, ef2 );
+	 EDGEFLAG_SET( ej1, ef1 );
+	 EDGEFLAG_SET( ej, ef );
+	 RESET_STIPPLE;
+      }
+   } else {
+      for (j=start+2; j<count ; j++, parity^=1) {
+	 RENDER_TRI( ELT(j-2+parity), ELT(j-1-parity), ELT(j) );
+      }
+   }
+   POSTFIX;
+}
+
+
+static void TAG(render_tri_fan)( GLcontext *ctx,
+				 GLuint start,
+				 GLuint count,
+				 GLuint flags )
+{
+   GLuint j;
+   LOCAL_VARS;
+   (void) flags;
+
+   INIT(GL_TRIANGLE_FAN);
+   if (NEED_EDGEFLAG_SETUP) {
+      for (j=start+2;j<count;j++) {
+	 /* For trifans, all edges are boundary.
+	  */
+	 GLuint ejs = ELT(start);
+	 GLuint ej1 = ELT(j-1);
+	 GLuint ej = ELT(j);
+	 GLboolean efs = EDGEFLAG_GET( ejs );
+	 GLboolean ef1 = EDGEFLAG_GET( ej1 );
+	 GLboolean ef = EDGEFLAG_GET( ej );
+	 EDGEFLAG_SET( ejs, GL_TRUE );
+	 EDGEFLAG_SET( ej1, GL_TRUE );
+	 EDGEFLAG_SET( ej, GL_TRUE );
+	 RENDER_TRI( ejs, ej1, ej);
+	 EDGEFLAG_SET( ejs, efs );
+	 EDGEFLAG_SET( ej1, ef1 );
+	 EDGEFLAG_SET( ej, ef );
+	 RESET_STIPPLE;
+      }
+   } else {
+      for (j=start+2;j<count;j++) {
+	 RENDER_TRI( ELT(start), ELT(j-1), ELT(j) );
+      }
+   }
+
+   POSTFIX;
+}
+
+
+static void TAG(render_poly)( GLcontext *ctx,
+			      GLuint start,
+			      GLuint count,
+			      GLuint flags )
+{
+   GLuint j = start+2;
+   LOCAL_VARS;
+   (void) flags;
+
+   INIT(GL_POLYGON);
+   if (NEED_EDGEFLAG_SETUP) {
+      GLboolean efstart = EDGEFLAG_GET( ELT(start) );
+      GLboolean efcount = EDGEFLAG_GET( ELT(count-1) );
+
+      /* If the primitive does not begin here, the first edge
+       * is non-boundary.
+       */
+      if (!TEST_PRIM_BEGIN(flags))
+	 EDGEFLAG_SET( ELT(start), GL_FALSE );
+
+      /* If the primitive does not end here, the final edge is
+       * non-boundary.
+       */
+      if (!TEST_PRIM_END(flags))
+	 EDGEFLAG_SET( ELT(count-1), GL_FALSE );
+
+      /* Draw the first triangles (possibly zero)
+       */
+      if (j<count-1) {
+	 GLboolean ef = EDGEFLAG_GET( ELT(j) );
+	 EDGEFLAG_SET( ELT(j), GL_FALSE );
+	 RENDER_TRI( ELT(j-1), ELT(j), ELT(start) );
+	 EDGEFLAG_SET( ELT(j), ef );
+	 j++;
+
+	 /* Don't render the first edge again:
+	  */
+	 EDGEFLAG_SET( ELT(start), GL_FALSE );
+
+	 for (;j<count-1;j++) {
+	    GLboolean efj = EDGEFLAG_GET( ELT(j) );
+	    EDGEFLAG_SET( ELT(j), GL_FALSE );
+	    RENDER_TRI( ELT(j-1), ELT(j), ELT(start) );
+	    EDGEFLAG_SET( ELT(j), efj );
+	 }
+      }
+
+      /* Draw the last or only triangle
+       */
+      if (j < count)
+	 RENDER_TRI( ELT(j-1), ELT(j), ELT(start) );
+
+      /* Restore the first and last edgeflags:
+       */
+      EDGEFLAG_SET( ELT(count-1), efcount );
+      EDGEFLAG_SET( ELT(start), efstart );
+
+      if (TEST_PRIM_END(flags)) {
+	 RESET_STIPPLE;
+      }
+   }
+   else {
+      for (j=start+2;j<count;j++) {
+	 RENDER_TRI( ELT(j-1), ELT(j), ELT(start) );
+      }
+   }
+   POSTFIX;
+}
+
+static void TAG(render_quads)( GLcontext *ctx,
+			       GLuint start,
+			       GLuint count,
+			       GLuint flags )
+{
+   GLuint j;
+   LOCAL_VARS;
+   (void) flags;
+
+   INIT(GL_QUADS);
+   if (NEED_EDGEFLAG_SETUP) {
+      for (j=start+3; j<count; j+=4) {
+	 /* Use user-specified edgeflags for quads.
+	  */
+	 RENDER_QUAD( ELT(j-3), ELT(j-2), ELT(j-1), ELT(j) );
+	 RESET_STIPPLE;
+      }
+   } else {
+      for (j=start+3; j<count; j+=4) {
+	 RENDER_QUAD( ELT(j-3), ELT(j-2), ELT(j-1), ELT(j) );
+      }
+   }
+   POSTFIX;
+}
+
+static void TAG(render_quad_strip)( GLcontext *ctx,
+				    GLuint start,
+				    GLuint count,
+				    GLuint flags )
+{
+   GLuint j;
+   LOCAL_VARS;
+   (void) flags;
+
+   INIT(GL_QUAD_STRIP);
+   if (NEED_EDGEFLAG_SETUP) {
+      for (j=start+3;j<count;j+=2) {
+	 /* All edges are boundary.  Set edgeflags to 1, draw the
+	  * quad, and restore them to the original values.
+	  */
+	 GLboolean ef3 = EDGEFLAG_GET( ELT(j-3) );
+	 GLboolean ef2 = EDGEFLAG_GET( ELT(j-2) );
+	 GLboolean ef1 = EDGEFLAG_GET( ELT(j-1) );
+	 GLboolean ef = EDGEFLAG_GET( ELT(j) );
+	 EDGEFLAG_SET( ELT(j-3), GL_TRUE );
+	 EDGEFLAG_SET( ELT(j-2), GL_TRUE );
+	 EDGEFLAG_SET( ELT(j-1), GL_TRUE );
+	 EDGEFLAG_SET( ELT(j), GL_TRUE );
+	 RENDER_QUAD( ELT(j-1), ELT(j-3), ELT(j-2), ELT(j) );
+	 EDGEFLAG_SET( ELT(j-3), ef3 );
+	 EDGEFLAG_SET( ELT(j-2), ef2 );
+	 EDGEFLAG_SET( ELT(j-1), ef1 );
+	 EDGEFLAG_SET( ELT(j), ef );
+	 RESET_STIPPLE;
+      }
+   } else {
+      for (j=start+3;j<count;j+=2) {
+	 RENDER_QUAD( ELT(j-1), ELT(j-3), ELT(j-2), ELT(j) );
+      }
+   }
+   POSTFIX;
+}
+
+static void TAG(render_noop)( GLcontext *ctx,
+			      GLuint start,
+			      GLuint count,
+			      GLuint flags )
+{
+   (void)(ctx && start && count && flags);
+}
+
+RENDER_TAB_QUALIFIER void (*TAG(render_tab)[GL_POLYGON+2])(GLcontext *,
+							   GLuint,
+							   GLuint,
+							   GLuint) =
+{
+   TAG(render_points),
+   TAG(render_lines),
+   TAG(render_line_loop),
+   TAG(render_line_strip),
+   TAG(render_triangles),
+   TAG(render_tri_strip),
+   TAG(render_tri_fan),
+   TAG(render_quads),
+   TAG(render_quad_strip),
+   TAG(render_poly),
+   TAG(render_noop),
+};
+
+
+
+#ifndef PRESERVE_VB_DEFS
+#undef RENDER_TRI
+#undef RENDER_QUAD
+#undef RENDER_LINE
+#undef RENDER_POINTS
+#undef LOCAL_VARS
+#undef INIT
+#undef POSTFIX
+#undef RESET_STIPPLE
+#undef DBG
+#undef ELT
+#undef RENDER_TAB_QUALIFIER
+#endif
+
+#ifndef PRESERVE_TAG
+#undef TAG
+#endif
+
+#undef PRESERVE_VB_DEFS
+#undef PRESERVE_TAG
diff --git a/src/mesa/tnl_dd/t_dd_triemit.h b/src/mesa/tnl_dd/t_dd_triemit.h
new file mode 100644
index 00000000000..2f5bcee7ec4
--- /dev/null
+++ b/src/mesa/tnl_dd/t_dd_triemit.h
@@ -0,0 +1,157 @@
+#ifndef DO_DEBUG_VERTS
+#define DO_DEBUG_VERTS 0
+#endif 
+
+#ifndef PRINT_VERTEX
+#define PRINT_VERTEX(x) 
+#endif
+
+#if defined(USE_X86_ASM)
+#define COPY_DWORDS( j, vb, vertsize, v )				\
+do {									\
+	int __tmp;							\
+	__asm__ __volatile__( "rep ; movsl"				\
+			      : "=%c" (j), "=D" (vb), "=S" (__tmp)	\
+			      : "0" (vertsize),				\
+			        "D" ((long)vb),				\
+			        "S" ((long)v) );			\
+} while (0)
+#else
+#define COPY_DWORDS( j, vb, vertsize, v )				\
+do {									\
+   for ( j = 0 ; j < vertsize ; j++ )					\
+      vb[j] = ((GLuint *)v)[j];						\
+   vb += vertsize;							\
+} while (0)
+#endif
+
+
+
+#if HAVE_QUADS
+static __inline void TAG(quad)( CTX_ARG,
+				VERTEX *v0,
+				VERTEX *v1,
+				VERTEX *v2,
+				VERTEX *v3 )
+{
+   GLuint vertsize = GET_VERTEX_DWORDS();
+   GLuint *vb = (GLuint *)ALLOC_VERTS( 4, vertsize);
+   GLuint j;
+
+   if (DO_DEBUG_VERTS) {
+      fprintf(stderr, "%s\n", __FUNCTION__);
+      PRINT_VERTEX(v0);
+      PRINT_VERTEX(v1);
+      PRINT_VERTEX(v2);
+      PRINT_VERTEX(v3);
+   }
+      
+   COPY_DWORDS( j, vb, vertsize, v0 );
+   COPY_DWORDS( j, vb, vertsize, v1 );
+   COPY_DWORDS( j, vb, vertsize, v2 );
+   COPY_DWORDS( j, vb, vertsize, v3 );
+}
+#else
+static __inline void TAG(quad)( CTX_ARG,
+				VERTEX *v0,
+				VERTEX *v1,
+				VERTEX *v2,
+				VERTEX *v3 )
+{
+   GLuint vertsize = GET_VERTEX_DWORDS();
+   GLuint *vb = (GLuint *)ALLOC_VERTS(  6, vertsize);
+   GLuint j;
+
+   if (DO_DEBUG_VERTS) {
+      fprintf(stderr, "%s\n", __FUNCTION__);
+      PRINT_VERTEX(v0);
+      PRINT_VERTEX(v1);
+      PRINT_VERTEX(v2);
+      PRINT_VERTEX(v3);
+   }
+ 
+   COPY_DWORDS( j, vb, vertsize, v0 );
+   COPY_DWORDS( j, vb, vertsize, v1 );
+   COPY_DWORDS( j, vb, vertsize, v3 );
+   COPY_DWORDS( j, vb, vertsize, v1 );
+   COPY_DWORDS( j, vb, vertsize, v2 );
+   COPY_DWORDS( j, vb, vertsize, v3 );
+}
+#endif
+
+
+static __inline void TAG(triangle)( CTX_ARG,
+				    VERTEX *v0,
+				    VERTEX *v1,
+				    VERTEX *v2 )
+{
+   GLuint vertsize = GET_VERTEX_DWORDS();
+   GLuint *vb = (GLuint *)ALLOC_VERTS( 3, vertsize);
+   GLuint j;
+
+   if (DO_DEBUG_VERTS) {
+      fprintf(stderr, "%s\n", __FUNCTION__);
+      PRINT_VERTEX(v0);
+      PRINT_VERTEX(v1);
+      PRINT_VERTEX(v2);
+   }
+
+   COPY_DWORDS( j, vb, vertsize, v0 );
+   COPY_DWORDS( j, vb, vertsize, v1 );
+   COPY_DWORDS( j, vb, vertsize, v2 );
+}
+
+
+#if HAVE_LINES
+static __inline void TAG(line)( CTX_ARG,
+				VERTEX *v0,
+				VERTEX *v1 )
+{
+   GLuint vertsize = GET_VERTEX_DWORDS();
+   GLuint *vb = (GLuint *)ALLOC_VERTS( 2, vertsize);
+   GLuint j;
+
+   COPY_DWORDS( j, vb, vertsize, v0 );
+   COPY_DWORDS( j, vb, vertsize, v1 );
+}
+#endif
+
+#if HAVE_POINTS
+static __inline void TAG(point)( CTX_ARG,
+				 VERTEX *v0 )
+{
+   GLuint vertsize = GET_VERTEX_DWORDS();
+   GLuint *vb = (GLuint *)ALLOC_VERTS( 1, vertsize);
+   int j;
+
+   COPY_DWORDS( j, vb, vertsize, v0 );
+}
+#endif
+
+
+static void TAG(fast_clipped_poly)( GLcontext *ctx, const GLuint *elts,
+				    GLuint n )
+{
+   LOCAL_VARS
+   GLuint vertsize = GET_VERTEX_DWORDS();
+   GLuint *vb = (GLuint *)ALLOC_VERTS( (n-2) * 3, vertsize );
+   const GLuint *start = (const GLuint *)VERT(elts[0]);
+   int i,j;
+
+   if (DO_DEBUG_VERTS) {
+      fprintf(stderr, "%s\n", __FUNCTION__);
+      PRINT_VERTEX(VERT(elts[0]));
+      PRINT_VERTEX(VERT(elts[1]));
+   }
+
+   for (i = 2 ; i < n ; i++) {
+      if (DO_DEBUG_VERTS) {
+	 PRINT_VERTEX(VERT(elts[i]));
+      }
+
+      COPY_DWORDS( j, vb, vertsize, VERT(elts[i-1]) );
+      COPY_DWORDS( j, vb, vertsize, VERT(elts[i]) );
+      COPY_DWORDS( j, vb, vertsize, start );
+   }
+}
+
diff --git a/src/mesa/tnl_dd/t_dd_tritmp.h b/src/mesa/tnl_dd/t_dd_tritmp.h
new file mode 100644
index 00000000000..cc47798d5e4
--- /dev/null
+++ b/src/mesa/tnl_dd/t_dd_tritmp.h
@@ -0,0 +1,739 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999-2001  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 <[email protected]>
+ */
+
+
+/* Template for building functions to plug into the driver interface
+ * of t_vb_render.c:
+ *     ctx->Driver.QuadFunc
+ *     ctx->Driver.TriangleFunc
+ *     ctx->Driver.LineFunc
+ *     ctx->Driver.PointsFunc
+ *
+ * DO_TWOSIDE:   Plug back-color values from the VB into backfacing triangles,
+ *               and restore vertices afterwards.
+ * DO_OFFSET:    Calculate offset for triangles and adjust vertices.  Restore
+ *               vertices after rendering.
+ * DO_FLAT:      For hardware without native flatshading, copy provoking colors
+ *               into the other vertices.  Restore after rendering.
+ * DO_UNFILLED:  Decompose triangles to lines and points where appropriate.
+ *
+ * HAVE_RGBA: Vertices have rgba values (otherwise index values).
+ * HAVE_SPEC: Vertices have secondary rgba values.
+ *
+ * VERT_X(v): Alias for vertex x value.
+ * VERT_Y(v): Alias for vertex y value.
+ * VERT_Z(v): Alias for vertex z value.
+ * DEPTH_SCALE: Scale for offset.
+ *
+ * VERTEX: Hardware vertex type.
+ * GET_VERTEX(n): Retreive vertex with index n.
+ * AREA_IS_CCW(a): Return true if triangle with signed area a is ccw.
+ *
+ * VERT_SET_RGBA: Assign vertex rgba from VB color.
+ * VERT_COPY_RGBA: Copy vertex rgba another vertex.
+ * VERT_SAVE_RGBA: Save vertex rgba to a local variable.
+ * VERT_RESTORE_RGBA: Restore vertex rgba from a local variable.
+ *   --> Similar for IND and SPEC.
+ *
+ * LOCAL_VARS(n): (At least) define local vars for save/restore rgba.
+ *
+ */
+
+#if HAVE_RGBA
+#define VERT_SET_IND( v, c ) (void) c
+#define VERT_COPY_IND( v0, v1 )
+#define VERT_SAVE_IND( idx )
+#define VERT_RESTORE_IND( idx )
+#if HAVE_BACK_COLORS
+#define VERT_SET_RGBA( v, c )
+#endif
+#else
+#define VERT_SET_RGBA( v, c ) (void) c
+#define VERT_COPY_RGBA( v0, v1 )
+#define VERT_SAVE_RGBA( idx )
+#define VERT_RESTORE_RGBA( idx )
+#if HAVE_BACK_COLORS
+#define VERT_SET_IND( v, c )
+#endif
+#endif
+
+#if !HAVE_SPEC
+#define VERT_SET_SPEC( v, c ) (void) c
+#define VERT_COPY_SPEC( v0, v1 )
+#define VERT_SAVE_SPEC( idx )
+#define VERT_RESTORE_SPEC( idx )
+#if HAVE_BACK_COLORS
+#define VERT_COPY_SPEC1( v )
+#endif
+#else
+#if HAVE_BACK_COLORS
+#define VERT_SET_SPEC( v, c )
+#endif
+#endif
+
+#if !HAVE_BACK_COLORS
+#define VERT_COPY_SPEC1( v )
+#define VERT_COPY_IND1( v )
+#define VERT_COPY_RGBA1( v )
+#endif
+
+#ifndef INSANE_VERTICES
+#define VERT_SET_Z(v,val) VERT_Z(v) = val
+#define VERT_Z_ADD(v,val) VERT_Z(v) += val
+#endif
+
+#if DO_TRI
+static void TAG(triangle)( GLcontext *ctx, GLuint e0, GLuint e1, GLuint e2 )
+{
+   struct vertex_buffer *VB = &TNL_CONTEXT( ctx )->vb;
+   VERTEX *v[3];
+   GLfloat offset;
+   GLfloat z[3];
+   GLenum mode = GL_FILL;
+   GLuint facing;
+   LOCAL_VARS(3);
+
+/*     fprintf(stderr, "%s\n", __FUNCTION__); */
+
+   v[0] = (VERTEX *)GET_VERTEX(e0);
+   v[1] = (VERTEX *)GET_VERTEX(e1);
+   v[2] = (VERTEX *)GET_VERTEX(e2);
+
+   if (DO_TWOSIDE || DO_OFFSET || DO_UNFILLED)
+   {
+      GLfloat ex = VERT_X(v[0]) - VERT_X(v[2]);
+      GLfloat ey = VERT_Y(v[0]) - VERT_Y(v[2]);
+      GLfloat fx = VERT_X(v[1]) - VERT_X(v[2]);
+      GLfloat fy = VERT_Y(v[1]) - VERT_Y(v[2]);
+      GLfloat cc = ex*fy - ey*fx;
+
+      if (DO_TWOSIDE || DO_UNFILLED)
+      {
+	 facing = AREA_IS_CCW( cc ) ^ ctx->Polygon._FrontBit;
+
+	 if (DO_UNFILLED) {
+	    if (facing) {
+	       mode = ctx->Polygon.BackMode;
+	       if (ctx->Polygon.CullFlag &&
+		   ctx->Polygon.CullFaceMode != GL_FRONT) {
+		  return;
+	       }
+	    } else {
+	       mode = ctx->Polygon.FrontMode;
+	       if (ctx->Polygon.CullFlag &&
+		   ctx->Polygon.CullFaceMode != GL_BACK) {
+		  return;
+	       }
+	    }
+	 }
+
+	 if (DO_TWOSIDE && facing == 1)
+	 {
+	    if (HAVE_RGBA) {
+	       if (HAVE_BACK_COLORS) {
+		  if (!DO_FLAT) {
+		     VERT_SAVE_RGBA( 0 );
+		     VERT_SAVE_RGBA( 1 );
+		     VERT_COPY_RGBA1( v[0] );
+		     VERT_COPY_RGBA1( v[1] );
+		  }
+		  VERT_SAVE_RGBA( 2 );
+		  VERT_COPY_RGBA1( v[2] );
+		  if (HAVE_SPEC) {
+		     if (!DO_FLAT) {
+			VERT_SAVE_SPEC( 0 );
+			VERT_SAVE_SPEC( 1 );
+			VERT_COPY_SPEC1( v[0] );
+			VERT_COPY_SPEC1( v[1] );
+		     }
+		     VERT_SAVE_SPEC( 2 );
+		     VERT_COPY_SPEC1( v[2] );
+		  }
+	       }
+	       else {
+		  GLchan (*vbcolor)[4] = VB->ColorPtr[1]->Ptr;
+		  ASSERT(VB->ColorPtr[1]->Type == CHAN_TYPE);
+		  ASSERT(VB->ColorPtr[1]->StrideB == 4*sizeof(GLchan));
+		  (void) vbcolor;
+
+		  if (!DO_FLAT) {
+		     VERT_SAVE_RGBA( 0 );
+		     VERT_SAVE_RGBA( 1 );
+		     VERT_SET_RGBA( v[0], vbcolor[e0] );
+		     VERT_SET_RGBA( v[1], vbcolor[e1] );
+		  }
+		  VERT_SAVE_RGBA( 2 );
+		  VERT_SET_RGBA( v[2], vbcolor[e2] );
+
+		  if (HAVE_SPEC && VB->SecondaryColorPtr[1]) {
+		     GLchan (*vbspec)[4] = VB->SecondaryColorPtr[1]->Ptr;
+
+		     if (!DO_FLAT) {
+			VERT_SAVE_SPEC( 0 );
+			VERT_SAVE_SPEC( 1 );
+			VERT_SET_SPEC( v[0], vbspec[e0] );
+			VERT_SET_SPEC( v[1], vbspec[e1] );
+		     }
+		     VERT_SAVE_SPEC( 2 );
+		     VERT_SET_SPEC( v[2], vbspec[e2] );
+		  }
+	       }
+	    }
+	    else {
+	       GLuint *vbindex = VB->IndexPtr[1]->data;
+	       if (!DO_FLAT) {
+		  VERT_SET_IND( v[0], vbindex[e0] );
+		  VERT_SET_IND( v[1], vbindex[e1] );
+	       }
+	       VERT_SET_IND( v[2], vbindex[e2] );
+	    }
+	 }
+      }
+
+
+      if (DO_OFFSET)
+      {
+	 offset = ctx->Polygon.OffsetUnits * DEPTH_SCALE;
+	 z[0] = VERT_Z(v[0]);
+	 z[1] = VERT_Z(v[1]);
+	 z[2] = VERT_Z(v[2]);
+	 if (cc * cc > 1e-16) {
+	    GLfloat ic	= 1.0 / cc;
+	    GLfloat ez	= z[0] - z[2];
+	    GLfloat fz	= z[1] - z[2];
+	    GLfloat a	= ey*fz - ez*fy;
+	    GLfloat b	= ez*fx - ex*fz;
+	    GLfloat ac	= a * ic;
+	    GLfloat bc	= b * ic;
+	    if ( ac < 0.0f ) ac = -ac;
+	    if ( bc < 0.0f ) bc = -bc;
+	    offset += MAX2( ac, bc ) * ctx->Polygon.OffsetFactor;
+	 }
+	 offset *= ctx->MRD;
+      }
+   }
+
+   if (DO_FLAT) {
+      if (HAVE_RGBA) {
+	 VERT_SAVE_RGBA( 0 );
+	 VERT_SAVE_RGBA( 1 );
+	 VERT_COPY_RGBA( v[0], v[2] );
+	 VERT_COPY_RGBA( v[1], v[2] );
+	 if (HAVE_SPEC && VB->SecondaryColorPtr[0]) {
+	    VERT_SAVE_SPEC( 0 );
+	    VERT_SAVE_SPEC( 1 );
+	    VERT_COPY_SPEC( v[0], v[2] );
+	    VERT_COPY_SPEC( v[1], v[2] );
+	 }
+      }
+      else {
+	 VERT_SAVE_IND( 0 );
+	 VERT_SAVE_IND( 1 );
+	 VERT_COPY_IND( v[0], v[2] );
+	 VERT_COPY_IND( v[1], v[2] );
+      }
+   }
+
+   if (mode == GL_POINT) {
+      if (DO_OFFSET && ctx->Polygon.OffsetPoint) {
+	 VERT_Z_ADD(v[0], offset);
+	 VERT_Z_ADD(v[1], offset);
+	 VERT_Z_ADD(v[2], offset);
+      }
+      UNFILLED_TRI( ctx, GL_POINT, e0, e1, e2 );
+   } else if (mode == GL_LINE) {
+      if (DO_OFFSET && ctx->Polygon.OffsetLine) {
+	 VERT_Z_ADD(v[0], offset);
+	 VERT_Z_ADD(v[1], offset);
+	 VERT_Z_ADD(v[2], offset);
+      }
+      UNFILLED_TRI( ctx, GL_LINE, e0, e1, e2 );
+   } else {
+      if (DO_OFFSET && ctx->Polygon.OffsetFill) {
+	 VERT_Z_ADD(v[0], offset);
+	 VERT_Z_ADD(v[1], offset);
+	 VERT_Z_ADD(v[2], offset);
+      }
+      if (DO_UNFILLED)
+	 RASTERIZE( GL_TRIANGLES );
+      TRI( v[0], v[1], v[2] );
+   }
+
+   if (DO_OFFSET)
+   {
+      VERT_SET_Z(v[0], z[0]);
+      VERT_SET_Z(v[1], z[1]);
+      VERT_SET_Z(v[2], z[2]);
+   }
+
+   if (DO_TWOSIDE && facing == 1)
+   {
+      if (HAVE_RGBA) {
+	 if (!DO_FLAT) {
+	    VERT_RESTORE_RGBA( 0 );
+	    VERT_RESTORE_RGBA( 1 );
+	 }
+	 VERT_RESTORE_RGBA( 2 );
+	 if (HAVE_SPEC) {
+	    if (!DO_FLAT) {
+	       VERT_RESTORE_SPEC( 0 );
+	       VERT_RESTORE_SPEC( 1 );
+	    }
+	    VERT_RESTORE_SPEC( 2 );
+	 }
+      }
+      else {
+	 GLuint *vbindex = VB->IndexPtr[0]->data;
+	 if (!DO_FLAT) {
+	    VERT_SET_IND( v[0], vbindex[e0] );
+	    VERT_SET_IND( v[1], vbindex[e1] );
+	 }
+	 VERT_SET_IND( v[2], vbindex[e2] );
+      }
+   }
+
+
+   if (DO_FLAT) {
+      if (HAVE_RGBA) {
+	 VERT_RESTORE_RGBA( 0 );
+	 VERT_RESTORE_RGBA( 1 );
+	 if (HAVE_SPEC && VB->SecondaryColorPtr[0]) {
+	    VERT_RESTORE_SPEC( 0 );
+	    VERT_RESTORE_SPEC( 1 );
+	 }
+      }
+      else {
+	 VERT_RESTORE_IND( 0 );
+	 VERT_RESTORE_IND( 1 );
+      }
+   }
+}
+#endif
+
+#if DO_QUAD
+#if DO_FULL_QUAD
+static void TAG(quad)( GLcontext *ctx,
+		       GLuint e0, GLuint e1, GLuint e2, GLuint e3 )
+{
+   struct vertex_buffer *VB = &TNL_CONTEXT( ctx )->vb;
+   VERTEX *v[4];
+   GLfloat offset;
+   GLfloat z[4];
+   GLenum mode = GL_FILL;
+   GLuint facing;
+   LOCAL_VARS(4);
+
+   v[0] = (VERTEX *)GET_VERTEX(e0);
+   v[1] = (VERTEX *)GET_VERTEX(e1);
+   v[2] = (VERTEX *)GET_VERTEX(e2);
+   v[3] = (VERTEX *)GET_VERTEX(e3);
+
+   if (DO_TWOSIDE || DO_OFFSET || DO_UNFILLED)
+   {
+      GLfloat ex = VERT_X(v[2]) - VERT_X(v[0]);
+      GLfloat ey = VERT_Y(v[2]) - VERT_Y(v[0]);
+      GLfloat fx = VERT_X(v[3]) - VERT_X(v[1]);
+      GLfloat fy = VERT_Y(v[3]) - VERT_Y(v[1]);
+      GLfloat cc = ex*fy - ey*fx;
+
+      if (DO_TWOSIDE || DO_UNFILLED)
+      {
+	 facing = AREA_IS_CCW( cc ) ^ ctx->Polygon._FrontBit;
+
+	 if (DO_UNFILLED) {
+	    if (facing) {
+	       mode = ctx->Polygon.BackMode;
+	       if (ctx->Polygon.CullFlag &&
+		   ctx->Polygon.CullFaceMode != GL_FRONT) {
+		  return;
+	       }
+	    } else {
+	       mode = ctx->Polygon.FrontMode;
+	       if (ctx->Polygon.CullFlag &&
+		   ctx->Polygon.CullFaceMode != GL_BACK) {
+		  return;
+	       }
+	    }
+	 }
+
+	 if (DO_TWOSIDE && facing == 1)
+	 {
+	    if (HAVE_RGBA) {
+	       GLchan (*vbcolor)[4] = VB->ColorPtr[1]->Ptr;
+	       (void)vbcolor;
+
+	       if (HAVE_BACK_COLORS) {
+                  if (!DO_FLAT) {
+                     VERT_SAVE_RGBA( 0 );
+                     VERT_SAVE_RGBA( 1 );
+                     VERT_SAVE_RGBA( 2 );
+		     VERT_COPY_RGBA1( v[0] );
+		     VERT_COPY_RGBA1( v[1] );
+		     VERT_COPY_RGBA1( v[2] );
+		  }
+		  VERT_SAVE_RGBA( 3 );
+		  VERT_COPY_RGBA1( v[3] );
+		  if (HAVE_SPEC) {
+                     if (!DO_FLAT) {
+                        VERT_SAVE_SPEC( 0 );
+                        VERT_SAVE_SPEC( 1 );
+                        VERT_SAVE_SPEC( 2 );
+			VERT_COPY_SPEC1( v[0] );
+			VERT_COPY_SPEC1( v[1] );
+			VERT_COPY_SPEC1( v[2] );
+		     }
+		     VERT_SAVE_SPEC( 3 );
+		     VERT_COPY_SPEC1( v[3] );
+		  }
+	       }
+	       else {
+	          if (!DO_FLAT) {
+		     VERT_SAVE_RGBA( 0 );
+		     VERT_SAVE_RGBA( 1 );
+		     VERT_SAVE_RGBA( 2 );
+		     VERT_SET_RGBA( v[0], vbcolor[e0] );
+		     VERT_SET_RGBA( v[1], vbcolor[e1] );
+		     VERT_SET_RGBA( v[2], vbcolor[e2] );
+	          }
+	          VERT_SAVE_RGBA( 3 );
+	          VERT_SET_RGBA( v[3], vbcolor[e3] );
+
+	          if (HAVE_SPEC && VB->SecondaryColorPtr[1]) {
+		     GLchan (*vbspec)[4] = VB->SecondaryColorPtr[1]->Ptr;
+		     ASSERT(VB->SecondaryColorPtr[1]->StrideB==4*sizeof(GLchan));
+
+		     if (!DO_FLAT) {
+		        VERT_SAVE_SPEC( 0 );
+		        VERT_SAVE_SPEC( 1 );
+		        VERT_SAVE_SPEC( 2 );
+		        VERT_SET_SPEC( v[0], vbspec[e0] );
+		        VERT_SET_SPEC( v[1], vbspec[e1] );
+		        VERT_SET_SPEC( v[2], vbspec[e2] );
+		     }
+		     VERT_SAVE_SPEC( 3 );
+		     VERT_SET_SPEC( v[3], vbspec[e3] );
+	          }
+	       }
+	    }
+	    else {
+	       GLuint *vbindex = VB->IndexPtr[1]->data;
+	       if (!DO_FLAT) {
+		  VERT_SET_IND( v[0], vbindex[e0] );
+		  VERT_SET_IND( v[1], vbindex[e1] );
+		  VERT_SET_IND( v[2], vbindex[e2] );
+	       }
+	       VERT_SET_IND( v[3], vbindex[e3] );
+	    }
+	 }
+      }
+
+
+      if (DO_OFFSET)
+      {
+	 offset = ctx->Polygon.OffsetUnits * DEPTH_SCALE;
+	 z[0] = VERT_Z(v[0]);
+	 z[1] = VERT_Z(v[1]);
+	 z[2] = VERT_Z(v[2]);
+	 z[3] = VERT_Z(v[3]);
+	 if (cc * cc > 1e-16) {
+	    GLfloat ez = z[2] - z[0];
+	    GLfloat fz = z[3] - z[1];
+	    GLfloat a	= ey*fz - ez*fy;
+	    GLfloat b	= ez*fx - ex*fz;
+	    GLfloat ic	= 1.0 / cc;
+	    GLfloat ac	= a * ic;
+	    GLfloat bc	= b * ic;
+	    if ( ac < 0.0f ) ac = -ac;
+	    if ( bc < 0.0f ) bc = -bc;
+	    offset += MAX2( ac, bc ) * ctx->Polygon.OffsetFactor;
+	 }
+	 offset *= ctx->MRD;
+      }
+   }
+
+   if (DO_FLAT) {
+      if (HAVE_RGBA) {
+	 VERT_SAVE_RGBA( 0 );
+	 VERT_SAVE_RGBA( 1 );
+	 VERT_SAVE_RGBA( 2 );
+	 VERT_COPY_RGBA( v[0], v[3] );
+	 VERT_COPY_RGBA( v[1], v[3] );
+	 VERT_COPY_RGBA( v[2], v[3] );
+	 if (HAVE_SPEC && VB->SecondaryColorPtr[0]) {
+	    VERT_SAVE_SPEC( 0 );
+	    VERT_SAVE_SPEC( 1 );
+	    VERT_SAVE_SPEC( 2 );
+	    VERT_COPY_SPEC( v[0], v[3] );
+	    VERT_COPY_SPEC( v[1], v[3] );
+	    VERT_COPY_SPEC( v[2], v[3] );
+	 }
+      }
+      else {
+	 VERT_SAVE_IND( 0 );
+	 VERT_SAVE_IND( 1 );
+	 VERT_SAVE_IND( 2 );
+	 VERT_COPY_IND( v[0], v[3] );
+	 VERT_COPY_IND( v[1], v[3] );
+	 VERT_COPY_IND( v[2], v[3] );
+      }
+   }
+
+   if (mode == GL_POINT) {
+      if (( DO_OFFSET) && ctx->Polygon.OffsetPoint) {
+	 VERT_Z_ADD(v[0], offset);
+	 VERT_Z_ADD(v[1], offset);
+	 VERT_Z_ADD(v[2], offset);
+	 VERT_Z_ADD(v[3], offset);
+      }
+      UNFILLED_QUAD( ctx, GL_POINT, e0, e1, e2, e3 );
+   } else if (mode == GL_LINE) {
+      if (DO_OFFSET && ctx->Polygon.OffsetLine) {
+	 VERT_Z_ADD(v[0], offset);
+	 VERT_Z_ADD(v[1], offset);
+	 VERT_Z_ADD(v[2], offset);
+	 VERT_Z_ADD(v[3], offset);
+      }
+      UNFILLED_QUAD( ctx, GL_LINE, e0, e1, e2, e3 );
+   } else {
+      if (DO_OFFSET && ctx->Polygon.OffsetFill) {
+	 VERT_Z_ADD(v[0], offset);
+	 VERT_Z_ADD(v[1], offset);
+	 VERT_Z_ADD(v[2], offset);
+	 VERT_Z_ADD(v[3], offset);
+      }
+      RASTERIZE( GL_TRIANGLES );
+      QUAD( (v[0]), (v[1]), (v[2]), (v[3]) );
+   }
+
+   if (DO_OFFSET)
+   {
+      VERT_SET_Z(v[0], z[0]);
+      VERT_SET_Z(v[1], z[1]);
+      VERT_SET_Z(v[2], z[2]);
+      VERT_SET_Z(v[3], z[3]);
+   }
+
+   if (DO_TWOSIDE && facing == 1)
+   {
+      if (HAVE_RGBA) {
+	 if (!DO_FLAT) {
+	    VERT_RESTORE_RGBA( 0 );
+	    VERT_RESTORE_RGBA( 1 );
+	    VERT_RESTORE_RGBA( 2 );
+	 }
+	 VERT_RESTORE_RGBA( 3 );
+	 if (HAVE_SPEC) {
+	    if (!DO_FLAT) {
+	       VERT_RESTORE_SPEC( 0 );
+	       VERT_RESTORE_SPEC( 1 );
+	       VERT_RESTORE_SPEC( 2 );
+	    }
+	    VERT_RESTORE_SPEC( 3 );
+	 }
+      }
+      else {
+	 GLuint *vbindex = VB->IndexPtr[0]->data;
+	 if (!DO_FLAT) {
+	    VERT_SET_IND( v[0], vbindex[e0] );
+	    VERT_SET_IND( v[1], vbindex[e1] );
+	    VERT_SET_IND( v[2], vbindex[e2] );
+	 }
+	 VERT_SET_IND( v[3], vbindex[e3] );
+      }
+   }
+
+
+   if (DO_FLAT) {
+      if (HAVE_RGBA) {
+	 VERT_RESTORE_RGBA( 0 );
+	 VERT_RESTORE_RGBA( 1 );
+	 VERT_RESTORE_RGBA( 2 );
+	 if (HAVE_SPEC && VB->SecondaryColorPtr[0]) {
+	    VERT_RESTORE_SPEC( 0 );
+	    VERT_RESTORE_SPEC( 1 );
+	    VERT_RESTORE_SPEC( 2 );
+	 }
+      }
+      else {
+	 VERT_RESTORE_IND( 0 );
+	 VERT_RESTORE_IND( 1 );
+	 VERT_RESTORE_IND( 2 );
+      }
+   }
+}
+#else
+static void TAG(quad)( GLcontext *ctx, GLuint e0,
+		       GLuint e1, GLuint e2, GLuint e3 )
+{
+   if (DO_UNFILLED) {
+      struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+      GLubyte ef1 = VB->EdgeFlag[e1];
+      GLubyte ef3 = VB->EdgeFlag[e3];
+      VB->EdgeFlag[e1] = 0;
+      TAG(triangle)( ctx, e0, e1, e3 );
+      VB->EdgeFlag[e1] = ef1;
+      VB->EdgeFlag[e3] = 0;
+      TAG(triangle)( ctx, e1, e2, e3 );
+      VB->EdgeFlag[e3] = ef3;
+   } else {
+      TAG(triangle)( ctx, e0, e1, e3 );
+      TAG(triangle)( ctx, e1, e2, e3 );
+   }
+}
+#endif
+#endif
+
+#if DO_LINE
+static void TAG(line)( GLcontext *ctx, GLuint e0, GLuint e1 )
+{
+   TNLvertexbuffer *VB = &TNL_CONTEXT(ctx)->vb;
+   VERTEX *v[2];
+   LOCAL_VARS(2);
+
+   v[0] = (VERTEX *)GET_VERTEX(e0);
+   v[1] = (VERTEX *)GET_VERTEX(e1);
+
+   if (DO_FLAT) {
+      if (HAVE_RGBA) {
+	 VERT_SAVE_RGBA( 0 );
+	 VERT_COPY_RGBA( v[0], v[1] );
+	 if (HAVE_SPEC && VB->SecondaryColorPtr[0]) {
+	    VERT_SAVE_SPEC( 0 );
+	    VERT_COPY_SPEC( v[0], v[1] );
+	 }
+      }
+      else {
+	 VERT_SAVE_IND( 0 );
+	 VERT_COPY_IND( v[0], v[1] );
+      }
+   }
+
+   LINE( v[0], v[1] );
+
+   if (DO_FLAT) {
+      if (HAVE_RGBA) {
+	 VERT_RESTORE_RGBA( 0 );
+
+	 if (HAVE_SPEC && VB->SecondaryColorPtr[0]) {
+	    VERT_RESTORE_SPEC( 0 );
+	 }
+      }
+      else {
+	 VERT_RESTORE_IND( 0 );
+      }
+   }
+}
+#endif
+
+#if DO_POINTS
+static void TAG(points)( GLcontext *ctx, GLuint first, GLuint last )
+{
+   struct vertex_buffer *VB = &TNL_CONTEXT( ctx )->vb;
+   int i;
+   LOCAL_VARS(1);
+
+   if (VB->Elts == 0) {
+      for ( i = first ; i < last ; i++ ) {
+	 if ( VB->ClipMask[i] == 0 ) {
+	    VERTEX *v = (VERTEX *)GET_VERTEX(i);
+	    POINT( v );
+	 }
+      }
+   } else {
+      for ( i = first ; i < last ; i++ ) {
+	 GLuint e = VB->Elts[i];
+	 if ( VB->ClipMask[e] == 0 ) {
+	    VERTEX *v = (VERTEX *)GET_VERTEX(e);
+	    POINT( v );
+	 }
+      }
+   }
+}
+#endif
+
+static void TAG(init)( void )
+{
+#if DO_QUAD
+   TAB[IND].quad = TAG(quad);
+#endif
+#if DO_TRI
+   TAB[IND].triangle = TAG(triangle);
+#endif
+#if DO_LINE
+   TAB[IND].line = TAG(line);
+#endif
+#if DO_POINTS
+   TAB[IND].points = TAG(points);
+#endif
+}
+
+#undef IND
+#undef TAG
+
+#if HAVE_RGBA
+#undef VERT_SET_IND
+#undef VERT_COPY_IND
+#undef VERT_SAVE_IND
+#undef VERT_RESTORE_IND
+#if HAVE_BACK_COLORS
+#undef VERT_SET_RGBA
+#endif
+#else
+#undef VERT_SET_RGBA
+#undef VERT_COPY_RGBA
+#undef VERT_SAVE_RGBA
+#undef VERT_RESTORE_RGBA
+#if HAVE_BACK_COLORS
+#undef VERT_SET_IND
+#endif
+#endif
+
+#if !HAVE_SPEC
+#undef VERT_SET_SPEC
+#undef VERT_COPY_SPEC
+#undef VERT_SAVE_SPEC
+#undef VERT_RESTORE_SPEC
+#if HAVE_BACK_COLORS
+#undef VERT_COPY_SPEC1
+#endif
+#else
+#if HAVE_BACK_COLORS
+#undef VERT_SET_SPEC
+#endif
+#endif
+
+#if !HAVE_BACK_COLORS
+#undef VERT_COPY_SPEC1
+#undef VERT_COPY_IND1
+#undef VERT_COPY_RGBA1
+#endif
+
+#ifndef INSANE_VERTICES
+#undef VERT_SET_Z
+#undef VERT_Z_ADD
+#endif
diff --git a/src/mesa/tnl_dd/t_dd_unfilled.h b/src/mesa/tnl_dd/t_dd_unfilled.h
new file mode 100644
index 00000000000..46415ea5f68
--- /dev/null
+++ b/src/mesa/tnl_dd/t_dd_unfilled.h
@@ -0,0 +1,212 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999-2001  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 <[email protected]>
+ */
+
+#if HAVE_RGBA
+#define VERT_SET_IND( v, c )
+#define VERT_COPY_IND( v0, v1 )
+#define VERT_SAVE_IND( idx )
+#define VERT_RESTORE_IND( idx )
+#endif
+
+#if !HAVE_SPEC
+#define VERT_SET_SPEC( v, c )
+#define VERT_COPY_SPEC( v0, v1 )
+#define VERT_SAVE_SPEC( idx )
+#define VERT_RESTORE_SPEC( idx )
+#endif
+
+static void TAG(unfilled_tri)( GLcontext *ctx,
+			       GLenum mode,
+			       GLuint e0, GLuint e1, GLuint e2 )
+{
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+   GLubyte *ef = VB->EdgeFlag;
+   VERTEX *v[3];
+   LOCAL_VARS(3);
+
+   v[0] = (VERTEX *)GET_VERTEX(e0);
+   v[1] = (VERTEX *)GET_VERTEX(e1);
+   v[2] = (VERTEX *)GET_VERTEX(e2);
+
+   if ((ctx->_TriangleCaps & DD_FLATSHADE) && HAVE_HW_FLATSHADE) {
+      if (HAVE_RGBA) {
+	 VERT_SAVE_RGBA(0);
+	 VERT_SAVE_RGBA(1);
+	 VERT_COPY_RGBA(v[0], v[2]);
+	 VERT_COPY_RGBA(v[1], v[2]);
+
+	 if (HAVE_SPEC) {
+	    VERT_SAVE_SPEC(0);
+	    VERT_SAVE_SPEC(1);
+	    VERT_COPY_SPEC(v[0], v[2]);
+	    VERT_COPY_SPEC(v[1], v[2]);
+	 }
+      } else {
+	 VERT_SAVE_IND(0);
+	 VERT_SAVE_IND(1);
+	 VERT_COPY_IND(v[0], v[2]);
+	 VERT_COPY_IND(v[1], v[2]);
+      }
+   }
+
+/*     fprintf(stderr, "%s %s %d %d %d\n", __FUNCTION__, */
+/*  	   _mesa_lookup_enum_by_nr( mode ), */
+/*  	   ef[e0], ef[e1], ef[e2]); */
+
+   if (mode == GL_POINT) {
+      RASTERIZE(GL_POINTS);
+      if (ef[e0]) POINT( v[0] );
+      if (ef[e1]) POINT( v[1] );
+      if (ef[e2]) POINT( v[2] );
+   }
+   else {
+      RASTERIZE(GL_LINES);
+      if (RENDER_PRIMITIVE == GL_POLYGON) {
+	 if (ef[e2]) LINE( v[2], v[0] );
+	 if (ef[e0]) LINE( v[0], v[1] );
+	 if (ef[e1]) LINE( v[1], v[2] );
+      }
+      else {
+	 if (ef[e0]) LINE( v[0], v[1] );
+	 if (ef[e1]) LINE( v[1], v[2] );
+	 if (ef[e2]) LINE( v[2], v[0] );
+      }
+   }
+
+   if ((ctx->_TriangleCaps & DD_FLATSHADE) && HAVE_HW_FLATSHADE) {
+      if (HAVE_RGBA) {
+	 VERT_RESTORE_RGBA(0);
+	 VERT_RESTORE_RGBA(1);
+
+	 if (HAVE_SPEC) {
+	    VERT_RESTORE_SPEC(0);
+	    VERT_RESTORE_SPEC(1);
+	 }
+      } else {
+	 VERT_RESTORE_IND(0);
+	 VERT_RESTORE_IND(1);
+      }
+   }
+}
+
+
+static void TAG(unfilled_quad)( GLcontext *ctx,
+				GLenum mode,
+				GLuint e0, GLuint e1,
+				GLuint e2, GLuint e3 )
+{
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+   GLubyte *ef = VB->EdgeFlag;
+   VERTEX *v[4];
+   LOCAL_VARS(4);
+
+   v[0] = (VERTEX *)GET_VERTEX(e0);
+   v[1] = (VERTEX *)GET_VERTEX(e1);
+   v[2] = (VERTEX *)GET_VERTEX(e2);
+   v[3] = (VERTEX *)GET_VERTEX(e3);
+
+   /* Hardware flatshading breaks down here.  If the hardware doesn't
+    * support flatshading, this will already have been done:
+    */
+   if ((ctx->_TriangleCaps & DD_FLATSHADE) && HAVE_HW_FLATSHADE) {
+      if (HAVE_RGBA) {
+	 VERT_SAVE_RGBA(0);
+	 VERT_SAVE_RGBA(1);
+	 VERT_SAVE_RGBA(2);
+	 VERT_COPY_RGBA(v[0], v[3]);
+	 VERT_COPY_RGBA(v[1], v[3]);
+	 VERT_COPY_RGBA(v[2], v[3]);
+
+	 if (HAVE_SPEC) {
+	    VERT_SAVE_SPEC(0);
+	    VERT_SAVE_SPEC(1);
+	    VERT_SAVE_SPEC(2);
+	    VERT_COPY_SPEC(v[0], v[3]);
+	    VERT_COPY_SPEC(v[1], v[3]);
+	    VERT_COPY_SPEC(v[2], v[3]);
+	 }
+      } else {
+	 VERT_SAVE_IND(0);
+	 VERT_SAVE_IND(1);
+	 VERT_SAVE_IND(2);
+	 VERT_COPY_IND(v[0], v[3]);
+	 VERT_COPY_IND(v[1], v[3]);
+	 VERT_COPY_IND(v[2], v[3]);
+      }
+   }
+
+   if (mode == GL_POINT) {
+      RASTERIZE(GL_POINTS);
+      if (ef[e0]) POINT( v[0] );
+      if (ef[e1]) POINT( v[1] );
+      if (ef[e2]) POINT( v[2] );
+      if (ef[e3]) POINT( v[3] );
+   }
+   else {
+      RASTERIZE(GL_LINES);
+      if (ef[e0]) LINE( v[0], v[1] );
+      if (ef[e1]) LINE( v[1], v[2] );
+      if (ef[e2]) LINE( v[2], v[3] );
+      if (ef[e3]) LINE( v[3], v[0] );
+   }
+
+   if ((ctx->_TriangleCaps & DD_FLATSHADE) && HAVE_HW_FLATSHADE) {
+      if (HAVE_RGBA) {
+	 VERT_RESTORE_RGBA(0);
+	 VERT_RESTORE_RGBA(1);
+	 VERT_RESTORE_RGBA(2);
+
+	 if (HAVE_SPEC) {
+	    VERT_RESTORE_SPEC(0);
+	    VERT_RESTORE_SPEC(1);
+	    VERT_RESTORE_SPEC(2);
+	 }
+      } else {
+	 VERT_RESTORE_IND(0);
+	 VERT_RESTORE_IND(1);
+	 VERT_RESTORE_IND(2);
+      }
+   }
+}
+
+
+#if HAVE_RGBA
+#undef VERT_SET_IND
+#undef VERT_COPY_IND
+#undef VERT_SAVE_IND
+#undef VERT_RESTORE_IND
+#endif
+
+#if !HAVE_SPEC
+#undef VERT_SET_SPEC
+#undef VERT_COPY_SPEC
+#undef VERT_SAVE_SPEC
+#undef VERT_RESTORE_SPEC
+#endif
+
+#undef TAG
diff --git a/src/mesa/tnl_dd/t_dd_vb.c b/src/mesa/tnl_dd/t_dd_vb.c
new file mode 100644
index 00000000000..4a742bcbbee
--- /dev/null
+++ b/src/mesa/tnl_dd/t_dd_vb.c
@@ -0,0 +1,392 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  3.5
+ *
+ * Copyright (C) 1999-2001  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 <[email protected]>
+ */
+#include "math/m_translate.h"
+
+#if (HAVE_HW_VIEWPORT)
+#define UNVIEWPORT_VARS
+#define UNVIEWPORT_X(x) x
+#define UNVIEWPORT_Y(x) x
+#define UNVIEWPORT_Z(x) x
+#endif
+
+#ifndef LOCALVARS
+#define LOCALVARS
+#endif
+
+#ifndef CHECK_HW_DIVIDE
+#define CHECK_HW_DIVIDE 1
+#endif
+
+/* These don't need to be duplicated, but there's currently nowhere
+ * really convenient to put them.  Need to build some actual .o files in
+ * this directory?
+ */
+static void copy_pv_rgba4_spec5( GLcontext *ctx, GLuint edst, GLuint esrc )
+{
+   LOCALVARS   
+   GLubyte *verts = GET_VERTEX_STORE();
+   GLuint shift = GET_VERTEX_STRIDE_SHIFT();
+   GLuint *dst = (GLuint *)(verts + (edst << shift));
+   GLuint *src = (GLuint *)(verts + (esrc << shift));
+   dst[4] = src[4];
+   dst[5] = src[5];
+}
+
+static void copy_pv_rgba4( GLcontext *ctx, GLuint edst, GLuint esrc )
+{
+   LOCALVARS
+   GLubyte *verts = GET_VERTEX_STORE();
+   GLuint shift = GET_VERTEX_STRIDE_SHIFT();
+   GLuint *dst = (GLuint *)(verts + (edst << shift));
+   GLuint *src = (GLuint *)(verts + (esrc << shift));
+   dst[4] = src[4];
+}
+
+static void copy_pv_rgba3( GLcontext *ctx, GLuint edst, GLuint esrc )
+{
+   LOCALVARS
+   GLubyte *verts = GET_VERTEX_STORE();
+   GLuint shift = GET_VERTEX_STRIDE_SHIFT();
+   GLuint *dst = (GLuint *)(verts + (edst << shift));
+   GLuint *src = (GLuint *)(verts + (esrc << shift));
+   dst[3] = src[3];
+}
+
+
+void TAG(translate_vertex)(GLcontext *ctx,
+			   const VERTEX *src,
+			   SWvertex *dst)
+{
+   LOCALVARS
+   GLuint format = GET_VERTEX_FORMAT();
+   GLfloat *s = ctx->Viewport._WindowMap.m;
+   UNVIEWPORT_VARS;
+
+   if (format == TINY_VERTEX_FORMAT) {
+      if (HAVE_HW_VIEWPORT) {
+	 dst->win[0] = s[0]  * src->v.x + s[12];
+	 dst->win[1] = s[5]  * src->v.y + s[13];
+	 dst->win[2] = s[10] * src->v.z + s[14];
+	 dst->win[3] = 1.0;
+      } else {
+	 dst->win[0] = UNVIEWPORT_X( src->v.x );
+	 dst->win[1] = UNVIEWPORT_Y( src->v.y );
+	 dst->win[2] = UNVIEWPORT_Z( src->v.z );
+	 dst->win[3] = 1.0;
+      }
+
+      dst->color[0] = src->tv.color.red;
+      dst->color[1] = src->tv.color.green;
+      dst->color[2] = src->tv.color.blue;
+      dst->color[3] = src->tv.color.alpha;
+   }
+   else {
+      if (HAVE_HW_VIEWPORT) {
+	 if (HAVE_HW_DIVIDE && CHECK_HW_DIVIDE) {
+	    GLfloat oow = 1.0 / src->v.w;
+	    dst->win[0] = s[0]  * src->v.x * oow + s[12];
+	    dst->win[1] = s[5]  * src->v.y * oow + s[13];
+	    dst->win[2] = s[10] * src->v.z * oow + s[14];
+	    dst->win[3] = oow;
+	 } else {
+	    dst->win[0] = s[0]  * src->v.x + s[12];
+	    dst->win[1] = s[5]  * src->v.y + s[13];
+	    dst->win[2] = s[10] * src->v.z + s[14];
+	    dst->win[3] = src->v.w;
+	 }
+      } else {
+	 dst->win[0] = UNVIEWPORT_X( src->v.x );
+	 dst->win[1] = UNVIEWPORT_Y( src->v.y );
+	 dst->win[2] = UNVIEWPORT_Z( src->v.z );
+	 dst->win[3] = src->v.w;
+      }
+
+      dst->color[0] = src->v.color.red;
+      dst->color[1] = src->v.color.green;
+      dst->color[2] = src->v.color.blue;
+      dst->color[3] = src->v.color.alpha;
+
+      dst->specular[0] = src->v.specular.red;
+      dst->specular[1] = src->v.specular.green;
+      dst->specular[2] = src->v.specular.blue;
+
+      dst->fog = src->v.specular.alpha/255.0;
+
+      if (HAVE_PTEX_VERTICES &&
+	  ((HAVE_TEX2_VERTICES && format == PROJ_TEX3_VERTEX_FORMAT) ||
+	   (format == PROJ_TEX1_VERTEX_FORMAT))) {
+
+	 dst->texcoord[0][0] = src->pv.u0;
+	 dst->texcoord[0][1] = src->pv.v0;
+	 dst->texcoord[0][3] = src->pv.q0;
+
+	 dst->texcoord[1][0] = src->pv.u1;
+	 dst->texcoord[1][1] = src->pv.v1;
+	 dst->texcoord[1][3] = src->pv.q1;
+
+	 if (HAVE_TEX2_VERTICES) {
+	    dst->texcoord[2][0] = src->pv.u2;
+	    dst->texcoord[2][1] = src->pv.v2;
+	    dst->texcoord[2][3] = src->pv.q2;
+	 }
+
+	 if (HAVE_TEX3_VERTICES) {
+	    dst->texcoord[3][0] = src->pv.u3;
+	    dst->texcoord[3][1] = src->pv.v3;
+	    dst->texcoord[3][3] = src->pv.q3;
+	 }
+      }
+      else {
+	 dst->texcoord[0][0] = src->v.u0;
+	 dst->texcoord[0][1] = src->v.v0;
+	 dst->texcoord[0][3] = 1.0;
+
+	 dst->texcoord[1][0] = src->v.u1;
+	 dst->texcoord[1][1] = src->v.v1;
+	 dst->texcoord[1][3] = 1.0;
+
+	 if (HAVE_TEX2_VERTICES) {
+	    dst->texcoord[2][0] = src->v.u2;
+	    dst->texcoord[2][1] = src->v.v2;
+	    dst->texcoord[2][3] = 1.0;
+	 }
+
+	 if (HAVE_TEX3_VERTICES) {
+	    dst->texcoord[3][0] = src->v.u3;
+	    dst->texcoord[3][1] = src->v.v3;
+	    dst->texcoord[3][3] = 1.0;
+	 }
+      }
+   }
+
+   dst->pointSize = ctx->Point._Size;
+}
+
+
+
+void TAG(print_vertex)( GLcontext *ctx, const VERTEX *v )
+{
+   LOCALVARS
+   GLuint format = GET_VERTEX_FORMAT();
+
+   fprintf(stderr, "(%x) ", format);
+
+   switch (format) {
+#if HAVE_TINY_VERTICES
+   case TINY_VERTEX_FORMAT:
+      fprintf(stderr, "xyz %.4f,%.4f,%.4f rgba %x:%x:%x:%x\n",
+	      v->v.x, v->v.y, v->v.z,
+	      v->tv.color.red,
+	      v->tv.color.green,
+	      v->tv.color.blue,
+	      v->tv.color.alpha);
+      break;
+#endif
+#if HAVE_NOTEX_VERTICES
+   case NOTEX_VERTEX_FORMAT:
+      fprintf(stderr, "xyzw %.4f,%.4f,%.4f,%.4f rgba %x:%x:%x:%x spec %x:%x:%x:%x\n",
+	      v->v.x, v->v.y, v->v.z, v->v.w,
+	      v->v.color.red,
+	      v->v.color.green,
+	      v->v.color.blue,
+	      v->v.color.alpha,
+	      v->v.specular.red,
+	      v->v.specular.green,
+	      v->v.specular.blue,
+	      v->v.specular.alpha);
+      break;
+#endif
+#if HAVE_TEX0_VERTICES
+   case TEX0_VERTEX_FORMAT:
+      fprintf(stderr, "xyzw %.4f,%.4f,%.4f,%.4f rgba %x:%x:%x:%x st %.4f,%.4f\n",
+	      v->v.x, v->v.y, v->v.z, v->v.w,
+	      v->v.color.red,
+	      v->v.color.green,
+	      v->v.color.blue,
+	      v->v.color.alpha,
+	      v->v.u0,
+	      v->v.v0);
+      break;
+#endif
+#if HAVE_TEX1_VERTICES
+   case TEX1_VERTEX_FORMAT:
+      fprintf(stderr, "xyzw %.4f,%.4f,%.4f,%.4f rgba %x:%x:%x:%x st %.4f,%.4f st %.4f,%.4f\n",
+	      v->v.x, v->v.y, v->v.z, v->v.w,
+	      v->v.color.red,
+	      v->v.color.green,
+	      v->v.color.blue,
+	      v->v.color.alpha,
+	      v->v.u0,
+	      v->v.v0,
+	      v->v.u1,
+	      v->v.u2);
+      break;
+#endif
+#if HAVE_PTEX_VERTICES
+   case PROJ_TEX1_VERTEX_FORMAT:
+      fprintf(stderr, "xyzw %.4f,%.4f,%.4f,%.4f rgba %x:%x:%x:%x stq %.4f,%.4f,%.4f stq %.4f,%.4f,%.4f\n",
+	      v->v.x, v->v.y, v->v.z, v->v.w,
+	      v->v.color.red,
+	      v->v.color.green,
+	      v->v.color.blue,
+	      v->v.color.alpha,
+	      v->pv.u0,
+	      v->pv.v0,
+	      v->pv.q0,
+	      v->pv.u1,
+	      v->pv.v1,
+	      v->pv.q1);
+      break;
+#endif      
+   default:
+      fprintf(stderr, "???\n");
+      break;
+   }
+
+   fprintf(stderr, "\n");
+}
+
+static void do_import( struct vertex_buffer *VB,
+		       struct gl_client_array *to,
+		       struct gl_client_array *from )
+{
+   GLuint count = VB->Count;
+
+   if (!to->Ptr) {
+      to->Ptr = ALIGN_MALLOC( VB->Size * 4 * sizeof(GLubyte), 32 );
+      to->Type = GL_UNSIGNED_BYTE;
+   }
+
+   /* No need to transform the same value 3000 times.
+    */
+   if (!from->StrideB) {
+      to->StrideB = 0;
+      count = 1;
+   }
+   else
+      to->StrideB = 4 * sizeof(GLubyte);
+   
+   _math_trans_4ub( (GLubyte (*)[4]) to->Ptr,
+		    from->Ptr,
+		    from->StrideB,
+		    from->Type,
+		    from->Size,
+		    0,
+		    count);
+}
+
+#ifndef IMPORT_QUALIFIER
+#define IMPORT_QUALIFIER static
+#endif
+
+IMPORT_QUALIFIER void TAG(import_float_colors)( GLcontext *ctx )
+{
+   LOCALVARS
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+   struct gl_client_array *to = GET_UBYTE_COLOR_STORE();
+   do_import( VB, to, VB->ColorPtr[0] );
+   VB->ColorPtr[0] = to;
+}
+
+IMPORT_QUALIFIER void TAG(import_float_spec_colors)( GLcontext *ctx )
+{
+   LOCALVARS
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+   struct gl_client_array *to = GET_UBYTE_SPEC_COLOR_STORE();
+   do_import( VB, to, VB->SecondaryColorPtr[0] );
+   VB->SecondaryColorPtr[0] = to;
+}
+
+/* Interpolate the elements of the VB not included in typical hardware
+ * vertices.  
+ *
+ * NOTE: All these arrays are guarenteed by tnl to be writeable and
+ * have good stride.
+ */
+#ifndef INTERP_QUALIFIER 
+#define INTERP_QUALIFIER static
+#endif
+
+#define GET_COLOR(ptr, idx) (((GLchan (*)[4])((ptr)->Ptr))[idx])
+
+
+INTERP_QUALIFIER void TAG(interp_extras)( GLcontext *ctx,
+					  GLfloat t,
+					  GLuint dst, GLuint out, GLuint in,
+					  GLboolean force_boundary )
+{
+   LOCALVARS
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+
+   if (VB->ColorPtr[1]) {
+      INTERP_4CHAN( t,
+		    GET_COLOR(VB->ColorPtr[1], dst),
+		    GET_COLOR(VB->ColorPtr[1], out),
+		    GET_COLOR(VB->ColorPtr[1], in) );
+
+      if (VB->SecondaryColorPtr[1]) {
+	 INTERP_3CHAN( t,
+		       GET_COLOR(VB->SecondaryColorPtr[1], dst),
+		       GET_COLOR(VB->SecondaryColorPtr[1], out),
+		       GET_COLOR(VB->SecondaryColorPtr[1], in) );
+      }
+   }
+
+   if (VB->EdgeFlag) {
+      VB->EdgeFlag[dst] = VB->EdgeFlag[out] || force_boundary;
+   }
+
+   INTERP_VERTEX(ctx, t, dst, out, in, force_boundary);
+}
+
+INTERP_QUALIFIER void TAG(copy_pv_extras)( GLcontext *ctx, 
+					   GLuint dst, GLuint src )
+{
+   LOCALVARS
+      struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+
+   if (VB->ColorPtr[1]) {
+      COPY_CHAN4( GET_COLOR(VB->ColorPtr[1], dst), 
+		  GET_COLOR(VB->ColorPtr[1], src) );
+
+      if (VB->SecondaryColorPtr[1]) {
+	 COPY_CHAN4( GET_COLOR(VB->SecondaryColorPtr[1], dst), 
+		     GET_COLOR(VB->SecondaryColorPtr[1], src) );
+      }
+   }
+
+   COPY_PV_VERTEX(ctx, dst, src);
+}
+
+
+#undef INTERP_QUALIFIER
+#undef IMPORT_QUALIFIER
+#undef GET_COLOR
+
+#undef IND
+#undef TAG
diff --git a/src/mesa/tnl_dd/t_dd_vbtmp.h b/src/mesa/tnl_dd/t_dd_vbtmp.h
new file mode 100644
index 00000000000..5aad0b56222
--- /dev/null
+++ b/src/mesa/tnl_dd/t_dd_vbtmp.h
@@ -0,0 +1,834 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  5.0.1
+ *
+ * Copyright (C) 1999-2003  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 <[email protected]>
+ */
+
+
+/* Unlike the other templates here, this assumes quite a bit about the
+ * underlying hardware.  Specifically it assumes a d3d-like vertex
+ * format, with a layout more or less constrained to look like the
+ * following:
+ *
+ * union {
+ *    struct {
+ *        float x, y, z, w;
+ *        struct { char r, g, b, a; } color;
+ *        struct { char r, g, b, fog; } spec;
+ *        float u0, v0;
+ *        float u1, v1;
+ *        float u2, v2;
+ *        float u3, v3;
+ *    } v;
+ *    struct {
+ *        float x, y, z, w;
+ *        struct { char r, g, b, a; } color;
+ *        struct { char r, g, b, fog; } spec;
+ *        float u0, v0, q0;
+ *        float u1, v1, q1;
+ *        float u2, v2, q2;
+ *        float u3, v3, q3;
+ *    } pv;
+ *    struct {
+ *        float x, y, z;
+ *        struct { char r, g, b, a; } color;
+ *    } tv;
+ *    float f[16];
+ *    unsigned int ui[16];
+ *    unsigned char ub4[4][16];
+ * }
+ *
+ 
+ * VERTEX:   hw vertex type as above
+ * VERTEX_COLOR: hw color struct type in VERTEX
+ *
+ * DO_XYZW:  Emit xyz and maybe w coordinates.
+ * DO_RGBA:  Emit color.
+ * DO_SPEC:  Emit specular color.
+ * DO_FOG:   Emit fog coordinate in specular alpha.
+ * DO_TEX0:  Emit tex0 u,v coordinates.
+ * DO_TEX1:  Emit tex1 u,v coordinates.
+ * DO_TEX2:  Emit tex2 u,v coordinates.
+ * DO_TEX3:  Emit tex3 u,v coordinates.
+ * DO_PTEX:  Emit tex0,1,2,3 q coordinates where possible.
+ *
+ * HAVE_RGBA_COLOR: Hardware takes color in rgba order (else bgra).
+ *
+ * HAVE_HW_VIEWPORT:  Hardware performs viewport transform.
+ * HAVE_HW_DIVIDE:  Hardware performs perspective divide.
+ *
+ * HAVE_TINY_VERTICES:  Hardware understands v.tv format.
+ * HAVE_PTEX_VERTICES:  Hardware understands v.pv format.
+ * HAVE_NOTEX_VERTICES:  Hardware understands v.v format with texcount 0.
+ *
+ * Additionally, this template assumes it is emitting *transformed*
+ * vertices; the modifications to emit untransformed vertices (ie. to
+ * t&l hardware) are probably too great to cooexist with the code
+ * already in this file.
+ *
+ * NOTE: The PTEX vertex format always includes TEX0 and TEX1, even if
+ * only TEX0 is enabled, in order to maintain a vertex size which is
+ * an exact number of quadwords.
+ */
+
+#if (HAVE_HW_VIEWPORT)
+#define VIEWPORT_X(dst,x) dst = x
+#define VIEWPORT_Y(dst,y) dst = y
+#define VIEWPORT_Z(dst,z) dst = z
+#else
+#define VIEWPORT_X(dst,x) dst = s[0]  * x + s[12]
+#define VIEWPORT_Y(dst,y) dst = s[5]  * y + s[13]
+#define VIEWPORT_Z(dst,z) dst = s[10] * z + s[14]
+#endif
+
+#if (HAVE_HW_DIVIDE && !HAVE_PTEX_VERTICES)
+#error "can't cope with this combination" 
+#endif 
+
+#ifndef LOCALVARS
+#define LOCALVARS
+#endif
+
+#ifndef CHECK_HW_DIVIDE
+#define CHECK_HW_DIVIDE 1
+#endif
+
+#if (HAVE_HW_DIVIDE || DO_SPEC || DO_TEX0 || DO_FOG || !HAVE_TINY_VERTICES)
+
+static void TAG(emit)( GLcontext *ctx,
+		       GLuint start, GLuint end,
+		       void *dest,
+		       GLuint stride )
+{
+   LOCALVARS
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+   GLfloat (*tc0)[4], (*tc1)[4], (*fog)[4];
+   GLfloat (*tc2)[4], (*tc3)[4];
+   GLubyte (*col)[4], (*spec)[4];
+   GLuint tc0_stride, tc1_stride, col_stride, spec_stride, fog_stride;
+   GLuint tc2_stride, tc3_stride;
+   GLuint tc0_size, tc1_size;
+   GLuint tc2_size, tc3_size;
+   GLfloat (*coord)[4];
+   GLuint coord_stride;
+   VERTEX *v = (VERTEX *)dest;
+   const GLfloat *s = GET_VIEWPORT_MAT();
+   const GLubyte *mask = VB->ClipMask;
+   int i;
+
+/*     fprintf(stderr, "%s(big) importable %d %d..%d\n",  */
+/*  	   __FUNCTION__, VB->importable_data, start, end); */
+
+   if (HAVE_HW_VIEWPORT && HAVE_HW_DIVIDE && CHECK_HW_DIVIDE) {
+      (void) s;
+      coord = VB->ClipPtr->data;
+      coord_stride = VB->ClipPtr->stride;
+   }
+   else {
+      coord = VB->NdcPtr->data;
+      coord_stride = VB->NdcPtr->stride;
+   }
+
+   if (DO_TEX3) {
+      const GLuint t3 = GET_TEXSOURCE(3);
+      tc3 = VB->TexCoordPtr[t3]->data;
+      tc3_stride = VB->TexCoordPtr[t3]->stride;
+      if (DO_PTEX)
+	 tc3_size = VB->TexCoordPtr[t3]->size;
+   }
+
+   if (DO_TEX2) {
+      const GLuint t2 = GET_TEXSOURCE(2);
+      tc2 = VB->TexCoordPtr[t2]->data;
+      tc2_stride = VB->TexCoordPtr[t2]->stride;
+      if (DO_PTEX)
+	 tc2_size = VB->TexCoordPtr[t2]->size;
+   }
+
+   if (DO_TEX1) {
+      const GLuint t1 = GET_TEXSOURCE(1);
+      tc1 = VB->TexCoordPtr[t1]->data;
+      tc1_stride = VB->TexCoordPtr[t1]->stride;
+      if (DO_PTEX)
+	 tc1_size = VB->TexCoordPtr[t1]->size;
+   }
+
+   if (DO_TEX0) {
+      const GLuint t0 = GET_TEXSOURCE(0);
+      tc0_stride = VB->TexCoordPtr[t0]->stride;
+      tc0 = VB->TexCoordPtr[t0]->data;
+      if (DO_PTEX) 
+	 tc0_size = VB->TexCoordPtr[t0]->size;
+   }
+
+   if (DO_RGBA) {
+      if (VB->ColorPtr[0]->Type != GL_UNSIGNED_BYTE)
+	 IMPORT_FLOAT_COLORS( ctx );
+      col = (GLubyte (*)[4])VB->ColorPtr[0]->Ptr;
+      col_stride = VB->ColorPtr[0]->StrideB;
+   }
+
+   if (DO_SPEC) {
+      if (VB->SecondaryColorPtr[0]) {
+	 if (VB->SecondaryColorPtr[0]->Type != GL_UNSIGNED_BYTE)
+	    IMPORT_FLOAT_SPEC_COLORS( ctx );
+	 spec = (GLubyte (*)[4])VB->SecondaryColorPtr[0]->Ptr;
+	 spec_stride = VB->SecondaryColorPtr[0]->StrideB;
+      } else {
+	 GLubyte tmp[4];
+	 spec = &tmp;
+	 spec_stride = 0;
+      }
+   }
+
+   if (DO_FOG) {
+      if (VB->FogCoordPtr) {
+	 fog = VB->FogCoordPtr->data;
+	 fog_stride = VB->FogCoordPtr->stride;
+      }
+      else {
+	 static GLfloat tmp[4] = {0, 0, 0, 0};
+	 fog = &tmp;
+	 fog_stride = 0;
+      }
+   }
+
+   if (VB->importable_data || (DO_SPEC && !spec_stride) || (DO_FOG && !fog_stride)) {
+      /* May have nonstandard strides:
+       */
+      if (start) {
+	 coord =  (GLfloat (*)[4])((GLubyte *)coord + start * coord_stride);
+	 if (DO_TEX0)
+	    tc0 =  (GLfloat (*)[4])((GLubyte *)tc0 + start * tc0_stride);
+	 if (DO_TEX1) 
+	    tc1 =  (GLfloat (*)[4])((GLubyte *)tc1 + start * tc1_stride);
+	 if (DO_TEX2) 
+	    tc2 =  (GLfloat (*)[4])((GLubyte *)tc2 + start * tc2_stride);
+	 if (DO_TEX3) 
+	    tc3 =  (GLfloat (*)[4])((GLubyte *)tc3 + start * tc3_stride);
+	 if (DO_RGBA) 
+	    STRIDE_4UB(col, start * col_stride);
+	 if (DO_SPEC)
+	    STRIDE_4UB(spec, start * spec_stride);
+	 if (DO_FOG)
+	    /*STRIDE_F(fog, start * fog_stride);*/
+	    fog =  (GLfloat (*)[4])((GLubyte *)fog + start * fog_stride);
+      }
+
+      for (i=start; i < end; i++, v = (VERTEX *)((GLubyte *)v + stride)) {
+	 if (DO_XYZW) {
+	    if (HAVE_HW_VIEWPORT || mask[i] == 0) {
+	       VIEWPORT_X(v->v.x, coord[0][0]);
+	       VIEWPORT_Y(v->v.y, coord[0][1]);
+	       VIEWPORT_Z(v->v.z, coord[0][2]);
+	       v->v.w = coord[0][3];
+	    }
+/*  	    fprintf(stderr, "vert %d: %.2f %.2f %.2f %.2f\n",  */
+/*  		    i, v->v.x, v->v.y, v->v.z, v->v.w); */
+	    coord =  (GLfloat (*)[4])((GLubyte *)coord +  coord_stride);
+	 }
+	 if (DO_RGBA) {
+	    if (HAVE_RGBA_COLOR) {
+	       *(GLuint *)&v->v.color = LE32_TO_CPU(*(GLuint *)&col[0]);
+	       STRIDE_4UB(col, col_stride);
+	    } else {
+	       v->v.color.blue  = col[0][2];
+	       v->v.color.green = col[0][1];
+	       v->v.color.red   = col[0][0];
+	       v->v.color.alpha = col[0][3];
+	       STRIDE_4UB(col, col_stride);
+	    }
+	 }
+	 if (DO_SPEC) {
+	    v->v.specular.red = spec[0][0];
+	    v->v.specular.green = spec[0][1];
+	    v->v.specular.blue = spec[0][2];
+	    STRIDE_4UB(spec, spec_stride);
+	 }
+	 if (DO_FOG) {
+	    v->v.specular.alpha = fog[0][0] * 255.0;
+	    /*STRIDE_F(fog, fog_stride);*/
+	    fog =  (GLfloat (*)[4])((GLubyte *)fog + fog_stride);
+	 }
+	 if (DO_TEX0) {
+	    v->v.u0 = tc0[0][0];
+	    v->v.v0 = tc0[0][1];
+	    if (DO_PTEX) {
+	       if (HAVE_PTEX_VERTICES) {
+		  if (tc0_size == 4) 
+		     v->pv.q0 = tc0[0][3];
+		  else
+		     v->pv.q0 = 1.0;
+	       } 
+	       else if (tc0_size == 4) {
+		  float rhw = 1.0 / tc0[0][3];
+		  v->v.w *= tc0[0][3];
+		  v->v.u0 *= rhw;
+		  v->v.v0 *= rhw;
+	       } 
+	    } 
+	    tc0 =  (GLfloat (*)[4])((GLubyte *)tc0 +  tc0_stride);
+	 }
+	 if (DO_TEX1) {
+	    if (DO_PTEX) {
+	       v->pv.u1 = tc1[0][0];
+	       v->pv.v1 = tc1[0][1];
+	       if (tc1_size == 4) 
+		  v->pv.q1 = tc1[0][3];
+	       else
+		  v->pv.q1 = 1.0;
+	    } 
+	    else {
+	       v->v.u1 = tc1[0][0];
+	       v->v.v1 = tc1[0][1];
+	    }
+	    tc1 =  (GLfloat (*)[4])((GLubyte *)tc1 +  tc1_stride);
+	 } 
+	 else if (DO_PTEX) {
+	    *(GLuint *)&v->pv.q1 = 0;	/* avoid culling on radeon */
+	 }
+	 if (DO_TEX2) {
+	    if (DO_PTEX) {
+	       v->pv.u2 = tc2[0][0];
+	       v->pv.v2 = tc2[0][1];
+	       if (tc2_size == 4) 
+		  v->pv.q2 = tc2[0][3];
+	       else
+		  v->pv.q2 = 1.0;
+	    } 
+	    else {
+	       v->v.u2 = tc2[0][0];
+	       v->v.v2 = tc2[0][1];
+	    }
+	    tc2 =  (GLfloat (*)[4])((GLubyte *)tc2 +  tc2_stride);
+	 } 
+	 if (DO_TEX3) {
+	    if (DO_PTEX) {
+	       v->pv.u3 = tc3[0][0];
+	       v->pv.v3 = tc3[0][1];
+	       if (tc3_size == 4) 
+		  v->pv.q3 = tc3[0][3];
+	       else
+		  v->pv.q3 = 1.0;
+	    } 
+	    else {
+	       v->v.u3 = tc3[0][0];
+	       v->v.v3 = tc3[0][1];
+	    }
+	    tc3 =  (GLfloat (*)[4])((GLubyte *)tc3 +  tc3_stride);
+	 } 
+      }
+   }
+   else {
+      for (i=start; i < end; i++, v = (VERTEX *)((GLubyte *)v + stride)) {
+	 if (DO_XYZW) {
+	    if (HAVE_HW_VIEWPORT || mask[i] == 0) {
+	       VIEWPORT_X(v->v.x, coord[i][0]);
+	       VIEWPORT_Y(v->v.y, coord[i][1]);
+	       VIEWPORT_Z(v->v.z, coord[i][2]);
+	       v->v.w = coord[i][3];
+	    }
+	 }
+	 if (DO_RGBA) {
+	    if (HAVE_RGBA_COLOR) {
+	       *(GLuint *)&v->v.color = LE32_TO_CPU(*(GLuint *)&col[i]);
+	    }
+	    else {
+	       v->v.color.blue  = col[i][2];
+	       v->v.color.green = col[i][1];
+	       v->v.color.red   = col[i][0];
+	       v->v.color.alpha = col[i][3];
+	    }
+	 }
+	 if (DO_SPEC) {
+	    v->v.specular.red   = spec[i][0];
+	    v->v.specular.green = spec[i][1];
+	    v->v.specular.blue  = spec[i][2];
+	 }
+	 if (DO_FOG) {
+	    v->v.specular.alpha = fog[i][0] * 255.0;
+	 }
+	 if (DO_TEX0) {
+	    v->v.u0 = tc0[i][0];
+	    v->v.v0 = tc0[i][1];
+	    if (DO_PTEX) {
+	       if (HAVE_PTEX_VERTICES) {
+		  if (tc0_size == 4) 
+		     v->pv.q0 = tc0[i][3];
+		  else
+		     v->pv.q0 = 1.0;
+
+		  v->pv.q1 = 0;	/* radeon */
+	       } 
+	       else if (tc0_size == 4) {
+		  float rhw = 1.0 / tc0[i][3];
+		  v->v.w *= tc0[i][3];
+		  v->v.u0 *= rhw;
+		  v->v.v0 *= rhw;
+	       } 
+	    } 
+	 }
+	 if (DO_TEX1) {
+	    if (DO_PTEX) {
+	       v->pv.u1 = tc1[i][0];
+	       v->pv.v1 = tc1[i][1];
+	       if (tc1_size == 4) 
+		  v->pv.q1 = tc1[i][3];
+	       else
+		  v->pv.q1 = 1.0;
+	    } 
+	    else {
+	       v->v.u1 = tc1[i][0];
+	       v->v.v1 = tc1[i][1];
+	    }
+	 }
+      }
+   }
+}
+#else
+#if DO_XYZW
+
+#if HAVE_HW_DIVIDE
+#error "cannot use tiny vertices with hw perspective divide"
+#endif
+
+static void TAG(emit)( GLcontext *ctx, GLuint start, GLuint end,
+		       void *dest, GLuint stride )
+{
+   LOCALVARS
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+   GLubyte (*col)[4];
+   GLuint col_stride;
+   GLfloat (*coord)[4] = VB->NdcPtr->data;
+   GLuint coord_stride = VB->NdcPtr->stride;
+   GLfloat *v = (GLfloat *)dest;
+   const GLubyte *mask = VB->ClipMask;
+   const GLfloat *s = GET_VIEWPORT_MAT();
+   int i;
+
+   (void) s;
+
+   ASSERT(stride == 4);
+
+   if (VB->ColorPtr[0]->Type != GL_UNSIGNED_BYTE)
+      IMPORT_FLOAT_COLORS( ctx );
+
+   col = (GLubyte (*)[4])VB->ColorPtr[0]->Ptr;
+   col_stride = VB->ColorPtr[0]->StrideB;
+   ASSERT(VB->ColorPtr[0]->Type == GL_UNSIGNED_BYTE);
+
+/*     fprintf(stderr, "%s(small) importable %x\n",  */
+/*  	   __FUNCTION__, VB->importable_data); */
+
+   /* Pack what's left into a 4-dword vertex.  Color is in a different
+    * place, and there is no 'w' coordinate.
+    */
+   if (VB->importable_data) {
+      if (start) {
+	 coord =  (GLfloat (*)[4])((GLubyte *)coord + start * coord_stride);
+	 STRIDE_4UB(col, start * col_stride);
+      }
+
+      for (i=start; i < end; i++, v+=4) {
+	 if (HAVE_HW_VIEWPORT || mask[i] == 0) {
+	    VIEWPORT_X(v[0], coord[0][0]);
+	    VIEWPORT_Y(v[1], coord[0][1]);
+	    VIEWPORT_Z(v[2], coord[0][2]);
+	 }
+	 coord =  (GLfloat (*)[4])((GLubyte *)coord +  coord_stride);
+	 if (DO_RGBA) {
+	    if (HAVE_RGBA_COLOR) {
+	       *(GLuint *)&v[3] = LE32_TO_CPU(*(GLuint *)col);
+	    }
+	    else {
+	       VERTEX_COLOR *c = (VERTEX_COLOR *)&v[3];
+	       c->blue  = col[0][2];
+	       c->green = col[0][1];
+	       c->red   = col[0][0];
+	       c->alpha = col[0][3];
+	    }
+	    STRIDE_4UB( col, col_stride );
+	 }
+/*  	 fprintf(stderr, "vert %d: %.2f %.2f %.2f %x\n",  */
+/*  		 i, v[0], v[1], v[2], *(int *)&v[3]); */
+      }
+   }
+   else {
+      for (i=start; i < end; i++, v+=4) {
+	 if (HAVE_HW_VIEWPORT || mask[i] == 0) {
+	    VIEWPORT_X(v[0], coord[i][0]);
+	    VIEWPORT_Y(v[1], coord[i][1]);
+	    VIEWPORT_Z(v[2], coord[i][2]);
+	 }
+	 if (DO_RGBA) {
+	    if (HAVE_RGBA_COLOR) {
+	       *(GLuint *)&v[3] = LE32_TO_CPU(*(GLuint *)&col[i]);
+	    }
+	    else {
+	       VERTEX_COLOR *c = (VERTEX_COLOR *)&v[3];
+	       c->blue  = col[i][2];
+	       c->green = col[i][1];
+	       c->red   = col[i][0];
+	       c->alpha = col[i][3];
+	    }
+	 }
+/*  	 fprintf(stderr, "vert %d: %.2f %.2f %.2f %x\n",  */
+/*  		 i, v[0], v[1], v[2], *(int *)&v[3]); */
+
+      }
+   }
+}
+#else
+static void TAG(emit)( GLcontext *ctx, GLuint start, GLuint end,
+		       void *dest, GLuint stride )
+{
+   LOCALVARS
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+   GLubyte (*col)[4];
+   GLuint col_stride;
+   GLfloat *v = (GLfloat *)dest;
+   int i;
+
+   if (VB->ColorPtr[0]->Type != GL_UNSIGNED_BYTE)
+      IMPORT_FLOAT_COLORS( ctx );
+
+   col = VB->ColorPtr[0]->Ptr;
+   col_stride = VB->ColorPtr[0]->StrideB;
+
+   if (start)
+      STRIDE_4UB(col, col_stride * start);
+
+   /* Need to figure out where color is:
+    */
+   if (GET_VERTEX_FORMAT() == TINY_VERTEX_FORMAT)
+      v += 3;
+   else
+      v += 4;
+
+   for (i=start; i < end; i++, STRIDE_F(v, stride)) {
+      if (HAVE_RGBA_COLOR) {
+	 *(GLuint *)v = LE32_TO_CPU(*(GLuint *)col[0]);
+      }
+      else {
+	 VERTEX_COLOR *c = (VERTEX_COLOR *)v;
+	 c->blue  = col[0][2];
+	 c->green = col[0][1];
+	 c->red   = col[0][0];
+	 c->alpha = col[0][3];
+      }
+      STRIDE_4UB( col, col_stride );
+   }
+}
+#endif /* emit */
+#endif /* emit */
+
+#if (DO_XYZW) && (DO_RGBA)
+
+
+#if (HAVE_PTEX_VERTICES)
+static GLboolean TAG(check_tex_sizes)( GLcontext *ctx )
+{
+   LOCALVARS
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+
+   /* Force 'missing' texcoords to something valid.
+    */
+   if (DO_TEX3 && VB->TexCoordPtr[2] == 0)
+      VB->TexCoordPtr[2] = VB->TexCoordPtr[3];
+
+   if (DO_TEX2 && VB->TexCoordPtr[1] == 0)
+      VB->TexCoordPtr[1] = VB->TexCoordPtr[2];
+
+   if (DO_TEX1 && VB->TexCoordPtr[0] == 0)
+      VB->TexCoordPtr[0] = VB->TexCoordPtr[1];
+
+   if (DO_PTEX)
+      return GL_TRUE;
+   
+   if ((DO_TEX3 && VB->TexCoordPtr[GET_TEXSOURCE(3)]->size == 4) ||
+       (DO_TEX2 && VB->TexCoordPtr[GET_TEXSOURCE(2)]->size == 4) ||
+       (DO_TEX1 && VB->TexCoordPtr[GET_TEXSOURCE(1)]->size == 4) ||
+       (DO_TEX0 && VB->TexCoordPtr[GET_TEXSOURCE(0)]->size == 4))
+      return GL_FALSE;
+
+   return GL_TRUE;
+}
+#else
+static GLboolean TAG(check_tex_sizes)( GLcontext *ctx )
+{
+   LOCALVARS
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+
+   /* Force 'missing' texcoords to something valid.
+    */
+   if (DO_TEX3 && VB->TexCoordPtr[2] == 0)
+      VB->TexCoordPtr[2] = VB->TexCoordPtr[3];
+
+   if (DO_TEX2 && VB->TexCoordPtr[1] == 0)
+      VB->TexCoordPtr[1] = VB->TexCoordPtr[2];
+
+   if (DO_TEX1 && VB->TexCoordPtr[0] == 0)
+      VB->TexCoordPtr[0] = VB->TexCoordPtr[1];
+
+   if (DO_PTEX)
+      return GL_TRUE;
+
+   /* No hardware support for projective texture.  Can fake it for
+    * TEX0 only.
+    */
+   if ((DO_TEX3 && VB->TexCoordPtr[GET_TEXSOURCE(3)]->size == 4) ||
+       (DO_TEX2 && VB->TexCoordPtr[GET_TEXSOURCE(2)]->size == 4) ||
+       (DO_TEX1 && VB->TexCoordPtr[GET_TEXSOURCE(1)]->size == 4)) {
+      PTEX_FALLBACK();
+      return GL_FALSE;
+   }
+
+   if (DO_TEX0 && VB->TexCoordPtr[GET_TEXSOURCE(0)]->size == 4) {
+      if (DO_TEX1 || DO_TEX2 || DO_TEX3) {
+	 PTEX_FALLBACK();
+      }
+      return GL_FALSE;
+   }
+
+   return GL_TRUE;
+}
+#endif /* ptex */
+
+
+static void TAG(interp)( GLcontext *ctx,
+			 GLfloat t,
+			 GLuint edst, GLuint eout, GLuint ein,
+			 GLboolean force_boundary )
+{
+   LOCALVARS
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+   GLubyte *ddverts = GET_VERTEX_STORE();
+   GLuint shift = GET_VERTEX_STRIDE_SHIFT();
+   const GLfloat *dstclip = VB->ClipPtr->data[edst];
+   GLfloat w;
+   const GLfloat *s = GET_VIEWPORT_MAT();
+
+   VERTEX *dst = (VERTEX *)(ddverts + (edst << shift));
+   VERTEX *in  = (VERTEX *)(ddverts + (ein << shift));
+   VERTEX *out = (VERTEX *)(ddverts + (eout << shift));
+
+   (void)s;
+
+   if (HAVE_HW_DIVIDE && CHECK_HW_DIVIDE) {
+      VIEWPORT_X( dst->v.x, dstclip[0] );
+      VIEWPORT_Y( dst->v.y, dstclip[1] );
+      VIEWPORT_Z( dst->v.z, dstclip[2] );
+      w = dstclip[3];
+   }
+   else {
+      w = 1.0 / dstclip[3];
+      VIEWPORT_X( dst->v.x, dstclip[0] * w );
+      VIEWPORT_Y( dst->v.y, dstclip[1] * w );
+      VIEWPORT_Z( dst->v.z, dstclip[2] * w );
+   }
+
+   if ((HAVE_HW_DIVIDE && CHECK_HW_DIVIDE) || 
+       DO_FOG || DO_SPEC || DO_TEX0 || DO_TEX1 ||
+       DO_TEX2 || DO_TEX3 || !HAVE_TINY_VERTICES) {
+
+      dst->v.w = w;
+
+      INTERP_UB( t, dst->ub4[4][0], out->ub4[4][0], in->ub4[4][0] );
+      INTERP_UB( t, dst->ub4[4][1], out->ub4[4][1], in->ub4[4][1] );
+      INTERP_UB( t, dst->ub4[4][2], out->ub4[4][2], in->ub4[4][2] );
+      INTERP_UB( t, dst->ub4[4][3], out->ub4[4][3], in->ub4[4][3] );
+
+      if (DO_SPEC) {
+	 INTERP_UB( t, dst->v.specular.red,   out->v.specular.red,   in->v.specular.red );
+	 INTERP_UB( t, dst->v.specular.green, out->v.specular.green, in->v.specular.green );
+	 INTERP_UB( t, dst->v.specular.blue,  out->v.specular.blue,  in->v.specular.blue );
+      }
+      if (DO_FOG) {
+	 INTERP_UB( t, dst->v.specular.alpha, out->v.specular.alpha, in->v.specular.alpha );
+      }
+      if (DO_TEX0) {
+	 if (DO_PTEX) {
+	    if (HAVE_PTEX_VERTICES) {
+	       INTERP_F( t, dst->pv.u0, out->pv.u0, in->pv.u0 );
+	       INTERP_F( t, dst->pv.v0, out->pv.v0, in->pv.v0 );
+	       INTERP_F( t, dst->pv.q0, out->pv.q0, in->pv.q0 );
+	    } else {
+	       GLfloat wout = VB->NdcPtr->data[eout][3];
+	       GLfloat win = VB->NdcPtr->data[ein][3];
+	       GLfloat qout = out->pv.w / wout;
+	       GLfloat qin = in->pv.w / win;
+	       GLfloat qdst, rqdst;
+
+	       ASSERT( !HAVE_HW_DIVIDE );
+
+	       INTERP_F( t, dst->v.u0, out->v.u0 * qout, in->v.u0 * qin );
+	       INTERP_F( t, dst->v.v0, out->v.v0 * qout, in->v.v0 * qin );
+	       INTERP_F( t, qdst, qout, qin );
+
+	       rqdst = 1.0 / qdst;
+	       dst->v.u0 *= rqdst;
+	       dst->v.v0 *= rqdst;
+	       dst->v.w *= rqdst;
+	    }
+	 }
+	 else {
+	    INTERP_F( t, dst->v.u0, out->v.u0, in->v.u0 );
+	    INTERP_F( t, dst->v.v0, out->v.v0, in->v.v0 );
+	 }
+      }
+      if (DO_TEX1) {
+	 if (DO_PTEX) {
+	    INTERP_F( t, dst->pv.u1, out->pv.u1, in->pv.u1 );
+	    INTERP_F( t, dst->pv.v1, out->pv.v1, in->pv.v1 );
+	    INTERP_F( t, dst->pv.q1, out->pv.q1, in->pv.q1 );
+	 } else {
+	    INTERP_F( t, dst->v.u1, out->v.u1, in->v.u1 );
+	    INTERP_F( t, dst->v.v1, out->v.v1, in->v.v1 );
+	 }
+      }
+      else if (DO_PTEX) {
+	 dst->pv.q1 = 0.0;	/* must be a valid float on radeon */
+      }
+      if (DO_TEX2) {
+	 if (DO_PTEX) {
+	    INTERP_F( t, dst->pv.u2, out->pv.u2, in->pv.u2 );
+	    INTERP_F( t, dst->pv.v2, out->pv.v2, in->pv.v2 );
+	    INTERP_F( t, dst->pv.q2, out->pv.q2, in->pv.q2 );
+	 } else {
+	    INTERP_F( t, dst->v.u2, out->v.u2, in->v.u2 );
+	    INTERP_F( t, dst->v.v2, out->v.v2, in->v.v2 );
+	 }
+      }
+      if (DO_TEX3) {
+	 if (DO_PTEX) {
+	    INTERP_F( t, dst->pv.u3, out->pv.u3, in->pv.u3 );
+	    INTERP_F( t, dst->pv.v3, out->pv.v3, in->pv.v3 );
+	    INTERP_F( t, dst->pv.q3, out->pv.q3, in->pv.q3 );
+	 } else {
+	    INTERP_F( t, dst->v.u3, out->v.u3, in->v.u3 );
+	    INTERP_F( t, dst->v.v3, out->v.v3, in->v.v3 );
+	 }
+      }
+   } else {
+      /* 4-dword vertex.  Color is in v[3] and there is no oow coordinate.
+       */
+      INTERP_UB( t, dst->ub4[3][0], out->ub4[3][0], in->ub4[3][0] );
+      INTERP_UB( t, dst->ub4[3][1], out->ub4[3][1], in->ub4[3][1] );
+      INTERP_UB( t, dst->ub4[3][2], out->ub4[3][2], in->ub4[3][2] );
+      INTERP_UB( t, dst->ub4[3][3], out->ub4[3][3], in->ub4[3][3] );
+   }
+}
+
+#endif /* rgba && xyzw */
+
+
+static void TAG(init)( void )
+{
+   setup_tab[IND].emit = TAG(emit);
+
+#if (DO_XYZW && DO_RGBA)
+   setup_tab[IND].check_tex_sizes = TAG(check_tex_sizes);
+   setup_tab[IND].interp = TAG(interp);
+#endif
+
+   if (DO_SPEC)
+      setup_tab[IND].copy_pv = copy_pv_rgba4_spec5;
+   else if (HAVE_HW_DIVIDE || DO_SPEC || DO_FOG || DO_TEX0 || DO_TEX1 ||
+	    DO_TEX2 || DO_TEX3 || !HAVE_TINY_VERTICES)
+      setup_tab[IND].copy_pv = copy_pv_rgba4;
+   else
+      setup_tab[IND].copy_pv = copy_pv_rgba3;
+
+   if (DO_TEX3) {
+      if (DO_PTEX) {
+	 ASSERT(HAVE_PTEX_VERTICES);
+	 setup_tab[IND].vertex_format = PROJ_TEX3_VERTEX_FORMAT;
+	 setup_tab[IND].vertex_size = 18;
+	 setup_tab[IND].vertex_stride_shift = 7;
+      }
+      else {
+	 setup_tab[IND].vertex_format = TEX3_VERTEX_FORMAT;
+	 setup_tab[IND].vertex_size = 14;
+	 setup_tab[IND].vertex_stride_shift = 6;
+      }
+   }
+   else if (DO_TEX2) {
+      if (DO_PTEX) {
+	 ASSERT(HAVE_PTEX_VERTICES);
+	 setup_tab[IND].vertex_format = PROJ_TEX3_VERTEX_FORMAT;
+	 setup_tab[IND].vertex_size = 18;
+	 setup_tab[IND].vertex_stride_shift = 7;
+      }
+      else {
+	 setup_tab[IND].vertex_format = TEX2_VERTEX_FORMAT;
+	 setup_tab[IND].vertex_size = 12;
+	 setup_tab[IND].vertex_stride_shift = 6;
+      }
+   }
+   else if (DO_TEX1) {
+      if (DO_PTEX) {
+	 ASSERT(HAVE_PTEX_VERTICES);
+	 setup_tab[IND].vertex_format = PROJ_TEX1_VERTEX_FORMAT;
+	 setup_tab[IND].vertex_size = 12;
+	 setup_tab[IND].vertex_stride_shift = 6;
+      }
+      else {
+	 setup_tab[IND].vertex_format = TEX1_VERTEX_FORMAT;
+	 setup_tab[IND].vertex_size = 10;
+	 setup_tab[IND].vertex_stride_shift = 6;
+      }
+   }
+   else if (DO_TEX0) {
+      if (DO_PTEX && HAVE_PTEX_VERTICES) {
+	 setup_tab[IND].vertex_format = PROJ_TEX1_VERTEX_FORMAT;
+	 setup_tab[IND].vertex_size = 12;
+	 setup_tab[IND].vertex_stride_shift = 6;
+      } else {
+	 setup_tab[IND].vertex_format = TEX0_VERTEX_FORMAT;
+	 setup_tab[IND].vertex_size = 8;
+	 setup_tab[IND].vertex_stride_shift = 5;
+      }
+   }
+   else if (!HAVE_HW_DIVIDE && !DO_SPEC && !DO_FOG && HAVE_TINY_VERTICES) {
+      setup_tab[IND].vertex_format = TINY_VERTEX_FORMAT;
+      setup_tab[IND].vertex_size = 4;
+      setup_tab[IND].vertex_stride_shift = 4;
+   } else if (HAVE_NOTEX_VERTICES) {
+      setup_tab[IND].vertex_format = NOTEX_VERTEX_FORMAT;
+      setup_tab[IND].vertex_size = 6;
+      setup_tab[IND].vertex_stride_shift = 5;
+   } else {
+      setup_tab[IND].vertex_format = TEX0_VERTEX_FORMAT;
+      setup_tab[IND].vertex_size = 8;
+      setup_tab[IND].vertex_stride_shift = 5;
+   }
+
+   assert(setup_tab[IND].vertex_size * 4 <=
+          1 << setup_tab[IND].vertex_stride_shift);
+}
+
+
+#undef IND
+#undef TAG
diff --git a/src/mesa/tnl_dd/t_dd_vertex.h b/src/mesa/tnl_dd/t_dd_vertex.h
new file mode 100644
index 00000000000..d45dd09fbd7
--- /dev/null
+++ b/src/mesa/tnl_dd/t_dd_vertex.h
@@ -0,0 +1,78 @@
+
+/*
+ * Mesa 3-D graphics library
+ * Version:  4.0.3
+ *
+ * Copyright (C) 1999-2002  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 <[email protected]>
+ */
+
+typedef struct {
+   GLfloat x, y, z, w;
+} TAG(_coord_t);
+
+#ifdef COLOR_IS_RGBA
+typedef struct {
+#if defined(BYTE_ORDER) && defined(BIG_ENDIAN) && BYTE_ORDER == BIG_ENDIAN
+   GLubyte alpha, blue, green, red;
+#else
+   GLubyte red, green, blue, alpha;
+#endif
+} TAG(_color_t);
+#else
+typedef struct {
+#if defined(BYTE_ORDER) && defined(BIG_ENDIAN) && BYTE_ORDER == BIG_ENDIAN
+   GLubyte alpha, red, green, blue;
+#else
+   GLubyte blue, green, red, alpha;
+#endif
+} TAG(_color_t);
+#endif
+
+typedef union {
+   struct {
+      GLfloat x, y, z, w;
+      TAG(_color_t) color;
+      TAG(_color_t) specular;
+      GLfloat u0, v0;
+      GLfloat u1, v1;
+      GLfloat u2, v2;
+      GLfloat u3, v3;
+   } v;
+   struct {
+      GLfloat x, y, z, w;
+      TAG(_color_t) color;
+      TAG(_color_t) specular;
+      GLfloat u0, v0, q0;
+      GLfloat u1, v1, q1;
+      GLfloat u2, v2, q2;
+      GLfloat u3, v3, q3;
+   } pv;
+   struct {
+      GLfloat x, y, z;
+      TAG(_color_t) color;
+   } tv;
+   GLfloat f[24];
+   GLuint  ui[24];
+   GLubyte ub4[24][4];
+} TAG(Vertex), *TAG(VertexPtr);
+