Reduce the amount of debug output, while still printing important messages.

Make guessing texture formats easier.

Makefile

@@ -33,7 +33,11 @@ realclean:
 
 install:
 	@echo "Installing"
-	$(TOP)/bin/installmesa 
+	$(TOP)/bin/installmesa
+
+# DirectFBGL module installation
+linux-directfb-install:
+	cd src/mesa/drivers/directfb && $(MAKE) install
 
 # If there's no current configuration file
 $(TOP)/configs/current:
@@ -78,6 +82,7 @@ linux \
 linux-alpha \
 linux-alpha-static \
 linux-debug \
+linux-directfb \
 linux-dri \
 linux-dri-x86 \
 linux-dri-x86-64 \
@@ -122,9 +127,9 @@ ultrix-gcc:
 
 # Rules for making release tarballs
 
-DIRECTORY = Mesa-6.2
-LIB_NAME = MesaLib-6.2
-DEMO_NAME = MesaDemos-6.2
+DIRECTORY = Mesa-6.3
+LIB_NAME = MesaLib-6.3
+DEMO_NAME = MesaDemos-6.3
 
 LIB_FILES =	\
 	$(DIRECTORY)/Makefile*						\
@@ -201,6 +206,8 @@ LIB_FILES =	\
 	$(DIRECTORY)/src/mesa/drivers/beos/Makefile			\
 	$(DIRECTORY)/src/mesa/drivers/common/*.[ch]			\
 	$(DIRECTORY)/src/mesa/drivers/common/descrip.mms		\
+	$(DIRECTORY)/src/mesa/drivers/directfb/*.[ch]		\
+	$(DIRECTORY)/src/mesa/drivers/directfb/Makefile		\
 	$(DIRECTORY)/src/mesa/drivers/dos/*.[chS]			\
 	$(DIRECTORY)/src/mesa/drivers/dri/common/*.[ch]			\
 	$(DIRECTORY)/src/mesa/drivers/fbdev/glfbdev.c			\
@@ -270,7 +277,7 @@ LIB_FILES =	\
 	$(DIRECTORY)/vms/xlib_share.opt
 
 
-DEMO_FILES =	\
+GLUT_FILES = \
 	$(DIRECTORY)/include/GL/glut.h			\
 	$(DIRECTORY)/include/GL/glutf90.h		\
 	$(DIRECTORY)/src/glut/glx/Makefile*		\
@@ -287,7 +294,10 @@ DEMO_FILES =	\
 	$(DIRECTORY)/src/glut/dos/Makefile.DJ		\
 	$(DIRECTORY)/src/glut/dos/PC_HW/*.[chS]		\
 	$(DIRECTORY)/src/glut/ggi/*.[ch]		\
-	$(DIRECTORY)/src/glut/ggi/Makefile		\
+	$(DIRECTORY)/src/glut/ggi/Makefile
+
+
+DEMO_FILES = \
 	$(DIRECTORY)/progs/beos/*.cpp			\
 	$(DIRECTORY)/progs/beos/Makefile		\
 	$(DIRECTORY)/progs/images/*.rgb			\
@@ -344,7 +354,7 @@ lib_gz:
 
 demo_gz:
 	cd .. ; \
-	tar -cvf $(DEMO_NAME).tar $(DEMO_FILES) ; \
+	tar -cvf $(DEMO_NAME).tar $(DEMO_FILES) $(GLUT_FILES) ; \
 	gzip $(DEMO_NAME).tar ; \
 	mv $(DEMO_NAME).tar.gz $(DIRECTORY)
 
@@ -357,7 +367,7 @@ lib_bz2:
 
 demo_bz2:
 	cd .. ; \
-	tar -cvf $(DEMO_NAME).tar $(DEMO_FILES) ; \
+	tar -cvf $(DEMO_NAME).tar $(DEMO_FILES) $(GLUT_FILES) ; \
 	bzip2 $(DEMO_NAME).tar ; \
 	mv $(DEMO_NAME).tar.bz2 $(DIRECTORY)
 
@@ -371,7 +381,7 @@ lib_zip:
 demo_zip:
 	-rm $(DEMO_NAME).zip ; \
 	cd .. ; \
-	zip -r $(DEMO_NAME).zip $(DEMO_FILES) ; \
+	zip -r $(DEMO_NAME).zip $(DEMO_FILES) $(GLUT_FILES) ; \
 	mv $(DEMO_NAME).zip $(DIRECTORY)
 
 md5:

configs/linux-directfb

@@ -0,0 +1,29 @@
+# Configuration for DirectFB
+
+include $(TOP)/configs/default
+
+CONFIG_NAME = linux-directfb
+
+# Compiler and flags
+CC  = gcc
+CXX = g++
+
+CFLAGS   = -Wall -O3 -ffast-math -fPIC -std=c99 -D_GNU_SOURCE -D_POSIX_SOURCE -D_SVID_SOURCE \
+         -D_POSIX_C_SOURCE=199309L -D_BSD_SOURCE -DPTHREADS
+
+CXXFLAGS = -Wall -O3 -fPIC -D_POSIX_SOURCE -D_POSIX_C_SOURCE=199309L -D_SVID_SOURCE -D_BSD_SOURCE   
+
+HAVE_X86 = $(shell uname -m | grep 'i[3-6]86' >/dev/null && echo yes)
+ifeq ($(HAVE_X86), yes)
+     CFLAGS   += -DUSE_X86_ASM -DUSE_MMX_ASM -DUSE_3DNOW_ASM -DUSE_SSE_ASM
+     CXXFLAGS += -DUSE_X86_ASM -DUSE_MMX_ASM -DUSE_3DNOW_ASM -DUSE_SSE_ASM
+     ASM_SOURCES = $(X86_SOURCES) $(X86_API)
+endif
+
+# Directories
+SRC_DIRS     = mesa glu
+DRIVER_DIRS  = directfb
+PROGRAM_DIRS = # disabled
+
+# Library/program dependencies
+GL_LIB_DEPS  = -lm -lpthread

docs/README.directfb

@@ -0,0 +1,28 @@
+                         
+                         Mesa DirectFB Information
+
+
+Requirements
+============
+
+  To build Mesa with DirectFB (DirectFBGL) support you need:
+     - DirectFB at least 0.9.21 (http://directfb.org)
+     - pkg-config at least 0.9 (http://pkgconfig.sf.net)
+
+
+Installation
+============
+  Run
+     
+     make linux-directfb
+
+  to build Mesa and DirectFBGL module,
+
+     make install
+
+  to install OpenGL libraries and
+
+     make linux-directfb-install
+
+  to install DirectFBGL module in the proper location.
+

docs/RELNOTES-6.3

@@ -25,6 +25,19 @@ GL_ARB_draw_buffers - allows a fragment program to write to a number of
 GL_OES_read_format - allows one to query the fastest glReadPixels format
   and datatype.
 
+GL_ARB_pixel_buffer_object - buffer objects for pixel read/write functions.
+
+DirectFB driver, contributed by Claudio Ciccani.  See docs/README.directfb
+for details.
+
+
+
+Vertex/Fragment Program PRINT Instruction
+-----------------------------------------
+
+The GL_NV_vertex_program and GL_NV_fragment_program languages have been
+extended with a PRINT instruction.
+
 
 
 To Do before release
@@ -53,4 +66,4 @@ D3D			needs updating
 
 
 ----------------------------------------------------------------------
-$Id: RELNOTES-6.3,v 3.6 2004/12/09 23:21:36 brianp Exp $
+$Id: RELNOTES-6.3,v 3.7 2005/01/03 15:55:51 brianp Exp $

docs/VERSIONS

@@ -1317,11 +1317,12 @@ Mesa Version History
 6.3  Month day, 2004
     New:
 	- GL_ARB_draw_buffers extension
+	- GL_ARB_pixel_buffer_object extension
 	- GL_OES_read_format extension (Ian Romanick)
-	- full support for GL_EXT_pixel_buffer_object (convolution filters,
-	  polygon stipple, colormaps, etc)
+	- DirectFB driver (Claudio Ciccani)
     Changes:
 	- added -stereo option for glxgears demo (Jacek Rosik)
     Bug fixes:
-	fixes from 6.2.1, plus:
-	- 
+	- some functions didn't support PBO functionality
+	- glGetTexImage didn't convert color index images to RGBA as required
+

include/GL/directfbgl.h

@@ -0,0 +1,89 @@
+/*
+   (c) Copyright 2001  convergence integrated media GmbH.
+   All rights reserved.
+
+   Written by Denis Oliver Kropp <dok@convergence.de> and
+              Andreas Hundt <andi@convergence.de>.
+
+   This library is free software; you can redistribute it and/or
+   modify it under the terms of the GNU Lesser General Public
+   License as published by the Free Software Foundation; either
+   version 2 of the License, or (at your option) any later version.
+
+   This library is distributed in the hope that it will be useful,
+   but WITHOUT ANY WARRANTY; without even the implied warranty of
+   MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+   Lesser General Public License for more details.
+
+   You should have received a copy of the GNU Lesser General Public
+   License along with this library; if not, write to the
+   Free Software Foundation, Inc., 59 Temple Place - Suite 330,
+   Boston, MA 02111-1307, USA.
+*/
+
+#ifndef __DIRECTFBGL_H__
+#define __DIRECTFBGL_H__
+
+#include <directfb.h>
+
+#ifdef __cplusplus
+extern "C"
+{
+#endif
+
+
+typedef struct {
+     int        buffer_size;
+     int        depth_size;
+     int        stencil_size;
+     int        aux_buffers;
+    
+     int        red_size;
+     int        green_size;
+     int        blue_size;
+     int        alpha_size;
+
+     int        accum_red_size;
+     int        accum_green_size;
+     int        accum_blue_size;
+     int        accum_alpha_size;
+
+     DFBBoolean double_buffer;
+     DFBBoolean stereo;
+} DFBGLAttributes;
+
+
+DEFINE_INTERFACE(   IDirectFBGL,
+
+   /** Context handling **/
+
+     /*
+      * Acquire the hardware lock.
+      */
+     DFBResult (*Lock) (
+          IDirectFBGL              *thiz
+     );
+
+     /*
+      * Release the lock.
+      */
+     DFBResult (*Unlock) (
+          IDirectFBGL              *thiz
+     );
+
+     /*
+      * Query the OpenGL attributes.
+      */
+     DFBResult (*GetAttributes) (
+          IDirectFBGL              *thiz,
+          DFBGLAttributes          *attributes
+     );
+)
+
+
+#ifdef __cplusplus
+}
+#endif
+
+#endif
+

include/GL/glx.h

@@ -1,8 +1,8 @@
 /*
  * Mesa 3-D graphics library
- * Version:  6.2
+ * Version:  6.3
  * 
- * Copyright (C) 1999-2004  Brian Paul   All Rights Reserved.
+ * Copyright (C) 1999-2005  Brian Paul   All Rights Reserved.
  * 
  * Permission is hereby granted, free of charge, to any person obtaining a
  * copy of this software and associated documentation files (the "Software"),
@@ -495,6 +495,17 @@ extern Bool glXDrawableAttribARB(Display *dpy, GLXDrawable draw, const int *attr
 #endif /* GLX_ARB_render_texture */
 
 
+/*
+ * Remove this when glxext.h is updated.
+ */
+#ifndef GLX_NV_float_buffer
+#define GLX_NV_float_buffer 1
+
+#define GLX_FLOAT_COMPONENTS_NV         0x20B0
+
+#endif /* GLX_NV_float_buffer */
+
+
 
 #ifdef __cplusplus
 }

progs/xdemos/pbdemo.c

@@ -41,22 +41,23 @@
 
 
 /* Some ugly global vars */
-static GLXFBConfigSGIX gFBconfig = 0;
 static Display *gDpy = NULL;
 static int gScreen = 0;
-static GLXPbufferSGIX gPBuffer = 0;
+static FBCONFIG gFBconfig = 0;
+static PBUFFER gPBuffer = 0;
 static int gWidth, gHeight;
+static GLXContext glCtx;
 
 
 
 /*
- * Create the pbuffer and return a GLXPbufferSGIX handle.
+ * Create the pbuffer and return a GLXPbuffer handle.
  *
  * We loop over a list of fbconfigs trying to create
  * a pixel buffer.  We return the first pixel buffer which we successfully
  * create.
  */
-static GLXPbufferSGIX
+static PBUFFER
 MakePbuffer( Display *dpy, int screen, int width, int height )
 {
 #define NUM_FB_CONFIGS 4
@@ -69,8 +70,8 @@ MakePbuffer( Display *dpy, int screen, int width, int height )
    int fbAttribs[NUM_FB_CONFIGS][100] = {
       {
          /* Single buffered, with depth buffer */
-         GLX_RENDER_TYPE_SGIX, GLX_RGBA_BIT_SGIX,
-         GLX_DRAWABLE_TYPE_SGIX, GLX_PBUFFER_BIT_SGIX,
+         GLX_RENDER_TYPE, GLX_RGBA_BIT,
+         GLX_DRAWABLE_TYPE, GLX_PBUFFER_BIT,
          GLX_RED_SIZE, 1,
          GLX_GREEN_SIZE, 1,
          GLX_BLUE_SIZE, 1,
@@ -81,8 +82,8 @@ MakePbuffer( Display *dpy, int screen, int width, int height )
       },
       {
          /* Double buffered, with depth buffer */
-         GLX_RENDER_TYPE_SGIX, GLX_RGBA_BIT_SGIX,
-         GLX_DRAWABLE_TYPE_SGIX, GLX_PBUFFER_BIT_SGIX,
+         GLX_RENDER_TYPE, GLX_RGBA_BIT,
+         GLX_DRAWABLE_TYPE, GLX_PBUFFER_BIT,
          GLX_RED_SIZE, 1,
          GLX_GREEN_SIZE, 1,
          GLX_BLUE_SIZE, 1,
@@ -93,8 +94,8 @@ MakePbuffer( Display *dpy, int screen, int width, int height )
       },
       {
          /* Single bufferd, without depth buffer */
-         GLX_RENDER_TYPE_SGIX, GLX_RGBA_BIT_SGIX,
-         GLX_DRAWABLE_TYPE_SGIX, GLX_PBUFFER_BIT_SGIX,
+         GLX_RENDER_TYPE, GLX_RGBA_BIT,
+         GLX_DRAWABLE_TYPE, GLX_PBUFFER_BIT,
          GLX_RED_SIZE, 1,
          GLX_GREEN_SIZE, 1,
          GLX_BLUE_SIZE, 1,
@@ -105,8 +106,8 @@ MakePbuffer( Display *dpy, int screen, int width, int height )
       },
       {
          /* Double bufferd, without depth buffer */
-         GLX_RENDER_TYPE_SGIX, GLX_RGBA_BIT_SGIX,
-         GLX_DRAWABLE_TYPE_SGIX, GLX_PBUFFER_BIT_SGIX,
+         GLX_RENDER_TYPE, GLX_RGBA_BIT,
+         GLX_DRAWABLE_TYPE, GLX_PBUFFER_BIT,
          GLX_RED_SIZE, 1,
          GLX_GREEN_SIZE, 1,
          GLX_BLUE_SIZE, 1,
@@ -116,13 +117,10 @@ MakePbuffer( Display *dpy, int screen, int width, int height )
          None
       }
    };
-   int pbAttribs[] = {
-      GLX_LARGEST_PBUFFER_SGIX, True,
-      GLX_PRESERVED_CONTENTS_SGIX, False,
-      None
-   };
-   GLXFBConfigSGIX *fbConfigs;
-   GLXPbufferSGIX pBuffer = None;
+   Bool largest = True;
+   Bool preserve = False;
+   FBCONFIG *fbConfigs;
+   PBUFFER pBuffer = None;
    int nConfigs;
    int i;
    int attempt;
@@ -130,23 +128,23 @@ MakePbuffer( Display *dpy, int screen, int width, int height )
    for (attempt=0; attempt<NUM_FB_CONFIGS; attempt++) {
 
       /* Get list of possible frame buffer configurations */
-      fbConfigs = glXChooseFBConfigSGIX(dpy, screen, fbAttribs[attempt], &nConfigs);
+      fbConfigs = ChooseFBConfig(dpy, screen, fbAttribs[attempt], &nConfigs);
       if (nConfigs==0 || !fbConfigs) {
-         printf("Error: glxChooseFBConfigSGIX failed\n");
+         printf("Error: glXChooseFBConfig failed\n");
          XCloseDisplay(dpy);
          return 0;
       }
 
-#ifdef DEBUG
+#if 0 /*DEBUG*/
       for (i=0;i<nConfigs;i++) {
          printf("Config %d\n", i);
-         PrintFBConfigInfo(dpy, fbConfigs[i], 0);
+         PrintFBConfigInfo(dpy, screen, fbConfigs[i], 0);
       }
 #endif
 
       /* Create the pbuffer using first fbConfig in the list that works. */
       for (i=0;i<nConfigs;i++) {
-         pBuffer = CreatePbuffer(dpy, fbConfigs[i], width, height, pbAttribs);
+         pBuffer = CreatePbuffer(dpy, screen, fbConfigs[i], width, height, preserve, largest);
          if (pBuffer) {
             gFBconfig = fbConfigs[i];
             gWidth = width;
@@ -178,9 +176,8 @@ MakePbuffer( Display *dpy, int screen, int width, int height )
 static int
 Setup(int width, int height)
 {
-#if defined(GLX_SGIX_fbconfig) && defined(GLX_SGIX_pbuffer)
+   int pbSupport;
    XVisualInfo *visInfo;
-   GLXContext glCtx;
 
    /* Open the X display */
    gDpy = XOpenDisplay(NULL);
@@ -193,7 +190,14 @@ Setup(int width, int height)
    gScreen = DefaultScreen(gDpy);
 
    /* Test that pbuffers are available */
-   if (!QueryPbuffers(gDpy, gScreen)) {
+   pbSupport = QueryPbuffers(gDpy, gScreen);
+   if (pbSupport == 1) {
+      printf("Using GLX 1.3 Pbuffers\n");
+   }
+   else if (pbSupport == 2) {
+      printf("Using SGIX Pbuffers\n");
+   }
+   else {
       printf("Error: pbuffers not available on this screen\n");
       XCloseDisplay(gDpy);
       return 0;
@@ -208,7 +212,7 @@ Setup(int width, int height)
    }
 
    /* Get corresponding XVisualInfo */
-   visInfo = glXGetVisualFromFBConfigSGIX(gDpy, gFBconfig);
+   visInfo = GetVisualFromFBConfig(gDpy, gScreen, gFBconfig);
    if (!visInfo) {
       printf("Error: can't get XVisualInfo from FBconfig\n");
       XCloseDisplay(gDpy);
@@ -240,11 +244,6 @@ Setup(int width, int height)
    }
 
    return 1;  /* Success!! */
-#else
-   printf("Error: GLX_SGIX_fbconfig and/or GLX_SGIX_pbuffer extensions not"
-                  " available at compile-time.\n");
-   return 0;
-#endif
 }
 
 
@@ -360,7 +359,6 @@ Render(void)
    int NumBoxes = 100;
    int i;
 
-   InitGL();
    glClearColor(0.2, 0.2, 0.9, 0.0);
    glClear(GL_COLOR_BUFFER_BIT | GL_DEPTH_BUFFER_BIT);
 
@@ -470,9 +468,10 @@ main(int argc, char *argv[])
       if (!Setup(width, height)) {
          return 1;
       }
+      InitGL();
       Render();
       WriteFile(fileName);
-      glXDestroyGLXPbufferSGIX( gDpy, gPBuffer );
+      DestroyPbuffer(gDpy, gScreen, gPBuffer);
    }
    return 0;
 }

progs/xdemos/pbinfo.c

@@ -20,18 +20,18 @@
 static void
 PrintConfigs(Display *dpy, int screen, Bool horizFormat)
 {
-   GLXFBConfigSGIX *fbConfigs;
+   FBCONFIG *fbConfigs;
    int nConfigs;
    int i;
    /* Note: you may want to tweek the attribute list to select a different
     * set of fbconfigs.
     */
    int fbAttribs[] = {
-                      GLX_RENDER_TYPE_SGIX, 0,
-		      GLX_DRAWABLE_TYPE_SGIX, 0,
+                      GLX_RENDER_TYPE, 0,
+		      GLX_DRAWABLE_TYPE, 0,
 #if 0
-                      GLX_RENDER_TYPE_SGIX, GLX_RGBA_BIT_SGIX,
-		      GLX_DRAWABLE_TYPE_SGIX, GLX_PIXMAP_BIT_SGIX,
+                      GLX_RENDER_TYPE, GLX_RGBA_BIT,
+		      GLX_DRAWABLE_TYPE, GLX_PIXMAP_BIT,
 		      GLX_RED_SIZE, 1,
 		      GLX_GREEN_SIZE, 1,
 		      GLX_BLUE_SIZE, 1,
@@ -43,17 +43,9 @@ PrintConfigs(Display *dpy, int screen, Bool horizFormat)
 
 
    /* Get list of possible frame buffer configurations */
-#if 0
-   /* SGIX method */
-   fbConfigs = glXChooseFBConfigSGIX(dpy, screen, fbAttribs, &nConfigs);
-#else
-   /* GLX 1.3 method */
-   (void) fbAttribs;
-   fbConfigs = glXGetFBConfigs(dpy, screen, &nConfigs);
-#endif
-
-   if (nConfigs==0 || !fbConfigs) {
-      printf("Error: glxChooseFBConfigSGIX failed\n");
+   fbConfigs = ChooseFBConfig(dpy, screen, fbAttribs, &nConfigs);
+   if (!nConfigs || !fbConfigs) {
+      printf("Error: glxChooseFBConfig failed\n");
       return;
    }
 
@@ -61,12 +53,12 @@ PrintConfigs(Display *dpy, int screen, Bool horizFormat)
 
    if (horizFormat) {
       printf("  ID  VisualType  Depth Lvl RGB CI DB Stereo  R  G  B  A");
-      printf("   Z  S  AR AG AB AA  MSbufs MSnum  Pbuffer\n");
+      printf("   Z  S  AR AG AB AA  MSbufs MSnum  Pbuffer  Float\n");
    }
 
    /* Print config info */
    for (i=0;i<nConfigs;i++) {
-      PrintFBConfigInfo(dpy, fbConfigs[i], horizFormat);
+      PrintFBConfigInfo(dpy, screen, fbConfigs[i], horizFormat);
    }
 
    /* free the list */

progs/xdemos/pbutil.c

@@ -3,8 +3,10 @@
  * OpenGL pbuffers utility functions.
  *
  * Brian Paul
- * April 1997
+ * Original code: April 1997
  * Updated on 5 October 2002
+ * Updated again on 3 January 2005 to use GLX 1.3 functions in preference
+ * to the GLX_SGIX_fbconfig/pbuffer extensions.
  */
 
 
@@ -13,51 +15,140 @@
 #include "pbutil.h"
 
 
-
-/*
+/**
  * Test if we pixel buffers are available for a particular X screen.
  * Input:  dpy - the X display
  *         screen - screen number
  * Return:  0 = pixel buffers not available.
- *          1 = pixel buffers are available.
+ *          1 = pixel buffers are available via GLX 1.3.
+ *          2 = pixel buffers are available via GLX_SGIX_fbconfig/pbuffer.
  */
 int
 QueryPbuffers(Display *dpy, int screen)
 {
-#if defined(GLX_SGIX_fbconfig) && defined(GLX_SGIX_pbuffer)
-   char *extensions;
-
-   extensions = (char *) glXQueryServerString(dpy, screen, GLX_EXTENSIONS);
-   if (!strstr(extensions,"GLX_SGIX_fbconfig")) {
-      return 0;
+#if defined(GLX_VERSION_1_3)
+   {
+      /* GLX 1.3 supports pbuffers */
+      int glxVersionMajor, glxVersionMinor;
+      if (!glXQueryVersion(dpy, &glxVersionMajor, &glxVersionMinor)) {
+         /* GLX not available! */
+         return 0;
+      }
+      if (glxVersionMajor * 100 + glxVersionMinor >= 103) {
+         return 1;
+      }
+      /* fall-through */
    }
-   if (!strstr(extensions,"GLX_SGIX_pbuffer")) {
-      return 0;
-   }
-   return 1;
-#else
-   return 0;
 #endif
+
+#if defined(GLX_SGIX_fbconfig) && defined(GLX_SGIX_pbuffer)
+   /* Try the SGIX extensions */
+   {
+      char *extensions;
+      extensions = (char *) glXQueryServerString(dpy, screen, GLX_EXTENSIONS);
+      if (!extensions ||
+          !strstr(extensions,"GLX_SGIX_fbconfig") ||
+          !strstr(extensions,"GLX_SGIX_pbuffer")) {
+         return 0;
+      }
+      return 2;
+   }
+#endif
+
+   return 0;
 }
 
 
 
-#ifdef GLX_SGIX_fbconfig
+FBCONFIG *
+ChooseFBConfig(Display *dpy, int screen, const int attribs[], int *nConfigs)
+{
+   int pbSupport = QueryPbuffers(dpy, screen);
+#if defined(GLX_VERSION_1_3)
+   if (pbSupport == 1) {
+      return glXChooseFBConfig(dpy, screen, attribs, nConfigs);
+   }
+#endif
+#if defined(GLX_SGIX_fbconfig) && defined(GLX_SGIX_pbuffer)
+   if (pbSupport == 2) {
+      return glXChooseFBConfigSGIX(dpy, screen, (int *) attribs, nConfigs);
+   }
+#endif
+   return NULL;
+}
 
 
-/*
+
+XVisualInfo *
+GetVisualFromFBConfig(Display *dpy, int screen, FBCONFIG config)
+{
+   int pbSupport = QueryPbuffers(dpy, screen);
+#if defined(GLX_VERSION_1_3)
+   if (pbSupport == 1) {
+      return glXGetVisualFromFBConfig(dpy, config);
+   }
+#endif
+#if defined(GLX_SGIX_fbconfig) && defined(GLX_SGIX_pbuffer)
+   if (pbSupport == 2) {
+      return glXGetVisualFromFBConfigSGIX(dpy, config);
+   }
+#endif
+   return NULL;
+}
+
+
+/**
+ * Either use glXGetFBConfigAttrib() or glXGetFBConfigAttribSGIX()
+ * to query an fbconfig attribute.
+ */
+static int
+GetFBConfigAttrib(Display *dpy,
+#if defined(GLX_VERSION_1_3)
+                  const GLXFBConfig config,
+#elif defined(GLX_SGIX_fbconfig)
+                  const GLXFBConfigSGIX config,
+#endif
+                  int attrib
+                  )
+{
+   int value;
+
+#if defined(GLX_VERSION_1_3)
+   int glxVersionMajor, glxVersionMinor;
+   if (glXQueryVersion(dpy, &glxVersionMajor, &glxVersionMinor)
+       && glxVersionMajor * 100 + glxVersionMinor >= 103) {
+      /* ok */
+      if (glXGetFBConfigAttrib(dpy, config, attrib, &value) != 0) {
+         value = 0;
+      }
+      return value;
+   }
+   /* fall-through */
+#endif
+
+#if defined(GLX_SGIX_fbconfig) && defined(GLX_SGIX_pbuffer)
+   if (glXGetFBConfigAttribSGIX(dpy, config, attrib, &value) != 0) {
+      value = 0;
+   }
+   return value;
+#endif
+   
+   return 0;
+}
+
+
+
+/**
  * Print parameters for a GLXFBConfig to stdout.
  * Input:  dpy - the X display
+ *         screen - the X screen number
  *         fbConfig - the fbconfig handle
  *         horizFormat - if true, print in horizontal format
  */
 void
-PrintFBConfigInfo(Display *dpy, GLXFBConfigSGIX fbConfig, Bool horizFormat)
+PrintFBConfigInfo(Display *dpy, int screen, FBCONFIG config, Bool horizFormat)
 {
-   int pbAttribs[] = {GLX_LARGEST_PBUFFER_SGIX, True,
-		      GLX_PRESERVED_CONTENTS_SGIX, False,
-		      None};
-   GLXPbufferSGIX pBuffer;
+   PBUFFER pBuffer;
    int width=2, height=2;
    int bufferSize, level, doubleBuffer, stereo, auxBuffers;
    int redSize, greenSize, blueSize, alphaSize;
@@ -67,44 +158,49 @@ PrintFBConfigInfo(Display *dpy, GLXFBConfigSGIX fbConfig, Bool horizFormat)
    int drawableType, renderType, xRenderable, xVisual, id;
    int maxWidth, maxHeight, maxPixels;
    int optWidth, optHeight;
+   int floatComponents;
 
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_BUFFER_SIZE, &bufferSize);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_LEVEL, &level);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_DOUBLEBUFFER, &doubleBuffer);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_STEREO, &stereo);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_AUX_BUFFERS, &auxBuffers);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_RED_SIZE, &redSize);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_GREEN_SIZE, &greenSize);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_BLUE_SIZE, &blueSize);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_ALPHA_SIZE, &alphaSize);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_DEPTH_SIZE, &depthSize);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_STENCIL_SIZE, &stencilSize);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_ACCUM_RED_SIZE, &accumRedSize);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_ACCUM_GREEN_SIZE, &accumGreenSize);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_ACCUM_BLUE_SIZE, &accumBlueSize);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_ACCUM_ALPHA_SIZE, &accumAlphaSize);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_SAMPLE_BUFFERS_SGIS, &sampleBuffers);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_SAMPLES_SGIS, &samples);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_DRAWABLE_TYPE_SGIX, &drawableType);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_RENDER_TYPE_SGIX, &renderType);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_X_RENDERABLE_SGIX, &xRenderable);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_X_VISUAL_TYPE_EXT, &xVisual);
+   /* do queries using the GLX 1.3 tokens (same as the SGIX tokens) */
+   bufferSize     = GetFBConfigAttrib(dpy, config, GLX_BUFFER_SIZE);
+   level          = GetFBConfigAttrib(dpy, config, GLX_LEVEL);
+   doubleBuffer   = GetFBConfigAttrib(dpy, config, GLX_DOUBLEBUFFER);
+   stereo         = GetFBConfigAttrib(dpy, config, GLX_STEREO);
+   auxBuffers     = GetFBConfigAttrib(dpy, config, GLX_AUX_BUFFERS);
+   redSize        = GetFBConfigAttrib(dpy, config, GLX_RED_SIZE);
+   greenSize      = GetFBConfigAttrib(dpy, config, GLX_GREEN_SIZE);
+   blueSize       = GetFBConfigAttrib(dpy, config, GLX_BLUE_SIZE);
+   alphaSize      = GetFBConfigAttrib(dpy, config, GLX_ALPHA_SIZE);
+   depthSize      = GetFBConfigAttrib(dpy, config, GLX_DEPTH_SIZE);
+   stencilSize    = GetFBConfigAttrib(dpy, config, GLX_STENCIL_SIZE);
+   accumRedSize   = GetFBConfigAttrib(dpy, config, GLX_ACCUM_RED_SIZE);
+   accumGreenSize = GetFBConfigAttrib(dpy, config, GLX_ACCUM_GREEN_SIZE);
+   accumBlueSize  = GetFBConfigAttrib(dpy, config, GLX_ACCUM_BLUE_SIZE);
+   accumAlphaSize = GetFBConfigAttrib(dpy, config, GLX_ACCUM_ALPHA_SIZE);
+   sampleBuffers  = GetFBConfigAttrib(dpy, config, GLX_SAMPLE_BUFFERS);
+   samples        = GetFBConfigAttrib(dpy, config, GLX_SAMPLES);
+   drawableType   = GetFBConfigAttrib(dpy, config, GLX_DRAWABLE_TYPE);
+   renderType     = GetFBConfigAttrib(dpy, config, GLX_RENDER_TYPE);
+   xRenderable    = GetFBConfigAttrib(dpy, config, GLX_X_RENDERABLE);
+   xVisual        = GetFBConfigAttrib(dpy, config, GLX_X_VISUAL_TYPE);
    if (!xRenderable || !(drawableType & GLX_WINDOW_BIT_SGIX))
       xVisual = -1;
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_FBCONFIG_ID_SGIX, &id);
 
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_MAX_PBUFFER_WIDTH_SGIX,
-			    &maxWidth);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_MAX_PBUFFER_HEIGHT_SGIX,
-			    &maxHeight);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_MAX_PBUFFER_PIXELS_SGIX,
-			    &maxPixels);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_OPTIMAL_PBUFFER_WIDTH_SGIX,
-			    &optWidth);
-   glXGetFBConfigAttribSGIX(dpy, fbConfig, GLX_OPTIMAL_PBUFFER_HEIGHT_SGIX,
-			    &optHeight);
+   id        = GetFBConfigAttrib(dpy, config, GLX_FBCONFIG_ID);
+   maxWidth  = GetFBConfigAttrib(dpy, config, GLX_MAX_PBUFFER_WIDTH);
+   maxHeight = GetFBConfigAttrib(dpy, config, GLX_MAX_PBUFFER_HEIGHT);
+   maxPixels = GetFBConfigAttrib(dpy, config, GLX_MAX_PBUFFER_PIXELS);
+#if defined(GLX_SGIX_pbuffer)
+   optWidth  = GetFBConfigAttrib(dpy, config, GLX_OPTIMAL_PBUFFER_WIDTH_SGIX);
+   optHeight = GetFBConfigAttrib(dpy, config, GLX_OPTIMAL_PBUFFER_HEIGHT_SGIX);
+#else
+   optWidth = optHeight = 0;
+#endif
+#if defined(GLX_NV_float_buffer)
+   floatComponents = GetFBConfigAttrib(dpy, config, GLX_FLOAT_COMPONENTS_NV);
+#endif
 
-   pBuffer = CreatePbuffer(dpy, fbConfig, width, height, pbAttribs);
+   /* See if we can create a pbuffer with this config */
+   pBuffer = CreatePbuffer(dpy, screen, config, width, height, False, False);
 
    if (horizFormat) {
       printf("0x%03x ", id);
@@ -125,7 +221,8 @@ PrintFBConfigInfo(Display *dpy, GLXFBConfigSGIX fbConfig, Bool horizFormat)
       printf("%2d %2d %2d %2d", accumRedSize, accumGreenSize, accumBlueSize,
 	     accumAlphaSize);
       printf("    %2d    %2d", sampleBuffers, samples);
-      printf("       %s", pBuffer ? "y" : "n");
+      printf("       %s       %c", pBuffer ? "y" : "n",
+             "ny"[floatComponents]);
       printf("\n");
    }
    else {
@@ -148,27 +245,28 @@ PrintFBConfigInfo(Display *dpy, GLXFBConfigSGIX fbConfig, Bool horizFormat)
       printf("  Sample Buffers: %d\n", sampleBuffers);
       printf("  Samples/Pixel: %d\n", samples);
       printf("  Drawable Types: ");
-      if (drawableType & GLX_WINDOW_BIT_SGIX)  printf("Window ");
-      if (drawableType & GLX_PIXMAP_BIT_SGIX)  printf("Pixmap ");
-      if (drawableType & GLX_PBUFFER_BIT_SGIX)  printf("PBuffer");
+      if (drawableType & GLX_WINDOW_BIT)  printf("Window ");
+      if (drawableType & GLX_PIXMAP_BIT)  printf("Pixmap ");
+      if (drawableType & GLX_PBUFFER_BIT)  printf("PBuffer");
       printf("\n");
       printf("  Render Types: ");
       if (renderType & GLX_RGBA_BIT_SGIX)  printf("RGBA ");
       if (renderType & GLX_COLOR_INDEX_BIT_SGIX)  printf("CI ");
       printf("\n");
       printf("  X Renderable: %s\n", xRenderable ? "yes" : "no");
-      /*
-      printf("  Max width: %d\n", maxWidth);
-      printf("  Max height: %d\n", maxHeight);
-      printf("  Max pixels: %d\n", maxPixels);
-      printf("  Optimum width: %d\n", optWidth);
-      printf("  Optimum height: %d\n", optHeight);
-      */
+
       printf("  Pbuffer: %s\n", pBuffer ? "yes" : "no");
+      printf("  Max Pbuffer width: %d\n", maxWidth);
+      printf("  Max Pbuffer height: %d\n", maxHeight);
+      printf("  Max Pbuffer pixels: %d\n", maxPixels);
+      printf("  Optimum Pbuffer width: %d\n", optWidth);
+      printf("  Optimum Pbuffer height: %d\n", optHeight);
+
+      printf("  Float Components: %s\n", floatComponents ? "yes" : "no");
    }
 
    if (pBuffer) {
-      glXDestroyGLXPbufferSGIX(dpy, pBuffer);
+      DestroyPbuffer(dpy, screen, pBuffer);
    }
 }
 
@@ -176,42 +274,73 @@ PrintFBConfigInfo(Display *dpy, GLXFBConfigSGIX fbConfig, Bool horizFormat)
 
 /* This is only used by CreatePbuffer() */
 static int XErrorFlag = 0;
-static int HandleXError( Display *dpy, XErrorEvent *event )
+static int HandleXError(Display *dpy, XErrorEvent *event)
 {
     XErrorFlag = 1;
     return 0;
 }
 
 
-/*
+/**
  * Create a Pbuffer.  Use an X error handler to deal with potential
  * BadAlloc errors.
  *
  * Input:  dpy - the X display
  *         fbConfig - an FBConfig as returned by glXChooseFBConfigSGIX().
  *         width, height - size of pixel buffer to request, in pixels.
- *         pbAttribs - list of pixel buffer attributes as used by
- *                     glXCreateGLXPbufferSGIX().
- * Return:  a pixel buffer or None.
+ *         pbAttribs - list of optional pixel buffer attributes
+ * Return:  a Pbuffer or None.
  */
-GLXPbufferSGIX
-CreatePbuffer( Display *dpy, GLXFBConfigSGIX fbConfig,
-	       int width, int height, int *pbAttribs )
+PBUFFER
+CreatePbuffer(Display *dpy, int screen, FBCONFIG config,
+              int width, int height, Bool largest, Bool preserve)
 {
-   int (*oldHandler)( Display *, XErrorEvent * );
-   GLXPbufferSGIX pBuffer = None;
+   int (*oldHandler)(Display *, XErrorEvent *);
+   PBUFFER pBuffer = None;
+   int pbSupport = QueryPbuffers(dpy, screen);
 
    /* Catch X protocol errors with our own error handler */
-   oldHandler = XSetErrorHandler( HandleXError );
-
+   oldHandler = XSetErrorHandler(HandleXError);
    XErrorFlag = 0;
-   pBuffer = glXCreateGLXPbufferSGIX(dpy, fbConfig, width, height, pbAttribs);
+
+#if defined(GLX_VERSION_1_3)
+   if (pbSupport == 1) {
+      /* GLX 1.3 */
+      int attribs[100], i = 0;
+      attribs[i++] = GLX_PBUFFER_WIDTH;
+      attribs[i++] = width;
+      attribs[i++] = GLX_PBUFFER_HEIGHT;
+      attribs[i++] = height;
+      attribs[i++] = GLX_PRESERVED_CONTENTS;
+      attribs[i++] = preserve;
+      attribs[i++] = GLX_LARGEST_PBUFFER;
+      attribs[i++] = largest;
+      attribs[i++] = 0;
+      pBuffer = glXCreatePbuffer(dpy, config, attribs);
+   }
+   else
+#endif
+#if defined(GLX_SGIX_fbconfig) && defined(GLX_SGIX_pbuffer)
+   if (pbSupport == 2) {
+      int attribs[100], i = 0;
+      attribs[i++] = GLX_PRESERVED_CONTENTS;
+      attribs[i++] = preserve;
+      attribs[i++] = GLX_LARGEST_PBUFFER;
+      attribs[i++] = largest;
+      attribs[i++] = 0;
+      pBuffer = glXCreateGLXPbufferSGIX(dpy, config, width, height, attribs);
+   }
+   else
+#endif
+   {
+      pBuffer = None;
+   }
 
    /* Restore original X error handler */
-   (void) XSetErrorHandler( oldHandler );
+   (void) XSetErrorHandler(oldHandler);
 
    /* Return pbuffer (may be None) */
-   if (!XErrorFlag && pBuffer!=None) {
+   if (!XErrorFlag && pBuffer != None) {
       /*printf("config %d worked!\n", i);*/
       return pBuffer;
    }
@@ -221,7 +350,20 @@ CreatePbuffer( Display *dpy, GLXFBConfigSGIX fbConfig,
 }
 
 
-
-#endif  /*GLX_SGIX_fbconfig*/
-
-
+void
+DestroyPbuffer(Display *dpy, int screen, PBUFFER pbuffer)
+{
+   int pbSupport = QueryPbuffers(dpy, screen);
+#if defined(GLX_VERSION_1_3)
+   if (pbSupport == 1) {
+      glXDestroyPbuffer(dpy, pbuffer);
+      return;
+   }
+#endif
+#if defined(GLX_SGIX_fbconfig) && defined(GLX_SGIX_pbuffer)
+   if (pbSupport == 2) {
+      glXDestroyGLXPbufferSGIX(dpy, pbuffer);
+      return;
+   }
+#endif
+}

progs/xdemos/pbutil.h

@@ -1,4 +1,3 @@
-
 /*
  * OpenGL pbuffers utility functions.
  *
@@ -15,23 +14,41 @@
 #include <GL/glx.h>
 
 
+#if defined(GLX_VERSION_1_3)
+#define PBUFFER GLXPbuffer
+#define FBCONFIG GLXFBConfig
+#elif defined(GLX_SGIX_fbconfig) && defined(GLX_SGIX_pbuffer)
+#define PBUFFER GLXPbufferSGIX
+#define FBCONFIG GLXFBConfigSGIX
+#else
+#define PBUFFER int
+#define FBCONFIG int
+#endif
+
+
 extern int
 QueryPbuffers(Display *dpy, int screen);
 
 
-#ifdef GLX_SGIX_fbconfig
+extern void
+PrintFBConfigInfo(Display *dpy, int screen, FBCONFIG config, Bool horizFormat);
+
+
+extern FBCONFIG *
+ChooseFBConfig(Display *dpy, int screen, const int attribs[], int *nConfigs);
+
+
+extern XVisualInfo *
+GetVisualFromFBConfig(Display *dpy, int screen, FBCONFIG config);
+
+
+extern PBUFFER
+CreatePbuffer(Display *dpy, int screen, FBCONFIG config,
+	      int width, int height, Bool preserve, Bool largest);
 
 
 extern void
-PrintFBConfigInfo(Display *dpy, GLXFBConfigSGIX fbConfig, Bool horizFormat);
-
-
-extern GLXPbufferSGIX
-CreatePbuffer( Display *dpy, GLXFBConfigSGIX fbConfig,
-	       int width, int height, int *pbAttribs );
-
-
-#endif  /*GLX_SGIX_fbconfig*/
+DestroyPbuffer(Display *dpy, int screen, PBUFFER pbuffer);
 
 
 #endif  /*PBUTIL_H*/

src/glx/x11/indirect.c

@@ -276,8 +276,8 @@ __indirect_glCallLists(GLsizei n, GLenum type, const GLvoid * lists)
             const GLint op = X_GLrop_CallLists;
             const GLuint cmdlenLarge = cmdlen + 4;
             GLubyte * const pc = __glXFlushRenderBuffer(gc, gc->pc);
-            (void) memcpy((void *)(pc + 0), (void *)(&op), 4);
-            (void) memcpy((void *)(pc + 4), (void *)(&cmdlenLarge), 4);
+            (void) memcpy((void *)(pc + 0), (void *)(&cmdlenLarge), 4);
+            (void) memcpy((void *)(pc + 4), (void *)(&op), 4);
             (void) memcpy((void *)(pc + 8), (void *)(&n), 4);
             (void) memcpy((void *)(pc + 12), (void *)(&type), 4);
             __glXSendLargeCommand(gc, pc, 16, lists, (compsize * n));
@@ -2992,8 +2992,8 @@ __indirect_glPixelMapfv(GLenum map, GLsizei mapsize, const GLfloat * values)
             const GLint op = X_GLrop_PixelMapfv;
             const GLuint cmdlenLarge = cmdlen + 4;
             GLubyte * const pc = __glXFlushRenderBuffer(gc, gc->pc);
-            (void) memcpy((void *)(pc + 0), (void *)(&op), 4);
-            (void) memcpy((void *)(pc + 4), (void *)(&cmdlenLarge), 4);
+            (void) memcpy((void *)(pc + 0), (void *)(&cmdlenLarge), 4);
+            (void) memcpy((void *)(pc + 4), (void *)(&op), 4);
             (void) memcpy((void *)(pc + 8), (void *)(&map), 4);
             (void) memcpy((void *)(pc + 12), (void *)(&mapsize), 4);
             __glXSendLargeCommand(gc, pc, 16, values, (mapsize * 4));
@@ -3023,8 +3023,8 @@ __indirect_glPixelMapuiv(GLenum map, GLsizei mapsize, const GLuint * values)
             const GLint op = X_GLrop_PixelMapuiv;
             const GLuint cmdlenLarge = cmdlen + 4;
             GLubyte * const pc = __glXFlushRenderBuffer(gc, gc->pc);
-            (void) memcpy((void *)(pc + 0), (void *)(&op), 4);
-            (void) memcpy((void *)(pc + 4), (void *)(&cmdlenLarge), 4);
+            (void) memcpy((void *)(pc + 0), (void *)(&cmdlenLarge), 4);
+            (void) memcpy((void *)(pc + 4), (void *)(&op), 4);
             (void) memcpy((void *)(pc + 8), (void *)(&map), 4);
             (void) memcpy((void *)(pc + 12), (void *)(&mapsize), 4);
             __glXSendLargeCommand(gc, pc, 16, values, (mapsize * 4));
@@ -3054,8 +3054,8 @@ __indirect_glPixelMapusv(GLenum map, GLsizei mapsize, const GLushort * values)
             const GLint op = X_GLrop_PixelMapusv;
             const GLuint cmdlenLarge = cmdlen + 4;
             GLubyte * const pc = __glXFlushRenderBuffer(gc, gc->pc);
-            (void) memcpy((void *)(pc + 0), (void *)(&op), 4);
-            (void) memcpy((void *)(pc + 4), (void *)(&cmdlenLarge), 4);
+            (void) memcpy((void *)(pc + 0), (void *)(&cmdlenLarge), 4);
+            (void) memcpy((void *)(pc + 4), (void *)(&op), 4);
             (void) memcpy((void *)(pc + 8), (void *)(&map), 4);
             (void) memcpy((void *)(pc + 12), (void *)(&mapsize), 4);
             __glXSendLargeCommand(gc, pc, 16, values, (mapsize * 2));

src/mesa/Makefile

@@ -26,6 +26,8 @@ default:
 		$(MAKE) osmesa-only ; \
 	elif [ "$(DRIVER_DIRS)" = "beos" ]; then \
 		$(MAKE) beos ; \
+	elif [ "$(DRIVER_DIRS)" = "directfb" ]; then \
+		$(MAKE) directfb; \
 	else \
 		$(MAKE) stand-alone ; \
 	fi
@@ -69,6 +71,30 @@ drivers-dri:
 	cd drivers/dri ; $(MAKE)
 
 
+
+#####################################################################
+# Stand-alone Mesa libGL, no built-in drivers (DirectFB)
+
+LIBGL_CORE_SOURCES =	\
+	$(CORE_SOURCES) 	\
+	$(ASM_SOURCES)
+
+LIBGL_CORE_OBJECTS = 	\
+	$(CORE_OBJECTS) 	\
+	$(ASM_SOURCES:.S=.o)
+
+# Make libGL from core object files
+libgl-core: $(LIBGL_CORE_OBJECTS)
+	@ CC=$(CC) CXX=$(CXX) $(TOP)/bin/mklib -o $(GL_LIB) \
+		-major $(MESA_MAJOR) -minor $(MESA_MINOR) -patch $(MESA_TINY) \
+		-install $(LIB_DIR) $(MKLIB_OPTIONS) $(LIBGL_CORE_OBJECTS) $(GL_LIB_DEPS)
+
+# DirectFB driver target
+directfb: depend subdirs libgl-core
+	cd drivers/directfb; $(MAKE)
+
+
+
 ######################################################################
 # Stand-alone Mesa libGL and libOSMesa
 

src/mesa/drivers/directfb/Makefile

@@ -0,0 +1,53 @@
+# src/mesa/drivers/directfb/Makefile
+
+TOP = ../../../..
+
+include $(TOP)/configs/current
+
+
+INCLUDE_DIRS = \
+	-I$(TOP)/include \
+	-I$(TOP)/src/mesa \
+	-I$(TOP)/src/mesa/main \
+	-I$(TOP)/src/mesa/glapi \
+	-I$(TOP)/src/mesa/math \
+	-I$(TOP)/src/mesa/tnl \
+	-I$(TOP)/src/mesa/shader \
+	-I$(TOP)/src/mesa/swrast \
+	-I$(TOP)/src/mesa/swrast_setup
+
+DFB_CFLAGS    = $(shell pkg-config --cflags directfb)
+DFB_MODULEDIR = $(shell pkg-config --variable=moduledir directfb-internal)
+
+DIRECTFBGL_MESA_SOURCES = ../common/driverfuncs.c idirectfbgl_mesa.c
+
+DIRECTFBGL_MESA_OBJECTS = $(DIRECTFBGL_MESA_SOURCES:.c=.o)
+
+DIRECTFBGL_MESA = libidirectfbgl_mesa.so
+
+.c.o:
+	$(CC) -c $(INCLUDE_DIRS) $(CFLAGS) $(DFB_CFLAGS) $< -o $@
+
+
+default: directfbgl_mesa
+
+# Mesa DirectFBGL module
+directfbgl_mesa: $(DIRECTFBGL_MESA_OBJECTS)
+	$(CC) -shared $(CFLAGS) $(DIRECTFBGL_MESA_OBJECTS) -o $(DIRECTFBGL_MESA) \
+	-Wl,-soname -Wl,$(DIRECTFBGL_MESA) -L$(TOP)/lib -lGL -lm
+
+
+install:
+	@if test -d $(DFB_MODULEDIR); then \
+		echo "Installing DirectFBGL module."; \
+	else \
+		echo "*** Failed to determine DirectFB module's directory."; \
+		echo "*** Installation aborted."; \
+		exit 1; \
+	fi;
+	install -m 0755 $(DIRECTFBGL_MESA) $(DFB_MODULEDIR)/interfaces/IDirectFBGL/
+
+
+clean:
+	rm -f *.o *.so
+

src/mesa/drivers/directfb/idirectfbgl_mesa.c

@@ -0,0 +1,642 @@
+/*
+ * Copyright (C) 2004 Claudio Ciccani <klan@users.sf.net>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful,
+ * but WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
+ * GNU General Public License for more details.
+ *
+ * You should have received a copy of the GNU General Public License
+ * along with this program; if not, write to the Free Software
+ * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA  02111-1307  USA
+ *
+ * 
+ * Based on glfbdev.c, written by Brian Paul.
+ *
+ */
+
+#include <stdio.h>
+#include <stdlib.h>
+#include <unistd.h>
+
+#include <directfb.h>
+
+#include <direct/messages.h>
+#include <direct/interface.h>
+#include <direct/mem.h>
+
+#ifdef CLAMP
+# undef CLAMP
+#endif 
+
+#include "GL/directfbgl.h"
+#include "glheader.h"
+#include "buffers.h"
+#include "context.h"
+#include "extensions.h"
+#include "imports.h"
+#include "texformat.h"
+#include "teximage.h"
+#include "texstore.h"
+#include "array_cache/acache.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 "drivers/common/driverfuncs.h"
+
+
+static DFBResult
+Probe( void *data );
+
+static DFBResult
+Construct( IDirectFBGL      *thiz,
+           IDirectFBSurface *surface );
+
+#include <direct/interface_implementation.h>
+
+DIRECT_INTERFACE_IMPLEMENTATION( IDirectFBGL, Mesa )
+
+/*
+ * private data struct of IDirectFBGL
+ */
+typedef struct {
+     int                     ref;       /* reference counter */
+     
+     bool                    locked;
+     
+     IDirectFBSurface       *surface;
+     DFBSurfacePixelFormat   format;
+     int                     width;
+     int                     height;
+     
+     struct {
+          __u8              *start;
+          __u8              *end;
+          int                pitch;
+     } video;
+
+     GLvisual                visual;
+     GLframebuffer           framebuffer;
+     GLcontext               context;
+} IDirectFBGL_data;
+
+
+static bool  dfb_mesa_setup_visual   ( GLvisual              *visual,
+                                       DFBSurfacePixelFormat  format );
+static bool  dfb_mesa_create_context ( GLcontext             *context,
+                                       GLframebuffer         *framebuffer,
+                                       GLvisual              *visual,
+                                       DFBSurfacePixelFormat  format,
+                                       void                  *data );
+static void  dfb_mesa_destroy_context( GLcontext             *context,
+                                       GLframebuffer         *framebuffer );
+
+
+static void
+IDirectFBGL_Destruct( IDirectFBGL *thiz )
+{
+     IDirectFBGL_data *data = (IDirectFBGL_data*) thiz->priv;
+
+     dfb_mesa_destroy_context( &data->context, &data->framebuffer );
+     
+     data->surface->Release( data->surface );
+
+     DIRECT_DEALLOCATE_INTERFACE( thiz );
+}
+
+static DFBResult
+IDirectFBGL_AddRef( IDirectFBGL *thiz )
+{
+     DIRECT_INTERFACE_GET_DATA( IDirectFBGL );
+
+     data->ref++;
+
+     return DFB_OK;
+}
+
+static DFBResult
+IDirectFBGL_Release( IDirectFBGL *thiz )
+{
+     DIRECT_INTERFACE_GET_DATA( IDirectFBGL )
+
+     if (--data->ref == 0) {
+          IDirectFBGL_Destruct( thiz );
+     }
+
+     return DFB_OK;
+}
+
+static DFBResult
+IDirectFBGL_Lock( IDirectFBGL *thiz )
+{
+     IDirectFBSurface *surface;
+     int               width   = 0;
+     int               height  = 0;
+     DFBResult         err;
+     
+     DIRECT_INTERFACE_GET_DATA( IDirectFBGL );
+
+     if (data->locked)
+          return DFB_LOCKED;
+
+     surface = data->surface;
+     surface->GetSize( surface, &width, &height );
+     
+     err = surface->Lock( surface, DSLF_READ | DSLF_WRITE, 
+                          (void**) &data->video.start, &data->video.pitch );
+     if (err != DFB_OK) {
+          D_ERROR( "DirectFBGL/Mesa: couldn't lock surface.\n" );
+          return err;
+     }
+     data->video.end = data->video.start + (height-1) * data->video.pitch;
+     
+     if (data->width != width || data->height != height) {
+          data->width  = width;
+          data->height = height;
+          _mesa_ResizeBuffersMESA();
+     }
+     
+     data->locked = true;
+     
+     return DFB_OK;
+}
+
+static DFBResult
+IDirectFBGL_Unlock( IDirectFBGL *thiz )
+{     
+     DIRECT_INTERFACE_GET_DATA( IDirectFBGL );
+
+     if (!data->locked)
+          return DFB_OK;
+
+     data->surface->Unlock( data->surface );
+     data->video.start = NULL;
+     data->video.end   = NULL;
+    
+     data->locked = false;
+
+     return DFB_OK;
+}
+
+static DFBResult
+IDirectFBGL_GetAttributes( IDirectFBGL     *thiz,
+                           DFBGLAttributes *attributes )
+{
+     GLvisual *visual;
+     
+     DIRECT_INTERFACE_GET_DATA( IDirectFBGL );
+
+     if (!attributes)
+          return DFB_INVARG;
+
+     visual = &data->visual;
+     
+     attributes->buffer_size      = visual->rgbBits ? : visual->indexBits;
+     attributes->depth_size       = visual->depthBits;
+     attributes->stencil_size     = visual->stencilBits;
+     attributes->aux_buffers      = visual->numAuxBuffers;
+     attributes->red_size         = visual->redBits;
+     attributes->green_size       = visual->greenBits;
+     attributes->blue_size        = visual->blueBits;
+     attributes->alpha_size       = visual->alphaBits;
+     attributes->accum_red_size   = visual->accumRedBits;
+     attributes->accum_green_size = visual->accumGreenBits;
+     attributes->accum_blue_size  = visual->accumBlueBits;
+     attributes->accum_alpha_size = visual->accumAlphaBits;
+     attributes->double_buffer    = (visual->doubleBufferMode != 0);
+     attributes->stereo           = (visual->stereoMode != 0);
+
+     return DFB_OK;
+}
+
+
+/* exported symbols */
+
+static DFBResult
+Probe( void *data )
+{
+     return DFB_OK;
+}
+
+static DFBResult
+Construct( IDirectFBGL      *thiz,
+           IDirectFBSurface *surface )
+{ 
+     /* Allocate interface data. */
+     DIRECT_ALLOCATE_INTERFACE_DATA( thiz, IDirectFBGL );
+
+     /* Initialize interface data. */
+     data->ref     = 1;
+     data->surface = surface;
+
+     surface->AddRef( surface );
+     surface->GetPixelFormat( surface, &data->format );
+     surface->GetSize( surface, &data->width, &data->height );
+
+     /* Configure visual. */
+     if (!dfb_mesa_setup_visual( &data->visual, data->format )) {
+          D_ERROR( "DirectFBGL/Mesa: failed to initialize visual.\n" );
+          surface->Release( surface );
+          return DFB_UNSUPPORTED;
+     }
+     
+     /* Create context. */
+     if (!dfb_mesa_create_context( &data->context, &data->framebuffer,
+                                   &data->visual, data->format, (void*) data )) {
+          D_ERROR( "DirectFBGL/Mesa: failed to create context.\n" );
+          surface->Release( surface );
+          return DFB_UNSUPPORTED;
+     }
+
+     /* Assign interface pointers. */
+     thiz->AddRef        = IDirectFBGL_AddRef;
+     thiz->Release       = IDirectFBGL_Release;
+     thiz->Lock          = IDirectFBGL_Lock;
+     thiz->Unlock        = IDirectFBGL_Unlock;
+     thiz->GetAttributes = IDirectFBGL_GetAttributes;
+
+     return DFB_OK;
+}
+
+
+/* internal functions */
+
+static const GLubyte*
+get_string( GLcontext *ctx, GLenum pname )
+{
+     switch (pname) {
+          case GL_VENDOR:
+               return "Claudio Ciccani";
+          case GL_VERSION:
+               return "1.0";
+          default:
+               return NULL;
+     }
+}
+
+static void
+update_state( 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 );
+}
+
+static void
+get_buffer_size( GLframebuffer *buffer, GLuint *width, GLuint *height )
+{
+     GLcontext        *ctx  = _mesa_get_current_context();
+     IDirectFBGL_data *data = (IDirectFBGL_data*) ctx->DriverCtx;
+
+     *width  = (GLuint) data->width;
+     *height = (GLuint) data->height;
+}
+
+static void
+set_viewport( GLcontext *ctx, GLint x, GLint y, GLsizei w, GLsizei h )
+{
+     _mesa_ResizeBuffersMESA();
+}
+
+/* required but not used */
+static void
+set_buffer( GLcontext *ctx, GLframebuffer *buffer, GLuint bufferBit )
+{
+     return;
+}
+
+
+/* RGB332 */
+#define NAME(PREFIX) PREFIX##_RGB332
+#define SPAN_VARS \
+   IDirectFBGL_data *data = (IDirectFBGL_data*) ctx->DriverCtx;
+#define INIT_PIXEL_PTR(P, X, Y) \
+   GLubyte *P = data->video.end - (Y) * data->video.pitch + (X)
+#define INC_PIXEL_PTR(P) P += 1
+#define STORE_RGB_PIXEL(P, X, Y, R, G, B) \
+   *P = ( (((R) & 0xe0)) | (((G) & 0xe0) >> 3) | ((B) >> 6) )
+#define STORE_RGBA_PIXEL(P, X, Y, R, G, B, A) \
+   *P = ( (((R) & 0xe0)) | (((G) & 0xe0) >> 3) | ((B) >> 6) )
+#define FETCH_RGBA_PIXEL(R, G, B, A, P) \
+   R = ((*P & 0xe0)     ); \
+   G = ((*P & 0x1c) << 3); \
+   B = ((*P & 0x03) << 6); \
+   A = CHAN_MAX
+
+#include "swrast/s_spantemp.h"
+
+/* ARGB1555 */
+#define NAME(PREFIX) PREFIX##_ARGB1555
+#define SPAN_VARS \
+   IDirectFBGL_data *data = (IDirectFBGL_data*) ctx->DriverCtx;
+#define INIT_PIXEL_PTR(P, X, Y) \
+   GLushort *P = (GLushort *) (data->video.end - (Y) * data->video.pitch + (X) * 2)
+#define INC_PIXEL_PTR(P) P += 1
+#define STORE_RGB_PIXEL(P, X, Y, R, G, B) \
+   *P = ( (((R) & 0xf8) << 7) | (((G) & 0xf8) << 2) | ((B) >> 3) )
+#define STORE_RGBA_PIXEL(P, X, Y, R, G, B, A) \
+   *P = ( (((A) & 0x80) << 8) | (((R) & 0xf8) << 7) | (((G) & 0xf8) << 2) | ((B) >> 3) )
+#define FETCH_RGBA_PIXEL(R, G, B, A, P) \
+   R = ((*P & 0x7c00) >> 7); \
+   G = ((*P & 0x03e0) >> 2); \
+   B = ((*P & 0x001f) << 3); \
+   A = ((*P & 0x8000) ? 0xff : 0)
+
+#include "swrast/s_spantemp.h"
+
+/* RGB16 */
+#define NAME(PREFIX) PREFIX##_RGB16
+#define SPAN_VARS \
+   IDirectFBGL_data *data = (IDirectFBGL_data*) ctx->DriverCtx;
+#define INIT_PIXEL_PTR(P, X, Y) \
+   GLushort *P = (GLushort *) (data->video.end - (Y) * data->video.pitch + (X) * 2)
+#define INC_PIXEL_PTR(P) P += 1
+#define STORE_RGB_PIXEL(P, X, Y, R, G, B) \
+   *P = ( (((R) & 0xf8) << 8) | (((G) & 0xfc) << 3) | ((B) >> 3) )
+#define STORE_RGBA_PIXEL(P, X, Y, R, G, B, A) \
+   *P = ( (((R) & 0xf8) << 8) | (((G) & 0xfc) << 3) | ((B) >> 3) )
+#define FETCH_RGBA_PIXEL(R, G, B, A, P) \
+   R = ((*P & 0xf800) >> 8); \
+   G = ((*P & 0x07e0) >> 3); \
+   B = ((*P & 0x001f) << 3); \
+   A = CHAN_MAX
+
+#include "swrast/s_spantemp.h"
+
+/* RGB24 */
+#define NAME(PREFIX) PREFIX##_RGB24
+#define SPAN_VARS \
+   IDirectFBGL_data *data = ctx->DriverCtx;
+#define INIT_PIXEL_PTR(P, X, Y) \
+   GLubyte *P = data->video.end - (Y) * data->video.pitch + (X) * 3
+#define INC_PIXEL_PTR(P) P += 3
+#define STORE_RGB_PIXEL(P, X, Y, R, G, B) \
+   P[0] = B;  P[1] = G;  P[2] = R
+#define STORE_RGBA_PIXEL(P, X, Y, R, G, B, A) \
+   P[0] = B;  P[1] = G;  P[2] = R
+#define FETCH_RGBA_PIXEL(R, G, B, A, P) \
+   R = P[2];  G = P[1];  B = P[0];  A = CHAN_MAX
+
+#include "swrast/s_spantemp.h"
+
+/* RGB32 */
+#define NAME(PREFIX) PREFIX##_RGB32
+#define SPAN_VARS \
+   IDirectFBGL_data *data = (IDirectFBGL_data*) ctx->DriverCtx;
+#define INIT_PIXEL_PTR(P, X, Y) \
+   GLuint *P = (GLuint*) (data->video.end - (Y) * data->video.pitch + (X) * 4)
+#define INC_PIXEL_PTR(P) P += 1
+#define STORE_RGB_PIXEL(P, X, Y, R, G, B) \
+   *P = ( ((R) << 16) | ((G) << 8) | (B) )
+#define STORE_RGBA_PIXEL(P, X, Y, R, G, B, A) \
+   *P = ( ((R) << 16) | ((G) << 8) | (B) )
+#define FETCH_RGBA_PIXEL(R, G, B, A, P) \
+   R = ((*P & 0x00ff0000) >> 16); \
+   G = ((*P & 0x0000ff00) >>  8); \
+   B = ((*P & 0x000000ff)      ); \
+   A = CHAN_MAX
+
+#include "swrast/s_spantemp.h"
+   
+/* ARGB */
+#define NAME(PREFIX) PREFIX##_ARGB
+#define SPAN_VARS \
+   IDirectFBGL_data *data = (IDirectFBGL_data*) ctx->DriverCtx;
+#define INIT_PIXEL_PTR(P, X, Y) \
+   GLuint *P = (GLuint*) (data->video.end - (Y) * data->video.pitch + (X) * 4)
+#define INC_PIXEL_PTR(P) P += 1
+#define STORE_RGB_PIXEL(P, X, Y, R, G, B) \
+   *P = ( 0xff000000  | ((R) << 16) | ((G) << 8) | (B) )
+#define STORE_RGBA_PIXEL(P, X, Y, R, G, B, A) \
+   *P = ( ((A) << 24) | ((R) << 16) | ((G) << 8) | (B) )
+#define FETCH_RGBA_PIXEL(R, G, B, A, P) \
+   R = ((*P & 0x00ff0000) >> 16); \
+   G = ((*P & 0x0000ff00) >>  8); \
+   B = ((*P & 0x000000ff)      ); \
+   A = ((*P & 0xff000000) >> 24)
+
+#include "swrast/s_spantemp.h"
+
+/* AiRGB */
+#define NAME(PREFIX) PREFIX##_AiRGB
+#define SPAN_VARS \
+   IDirectFBGL_data *data = (IDirectFBGL_data*) ctx->DriverCtx;
+#define INIT_PIXEL_PTR(P, X, Y) \
+   GLuint *P = (GLuint*) (data->video.end - (Y) * data->video.pitch + (X) * 4)
+#define INC_PIXEL_PTR(P) P += 1
+#define STORE_RGB_PIXEL(P, X, Y, R, G, B) \
+   *P = ( ((R) << 16) | ((G) << 8) | (B) )
+#define STORE_RGBA_PIXEL(P, X, Y, R, G, B, A) \
+   *P = ( ((0xff - (A)) << 24) | ((R) << 16) | ((G) << 8) | (B) )
+#define FETCH_RGBA_PIXEL(R, G, B, A, P) \
+   R =         ((*P & 0x00ff0000) >> 16); \
+   G =         ((*P & 0x0000ff00) >>  8); \
+   B =         ((*P & 0x000000ff)      ); \
+   A = (0xff - ((*P & 0xff000000) >> 24))
+
+#include "swrast/s_spantemp.h"
+
+
+static bool
+dfb_mesa_setup_visual( GLvisual              *visual,
+                       DFBSurfacePixelFormat  format )
+{
+     GLboolean  rgbFlag        = GL_TRUE;
+     GLboolean  dbFlag         = GL_FALSE;
+     GLboolean  stereoFlag     = GL_FALSE;
+     GLint      redBits        = 0;
+     GLint      blueBits       = 0;
+     GLint      greenBits      = 0;
+     GLint      alphaBits      = 0;
+     GLint      indexBits      = 0;
+     GLint      depthBits      = 0;
+     GLint      stencilBits    = 0;
+     GLint      accumRedBits   = 0;
+     GLint      accumGreenBits = 0;
+     GLint      accumBlueBits  = 0;
+     GLint      accumAlphaBits = 0;
+     GLint      numSamples     = 0;
+
+     /* FIXME: LUT8 support. */
+     switch (format) {
+          case DSPF_RGB332:
+               redBits   = 3;
+               greenBits = 3;
+               blueBits  = 2;
+               break;
+          case DSPF_ARGB1555:
+               redBits   = 5;
+               greenBits = 5;
+               blueBits  = 5;
+               alphaBits = 1;
+               break;
+          case DSPF_RGB16:
+               redBits   = 5;
+               greenBits = 6;
+               blueBits  = 5;
+               break;
+          case DSPF_ARGB:
+          case DSPF_AiRGB:
+               alphaBits = 8;
+          case DSPF_RGB24:
+          case DSPF_RGB32:
+               redBits   = 8;
+               greenBits = 8;
+               blueBits  = 8;
+               break;
+          default:
+               D_WARN( "unsupported pixelformat" );
+               return false;
+     }
+
+     if (rgbFlag) {
+          accumRedBits   = redBits;
+          accumGreenBits = greenBits;
+          accumBlueBits  = blueBits;
+          accumAlphaBits = alphaBits;
+          depthBits      = redBits + greenBits + blueBits;
+          stencilBits    = alphaBits;
+     } else
+          depthBits      = 8;
+
+     return _mesa_initialize_visual( visual,
+                                     rgbFlag, dbFlag, stereoFlag,
+                                     redBits, greenBits, blueBits, alphaBits,
+                                     indexBits, depthBits, stencilBits,
+                                     accumRedBits, accumGreenBits,
+                                     accumBlueBits, accumAlphaBits,
+                                     numSamples );
+}
+
+static bool
+dfb_mesa_create_context( GLcontext             *context,
+                         GLframebuffer         *framebuffer,
+                         GLvisual              *visual,
+                         DFBSurfacePixelFormat  format,
+                         void                  *data )
+{
+     struct dd_function_table     functions;
+     struct swrast_device_driver *swdd;
+     
+     _mesa_initialize_framebuffer( framebuffer, visual, 
+                                   visual->haveDepthBuffer,
+                                   visual->haveStencilBuffer,
+                                   visual->haveAccumBuffer,
+                                   visual->accumAlphaBits > 0 );
+     
+     _mesa_init_driver_functions( &functions );
+     functions.GetString     = get_string;
+     functions.UpdateState   = update_state;
+     functions.GetBufferSize = get_buffer_size;
+     functions.Viewport      = set_viewport;
+     
+     if (!_mesa_initialize_context( context, visual, NULL,
+                                    &functions, data )) {
+          D_DEBUG( "DirectFBGL/Mesa: _mesa_initialize_context() failed.\n" );
+          _mesa_free_framebuffer_data( framebuffer );
+          return false;
+     }
+
+     _swrast_CreateContext( context );
+     _ac_CreateContext( context );
+     _tnl_CreateContext( context );
+     _swsetup_CreateContext( context );
+     _swsetup_Wakeup( context );
+     
+     swdd = _swrast_GetDeviceDriverReference( context );
+     swdd->SetBuffer = set_buffer;
+     switch (format) {
+          case DSPF_RGB332: 
+               swdd->WriteRGBASpan       = write_rgba_span_RGB332;
+               swdd->WriteRGBSpan        = write_rgb_span_RGB332;
+               swdd->WriteMonoRGBASpan   = write_monorgba_span_RGB332;
+               swdd->WriteRGBAPixels     = write_rgba_pixels_RGB332;
+               swdd->WriteMonoRGBAPixels = write_monorgba_pixels_RGB332;
+               swdd->ReadRGBASpan        = read_rgba_span_RGB332;
+               swdd->ReadRGBAPixels      = read_rgba_pixels_RGB332;
+               break;
+          case DSPF_ARGB1555:
+               swdd->WriteRGBASpan       = write_rgba_span_ARGB1555;
+               swdd->WriteRGBSpan        = write_rgb_span_ARGB1555;
+               swdd->WriteMonoRGBASpan   = write_monorgba_span_ARGB1555;
+               swdd->WriteRGBAPixels     = write_rgba_pixels_ARGB1555;
+               swdd->WriteMonoRGBAPixels = write_monorgba_pixels_ARGB1555;
+               swdd->ReadRGBASpan        = read_rgba_span_ARGB1555;
+               swdd->ReadRGBAPixels      = read_rgba_pixels_ARGB1555;
+               break;
+          case DSPF_RGB16:
+               swdd->WriteRGBASpan       = write_rgba_span_RGB16;
+               swdd->WriteRGBSpan        = write_rgb_span_RGB16;
+               swdd->WriteMonoRGBASpan   = write_monorgba_span_RGB16;
+               swdd->WriteRGBAPixels     = write_rgba_pixels_RGB16;
+               swdd->WriteMonoRGBAPixels = write_monorgba_pixels_RGB16;
+               swdd->ReadRGBASpan        = read_rgba_span_RGB16;
+               swdd->ReadRGBAPixels      = read_rgba_pixels_RGB16;
+               break;
+          case DSPF_RGB24:   
+               swdd->WriteRGBASpan       = write_rgba_span_RGB24;
+               swdd->WriteRGBSpan        = write_rgb_span_RGB24;
+               swdd->WriteMonoRGBASpan   = write_monorgba_span_RGB24;
+               swdd->WriteRGBAPixels     = write_rgba_pixels_RGB24;
+               swdd->WriteMonoRGBAPixels = write_monorgba_pixels_RGB24;
+               swdd->ReadRGBASpan        = read_rgba_span_RGB24;
+               swdd->ReadRGBAPixels      = read_rgba_pixels_RGB24;
+               break;
+          case DSPF_RGB32:
+               swdd->WriteRGBASpan       = write_rgba_span_RGB32;
+               swdd->WriteRGBSpan        = write_rgb_span_RGB32;
+               swdd->WriteMonoRGBASpan   = write_monorgba_span_RGB32;
+               swdd->WriteRGBAPixels     = write_rgba_pixels_RGB32;
+               swdd->WriteMonoRGBAPixels = write_monorgba_pixels_RGB32;
+               swdd->ReadRGBASpan        = read_rgba_span_RGB32;
+               swdd->ReadRGBAPixels      = read_rgba_pixels_RGB32;
+               break;
+          case DSPF_ARGB:
+               swdd->WriteRGBASpan       = write_rgba_span_ARGB;
+               swdd->WriteRGBSpan        = write_rgb_span_ARGB;
+               swdd->WriteMonoRGBASpan   = write_monorgba_span_ARGB;
+               swdd->WriteRGBAPixels     = write_rgba_pixels_ARGB;
+               swdd->WriteMonoRGBAPixels = write_monorgba_pixels_ARGB;
+               swdd->ReadRGBASpan        = read_rgba_span_ARGB;
+               swdd->ReadRGBAPixels      = read_rgba_pixels_ARGB;
+               break;
+          case DSPF_AiRGB:
+               swdd->WriteRGBASpan       = write_rgba_span_AiRGB;
+               swdd->WriteRGBSpan        = write_rgb_span_AiRGB;
+               swdd->WriteMonoRGBASpan   = write_monorgba_span_AiRGB;
+               swdd->WriteRGBAPixels     = write_rgba_pixels_AiRGB;
+               swdd->WriteMonoRGBAPixels = write_monorgba_pixels_AiRGB;
+               swdd->ReadRGBASpan        = read_rgba_span_AiRGB;
+               swdd->ReadRGBAPixels      = read_rgba_pixels_AiRGB;
+               break;
+          default:
+               D_BUG( "unexpected pixelformat" );
+               return false;
+     }
+
+     TNL_CONTEXT( context )->Driver.RunPipeline = _tnl_run_pipeline;
+
+     _mesa_enable_sw_extensions( context );
+
+     _mesa_make_current( context, framebuffer );
+     
+     return true;
+}
+
+static void
+dfb_mesa_destroy_context( GLcontext     *context,
+                          GLframebuffer *framebuffer )
+{
+     _mesa_make_current( NULL, NULL );
+     _mesa_free_framebuffer_data( framebuffer );
+     _mesa_notifyDestroy( context );
+     _mesa_free_context_data( context );
+}    
+ 

src/mesa/drivers/dri/Makefile.template

@@ -18,7 +18,7 @@ WINOBJ=
 WINLIB=-L$(MESA)/src/glx/mini
 MINIGLX_INCLUDES = -I$(TOP)/src/glx/mini
 INCLUDES = $(MINIGLX_INCLUDES) \
-	   -I$(DRM_SOURCE_PATH)/shared \
+	   -I$(DRM_SOURCE_PATH)/shared-core \
 	   -I$(DRM_SOURCE_PATH)/libdrm \
 	   $(SHARED_INCLUDES)
 
@@ -36,8 +36,7 @@ SHARED_INCLUDES = \
 	-I. \
 	-I$(TOP)/src/mesa/drivers/dri/common \
 	-Iserver \
-	-I$(DRM_SOURCE_PATH)/shared \
-	-I$(DRM_SOURCE_PATH)/linux \
+	-I$(DRM_SOURCE_PATH)/shared-core \
 	-I$(TOP)/include \
 	-I$(TOP)/include/GL/internal \
 	-I$(TOP)/src/mesa \
@@ -95,7 +94,7 @@ tags:
 
 # Remove .o and backup files
 clean:
-	-rm -f *.o */*.o *~ *.o *~ *.so server/*.o $(SYMLINKS)
+	-rm -f *.o */*.o *~ *.so *~ server/*.o $(SYMLINKS)
 	-rm -f depend depend.bak
 
 include depend

src/mesa/drivers/dri/common/xmlpool.h

@@ -210,6 +210,12 @@ DRI_CONF_OPT_BEGIN_V(dither_mode,enum,def,"0:2") \
         DRI_CONF_DESC_END \
 DRI_CONF_OPT_END
 
+#define DRI_CONF_FLOAT_DEPTH(def) \
+DRI_CONF_OPT_BEGIN(float_depth,bool,def) \
+        DRI_CONF_DESC(en,"Floating point depth buffer") \
+        DRI_CONF_DESC(de,"Fließkomma z-Puffer") \
+DRI_CONF_OPT_END
+
 /** \brief Performance-related options */
 #define DRI_CONF_SECTION_PERFORMANCE \
 DRI_CONF_SECTION_BEGIN \

src/mesa/drivers/dri/r300/Makefile

@@ -33,6 +33,9 @@ DRIVER_SOURCES = \
 		 r300_state.c \
 		 r300_render.c \
 		 r300_lib.c \
+		 r300_texmem.c \
+		 r300_tex.c \
+		 r300_texstate.c \
 		 \
 		 r200_context.c \
 		 r200_ioctl.c \
@@ -52,6 +55,7 @@ DRIVER_SOURCES = \
 		 r200_vtxfmt_sse.c \
 		 r200_vtxfmt_x86.c
 
+
 C_SOURCES = $(COMMON_SOURCES) $(DRIVER_SOURCES)
 
 X86_SOURCES = r200_vtxtmp_x86.S

src/mesa/drivers/dri/r300/r300_cmdbuf.c

@@ -46,8 +46,10 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "radeon_ioctl.h"
 #include "r300_context.h"
 #include "r300_ioctl.h"
+#include "radeon_reg.h"
 #include "r300_reg.h"
 #include "r300_cmdbuf.h"
+#include "r300_emit.h"
 
 
 // Set this to 1 for extremely verbose debugging of command buffers
@@ -212,6 +214,7 @@ void r300EmitState(r300ContextPtr r300)
 	r300->hw.all_dirty = GL_FALSE;
 }
 
+#if 0
 
 static __inline__ uint32_t cmducs(int reg, int count)
 {
@@ -236,6 +239,7 @@ static __inline__ uint32_t cmdvpu(int addr, int count)
 
 	return cmd.u;
 }
+#endif
 
 #define CHECK( NM, COUNT )				\
 static int check_##NM( r300ContextPtr r300, 		\
@@ -273,9 +277,12 @@ CHECK( vpu, vpucount(atom->cmd) ? (1 + vpucount(atom->cmd)*4) : 0 )
  */
 void r300InitCmdBuf(r300ContextPtr r300)
 {
-	int size;
+	int size, i, mtu;
 
 	r300->hw.max_state_size = 0;
+	
+	mtu = r300->radeon.glCtx->Const.MaxTextureUnits;
+	fprintf(stderr, "Using %d maximum texture units..\n", mtu);
 
 	/* Initialize state atoms */
 	ALLOC_STATE( vpt, always, R300_VPT_CMDSIZE, "vpt", 0 );
@@ -284,8 +291,8 @@ void r300InitCmdBuf(r300ContextPtr r300)
 		r300->hw.unk2080.cmd[0] = cmducs(0x2080, 1);
 	ALLOC_STATE( ovf, always, R300_OVF_CMDSIZE, "ovf", 0 );
 		r300->hw.ovf.cmd[R300_OVF_CMD_0] = cmducs(R300_VAP_OUTPUT_VTX_FMT_0, 2);
-	ALLOC_STATE( unk20B0, always, 3, "unk20B0", 0 );
-		r300->hw.unk20B0.cmd[0] = cmducs(0x20B0, 2);
+	ALLOC_STATE( vte, always, 3, "vte", 0 );
+		r300->hw.vte.cmd[0] = cmducs(R300_SE_VTE_CNTL, 2);
 	ALLOC_STATE( unk2134, always, 3, "unk2134", 0 );
 		r300->hw.unk2134.cmd[0] = cmducs(0x2134, 2);
 	ALLOC_STATE( unk2140, always, 2, "unk2140", 0 );
@@ -408,6 +415,29 @@ void r300InitCmdBuf(r300ContextPtr r300)
 	ALLOC_STATE( vps, vpu, R300_VPS_CMDSIZE, "vps", 0 );
 		r300->hw.vps.cmd[R300_VPS_CMD_0] = cmdvpu(R300_PVS_UPLOAD_POINTSIZE, 1);
 
+	/* Textures */
+	ALLOC_STATE( tex.filter, variable, mtu+1, "tex_filter", 0 );
+		r300->hw.tex.filter.cmd[R300_TEX_CMD_0] = cmducs(R300_TX_FILTER_0, 0);
+		
+	ALLOC_STATE( tex.unknown1, variable, mtu+1, "tex_unknown1", 0 );
+		r300->hw.tex.unknown1.cmd[R300_TEX_CMD_0] = cmducs(R300_TX_UNK1_0, 0);
+		
+	ALLOC_STATE( tex.size, variable, mtu+1, "tex_size", 0 );
+		r300->hw.tex.size.cmd[R300_TEX_CMD_0] = cmducs(R300_TX_SIZE_0, 0);
+		
+	ALLOC_STATE( tex.format, variable, mtu+1, "tex_format", 0 );
+		r300->hw.tex.format.cmd[R300_TEX_CMD_0] = cmducs(R300_TX_FORMAT_0, 0);
+		
+	ALLOC_STATE( tex.offset, variable, mtu+1, "tex_offset", 0 );
+		r300->hw.tex.offset.cmd[R300_TEX_CMD_0] = cmducs(R300_TX_OFFSET_0, 0);
+		
+	ALLOC_STATE( tex.unknown4, variable, mtu+1, "tex_unknown4", 0 );
+		r300->hw.tex.unknown4.cmd[R300_TEX_CMD_0] = cmducs(R300_TX_UNK4_0, 0);
+		
+	ALLOC_STATE( tex.unknown5, variable, mtu+1, "tex_unknown5", 0 );
+		r300->hw.tex.unknown5.cmd[R300_TEX_CMD_0] = cmducs(R300_TX_UNK5_0, 0);
+	
+
 	/* Setup the atom linked list */
 	make_empty_list(&r300->hw.atomlist);
 	r300->hw.atomlist.name = "atom-list";
@@ -415,7 +445,7 @@ void r300InitCmdBuf(r300ContextPtr r300)
 	insert_at_tail(&r300->hw.atomlist, &r300->hw.vpt);
 	insert_at_tail(&r300->hw.atomlist, &r300->hw.unk2080);
 	insert_at_tail(&r300->hw.atomlist, &r300->hw.ovf);
-	insert_at_tail(&r300->hw.atomlist, &r300->hw.unk20B0);
+	insert_at_tail(&r300->hw.atomlist, &r300->hw.vte);
 	insert_at_tail(&r300->hw.atomlist, &r300->hw.unk2134);
 	insert_at_tail(&r300->hw.atomlist, &r300->hw.unk2140);
 	insert_at_tail(&r300->hw.atomlist, &r300->hw.vir[0]);
@@ -477,6 +507,14 @@ void r300InitCmdBuf(r300ContextPtr r300)
 	insert_at_tail(&r300->hw.atomlist, &r300->hw.vpp);
 	insert_at_tail(&r300->hw.atomlist, &r300->hw.vps);
 
+	insert_at_tail(&r300->hw.atomlist, &r300->hw.tex.filter);	
+	insert_at_tail(&r300->hw.atomlist, &r300->hw.tex.unknown1);
+	insert_at_tail(&r300->hw.atomlist, &r300->hw.tex.size);
+	insert_at_tail(&r300->hw.atomlist, &r300->hw.tex.format);
+	insert_at_tail(&r300->hw.atomlist, &r300->hw.tex.offset);
+	insert_at_tail(&r300->hw.atomlist, &r300->hw.tex.unknown4);
+	insert_at_tail(&r300->hw.atomlist, &r300->hw.tex.unknown5);
+
 	r300->hw.is_dirty = GL_TRUE;
 	r300->hw.all_dirty = GL_TRUE;
 
@@ -511,3 +549,206 @@ void r300DestroyCmdBuf(r300ContextPtr r300)
 	}
 }
 
+void r300EmitBlit(r300ContextPtr 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)
+{
+	drm_radeon_cmd_header_t *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 =
+	    (drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa, 8 * sizeof(int),
+							__FUNCTION__);
+
+	cmd[0].header.cmd_type = RADEON_CMD_PACKET3;
+	cmd[1].i = R200_CP_CMD_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 << 8) |
+		    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 r300EmitWait(r300ContextPtr rmesa, GLuint flags)
+{
+	if (rmesa->radeon.dri.drmMinor >= 6) {
+		drm_radeon_cmd_header_t *cmd;
+
+		assert(!(flags & ~(RADEON_WAIT_2D | RADEON_WAIT_3D)));
+
+		cmd =
+		    (drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa,
+								1 * sizeof(int),
+								__FUNCTION__);
+		cmd[0].i = 0;
+		cmd[0].wait.cmd_type = RADEON_CMD_WAIT;
+		cmd[0].wait.flags = flags;
+	}
+}
+
+void r300EmitLOAD_VBPNTR(r300ContextPtr rmesa, int start)
+{
+int i, a, count;
+GLuint dw;
+LOCAL_VARS
+
+count=rmesa->state.aos_count;
+
+a=1+(count>>1)*3+(count & 1)*2;
+start_packet3(RADEON_CP_PACKET3_3D_LOAD_VBPNTR, a-1);
+e32(count);
+for(i=0;i+1<count;i+=2){
+	e32(  (rmesa->state.aos[i].element_size << 0) 
+	     |(rmesa->state.aos[i].stride << 8)
+	     |(rmesa->state.aos[i+1].element_size << 16)
+	     |(rmesa->state.aos[i+1].stride << 24)
+	    );
+	e32(rmesa->state.aos[i].offset+start*4*rmesa->state.aos[i].stride);
+	e32(rmesa->state.aos[i+1].offset+start*4*rmesa->state.aos[i+1].stride);
+	}
+if(count & 1){
+	e32(  (rmesa->state.aos[count-1].element_size << 0) 
+	     |(rmesa->state.aos[count-1].stride << 8)
+	    );
+	e32(rmesa->state.aos[count-1].offset+start*4*rmesa->state.aos[count-1].stride);	
+	}
+
+/* delay ? */
+#if 0
+e32(RADEON_CP_PACKET2);
+e32(RADEON_CP_PACKET2);
+#endif
+}
+
+void static inline upload_vertex_shader_fragment(PREFIX int dest, struct r300_vertex_shader_fragment *vsf)
+{
+int i;
+LOCAL_VARS
+
+if(vsf->length==0)return;
+
+if(vsf->length & 0x3){
+	fprintf(stderr,"VERTEX_SHADER_FRAGMENT must have length divisible by 4\n");
+	exit(-1);
+	}
+
+vsf_start_fragment(dest, vsf->length);
+for(i=0;i<vsf->length;i++)
+	e32(vsf->body.d[i]);
+
+}
+
+void r300EmitVertexShader(r300ContextPtr rmesa)
+{
+LOCAL_VARS
+
+upload_vertex_shader_fragment(PASS_PREFIX VSF_DEST_PROGRAM, &(rmesa->state.vertex_shader.program));
+
+upload_vertex_shader_fragment(PASS_PREFIX VSF_DEST_MATRIX0, &(rmesa->state.vertex_shader.matrix[0]));
+upload_vertex_shader_fragment(PASS_PREFIX VSF_DEST_MATRIX1, &(rmesa->state.vertex_shader.matrix[0]));
+upload_vertex_shader_fragment(PASS_PREFIX VSF_DEST_MATRIX2, &(rmesa->state.vertex_shader.matrix[0]));
+
+upload_vertex_shader_fragment(PASS_PREFIX VSF_DEST_VECTOR0, &(rmesa->state.vertex_shader.vector[0]));
+upload_vertex_shader_fragment(PASS_PREFIX VSF_DEST_VECTOR1, &(rmesa->state.vertex_shader.vector[1]));
+
+upload_vertex_shader_fragment(PASS_PREFIX VSF_DEST_UNKNOWN1, &(rmesa->state.vertex_shader.unknown1));
+upload_vertex_shader_fragment(PASS_PREFIX VSF_DEST_UNKNOWN2, &(rmesa->state.vertex_shader.unknown2));
+
+reg_start(R300_VAP_PVS_CNTL_1, 2);
+e32(      (rmesa->state.vertex_shader.program_start << R300_PVS_CNTL_1_PROGRAM_START_SHIFT)
+ 	| (rmesa->state.vertex_shader.unknown_ptr1 << R300_PVS_CNTL_1_UNKNOWN_SHIFT)
+ 	| (rmesa->state.vertex_shader.program_end << R300_PVS_CNTL_1_PROGRAM_END_SHIFT)
+    );
+e32(      (rmesa->state.vertex_shader.param_offset << R300_PVS_CNTL_2_PARAM_OFFSET_SHIFT)
+ 	| (rmesa->state.vertex_shader.param_count << R300_PVS_CNTL_2_PARAM_COUNT_SHIFT)
+    );
+e32( (rmesa->state.vertex_shader.unknown_ptr2 << R300_PVS_CNTL_3_PROGRAM_UNKNOWN_SHIFT)
+   | (rmesa->state.vertex_shader.unknown_ptr3 << 0));
+
+reg_start(R300_VAP_PVS_WAITIDLE,0);
+	e32(0x00000000);
+}
+
+void r300EmitPixelShader(r300ContextPtr rmesa)
+{
+int i,k;
+LOCAL_VARS
+
+if(rmesa->state.pixel_shader.program.tex.length>0){
+	reg_start(R300_PFS_TEXI_0, rmesa->state.pixel_shader.program.tex.length-1);
+	for(i=0;i<rmesa->state.pixel_shader.program.tex.length;i++)
+		e32(rmesa->state.pixel_shader.program.tex.inst[i]);
+	}
+
+if(rmesa->state.pixel_shader.program.alu.length>0){
+	#define OUTPUT_FIELD(reg, field)  \
+		reg_start(reg,rmesa->state.pixel_shader.program.alu.length-1); \
+		for(i=0;i<rmesa->state.pixel_shader.program.alu.length;i++) \
+			e32(rmesa->state.pixel_shader.program.alu.inst[i].field);
+	
+	OUTPUT_FIELD(R300_PFS_INSTR0_0, inst0);
+	OUTPUT_FIELD(R300_PFS_INSTR1_0, inst1);
+	OUTPUT_FIELD(R300_PFS_INSTR2_0, inst2);
+	OUTPUT_FIELD(R300_PFS_INSTR3_0, inst3);
+	#undef OUTPUT_FIELD
+	}
+
+reg_start(R300_PFS_NODE_0, 3);
+for(i=0;i<4;i++){
+	e32(  (rmesa->state.pixel_shader.program.node[i].alu_offset << R300_PFS_NODE_ALU_OFFSET_SHIFT)
+	    | (rmesa->state.pixel_shader.program.node[i].alu_end  << R300_PFS_NODE_ALU_END_SHIFT)
+	    | (rmesa->state.pixel_shader.program.node[i].tex_offset << R300_PFS_NODE_TEX_OFFSET_SHIFT)
+	    | (rmesa->state.pixel_shader.program.node[i].tex_end  << R300_PFS_NODE_TEX_END_SHIFT)
+	    | ( (i==3) ? R300_PFS_NODE_LAST_NODE : 0)
+	    );
+	}
+
+reg_start(R300_PFS_CNTL_0, 2);
+	/*  PFS_CNTL_0 */
+e32((rmesa->state.pixel_shader.program.active_nodes-1) | (rmesa->state.pixel_shader.program.first_node_has_tex<<3));
+	/* PFS_CNTL_1 */
+e32(rmesa->state.pixel_shader.program.temp_register_count);
+	/* PFS_CNTL_2 */
+e32( 	  (rmesa->state.pixel_shader.program.alu_offset << R300_PFS_CNTL_ALU_OFFSET_SHIFT)
+	| (rmesa->state.pixel_shader.program.alu_end << R300_PFS_CNTL_ALU_END_SHIFT)
+	| (rmesa->state.pixel_shader.program.tex_offset << R300_PFS_CNTL_TEX_OFFSET_SHIFT)
+	| (rmesa->state.pixel_shader.program.tex_end << R300_PFS_CNTL_TEX_END_SHIFT) 
+   );
+	
+if(rmesa->state.pixel_shader.param_length>0){
+	reg_start(R300_PFS_PARAM_0_X, rmesa->state.pixel_shader.param_length*4-1);
+	for(i=0;i<rmesa->state.pixel_shader.param_length;i++){
+		efloat(rmesa->state.pixel_shader.param[i].x);
+		efloat(rmesa->state.pixel_shader.param[i].y);
+		efloat(rmesa->state.pixel_shader.param[i].z);
+		efloat(rmesa->state.pixel_shader.param[i].w);
+		}
+	}
+	
+}

src/mesa/drivers/dri/r300/r300_cmdbuf.h

@@ -98,5 +98,18 @@ static __inline__ uint32_t* r300AllocCmdBuf(r300ContextPtr r300,
 	return ptr;
 }
 
+extern void r300EmitBlit(r300ContextPtr 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 r300EmitWait(r300ContextPtr rmesa, GLuint flags);
+extern void r300EmitLOAD_VBPNTR(r300ContextPtr rmesa, int start);
+extern void r300EmitVertexShader(r300ContextPtr rmesa);
+extern void r300EmitPixelShader(r300ContextPtr rmesa);
 
 #endif /* __R300_CMDBUF_H__ */

src/mesa/drivers/dri/r300/r300_context.c

@@ -57,6 +57,7 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r300_cmdbuf.h"
 #include "r300_state.h"
 #include "r300_ioctl.h"
+#include "r300_tex.h"
 
 #include "vblank.h"
 #include "utils.h"
@@ -147,7 +148,7 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual,
 	struct dd_function_table functions;
 	r300ContextPtr r300;
 	GLcontext *ctx;
-	int tcl_mode;
+	int tcl_mode, i;
 
 	assert(glVisual);
 	assert(driContextPriv);
@@ -171,7 +172,7 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual,
 	_mesa_init_driver_functions(&functions);
 	r300InitIoctlFuncs(&functions);
 	r300InitStateFuncs(&functions);
-	//r200InitTextureFuncs(&functions);
+	r300InitTextureFuncs(&functions);
 
 	if (!radeonInitContext(&r300->radeon, &functions,
 			       glVisual, driContextPriv, sharedContextPrivate)) {
@@ -180,6 +181,35 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual,
 	}
 
 	/* Init r300 context data */
+	r300->dma.buf0_address = r300->radeon.radeonScreen->buffers->list[0].address;
+
+	(void)memset(r300->texture_heaps, 0, sizeof(r300->texture_heaps));
+	make_empty_list(&r300->swapped);
+
+	r300->nr_heaps = 1 /* screen->numTexHeaps */ ;
+	assert(r300->nr_heaps < R200_NR_TEX_HEAPS);
+	for (i = 0; i < r300->nr_heaps; i++) {
+		r300->texture_heaps[i] = driCreateTextureHeap(i, r300,
+							       screen->
+							       texSize[i], 12,
+							       RADEON_NR_TEX_REGIONS,
+							       (drmTextureRegionPtr)
+							       r300->radeon.sarea->
+							       tex_list[i],
+							       &r300->radeon.sarea->
+							       tex_age[i],
+							       &r300->swapped,
+							       sizeof
+							       (r300TexObj),
+							       (destroy_texture_object_t
+								*)
+							       r300DestroyTexObj);
+	}
+	r300->texture_depth = driQueryOptioni(&r300->radeon.optionCache,
+					       "texture_depth");
+	if (r300->texture_depth == DRI_CONF_TEXTURE_DEPTH_FB)
+		r300->texture_depth = (screen->cpp == 4) ?
+		    DRI_CONF_TEXTURE_DEPTH_32 : DRI_CONF_TEXTURE_DEPTH_16;
 
 	/* 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
@@ -239,8 +269,8 @@ GLboolean r300CreateContext(const __GLcontextModes * glVisual,
 #if 0
 	/* plug in a few more device driver functions */
 	/* XXX these should really go right after _mesa_init_driver_functions() */
-	r200InitPixelFuncs(ctx);
-	r200InitSwtcl(ctx);
+	r300InitPixelFuncs(ctx);
+	r300InitSwtcl(ctx);
 #endif
 	TNL_CONTEXT(ctx)->Driver.RunPipeline = _tnl_run_pipeline;
 

src/mesa/drivers/dri/r300/r300_context.h

@@ -54,9 +54,16 @@ typedef struct r300_context *r300ContextPtr;
 #include "radeon_lock.h"
 #include "mm.h"
 
+
 typedef GLuint uint32_t;
 typedef GLubyte uint8_t;
 
+  /* We should probably change types within vertex_shader 
+      and pixel_shader structure later on */
+#define CARD32 GLuint
+#include "vertex_shader.h"
+#include "pixel_shader.h"
+#undef CARD32
 
 static __inline__ uint32_t r300PackFloat32(float fl)
 {
@@ -66,6 +73,114 @@ static __inline__ uint32_t r300PackFloat32(float fl)
 	return u.u;
 }
 
+
+/************ DMA BUFFERS **************/
+
+/* Need refcounting on dma buffers:
+ */
+struct r300_dma_buffer {
+	int refcount;		/* the number of retained regions in buf */
+	drmBufPtr buf;
+};
+
+#define GET_START(rvb) (rmesa->radeon.radeonScreen->gart_buffer_offset +		\
+			(rvb)->address - rmesa->dma.buf0_address +	\
+			(rvb)->start)
+
+/* A retained region, eg vertices for indexed vertices.
+ */
+struct r300_dma_region {
+	struct r300_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 r300_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 r300_dma_region current;
+
+	void (*flush) (r300ContextPtr);
+
+	char *buf0_address;	/* start of buf[0], for index calcs */
+	GLuint nr_released_bufs;	/* flush after so many buffers released */
+};
+
+       /* Texture related */
+
+#define TEX_0   0x1
+#define TEX_1   0x2
+#define TEX_2	0x4
+#define TEX_3	0x8
+#define TEX_4	0x10
+#define TEX_5	0x20
+#define TEX_6	0x40
+#define TEX_7	0x80
+#define TEX_ALL 0xff
+
+typedef struct r300_tex_obj r300TexObj, *r300TexObjPtr;
+
+/* Texture object in locally shared texture space.
+ */
+struct r300_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. */
+
+	drm_radeon_tex_image_t image[6][RADEON_MAX_TEXTURE_LEVELS];
+	/* Six, for the cube faces */
+
+
+	/* hardware register values */
+	/* Note that R200 has 8 registers per texture and R300 only 7 */
+	GLuint filter;	
+	GLuint pitch; /* one of the unknown registers.. unknown 1 ?*/
+	GLuint size;	/* npot only */
+	GLuint format;
+	GLuint offset;	/* Image location in texmem.
+				   All cube faces follow. */
+	GLuint unknown4;
+	GLuint unknown5; 
+	/* end hardware registers */
+	
+	/* registers computed by r200 code - keep them here to 
+	   compare against what is actually written.
+	   
+	   to be removed later.. */
+	GLuint pp_border_color;
+	GLuint pp_cubic_faces;	/* cube face 1,2,3,4 log2 sizes */
+	GLuint format_x;
+	
+	
+	GLboolean border_fallback;
+};
+
+struct r300_texture_env_state {
+	r300TexObjPtr texobj;
+	GLenum format;
+	GLenum envMode;
+};
+
+#define R300_MAX_TEXTURE_UNITS 8
+
+struct r300_texture_state {
+	struct r300_texture_env_state unit[R300_MAX_TEXTURE_UNITS];
+	int tc_count; /* number of incoming texture coordinates from VAP */
+};
+
 /**
  * A block of hardware state.
  *
@@ -233,6 +348,13 @@ struct r300_state_atom {
 #define R300_VPS_ZERO_3		4
 #define R300_VPS_CMDSIZE	5
 
+	/* the layout is common for all fields inside tex */
+#define R300_TEX_CMD_0		0
+#define R300_TEX_VALUE_0	1
+/* We don't really use this, instead specify mtu+1 dynamically 
+#define R300_TEX_CMDSIZE	(MAX_TEXTURE_UNITS+1)
+*/
+
 /**
  * Cache for hardware register state.
  */
@@ -246,7 +368,7 @@ struct r300_hw_state {
 	struct r300_state_atom vpt;	/* viewport (1D98) */
 	struct r300_state_atom unk2080;	/* (2080) */
 	struct r300_state_atom ovf;	/* output vertex format (2090) */
-	struct r300_state_atom unk20B0;	/* (20B0) */
+	struct r300_state_atom vte;	/* (20B0) */
 	struct r300_state_atom unk2134;	/* (2134) */
 	struct r300_state_atom unk2140;	/* (2140) */
 	struct r300_state_atom vir[2];	/* vap input route (2150/21E0) */
@@ -259,7 +381,6 @@ struct r300_hw_state {
 	struct r300_state_atom vof;     /* VAP output format register 0x4000 */
 	struct r300_state_atom gb_enable; /* (4008) */
 	struct r300_state_atom gb_misc; /* Multisampling position shifts ? (4010) */
-	struct r300_state_atom txe;	/* tex enable (4104) */
 	struct r300_state_atom unk4200; /* (4200) */
 	struct r300_state_atom unk4214; /* (4214) */
 	struct r300_state_atom ps;	/* pointsize (421C) */
@@ -303,6 +424,22 @@ struct r300_hw_state {
 	struct r300_state_atom vpi;	/* vp instructions */
 	struct r300_state_atom vpp;	/* vp parameters */
 	struct r300_state_atom vps;	/* vertex point size (?) */
+
+		/* 8 texture units */
+		/* the state is grouped by function and not by 
+		   texture unit. This makes single unit updates
+		   really awkward - we are much better off 
+		   updating the whole thing at once */
+	struct {
+		struct r300_state_atom filter;
+		struct r300_state_atom unknown1;
+		struct r300_state_atom size;
+		struct r300_state_atom format;
+		struct r300_state_atom offset;
+		struct r300_state_atom unknown4;
+		struct r300_state_atom unknown5;		
+		} tex;
+	struct r300_state_atom txe;	/* tex enable (4104) */
 };
 
 
@@ -329,8 +466,151 @@ struct r300_depthbuffer_state {
 	GLfloat scale;
 };
 
+struct r300_vap_reg_state {
+	   /* input register assigments */
+	   int i_coords;
+	   int i_color[2];
+	   int i_tex[R300_MAX_TEXTURE_UNITS];
+	};
+
+/* Vertex shader state */
+
+/* 64 appears to be the maximum */
+#define VSF_MAX_FRAGMENT_LENGTH 64
+
+
+struct r300_vertex_shader_fragment {
+	int length;
+	union {
+		GLuint d[VSF_MAX_FRAGMENT_LENGTH];  
+		float f[VSF_MAX_FRAGMENT_LENGTH];
+		VERTEX_SHADER_INSTRUCTION i[VSF_MAX_FRAGMENT_LENGTH/4];
+		} body;
+	};
+
+#define VSF_DEST_PROGRAM	0x0
+#define VSF_DEST_MATRIX0	0x200
+#define VSF_DEST_MATRIX1	0x204
+#define VSF_DEST_MATRIX2	0x208
+#define VSF_DEST_VECTOR0	0x20c
+#define VSF_DEST_VECTOR1	0x20d
+#define VSF_DEST_UNKNOWN1	0x400
+#define VSF_DEST_UNKNOWN2	0x406
+
+struct r300_vertex_shader_state {
+	struct r300_vertex_shader_fragment program;
+
+	/* a bit of a waste - each uses only a subset of allocated space..
+	    but easier to program */
+	struct r300_vertex_shader_fragment matrix[3];
+	struct r300_vertex_shader_fragment vector[2];
+		
+	struct r300_vertex_shader_fragment unknown1;
+	struct r300_vertex_shader_fragment unknown2;
+		
+	int program_start;
+	int unknown_ptr1;  /* pointer within program space */
+	int program_end;
+		
+	int param_offset;
+	int param_count;
+		
+	int unknown_ptr2;  /* pointer within program space */
+	int unknown_ptr3;  /* pointer within program space */
+	};
+
+/* 64 appears to be the maximum */
+#define PSF_MAX_PROGRAM_LENGTH 64
+
+struct r300_pixel_shader_program {
+	struct {
+		int length;
+		GLuint inst[PSF_MAX_PROGRAM_LENGTH];
+		} tex;
+	
+	/* ALU intructions (logic and integer) */
+	struct {
+		int length;
+		struct {
+			GLuint inst0;
+			GLuint inst1;
+			GLuint inst2;
+			GLuint inst3;
+			} inst[PSF_MAX_PROGRAM_LENGTH];
+		} alu;
+	
+	/* node information */
+	/* nodes are used to synchronize ALU and TEX streams */
+	/* There could be up to 4 nodes each consisting of
+	   a number of TEX instructions followed by some ALU
+	   instructions */
+	/* the last node of a program should always be node3 */
+	struct {
+		int tex_offset;
+		int tex_end;		
+		int alu_offset;
+		int alu_end;
+		} node[4];
+		
+	int active_nodes;	/* must be between 1 and 4, inclusive */
+	int first_node_has_tex;  /* other nodes always have it */
+	
+	int temp_register_count;  /* magic value goes into PFS_CNTL_1 */
+	
+	/* entire program */
+	int tex_offset;
+	int tex_end;
+	int alu_offset;
+	int alu_end;
+	
+	};
+	
+	#define MAX_PIXEL_SHADER_PARAMS 32
+struct r300_pixel_shader_state {
+	struct r300_pixel_shader_program program;
+	
+	/* parameters */		
+	int param_length;  /* to limit the number of unnecessary writes */
+	struct {
+		float x;
+		float y;
+		float z;
+		float w;
+		} param[MAX_PIXEL_SHADER_PARAMS];
+	};
+	
+/* 8 is somewhat bogus... it is probably something like 24 */
+#define R300_MAX_AOS_ARRAYS		8
+
+struct r300_aos_rec {
+	GLuint offset;
+	int element_size; /* in dwords */
+	int stride;       /* distance between elements, in dwords */
+	
+	#define AOS_FORMAT_FLOAT	1
+	#define AOS_FORMAT_UBYTE	2
+	#define AOS_FORMAT_FLOAT_COLOR	3
+	int format;	
+	
+	int ncomponents; /* number of components - between 1 and 4, inclusive */
+
+ 	 /* just guesses */
+	#define REG_COORDS	0
+	#define REG_COLOR0	1
+	#define REG_TEX0	2
+	int reg; /* which register they are assigned to. */
+			
+	};
+
 struct r300_state {
 	struct r300_depthbuffer_state depth;
+	struct r300_texture_state texture;
+	struct r300_vap_reg_state vap_reg;
+	struct r300_vertex_shader_state vertex_shader;
+	struct r300_pixel_shader_state pixel_shader;
+	struct r300_aos_rec aos[R300_MAX_AOS_ARRAYS];
+	int aos_count;
+	
 };
 
 
@@ -348,6 +628,38 @@ struct r300_context {
 	int elt_count;  /* size of the buffer for vertices */
 	int attrib_count; /* size of the buffer for vertex attributes.. Somehow it can be different ? */
 	
+
+	/* Vertex buffers
+	 */
+	#if 0 /* we'll need it later, but not now */
+	struct r300_ioctl ioctl;
+	#endif
+	struct r300_dma dma;
+	GLboolean save_on_next_unlock;
+
+	/* Texture object bookkeeping
+	 */
+	unsigned nr_heaps;
+	driTexHeap *texture_heaps[R200_NR_TEX_HEAPS];
+	driTextureObject swapped;
+	int texture_depth;
+	float initialMaxAnisotropy;
+
+	/* Clientdata textures;
+	 */
+	GLuint prefer_gart_client_texturing;
+
+	/* TCL stuff
+	 */
+	GLmatrix TexGenMatrix[R300_MAX_TEXTURE_UNITS];
+	GLboolean recheck_texgen[R300_MAX_TEXTURE_UNITS];
+	GLboolean TexGenNeedNormals[R300_MAX_TEXTURE_UNITS];
+	GLuint TexMatEnabled;
+	GLuint TexMatCompSel;
+	GLuint TexGenEnabled;
+	GLuint TexGenInputs;
+	GLuint TexGenCompSel;
+	GLmatrix tmpmat;
 };
 
 #define R300_CONTEXT(ctx)		((r300ContextPtr)(ctx->DriverCtx))

src/mesa/drivers/dri/r300/r300_emit.h

@@ -96,6 +96,16 @@ static __inline__ uint32_t cmdpacket3(int packet)
 	return cmd.u;
 }
 
+static __inline__ uint32_t cmdcpdelay(unsigned short count)
+{
+	drm_r300_cmd_header_t cmd;
+
+	cmd.delay.cmd_type = R300_CMD_CP_DELAY;
+	cmd.delay.count = count;
+
+	return cmd.u;
+}
+
 /* Prepare to write a register value to register at address reg.
    If num_extra > 0 then the following extra values are written
    to registers with address +4, +8 and so on.. */
@@ -181,4 +191,15 @@ cmd=(drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa, \
 cmd[0].header.cmd_type=R300_CMD_END3D;
 }
 
+void static inline cp_delay(PREFIX unsigned short count)
+{
+LOCAL_VARS
+
+cmd=(drm_radeon_cmd_header_t *) r300AllocCmdBuf(rmesa, \
+					0, \
+					__FUNCTION__); \
+
+cmd[0].i=cmdcpdelay(count);
+}
+
 #endif

src/mesa/drivers/dri/r300/r300_ioctl.c

@@ -81,6 +81,26 @@ static void r300ClearBuffer(r300ContextPtr r300, int flags, int buffer)
 
 	cboffset += r300->radeon.radeonScreen->fbLocation;
 
+	R300_STATECHANGE(r300, vir[0]);
+	((drm_r300_cmd_header_t*)r300->hw.vir[0].cmd)->unchecked_state.count = 1;
+	r300->hw.vir[0].cmd[1] = 0x21030003;
+
+	R300_STATECHANGE(r300, vir[1]);
+	((drm_r300_cmd_header_t*)r300->hw.vir[1].cmd)->unchecked_state.count = 1;
+	r300->hw.vir[1].cmd[1] = 0xF688F688;
+
+	R300_STATECHANGE(r300, vic);
+	r300->hw.vic.cmd[R300_VIR_CNTL_0] = 0x00000001;
+	r300->hw.vic.cmd[R300_VIR_CNTL_1] = 0x00000405;
+	
+	R300_STATECHANGE(r300, vof);
+	r300->hw.vof.cmd[R300_VOF_CNTL_0] = R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT
+				| R300_VAP_OUTPUT_VTX_FMT_0__COLOR_PRESENT;
+	r300->hw.vof.cmd[R300_VOF_CNTL_1] = 0; /* no textures */
+	
+	R300_STATECHANGE(r300, txe);
+	r300->hw.txe.cmd[R300_TXE_ENABLE] = 0;
+	
 	R300_STATECHANGE(r300, vpt);
 	r300->hw.vpt.cmd[R300_VPT_XSCALE] = r300PackFloat32(1.0);
 	r300->hw.vpt.cmd[R300_VPT_XOFFSET] = r300PackFloat32(dPriv->x);
@@ -89,6 +109,13 @@ static void r300ClearBuffer(r300ContextPtr r300, int flags, int buffer)
 	r300->hw.vpt.cmd[R300_VPT_ZSCALE] = r300PackFloat32(1.0);
 	r300->hw.vpt.cmd[R300_VPT_ZOFFSET] = r300PackFloat32(0.0);
 
+	R300_STATECHANGE(r300, at);
+	r300->hw.at.cmd[R300_AT_ALPHA_TEST] = 0;
+	
+	R300_STATECHANGE(r300, bld);
+	r300->hw.bld.cmd[R300_BLD_CBLEND] = 0;
+	r300->hw.bld.cmd[R300_BLD_ABLEND] = 0;
+	
 	R300_STATECHANGE(r300, cb);
 	r300->hw.cb.cmd[R300_CB_OFFSET] = cboffset;
 	r300->hw.cb.cmd[R300_CB_PITCH] = cbpitch | R300_COLOR_UNKNOWN_22_23;
@@ -105,6 +132,11 @@ static void r300ClearBuffer(r300ContextPtr r300, int flags, int buffer)
 	for(i = 1; i <= 8; ++i)
 		r300->hw.ri.cmd[i] = R300_RS_INTERP_USED;
 
+	R300_STATECHANGE(r300, rc);
+	/* The second constant is needed to get glxgears display anything .. */
+	r300->hw.rc.cmd[1] = R300_RS_CNTL_0_UNKNOWN_7 | R300_RS_CNTL_0_UNKNOWN_18;
+	r300->hw.rc.cmd[2] = 0;
+	
 	R300_STATECHANGE(r300, rr);
 	((drm_r300_cmd_header_t*)r300->hw.rr.cmd)->unchecked_state.count = 1;
 	r300->hw.rr.cmd[1] = 0x00004000;
@@ -282,6 +314,195 @@ void r300Flush(GLcontext * ctx)
 		r300FlushCmdBuf(r300, __FUNCTION__);
 }
 
+void r300RefillCurrentDmaRegion(r300ContextPtr rmesa)
+{
+	struct r200_dma_buffer *dmabuf;
+	int fd = rmesa->radeon.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)
+		r300ReleaseDmaRegion(rmesa, &rmesa->dma.current, __FUNCTION__);
+
+	if (rmesa->dma.nr_released_bufs > 4)
+		r300FlushCmdBuf(rmesa, __FUNCTION__);
+
+	dma.context = rmesa->radeon.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->radeon);	/* no need to validate */
+
+	while (1) {
+		ret = drmDMA(fd, &dma);
+		if (ret == 0)
+			break;
+
+		if (rmesa->dma.nr_released_bufs) {
+			r200FlushCmdBufLocked(rmesa, __FUNCTION__);
+		}
+
+		if (rmesa->radeon.do_usleeps) {
+			UNLOCK_HARDWARE(&rmesa->radeon);
+			DO_USLEEP(1);
+			LOCK_HARDWARE(&rmesa->radeon);
+		}
+	}
+
+	UNLOCK_HARDWARE(&rmesa->radeon);
+
+	if (RADEON_DEBUG & DEBUG_DMA)
+		fprintf(stderr, "Allocated buffer %d\n", index);
+
+	dmabuf = CALLOC_STRUCT(r300_dma_buffer);
+	dmabuf->buf = &rmesa->radeon.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;
+}
+
+void r300ReleaseDmaRegion(r300ContextPtr rmesa,
+			  struct r300_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) {
+		drm_radeon_cmd_header_t *cmd;
+
+		if (RADEON_DEBUG & (DEBUG_IOCTL | DEBUG_DMA))
+			fprintf(stderr, "%s -- DISCARD BUF %d\n", __FUNCTION__,
+				region->buf->buf->idx);
+
+		cmd =
+		    (drm_radeon_cmd_header_t *) r300AllocCmdBuf(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 r300AllocDmaRegion(r300ContextPtr rmesa,
+			struct r300_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)
+		r300ReleaseDmaRegion(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)
+		r300RefillCurrentDmaRegion(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;
+
+	assert(rmesa->dma.current.ptr <= rmesa->dma.current.end);
+}
+
+/* Called via glXGetMemoryOffsetMESA() */
+GLuint r300GetMemoryOffsetMESA(__DRInativeDisplay * dpy, int scrn,
+			       const GLvoid * pointer)
+{
+	GET_CURRENT_CONTEXT(ctx);
+	r300ContextPtr rmesa;
+	GLuint card_offset;
+
+	if (!ctx || !(rmesa = R300_CONTEXT(ctx))) {
+		fprintf(stderr, "%s: no context\n", __FUNCTION__);
+		return ~0;
+	}
+
+	if (!r300IsGartMemory(rmesa, pointer, 0))
+		return ~0;
+
+	if (rmesa->radeon.dri.drmMinor < 6)
+		return ~0;
+
+	card_offset = r300GartOffsetFromVirtual(rmesa, pointer);
+
+	return card_offset - rmesa->radeon.radeonScreen->gart_base;
+}
+
+GLboolean r300IsGartMemory(r300ContextPtr rmesa, const GLvoid * pointer,
+			   GLint size)
+{
+	int offset =
+	    (char *)pointer - (char *)rmesa->radeon.radeonScreen->gartTextures.map;
+	int valid = (size >= 0 && offset >= 0
+		     && offset + size < rmesa->radeon.radeonScreen->gartTextures.size);
+
+	if (RADEON_DEBUG & DEBUG_IOCTL)
+		fprintf(stderr, "r300IsGartMemory( %p ) : %d\n", pointer,
+			valid);
+
+	return valid;
+}
+
+GLuint r300GartOffsetFromVirtual(r300ContextPtr rmesa, const GLvoid * pointer)
+{
+	int offset =
+	    (char *)pointer - (char *)rmesa->radeon.radeonScreen->gartTextures.map;
+
+	fprintf(stderr, "offset=%08x\n", offset);
+
+	if (offset < 0 || offset > rmesa->radeon.radeonScreen->gartTextures.size)
+		return ~0;
+	else
+		return rmesa->radeon.radeonScreen->gart_texture_offset + offset;
+}
+
 void r300InitIoctlFuncs(struct dd_function_table *functions)
 {
 	functions->Clear = r300Clear;

src/mesa/drivers/dri/r300/r300_ioctl.h

@@ -36,7 +36,27 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #ifndef __R300_IOCTL_H__
 #define __R300_IOCTL_H__
 
+#include "r300_context.h"
+#include "radeon_drm.h"
+
+extern GLuint r300GetMemoryOffsetMESA(__DRInativeDisplay * dpy, int scrn,
+				      const GLvoid * pointer);
+
+extern GLboolean r300IsGartMemory(r300ContextPtr rmesa, const GLvoid * pointer,
+				  GLint size);
+
+extern GLuint r300GartOffsetFromVirtual(r300ContextPtr rmesa,
+					const GLvoid * pointer);
+
 extern void r300Flush(GLcontext * ctx);
+
+extern void r300RefillCurrentDmaRegion(r300ContextPtr rmesa);
+extern void r300ReleaseDmaRegion(r300ContextPtr rmesa,
+			  struct r300_dma_region *region, const char *caller);
+extern void r300AllocDmaRegion(r300ContextPtr rmesa,
+			struct r300_dma_region *region,
+			int bytes, int alignment);
+
 extern void r300InitIoctlFuncs(struct dd_function_table *functions);
 
 #endif				/* __R300_IOCTL_H__ */

src/mesa/drivers/dri/r300/r300_reg.h

@@ -70,6 +70,19 @@ I am fairly certain that they are correct unless stated otherwise in comments.
 #       define R300_VAP_OUTPUT_VTX_FMT_1__TEX_7_COMP_CNT_SHIFT 21
 /* END */
 
+#define R300_SE_VTE_CNTL                  0x20b0
+#	define     R300_VPORT_X_SCALE_ENA                0x00000001
+#	define     R300_VPORT_X_OFFSET_ENA               0x00000002
+#	define     R300_VPORT_Y_SCALE_ENA                0x00000004
+#	define     R300_VPORT_Y_OFFSET_ENA               0x00000008
+#	define     R300_VPORT_Z_SCALE_ENA                0x00000010
+#	define     R300_VPORT_Z_OFFSET_ENA               0x00000020
+#	define     R300_VTX_XY_FMT                       0x00000100
+#	define     R300_VTX_Z_FMT                        0x00000200
+#	define     R300_VTX_W0_FMT                       0x00000400
+#	define     R300_VTX_W0_NORMALIZE                 0x00000800
+#	define     R300_VTX_ST_DENORMALIZED              0x00001000
+
 /* BEGIN: Vertex data assembly - lots of uncertainties */
 /* gap */
 /* Where do we get our vertex data?
@@ -574,6 +587,11 @@ I am fairly certain that they are correct unless stated otherwise in comments.
 #       define R300_TX_SIZE_SHIFT                26 /* largest of width, height */
 #       define R300_TX_SIZE_MASK                 (15 << 26)
 #define R300_TX_FORMAT_0                    0x44C0
+	/* Note - other bits are not known yet.. */
+#	define R300_TX_FORMAT_WIDTH_SHIFT	12
+#	define R300_TX_FORMAT_WIDTH_MASK	(0xf<<R300_TX_FORMAT_WIDTH_SHIFT)
+#	define R300_TX_FORMAT_HEIGHT_SHIFT	16
+#	define R300_TX_FORMAT_HEIGHT_MASK	(0xf<<R300_TX_FORMAT_HEIGHT_SHIFT)
 #define R300_TX_OFFSET_0                    0x4540
 /* BEGIN: Guess from R200 */
 #       define R300_TXO_ENDIAN_NO_SWAP           (0 << 0)

src/mesa/drivers/dri/r300/r300_render.c

@@ -45,6 +45,8 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "array_cache/acache.h"
 #include "tnl/tnl.h"
 
+#include "radeon_reg.h"
+#include "radeon_macros.h"
 #include "radeon_ioctl.h"
 #include "radeon_state.h"
 #include "r300_context.h"
@@ -52,6 +54,7 @@ USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r300_state.h"
 #include "r300_reg.h"
 #include "r300_program.h"
+#include "r300_tex.h"
 
 #include "r300_lib.h"
 
@@ -72,84 +75,90 @@ static int r300_get_primitive_type(r300ContextPtr rmesa,
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    struct vertex_buffer *VB = &tnl->vb;
    GLuint i;
-   int type=-1;
+   int type=-1, min_vertices=0;
+   char *name="UNKNOWN";
    
    if(end<=start)return -1; /* do we need to watch for this ? */
    
-	fprintf(stderr, "[%d-%d]", start, end);
 	switch (prim & PRIM_MODE_MASK) {
-		case GL_LINES:
-   		fprintf(stderr, "L ");
+	case GL_POINTS:
+   		name="P";
+	        type=R300_VAP_VF_CNTL__PRIM_POINTS;
+		min_vertices=1;
+      		break;		
+	case GL_LINES:
+   		name="L";
 	        type=R300_VAP_VF_CNTL__PRIM_LINES;
-		if(end<start+2){
-			fprintf(stderr, "Not enough vertices\n");
-			return -1; /* need enough vertices for Q */
-			}
-      		break;
-		case GL_LINE_STRIP:
-   		fprintf(stderr, "LS ");
+		min_vertices=2;
+      		break;		
+	case GL_LINE_STRIP:
+   		name="LS";
 	        type=R300_VAP_VF_CNTL__PRIM_LINE_STRIP;
-		if(end<start+2){
-			fprintf(stderr, "Not enough vertices\n");
-			return -1; /* need enough vertices for Q */
-			}
+		min_vertices=2;
       		break;
-		case GL_LINE_LOOP:
-   		fprintf(stderr, "LL ");
+	case GL_LINE_LOOP:
+   		name="LL";
+		min_vertices=2;
 		return -1;
-		if(end<start+2){
-			fprintf(stderr, "Not enough vertices\n");
-			return -1; /* need enough vertices for Q */
-			}
       		break;
-    		case GL_TRIANGLES:
-   		fprintf(stderr, "T ");
+    	case GL_TRIANGLES:
+   		name="T";
 	        type=R300_VAP_VF_CNTL__PRIM_TRIANGLES;
-		if(end<start+3){
-			fprintf(stderr, "Not enough vertices\n");
-			return -1; /* need enough vertices for Q */
-			}
+		min_vertices=3;
       		break;
-   		case GL_TRIANGLE_STRIP:
-   		fprintf(stderr, "TS ");
+   	case GL_TRIANGLE_STRIP:
+   		name="TS";
 	        type=R300_VAP_VF_CNTL__PRIM_TRIANGLE_STRIP;
-		if(end<start+3){
-			fprintf(stderr, "Not enough vertices\n");
-			return -1; /* need enough vertices for Q */
-			}
+		min_vertices=3;
       		break;
-   		case GL_TRIANGLE_FAN:
-   		fprintf(stderr, "TF ");
+   	case GL_TRIANGLE_FAN:
+   		name="TF";
 	        type=R300_VAP_VF_CNTL__PRIM_TRIANGLE_FAN;
-		if(end<start+3){
-			fprintf(stderr, "Not enough vertices\n");
-			return -1; /* need enough vertices for Q */
-			}
+		min_vertices=3;
       		break;
-		case GL_QUADS:
-   		fprintf(stderr, "Q ");
+	case GL_QUADS:
+   		name="Q";
 	        type=R300_VAP_VF_CNTL__PRIM_QUADS;
-		if(end<start+4){
-			fprintf(stderr, "Not enough vertices\n");
-			return -1; /* need enough vertices for Q */
-			}
+		min_vertices=4;
       		break;
-		case GL_QUAD_STRIP:
-   		fprintf(stderr, "QS ");
+	case GL_QUAD_STRIP:
+   		name="QS";
 	        type=R300_VAP_VF_CNTL__PRIM_QUAD_STRIP;
-		if(end<start+4){
-			fprintf(stderr, "Not enough vertices\n");
-			return -1; /* need enough vertices for Q */
-			}
+		min_vertices=4;
       		break;
-   		default:
-   		fprintf(stderr, "Cannot handle primitive %02x ", prim & PRIM_MODE_MASK);
+   	default:
+ 		fprintf(stderr, "%s:%s Do not know how to handle primitive %02x - help me !\n",
+			__FILE__, __FUNCTION__,
+			prim & PRIM_MODE_MASK);
 		return -1;
          	break;
-   		}
+   	}
+   #if 0
+   fprintf(stderr, "[%d-%d]%s ", start, end, name);
+   #endif
+   if(start+min_vertices>=end){
+	static int warn_once=1;
+	if(warn_once){
+		fprintf(stderr, "%s:%s Not enough vertices to draw primitive %02x - help me !\n",
+				__FILE__, __FUNCTION__,
+				prim & PRIM_MODE_MASK);
+			warn_once=0;
+			}
+	return -1;
+   	}
    return type;
 }
 
+/* This function compiles GL context into state registers that 
+   describe data routing inside of R300 pipeline.
+   
+   In particular, it programs input_route, output_vtx_fmt, texture
+   unit configuration and gb_output_vtx_fmt
+   
+   This function encompasses setup_AOS() from r300_lib.c
+*/
+
+
 
 
 /* Immediate implementation - vertex data is sent via command stream */
@@ -167,7 +176,9 @@ static GLfloat default_vector[4]={0.0, 0.0, 0.0, 1.0};
 			} \
 	}
 
-static void r300_render_flat_primitive(r300ContextPtr rmesa, 
+/* Immediate implementation - vertex data is sent via command stream */
+
+static void r300_render_immediate_primitive(r300ContextPtr rmesa, 
 	GLcontext *ctx,
 	int start,
 	int end,
@@ -180,10 +191,20 @@ static void r300_render_flat_primitive(r300ContextPtr rmesa,
    LOCAL_VARS
        	
    type=r300_get_primitive_type(rmesa, ctx, start, end, prim);
+		
+   		#if 0
+		fprintf(stderr,"ObjPtr: size=%d stride=%d\n", 
+			VB->ObjPtr->size, VB->ObjPtr->stride);
+		fprintf(stderr,"ColorPtr[0]: size=%d stride=%d\n", 
+			VB->ColorPtr[0]->size, VB->ColorPtr[0]->stride);
+		fprintf(stderr,"TexCoordPtr[0]: size=%d stride=%d\n", 
+			VB->TexCoordPtr[0]->size, VB->TexCoordPtr[0]->stride);
+		#endif
+   
    if(type<0)return;
 
 
-   start_immediate_packet(end-start, type, 8);
+   start_immediate_packet(end-start, type, 8+4*rmesa->state.texture.tc_count);
 
 	for(i=start;i<end;i++){
 		#if 0
@@ -206,44 +227,58 @@ static void r300_render_flat_primitive(r300ContextPtr rmesa,
 		
 		/* color components */
 		output_vector(VB->ColorPtr[0], i);
+		
+		/* texture coordinates */
+		for(k=0;k < ctx->Const.MaxTextureUnits;k++)
+			if(ctx->Texture.Unit[k].Enabled)
+				output_vector(VB->TexCoordPtr[k], i);
 		}
 
 }
 
-static GLboolean r300_run_flat_render(GLcontext *ctx,
+
+static void assign_pipeline(r300ContextPtr rmesa, R300_PIPELINE *p)
+{
+   /* Watch out ! This is buggy .. but will do for now */
+   
+   /* At least one sanity check is in order */
+   if(sizeof(rmesa->state.vertex_shader) != sizeof(p->vertex_shader)){
+   	fprintf(stderr, "Aieee ! vertex_shader sizes don't match.\n");
+	exit(-1);
+   	}
+   if(sizeof(rmesa->state.pixel_shader) != sizeof(p->pixel_shader)){
+   	fprintf(stderr, "Aieee ! vertex_shader sizes don't match.\n");
+	exit(-1);
+   	}
+   
+   memcpy(&rmesa->state.vertex_shader, &(p->vertex_shader), sizeof(rmesa->state.vertex_shader));
+   memcpy(&rmesa->state.pixel_shader, &(p->pixel_shader), sizeof(rmesa->state.pixel_shader));
+
+}
+
+static GLboolean r300_run_immediate_render(GLcontext *ctx,
 				 struct tnl_pipeline_stage *stage)
 {
    r300ContextPtr rmesa = R300_CONTEXT(ctx);
    TNLcontext *tnl = TNL_CONTEXT(ctx);
    struct vertex_buffer *VB = &tnl->vb;
    GLuint i;
-   AOS_DATA vb_arrays[2];
+   /* Only do 2d textures */
+   struct gl_texture_object *to=ctx->Texture.Unit[0].Current2D;
+   r300TexObjPtr t=to->DriverData;
    LOCAL_VARS
 	
+   
+   /* Update texture state - needs to be done only when actually changed..
+      All the time for now.. */
+   /* Flush state - make sure command buffer is nice and large */
+   r300Flush(ctx);
+
+
 	if (RADEON_DEBUG == DEBUG_PRIMS)
 		fprintf(stderr, "%s\n", __FUNCTION__);
 
-   /* setup array of structures data */
-
-   /* Note: immediate vertex data includes all coordinates.
-     To save bandwidth use either VBUF or state-based vertex generation */
-    /* xyz */
-   vb_arrays[0].element_size=4;
-   vb_arrays[0].stride=4;
-   vb_arrays[0].offset=0; /* Not used */
-   vb_arrays[0].format=AOS_FORMAT_FLOAT;
-   vb_arrays[0].ncomponents=4;
-   vb_arrays[0].reg=REG_COORDS;
-
-    /* color */
-   vb_arrays[1].element_size=4;
-   vb_arrays[1].stride=4;
-   vb_arrays[1].offset=0; /* Not used */
-   vb_arrays[1].format=AOS_FORMAT_FLOAT_COLOR;
-   vb_arrays[1].ncomponents=4;
-   vb_arrays[1].reg=REG_COLOR0;
-
-   
+     
    /* needed before starting 3d operation .. */
    reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
 	e32(0x0000000a);
@@ -251,43 +286,75 @@ static GLboolean r300_run_flat_render(GLcontext *ctx,
    reg_start(0x4f18,0);
 	e32(0x00000003);
 	
+   
+	#if 0 /* looks like the Z offset issue got fixed */
+   rmesa->hw.vte.cmd[1] = R300_VPORT_X_SCALE_ENA
+				| R300_VPORT_X_OFFSET_ENA
+				| R300_VPORT_Y_SCALE_ENA
+				| R300_VPORT_Y_OFFSET_ENA
+				| R300_VTX_W0_FMT;
+	#endif
+   R300_STATECHANGE(rmesa, vte);
+   
    r300EmitState(rmesa);
-   
-   reg_start(0x20b0,0);
-	e32(0x0000043f);
-   
-   FLAT_COLOR_PIPELINE.vertex_shader.matrix[0].length=16;
-   memcpy(FLAT_COLOR_PIPELINE.vertex_shader.matrix[0].body.f, ctx->_ModelProjectMatrix.m, 16*4);
+      
+   /* Magic register - note it is right after 20b0 */
 
-   FLAT_COLOR_PIPELINE.vertex_shader.unknown2.length=4;
-   FLAT_COLOR_PIPELINE.vertex_shader.unknown2.body.f[0]=0.0;
-   FLAT_COLOR_PIPELINE.vertex_shader.unknown2.body.f[1]=0.0;
-   FLAT_COLOR_PIPELINE.vertex_shader.unknown2.body.f[2]=1.0;
-   FLAT_COLOR_PIPELINE.vertex_shader.unknown2.body.f[3]=0.0;
+   if(rmesa->state.texture.tc_count>0){
+   	reg_start(0x20b4,0);
+		e32(0x0000000c);
    
-   program_pipeline(PASS_PREFIX &FLAT_COLOR_PIPELINE);
+	   assign_pipeline(rmesa, &SINGLE_TEXTURE_PIPELINE);
+	} else {
+	   assign_pipeline(rmesa, &FLAT_COLOR_PIPELINE);
+	}
+   
+   rmesa->state.vertex_shader.matrix[0].length=16;
+   memcpy(rmesa->state.vertex_shader.matrix[0].body.f, ctx->_ModelProjectMatrix.m, 16*4);
+
+   rmesa->state.vertex_shader.unknown2.length=4;
+   rmesa->state.vertex_shader.unknown2.body.f[0]=0.0;
+   rmesa->state.vertex_shader.unknown2.body.f[1]=0.0;
+   rmesa->state.vertex_shader.unknown2.body.f[2]=1.0;
+   rmesa->state.vertex_shader.unknown2.body.f[3]=0.0;
+	
+   
+   r300EmitVertexShader(rmesa);
+   r300EmitPixelShader(rmesa);
+   
+   #if 0
+   reg_start(R300_RB3D_COLORMASK, 0);
+   	e32(0xf);
+   #endif
+   /* ----------------------------------- */
    
    /* We need LOAD_VBPNTR to setup AOS_ATTR fields.. the offsets are irrelevant */
-   setup_AOS(PASS_PREFIX vb_arrays, 2);
+   r300EmitLOAD_VBPNTR(rmesa, 0);
    
    for(i=0; i < VB->PrimitiveCount; i++){
        GLuint prim = VB->Primitive[i].mode;
        GLuint start = VB->Primitive[i].start;
        GLuint length = VB->Primitive[i].count;
-	r300_render_flat_primitive(rmesa, ctx, start, start + length, prim);
+	r300_render_immediate_primitive(rmesa, ctx, start, start + length, prim);
    	}
 	
-   end_3d(PASS_PREFIX_VOID);
+    /* This sequence is required after any 3d drawing packet
+      I suspect it work arounds a bug (or deficiency) in hardware */
+  
+   reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
+	e32(0x0000000a);
    
-   fprintf(stderr, "\n");
+   reg_start(0x4f18,0);
+	e32(0x00000003);
+         
    return GL_FALSE;
 }
 
+
 /* vertex buffer implementation */
 
 /* We use the start part of GART texture buffer for vertices */
 
-#define R300_MAX_AOS_ARRAYS		16
 
 static void upload_vertex_buffer(r300ContextPtr rmesa, 
 	GLcontext *ctx, AOS_DATA *array, int *n_arrays)
@@ -312,7 +379,7 @@ static void upload_vertex_buffer(r300ContextPtr rmesa,
 			} else { \
 			for(i=0;i<VB->Count;i++){ \
 				/* copy one vertex at a time*/ \
-				memcpy(rsp->gartTextures.map+offset, VEC_ELT(v, GLfloat, i), v->size*4); \
+				memcpy(rsp->gartTextures.map+offset+i*v->size*4, VEC_ELT(v, GLfloat, i), v->size*4); \
 				} \
 			} \
 		/* v->flags &= ~((1<<v->size)-1);*/ \
@@ -325,24 +392,8 @@ static void upload_vertex_buffer(r300ContextPtr rmesa,
 	array[idx].reg=r; \
 	offset+=v->size*4*VB->Count; \
 	idx++; \
-	/* Fill in the rest with the components of default_vector */\
-	/* \
-	if(v->size<4){ \
-		array[idx].element_size=4-v->size; \
-		array[idx].stride=0; \
-		array[idx].format=(f); \
-		array[idx].ncomponents=4-v->size; \
-		array[idx].offset=rsp->gartTextures.handle+v->size*4;\
-		array[idx].reg=r; \
-		idx++; \
-		} \
-	*/\
 	}
 	
-/* Put a copy of default vector */
-memcpy(rsp->gartTextures.map, default_vector, 16);
-offset+=16;
-		
 UPLOAD_VECTOR(VB->ObjPtr, REG_COORDS, AOS_FORMAT_FLOAT);
 UPLOAD_VECTOR(VB->ColorPtr[0], REG_COLOR0, AOS_FORMAT_FLOAT_COLOR);
 
@@ -373,13 +424,6 @@ static void r300_render_vb_flat_primitive(r300ContextPtr rmesa,
    fire_AOS(PASS_PREFIX end-start, type);
 }
 
-static VERTEX_SHADER_FRAGMENT default_vector_vsf={
-	length: 4,
-	body: { 
-		f: {0.0, 0.0, 0.0, 1.0}
-		}
-	};
-
 static GLboolean r300_run_vb_flat_render(GLcontext *ctx,
 				 struct tnl_pipeline_stage *stage)
 {
@@ -405,11 +449,10 @@ static GLboolean r300_run_vb_flat_render(GLcontext *ctx,
    reg_start(0x4f18,0);
 	e32(0x00000003);
    
+   r300_setup_routing(rmesa, ctx, GL_FALSE);
+   
    r300EmitState(rmesa);
    
-   reg_start(0x20b0,0);
-	e32(0x0000043f);
- 
    FLAT_COLOR_PIPELINE.vertex_shader.matrix[0].length=16;
    memcpy(FLAT_COLOR_PIPELINE.vertex_shader.matrix[0].body.f, ctx->_ModelProjectMatrix.m, 16*4);
 
@@ -418,12 +461,9 @@ static GLboolean r300_run_vb_flat_render(GLcontext *ctx,
    FLAT_COLOR_PIPELINE.vertex_shader.unknown2.body.f[1]=0.0;
    FLAT_COLOR_PIPELINE.vertex_shader.unknown2.body.f[2]=1.0;
    FLAT_COLOR_PIPELINE.vertex_shader.unknown2.body.f[3]=0.0;
-   
+
    program_pipeline(PASS_PREFIX &FLAT_COLOR_PIPELINE);
       
-   reg_start(R300_RE_OCCLUSION_CNTL, 0);
-	     e32(R300_OCCLUSION_ON);
-	     
    set_quad0(PASS_PREFIX 1.0,1.0,1.0,1.0);
    set_init21(PASS_PREFIX 0.0,1.0);
    
@@ -435,7 +475,7 @@ static GLboolean r300_run_vb_flat_render(GLcontext *ctx,
         /* copy arrays */
         memcpy(vb_arrays2, vb_arrays, sizeof(AOS_DATA)*n_arrays);
 	for(j=0;j<n_arrays;j++){
-		vb_arrays2[j].offset+=vb_arrays2[j].stride*start;
+		vb_arrays2[j].offset+=vb_arrays2[j].stride*start*4;
 		}
 		
         setup_AOS(PASS_PREFIX vb_arrays2, n_arrays);
@@ -443,6 +483,15 @@ static GLboolean r300_run_vb_flat_render(GLcontext *ctx,
 	r300_render_vb_flat_primitive(rmesa, ctx, start, start + length, prim);
    	}
 	
+    /* This sequence is required after any 3d drawing packet
+      I suspect it work arounds a bug (or deficiency) in hardware */
+  
+  reg_start(R300_RB3D_DSTCACHE_CTLSTAT,0);
+	e32(0x0000000a);
+   
+   reg_start(0x4f18,0);
+	e32(0x00000003);
+   
    end_3d(PASS_PREFIX_VOID);
    
    /* Flush state - we are done drawing.. */
@@ -451,6 +500,7 @@ static GLboolean r300_run_vb_flat_render(GLcontext *ctx,
    return GL_FALSE;
 }
 
+
 /**
  * Called by the pipeline manager to render a batch of primitives.
  * We can return true to pass on to the next stage (i.e. software
@@ -468,8 +518,10 @@ static GLboolean r300_run_render(GLcontext *ctx,
 	if (RADEON_DEBUG == DEBUG_PRIMS)
 		fprintf(stderr, "%s\n", __FUNCTION__);
 
+		
    #if 1
-        return r300_run_flat_render(ctx, stage);
+	
+        return r300_run_immediate_render(ctx, stage);
    #else
 	return GL_TRUE;
    #endif
@@ -546,8 +598,10 @@ static void r300_check_render(GLcontext *ctx, struct tnl_pipeline_stage *stage)
 	//FALLBACK_IF(ctx->Color.DitherFlag);
 
 	/* I'm almost certain I forgot something here */
+	#if 0 /* This should work now.. */
 	FALLBACK_IF(ctx->Color.AlphaEnabled); // GL_ALPHA_TEST
-	FALLBACK_IF(ctx->Color.BlendEnabled); // GL_BLEND
+	#endif
+	//FALLBACK_IF(ctx->Color.BlendEnabled); // GL_BLEND
 	FALLBACK_IF(ctx->Fog.Enabled); // GL_FOG
 	FALLBACK_IF(ctx->Line.SmoothFlag); // GL_LINE_SMOOTH
 	FALLBACK_IF(ctx->Line.StippleFlag); // GL_LINE_STIPPLE
@@ -562,7 +616,8 @@ static void r300_check_render(GLcontext *ctx, struct tnl_pipeline_stage *stage)
 	FALLBACK_IF(ctx->Stencil.Enabled); // GL_STENCIL_TEST
 	FALLBACK_IF(ctx->Multisample.Enabled); // GL_MULTISAMPLE_ARB
 
-	for (i = 0; i < ctx->Const.MaxTextureUnits; i++)
+	/* One step at a time - let one texture pass.. */
+	for (i = 1; i < ctx->Const.MaxTextureUnits; i++)
 		FALLBACK_IF(ctx->Texture.Unit[i].Enabled);
 
 

src/mesa/drivers/dri/r300/r300_state.c

@@ -57,6 +57,315 @@ WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
 #include "r300_reg.h"
 #include "r300_program.h"
 
+static void r300AlphaFunc(GLcontext * ctx, GLenum func, GLfloat ref)
+{
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	int pp_misc = rmesa->hw.at.cmd[R300_AT_ALPHA_TEST];
+	GLubyte refByte;
+
+	CLAMPED_FLOAT_TO_UBYTE(refByte, ref);
+
+	R300_STATECHANGE(rmesa, at);
+
+	pp_misc &= ~(R300_ALPHA_TEST_OP_MASK | R300_REF_ALPHA_MASK);
+	pp_misc |= (refByte & R300_REF_ALPHA_MASK);
+
+	switch (func) {
+	case GL_NEVER:
+		pp_misc |= R300_ALPHA_TEST_FAIL;
+		break;
+	case GL_LESS:
+		pp_misc |= R300_ALPHA_TEST_LESS;
+		break;
+	case GL_EQUAL:
+		pp_misc |= R300_ALPHA_TEST_EQUAL;
+		break;
+	case GL_LEQUAL:
+		pp_misc |= R300_ALPHA_TEST_LEQUAL;
+		break;
+	case GL_GREATER:
+		pp_misc |= R300_ALPHA_TEST_GREATER;
+		break;
+	case GL_NOTEQUAL:
+		pp_misc |= R300_ALPHA_TEST_NEQUAL;
+		break;
+	case GL_GEQUAL:
+		pp_misc |= R300_ALPHA_TEST_GEQUAL;
+		break;
+	case GL_ALWAYS:
+		pp_misc |= R300_ALPHA_TEST_PASS;
+		break;
+	}
+
+	rmesa->hw.at.cmd[R300_AT_ALPHA_TEST] = pp_misc;
+}
+
+static void r300BlendColor(GLcontext * ctx, const GLfloat cf[4])
+{
+	GLubyte color[4];
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	fprintf(stderr, "%s:%s is not implemented yet. Fixme !\n", __FILE__, __FUNCTION__);
+	#if 0
+	R200_STATECHANGE(rmesa, ctx);
+	CLAMPED_FLOAT_TO_UBYTE(color[0], cf[0]);
+	CLAMPED_FLOAT_TO_UBYTE(color[1], cf[1]);
+	CLAMPED_FLOAT_TO_UBYTE(color[2], cf[2]);
+	CLAMPED_FLOAT_TO_UBYTE(color[3], cf[3]);
+	if (rmesa->radeon.radeonScreen->drmSupportsBlendColor)
+		rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCOLOR] =
+		    radeonPackColor(4, color[0], color[1], color[2], color[3]);
+	#endif
+}
+
+/**
+ * Calculate the hardware blend factor setting.  This same function is used
+ * for source and destination of both alpha and RGB.
+ *
+ * \returns
+ * The hardware register value for the specified blend factor.  This value
+ * will need to be shifted into the correct position for either source or
+ * destination factor.
+ *
+ * \todo
+ * Since the two cases where source and destination are handled differently
+ * are essentially error cases, they should never happen.  Determine if these
+ * cases can be removed.
+ */
+static int blend_factor(GLenum factor, GLboolean is_src)
+{
+	int func;
+
+	switch (factor) {
+	case GL_ZERO:
+		func = R200_BLEND_GL_ZERO;
+		break;
+	case GL_ONE:
+		func = R200_BLEND_GL_ONE;
+		break;
+	case GL_DST_COLOR:
+		func = R200_BLEND_GL_DST_COLOR;
+		break;
+	case GL_ONE_MINUS_DST_COLOR:
+		func = R200_BLEND_GL_ONE_MINUS_DST_COLOR;
+		break;
+	case GL_SRC_COLOR:
+		func = R200_BLEND_GL_SRC_COLOR;
+		break;
+	case GL_ONE_MINUS_SRC_COLOR:
+		func = R200_BLEND_GL_ONE_MINUS_SRC_COLOR;
+		break;
+	case GL_SRC_ALPHA:
+		func = R200_BLEND_GL_SRC_ALPHA;
+		break;
+	case GL_ONE_MINUS_SRC_ALPHA:
+		func = R200_BLEND_GL_ONE_MINUS_SRC_ALPHA;
+		break;
+	case GL_DST_ALPHA:
+		func = R200_BLEND_GL_DST_ALPHA;
+		break;
+	case GL_ONE_MINUS_DST_ALPHA:
+		func = R200_BLEND_GL_ONE_MINUS_DST_ALPHA;
+		break;
+	case GL_SRC_ALPHA_SATURATE:
+		func =
+		    (is_src) ? R200_BLEND_GL_SRC_ALPHA_SATURATE :
+		    R200_BLEND_GL_ZERO;
+		break;
+	case GL_CONSTANT_COLOR:
+		func = R200_BLEND_GL_CONST_COLOR;
+		break;
+	case GL_ONE_MINUS_CONSTANT_COLOR:
+		func = R200_BLEND_GL_ONE_MINUS_CONST_COLOR;
+		break;
+	case GL_CONSTANT_ALPHA:
+		func = R200_BLEND_GL_CONST_ALPHA;
+		break;
+	case GL_ONE_MINUS_CONSTANT_ALPHA:
+		func = R200_BLEND_GL_ONE_MINUS_CONST_ALPHA;
+		break;
+	default:
+		func = (is_src) ? R200_BLEND_GL_ONE : R200_BLEND_GL_ZERO;
+	}
+	return func;
+}
+
+/**
+ * Sets both the blend equation and the blend function.
+ * This is done in a single
+ * function because some blend equations (i.e., \c GL_MIN and \c GL_MAX)
+ * change the interpretation of the blend function.
+ * Also, make sure that blend function and blend equation are set to their default
+ * value if color blending is not enabled, since at least blend equations GL_MIN
+ * and GL_FUNC_REVERSE_SUBTRACT will cause wrong results otherwise for
+ * unknown reasons.
+ */
+static void r300_set_blend_state(GLcontext * ctx)
+{
+	r300ContextPtr rmesa = R300_CONTEXT(ctx);
+	#if 0
+	GLuint cntl = rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] &
+	    ~(R300_ROP_ENABLE | R300_ALPHA_BLEND_ENABLE |
+	      R300_SEPARATE_ALPHA_ENABLE);
+	#endif
+
+	int func = (R200_BLEND_GL_ONE << R200_SRC_BLEND_SHIFT) |
+	    (R200_BLEND_GL_ZERO << R200_DST_BLEND_SHIFT);
+	int eqn = R200_COMB_FCN_ADD_CLAMP;
+	int funcA = (R200_BLEND_GL_ONE << R200_SRC_BLEND_SHIFT) |
+	    (R200_BLEND_GL_ZERO << R200_DST_BLEND_SHIFT);
+	int eqnA = R200_COMB_FCN_ADD_CLAMP;
+
+	R300_STATECHANGE(rmesa, bld);
+
+	if (rmesa->radeon.radeonScreen->drmSupportsBlendColor) {
+		if (ctx->Color._LogicOpEnabled) {
+			#if 0
+			rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] =
+			    cntl | R300_ROP_ENABLE;
+			#endif
+			rmesa->hw.bld.cmd[R300_BLD_ABLEND] = eqn | func;
+			rmesa->hw.bld.cmd[R300_BLD_CBLEND] = eqn | func;
+			return;
+		} else if (ctx->Color.BlendEnabled) {
+			#if 0
+			rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] =
+			    cntl | R300_ALPHA_BLEND_ENABLE |
+			    R300_SEPARATE_ALPHA_ENABLE;
+			#endif
+		} else {
+			#if 0
+			rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] = cntl;
+			#endif
+			rmesa->hw.bld.cmd[R300_BLD_ABLEND] = eqn | func;
+			rmesa->hw.bld.cmd[R300_BLD_CBLEND] = eqn | func;
+			return;
+		}
+	} else {
+		if (ctx->Color._LogicOpEnabled) {
+			#if 0
+			rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] =
+			    cntl | R300_ROP_ENABLE;
+			rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCNTL] = eqn | func;
+			#endif
+			return;
+		} else if (ctx->Color.BlendEnabled) {
+			#if 0
+			rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] =
+			    cntl | R300_ALPHA_BLEND_ENABLE;
+			#endif
+		} else {
+			#if 0
+			rmesa->hw.ctx.cmd[CTX_RB3D_CNTL] = cntl;
+			rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCNTL] = eqn | func;
+			#endif
+			return;
+		}
+	}
+
+	func =
+	    (blend_factor(ctx->Color.BlendSrcRGB, GL_TRUE) <<
+	     R200_SRC_BLEND_SHIFT) | (blend_factor(ctx->Color.BlendDstRGB,
+						   GL_FALSE) <<
+				      R200_DST_BLEND_SHIFT);
+
+	switch (ctx->Color.BlendEquationRGB) {
+	case GL_FUNC_ADD:
+		eqn = R300_COMB_FCN_ADD_CLAMP;
+		break;
+
+	case GL_FUNC_SUBTRACT:
+		eqn = R300_COMB_FCN_SUB_CLAMP;
+		break;
+
+	case GL_FUNC_REVERSE_SUBTRACT:
+		eqn = R200_COMB_FCN_RSUB_CLAMP;
+		break;
+
+	case GL_MIN:
+		eqn = R200_COMB_FCN_MIN;
+		func = (R200_BLEND_GL_ONE << R200_SRC_BLEND_SHIFT) |
+		    (R200_BLEND_GL_ONE << R200_DST_BLEND_SHIFT);
+		break;
+
+	case GL_MAX:
+		eqn = R200_COMB_FCN_MAX;
+		func = (R200_BLEND_GL_ONE << R200_SRC_BLEND_SHIFT) |
+		    (R200_BLEND_GL_ONE << R200_DST_BLEND_SHIFT);
+		break;
+
+	default:
+		fprintf(stderr,
+			"[%s:%u] Invalid RGB blend equation (0x%04x).\n",
+			__func__, __LINE__, ctx->Color.BlendEquationRGB);
+		return;
+	}
+
+	if (!rmesa->radeon.radeonScreen->drmSupportsBlendColor) {
+		#if 0
+		rmesa->hw.ctx.cmd[CTX_RB3D_BLENDCNTL] = eqn | func;
+		#endif
+		return;
+	}
+
+	funcA =
+	    (blend_factor(ctx->Color.BlendSrcA, GL_TRUE) <<
+	     R200_SRC_BLEND_SHIFT) | (blend_factor(ctx->Color.BlendDstA,
+						   GL_FALSE) <<
+				      R200_DST_BLEND_SHIFT);
+
+	switch (ctx->Color.BlendEquationA) {
+	case GL_FUNC_ADD:
+		eqnA = R300_COMB_FCN_ADD_CLAMP;
+		break;
+
+	case GL_FUNC_SUBTRACT:
+		eqnA = R300_COMB_FCN_SUB_CLAMP;
+		break;
+
+	case GL_FUNC_REVERSE_SUBTRACT:
+		eqnA = R200_COMB_FCN_RSUB_CLAMP;
+		break;
+
+	case GL_MIN:
+		eqnA = R200_COMB_FCN_MIN;
+		funcA = (R200_BLEND_GL_ONE << R200_SRC_BLEND_SHIFT) |
+		    (R200_BLEND_GL_ONE << R200_DST_BLEND_SHIFT);
+		break;
+
+	case GL_MAX:
+		eqnA = R200_COMB_FCN_MAX;
+		funcA = (R200_BLEND_GL_ONE << R200_SRC_BLEND_SHIFT) |
+		    (R200_BLEND_GL_ONE << R200_DST_BLEND_SHIFT);
+		break;
+
+	default:
+		fprintf(stderr, "[%s:%u] Invalid A blend equation (0x%04x).\n",
+			__func__, __LINE__, ctx->Color.BlendEquationA);
+		return;
+	}
+
+	rmesa->hw.bld.cmd[R300_BLD_ABLEND] = eqnA | funcA;
+	rmesa->hw.bld.cmd[R300_BLD_CBLEND] = eqn | func ;
+	if(rmesa->hw.bld.cmd[R300_BLD_ABLEND] == rmesa->hw.bld.cmd[R300_BLD_CBLEND]){
+		rmesa->hw.bld.cmd[R300_BLD_CBLEND] |= R300_BLEND_UNKNOWN | R300_BLEND_ENABLE | R300_BLEND_NO_SEPARATE;
+		} else {
+		rmesa->hw.bld.cmd[R300_BLD_CBLEND] |= R300_BLEND_UNKNOWN | R300_BLEND_ENABLE;
+		}
+
+}
+
+static void r300BlendEquationSeparate(GLcontext * ctx,
+				      GLenum modeRGB, GLenum modeA)
+{
+	r300_set_blend_state(ctx);
+}
+
+static void r300BlendFuncSeparate(GLcontext * ctx,
+				  GLenum sfactorRGB, GLenum dfactorRGB,
+				  GLenum sfactorA, GLenum dfactorA)
+{
+	r300_set_blend_state(ctx);
+}
 
 /**
  * Update our tracked culling state based on Mesa's state.
@@ -225,6 +534,18 @@ static void r300ColorMask(GLcontext* ctx,
 	}
 }
 
+/* =============================================================
+ * Point state
+ */
+static void r300PointSize(GLcontext * ctx, GLfloat size)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	
+	/* This might need fixing later */
+	R300_STATECHANGE(r300, vps);
+	r300->hw.vps.cmd[R300_VPS_POINTSIZE] = r300PackFloat32(1.0);
+}
+
 /* =============================================================
  * Window position and viewport transformation
  */
@@ -277,6 +598,331 @@ static void r300DepthRange(GLcontext * ctx, GLclampd nearval, GLclampd farval)
 	r300UpdateWindow(ctx);
 }
 
+/* Routing and texture-related */
+
+void r300_setup_routing(GLcontext *ctx, GLboolean immediate)
+{
+	int i, count=0,reg=0;
+	GLuint dw, mask;
+	TNLcontext *tnl = TNL_CONTEXT(ctx);
+	struct vertex_buffer *VB = &tnl->vb;
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	
+	
+	/* Stage 1 - input to VAP */
+	
+	/* Assign register number automatically, retaining it in rmesa->state.reg */
+	
+	/* Note: immediate vertex data includes all coordinates.
+	To save bandwidth use either VBUF or state-based vertex generation */
+	
+	#define CONFIGURE_AOS(v, o, r, f) \
+		{\
+		if(immediate){ \
+			r300->state.aos[count].element_size=4; \
+			r300->state.aos[count].stride=4; \
+			r300->state.aos[count].ncomponents=4; \
+			} else { \
+			r300->state.aos[count].element_size=v->size; \
+			r300->state.aos[count].stride=v->size; \
+			r300->state.aos[count].ncomponents=v->size; \
+			} \
+		r300->state.aos[count].offset=o; \
+		r300->state.aos[count].reg=reg; \
+		r300->state.aos[count].format=(f); \
+		r300->state.vap_reg.r=reg; \
+		count++; \
+		reg++; \
+		}
+	
+		/* All offsets are 0 - for use by immediate mode. 
+		Should change later to handle vertex buffers */
+	CONFIGURE_AOS(VB->ObjPtr, 0, i_coords, AOS_FORMAT_FLOAT);
+	CONFIGURE_AOS(VB->ColorPtr[0], 0, i_color[0], AOS_FORMAT_FLOAT_COLOR);
+	for(i=0;i < ctx->Const.MaxTextureUnits;i++)
+		if(ctx->Texture.Unit[i].Enabled)
+			CONFIGURE_AOS(VB->TexCoordPtr[i], 0, i_tex[i], AOS_FORMAT_FLOAT);
+			
+	r300->state.aos_count=count;
+	
+	if (RADEON_DEBUG & DEBUG_STATE)
+		fprintf(stderr, "aos_count=%d\n", count);
+	
+	if(count>R300_MAX_AOS_ARRAYS){
+		fprintf(stderr, "Aieee ! AOS array count exceeded !\n");
+		exit(-1);
+		}
+			
+	/* Implement AOS */
+	
+	/* setup INPUT_ROUTE */
+	R300_STATECHANGE(r300, vir[0]);
+	for(i=0;i+1<count;i+=2){
+		dw=(r300->state.aos[i].ncomponents-1) 
+		| ((r300->state.aos[i].reg)<<8)
+		| (r300->state.aos[i].format<<14)
+		| (((r300->state.aos[i+1].ncomponents-1) 
+		| ((r300->state.aos[i+1].reg)<<8)
+		| (r300->state.aos[i+1].format<<14))<<16);
+		
+		if(i+2==count){
+			dw|=(1<<(13+16));
+			}
+		r300->hw.vir[0].cmd[R300_VIR_CNTL_0+(i>>1)]=dw;
+		}
+	if(count & 1){
+		dw=(r300->state.aos[count-1].ncomponents-1)
+		| (r300->state.aos[count-1].format<<14)
+		| ((r300->state.aos[count-1].reg)<<8)
+		| (1<<13);
+		r300->hw.vir[0].cmd[R300_VIR_CNTL_0+(count>>1)]=dw;
+		//fprintf(stderr, "vir0 dw=%08x\n", dw);
+		}
+	/* Set the rest of INPUT_ROUTE_0 to 0 */
+	//for(i=((count+1)>>1); i<8; i++)r300->hw.vir[0].cmd[R300_VIR_CNTL_0+i]=(0x0);
+	((drm_r300_cmd_header_t*)r300->hw.vir[0].cmd)->unchecked_state.count = (count+1)>>1;
+	
+	
+	/* Mesa assumes that all missing components are from (0, 0, 0, 1) */
+	#define ALL_COMPONENTS ((R300_INPUT_ROUTE_SELECT_X<<R300_INPUT_ROUTE_X_SHIFT) \
+		| (R300_INPUT_ROUTE_SELECT_Y<<R300_INPUT_ROUTE_Y_SHIFT) \
+		| (R300_INPUT_ROUTE_SELECT_Z<<R300_INPUT_ROUTE_Z_SHIFT) \
+		| (R300_INPUT_ROUTE_SELECT_W<<R300_INPUT_ROUTE_W_SHIFT))
+	
+	#define ALL_DEFAULT ((R300_INPUT_ROUTE_SELECT_ZERO<<R300_INPUT_ROUTE_X_SHIFT) \
+		| (R300_INPUT_ROUTE_SELECT_ZERO<<R300_INPUT_ROUTE_Y_SHIFT) \
+		| (R300_INPUT_ROUTE_SELECT_ZERO<<R300_INPUT_ROUTE_Z_SHIFT) \
+		| (R300_INPUT_ROUTE_SELECT_ONE<<R300_INPUT_ROUTE_W_SHIFT))
+	
+	R300_STATECHANGE(r300, vir[1]);
+		
+	for(i=0;i+1<count;i+=2){
+		/* do i first.. */
+		mask=(1<<(r300->state.aos[i].ncomponents*3))-1;
+		dw=(ALL_COMPONENTS & mask)
+		| (ALL_DEFAULT & ~mask)
+		| R300_INPUT_ROUTE_ENABLE;
+		
+		/* i+1 */
+		mask=(1<<(r300->state.aos[i+1].ncomponents*3))-1;
+		dw|=( 
+		(ALL_COMPONENTS & mask)
+		| (ALL_DEFAULT & ~mask)
+		| R300_INPUT_ROUTE_ENABLE
+		)<<16;
+	
+		r300->hw.vir[1].cmd[R300_VIR_CNTL_0+(i>>1)]=dw;
+		}
+	if(count & 1){
+		mask=(1<<(r300->state.aos[count-1].ncomponents*3))-1;
+		dw=(ALL_COMPONENTS & mask)
+		| (ALL_DEFAULT & ~mask)
+		| R300_INPUT_ROUTE_ENABLE;
+		r300->hw.vir[1].cmd[R300_VIR_CNTL_0+(count>>1)]=dw;
+		//fprintf(stderr, "vir1 dw=%08x\n", dw);
+		}
+	/* Set the rest of INPUT_ROUTE_1 to 0 */
+	//for(i=((count+1)>>1); i<8; i++)r300->hw.vir[1].cmd[R300_VIR_CNTL_0+i]=0x0;
+	((drm_r300_cmd_header_t*)r300->hw.vir[1].cmd)->unchecked_state.count = (count+1)>>1;
+	
+	/* Set up input_cntl */
+	
+	R300_STATECHANGE(r300, vic);
+	r300->hw.vic.cmd[R300_VIC_CNTL_0]=0x5555;  /* Hard coded value, no idea what it means */
+	
+	r300->hw.vic.cmd[R300_VIC_CNTL_1]=R300_INPUT_CNTL_POS
+					| R300_INPUT_CNTL_COLOR;
+	
+	for(i=0;i < ctx->Const.MaxTextureUnits;i++)
+		if(ctx->Texture.Unit[i].Enabled)
+			r300->hw.vic.cmd[R300_VIC_CNTL_1]|=(R300_INPUT_CNTL_TC0<<i);
+	
+	/* Stage 3: VAP output */
+	R300_STATECHANGE(r300, vof);
+	r300->hw.vof.cmd[R300_VOF_CNTL_0]=R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT
+					| R300_VAP_OUTPUT_VTX_FMT_0__COLOR_PRESENT;
+	
+	r300->hw.vof.cmd[R300_VOF_CNTL_1]=0;
+	for(i=0;i < ctx->Const.MaxTextureUnits;i++)
+		if(ctx->Texture.Unit[i].Enabled)
+			r300->hw.vof.cmd[R300_VOF_CNTL_1]|=(4<<(3*i));
+	
+}
+
+static r300TexObj default_tex_obj={
+	filter:R300_TX_MAG_FILTER_LINEAR | R300_TX_MIN_FILTER_LINEAR,
+	pitch: 0x8000,
+	size: (0xff << R300_TX_WIDTHMASK_SHIFT) 
+	      | (0xff << R300_TX_HEIGHTMASK_SHIFT)
+	      | (0x8 << R300_TX_SIZE_SHIFT),
+	format: 0x88a0c,
+	offset: 0x0,
+	unknown4: 0x0,
+	unknown5: 0x0
+	};
+
+	/* there is probably a system to these value, but, for now, 
+	   we just try by hand */
+static GLuint translate_texture_format(GLuint format)
+{
+	switch(format){
+	case 0x88047:
+	case 0x55047:
+		return 0x53a0c;
+	default:
+		{
+		static int warn_once=1;
+		if(warn_once){
+			fprintf(stderr, "%s:%s Do not know how to translate texture format %08x - help me !\n",
+				__FILE__, __FUNCTION__,
+				format);
+			warn_once=0;
+			}
+		}
+		return 0;
+	}
+}
+	
+void r300_setup_textures(GLcontext *ctx)
+{
+	int i, mtu;
+	struct r300_tex_obj *t;
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	int max_texture_unit=-1; /* -1 translates into no setup costs for fields */
+	
+	R300_STATECHANGE(r300, txe);
+	R300_STATECHANGE(r300, tex.filter);
+	R300_STATECHANGE(r300, tex.unknown1);
+	R300_STATECHANGE(r300, tex.size);
+	R300_STATECHANGE(r300, tex.format);
+	R300_STATECHANGE(r300, tex.offset);
+	R300_STATECHANGE(r300, tex.unknown4);
+	R300_STATECHANGE(r300, tex.unknown5);
+	
+	r300->state.texture.tc_count=0;
+	
+	r300->hw.txe.cmd[R300_TXE_ENABLE]=0x0;
+	
+	mtu = r300->radeon.glCtx->Const.MaxTextureUnits;
+	if (RADEON_DEBUG & DEBUG_STATE)
+		fprintf(stderr, "mtu=%d\n", mtu);
+	
+	if(mtu>R300_MAX_TEXTURE_UNITS){
+		fprintf(stderr, "Aiiee ! mtu=%d is greater than R300_MAX_TEXTURE_UNITS=%d\n", 
+			mtu, R300_MAX_TEXTURE_UNITS);
+		exit(-1);
+		}
+	for(i=0;i<mtu;i++){
+		if(ctx->Texture.Unit[i].Enabled){
+			t=r300->state.texture.unit[i].texobj;
+			r300->state.texture.tc_count++;
+			if(t==NULL){
+				fprintf(stderr, "Texture unit %d enabled, but corresponding texobj is NULL, using default object.\n", i);
+				//exit(-1);
+				t=&default_tex_obj;
+				}
+			if (RADEON_DEBUG & DEBUG_STATE)
+				fprintf(stderr, "Activating texture unit %d\n", i);
+			max_texture_unit=i;
+			r300->hw.txe.cmd[R300_TXE_ENABLE]|=(1<<i);
+			
+			r300->hw.tex.filter.cmd[R300_TEX_VALUE_0+i]=t->filter;
+			r300->hw.tex.unknown1.cmd[R300_TEX_VALUE_0+i]=t->pitch;
+			r300->hw.tex.size.cmd[R300_TEX_VALUE_0+i]=t->size;
+			r300->hw.tex.format.cmd[R300_TEX_VALUE_0+i]=t->format;
+			r300->hw.tex.offset.cmd[R300_TEX_VALUE_0+i]=r300->radeon.radeonScreen->fbLocation+t->offset;
+			r300->hw.tex.unknown4.cmd[R300_TEX_VALUE_0+i]=0x0;
+			r300->hw.tex.unknown5.cmd[R300_TEX_VALUE_0+i]=0x0;
+			
+			
+			/* We don't know how to set this yet */
+			//value from r300_lib.c for RGB24
+			//r300->hw.tex.format.cmd[R300_TEX_VALUE_0+i]=0x88a0c; 
+			r300->hw.tex.format.cmd[R300_TEX_VALUE_0+i]=translate_texture_format(t->format);
+			/* Use the code below to quickly find matching texture
+			   formats. Requires an app that displays the same texture
+			   repeatedly  */
+			      #if 1
+				if(r300->hw.tex.format.cmd[R300_TEX_VALUE_0+i]==0){ 
+					static int fmt=0;
+					static int k=0;
+					k++;
+					if(k>200){
+						k=0;
+						fmt++;
+						fprintf(stderr, "Want to set format %08x\n", t->format);
+						if(fmt>0xff){
+							//exit(-1);
+							fmt=0;
+							}
+						//sleep(1);
+						fprintf(stderr, "Instead trying format %08x\n", 
+							0x00a0c | (fmt<<12));
+						}
+					r300->hw.tex.format.cmd[R300_TEX_VALUE_0+i]=0x00a0c | (fmt<<12);
+					}
+			      #endif
+			
+			}
+			
+		}
+	((drm_r300_cmd_header_t*)r300->hw.tex.filter.cmd)->unchecked_state.count = max_texture_unit+1;
+	((drm_r300_cmd_header_t*)r300->hw.tex.unknown1.cmd)->unchecked_state.count = max_texture_unit+1;
+	((drm_r300_cmd_header_t*)r300->hw.tex.size.cmd)->unchecked_state.count = max_texture_unit+1;
+	((drm_r300_cmd_header_t*)r300->hw.tex.format.cmd)->unchecked_state.count = max_texture_unit+1;
+	((drm_r300_cmd_header_t*)r300->hw.tex.offset.cmd)->unchecked_state.count = max_texture_unit+1;
+	((drm_r300_cmd_header_t*)r300->hw.tex.unknown4.cmd)->unchecked_state.count = max_texture_unit+1;
+	((drm_r300_cmd_header_t*)r300->hw.tex.unknown5.cmd)->unchecked_state.count = max_texture_unit+1;
+	
+	if (RADEON_DEBUG & DEBUG_STATE)
+		fprintf(stderr, "TX_ENABLE: %08x  max_texture_unit=%d\n", r300->hw.txe.cmd[R300_TXE_ENABLE], max_texture_unit);
+}
+
+void r300_setup_rs_unit(GLcontext *ctx)
+{
+	r300ContextPtr r300 = R300_CONTEXT(ctx);
+	int i;
+	
+	/* This needs to be rewritten - it is a hack at best */
+	
+	R300_STATECHANGE(r300, ri);
+	R300_STATECHANGE(r300, rc);
+	R300_STATECHANGE(r300, rr);
+	
+	for(i = 1; i <= 8; ++i)
+		r300->hw.ri.cmd[i] = 0x00d10000;
+	r300->hw.ri.cmd[R300_RI_INTERP_1] |= R300_RS_INTERP_1_UNKNOWN;
+	r300->hw.ri.cmd[R300_RI_INTERP_2] |= R300_RS_INTERP_2_UNKNOWN;
+	r300->hw.ri.cmd[R300_RI_INTERP_3] |= R300_RS_INTERP_3_UNKNOWN;
+	
+	for(i = 1; i <= 8; ++i)
+		r300->hw.rr.cmd[i] = 0;
+	/* textures enabled ? */
+	if(r300->state.texture.tc_count>0){
+	
+		/* This code only really works with one set of texture coordinates */
+		
+		/* The second constant is needed to get glxgears display anything .. */
+		r300->hw.rc.cmd[1] = R300_RS_CNTL_0_UNKNOWN_7 
+				| R300_RS_CNTL_0_UNKNOWN_18 
+				| (r300->state.texture.tc_count<<R300_RS_CNTL_TC_CNT_SHIFT);
+		r300->hw.rc.cmd[2] = 0xc0;
+	
+	
+		((drm_r300_cmd_header_t*)r300->hw.rr.cmd)->unchecked_state.count = 1;
+		r300->hw.rr.cmd[R300_RR_ROUTE_0] = 0x24008;
+		
+		} else {
+		
+		/* The second constant is needed to get glxgears display anything .. */
+		r300->hw.rc.cmd[1] = R300_RS_CNTL_0_UNKNOWN_7 | R300_RS_CNTL_0_UNKNOWN_18;
+		r300->hw.rc.cmd[2] = 0;
+		
+		((drm_r300_cmd_header_t*)r300->hw.rr.cmd)->unchecked_state.count = 1;
+		r300->hw.rr.cmd[R300_RR_ROUTE_0] = 0x4000;
+		
+		}
+}
 
 /**
  * Called by Mesa after an internal state update.
@@ -320,6 +966,16 @@ void r300ResetHwState(r300ContextPtr r300)
 	r300DepthFunc(ctx, ctx->Depth.Func);
 
 	r300UpdateCulling(ctx);
+        
+	r300_setup_routing(ctx, GL_TRUE);
+	
+	r300UpdateTextureState(ctx);
+	r300_setup_textures(ctx);
+	r300_setup_rs_unit(ctx);
+	
+	
+	r300_set_blend_state(ctx);
+	r300AlphaFunc(ctx, ctx->Color.AlphaFunc, ctx->Color.AlphaRef);
 
 //BEGIN: TODO
 	r300->hw.unk2080.cmd[1] = 0x0030045A;
@@ -327,14 +983,21 @@ void r300ResetHwState(r300ContextPtr r300)
 	r300->hw.ovf.cmd[R300_OVF_FMT_0] = 0x00000003;
 	r300->hw.ovf.cmd[R300_OVF_FMT_1] = 0x00000000;
 
-	r300->hw.unk20B0.cmd[1] = 0x0000040A;
-	r300->hw.unk20B0.cmd[2] = 0x00000008;
+	r300->hw.vte.cmd[1] = R300_VPORT_X_SCALE_ENA
+				| R300_VPORT_X_OFFSET_ENA
+				| R300_VPORT_Y_SCALE_ENA
+				| R300_VPORT_Y_OFFSET_ENA
+				| R300_VPORT_Z_SCALE_ENA
+				| R300_VPORT_Z_OFFSET_ENA
+				| R300_VTX_W0_FMT;
+	r300->hw.vte.cmd[2] = 0x00000008;
 
 	r300->hw.unk2134.cmd[1] = 0x00FFFFFF;
 	r300->hw.unk2134.cmd[2] = 0x00000000;
 
 	r300->hw.unk2140.cmd[1] = 0x00000000;
 
+	#if 0 /* Done in setup routing */
 	((drm_r300_cmd_header_t*)r300->hw.vir[0].cmd)->unchecked_state.count = 1;
 	r300->hw.vir[0].cmd[1] = 0x21030003;
 
@@ -343,7 +1006,8 @@ void r300ResetHwState(r300ContextPtr r300)
 
 	r300->hw.vic.cmd[R300_VIR_CNTL_0] = 0x00000001;
 	r300->hw.vic.cmd[R300_VIR_CNTL_1] = 0x00000405;
-
+	#endif
+	
 	r300->hw.unk21DC.cmd[1] = 0xAAAAAAAA;
 
 	r300->hw.unk221C.cmd[1] = R300_221C_NORMAL;
@@ -358,10 +1022,12 @@ void r300ResetHwState(r300ContextPtr r300)
 	else
 		r300->hw.unk2288.cmd[1] = R300_2288_RV350;
 
+	#if 0
 	r300->hw.vof.cmd[R300_VOF_CNTL_0] = R300_VAP_OUTPUT_VTX_FMT_0__POS_PRESENT
 				| R300_VAP_OUTPUT_VTX_FMT_0__COLOR_PRESENT;
 	r300->hw.vof.cmd[R300_VOF_CNTL_1] = 0; /* no textures */
-		
+	#endif	
+	
 	r300->hw.pvs.cmd[R300_PVS_CNTL_1] = 0;
 	r300->hw.pvs.cmd[R300_PVS_CNTL_2] = 0;
 	r300->hw.pvs.cmd[R300_PVS_CNTL_3] = 0;
@@ -383,7 +1049,7 @@ void r300ResetHwState(r300ContextPtr r300)
 	r300->hw.gb_misc.cmd[R300_GB_MISC_SELECT] = 0x00000000;
 	r300->hw.gb_misc.cmd[R300_GB_MISC_AA_CONFIG] = 0x00000000; /* No antialiasing */
 
-	r300->hw.txe.cmd[R300_TXE_ENABLE] = 0;
+	//r300->hw.txe.cmd[R300_TXE_ENABLE] = 0;
 
 	r300->hw.unk4200.cmd[1] = r300PackFloat32(0.0);
 	r300->hw.unk4200.cmd[2] = r300PackFloat32(0.0);
@@ -421,20 +1087,6 @@ void r300ResetHwState(r300ContextPtr r300)
 	r300->hw.unk42C0.cmd[1] = 0x4B7FFFFF;
 	r300->hw.unk42C0.cmd[2] = 0x00000000;
 
-	/* The second constant is needed to get glxgears display anything .. */
-	r300->hw.rc.cmd[1] = R300_RS_CNTL_0_UNKNOWN_7 | R300_RS_CNTL_0_UNKNOWN_18;
-	r300->hw.rc.cmd[2] = 0;
-
-	for(i = 1; i <= 8; ++i)
-		r300->hw.ri.cmd[i] = 0x00d10000;
-	r300->hw.ri.cmd[R300_RI_INTERP_1] |= R300_RS_INTERP_1_UNKNOWN;
-	r300->hw.ri.cmd[R300_RI_INTERP_2] |= R300_RS_INTERP_2_UNKNOWN;
-	r300->hw.ri.cmd[R300_RI_INTERP_3] |= R300_RS_INTERP_3_UNKNOWN;
-
-	((drm_r300_cmd_header_t*)r300->hw.rr.cmd)->unchecked_state.count = 1;
-	for(i = 1; i <= 8; ++i)
-		r300->hw.rr.cmd[i] = 0;
-	r300->hw.rr.cmd[R300_RR_ROUTE_0] = 0x4000;
 
 	r300->hw.unk43A4.cmd[1] = 0x0000001C;
 	r300->hw.unk43A4.cmd[2] = 0x2DA49525;
@@ -469,14 +1121,18 @@ void r300ResetHwState(r300ContextPtr r300)
 	r300->hw.unk4BC8.cmd[2] = 0;
 	r300->hw.unk4BC8.cmd[3] = 0;
 
+	#if 0
 	r300->hw.at.cmd[R300_AT_ALPHA_TEST] = 0;
+	#endif
 
 	r300->hw.unk4BD8.cmd[1] = 0;
 
 	r300->hw.unk4E00.cmd[1] = 0;
 
+	#if 0
 	r300->hw.bld.cmd[R300_BLD_CBLEND] = 0;
 	r300->hw.bld.cmd[R300_BLD_ABLEND] = 0;
+	#endif
 
 	r300->hw.unk4E10.cmd[1] = 0;
 	r300->hw.unk4E10.cmd[2] = 0;
@@ -541,8 +1197,9 @@ void r300ResetHwState(r300ContextPtr r300)
 	r300->hw.vps.cmd[R300_VPS_ZERO_1] = 0;
 	r300->hw.vps.cmd[R300_VPS_POINTSIZE] = r300PackFloat32(1.0);
 	r300->hw.vps.cmd[R300_VPS_ZERO_3] = 0;
+	
 //END: TODO
-
+	
 	r300->hw.all_dirty = GL_TRUE;
 }
 
@@ -555,9 +1212,29 @@ void r300ResetHwState(r300ContextPtr r300)
  */
 void r300InitState(r300ContextPtr r300)
 {
+	GLcontext *ctx = r300->radeon.glCtx;
+	GLuint depth_fmt;
+
 	radeonInitState(&r300->radeon);
 	
-	r300->state.depth.scale = 1.0 / (GLfloat) 0xffff;
+	switch (ctx->Visual.depthBits) {
+	case 16:
+		r300->state.depth.scale = 1.0 / (GLfloat) 0xffff;
+		depth_fmt = R200_DEPTH_FORMAT_16BIT_INT_Z;
+		//r300->state.stencil.clear = 0x00000000;
+		break;
+	case 24:
+		r300->state.depth.scale = 1.0 / (GLfloat) 0xffffff;
+		depth_fmt = R200_DEPTH_FORMAT_24BIT_INT_Z;
+		//r300->state.stencil.clear = 0xff000000;
+		break;
+	default:
+		fprintf(stderr, "Error: Unsupported depth %d... exiting\n",
+			ctx->Visual.depthBits);
+		exit(-1);
+	}
+	
+	memset(&(r300->state.texture), 0, sizeof(r300->state.texture));
 
 	r300ResetHwState(r300);
 }
@@ -572,6 +1249,10 @@ void r300InitStateFuncs(struct dd_function_table* functions)
 	radeonInitStateFuncs(functions);
 
 	functions->UpdateState = r300InvalidateState;
+	//functions->AlphaFunc = r300AlphaFunc;
+	functions->BlendColor = r300BlendColor;
+	functions->BlendEquationSeparate = r300BlendEquationSeparate;
+	functions->BlendFuncSeparate = r300BlendFuncSeparate;
 	functions->Enable = r300Enable;
 	functions->ColorMask = r300ColorMask;
 	functions->DepthFunc = r300DepthFunc;
@@ -582,5 +1263,6 @@ void r300InitStateFuncs(struct dd_function_table* functions)
 	/* Viewport related */
 	functions->Viewport = r300Viewport;
 	functions->DepthRange = r300DepthRange;
+	functions->PointSize = r300PointSize;
 }
 

src/mesa/drivers/dri/r300/r300_tex.h

@@ -0,0 +1,53 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r300/r300_tex.h,v 1.1 2002/10/30 12:51:53 alanh Exp $ */
+/*
+Copyright (C) The Weather Channel, Inc.  2002.  All Rights Reserved.
+
+The Weather Channel (TM) funded Tungsten Graphics to develop the
+initial release of the Radeon 8500 driver under the XFree86 license.
+This notice must be preserved.
+
+Permission is hereby granted, free of charge, to any person obtaining
+a copy of this software and associated documentation files (the
+"Software"), to deal in the Software without restriction, including
+without limitation the rights to use, copy, modify, merge, publish,
+distribute, sublicense, and/or sell copies of the Software, and to
+permit persons to whom the Software is furnished to do so, subject to
+the following conditions:
+
+The above copyright notice and this permission notice (including the
+next paragraph) shall be included in all copies or substantial
+portions of the Software.
+
+THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND,
+EXPRESS OR IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF
+MERCHANTABILITY, FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT.
+IN NO EVENT SHALL THE COPYRIGHT OWNER(S) AND/OR ITS SUPPLIERS BE
+LIABLE FOR ANY CLAIM, DAMAGES OR OTHER LIABILITY, WHETHER IN AN ACTION
+OF CONTRACT, TORT OR OTHERWISE, ARISING FROM, OUT OF OR IN CONNECTION
+WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE SOFTWARE.
+
+**************************************************************************/
+
+/*
+ * Authors:
+ *   Keith Whitwell <keith@tungstengraphics.com>
+ */
+
+#ifndef __r300_TEX_H__
+#define __r300_TEX_H__
+
+#ifdef GLX_DIRECT_RENDERING
+
+extern void r300UpdateTextureState(GLcontext * ctx);
+
+extern int r300UploadTexImages(r300ContextPtr rmesa, r300TexObjPtr t,
+			       GLuint face);
+
+extern void r300DestroyTexObj(r300ContextPtr rmesa, r300TexObjPtr t);
+
+extern void r300InitTextureFuncs(struct dd_function_table *functions);
+
+extern void r300UpdateTextureState(GLcontext * ctx);
+
+#endif
+#endif				/* __r300_TEX_H__ */

src/mesa/drivers/dri/r300/r300_texmem.c

@@ -0,0 +1,490 @@
+/* $XFree86: xc/lib/GL/mesa/src/drv/r300/r300_texmem.c,v 1.5 2002/12/17 00:32:56 dawes Exp $ */
+/**************************************************************************
+
+Copyright (C) Tungsten Graphics 2002.  All Rights Reserved.
+The Weather Channel, Inc. funded Tungsten Graphics to develop the
+initial release of the Radeon 8500 driver under the XFree86
+license. This notice must be preserved.
+
+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 <martin@valinux.com>
+ *   Gareth Hughes <gareth@valinux.com>
+ *
+ */
+
+#include <errno.h>
+
+#include "glheader.h"
+#include "imports.h"
+#include "context.h"
+#include "colormac.h"
+#include "macros.h"
+#include "simple_list.h"
+#include "radeon_reg.h"		/* gets definition for usleep */
+#include "r300_context.h"
+#include "r300_state.h"
+#include "radeon_ioctl.h"
+/*
+#include "r300_swtcl.h"
+*/
+#include "r300_tex.h"
+
+#include <unistd.h>		/* for usleep() */
+
+/**
+ * Destroy any device-dependent state associated with the texture.  This may
+ * include NULLing out hardware state that points to the texture.
+ */
+void r300DestroyTexObj(r300ContextPtr rmesa, r300TexObjPtr t)
+{
+	if (RADEON_DEBUG & DEBUG_TEXTURE) {
+		fprintf(stderr, "%s( %p, %p )\n", __FUNCTION__,
+			(void *)t, (void *)t->base.tObj);
+	}
+
+	if (rmesa != NULL) {
+		unsigned i;
+
+		for (i = 0; i < rmesa->radeon.glCtx->Const.MaxTextureUnits; i++) {
+			if (t == rmesa->state.texture.unit[i].texobj) {
+				rmesa->state.texture.unit[i].texobj = NULL;
+				/* This code below is meant to shorten state
+				   pushed to the hardware by not programming 
+				   unneeded units.
+				   
+				   This does not appear to be worthwhile on R300 */
+				#if 0
+				remove_from_list(&rmesa->hw.tex[i]);
+				make_empty_list(&rmesa->hw.tex[i]);
+				remove_from_list(&rmesa->hw.cube[i]);
+				make_empty_list(&rmesa->hw.cube[i]);
+				#endif
+			}
+		}
+	}
+}
+
+/* ------------------------------------------------------------
+ * Texture image conversions
+ */
+
+static void r300UploadGARTClientSubImage(r300ContextPtr rmesa,
+					 r300TexObjPtr t,
+					 struct gl_texture_image *texImage,
+					 GLint hwlevel,
+					 GLint x, GLint y,
+					 GLint width, GLint height)
+{
+	const struct gl_texture_format *texFormat = texImage->TexFormat;
+	GLuint srcPitch, dstPitch;
+	int blit_format;
+	int srcOffset;
+
+	/*
+	 * XXX it appears that we always upload the full image, not a subimage.
+	 * I.e. x==0, y==0, width=texWidth, height=texWidth.  If this is ever
+	 * changed, the src pitch will have to change.
+	 */
+	switch (texFormat->TexelBytes) {
+	case 1:
+		blit_format = R200_CP_COLOR_FORMAT_CI8;
+		srcPitch = t->image[0][0].width * texFormat->TexelBytes;
+		dstPitch = t->image[0][0].width * texFormat->TexelBytes;
+		break;
+	case 2:
+		blit_format = R200_CP_COLOR_FORMAT_RGB565;
+		srcPitch = t->image[0][0].width * texFormat->TexelBytes;
+		dstPitch = t->image[0][0].width * texFormat->TexelBytes;
+		break;
+	case 4:
+		blit_format = R200_CP_COLOR_FORMAT_ARGB8888;
+		srcPitch = t->image[0][0].width * texFormat->TexelBytes;
+		dstPitch = t->image[0][0].width * texFormat->TexelBytes;
+		break;
+	default:
+		return;
+	}
+
+	t->image[0][hwlevel].data = texImage->Data;
+	srcOffset = r300GartOffsetFromVirtual(rmesa, texImage->Data);
+
+	assert(srcOffset != ~0);
+
+	/* Don't currently need to cope with small pitches?
+	 */
+	width = texImage->Width;
+	height = texImage->Height;
+
+	r300EmitWait(rmesa, RADEON_WAIT_3D);
+
+	r300EmitBlit(rmesa, blit_format,
+		     srcPitch,
+		     srcOffset,
+		     dstPitch,
+		     t->bufAddr,
+		     x,
+		     y,
+		     t->image[0][hwlevel].x + x,
+		     t->image[0][hwlevel].y + y, width, height);
+
+	r300EmitWait(rmesa, RADEON_WAIT_2D);
+}
+
+static void r300UploadRectSubImage(r300ContextPtr rmesa,
+				   r300TexObjPtr 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 = R200_CP_COLOR_FORMAT_CI8;
+		break;
+	case 2:
+		blit_format = R200_CP_COLOR_FORMAT_RGB565;
+		break;
+	case 4:
+		blit_format = R200_CP_COLOR_FORMAT_ARGB8888;
+		break;
+	default:
+		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->pitch + 32;
+
+	if (rmesa->prefer_gart_client_texturing && texImage->IsClientData) {
+		/* In this case, could also use GART texturing.  This is
+		 * currently disabled, but has been tested & works.
+		 */
+		t->offset =
+		    r300GartOffsetFromVirtual(rmesa, texImage->Data);
+		t->pitch =
+		    texImage->RowStride * texFormat->TexelBytes - 32;
+
+		if (RADEON_DEBUG & DEBUG_TEXTURE)
+			fprintf(stderr,
+				"Using GART texturing for rectangular client texture\n");
+
+		/* Release FB memory allocated for this image:
+		 */
+		/* FIXME This may not be correct as driSwapOutTextureObject sets
+		 * FIXME dirty_images.  It may be fine, though.
+		 */
+		if (t->base.memBlock) {
+			driSwapOutTextureObject((driTextureObject *) t);
+		}
+	} else if (texImage->IsClientData) {
+		/* Data already in GART memory, with usable pitch.
+		 */
+		GLuint srcPitch;
+		srcPitch = texImage->RowStride * texFormat->TexelBytes;
+		r300EmitBlit(rmesa,
+			     blit_format,
+			     srcPitch,
+			     r300GartOffsetFromVirtual(rmesa, texImage->Data),
+			     dstPitch, t->bufAddr, 0, 0, 0, 0, width, height);
+	} else {
+		/* Data not in GART memory, or bad pitch.
+		 */
+		for (done = 0; done < height;) {
+			struct r300_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));
+			r300AllocDmaRegion(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 + region.start, tex,
+				       lines * src_pitch);
+			} else {
+				char *buf = region.address + region.start;
+				int i;
+				for (i = 0; i < lines; i++) {
+					memcpy(buf, tex, src_pitch);
+					buf += dstPitch;
+					tex += src_pitch;
+				}
+			}
+
+			r300EmitWait(rmesa, RADEON_WAIT_3D);
+
+			/* Blit to framebuffer
+			 */
+			r300EmitBlit(rmesa,
+				     blit_format,
+				     dstPitch, GET_START(&region),
+				     dstPitch, t->bufAddr,
+				     0, 0, 0, done, width, lines);
+
+			r300EmitWait(rmesa, RADEON_WAIT_2D);
+
+			r300ReleaseDmaRegion(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(r300ContextPtr rmesa, r300TexObjPtr 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;
+	drm_radeon_texture_t tex;
+	drm_radeon_tex_image_t 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__, (void *)t, (void *)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;
+	}
+
+	texImage = t->base.tObj->Image[face][level];
+
+	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__);
+		r300UploadRectSubImage(rmesa, t, texImage, x, y, width, height);
+		return;
+	} else if (texImage->IsClientData) {
+		if (RADEON_DEBUG & DEBUG_TEXTURE)
+			fprintf(stderr,
+				"%s: image data is in GART client storage\n",
+				__FUNCTION__);
+		r300UploadGARTClientSubImage(rmesa, t, texImage, hwlevel, x, y,
+					     width, height);
+		return;
+	} else if (RADEON_DEBUG & DEBUG_TEXTURE)
+		fprintf(stderr, "%s: image data is in normal memory\n",
+			__FUNCTION__);
+
+	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 / drm_radeon_texture_t 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 = R200_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(tmp));
+
+	LOCK_HARDWARE(&rmesa->radeon);
+	do {
+		ret = drmCommandWriteRead(rmesa->radeon.dri.fd, DRM_RADEON_TEXTURE,
+					  &tex, sizeof(drm_radeon_texture_t));
+		if (ret) {
+			if (RADEON_DEBUG & DEBUG_IOCTL)
+				fprintf(stderr,
+					"DRM_RADEON_TEXTURE:  again!\n");
+			usleep(1);
+		}
+	} while (ret && errno == EAGAIN);
+
+	UNLOCK_HARDWARE(&rmesa->radeon);
+
+	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 r300UploadTexImages(r300ContextPtr rmesa, r300TexObjPtr 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__,
+			(void *)rmesa->radeon.glCtx, (void *)t->base.tObj,
+			t->base.totalSize, t->base.firstLevel,
+			t->base.lastLevel);
+	}
+
+	if (!t || t->base.totalSize == 0)
+		return 0;
+
+	if (RADEON_DEBUG & DEBUG_SYNC) {
+		fprintf(stderr, "%s: Syncing\n", __FUNCTION__);
+		radeonFinish(rmesa->radeon.glCtx);
+	}
+
+	LOCK_HARDWARE(&rmesa->radeon);
+
+	if (t->base.memBlock == NULL) {
+		int heap;
+
+		heap = driAllocateTexture(rmesa->texture_heaps, rmesa->nr_heaps,
+					  (driTextureObject *) t);
+		if (heap == -1) {
+			UNLOCK_HARDWARE(&rmesa->radeon);
+			return -1;
+		}
+
+		/* Set the base offset of the texture image */
+		t->bufAddr = rmesa->radeon.radeonScreen->texOffset[heap]
+		    + t->base.memBlock->ofs;
+		t->offset = 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->radeon);
+
+	/* 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;
+	}
+
+	if (RADEON_DEBUG & DEBUG_SYNC) {
+		fprintf(stderr, "%s: Syncing\n", __FUNCTION__);
+		radeonFinish(rmesa->radeon.glCtx);
+	}
+
+	return 0;
+}

src/mesa/drivers/dri/s3v/s3v_context.c

@@ -16,7 +16,7 @@
 #include "matrix.h"
 #include "extensions.h"
 #if defined(USE_X86_ASM)
-#include "X86/common_x86_asm.h"
+#include "x86/common_x86_asm.h"
 #endif
 #include "simple_list.h"
 #include "mm.h"

src/mesa/drivers/dri/s3v/s3v_dd.c

@@ -6,7 +6,7 @@
 #include "s3v_vb.h"
 #include "s3v_lock.h"
 #if defined(USE_X86_ASM)
-#include "X86/common_x86_asm.h"
+#include "x86/common_x86_asm.h"
 #endif
 
 #include "context.h"

src/mesa/drivers/dri/savage/Makefile

@@ -25,8 +25,7 @@ DRIVER_SOURCES = \
 	savagetex.c \
 	savagetris.c \
 	savageioctl.c \
-	savagespan.c \
-	savagedma.c
+	savagespan.c
 
 C_SOURCES = \
 	$(COMMON_SOURCES) \

src/mesa/drivers/dri/savage/savage_bci.h

@@ -341,8 +341,8 @@ typedef enum
     TPS_256
 } TexPaletteSize;
 
-    #define MAX_MIPMAP_LOD_BIAS             255
-    #define MIN_MIPMAP_LOD_BIAS             -255
+#define MAX_MIPMAP_LOD_BIAS 255
+#define MIN_MIPMAP_LOD_BIAS -255
 
 typedef enum
 {
@@ -627,107 +627,6 @@ typedef union {
 #define SAVAGE_HW_NO_UV1        ((1<<6) | (1<<7))
 #define SAVAGE_HW_SKIPFLAGS     0x000000ff
 
-#define SAVAGE_HW_TRIANGLE_TYPE     (3UL<<25)
-#define SAVAGE_HW_TRIANGLE_CONT     (1UL<<24)
-#define SAVAGE_HW_TRIANGLE_LIST     (0<<25) 
-#define SAVAGE_HW_TRIANGLE_STRIP    (1<<25) 
-#define SAVAGE_HW_TRIANGLE_FAN      (2<<25) 
-#define SAVAGE_HW_QUAD              (3<<25) 
-
-#define __HW_TEXTURE_CHANGED       0x00002FE
-#define __HW_HAS_SCISSORS_CHANGED  0x00001800
-#define __HW_ALL_CHANGED           0x1FFFFFF           
-/*Frank 2001/11/14 Wait commands*/
-#define WAIT_3D_IDLE    0xC0010000
-#define WAIT_3D_2D_IDLE 0xC0030000
-
-#define SET_REGISTER(index, count) \
-    ((CMD_SetRegister << 27) | (0x6000000) | ((count) << 16) | (index))
-
-/*frank 2001/11/20 */
-#define MAXLOOP 0xFFFFFF
-/*#define MAXFIFO 0x7F00*/
-#define MAXFIFO 0x1FF00
-
-/* get eventtag from shadow status */
-/* here we use eventTag1 because eventTag0 is used by HWXvMC*/
-#define GET_EVENTTAG \
-    (((*(volatile GLuint *)(imesa->MMIO_BASE+0x48c04)) & 0xffff0000L)>>16)
-
-#define SHADOW_WAIT(imesa ) do \
-{ \
-    int loop=0; \
-    imesa->shadowCounter = (imesa->shadowCounter + 1) & 0xffff;\
-    if(imesa->shadowCounter == 0)\
-      imesa->shadowCounter = MAX_SHADOWCOUNTER;\
-    *(volatile GLuint *)imesa->BCIBase = imesa->shadowCounter | 0x98400000L;\
-    while(\
-	  (GET_EVENTTAG) != imesa->shadowCounter  &&\
-	  (loop++ < MAXLOOP));\
-}while(0);
-
-#define SHADOW_WAIT_IDLE(imesa ) do \
-{ \
-    int loop=0; \
-    imesa->shadowCounter = (imesa->shadowCounter + 1) & 0xffff;\
-    if(imesa->shadowCounter == 0)\
-      imesa->shadowCounter = MAX_SHADOWCOUNTER;\
-/*    *(volatile GLuint *)imesa->BCIBase = WAIT_3D_IDLE;\*/\
-    *(volatile GLuint *)imesa->BCIBase = imesa->shadowCounter | 0x98400000L;\
-    while ( \
-    (GET_EVENTTAG) != imesa->shadowCounter && \
-    (loop++ < MAXLOOP)); \
-}while(0);
-
-#if 0
-#define ALT_STATUS_WORD0 (* (volatile GLuint *)(imesa->MMIO_BASE+0x48c60))
-
-#define PAGE_PENDING(result) do{\
-result=((ALT_STATUS_WORD0 & 0x08000000)?GL_TRUE:GL_FALSE);\
-}while(0)
-
-#define WAIT_FOR_FIFO(count) do{\
-int loop = 0; \
-int slots = MAXFIFO-count; \
-while(((ALT_STATUS_WORD0 &0x001fffff)>slots)&&(loop++<MAXLOOP)); \
-}while(0)
-
-
-#define WAIT_IDLE_EMPTY do{\
-int loop = 0; \
- if (/*imesa->shadowStatus*/0)\
-   {\
-     SHADOW_WAIT_IDLE(imesa);\
-   }\
- else\
-   { \
-     while(((ALT_STATUS_WORD0 &0x00ffffff)!=0x00E00000L)&&(loop++<MAXLOOP));\
-   }\
-}while(0)
-
-#define WAIT_IDLE do{\
-int loop = 0; \
-if (imesa->shadowStatus)\
- while((((*imesa->shadowPointer) & 0x0E000000L)!=0x0E000000L)&&(loop++<MAXLOOP));\
-else\
-while(((ALT_STATUS_WORD0 &0x00E00000)!=0x00E00000L)&&(loop++<MAXLOOP)); \
-}while(0)
-#endif /* 0 */
-
-#define SAVAGE_DRAW_PRIMITIVE(count, typeandvertexSkip, isCont)  \
-        ( ((count)<<16) | (typeandvertexSkip) | (isCont | (1<<31)));
-
-static __inline volatile GLuint * SAVAGE_GET_BCI_POINTER(savageContextPtr imesa, GLuint count)   
-{ 
-  WAIT_FOR_FIFO(count);                                
-  return (volatile GLuint *)(imesa->BCIBase);      
-}
-
-/*use this set bci cmd now!*/
-#define WRITE_CMD(buf,cmd,type) do {\
-    *((type*)buf)=cmd;\
-    buf++;\
-  }while(0)
 #endif
 
 

src/mesa/drivers/dri/savage/savage_init.h

@@ -41,11 +41,6 @@ typedef struct {
 } savageRegion, *savageRegionPtr;
 
 typedef struct {
-   savageRegion front;
-   savageRegion back;
-   savageRegion depth;
-   savageRegion aperture;
-
    int chipset;
    int width;
    int height;
@@ -53,31 +48,34 @@ typedef struct {
 
    int cpp;			/* for front and back buffers */
    int zpp;
+
+   int agpMode;
+
+   unsigned int bufferSize;
+
 #if 0 
    int bitsPerPixel;
 #endif
    unsigned int frontFormat;
    unsigned int frontOffset;
-   unsigned int frontPitch;
-   unsigned int frontBitmapDesc;
-
    unsigned int backOffset;
-   unsigned int backBitmapDesc;
    unsigned int depthOffset;
-   unsigned int depthBitmapDesc;
 
-   unsigned int backPitch;
-   unsigned int backPitchBits;
+   unsigned int aperturePitch;
 
    unsigned int textureOffset[SAVAGE_NR_TEX_HEAPS];
    unsigned int textureSize[SAVAGE_NR_TEX_HEAPS];
    unsigned int logTextureGranularity[SAVAGE_NR_TEX_HEAPS];
    drmAddress texVirtual[SAVAGE_NR_TEX_HEAPS];
   
-  __DRIscreenPrivate *driScrnPriv;
-  drmBufMapPtr  bufs;
-  int use_copy_buf;
-  unsigned int sarea_priv_offset;
+   __DRIscreenPrivate *driScrnPriv;
+
+   savageRegion aperture;
+   savageRegion agpTextures;
+
+   drmBufMapPtr bufs;
+
+   unsigned int sarea_priv_offset;
 
    /* Configuration cache with default values for all contexts */
    driOptionCache optionCache;
@@ -87,7 +85,6 @@ typedef struct {
 #include "savagecontext.h"
 
 extern void savageGetLock( savageContextPtr imesa, GLuint flags );
-extern void savageEmitHwStateLocked( savageContextPtr imesa );
 extern void savageEmitScissorValues( savageContextPtr imesa, int box_nr, int emit );
 extern void savageEmitDrawingRectangle( savageContextPtr imesa );
 extern void savageXMesaSetBackClipRects( savageContextPtr imesa );

src/mesa/drivers/dri/savage/savage_xmesa.c

@@ -55,8 +55,6 @@
 
 #include "savage_dri.h"
 
-#include "savagedma.h"
-
 #include "xmlpool.h"
 
 /* Configuration
@@ -66,6 +64,7 @@ DRI_CONF_BEGIN
     DRI_CONF_SECTION_QUALITY
         DRI_CONF_TEXTURE_DEPTH(DRI_CONF_TEXTURE_DEPTH_FB)
         DRI_CONF_COLOR_REDUCTION(DRI_CONF_COLOR_REDUCTION_DITHER)
+        DRI_CONF_FLOAT_DEPTH(false)
     DRI_CONF_SECTION_END
     DRI_CONF_SECTION_PERFORMANCE
         DRI_CONF_MAX_TEXTURE_UNITS(2,1,2)
@@ -74,7 +73,7 @@ DRI_CONF_BEGIN
         DRI_CONF_NO_RAST(false)
     DRI_CONF_SECTION_END
 DRI_CONF_END;
-static const GLuint __driNConfigOptions = 4;
+static const GLuint __driNConfigOptions = 5;
 
 #ifdef USE_NEW_INTERFACE
 static PFNGLXCREATECONTEXTMODES create_context_modes = NULL;
@@ -85,6 +84,8 @@ static const struct dri_debug_control debug_control[] =
     { "fall",  DEBUG_FALLBACKS },
     { "api",   DEBUG_VERBOSE_API },
     { "lru",   DEBUG_VERBOSE_LRU },
+    { "verb",  DEBUG_VERBOSE_MSG },
+    { "dma",   DEBUG_DMA },
     { NULL,    0 }
 };
 #ifndef SAVAGE_DEBUG
@@ -131,7 +132,6 @@ savageInitDriver(__DRIscreenPrivate *sPriv)
   savageScreenPrivate *savageScreen;
   SAVAGEDRIPtr         gDRIPriv = (SAVAGEDRIPtr)sPriv->pDevPriv;
 
-
    /* Allocate the private area */
    savageScreen = (savageScreenPrivate *)Xmalloc(sizeof(savageScreenPrivate));
    if (!savageScreen)
@@ -146,23 +146,19 @@ savageInitDriver(__DRIscreenPrivate *sPriv)
    savageScreen->mem=gDRIPriv->mem;
    savageScreen->cpp=gDRIPriv->cpp;
    savageScreen->zpp=gDRIPriv->zpp;
-   savageScreen->frontPitch=gDRIPriv->frontPitch;
-   savageScreen->frontOffset=gDRIPriv->frontOffset;
-   savageScreen->frontBitmapDesc = gDRIPriv->frontBitmapDesc;
-   
+
+   savageScreen->agpMode=gDRIPriv->agpMode;
+
+   savageScreen->bufferSize=gDRIPriv->bufferSize;
+
    if (gDRIPriv->cpp == 4) 
        savageScreen->frontFormat = DV_PF_8888;
    else
        savageScreen->frontFormat = DV_PF_565;
-
+   savageScreen->frontOffset=gDRIPriv->frontOffset;
    savageScreen->backOffset = gDRIPriv->backOffset; 
-   savageScreen->backBitmapDesc = gDRIPriv->backBitmapDesc; 
    savageScreen->depthOffset=gDRIPriv->depthOffset;
-   savageScreen->depthBitmapDesc = gDRIPriv->depthBitmapDesc; 
-#if 0   
-   savageScreen->backPitch = gDRIPriv->auxPitch;
-   savageScreen->backPitchBits = gDRIPriv->auxPitchBits;
-#endif   
+
    savageScreen->textureOffset[SAVAGE_CARD_HEAP] = 
                                    gDRIPriv->textureOffset;
    savageScreen->textureSize[SAVAGE_CARD_HEAP] = 
@@ -171,57 +167,32 @@ savageInitDriver(__DRIscreenPrivate *sPriv)
                                    gDRIPriv->logTextureGranularity;
 
    savageScreen->textureOffset[SAVAGE_AGP_HEAP] = 
-                                   gDRIPriv->agpTextures.handle;
+                                   gDRIPriv->agpTextureHandle;
    savageScreen->textureSize[SAVAGE_AGP_HEAP] = 
-                                   gDRIPriv->agpTextures.size;
+                                   gDRIPriv->agpTextureSize;
    savageScreen->logTextureGranularity[SAVAGE_AGP_HEAP] =
                                    gDRIPriv->logAgpTextureGranularity;
-   
-   savageScreen->back.handle = gDRIPriv->backbuffer;
-   savageScreen->back.size = gDRIPriv->backbufferSize;
-   savageScreen->back.map = 
-       (drmAddress)(((unsigned int)sPriv->pFB)+gDRIPriv->backOffset);
-   
-   savageScreen->depth.handle = gDRIPriv->depthbuffer;
-   savageScreen->depth.size = gDRIPriv->depthbufferSize;
 
-   savageScreen->depth.map = 
-              (drmAddress)(((unsigned int)sPriv->pFB)+gDRIPriv->depthOffset);
-   
-   savageScreen->sarea_priv_offset = gDRIPriv->sarea_priv_offset;
+   savageScreen->agpTextures.handle = gDRIPriv->agpTextureHandle;
+   savageScreen->agpTextures.size   = gDRIPriv->agpTextureSize;
+   if (drmMap(sPriv->fd, 
+	      savageScreen->agpTextures.handle,
+	      savageScreen->agpTextures.size,
+	      (drmAddress *)&(savageScreen->agpTextures.map)) != 0) 
+   {
+      Xfree(savageScreen);
+      sPriv->private = NULL;
+      return GL_FALSE;
+   }
 
    savageScreen->texVirtual[SAVAGE_CARD_HEAP] = 
              (drmAddress)(((unsigned int)sPriv->pFB)+gDRIPriv->textureOffset);
-
-   if (drmMap(sPriv->fd, 
-	      gDRIPriv->registers.handle, 
-	      gDRIPriv->registers.size, 
-	      (drmAddress *)&(gDRIPriv->registers.map)) != 0) 
-   {
-      Xfree(savageScreen);
-      sPriv->private = NULL;
-      return GL_FALSE;
-   }
-   
-   if (drmMap(sPriv->fd, 
-	      gDRIPriv->agpTextures.handle, 
-	      gDRIPriv->agpTextures.size, 
-	      (drmAddress *)&(gDRIPriv->agpTextures.map)) != 0) 
-   {
-      Xfree(savageScreen);
-      sPriv->private = NULL;
-      return GL_FALSE;
-   }
-
-/* agp texture*/
    savageScreen->texVirtual[SAVAGE_AGP_HEAP] = 
-                        (drmAddress)(gDRIPriv->agpTextures.map);
+                        (drmAddress)(savageScreen->agpTextures.map);
 
-   gDRIPriv->BCIcmdBuf.map = (drmAddress *)
-                           ((unsigned int)gDRIPriv->registers.map+0x00010000);
-
-   savageScreen->aperture.handle = gDRIPriv->aperture.handle;
-   savageScreen->aperture.size   = gDRIPriv->aperture.size;
+   savageScreen->aperture.handle = gDRIPriv->apertureHandle;
+   savageScreen->aperture.size   = gDRIPriv->apertureSize;
+   savageScreen->aperturePitch   = gDRIPriv->aperturePitch;
    if (drmMap(sPriv->fd, 
 	      savageScreen->aperture.handle, 
 	      savageScreen->aperture.size, 
@@ -230,8 +201,12 @@ savageInitDriver(__DRIscreenPrivate *sPriv)
       Xfree(savageScreen);
       sPriv->private = NULL;
       return GL_FALSE;
-   } 
-      
+   }
+
+   savageScreen->bufs = drmMapBufs(sPriv->fd);
+
+   savageScreen->sarea_priv_offset = gDRIPriv->sarea_priv_offset;
+
    /* parse information in __driConfigOptions */
    driParseOptionInfo (&savageScreen->optionCache,
 		       __driConfigOptions, __driNConfigOptions);
@@ -251,6 +226,8 @@ savageDestroyScreen(__DRIscreenPrivate *sPriv)
 {
    savageScreenPrivate *savageScreen = (savageScreenPrivate *)sPriv->private;
 
+   drmUnmapBufs(savageScreen->bufs);
+
    /* free all option information */
    driDestroyOptionInfo (&savageScreen->optionCache);
 
@@ -295,7 +272,6 @@ savageCreateContext( const __GLcontextModes *mesaVis,
    savageContextPtr imesa;
    __DRIscreenPrivate *sPriv = driContextPriv->driScreenPriv;
    struct dd_function_table functions;
-   SAVAGEDRIPtr         gDRIPriv = (SAVAGEDRIPtr)sPriv->pDevPriv;
    savageScreenPrivate *savageScreen = (savageScreenPrivate *)sPriv->private;
    drm_savage_sarea_t *saPriv=(drm_savage_sarea_t *)(((char*)sPriv->pSAREA)+
 						 savageScreen->sarea_priv_offset);
@@ -323,10 +299,19 @@ savageCreateContext( const __GLcontextModes *mesaVis,
    }
    driContextPriv->driverPrivate = imesa;
 
+   imesa->cmdBuf.size = SAVAGE_CMDBUF_SIZE;
+   imesa->cmdBuf.base = imesa->cmdBuf.write =
+       malloc(SAVAGE_CMDBUF_SIZE * sizeof(drm_savage_cmd_header_t));
+   if (!imesa->cmdBuf.base)
+       return GL_FALSE;
+
    /* Parse configuration files */
    driParseConfigFiles (&imesa->optionCache, &savageScreen->optionCache,
                         sPriv->myNum, "savage");
 
+   imesa->float_depth = driQueryOptionb(&imesa->optionCache, "float_depth") &&
+       savageScreen->chipset >= S3_SAVAGE4;
+   imesa->no_rast = driQueryOptionb(&imesa->optionCache, "no_rast");
    imesa->texture_depth = driQueryOptioni (&imesa->optionCache,
 					   "texture_depth");
    if (imesa->texture_depth == DRI_CONF_TEXTURE_DEPTH_FB)
@@ -398,26 +383,14 @@ savageCreateContext( const __GLcontextModes *mesaVis,
    
    /* DMA buffer */
 
-   /*The shadow pointer*/
-   imesa->shadowPointer = 
-     (volatile GLuint *)((((GLuint)(&saPriv->shadow_status)) + 31) & 0xffffffe0L) ;
-   /* here we use eventTag1 because eventTag0 is used by HWXvMC*/
-   imesa->eventTag1 = (volatile GLuint *)(imesa->shadowPointer + 6);
-   /*   imesa->eventTag1=(volatile GLuint *)(imesa->MMIO_BASE+0x48c04);*/
-   imesa->shadowCounter = MAX_SHADOWCOUNTER;
-   imesa->shadowStatus = GL_TRUE;/*Will judge by 2d message */
-
-   imesa->MMIO_BASE = (GLuint)gDRIPriv->registers.map;
-   imesa->BCIBase= (GLuint)gDRIPriv->BCIcmdBuf.map;
    for(i=0;i<5;i++)
    {
        imesa->apertureBase[i] = ((GLuint)savageScreen->aperture.map + 
                                  0x01000000 * i );
    }
    
-   imesa->aperturePitch = gDRIPriv->aperturePitch;
-   
-   
+   imesa->aperturePitch = savageScreen->aperturePitch;
+
    /* change texHeap initialize to support two kind of texture heap*/
    /* here is some parts of initialization, others in InitDriver() */
     
@@ -438,9 +411,19 @@ savageCreateContext( const __GLcontextModes *mesaVis,
 
    imesa->hw_stencil = mesaVis->stencilBits && mesaVis->depthBits == 24;
    imesa->depth_scale = (imesa->savageScreen->zpp == 2) ?
-       (1.0F/0x10000):(1.0F/0x1000000);
+       (1.0F/0xffff):(1.0F/0xffffff);
 
-   imesa->vertex_dma_buffer = NULL;
+   imesa->bufferSize = savageScreen->bufferSize;
+   imesa->dmaVtxBuf.total = 0;
+   imesa->dmaVtxBuf.used = 0;
+   imesa->dmaVtxBuf.flushed = 0;
+
+   imesa->clientVtxBuf.total = 16384;
+   imesa->clientVtxBuf.used = 0;
+   imesa->clientVtxBuf.flushed = 0;
+   imesa->clientVtxBuf.buf = (u_int32_t *)malloc(16384*4);
+
+   imesa->vtxBuf = &imesa->clientVtxBuf;
 
    /* Uninitialized vertex format. Force setting the vertex state in
     * savageRenderStart.
@@ -450,6 +433,7 @@ savageCreateContext( const __GLcontextModes *mesaVis,
    /* Utah stuff
     */
    imesa->new_state = ~0;
+   imesa->new_gl_state = ~0;
    imesa->RenderIndex = ~0;
    imesa->dirty = ~0;
    imesa->lostContext = GL_TRUE;
@@ -483,8 +467,6 @@ savageCreateContext( const __GLcontextModes *mesaVis,
 
    ctx->DriverCtx = (void *) imesa;
    imesa->glCtx = ctx;
-   if (savageDMAInit(imesa) == GL_FALSE)
-       return GL_FALSE;  
 
 #ifndef SAVAGE_DEBUG
    SAVAGE_DEBUG = driParseDebugString( getenv( "SAVAGE_DEBUG" ),
@@ -501,9 +483,6 @@ savageCreateContext( const __GLcontextModes *mesaVis,
 
    savageDDInitState( imesa );
 
-   if (driQueryOptionb(&imesa->optionCache, "no_rast"))
-       FALLBACK(ctx, SAVAGE_FALLBACK_NORAST, GL_TRUE);
-
    driContextPriv->driverPrivate = (void *) imesa;
 
    return GL_TRUE;
@@ -529,8 +508,10 @@ savageDestroyContext(__DRIcontextPrivate *driContextPriv)
       }
       foreach_s (t, next_t, &(imesa->SwappedOut))
 	 savageDestroyTexObj(imesa, t);
-      /*free the dma buffer*/
-      savageDMAClose(imesa);
+
+      free(imesa->cmdBuf.base);
+      free(imesa->clientVtxBuf.buf);
+
       _swsetup_DestroyContext(imesa->glCtx );
       _tnl_DestroyContext( imesa->glCtx );
       _ac_DestroyContext( imesa->glCtx );
@@ -753,10 +734,13 @@ void savageGetLock( savageContextPtr imesa, GLuint flags )
     * more broken than usual.
     */
    if (sarea->ctxOwner != me) {
-      imesa->dirty |= (SAVAGE_UPLOAD_CTX |
-		       SAVAGE_UPLOAD_CLIPRECTS |
+      imesa->dirty |= (SAVAGE_UPLOAD_LOCAL |
+		       SAVAGE_UPLOAD_GLOBAL |
+		       SAVAGE_UPLOAD_FOGTBL |
 		       SAVAGE_UPLOAD_TEX0 |
-		       SAVAGE_UPLOAD_TEX1);
+		       SAVAGE_UPLOAD_TEX1 |
+		       SAVAGE_UPLOAD_TEXGLOBAL |
+		       SAVAGE_UPLOAD_CLIPRECTS);
       imesa->lostContext = GL_TRUE;
       sarea->ctxOwner = me;
    }
@@ -794,8 +778,6 @@ void savageGetLock( savageContextPtr imesa, GLuint flags )
 	      savageResetGlobalLRU( imesa , heap );
            }
 
-           imesa->dirty |= SAVAGE_UPLOAD_TEX0IMAGE;
-           imesa->dirty |= SAVAGE_UPLOAD_TEX1IMAGE;
            imesa->texAge[heap] = sarea->texAge[heap];
        }
    } /* end of for loop */ 
@@ -948,10 +930,9 @@ void * __driCreateNewScreen( __DRInativeDisplay *dpy, int scrn, __DRIscreen *psc
 			     
 {
    __DRIscreenPrivate *psp;
-   static const __DRIversion ddx_expected = { 1, 0, 0 };
+   static const __DRIversion ddx_expected = { 2, 0, 0 };
    static const __DRIversion dri_expected = { 4, 0, 0 };
-   static const __DRIversion drm_expected = { 1, 0, 0 };
-
+   static const __DRIversion drm_expected = { 2, 0, 0 };
 
    if ( ! driCheckDriDdxDrmVersions2( "Savage",
 				      dri_version, & dri_expected,

src/mesa/drivers/dri/savage/savagecontext.h

@@ -42,7 +42,6 @@ typedef struct savage_texture_object_t *savageTextureObjectPtr;
 #include "tnl/t_vertex.h"
 
 #include "savagetex.h"
-#include "savagedma.h"
 
 #include "xmlconfig.h"
 
@@ -69,10 +68,26 @@ typedef struct savage_texture_object_t *savageTextureObjectPtr;
 #define SAVAGE_NEW_TEXTURE 0x1
 #define SAVAGE_NEW_CULL    0x2
 
+/* What needs to be changed for the current vertex dma buffer?
+ * This will go away!
+ */
+#define SAVAGE_UPLOAD_LOCAL	0x1  /* DrawLocalCtrl (S4) or 
+					DrawCtrl and ZBufCtrl (S3D) */
+#define SAVAGE_UPLOAD_TEX0	0x2  /* texture unit 0 */
+#define SAVAGE_UPLOAD_TEX1	0x4  /* texture unit 1 (S4 only) */
+#define SAVAGE_UPLOAD_FOGTBL	0x8  /* fog table */
+#define SAVAGE_UPLOAD_GLOBAL	0x10 /* most global regs */
+#define SAVAGE_UPLOAD_TEXGLOBAL 0x20 /* TexBlendColor (S4 only) */
+#define SAVAGE_UPLOAD_CLIPRECTS 0x1000 /* FIXME: get rid of this */
 
 /*define the max numer of vertex in vertex buf*/
 #define SAVAGE_MAX_VERTEXS 0x10000
 
+/* Don't make it too big. We don't want to buffer up a whole frame
+ * that would force the application to wait later. */
+#define SAVAGE_CMDBUF_SIZE 1024
+#define SAVAGE_MAX_VERTS_PENDING 1024
+
 /* Use the templated vertex formats:
  */
 #define TAG(x) savage##x
@@ -113,6 +128,18 @@ typedef void (*savage_point_func)( savageContextPtr, savageVertex * );
 			imesa->savageScreen->deviceID == CHIP_S3TRISTAR64CDDR )
 
 
+struct savage_vtxbuf_t {
+    GLuint total, used, flushed; /* in 32 bit units */
+    GLuint idx;		/* for DMA buffers */
+    u_int32_t *buf;
+};
+
+struct savage_cmdbuf_t {
+    GLuint size; /* size in qwords */
+    drm_savage_cmd_header_t *base;  /* initial state starts here */
+    drm_savage_cmd_header_t *start; /* drawing/state commands start here */
+    drm_savage_cmd_header_t *write; /* append stuff here */
+};
 
 
 struct savage_context_t {
@@ -143,11 +170,12 @@ struct savage_context_t {
     GLuint new_gl_state;
     GLboolean ptexHack;
 
-    GLuint BCIBase;  
-    GLuint MMIO_BASE;
+    /* Command buffer */
+    struct savage_cmdbuf_t cmdBuf;
 
-    /* DMA command buffer */
-    DMABuffer_t DMABuf;
+    /* Vertex buffers */
+    struct savage_vtxbuf_t dmaVtxBuf, clientVtxBuf;
+    struct savage_vtxbuf_t *vtxBuf;
 
     /* aperture base */
     GLuint apertureBase[5];
@@ -179,7 +207,8 @@ struct savage_context_t {
    GLenum raster_primitive;
    GLenum render_primitive;
 
-   GLuint DrawPrimitiveCmd;
+   GLuint skip;
+   GLubyte HwPrim;
    GLuint HwVertexSize;
 
    /* Fallback rasterization functions 
@@ -194,8 +223,9 @@ struct savage_context_t {
     GLuint ClearColor;
     GLfloat depth_scale;
     GLfloat hw_viewport[16];
-    /* DRI stuff */  
-    drmBufPtr  vertex_dma_buffer;
+    /* DRI stuff */
+    GLuint bufferSize;
+    GLuint vertsPending;
 
     GLframebuffer *glBuffer;
    
@@ -225,8 +255,8 @@ struct savage_context_t {
     GLuint backup_streamFIFO;
     GLuint NotFirstFrame;
    
+    GLboolean inSwap;
     GLuint lastSwap;
-    GLuint secondLastSwap;
     GLuint ctxAge;
     GLuint dirtyAge;
     GLuint any_contend;		/* throttle me harder */
@@ -235,7 +265,7 @@ struct savage_context_t {
     GLboolean scissorChanged;
     drm_clip_rect_t draw_rect;
     drm_clip_rect_t scissor_rect;
-    drm_clip_rect_t tmp_boxes[2][SAVAGE_NR_SAREA_CLIPRECTS];
+
     /*Texture aging and DMA based aging*/
     unsigned int texAge[SAVAGE_NR_TEX_HEAPS]; 
 
@@ -259,16 +289,12 @@ struct savage_context_t {
 
     GLboolean hw_stencil;
 
-    /*shadow pointer*/
-    volatile GLuint  *shadowPointer;
-    volatile GLuint *eventTag1;
-    GLuint shadowCounter;
-    GLboolean shadowStatus;
-
     /* Configuration cache
      */
     driOptionCache optionCache;
-    int texture_depth;
+    GLint texture_depth;
+    GLboolean no_rast;
+    GLboolean float_depth;
 };
 
 #define SAVAGE_CONTEXT(ctx) ((savageContextPtr)(ctx->DriverCtx))
@@ -283,6 +309,8 @@ extern int SAVAGE_DEBUG;
 #define DEBUG_FALLBACKS      0x001
 #define DEBUG_VERBOSE_API    0x002
 #define DEBUG_VERBOSE_LRU    0x004
+#define DEBUG_VERBOSE_MSG    0x008
+#define DEBUG_DMA            0x010
 
 #define TARGET_FRONT    0x0
 #define TARGET_BACK     0x1

src/mesa/drivers/dri/savage/savagedd.c

@@ -38,6 +38,10 @@
 #include "savagecontext.h"
 #include "extensions.h"
 
+#include "utils.h"
+
+
+#define DRIVER_DATE "20050101"
 
 /***************************************
  * Mesa's Driver Functions
@@ -46,11 +50,33 @@
 
 static const GLubyte *savageDDGetString( GLcontext *ctx, GLenum name )
 {
+   static char *cardNames[S3_LAST] = {
+       "Unknown",
+       "Savage3D",
+       "Savage/MX/IX",
+       "Savage4",
+       "ProSavage",
+       "Twister",
+       "ProSavageDDR",
+       "SuperSavage",
+       "Savage2000"
+   };
+   static char buffer[128];
+   savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
+   savageScreenPrivate *screen = imesa->savageScreen;
+   enum S3CHIPTAGS chipset = screen->chipset;
+   unsigned offset;
+
+   if (chipset < S3_SAVAGE3D || chipset >= S3_LAST)
+      chipset = S3_UNKNOWN; /* should not happen */
+
    switch (name) {
    case GL_VENDOR:
       return (GLubyte *)"S3 Graphics Inc.";
    case GL_RENDERER:
-      return (GLubyte *)"Mesa DRI SAVAGE Linux_1.1.18";
+      offset = driGetRendererString( buffer, cardNames[chipset], DRIVER_DATE,
+				     screen->agpMode );
+      return (GLubyte *)buffer;
    default:
       return 0;
    }

src/mesa/drivers/dri/savage/savagedma.c

@@ -30,20 +30,7 @@
 #include <time.h>
 #include <unistd.h>
 
-/* Commit does not depend on whether we use real DMA or fake it via the BCI */
-void savageDMACommit (savageContextPtr imesa, void *endPtr) {
-    DMABufferPtr dmaBuff = &imesa->DMABuf;
-    GLuint end = (GLuint)endPtr;
-
-    /* make sure that enough space was allocated */
-    assert (end <= dmaBuff->allocEnd);
-
-    dmaBuff->allocEnd = dmaBuff->end = end;
-
-    /* TODO: check commands, either here or in flush */
-}
-
-#if SAVAGE_CMD_DMA
+#if 0
 /* flag =
          0  return -1 if no available page
 	 1  wait until a page be available */
@@ -219,139 +206,4 @@ int savageDMAClose (savageContextPtr imesa)
 
     return GL_TRUE;
 }
-#else
-/* Allocate space in faked DMA buffer */
-void *savageDMAAlloc (savageContextPtr imesa, GLuint size) {
-    DMABufferPtr dmaBuff = &imesa->DMABuf;
-
-    /* make sure that everything has been filled in and committed */
-    assert (dmaBuff->end == dmaBuff->allocEnd);
-
-    size *= sizeof (u_int32_t); /* size in bytes */
-    if (dmaBuff->end + size >= dmaBuff->buf->linear + DMA_PAGE_SIZE) {
-	/* need kick off */
-	savageDMAFlush (imesa);
-    }
-    dmaBuff->allocEnd = dmaBuff->end + size;
-    return (void *)dmaBuff->end;
-}
-
-/* Flush DMA buffer via BCI (faked DMA) */
-void savageDMAFlush(savageContextPtr imesa) {
-    volatile u_int32_t* BCIbase;
-    DMABufferPtr dmaBuff = &imesa->DMABuf;
-    u_int32_t *entry;
-
-    /* make sure that everything has been filled in and committed */
-    assert (dmaBuff->allocEnd == dmaBuff->end);
-
-    if (dmaBuff->start == dmaBuff->end) /* no command? */
-      return;
-
-    /* get bci base */
-    BCIbase = (volatile u_int32_t *)SAVAGE_GET_BCI_POINTER(
-	imesa, (dmaBuff->end - dmaBuff->start) / sizeof (u_int32_t));
-
-    for (entry = (u_int32_t *)dmaBuff->start;
-	 entry < (u_int32_t *)dmaBuff->end; ++entry)
-	*BCIbase = *entry;
-
-    dmaBuff->end = dmaBuff->allocEnd = dmaBuff->start;
-}
-
-/* Init faked DMA */
-int savageDMAInit (savageContextPtr imesa) {
-    DMABufferPtr dmaBuff = &imesa->DMABuf;
-    drm_savage_alloc_cont_mem_t * req;
-
-    req = (drm_savage_alloc_cont_mem_t *)
-	malloc (sizeof(drm_savage_alloc_cont_mem_t));
-    if (!req)
-	return GL_FALSE;
-
-    req->linear = (GLuint)malloc (DMA_PAGE_SIZE);
-    if (!req->linear) {
-	free (req);
-	return GL_FALSE;
-    }
-
-    dmaBuff->buf = req;
-
-    dmaBuff->start = dmaBuff->end = dmaBuff->allocEnd = req->linear;
-    dmaBuff->usingPage = 0;
-    dmaBuff->kickFlag = GL_FALSE;
-
-    return GL_TRUE;
-}
-
-/* Close faked DMA */
-int savageDMAClose (savageContextPtr imesa) {
-    DMABufferPtr dmaBuff = &imesa->DMABuf;
-    drm_savage_alloc_cont_mem_t * req = dmaBuff->buf;
-
-    free ((void *)req->linear);
-    free (req);
-
-    return GL_TRUE;
-}
-
 #endif
-
-/* Faked vertex buffers
- *
- * This is a dirty hack, knowing that it will go away soon when real
- * vertex DMA is implemented and eventually moved to the DRM.
- */
-
-static u_int32_t vertex_data[16384]; /* 64KB */
-static drmBuf vertex_buffer = {
-    0,                       /* idx */
-    65536,                   /* total = 64KB */
-    0,                       /* used */
-    (drmAddress)vertex_data  /* address */
-};
-
-void savageFakeVertices (savageContextPtr imesa, drmBufPtr buffer) {
-    GLuint vertexStride = imesa->HwVertexSize; /* stride in dwords */
-    GLuint vertexSize = imesa->HwVertexSize; /* the real vertex size in dwords */
-    GLuint nVertices = buffer->used / (vertexStride*4);
-    u_int32_t *data = (u_int32_t*)buffer->address;
-    u_int32_t vertexFormat = imesa->DrawPrimitiveCmd & SAVAGE_HW_SKIPFLAGS;
-    GLuint i, j, left;
-
-    /* we have the monopoly on vertex buffers ;-) */
-    assert (buffer == &vertex_buffer);
-    assert (buffer->used % (vertexStride*4) == 0); /* whole vertices */
-    assert (nVertices % 3 == 0);                   /* triangle lists */
-
-    /* Flush (pseodo) DMA before accessing the BCI directly. */
-    savageDMAFlush(imesa);
-
-    left = nVertices;
-    while (left != 0) {
-	/* Can emit up to 255 vertices (85 triangles) with one command. */
-	GLuint count = left > 255 ? 255 : left;
-	/* Don't go through another buffering mechanism, copy to BCI
-	 * directly. */
-	volatile u_int32_t *vb = SAVAGE_GET_BCI_POINTER(imesa,
-						       count*vertexSize + 1);
-
-	WRITE_CMD (vb, SAVAGE_DRAW_PRIMITIVE(
-		       count, SAVAGE_HW_TRIANGLE_LIST | vertexFormat, 0),
-		   u_int32_t);
-	for (i = 0; i < count; ++i) {
-	    for (j = 0; j < vertexSize; ++j)
-		WRITE_CMD (vb, data[j], u_int32_t);
-	    data += vertexStride;
-	}
-	left -= count;
-    }
-
-    /* clear the vertex buffer for the next set of vertices */
-    vertex_buffer.used = 0;
-}
-
-drmBufPtr savageFakeGetBuffer (savageContextPtr imesa) {
-    assert (vertex_buffer.used == 0); /* has been flushed */
-    return &vertex_buffer;
-}

src/mesa/drivers/dri/savage/savagedma.h

@@ -49,8 +49,4 @@ void savageDMAFlush (savageContextPtr imesa);
 int savageDMAInit (savageContextPtr imesa);
 int savageDMAClose (savageContextPtr);
 
-/* faked implementation of vertex buffers */
-void savageFakeVertices (savageContextPtr imesa, drmBufPtr buffer);
-drmBufPtr savageFakeGetBuffer (savageContextPtr imesa);
-
 #endif

src/mesa/drivers/dri/savage/savageioctl.c

@@ -37,16 +37,87 @@
 #include "savagecontext.h"
 #include "savageioctl.h"
 #include "savage_bci.h"
-#include "savagedma.h"
+#include "savagestate.h"
+#include "savagespan.h"
 
 #include "drm.h"
 #include <sys/ioctl.h>
 #include <sys/timeb.h>
 
-extern GLuint bcicount;
 #define DEPTH_SCALE_16 ((1<<16)-1)
 #define DEPTH_SCALE_24 ((1<<24)-1)
 
+
+void savageGetDMABuffer( savageContextPtr imesa )
+{
+   int idx = 0;
+   int size = 0;
+   drmDMAReq dma;
+   int retcode;
+   drmBufPtr buf;
+
+   if (SAVAGE_DEBUG & DEBUG_DMA)
+      fprintf(stderr,  "Getting dma buffer\n");
+
+   dma.context = imesa->hHWContext;
+   dma.send_count = 0;
+   dma.send_list = NULL;
+   dma.send_sizes = NULL;
+   dma.flags = 0;
+   dma.request_count = 1;
+   dma.request_size = imesa->bufferSize;
+   dma.request_list = &idx;
+   dma.request_sizes = &size;
+   dma.granted_count = 0;
+
+
+   if (SAVAGE_DEBUG & DEBUG_DMA)
+      fprintf(stderr, "drmDMA (get) ctx %d count %d size 0x%x\n",
+	   dma.context, dma.request_count,
+	   dma.request_size);
+
+   while (1) {
+      retcode = drmDMA(imesa->driFd, &dma);
+
+      if (SAVAGE_DEBUG & DEBUG_DMA)
+	 fprintf(stderr, "retcode %d sz %d idx %d count %d\n",
+		 retcode,
+		 dma.request_sizes[0],
+		 dma.request_list[0],
+		 dma.granted_count);
+
+      if (retcode == 0 &&
+	  dma.request_sizes[0] &&
+	  dma.granted_count)
+	 break;
+
+      if (SAVAGE_DEBUG & DEBUG_DMA)
+	 fprintf(stderr, "\n\nflush");
+   }
+
+   buf = &(imesa->savageScreen->bufs->list[idx]);
+
+   if (SAVAGE_DEBUG & DEBUG_DMA)
+      fprintf(stderr,
+	   "drmDMA (get) returns size[0] 0x%x idx[0] %d\n"
+	   "dma_buffer now: buf idx: %d size: %d used: %d addr %p\n",
+	   dma.request_sizes[0], dma.request_list[0],
+	   buf->idx, buf->total,
+	   buf->used, buf->address);
+
+   imesa->dmaVtxBuf.total = buf->total / 4;
+   imesa->dmaVtxBuf.used = 0;
+   imesa->dmaVtxBuf.flushed = 0;
+   imesa->dmaVtxBuf.idx = buf->idx;
+   imesa->dmaVtxBuf.buf = (u_int32_t *)buf->address;
+
+   if (SAVAGE_DEBUG & DEBUG_DMA)
+      fprintf(stderr, "finished getbuffer\n");
+}
+
+#if 0
+/* Still keeping this around because it demonstrates page flipping and
+ * automatic z-clear. */
 static void savage_BCI_clear(GLcontext *ctx, drm_savage_clear_t *pclear)
 {
 	savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
@@ -113,7 +184,7 @@ static void savage_BCI_clear(GLcontext *ctx, drm_savage_clear_t *pclear)
                             imesa->regs.s4.zBufCtrl.ni.frameID =
 				~imesa->regs.s4.zBufCtrl.ni.frameID;
                             
-                            imesa->dirty |= SAVAGE_UPLOAD_CTX;
+                            imesa->dirty |= SAVAGE_UPLOAD_GLOBAL;
 		        }
 		        else
 		        {
@@ -157,8 +228,6 @@ static void savage_BCI_clear(GLcontext *ctx, drm_savage_clear_t *pclear)
 	savageDMAFlush (imesa);
 }
 
-struct timeb a,b;
-
 static void savage_BCI_swap(savageContextPtr imesa)
 {
     int nbox = imesa->sarea->nbox;
@@ -186,7 +255,7 @@ static void savage_BCI_swap(savageContextPtr imesa)
         imesa->readMap = (char *)imesa->apertureBase[imesa->toggle];
         
         imesa->regs.s4.destCtrl.ni.offset = imesa->savageScreen->backOffset>>11;
-        imesa->dirty |= SAVAGE_UPLOAD_CTX;
+        imesa->dirty |= SAVAGE_UPLOAD_GLOBAL;
         bciptr = SAVAGE_GET_BCI_POINTER(imesa,3);
         *(bciptr) = 0x960100B0;
         *(bciptr) = (imesa->savageScreen->frontOffset); 
@@ -220,97 +289,120 @@ static void savage_BCI_swap(savageContextPtr imesa)
         
     }
 }
+#endif
+
+
+static GLboolean intersect_rect( drm_clip_rect_t *out,
+				 const drm_clip_rect_t *a,
+				 const drm_clip_rect_t *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;
+
+   return ((out->x1 < out->x2) && (out->y1 < out->y2));
+}
+
+
+static GLuint savageIntersectClipRects(drm_clip_rect_t *dest,
+				       const drm_clip_rect_t *src,
+				       GLuint nsrc,
+				       const drm_clip_rect_t *clip)
+{
+    GLuint i, ndest;
+
+    for (i = 0, ndest = 0; i < nsrc; ++i, ++src) {
+	if (intersect_rect(dest, src, clip)) {
+	    dest++;
+	    ndest++;
+	}
+    }
+
+    return ndest;
+}
 
 
 static void savageDDClear( GLcontext *ctx, GLbitfield mask, GLboolean all,
 			   GLint cx, GLint cy, GLint cw, GLint ch ) 
 {
   savageContextPtr imesa = SAVAGE_CONTEXT( ctx );
-   __DRIdrawablePrivate *dPriv = imesa->driDrawable;
-   const GLuint colorMask = *((GLuint *) &ctx->Color.ColorMask);
-   drm_savage_clear_t clear;
-   int i;
+   GLuint colorMask, depthMask, clearColor, clearDepth, flags;
 
-   clear.flags = 0;
-   clear.clear_color = imesa->ClearColor;
+   if (SAVAGE_DEBUG & DEBUG_VERBOSE_MSG)
+       fprintf (stderr, "%s\n", __FUNCTION__);
 
-   if(imesa->savageScreen->zpp == 2)
-       clear.clear_depth = (GLuint) (ctx->Depth.Clear * DEPTH_SCALE_16);
-   else
-       clear.clear_depth = (GLuint) (ctx->Depth.Clear * DEPTH_SCALE_24);
+   clearColor = imesa->ClearColor;
+   if (imesa->float_depth) {
+       if (imesa->savageScreen->zpp == 2)
+	   clearDepth = savageEncodeFloat16(1.0 - ctx->Depth.Clear);
+       else
+	   clearDepth = savageEncodeFloat24(1.0 - ctx->Depth.Clear);
+   } else {
+       if (imesa->savageScreen->zpp == 2)
+	   clearDepth = (GLuint) ((1.0 - ctx->Depth.Clear) * DEPTH_SCALE_16);
+       else
+	   clearDepth = (GLuint) ((1.0 - ctx->Depth.Clear) * DEPTH_SCALE_24);
+   }
 
-   FLUSH_BATCH( imesa );
+   colorMask = *((GLuint *) &ctx->Color.ColorMask);
+   depthMask = 0;
 
-   if ((mask & DD_FRONT_LEFT_BIT) && ((colorMask&0xffffffUL)==0xffffffUL) ){
-      clear.flags |= SAVAGE_FRONT;
+   flags = 0;
+
+   if (mask & DD_FRONT_LEFT_BIT) {
+      flags |= SAVAGE_FRONT;
       mask &= ~DD_FRONT_LEFT_BIT;
    }
 
-   if ((mask & DD_BACK_LEFT_BIT) && ((colorMask&0xffffffUL)==0xffffffUL) ) {
-      clear.flags |= SAVAGE_BACK;
+   if (mask & DD_BACK_LEFT_BIT) {
+      flags |= SAVAGE_BACK;
       mask &= ~DD_BACK_LEFT_BIT;
    }
 
    if ((mask & DD_DEPTH_BIT) && ctx->Depth.Mask) {
-      clear.flags |= SAVAGE_DEPTH;
+      flags |= SAVAGE_DEPTH;
+      depthMask |=
+	  (imesa->savageScreen->zpp == 2) ? 0xffffffff : 0x00ffffff;
       mask &= ~DD_DEPTH_BIT;
    }
    
    if((mask & DD_STENCIL_BIT) && imesa->hw_stencil)
    {
-       clear.flags |= SAVAGE_STENCIL;
-       mask &= ~DD_STENCIL_BIT;
+      flags |= SAVAGE_DEPTH;
+      depthMask |= 0xff000000;
+      mask &= ~DD_STENCIL_BIT;
    }
 
-   if (clear.flags) {
-       LOCK_HARDWARE( imesa );
+   savageFlushVertices(imesa);
 
-       /* flip top to bottom */
-       cy = dPriv->h-cy-ch;
-       cx += imesa->drawX;
-       cy += imesa->drawY;
-
-       for (i = 0 ; i < imesa->numClipRects ; ) { 	 
-	   int nr = MIN2(i + SAVAGE_NR_SAREA_CLIPRECTS, imesa->numClipRects);
-	   drm_clip_rect_t *box = imesa->pClipRects;	 
-	   drm_clip_rect_t *b = imesa->sarea->boxes;
-	   int 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++ = *(drm_clip_rect_t *)&box[i];
-		   n++;
-	       }
-	   }
-
-	   imesa->sarea->nbox = n;
-
-	   savage_BCI_clear(ctx,&clear);
+   if (flags) {
+       GLboolean depthCleared = GL_FALSE;
+       if (flags & (SAVAGE_FRONT|SAVAGE_BACK)) {
+	   drm_savage_cmd_header_t *cmd;
+	   cmd = savageAllocCmdBuf(imesa, sizeof(drm_savage_cmd_header_t));
+	   cmd[0].clear0.cmd = SAVAGE_CMD_CLEAR;
+	   if ((flags & SAVAGE_DEPTH) &&
+	       clearDepth == clearColor && depthMask == colorMask) {
+	       cmd[0].clear0.flags = flags;
+	       depthCleared = GL_TRUE;
+	   } else
+	       cmd[0].clear0.flags = flags & (SAVAGE_FRONT|SAVAGE_BACK);
+	   cmd[1].clear1.mask = colorMask;
+	   cmd[1].clear1.value = clearColor;
        }
 
-       UNLOCK_HARDWARE( imesa );
-       imesa->dirty |= SAVAGE_UPLOAD_CLIPRECTS|SAVAGE_UPLOAD_CTX;
+       if ((flags & SAVAGE_DEPTH) && !depthCleared) {
+	   drm_savage_cmd_header_t *cmd;
+	   cmd = savageAllocCmdBuf(imesa, sizeof(drm_savage_cmd_header_t));
+	   cmd[0].clear0.cmd = SAVAGE_CMD_CLEAR;
+	   cmd[0].clear0.flags = SAVAGE_DEPTH;
+	   cmd[1].clear1.mask = depthMask;
+	   cmd[1].clear1.value = clearDepth;
+       }
+
+       imesa->dirty |= SAVAGE_UPLOAD_CLIPRECTS;
    }
 
    if (mask) 
@@ -318,7 +410,11 @@ static void savageDDClear( GLcontext *ctx, GLbitfield mask, GLboolean all,
 }
 
 
-
+/* This is necessary as to prevent annyoing stuttering effects with
+ * some games, though it does reduce the frame rate (glxgears)
+ * slightly. I believe this is due to texture uploads which do not go
+ * through the Savage command pipeline yet. */
+#define SYNC_FRAMES 1
 
 /*
  * Copy the back buffer to the front buffer. 
@@ -326,11 +422,9 @@ static void savageDDClear( GLcontext *ctx, GLbitfield mask, GLboolean all,
 void savageSwapBuffers( __DRIdrawablePrivate *dPriv )
 {
    savageContextPtr imesa;
-   drm_clip_rect_t *pbox;
-   int nbox;
-   int i;
 
-   GLboolean pending;
+   if (SAVAGE_DEBUG & DEBUG_VERBOSE_MSG)
+       fprintf (stderr, "%s\n================================\n", __FUNCTION__);
 
    assert(dPriv);
    assert(dPriv->driContextPriv);
@@ -342,38 +436,30 @@ void savageSwapBuffers( __DRIdrawablePrivate *dPriv )
 
    FLUSH_BATCH(imesa);
 
-   LOCK_HARDWARE( imesa );
-   WAIT_IDLE_EMPTY;
-   PAGE_PENDING(pending);
+#if SYNC_FRAMES
+   imesa->lastSwap = savageEmitEvent( imesa, 0 );
+#endif
+   if (imesa->lastSwap != 0)
+       savageWaitEvent( imesa, imesa->lastSwap );
 
-   if(!pending)
    {
-   pbox = dPriv->pClipRects;
-   nbox = dPriv->numClipRects;
-
-   for (i = 0 ; i < nbox ; )
-   {
-      int nr = MIN2(i + SAVAGE_NR_SAREA_CLIPRECTS, dPriv->numClipRects);
-      drm_clip_rect_t *b = (drm_clip_rect_t *)imesa->sarea->boxes;
-
-      imesa->sarea->nbox = nr - i;
-
-      for ( ; i < nr ; i++) 
-	 *b++ = pbox[i];
-     savage_BCI_swap(imesa) ;
+       drm_savage_cmd_header_t *cmd = savageAllocCmdBuf(imesa, 0);
+       cmd->cmd.cmd = SAVAGE_CMD_SWAP;
+       imesa->inSwap = GL_TRUE; /* ignore scissors in savageFlushCmdBuf */
+       savageFlushCmdBuf(imesa, GL_FALSE);
+       imesa->inSwap = GL_FALSE;
    }
-   }
-   UNLOCK_HARDWARE( imesa );
-   
 
+#if !SYNC_FRAMES
+   imesa->lastSwap = savageEmitEvent( imesa, 0 );
+#endif
 }
 
 /* This waits for *everybody* to finish rendering -- overkill.
  */
 void savageDmaFinish( savageContextPtr imesa  ) 
 {
-    savageDMAFlush(imesa);
-    WAIT_IDLE_EMPTY;
+    savageWaitEvent( imesa, savageEmitEventLocked( imesa, SAVAGE_WAIT_3D ) );
 }
 
 
@@ -393,152 +479,189 @@ void savageWaitAge( savageContextPtr imesa, int age  )
 }
 
 
-
-void savageFlushVerticesLocked( savageContextPtr imesa )
+unsigned int savageEmitEventLocked( savageContextPtr imesa, unsigned int flags )
 {
-    drmBufPtr buffer = imesa->vertex_dma_buffer;
+    drm_savage_event_emit_t event;
+    int ret;
+    event.count = 0;
+    event.flags = flags;
+    ret = drmCommandWriteRead( imesa->driFd, DRM_SAVAGE_BCI_EVENT_EMIT,
+			       &event, sizeof(event) );
+    if (ret) {
+	fprintf (stderr, "emit event returned %d\n", ret);
+	exit (1);
+    }
+    return event.count;
+}
+unsigned int savageEmitEvent( savageContextPtr imesa, unsigned int flags )
+{
+    unsigned int ret;
+    LOCK_HARDWARE( imesa );
+    ret = savageEmitEventLocked( imesa, flags );
+    UNLOCK_HARDWARE( imesa );
+    return ret;
+}
 
-    if (!buffer)
-	return;
 
-    imesa->vertex_dma_buffer = NULL;
-
-    /* Lot's of stuff to do here. For now there is a fake DMA implementation
-     * in savagedma.c that emits drawing commands. Cliprects are not handled
-     * yet. */
-    if (buffer->used) {
-	/* State must be updated "per primitive" because hardware
-	 * culling must be disabled for unfilled primitives, points
-	 * and lines. */
-	savageEmitHwStateLocked (imesa);
-	savageFakeVertices (imesa, buffer);
+void savageWaitEvent( savageContextPtr imesa, unsigned int count )
+{
+    drm_savage_event_wait_t event;
+    int ret;
+    event.count = count;
+    event.flags = 0;
+    ret = drmCommandWriteRead( imesa->driFd, DRM_SAVAGE_BCI_EVENT_WAIT,
+			       &event, sizeof(event) );
+    if (ret) {
+	fprintf (stderr, "wait event returned %d\n", ret);
+	exit (1);
     }
 }
 
 
-void savageFlushVertices( savageContextPtr imesa ) 
+void savageFlushVertices( savageContextPtr imesa )
 {
+    struct savage_vtxbuf_t *buffer = imesa->vtxBuf;
+
+    if (SAVAGE_DEBUG & DEBUG_VERBOSE_MSG)
+	fprintf (stderr, "%s\n", __FUNCTION__);
+
+    if (!buffer->total)
+	return;
+
+    if (buffer->used > buffer->flushed) {
+	drm_savage_cmd_header_t *cmd;
+	/* State must be updated "per primitive" because hardware
+	 * culling must be disabled for unfilled primitives, points
+	 * and lines. */
+	savageEmitChangedState (imesa);
+	cmd = savageAllocCmdBuf(imesa, 0);
+	cmd->prim.cmd = buffer == &imesa->dmaVtxBuf ?
+	    SAVAGE_CMD_DMA_PRIM : SAVAGE_CMD_VB_PRIM;
+	cmd->prim.prim = imesa->HwPrim;
+	cmd->prim.skip = imesa->skip;
+	cmd->prim.start = buffer->flushed / imesa->HwVertexSize;
+	cmd->prim.count = buffer->used / imesa->HwVertexSize - cmd->prim.start;
+	buffer->flushed = buffer->used;
+	/* Make sure we don't buffer too many vertices without
+	 * telling the hardware. */
+	imesa->vertsPending += cmd->prim.count;
+	if (imesa->vertsPending > SAVAGE_MAX_VERTS_PENDING) {
+	    savageFlushCmdBuf(imesa, GL_FALSE);
+	    imesa->vertsPending = 0;
+	}
+    }
+}
+
+void savageFlushCmdBufLocked( savageContextPtr imesa, GLboolean discard )
+{
+    __DRIdrawablePrivate *dPriv = imesa->driDrawable;
+    drm_savage_cmdbuf_t cmdbuf;
+    drm_savage_cmd_header_t *start;
+    int ret;
+
+    /* If we lost the context we must restore the initial state (at
+     * the start of the command buffer). */
+    if (imesa->lostContext) {
+	start = imesa->cmdBuf.base;
+	imesa->lostContext = GL_FALSE;
+    } else
+	start = imesa->cmdBuf.start;
+
+    if (!imesa->dmaVtxBuf.total)
+	discard = GL_FALSE;
+
+    if ((SAVAGE_DEBUG & DEBUG_DMA) && discard)
+	fprintf (stderr, "Discarding DMA buffer, used=%u\n",
+		 imesa->dmaVtxBuf.used);
+
+    cmdbuf.dma_idx = imesa->dmaVtxBuf.idx;
+    cmdbuf.discard = discard;
+    cmdbuf.vb_addr = imesa->clientVtxBuf.buf;
+    cmdbuf.vb_size = imesa->clientVtxBuf.total*4;
+    cmdbuf.vb_stride = imesa->HwVertexSize;
+    cmdbuf.cmd_addr = start;
+    cmdbuf.size = (imesa->cmdBuf.write - start);
+    if (!imesa->inSwap && imesa->glCtx->Scissor.Enabled) {
+	drm_clip_rect_t *box = dPriv->pClipRects, *ibox;
+	GLuint nbox = dPriv->numClipRects, nibox;
+	ibox = malloc(dPriv->numClipRects*sizeof(drm_clip_rect_t));
+	if (!ibox) {
+	    fprintf(stderr, "Out of memory.\n");
+	    exit(1);
+	}
+	nibox = savageIntersectClipRects(ibox, box, nbox, &imesa->scissor_rect);
+	cmdbuf.nbox = nibox;
+	cmdbuf.box_addr = ibox;
+    } else {
+	cmdbuf.nbox = dPriv->numClipRects;
+	cmdbuf.box_addr = dPriv->pClipRects;
+    }
+
+    ret = drmCommandWrite( imesa->driFd, DRM_SAVAGE_BCI_CMDBUF,
+			   &cmdbuf, sizeof(cmdbuf) );
+    if (ret) {
+	fprintf (stderr, "cmdbuf ioctl returned %d\n", ret);
+	exit(1);
+    }
+
+    if (cmdbuf.box_addr != dPriv->pClipRects) {
+	free(cmdbuf.box_addr);
+    }
+
+    if (discard) {
+	imesa->dmaVtxBuf.total = 0;
+	imesa->dmaVtxBuf.used = 0;
+	imesa->dmaVtxBuf.flushed = 0;
+    }
+    imesa->clientVtxBuf.used = 0;
+    imesa->clientVtxBuf.flushed = 0;
+
+    imesa->cmdBuf.write = imesa->cmdBuf.base;
+
+    /* Save the current state at the start of the command buffer. That
+     * state will only be emitted, if the context was lost since the
+     * last command buffer. */
+    savageEmitOldState(imesa);
+    imesa->cmdBuf.start = imesa->cmdBuf.write;
+}
+
+
+void savageFlushCmdBuf( savageContextPtr imesa, GLboolean discard ) 
+{
+    if (SAVAGE_DEBUG & DEBUG_VERBOSE_MSG)
+	fprintf (stderr, "%s\n", __FUNCTION__);
     LOCK_HARDWARE(imesa);
-    savageFlushVerticesLocked (imesa);
+    savageFlushCmdBufLocked (imesa, discard);
     UNLOCK_HARDWARE(imesa);
 }
 
 
-int savage_check_copy(int fd)
-{
-    return 0;
-}
-
 static void savageDDFlush( GLcontext *ctx )
 {
+    if (SAVAGE_DEBUG & DEBUG_VERBOSE_MSG)
+	fprintf (stderr, "%s\n", __FUNCTION__);
     savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
+    savageFlushVertices (imesa);
     LOCK_HARDWARE(imesa);
-    savageFlushVerticesLocked (imesa);
-    savageDMAFlush (imesa);
+    savageFlushCmdBufLocked(imesa, GL_FALSE);
     UNLOCK_HARDWARE(imesa);
 }
 
 static void savageDDFinish( GLcontext *ctx  ) 
 {
+    if (SAVAGE_DEBUG & DEBUG_VERBOSE_MSG)
+	fprintf (stderr, "%s\n", __FUNCTION__);
     savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
+    savageFlushVertices (imesa);
     LOCK_HARDWARE(imesa);
-    savageFlushVerticesLocked (imesa);
+    savageFlushCmdBufLocked(imesa, GL_FALSE);
     savageDmaFinish (imesa);
     UNLOCK_HARDWARE(imesa);
 }
 
-#define ALT_STATUS_WORD0 (* (volatile GLuint *)(imesa->MMIO_BASE+0x48c60))
-#define STATUS_WORD0 (* (volatile GLuint *)(imesa->MMIO_BASE+0x48c00))
-#define MAXFIFO_S4  0x7F00
-#define MAXFIFO_S3D 0x7F00
-
-static GLboolean savagePagePending_s4( savageContextPtr imesa ) {
-    return (ALT_STATUS_WORD0 & 0x08000000) ? GL_TRUE : GL_FALSE;
-}
-static GLboolean savagePagePending_s3d( savageContextPtr imesa ) {
-    return GL_FALSE;
-}
-static void savageWaitForFIFO_s4( savageContextPtr imesa, unsigned count ) {
-    int loop = 0;
-    int slots = MAXFIFO_S4-count;
-    while((ALT_STATUS_WORD0 & 0x001fffff) > slots && loop++ < MAXLOOP);
-}
-static void savageWaitForFIFO_s3d( savageContextPtr imesa, unsigned count ) {
-    int loop = 0;
-    int slots = MAXFIFO_S3D-count;
-    while((STATUS_WORD0 & 0x0001ffff) > slots && loop++ < MAXLOOP);
-}
-static void savageWaitIdleEmpty_s4( savageContextPtr imesa ) {
-    int loop = 0;
-    while((ALT_STATUS_WORD0 & 0x00ffffff) != 0x00E00000L && loop++ < MAXLOOP);
-}
-static void savageWaitIdleEmpty_s3d( savageContextPtr imesa ) {
-    int loop = 0;
-    while((STATUS_WORD0 & 0x000fffff) != 0x000E0000L && loop++ < MAXLOOP);
-}
-
-GLboolean (*savagePagePending)( savageContextPtr imesa ) = NULL;
-void (*savageWaitForFIFO)( savageContextPtr imesa, unsigned count ) = NULL;
-void (*savageWaitIdleEmpty)( savageContextPtr imesa ) = NULL;
-
-
 void savageDDInitIoctlFuncs( GLcontext *ctx )
 {
    ctx->Driver.Clear = savageDDClear;
    ctx->Driver.Flush = savageDDFlush;
    ctx->Driver.Finish = savageDDFinish;
-   if (SAVAGE_CONTEXT( ctx )->savageScreen->chipset >= S3_SAVAGE4) {
-       savagePagePending = savagePagePending_s4;
-       savageWaitForFIFO = savageWaitForFIFO_s4;
-       savageWaitIdleEmpty = savageWaitIdleEmpty_s4;
-   } else {
-       savagePagePending = savagePagePending_s3d;
-       savageWaitForFIFO = savageWaitForFIFO_s3d;
-       savageWaitIdleEmpty = savageWaitIdleEmpty_s3d;
-   }
 }
-
-#if SAVAGE_CMD_DMA
-/* Alloc a continuous memory */
-/* return: 0 error when kernel alloc pages(can try a half memory size) 
-           >0 sucess
-	   <0 Other error*/
-int  savageAllocDMABuffer(savageContextPtr imesa,  drm_savage_alloc_cont_mem_t *req)
-{
-  int ret;
-  if (req ==NULL)
-    return 0;
-
-  if ((ret=ioctl(imesa->driFd, DRM_IOCTL_SAVAGE_ALLOC_CONTINUOUS_MEM, req)) <=0)    
-    return ret;
-  
-  return 1;
-  
-}
-
-/* get the physics address*/
-GLuint savageGetPhyAddress(savageContextPtr imesa,void * pointer)
-{
-
-  drm_savage_get_physcis_address_t req;
-  int ret;
-
-  req.v_address = (GLuint )pointer;
-  ret = ioctl(imesa->driFd, DRM_IOCTL_SAVAGE_GET_PHYSICS_ADDRESS,&req);
-
-  return req.p_address;
-}
-
-/* free the buffer got by savageAllocDMABuffe*/
-int  savageFreeDMABuffer(savageContextPtr imesa,  drm_savage_alloc_cont_mem_t *req)
-{
-  GLuint ret;
-  if (req ==NULL)
-    return 0;
-  
-  if ((ret=ioctl(imesa->driFd, DRM_IOCTL_SAVAGE_FREE_CONTINUOUS_MEM, req)) <=0)    
-    return ret;
-  return 1;
-  
-}
-#endif

src/mesa/drivers/dri/savage/savageioctl.h

@@ -27,17 +27,22 @@
 #define SAVAGE_IOCTL_H
 
 #include "savagecontext.h"
-#include "savagedma.h"
 
 void savageGetGeneralDmaBufferLocked( savageContextPtr mmesa ); 
 
 void savageFlushVertices( savageContextPtr mmesa ); 
-void savageFlushVerticesLocked( savageContextPtr mmesa );
 
 void savageFlushGeneralLocked( savageContextPtr imesa );
 void savageWaitAgeLocked( savageContextPtr imesa, int age );
 void savageWaitAge( savageContextPtr imesa, int age );
 
+unsigned int savageEmitEventLocked( savageContextPtr imesa, unsigned int flags );
+unsigned int savageEmitEvent( savageContextPtr imesa, unsigned int flags );
+void savageWaitEvent( savageContextPtr imesa, unsigned int event);
+
+void savageFlushCmdBufLocked( savageContextPtr imesa, GLboolean discard );
+void savageFlushCmdBuf( savageContextPtr imesa, GLboolean discard );
+
 void savageDmaFinish( savageContextPtr imesa );
 
 void savageRegetLockQuiescent( savageContextPtr imesa );
@@ -46,54 +51,69 @@ void savageDDInitIoctlFuncs( GLcontext *ctx );
 
 void savageSwapBuffers( __DRIdrawablePrivate *dPriv );
 
-int savage_check_copy(int fd);
-
-extern GLboolean (*savagePagePending)( savageContextPtr imesa );
-extern void (*savageWaitForFIFO)( savageContextPtr imesa, unsigned count );
-extern void (*savageWaitIdleEmpty)( savageContextPtr imesa );
-
-#define PAGE_PENDING(result) do { \
-    result = savagePagePending(imesa); \
-} while (0)
-#define WAIT_FOR_FIFO(count) do { \
-    savageWaitForFIFO(imesa, count); \
-} while (0)
 #define WAIT_IDLE_EMPTY do { \
-    savageWaitIdleEmpty(imesa); \
+    savageWaitEvent(imesa, \
+		    savageEmitEvent(imesa, SAVAGE_WAIT_3D|SAVAGE_WAIT_2D)); \
 } while (0)
 
-#if SAVAGE_CMD_DMA
-int  savageAllocDMABuffer(savageContextPtr imesa,  drm_savage_alloc_cont_mem_t *req);
-GLuint savageGetPhyAddress(savageContextPtr imesa,void * pointer);
-int  savageFreeDMABuffer(savageContextPtr, drm_savage_alloc_cont_mem_t*);
-#endif
-
 #define FLUSH_BATCH(imesa) do { \
-    if (imesa->vertex_dma_buffer) savageFlushVertices(imesa); \
+    if (SAVAGE_DEBUG & DEBUG_VERBOSE_MSG) \
+        fprintf (stderr, "FLUSH_BATCH in %s\n", __FUNCTION__); \
+    savageFlushVertices(imesa); \
+    savageFlushCmdBuf(imesa, GL_FALSE); \
 } while (0)
 
+extern void savageGetDMABuffer( savageContextPtr imesa );
+
 static __inline
-u_int32_t *savageAllocDmaLow( savageContextPtr imesa, GLuint bytes )
+u_int32_t *savageAllocVtxBuf( savageContextPtr imesa, GLuint words )
 {
+   struct savage_vtxbuf_t *buffer = imesa->vtxBuf;
    u_int32_t *head;
 
-   if (!imesa->vertex_dma_buffer) {
-      LOCK_HARDWARE(imesa);
-      imesa->vertex_dma_buffer = savageFakeGetBuffer (imesa);
-      UNLOCK_HARDWARE(imesa);
-   } else if (imesa->vertex_dma_buffer->used + bytes >
-	      imesa->vertex_dma_buffer->total) {
-      LOCK_HARDWARE(imesa);
-      savageFlushVerticesLocked( imesa );
-      imesa->vertex_dma_buffer = savageFakeGetBuffer (imesa);
-      UNLOCK_HARDWARE(imesa);
+   if (buffer == &imesa->dmaVtxBuf) {
+       if (!buffer->total) {
+	   LOCK_HARDWARE(imesa);
+	   savageGetDMABuffer(imesa);
+	   UNLOCK_HARDWARE(imesa);
+       } else if (buffer->used + words > buffer->total) {
+	   if (SAVAGE_DEBUG & DEBUG_VERBOSE_MSG)
+	       fprintf (stderr, "... flushing DMA buffer in %s\n",
+			__FUNCTION__);
+	   savageFlushVertices( imesa );
+	   LOCK_HARDWARE(imesa);
+	   savageFlushCmdBufLocked(imesa, GL_TRUE); /* discard DMA buffer */
+	   savageGetDMABuffer(imesa);
+	   UNLOCK_HARDWARE(imesa);
+       }
+   } else if (buffer->used + words > buffer->total) {
+       if (SAVAGE_DEBUG & DEBUG_VERBOSE_MSG)
+	   fprintf (stderr, "... flushing client vertex buffer in %s\n",
+		    __FUNCTION__);
+       savageFlushVertices( imesa );
+       LOCK_HARDWARE(imesa);
+       savageFlushCmdBufLocked(imesa, GL_FALSE); /* free clientVtxBuf */
+       UNLOCK_HARDWARE(imesa);
    }
 
-   head = (u_int32_t *)((u_int8_t *)imesa->vertex_dma_buffer->address +
-		       imesa->vertex_dma_buffer->used);
+   head = &buffer->buf[buffer->used];
 
-   imesa->vertex_dma_buffer->used += bytes;
+   buffer->used += words;
    return head;
 }
 
+static __inline
+drm_savage_cmd_header_t *savageAllocCmdBuf( savageContextPtr imesa, GLuint bytes )
+{
+    drm_savage_cmd_header_t *ret;
+    GLuint qwords = ((bytes + 7) >> 3) + 1; /* round up */
+    assert (qwords < imesa->cmdBuf.size);
+    if (imesa->cmdBuf.write - imesa->cmdBuf.base + qwords > imesa->cmdBuf.size) {
+	savageFlushCmdBuf(imesa, GL_FALSE);
+    }
+    ret = (drm_savage_cmd_header_t *)imesa->cmdBuf.write;
+    imesa->cmdBuf.write += qwords;
+    return ret;
+}
+
 #endif

src/mesa/drivers/dri/savage/savagespan.c

@@ -33,6 +33,7 @@
 #define DBG 0
 
 #define LOCAL_VARS					\
+   savageContextPtr imesa = SAVAGE_CONTEXT(ctx);	\
    __DRIdrawablePrivate *dPriv = imesa->mesa_drawable;	\
    savageScreenPrivate *savageScreen = imesa->savageScreen;	\
    GLuint cpp   = savageScreen->cpp;			\
@@ -48,6 +49,7 @@
    (void) read_buf; (void) buf; (void) p
 
 #define LOCAL_DEPTH_VARS				\
+   savageContextPtr imesa = SAVAGE_CONTEXT(ctx);	\
    __DRIdrawablePrivate *dPriv = imesa->mesa_drawable;	\
    savageScreenPrivate *savageScreen = imesa->savageScreen;	\
    GLuint zpp   = savageScreen->zpp;			\
@@ -77,8 +79,7 @@
 
 #define Y_FLIP(_y) (height - _y - 1)
 
-#define HW_LOCK() savageContextPtr imesa = SAVAGE_CONTEXT(ctx); \
-                  WAIT_IDLE_EMPTY;\
+#define HW_LOCK()
 
 #define HW_CLIPLOOP()						\
   do {								\
@@ -95,11 +96,7 @@
     }						\
   } while (0)
 
-#if 0
-#define HW_UNLOCK()				\
-    UNLOCK_HARDWARE(imesa);
-#endif
-#define HW_UNLOCK()	{ }
+#define HW_UNLOCK()
 
 
 /* 16 bit, 565 rgb color spanline and pixel functions
@@ -161,58 +158,91 @@ do {								\
 
 
 
-/* 16 bit depthbuffer functions.
+/* 16 bit integer depthbuffer functions
+ * Depth range is reversed. See also savageCalcViewport.
  */
 #define WRITE_DEPTH( _x, _y, d ) \
-do{							\
-    *(GLushort *)(buf + (_x<<1) + _y*pitch)  = d;	\
-}while(0)
-    
+    *(GLushort *)(buf + ((_x)<<1) + (_y)*pitch) = 0xFFFF - d
+
 #define READ_DEPTH( d, _x, _y ) \
-do{							\
-    d = *(GLushort *)(buf + (_x<<1) + _y*pitch);	\
-}while(0)
-       
-/*     d = 0xffff; */
-       
+    d = 0xFFFF - *(GLushort *)(buf + ((_x)<<1) + (_y)*pitch)
+
 #define TAG(x) savage##x##_16
 #include "depthtmp.h"
-       
 
 
 
 
-/* 8-bit stencil /24-bit depth depthbuffer functions.
+
+/* 16 bit float depthbuffer functions
  */
-#define WRITE_DEPTH( _x, _y, d ) {			\
-   GLuint tmp = *(GLuint *)(buf + (_x<<2) + _y*pitch);	\
-   tmp &= 0xFF000000;					\
-   tmp |= d;						\
+#define WRITE_DEPTH( _x, _y, d ) \
+    *(GLushort *)(buf + ((_x)<<1) + (_y)*pitch) = \
+        savageEncodeFloat16( 1.0 - (GLfloat)d/65535.0 )
+
+#define READ_DEPTH( d, _x, _y ) \
+    d = 65535 - \
+        savageDecodeFloat16( *(GLushort *)(buf + ((_x)<<1) + (_y)*pitch) ) * \
+	65535.0
+
+#define TAG(x) savage##x##_16f
+#include "depthtmp.h"
+
+
+
+
+
+/* 8-bit stencil /24-bit integer depth depthbuffer functions.
+ * Depth range is reversed. See also savageCalcViewport.
+ */
+#define WRITE_DEPTH( _x, _y, d ) do {				\
+   GLuint tmp = *(GLuint *)(buf + ((_x)<<2) + (_y)*pitch);	\
+   tmp &= 0xFF000000;						\
+   tmp |= 0x00FFFFFF - d;					\
    *(GLuint *)(buf + (_x<<2) + _y*pitch)  = tmp;		\
-}
+} while(0)
 
 #define READ_DEPTH( d, _x, _y )	\
-   d = *(GLuint *)(buf + (_x<<2) + _y*pitch) & 0x00FFFFFF;
-
-/*     d = 0x00ffffff; */
+   d = 0x00FFFFFF - (*(GLuint *)(buf + ((_x)<<2) + (_y)*pitch) & 0x00FFFFFF)
 
 #define TAG(x) savage##x##_8_24
 #include "depthtmp.h"
 
 
-#define WRITE_STENCIL( _x, _y, d ) {                    \
-   GLuint tmp = *(GLuint *)(buf + (_x<<2) + _y*pitch);     \
-   tmp &= 0x00FFFFFF;                                   \
-   tmp |= (((GLuint)d)<<24) & 0xFF000000;               \
-   *(GLuint *)(buf + (_x<<2) + _y*pitch) = tmp;            \
-}
-            
-#define READ_STENCIL( d, _x, _y )               \
-   d = (GLstencil)((*(GLuint *)(buf + (_x<<2) + _y*pitch) & 0xFF000000) >> 24);
-                
+
+
+
+/* 24 bit float depthbuffer functions
+ */
+#define WRITE_DEPTH( _x, _y, d ) do {				\
+    GLuint tmp = *(GLuint *)(buf + ((_x)<<2) + (_y)*pitch);	\
+    tmp &= 0xFF000000;						\
+    tmp |= savageEncodeFloat24( 1.0 - (GLfloat)d/16777215.0 );	\
+   *(GLuint *)(buf + (_x<<2) + _y*pitch)  = tmp;		\
+} while(0)
+
+#define READ_DEPTH( d, _x, _y )					\
+    d = 16777215 - savageDecodeFloat24(				\
+	*(GLuint *)(buf + ((_x)<<2) + (_y)*pitch) & 0x00FFFFFF)	\
+	* 16777215.0
+
+#define TAG(x) savage##x##_8_24f
+#include "depthtmp.h"
+
+
+#define WRITE_STENCIL( _x, _y, d ) do {				\
+   GLuint tmp = *(GLuint *)(buf + ((_x)<<2) + (_y)*pitch);	\
+   tmp &= 0x00FFFFFF;						\
+   tmp |= (((GLuint)d)<<24) & 0xFF000000;			\
+   *(GLuint *)(buf + ((_x)<<2) + (_y)*pitch) = tmp;		\
+} while(0)
+
+#define READ_STENCIL( d, _x, _y ) \
+   d = (GLstencil)((*(GLuint *)(buf + ((_x)<<2) + (_y)*pitch) & 0xFF000000) >> 24)
+
 #define TAG(x) savage##x##_8_24
 #include "stenciltmp.h"
-                
+
 
 /*
  * This function is called to specify which buffer to read and write
@@ -241,6 +271,58 @@ static void savageDDSetBuffer(GLcontext *ctx, GLframebuffer *buffer,
        ? imesa->driDrawable : imesa->driReadable;
 }
 
+/*
+ * Wrappers around _swrast_Copy/Draw/ReadPixels that make sure all
+ * primitives are flushed and the hardware is idle before accessing
+ * the frame buffer.
+ */
+static void
+savageCopyPixels( GLcontext *ctx,
+		  GLint srcx, GLint srcy, GLsizei width, GLsizei height,
+		  GLint destx, GLint desty,
+		  GLenum type )
+{
+    savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
+    FLUSH_BATCH(imesa);
+    WAIT_IDLE_EMPTY;
+    _swrast_CopyPixels(ctx, srcx, srcy, width, height, destx, desty, type);
+}
+static void
+savageDrawPixels( GLcontext *ctx,
+		  GLint x, GLint y,
+		  GLsizei width, GLsizei height,
+		  GLenum format, GLenum type,
+		  const struct gl_pixelstore_attrib *packing,
+		  const GLvoid *pixels )
+{
+    savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
+    FLUSH_BATCH(imesa);
+    WAIT_IDLE_EMPTY;
+    _swrast_DrawPixels(ctx, x, y, width, height, format, type, packing, pixels);
+}
+static void
+savageReadPixels( GLcontext *ctx,
+		  GLint x, GLint y, GLsizei width, GLsizei height,
+		  GLenum format, GLenum type,
+		  const struct gl_pixelstore_attrib *packing,
+		  GLvoid *pixels )
+{
+    savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
+    FLUSH_BATCH(imesa);
+    WAIT_IDLE_EMPTY;
+    _swrast_ReadPixels(ctx, x, y, width, height, format, type, packing, pixels);
+}
+
+/*
+ * Make sure the hardware is idle when span-rendering.
+ */
+static void savageSpanRenderStart( GLcontext *ctx )
+{
+   savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
+   FLUSH_BATCH(imesa);
+   WAIT_IDLE_EMPTY;
+}
+
 
 void savageDDInitSpanFuncs( GLcontext *ctx )
 {
@@ -274,18 +356,36 @@ void savageDDInitSpanFuncs( GLcontext *ctx )
 
    switch (imesa->savageScreen->zpp)
    {
-   case 2: 
-       swdd->ReadDepthSpan = savageReadDepthSpan_16;
-       swdd->WriteDepthSpan = savageWriteDepthSpan_16;
-       swdd->ReadDepthPixels = savageReadDepthPixels_16;
-       swdd->WriteDepthPixels = savageWriteDepthPixels_16;
+   case 2:
+       if (imesa->float_depth) {
+	   swdd->ReadDepthSpan = savageReadDepthSpan_16f;
+	   swdd->WriteDepthSpan = savageWriteDepthSpan_16f;
+	   swdd->WriteMonoDepthSpan = savageWriteMonoDepthSpan_16f;
+	   swdd->ReadDepthPixels = savageReadDepthPixels_16f;
+	   swdd->WriteDepthPixels = savageWriteDepthPixels_16f;
+       } else {
+	   swdd->ReadDepthSpan = savageReadDepthSpan_16;
+	   swdd->WriteDepthSpan = savageWriteDepthSpan_16;
+	   swdd->WriteMonoDepthSpan = savageWriteMonoDepthSpan_16;
+	   swdd->ReadDepthPixels = savageReadDepthPixels_16;
+	   swdd->WriteDepthPixels = savageWriteDepthPixels_16;
+       }
        
        break;
    case 4: 
-       swdd->ReadDepthSpan = savageReadDepthSpan_8_24;
-       swdd->WriteDepthSpan = savageWriteDepthSpan_8_24;
-       swdd->ReadDepthPixels = savageReadDepthPixels_8_24;
-       swdd->WriteDepthPixels = savageWriteDepthPixels_8_24;    
+       if (imesa->float_depth) {
+	   swdd->ReadDepthSpan = savageReadDepthSpan_8_24f;
+	   swdd->WriteDepthSpan = savageWriteDepthSpan_8_24f;
+	   swdd->WriteMonoDepthSpan = savageWriteMonoDepthSpan_8_24f;
+	   swdd->ReadDepthPixels = savageReadDepthPixels_8_24f;
+	   swdd->WriteDepthPixels = savageWriteDepthPixels_8_24f;    
+       } else {
+	   swdd->ReadDepthSpan = savageReadDepthSpan_8_24;
+	   swdd->WriteDepthSpan = savageWriteDepthSpan_8_24;
+	   swdd->WriteMonoDepthSpan = savageWriteMonoDepthSpan_8_24;
+	   swdd->ReadDepthPixels = savageReadDepthPixels_8_24;
+	   swdd->WriteDepthPixels = savageWriteDepthPixels_8_24;    
+       }
        swdd->ReadStencilSpan = savageReadStencilSpan_8_24;
        swdd->WriteStencilSpan = savageWriteStencilSpan_8_24;
        swdd->ReadStencilPixels = savageReadStencilPixels_8_24;
@@ -301,11 +401,13 @@ void savageDDInitSpanFuncs( GLcontext *ctx )
    swdd->ReadCI32Span        =NULL;
    swdd->ReadCI32Pixels      =NULL;
 
+   swdd->SpanRenderStart = savageSpanRenderStart;
+
    /* 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;
+   ctx->Driver.CopyPixels = savageCopyPixels;
+   ctx->Driver.DrawPixels = savageDrawPixels;
+   ctx->Driver.ReadPixels = savageReadPixels;
 }

src/mesa/drivers/dri/savage/savagespan.h

@@ -27,4 +27,102 @@
 
 extern void savageDDInitSpanFuncs( GLcontext *ctx );
 
+/*
+ * Savage 16-bit float depth format with zExpOffset=16:
+ *   4 bit unsigned exponent, 12 bit mantissa
+ *
+ * The meaning of the mantissa is different from IEEE floatint point
+ * formats. The same number can't be encoded with different exponents.
+ * So no bits are wasted.
+ *
+ * exponent | range encoded by mantissa | accuracy or mantissa
+ * ---------+---------------------------+---------------------
+ *       15 | 2^-1 .. 1                 | 2^-13
+ *       14 | 2^-2 .. 2^-1              | 2^-14
+ *       13 | 2^-3 .. 2^-2              | 2^-15
+ *      ... | ...                       |
+ *        2 | 2^-14 .. 2^-13            | 2^-27
+ *        1 | 2^-15 .. 2^-14            | 2^-27
+ *        0 | 2^-16 .. 2^-15            | 2^-28
+ *
+ * Note that there is no encoding for numbers < 2^-16.
+ */
+static __inline GLuint savageEncodeFloat16( GLdouble x )
+{
+    GLint r = (GLint)(x * 0x10000000);
+    GLint exp = 0;
+    if (r < 0x1000)
+	return 0;
+    while (r - 0x1000 > 0x0fff) {
+	r >>= 1;
+	exp++;
+    }
+    return exp > 0xf ? 0xffff : (r - 0x1000) | (exp << 12);
+}
+static __inline GLdouble savageDecodeFloat16( GLuint x )
+{
+    static const GLdouble pow2[16] = {
+	1.0/(1<<28), 1.0/(1<<27), 1.0/(1<<26), 1.0/(1<<25),
+	1.0/(1<<24), 1.0/(1<<23), 1.0/(1<<22), 1.0/(1<<21),
+	1.0/(1<<20), 1.0/(1<<19), 1.0/(1<<18), 1.0/(1<<17),
+	1.0/(1<<16), 1.0/(1<<15), 1.0/(1<<14), 1.0/(1<<13)
+    };
+    static const GLdouble bias[16] = {
+	1.0/(1<<16), 1.0/(1<<15), 1.0/(1<<14), 1.0/(1<<13),
+	1.0/(1<<12), 1.0/(1<<11), 1.0/(1<<10), 1.0/(1<< 9),
+	1.0/(1<< 8), 1.0/(1<< 7), 1.0/(1<< 6), 1.0/(1<< 5),
+	1.0/(1<< 4), 1.0/(1<< 3), 1.0/(1<< 2), 1.0/(1<< 1)
+    };
+    GLuint mant = x & 0x0fff;
+    GLuint exp = (x >> 12) & 0xf;
+    return bias[exp] + pow2[exp]*mant;
+}
+
+/*
+ * Savage 24-bit float depth format with zExpOffset=32:
+ *   5 bit unsigned exponent, 19 bit mantissa
+ *
+ * Details analogous to the 16-bit format.
+ */
+static __inline GLuint savageEncodeFloat24( GLdouble x )
+{
+    int64_t r = (int64_t)(x * ((int64_t)1 << (19+32)));
+    GLint exp = 0;
+    if (r < 0x80000)
+	return 0;
+    while (r - 0x80000 > 0x7ffff) {
+	r >>= 1;
+	exp++;
+    }
+    return exp > 0x1f ? 0xffffff : (r - 0x80000) | (exp << 19);
+}
+#define _1 (int64_t)1
+static __inline GLdouble savageDecodeFloat24( GLuint x )
+{
+    static const GLdouble pow2[32] = {
+	1.0/(_1<<51), 1.0/(_1<<50), 1.0/(_1<<49), 1.0/(_1<<48),
+	1.0/(_1<<47), 1.0/(_1<<46), 1.0/(_1<<45), 1.0/(_1<<44),
+	1.0/(_1<<43), 1.0/(_1<<42), 1.0/(_1<<41), 1.0/(_1<<40),
+	1.0/(_1<<39), 1.0/(_1<<38), 1.0/(_1<<37), 1.0/(_1<<36),
+	1.0/(_1<<35), 1.0/(_1<<34), 1.0/(_1<<33), 1.0/(_1<<32),
+	1.0/(_1<<31), 1.0/(_1<<30), 1.0/(_1<<29), 1.0/(_1<<28),
+	1.0/(_1<<27), 1.0/(_1<<26), 1.0/(_1<<25), 1.0/(_1<<24),
+	1.0/(_1<<23), 1.0/(_1<<22), 1.0/(_1<<21), 1.0/(_1<<20)
+    };
+    static const GLdouble bias[32] = {
+	1.0/(_1<<32), 1.0/(_1<<31), 1.0/(_1<<30), 1.0/(_1<<29),
+	1.0/(_1<<28), 1.0/(_1<<27), 1.0/(_1<<26), 1.0/(_1<<25),
+	1.0/(_1<<24), 1.0/(_1<<23), 1.0/(_1<<22), 1.0/(_1<<21),
+	1.0/(_1<<20), 1.0/(_1<<19), 1.0/(_1<<18), 1.0/(_1<<17),
+	1.0/(_1<<16), 1.0/(_1<<15), 1.0/(_1<<14), 1.0/(_1<<13),
+	1.0/(_1<<12), 1.0/(_1<<11), 1.0/(_1<<10), 1.0/(_1<< 9),
+	1.0/(_1<< 8), 1.0/(_1<< 7), 1.0/(_1<< 6), 1.0/(_1<< 5),
+	1.0/(_1<< 4), 1.0/(_1<< 3), 1.0/(_1<< 2), 1.0/(_1<< 1)
+    };
+    GLuint mant = x & 0x7ffff;
+    GLuint exp = (x >> 19) & 0x1f;
+    return bias[exp] + pow2[exp]*mant;
+}
+#undef _1
+
 #endif

src/mesa/drivers/dri/savage/savagestate.c

@@ -69,16 +69,10 @@ static __inline__ GLuint savagePackColor(GLuint format,
 
 static void savageDDAlphaFunc_s4(GLcontext *ctx, GLenum func, GLfloat ref)
 {
-    /* This can be done in BlendFunc*/
-    savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;
     savageBlendFunc_s4(ctx);
 }
 static void savageDDAlphaFunc_s3d(GLcontext *ctx, GLenum func, GLfloat ref)
 {
-    /* This can be done in BlendFunc*/
-    savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;
     savageBlendFunc_s3d(ctx);
 }
 
@@ -103,6 +97,9 @@ static void savageDDBlendEquationSeparate(GLcontext *ctx,
 static void savageBlendFunc_s4(GLcontext *ctx)
 {
     savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
+    u_int32_t drawLocalCtrl = imesa->regs.s4.drawLocalCtrl.ui;
+    u_int32_t drawCtrl0 = imesa->regs.s4.drawCtrl0.ui;
+    u_int32_t drawCtrl1 = imesa->regs.s4.drawCtrl1.ui;
 
     /* set up draw control register (including blending, alpha
      * test, and shading model)
@@ -266,11 +263,17 @@ static void savageBlendFunc_s4(GLcontext *ctx)
     /*imesa->regs.s4.drawLocalCtrl.ni.zUpdateEn =
         ~drawLocalCtrl.ni.wrZafterAlphaTst;*/
 
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;
+    if (drawLocalCtrl != imesa->regs.s4.drawLocalCtrl.ui)
+	imesa->dirty |= SAVAGE_UPLOAD_LOCAL;
+    if (drawCtrl0 != imesa->regs.s4.drawCtrl0.ui ||
+	drawCtrl1 != imesa->regs.s4.drawCtrl1.ui)
+	imesa->dirty |= SAVAGE_UPLOAD_GLOBAL;
 }
 static void savageBlendFunc_s3d(GLcontext *ctx)
 {
     savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
+    u_int32_t drawCtrl = imesa->regs.s3d.drawCtrl.ui;
+    u_int32_t zBufCtrl = imesa->regs.s3d.zBufCtrl.ui;
 
     /* set up draw control register (including blending, alpha
      * test, dithering, and shading model)
@@ -432,7 +435,9 @@ static void savageBlendFunc_s3d(GLcontext *ctx)
     imesa->regs.s3d.zBufCtrl.ni.wrZafterAlphaTst =
 	imesa->regs.s3d.drawCtrl.ni.alphaTestEn;
 
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;
+    if (drawCtrl != imesa->regs.s3d.drawCtrl.ui ||
+	zBufCtrl != imesa->regs.s3d.zBufCtrl.ui)
+	imesa->dirty |= SAVAGE_UPLOAD_LOCAL;
 }
 
 static void savageDDBlendFuncSeparate_s4( GLcontext *ctx, GLenum sfactorRGB, 
@@ -456,21 +461,23 @@ static void savageDDDepthFunc_s4(GLcontext *ctx, GLenum func)
 {
     savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
     ZCmpFunc zmode;
-#define depthIndex 0
+    u_int32_t drawLocalCtrl = imesa->regs.s4.drawLocalCtrl.ui;
+    u_int32_t zBufCtrl = imesa->regs.s4.zBufCtrl.ui;
+    u_int32_t zWatermarks = imesa->regs.s4.zWatermarks.ui; /* FIXME: in DRM */
 
     /* set up z-buffer control register (global)
      * set up z-buffer offset register (global)
      * set up z read/write watermarks register (global)
      */
 
-    switch(func)  { 
+    switch(func)  { /* reversed (see savageCalcViewport) */
     case GL_NEVER: zmode = CF_Never; break;
     case GL_ALWAYS: zmode = CF_Always; break;
-    case GL_LESS: zmode = CF_Less; break; 
-    case GL_LEQUAL: zmode = CF_LessEqual; break;
+    case GL_LESS: zmode = CF_Greater; break; 
+    case GL_LEQUAL: zmode = CF_GreaterEqual; break;
     case GL_EQUAL: zmode = CF_Equal; break;
-    case GL_GREATER: zmode = CF_Greater; break;
-    case GL_GEQUAL: zmode = CF_GreaterEqual; break;
+    case GL_GREATER: zmode = CF_Less; break;
+    case GL_GEQUAL: zmode = CF_LessEqual; break;
     case GL_NOTEQUAL: zmode = CF_NotEqual; break;
     default:return;
     } 
@@ -513,27 +520,33 @@ static void savageDDDepthFunc_s4(GLcontext *ctx, GLenum func)
 	imesa->regs.s4.drawLocalCtrl.ni.flushPdZbufWrites = GL_FALSE;
 	imesa->regs.s4.zWatermarks.ni.wLow        = 8;
     }
-  
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;
+
+    if (drawLocalCtrl != imesa->regs.s4.drawLocalCtrl.ui)
+	imesa->dirty |= SAVAGE_UPLOAD_LOCAL;
+    if (zBufCtrl != imesa->regs.s4.zBufCtrl.ui ||
+	zWatermarks != imesa->regs.s4.zWatermarks.ui)
+	imesa->dirty |= SAVAGE_UPLOAD_GLOBAL;
 }
 static void savageDDDepthFunc_s3d(GLcontext *ctx, GLenum func)
 {
     savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
     ZCmpFunc zmode;
-#define depthIndex 0
+    u_int32_t drawCtrl = imesa->regs.s3d.drawCtrl.ui;
+    u_int32_t zBufCtrl = imesa->regs.s3d.zBufCtrl.ui;
+    u_int32_t zWatermarks = imesa->regs.s3d.zWatermarks.ui; /* FIXME: in DRM */
 
     /* set up z-buffer control register (global)
      * set up z-buffer offset register (global)
      * set up z read/write watermarks register (global)
      */
-    switch(func)  { 
+    switch(func)  { /* reversed (see savageCalcViewport) */
     case GL_NEVER: zmode = CF_Never; break;
     case GL_ALWAYS: zmode = CF_Always; break;
-    case GL_LESS: zmode = CF_Less; break; 
-    case GL_LEQUAL: zmode = CF_LessEqual; break;
+    case GL_LESS: zmode = CF_Greater; break; 
+    case GL_LEQUAL: zmode = CF_GreaterEqual; break;
     case GL_EQUAL: zmode = CF_Equal; break;
-    case GL_GREATER: zmode = CF_Greater; break;
-    case GL_GEQUAL: zmode = CF_GreaterEqual; break;
+    case GL_GREATER: zmode = CF_Less; break;
+    case GL_GEQUAL: zmode = CF_LessEqual; break;
     case GL_NOTEQUAL: zmode = CF_NotEqual; break;
     default:return;
     } 
@@ -565,14 +578,18 @@ static void savageDDDepthFunc_s3d(GLcontext *ctx, GLenum func)
 	imesa->regs.s3d.zWatermarks.ni.wLow = 8;
     }
   
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;
+    if (drawCtrl != imesa->regs.s3d.drawCtrl.ui ||
+	zBufCtrl != imesa->regs.s3d.zBufCtrl.ui)
+	imesa->dirty |= SAVAGE_UPLOAD_LOCAL;
+    if (zWatermarks != imesa->regs.s3d.zWatermarks.ui)
+	imesa->dirty |= SAVAGE_UPLOAD_GLOBAL;
 }
 
 static void savageDDDepthMask_s4(GLcontext *ctx, GLboolean flag)
 {
     savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
+    u_int32_t drawLocalCtrl = imesa->regs.s4.drawLocalCtrl.ui;
 
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;
     if (flag)
     {
 	imesa->regs.s4.drawLocalCtrl.ni.flushPdZbufWrites = GL_TRUE;
@@ -582,12 +599,15 @@ static void savageDDDepthMask_s4(GLcontext *ctx, GLboolean flag)
 	imesa->regs.s4.drawLocalCtrl.ni.flushPdZbufWrites = GL_FALSE;
     }
     savageDDDepthFunc_s4(ctx,ctx->Depth.Func);
+
+    if (drawLocalCtrl != imesa->regs.s4.drawLocalCtrl.ui)
+	imesa->dirty |= SAVAGE_UPLOAD_LOCAL;
 }
 static void savageDDDepthMask_s3d(GLcontext *ctx, GLboolean flag)
 {
     savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
+    u_int32_t drawCtrl = imesa->regs.s3d.drawCtrl.ui;
 
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;
     if (flag)
     {
 	imesa->regs.s3d.drawCtrl.ni.flushPdZbufWrites = GL_TRUE;
@@ -597,6 +617,9 @@ static void savageDDDepthMask_s3d(GLcontext *ctx, GLboolean flag)
 	imesa->regs.s3d.drawCtrl.ni.flushPdZbufWrites = GL_FALSE;
     }
     savageDDDepthFunc_s3d(ctx,ctx->Depth.Func);
+
+    if (drawCtrl != imesa->regs.s3d.drawCtrl.ui)
+	imesa->dirty |= SAVAGE_UPLOAD_LOCAL;
 }
 
 
@@ -629,6 +652,7 @@ static void savageDDScissor( GLcontext *ctx, GLint x, GLint y,
 static void savageDDDrawBuffer(GLcontext *ctx, GLenum mode )
 {
     savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
+    u_int32_t destCtrl = imesa->regs.s4.destCtrl.ui;
 
     /*
      * _DrawDestMask is easier to cope with than <mode>.
@@ -641,7 +665,6 @@ static void savageDDDrawBuffer(GLcontext *ctx, GLenum mode )
 	imesa->readMap = (char *)imesa->apertureBase[TARGET_FRONT];
 	imesa->regs.s4.destCtrl.ni.offset = imesa->savageScreen->frontOffset>>11;
         imesa->NotFirstFrame = GL_FALSE;
-        imesa->dirty |= SAVAGE_UPLOAD_BUFFERS | SAVAGE_UPLOAD_CTX;
         savageXMesaSetFrontClipRects( imesa );
 	FALLBACK( ctx, SAVAGE_FALLBACK_DRAW_BUFFER, GL_FALSE );
 	break;
@@ -651,7 +674,6 @@ static void savageDDDrawBuffer(GLcontext *ctx, GLenum mode )
         imesa->readMap = (char *)imesa->apertureBase[TARGET_BACK];
 	imesa->regs.s4.destCtrl.ni.offset = imesa->savageScreen->backOffset>>11;
         imesa->NotFirstFrame = GL_FALSE;
-        imesa->dirty |= SAVAGE_UPLOAD_BUFFERS | SAVAGE_UPLOAD_CTX;
         savageXMesaSetBackClipRects( imesa );
 	FALLBACK( ctx, SAVAGE_FALLBACK_DRAW_BUFFER, GL_FALSE );
 	break;
@@ -664,6 +686,9 @@ static void savageDDDrawBuffer(GLcontext *ctx, GLenum mode )
      * gets called.
      */
     _swrast_DrawBuffer(ctx, mode);
+
+    if (destCtrl != imesa->regs.s4.destCtrl.ui)
+        imesa->dirty |= SAVAGE_UPLOAD_GLOBAL;
 }
 
 static void savageDDReadBuffer(GLcontext *ctx, GLenum mode )
@@ -691,14 +716,22 @@ static void savageCalcViewport( GLcontext *ctx )
    const GLfloat *v = ctx->Viewport._WindowMap.m;
    GLfloat *m = imesa->hw_viewport;
 
-   /* See also mga_translate_vertex.
-    */
    m[MAT_SX] =   v[MAT_SX];
    m[MAT_TX] =   v[MAT_TX] + imesa->drawX + SUBPIXEL_X;
    m[MAT_SY] = - v[MAT_SY];
    m[MAT_TY] = - v[MAT_TY] + imesa->driDrawable->h + imesa->drawY + SUBPIXEL_Y;
-   m[MAT_SZ] =   v[MAT_SZ] * imesa->depth_scale;
-   m[MAT_TZ] =   v[MAT_TZ] * imesa->depth_scale;
+   /* Depth range is reversed (far: 0, near: 1) so that float depth
+    * compensates for loss of accuracy of far coordinates. */
+   if (imesa->float_depth && imesa->savageScreen->zpp == 2) {
+       /* The Savage 16-bit floating point depth format can't encode
+	* numbers < 2^-16. Make sure all depth values stay greater
+	* than that. */
+       m[MAT_SZ] = - v[MAT_SZ] * imesa->depth_scale * (65535.0/65536.0);
+       m[MAT_TZ] = 1.0 - v[MAT_TZ] * imesa->depth_scale * (65535.0/65536.0);
+   } else {
+       m[MAT_SZ] = - v[MAT_SZ] * imesa->depth_scale;
+       m[MAT_TZ] = 1.0 - v[MAT_TZ] * imesa->depth_scale;
+   }
 
    imesa->SetupNewInputs = ~0;
 }
@@ -800,12 +833,12 @@ static void savageUpdateCull( GLcontext *ctx )
     if (imesa->savageScreen->chipset >= S3_SAVAGE4) {
 	if (imesa->regs.s4.drawCtrl1.ni.cullMode != cullMode) {
 	    imesa->regs.s4.drawCtrl1.ni.cullMode = cullMode;
-	    imesa->dirty |= SAVAGE_UPLOAD_CTX;
+	    imesa->dirty |= SAVAGE_UPLOAD_GLOBAL;
 	}
     } else {
 	if (imesa->regs.s3d.drawCtrl.ni.cullMode != cullMode) {
 	    imesa->regs.s3d.drawCtrl.ni.cullMode = cullMode;
-	    imesa->dirty |= SAVAGE_UPLOAD_CTX;
+	    imesa->dirty |= SAVAGE_UPLOAD_LOCAL;
 	}
     }
 #endif /* end  #if HW_CULL */
@@ -846,7 +879,7 @@ static void savageDDColorMask_s4(GLcontext *ctx,
     {
 	imesa->regs.s4.drawLocalCtrl.ni.drawUpdateEn = GL_FALSE;
     }
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;
+    imesa->dirty |= SAVAGE_UPLOAD_LOCAL;
     /* TODO: need a software fallback */
 }
 static void savageDDColorMask_s3d(GLcontext *ctx, 
@@ -873,7 +906,7 @@ static void savageDDColorMask_s3d(GLcontext *ctx,
     {
 	imesa->regs.s3d.zBufCtrl.ni.drawUpdateEn = GL_FALSE;
     }
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;
+    imesa->dirty |= SAVAGE_UPLOAD_LOCAL;
     /* TODO: need a software fallback */
 }
 
@@ -886,6 +919,7 @@ static void savageDDColorMask_s3d(GLcontext *ctx,
  */
 static void savageUpdateSpecular_s4(GLcontext *ctx) {
     savageContextPtr imesa = SAVAGE_CONTEXT( ctx );
+    u_int32_t drawLocalCtrl = imesa->regs.s4.drawLocalCtrl.ui;
 
     if (ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR &&
 	ctx->Light.Enabled) {
@@ -895,10 +929,13 @@ static void savageUpdateSpecular_s4(GLcontext *ctx) {
 	imesa->regs.s4.drawLocalCtrl.ni.specShadeEn = GL_FALSE;
 	/*FALLBACK (ctx, SAVAGE_FALLBACK_SPECULAR, GL_FALSE);*/
     }
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;
+
+    if (drawLocalCtrl != imesa->regs.s4.drawLocalCtrl.ui)
+	imesa->dirty |= SAVAGE_UPLOAD_LOCAL;
 }
 static void savageUpdateSpecular_s3d(GLcontext *ctx) {
     savageContextPtr imesa = SAVAGE_CONTEXT( ctx );
+    u_int32_t drawCtrl = imesa->regs.s3d.drawCtrl.ui;
 
     if (ctx->Light.Model.ColorControl == GL_SEPARATE_SPECULAR_COLOR &&
 	ctx->Light.Enabled) {
@@ -908,7 +945,9 @@ static void savageUpdateSpecular_s3d(GLcontext *ctx) {
 	imesa->regs.s3d.drawCtrl.ni.specShadeEn = GL_FALSE;
 	/*FALLBACK (ctx, SAVAGE_FALLBACK_SPECULAR, GL_FALSE);*/
     }
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;
+
+    if (drawCtrl != imesa->regs.s3d.drawCtrl.ui)
+	imesa->dirty |= SAVAGE_UPLOAD_LOCAL;
 }
 
 static void savageDDLightModelfv_s4(GLcontext *ctx, GLenum pname, 
@@ -925,6 +964,7 @@ static void savageDDLightModelfv_s3d(GLcontext *ctx, GLenum pname,
 static void savageDDShadeModel_s4(GLcontext *ctx, GLuint mod)
 {
     savageContextPtr imesa = SAVAGE_CONTEXT( ctx );
+    u_int32_t drawLocalCtrl = imesa->regs.s4.drawLocalCtrl.ui;
 
     if (mod == GL_SMOOTH)  
     {    
@@ -934,11 +974,14 @@ static void savageDDShadeModel_s4(GLcontext *ctx, GLuint mod)
     {
 	imesa->regs.s4.drawLocalCtrl.ni.flatShadeEn = GL_TRUE;
     }
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;
+
+    if (drawLocalCtrl != imesa->regs.s4.drawLocalCtrl.ui)
+	imesa->dirty |= SAVAGE_UPLOAD_LOCAL;
 }
 static void savageDDShadeModel_s3d(GLcontext *ctx, GLuint mod)
 {
     savageContextPtr imesa = SAVAGE_CONTEXT( ctx );
+    u_int32_t drawCtrl = imesa->regs.s3d.drawCtrl.ui;
 
     if (mod == GL_SMOOTH)  
     {    
@@ -948,7 +991,9 @@ static void savageDDShadeModel_s3d(GLcontext *ctx, GLuint mod)
     {
 	imesa->regs.s3d.drawCtrl.ni.flatShadeEn = GL_TRUE;
     }
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;
+
+    if (drawCtrl != imesa->regs.s3d.drawCtrl.ui)
+	imesa->dirty |= SAVAGE_UPLOAD_LOCAL;
 }
 
 
@@ -962,6 +1007,7 @@ static void savageDDFogfv(GLcontext *ctx, GLenum pname, const GLfloat *param)
 {
     savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
     GLuint  fogClr;
+    u_int32_t fogCtrl = imesa->regs.s4.fogCtrl.ui;
 
     /*if ((ctx->Fog.Enabled) &&(pname == GL_FOG_COLOR))*/
     if (ctx->Fog.Enabled)
@@ -981,17 +1027,19 @@ static void savageDDFogfv(GLcontext *ctx, GLenum pname, const GLfloat *param)
 	imesa->regs.s4.fogCtrl.ni.fogEn     = 0;
 	imesa->regs.s4.fogCtrl.ni.fogMode   = 0;
     }
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;      
+
+    if (fogCtrl != imesa->regs.s4.fogCtrl.ui)
+	imesa->dirty |= SAVAGE_UPLOAD_GLOBAL;
 }
 
 
-static void savageStencilFunc(GLcontext *);
-
 static void savageDDStencilFunc(GLcontext *ctx, GLenum func, GLint ref,
                                 GLuint mask)
 {
     savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
     SCmpFunc a=0;
+    u_int32_t zBufCtrl = imesa->regs.s4.zBufCtrl.ui;
+    u_int32_t stencilCtrl = imesa->regs.s4.stencilCtrl.ui;
 
     imesa->regs.s4.zBufCtrl.ni.stencilRefVal = ctx->Stencil.Ref[0];
     imesa->regs.s4.stencilCtrl.ni.readMask  = ctx->Stencil.ValueMask[0];
@@ -1012,22 +1060,26 @@ static void savageDDStencilFunc(GLcontext *ctx, GLenum func, GLint ref,
 
     imesa->regs.s4.stencilCtrl.ni.cmpFunc = a;
 
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;
+    if (zBufCtrl != imesa->regs.s4.zBufCtrl.ui ||
+	stencilCtrl != imesa->regs.s4.stencilCtrl.ui)
+	imesa->dirty |= SAVAGE_UPLOAD_GLOBAL;
 }
 
 static void savageDDStencilMask(GLcontext *ctx, GLuint mask)
 {
     savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
 
-    imesa->regs.s4.stencilCtrl.ni.writeMask = ctx->Stencil.WriteMask[0];
-
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;
+    if (imesa->regs.s4.stencilCtrl.ni.writeMask != ctx->Stencil.WriteMask[0]) {
+	imesa->regs.s4.stencilCtrl.ni.writeMask = ctx->Stencil.WriteMask[0];
+	imesa->dirty |= SAVAGE_UPLOAD_GLOBAL;
+    }
 }
 
 static void savageDDStencilOp(GLcontext *ctx, GLenum fail, GLenum zfail,
                               GLenum zpass)
 {
     savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
+    u_int32_t stencilCtrl = imesa->regs.s4.stencilCtrl.ui;
 
     switch (ctx->Stencil.FailFunc[0])
     {
@@ -1114,7 +1166,8 @@ static void savageDDStencilOp(GLcontext *ctx, GLenum fail, GLenum zfail,
 	break;
     }
 
-    imesa->dirty |= SAVAGE_UPLOAD_CTX;
+    if (stencilCtrl != imesa->regs.s4.stencilCtrl.ui)
+	imesa->dirty |= SAVAGE_UPLOAD_GLOBAL;
 }
 
 
@@ -1128,11 +1181,9 @@ static void savageDDEnable_s4(GLcontext *ctx, GLenum cap, GLboolean state)
     switch(cap) {
         case GL_ALPHA_TEST:
             /* we should consider the disable case*/
-            imesa->dirty |= SAVAGE_UPLOAD_CTX;
             savageBlendFunc_s4(ctx);
             break;
         case GL_BLEND:
-            imesa->dirty |= SAVAGE_UPLOAD_CTX;
             /*Can't find Enable bit in the 3D registers.*/ 
             /* For some reason enable(GL_BLEND) affects ColorLogicOpEnabled.
              */
@@ -1146,7 +1197,6 @@ static void savageDDEnable_s4(GLcontext *ctx, GLenum cap, GLboolean state)
             savageBlendFunc_s4(ctx);
             break;
         case GL_DEPTH_TEST:
-            imesa->dirty |= SAVAGE_UPLOAD_CTX;
             savageDDDepthFunc_s4(ctx,ctx->Depth.Func);
             break;
         case GL_SCISSOR_TEST:
@@ -1154,7 +1204,6 @@ static void savageDDEnable_s4(GLcontext *ctx, GLenum cap, GLboolean state)
             imesa->dirty |= SAVAGE_UPLOAD_CLIPRECTS;
             break;
         case GL_STENCIL_TEST:
-            imesa->dirty |= SAVAGE_UPLOAD_CTX;
 	    if (!imesa->hw_stencil)
 		FALLBACK (ctx, SAVAGE_FALLBACK_STENCIL, state);
 	    else {
@@ -1167,11 +1216,10 @@ static void savageDDEnable_s4(GLcontext *ctx, GLenum cap, GLboolean state)
 		    imesa->regs.s4.zBufCtrl.ni.zBufEn         = GL_TRUE;
 		    imesa->regs.s4.drawLocalCtrl.ni.zUpdateEn = GL_FALSE;
 		}
-		imesa->dirty |= SAVAGE_UPLOAD_CTX;
+		imesa->dirty |= SAVAGE_UPLOAD_GLOBAL | SAVAGE_UPLOAD_LOCAL;
 	    }
             break;
         case GL_FOG:
-            imesa->dirty |= SAVAGE_UPLOAD_CTX;
             savageDDFogfv(ctx,0,0);	
             break;
         case GL_CULL_FACE:
@@ -1188,7 +1236,6 @@ static void savageDDEnable_s4(GLcontext *ctx, GLenum cap, GLboolean state)
 #endif
             break;
         case GL_DITHER:
-            imesa->dirty |= SAVAGE_UPLOAD_CTX;
             if (state)
             {
                 if ( ctx->Color.DitherFlag )
@@ -1200,6 +1247,7 @@ static void savageDDEnable_s4(GLcontext *ctx, GLenum cap, GLboolean state)
             {
                 imesa->regs.s4.drawCtrl1.ni.ditherEn=GL_FALSE;
             }
+            imesa->dirty |= SAVAGE_UPLOAD_GLOBAL;
             break;
  
         case GL_LIGHTING:
@@ -1223,11 +1271,9 @@ static void savageDDEnable_s3d(GLcontext *ctx, GLenum cap, GLboolean state)
     switch(cap) {
         case GL_ALPHA_TEST:
             /* we should consider the disable case*/
-            imesa->dirty |= SAVAGE_UPLOAD_CTX;
             savageBlendFunc_s3d(ctx);
             break;
         case GL_BLEND:
-            imesa->dirty |= SAVAGE_UPLOAD_CTX;
             /*Can't find Enable bit in the 3D registers.*/ 
             /* For some reason enable(GL_BLEND) affects ColorLogicOpEnabled.
              */
@@ -1241,7 +1287,6 @@ static void savageDDEnable_s3d(GLcontext *ctx, GLenum cap, GLboolean state)
             savageBlendFunc_s3d(ctx);
             break;
         case GL_DEPTH_TEST:
-            imesa->dirty |= SAVAGE_UPLOAD_CTX;
             savageDDDepthFunc_s3d(ctx,ctx->Depth.Func);
             break;
         case GL_SCISSOR_TEST:
@@ -1252,7 +1297,6 @@ static void savageDDEnable_s3d(GLcontext *ctx, GLenum cap, GLboolean state)
 	    FALLBACK (ctx, SAVAGE_FALLBACK_STENCIL, state);
 	    break;
         case GL_FOG:
-            imesa->dirty |= SAVAGE_UPLOAD_CTX;
             savageDDFogfv(ctx,0,0);	
             break;
         case GL_CULL_FACE:
@@ -1269,7 +1313,6 @@ static void savageDDEnable_s3d(GLcontext *ctx, GLenum cap, GLboolean state)
 #endif
             break;
         case GL_DITHER:
-            imesa->dirty |= SAVAGE_UPLOAD_CTX;
             if (state)
             {
                 if ( ctx->Color.DitherFlag )
@@ -1281,6 +1324,7 @@ static void savageDDEnable_s3d(GLcontext *ctx, GLenum cap, GLboolean state)
             {
                 imesa->regs.s3d.drawCtrl.ni.ditherEn=GL_FALSE;
             }
+            imesa->dirty |= SAVAGE_UPLOAD_LOCAL;
             break;
  
         case GL_LIGHTING:
@@ -1303,8 +1347,7 @@ void savageDDUpdateHwState( GLcontext *ctx )
     savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
 
     if (imesa->new_state) {
-	FLUSH_BATCH(imesa);
-
+	savageFlushVertices(imesa);
 	if (imesa->new_state & SAVAGE_NEW_TEXTURE) {
 	    savageUpdateTextureState( ctx );
 	}
@@ -1368,20 +1411,20 @@ void savageEmitDrawingRectangle( savageContextPtr imesa )
          imesa->regs.ni.changed.ni.fDrawCtrl1Changed=GL_TRUE;*/
 
     savageCalcViewport (imesa->glCtx);
-
-    imesa->dirty |= SAVAGE_UPLOAD_BUFFERS;
 }
 
 
 static void savageDDPrintDirty( const char *msg, GLuint state )
 {
-    fprintf(stderr, "%s (0x%x): %s%s%s%s%s\n",	   
+    fprintf(stderr, "%s (0x%x): %s%s%s%s%s%s%s\n",	   
             msg,
             (unsigned int) state,
-            (state & SAVAGE_UPLOAD_TEX0IMAGE)  ? "upload-tex0, " : "",
-            (state & SAVAGE_UPLOAD_TEX1IMAGE)  ? "upload-tex1, " : "",
-            (state & SAVAGE_UPLOAD_CTX)        ? "upload-ctx, " : "",
-            (state & SAVAGE_UPLOAD_BUFFERS)    ? "upload-bufs, " : "",
+            (state & SAVAGE_UPLOAD_LOCAL)      ? "upload-local, " : "",
+            (state & SAVAGE_UPLOAD_TEX0)       ? "upload-tex0, " : "",
+            (state & SAVAGE_UPLOAD_TEX1)       ? "upload-tex1, " : "",
+            (state & SAVAGE_UPLOAD_FOGTBL)     ? "upload-fogtbl, " : "",
+            (state & SAVAGE_UPLOAD_GLOBAL)     ? "upload-global, " : "",
+            (state & SAVAGE_UPLOAD_TEXGLOBAL)  ? "upload-texglobal, " : "",
             (state & SAVAGE_UPLOAD_CLIPRECTS)  ? "upload-cliprects, " : ""
             );
 }
@@ -1400,18 +1443,32 @@ static GLboolean savageGlobalRegChanged (savageContextPtr imesa,
     }
     return GL_FALSE;
 }
+static void savageEmitOldRegs (savageContextPtr imesa,
+			       GLuint first, GLuint last, GLboolean global) {
+    GLuint n = last-first+1;
+    drm_savage_cmd_header_t *cmd = savageAllocCmdBuf(imesa, n*4);
+    cmd->state.cmd = SAVAGE_CMD_STATE;
+    cmd->state.global = global;
+    cmd->state.count = n;
+    cmd->state.start = first;
+    memcpy(cmd+1, &imesa->oldRegs.ui[first-SAVAGE_FIRST_REG], n*4);
+}
 static void savageEmitContiguousRegs (savageContextPtr imesa,
 				      GLuint first, GLuint last) {
     GLuint i;
-    u_int32_t *pBCIBase;
-    pBCIBase = savageDMAAlloc (imesa, last - first + 2);
-    WRITE_CMD (pBCIBase, SET_REGISTER(first, last - first + 1), u_int32_t);
-
-    for (i = first - SAVAGE_FIRST_REG; i <= last - SAVAGE_FIRST_REG; ++i) {
-	WRITE_CMD (pBCIBase, imesa->regs.ui[i], u_int32_t);
+    GLuint n = last-first+1;
+    drm_savage_cmd_header_t *cmd = savageAllocCmdBuf(imesa, n*4);
+    cmd->state.cmd = SAVAGE_CMD_STATE;
+    cmd->state.global = savageGlobalRegChanged(imesa, first, last);
+    cmd->state.count = n;
+    cmd->state.start = first;
+    memcpy(cmd+1, &imesa->regs.ui[first-SAVAGE_FIRST_REG], n*4);
+    /* savageAllocCmdBuf may need to flush the cmd buffer and backup
+     * the current hardware state. It should see the "old" (current)
+     * state that has actually been emitted to the hardware. Therefore
+     * this update is done *after* savageAllocCmdBuf. */
+    for (i = first - SAVAGE_FIRST_REG; i <= last - SAVAGE_FIRST_REG; ++i)
 	imesa->oldRegs.ui[i] = imesa->regs.ui[i];
-    }
-    savageDMACommit (imesa, pBCIBase);
 }
 static void savageEmitChangedRegs (savageContextPtr imesa,
 				   GLuint first, GLuint last) {
@@ -1445,8 +1502,6 @@ static void savageEmitChangedRegChunk (savageContextPtr imesa,
 }
 static void savageUpdateRegister_s4(savageContextPtr imesa)
 {
-    u_int32_t *pBCIBase;
-
     /*
      * Scissors updates drawctrl0 and drawctrl 1
      */
@@ -1470,23 +1525,12 @@ static void savageUpdateRegister_s4(savageContextPtr imesa)
 
     /* the savage4 uses the contiguous range of BCI registers 0x1e-0x39
      * 0x1e-0x27 are local, no need to check them for global changes */
-    if (imesa->lostContext || savageGlobalRegChanged (imesa, 0x28, 0x39)) {
-	pBCIBase = savageDMAAlloc (imesa, 1);
-        WRITE_CMD (pBCIBase, WAIT_3D_IDLE, u_int32_t);
-	savageDMACommit (imesa, pBCIBase);
-    }
-    if (imesa->lostContext)
-	savageEmitContiguousRegs (imesa, 0x1e, 0x39);
-    else
-	savageEmitChangedRegs (imesa, 0x1e, 0x39);
+    savageEmitContiguousRegs (imesa, 0x1e, 0x39);
 
     imesa->dirty=0;
-    imesa->lostContext = GL_FALSE;
 }
 static void savageUpdateRegister_s3d(savageContextPtr imesa)
 {
-    u_int32_t *pBCIBase;
-
     if (imesa->scissorChanged)
     {
         if(imesa->scissor)
@@ -1519,63 +1563,55 @@ static void savageUpdateRegister_s3d(savageContextPtr imesa)
     imesa->regs.s3d.drawCtrl.ni.flushPdZbufWrites = GL_TRUE;
 
     /* the savage3d uses two contiguous ranges of BCI registers:
-     * 0x18-0x1c and 0x20-0x38. The first range is local. */
-    if (imesa->lostContext || savageGlobalRegChanged (imesa, 0x20, 0x38)) {
-	pBCIBase = savageDMAAlloc (imesa, 1);
-        WRITE_CMD (pBCIBase, WAIT_3D_IDLE, u_int32_t);
-	savageDMACommit (imesa, pBCIBase);
-    }
+     * 0x18-0x1c and 0x20-0x38. Some texture registers need to be
+     * emitted in one chunk or we get some funky rendering errors. */
     /* FIXME: watermark registers aren't programmed correctly ATM */
-    if (imesa->lostContext) {
-	savageEmitContiguousRegs (imesa, 0x18, 0x1c);
-	savageEmitContiguousRegs (imesa, 0x20, 0x36);
-    } else {
-	/* On the Savage IX texture registers (at least some of them)
-	 * have to be emitted as one chunk. */
-	savageEmitChangedRegs (imesa, 0x18, 0x19);
-	savageEmitChangedRegChunk (imesa, 0x1a, 0x1c);
-	savageEmitChangedRegs (imesa, 0x20, 0x36);
-    }
+    savageEmitChangedRegs (imesa, 0x18, 0x19);
+    savageEmitChangedRegChunk (imesa, 0x1a, 0x1c);
+    savageEmitChangedRegs (imesa, 0x20, 0x36);
 
     imesa->dirty=0;
-    imesa->lostContext = GL_FALSE;
 }
 
 
+void savageEmitOldState( savageContextPtr imesa )
+{
+    assert(imesa->cmdBuf.write == imesa->cmdBuf.base);
+    if (imesa->savageScreen->chipset >= S3_SAVAGE4) {
+	savageEmitOldRegs (imesa, 0x1e, 0x39, GL_TRUE);
+    } else {
+	savageEmitOldRegs (imesa, 0x18, 0x1c, GL_TRUE);
+	savageEmitOldRegs (imesa, 0x20, 0x36, GL_FALSE);
+    }
+}
+
 
 /* Push the state into the sarea and/or texture memory.
  */
-void savageEmitHwStateLocked( savageContextPtr imesa )
+void savageEmitChangedState( savageContextPtr imesa )
 {
     if (SAVAGE_DEBUG & DEBUG_VERBOSE_API)
         savageDDPrintDirty( "\n\n\nsavageEmitHwStateLocked", imesa->dirty );
 
     if (imesa->dirty & ~SAVAGE_UPLOAD_CLIPRECTS)
     {
-        if (imesa->dirty & (SAVAGE_UPLOAD_CTX | SAVAGE_UPLOAD_TEX0  | \
-                            SAVAGE_UPLOAD_TEX1 | SAVAGE_UPLOAD_BUFFERS))
+        if (imesa->dirty & (SAVAGE_UPLOAD_GLOBAL | SAVAGE_UPLOAD_LOCAL |
+                            SAVAGE_UPLOAD_TEX0   | SAVAGE_UPLOAD_TEX1  |
+			    SAVAGE_UPLOAD_FOGTBL | SAVAGE_UPLOAD_TEXGLOBAL))
         {
-   
-            /*SAVAGE_STATE_COPY(imesa);*/
-            /* update state to hw*/
-            if (imesa->driDrawable &&imesa->driDrawable->numClipRects ==0 )
-            {
-                return ;
-            }
+	    if (SAVAGE_DEBUG & DEBUG_VERBOSE_MSG)
+		fprintf (stderr, "... emitting state\n");
 	    if (imesa->savageScreen->chipset >= S3_SAVAGE4)
 		savageUpdateRegister_s4(imesa);
 	    else
 		savageUpdateRegister_s3d(imesa);
         }
 
-        imesa->sarea->dirty |= (imesa->dirty & 
-                                ~(SAVAGE_UPLOAD_TEX1|SAVAGE_UPLOAD_TEX0));
         imesa->dirty &= SAVAGE_UPLOAD_CLIPRECTS;
     }
 }
 
 
-
 static void savageDDInitState_s4( savageContextPtr imesa )
 {
 #if 1
@@ -1584,7 +1620,14 @@ static void savageDDInitState_s4( savageContextPtr imesa )
 
     imesa->regs.s4.zBufCtrl.ni.zCmpFunc = CF_Less;
     imesa->regs.s4.zBufCtrl.ni.wToZEn               = GL_TRUE;
-    /*imesa->regs.s4.ZBufCtrl.ni.floatZEn          = GL_TRUE;*/
+    if (imesa->float_depth) {
+	imesa->regs.s4.zBufCtrl.ni.zExpOffset =
+	    imesa->savageScreen->zpp == 2 ? 16 : 32;
+	imesa->regs.s4.zBufCtrl.ni.floatZEn = GL_TRUE;
+    } else {
+	imesa->regs.s4.zBufCtrl.ni.zExpOffset = 0;
+	imesa->regs.s4.zBufCtrl.ni.floatZEn = GL_FALSE;
+    }
     imesa->regs.s4.texBlendCtrl[0].ui            = TBC_NoTexMap;
     imesa->regs.s4.texBlendCtrl[1].ui            = TBC_NoTexMap1;
     imesa->regs.s4.drawCtrl0.ui         = 0;
@@ -1719,7 +1762,6 @@ static void savageDDInitState_s3d( savageContextPtr imesa )
 }
 void savageDDInitState( savageContextPtr imesa ) {
     memset (imesa->regs.ui, 0, SAVAGE_NR_REGS*sizeof(u_int32_t));
-    memset (imesa->oldRegs.ui, 0, SAVAGE_NR_REGS*sizeof(u_int32_t));
     memset (imesa->globalRegMask.ui, 0xff, SAVAGE_NR_REGS*sizeof(u_int32_t));
     if (imesa->savageScreen->chipset >= S3_SAVAGE4)
 	savageDDInitState_s4 (imesa);
@@ -1789,6 +1831,13 @@ void savageDDInitState( savageContextPtr imesa ) {
             imesa->readMap = (char *)imesa->apertureBase[TARGET_BACK];
         }
     }
+
+    memcpy (imesa->oldRegs.ui, imesa->regs.ui, SAVAGE_NR_REGS*sizeof(u_int32_t));
+
+    /* Emit the initial state to the (empty) command buffer. */
+    assert (imesa->cmdBuf.write == imesa->cmdBuf.base);
+    savageEmitOldState(imesa);
+    imesa->cmdBuf.start = imesa->cmdBuf.write;
 }
 
 

src/mesa/drivers/dri/savage/savagestate.h

@@ -28,43 +28,13 @@
 
 #include "savagecontext.h"
 
+void savageEmitOldState( savageContextPtr imesa );
+void savageEmitChangedState( savageContextPtr imesa );
+
 extern void savageDDUpdateHwState( GLcontext *ctx );
 extern void savageDDInitState( savageContextPtr imesa );
 extern void savageDDInitStateFuncs( GLcontext *ctx );
 extern void savageDDRenderStart(GLcontext *ctx);
 extern void savageDDRenderEnd(GLcontext *ctx);
 
-/*frank 2001/11/13 add macro for sarea state copy*/
-#if 0
-#define SAVAGE_STATE_COPY(ctx) { \
-ctx->sarea->setup[0]=ctx->Registers.DrawLocalCtrl.ui; \
-ctx->sarea->setup[1]=ctx->Registers.TexPalAddr.ui; \
-ctx->sarea->setup[2]=ctx->Registers.TexCtrl[0].ui; \
-ctx->sarea->setup[3]=ctx->Registers.TexCtrl[1].ui; \
-ctx->sarea->setup[4]=ctx->Registers.TexAddr[0].ui; \
-ctx->sarea->setup[5]=ctx->Registers.TexAddr[1].ui; \
-ctx->sarea->setup[6]=ctx->Registers.TexBlendCtrl[0].ui; \
-ctx->sarea->setup[7]=ctx->Registers.TexBlendCtrl[1].ui; \
-ctx->sarea->setup[8]=ctx->Registers.TexXprClr.ui; \
-ctx->sarea->setup[9]=ctx->Registers.TexDescr.ui; \
-ctx->sarea->setup[10]=ctx->Registers.FogTable.ni.ulEntry[0]; \
-ctx->sarea->setup[11]=ctx->Registers.FogTable.ni.ulEntry[1]; \
-ctx->sarea->setup[12]=ctx->Registers.FogTable.ni.ulEntry[2]; \
-ctx->sarea->setup[13]=ctx->Registers.FogTable.ni.ulEntry[3]; \
-ctx->sarea->setup[14]=ctx->Registers.FogTable.ni.ulEntry[4]; \
-ctx->sarea->setup[15]=ctx->Registers.FogTable.ni.ulEntry[5]; \
-ctx->sarea->setup[16]=ctx->Registers.FogTable.ni.ulEntry[6]; \
-ctx->sarea->setup[17]=ctx->Registers.FogTable.ni.ulEntry[7]; \
-ctx->sarea->setup[18]=ctx->Registers.FogCtrl.ui; \
-ctx->sarea->setup[19]=ctx->Registers.StencilCtrl.ui; \
-ctx->sarea->setup[20]=ctx->Registers.ZBufCtrl.ui; \
-ctx->sarea->setup[21]=ctx->Registers.ZBufOffset.ui; \
-ctx->sarea->setup[22]=ctx->Registers.DestCtrl.ui; \
-ctx->sarea->setup[23]=ctx->Registers.DrawCtrl0.ui; \
-ctx->sarea->setup[24]=ctx->Registers.DrawCtrl1.ui; \
-ctx->sarea->setup[25]=ctx->Registers.ZWatermarks.ui; \
-ctx->sarea->setup[26]=ctx->Registers.DestTexWatermarks.ui; \
-ctx->sarea->setup[27]=ctx->Registers.TexBlendColor.ui; \
-}
-#endif
 #endif

src/mesa/drivers/dri/savage/savagetex.c

@@ -606,7 +606,7 @@ void savagePrintGlobalLRU( savageContextPtr imesa , GLuint heap)
 {
    int i, j;
 
-   drm_savage_tex_region_t *list = imesa->sarea->texList[heap];
+   drm_tex_region_t *list = imesa->sarea->texList[heap];
    
 
    for (i = 0, j = SAVAGE_NR_TEX_REGIONS ; i < SAVAGE_NR_TEX_REGIONS ; i++) {
@@ -628,7 +628,7 @@ void savagePrintGlobalLRU( savageContextPtr imesa , GLuint heap)
 
 void savageResetGlobalLRU( savageContextPtr imesa, GLuint heap )
 {
-    drm_savage_tex_region_t *list = imesa->sarea->texList[heap];
+    drm_tex_region_t *list = imesa->sarea->texList[heap];
    int sz = 1 << imesa->savageScreen->logTextureGranularity[heap];
    int i;
 
@@ -661,7 +661,7 @@ static void savageUpdateTexLRU( savageContextPtr imesa, savageTextureObjectPtr t
    int logsz = imesa->savageScreen->logTextureGranularity[heap];
    int start = t->MemBlock->ofs >> logsz;
    int end = (t->MemBlock->ofs + t->MemBlock->size - 1) >> logsz;
-   drm_savage_tex_region_t *list = imesa->sarea->texList[heap];
+   drm_tex_region_t *list = imesa->sarea->texList[heap];
    
    imesa->texAge[heap] = ++imesa->sarea->texAge[heap];
 
@@ -787,16 +787,19 @@ int savageUploadTexImages( savageContextPtr imesa, savageTextureObjectPtr t )
       ofs = t->MemBlock->ofs;
       t->texParams.hwPhysAddress = imesa->savageScreen->textureOffset[heap] + ofs;
       t->BufAddr = (char *)((GLuint) imesa->savageScreen->texVirtual[heap] + ofs);
-      imesa->dirty |= SAVAGE_UPLOAD_CTX;
+      imesa->dirty |= SAVAGE_UPLOAD_GLOBAL; /* FIXME: really needed? */
    }
 
    /* Let the world know we've used this memory recently.
     */
+   LOCK_HARDWARE(imesa);
    savageUpdateTexLRU( imesa, t );
+   UNLOCK_HARDWARE(imesa);
 
    if (t->dirty_images) {
+      savageFlushVertices (imesa);
       LOCK_HARDWARE(imesa);
-      savageFlushVerticesLocked (imesa);
+      savageFlushCmdBufLocked (imesa, GL_FALSE);
       savageDmaFinish (imesa);
       if (SAVAGE_DEBUG & DEBUG_VERBOSE_LRU)
 	 fprintf(stderr, "*");
@@ -1436,8 +1439,7 @@ static void savageUpdateTextureState_s4( GLcontext *ctx )
    imesa->CurrentTexObj[1] = 0;   
    savageUpdateTex0State_s4( ctx );
    savageUpdateTex1State_s4( ctx );
-   imesa->dirty |= (SAVAGE_UPLOAD_CTX |
-		    SAVAGE_UPLOAD_TEX0 | 
+   imesa->dirty |= (SAVAGE_UPLOAD_TEX0 | 
 		    SAVAGE_UPLOAD_TEX1);
 }
 static void savageUpdateTextureState_s3d( GLcontext *ctx )
@@ -1446,8 +1448,7 @@ static void savageUpdateTextureState_s3d( GLcontext *ctx )
     if (imesa->CurrentTexObj[0]) imesa->CurrentTexObj[0]->bound &= ~1;
     imesa->CurrentTexObj[0] = 0;
     savageUpdateTexState_s3d( ctx );
-    imesa->dirty |= (SAVAGE_UPLOAD_CTX |
-		     SAVAGE_UPLOAD_TEX0);
+    imesa->dirty |= (SAVAGE_UPLOAD_TEX0);
 }
 void savageUpdateTextureState( GLcontext *ctx)
 {

src/mesa/drivers/dri/savage/savagetris.c

@@ -100,7 +100,7 @@ static void __inline__ savage_draw_triangle (savageContextPtr imesa,
 					     savageVertexPtr v1,
 					     savageVertexPtr v2) {
    GLuint vertsize = imesa->HwVertexSize;
-   u_int32_t *vb = savageAllocDmaLow (imesa, 3*4*vertsize);
+   u_int32_t *vb = savageAllocVtxBuf (imesa, 3*vertsize);
    GLuint j;
 
    EMIT_VERT (j, vb, vertsize, 0, v0);
@@ -114,7 +114,7 @@ static void __inline__ savage_draw_quad (savageContextPtr imesa,
 					 savageVertexPtr v2,
 					 savageVertexPtr v3) {
    GLuint vertsize = imesa->HwVertexSize;
-   u_int32_t *vb = savageAllocDmaLow (imesa, 6*4*vertsize);
+   u_int32_t *vb = savageAllocVtxBuf (imesa, 6*vertsize);
    GLuint j;
 
    EMIT_VERT (j, vb, vertsize, 0, v0);
@@ -128,7 +128,7 @@ static void __inline__ savage_draw_quad (savageContextPtr imesa,
 static __inline__ void savage_draw_point (savageContextPtr imesa,
 					  savageVertexPtr tmp) {
    GLuint vertsize = imesa->HwVertexSize;
-   u_int32_t *vb = savageAllocDmaLow (imesa, 6*4*vertsize);
+   u_int32_t *vb = savageAllocVtxBuf (imesa, 6*vertsize);
    const GLfloat x = tmp->v.x;
    const GLfloat y = tmp->v.y;
    const GLfloat sz = imesa->glCtx->Point._Size * .5;
@@ -163,7 +163,7 @@ static __inline__ void savage_draw_line (savageContextPtr imesa,
 					 savageVertexPtr v0,
 					 savageVertexPtr v1 ) {
    GLuint vertsize = imesa->HwVertexSize;
-   u_int32_t *vb = savageAllocDmaLow (imesa, 6*4*vertsize);
+   u_int32_t *vb = savageAllocVtxBuf (imesa, 6*vertsize);
    GLfloat width = imesa->glCtx->Line._Width;
    GLfloat dx, dy, ix, iy;
    GLuint j;
@@ -220,7 +220,7 @@ static void __inline__ savage_ptex_tri (savageContextPtr imesa,
 					savageVertexPtr v1,
 					savageVertexPtr v2) {
    GLuint vertsize = imesa->HwVertexSize;
-   u_int32_t *vb = savageAllocDmaLow (imesa, 3*4*vertsize);
+   u_int32_t *vb = savageAllocVtxBuf (imesa, 3*vertsize);
    savageVertex tmp;
    GLuint j;
 
@@ -233,7 +233,7 @@ static __inline__ void savage_ptex_line (savageContextPtr imesa,
 					 savageVertexPtr v0,
 					 savageVertexPtr v1 ) {
    GLuint vertsize = imesa->HwVertexSize;
-   u_int32_t *vb = savageAllocDmaLow (imesa, 6*4*vertsize);
+   u_int32_t *vb = savageAllocVtxBuf (imesa, 6*vertsize);
    GLfloat width = imesa->glCtx->Line._Width;
    GLfloat dx, dy, ix, iy;
    savageVertex tmp0, tmp1;
@@ -278,7 +278,7 @@ static __inline__ void savage_ptex_line (savageContextPtr imesa,
 static __inline__ void savage_ptex_point (savageContextPtr imesa,
 					  savageVertexPtr v0) {
    GLuint vertsize = imesa->HwVertexSize;
-   u_int32_t *vb = savageAllocDmaLow (imesa, 6*4*vertsize);
+   u_int32_t *vb = savageAllocVtxBuf (imesa, 6*vertsize);
    const GLfloat x = v0->v.x;
    const GLfloat y = v0->v.y;
    const GLfloat sz = imesa->glCtx->Point._Size * .5;
@@ -428,9 +428,9 @@ do {								\
    savageContextPtr imesa = SAVAGE_CONTEXT(ctx);		\
    GLuint color[n], spec[n];					\
    GLuint coloroffset =						\
-      ((imesa->DrawPrimitiveCmd & SAVAGE_HW_NO_W) ? 3 : 4);	\
+      ((imesa->skip & SAVAGE_SKIP_W) ? 3 : 4);			\
    GLboolean specoffset =					\
-      ((imesa->DrawPrimitiveCmd & SAVAGE_HW_NO_CS) ? 0 : coloroffset+1);\
+      ((imesa->skip & SAVAGE_SKIP_C1) ? 0 : coloroffset+1);	\
    (void) color; (void) spec; (void) coloroffset; (void) specoffset;
 
 /***********************************************************************
@@ -756,6 +756,16 @@ static void savageChooseRenderState(GLcontext *ctx)
 
       imesa->RenderIndex = index;
    }
+
+   if (imesa->savageScreen->chipset < S3_SAVAGE4 && (flags & DD_FLATSHADE)) {
+      if (imesa->HwPrim != SAVAGE_PRIM_TRILIST_201)
+	 savageFlushVertices(imesa);
+      imesa->HwPrim = SAVAGE_PRIM_TRILIST_201;
+   } else {
+      if (imesa->HwPrim != SAVAGE_PRIM_TRILIST)
+	 savageFlushVertices(imesa);
+      imesa->HwPrim = SAVAGE_PRIM_TRILIST;
+   }
 }
 
 /**********************************************************************/
@@ -766,6 +776,9 @@ static void savageRunPipeline( GLcontext *ctx )
 {
    savageContextPtr imesa = SAVAGE_CONTEXT(ctx);
 
+   if (imesa->no_rast)
+      FALLBACK(ctx, SAVAGE_FALLBACK_NORAST, GL_TRUE);
+
    if (imesa->new_state)
       savageDDUpdateHwState( ctx );
 
@@ -777,6 +790,9 @@ static void savageRunPipeline( GLcontext *ctx )
    }
 
    _tnl_run_pipeline( ctx );
+
+   if (imesa->no_rast)
+      FALLBACK(ctx, SAVAGE_FALLBACK_NORAST, GL_FALSE);
 }
 
 /**********************************************************************/
@@ -858,7 +874,7 @@ do {									\
    imesa->vertex_attrs[imesa->vertex_attr_count].format = (STYLE);	\
    imesa->vertex_attr_count++;						\
    setupIndex |= (INDEX);						\
-   drawCmd &= ~(SKIP);							\
+   skip &= ~(SKIP);							\
 } while (0)
 
 #define EMIT_PAD( N )							\
@@ -891,11 +907,8 @@ static void savageRenderStart( GLcontext *ctx )
    struct vertex_buffer *VB = &tnl->vb;
    GLuint index = tnl->render_inputs;
    GLuint setupIndex = SAVAGE_EMIT_XYZ;
-   GLuint drawCmd = SAVAGE_HW_SKIPFLAGS;
+   GLubyte skip;
    GLboolean ptexHack;
-   if (imesa->savageScreen->chipset < S3_SAVAGE4)
-      drawCmd &= ~SAVAGE_HW_NO_UV1;
-   drawCmd &= ~SAVAGE_HW_NO_Z; /* all mesa vertices have a z coordinate */
 
    /* Check if we need to apply the ptex hack. Choose a new render
     * state if necessary. (Note: this can't be done in
@@ -917,66 +930,115 @@ static void savageRenderStart( GLcontext *ctx )
    VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
    imesa->vertex_attr_count = 0;
  
-   /* EMIT_ATTR's must be in order as they tell t_vertex.c how to
-    * build up a hardware vertex.
-    */
-   if ((index & _TNL_BITS_TEX_ANY) || !(ctx->_TriangleCaps & DD_FLATSHADE)) {
-      EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, SAVAGE_EMIT_W, SAVAGE_HW_NO_W );
-   }
-   else {
-      EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, 0, 0 );
-   }
+   if (imesa->savageScreen->chipset < S3_SAVAGE4) {
+      skip = SAVAGE_SKIP_ALL_S3D;
+      skip &= ~SAVAGE_SKIP_Z; /* all mesa vertices have a z coordinate */
 
-   /* t_context.c always includes a diffuse color */
-   EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, SAVAGE_EMIT_C0, SAVAGE_HW_NO_CD );
+      /* EMIT_ATTR's must be in order as they tell t_vertex.c how to
+       * build up a hardware vertex.
+       */
+      if ((index & _TNL_BITS_TEX_ANY) || !(ctx->_TriangleCaps & DD_FLATSHADE)) {
+	 EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, SAVAGE_EMIT_W, SAVAGE_SKIP_W );
+      }
+      else {
+	 EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, 0, 0 );
+	 EMIT_PAD( 4 );
+	 skip &= ~SAVAGE_SKIP_W;
+      }
+
+      /* t_context.c always includes a diffuse color */
+      EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, SAVAGE_EMIT_C0, SAVAGE_SKIP_C0 );
       
-   if (index & (_TNL_BIT_COLOR1|_TNL_BIT_FOG)) {
       if ((index & _TNL_BIT_COLOR1))
-	 EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_3UB_3F_BGR, SAVAGE_EMIT_C1, SAVAGE_HW_NO_CS );
+	 EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_3UB_3F_BGR, SAVAGE_EMIT_C1, SAVAGE_SKIP_C1 );
       else
 	 EMIT_PAD( 3 );
       if ((index & _TNL_BIT_FOG))
-	 EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1UB_1F, SAVAGE_EMIT_FOG, SAVAGE_HW_NO_CS );
+	 EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1UB_1F, SAVAGE_EMIT_FOG, SAVAGE_SKIP_C1 );
       else
 	 EMIT_PAD( 1 );
-   }
+      skip &= ~SAVAGE_SKIP_C1;
 
-   if (index & _TNL_BIT_TEX(0)) {
-      if (ptexHack)
-	 EMIT_ATTR( _TNL_ATTRIB_TEX0, EMIT_3F_XYW, SAVAGE_EMIT_STQ0, SAVAGE_HW_NO_UV0);
-      else if (VB->TexCoordPtr[0]->size == 4)
-	 assert (0); /* should be caught by savageCheckPTexHack */
-      else if (VB->TexCoordPtr[0]->size >= 2)
-	 /* The chromium menu emits some 3D tex coords even though no
-	  * 3D texture is enabled. Ignore the 3rd coordinate. */
-	 EMIT_ATTR( _TNL_ATTRIB_TEX0, EMIT_2F, SAVAGE_EMIT_ST0, SAVAGE_HW_NO_UV0 );
+      if (index & _TNL_BIT_TEX(0)) {
+	 if (ptexHack)
+	    EMIT_ATTR( _TNL_ATTRIB_TEX0, EMIT_3F_XYW, SAVAGE_EMIT_STQ0, SAVAGE_SKIP_ST0);
+	 else if (VB->TexCoordPtr[0]->size == 4)
+	    assert (0); /* should be caught by savageCheckPTexHack */
+	 else if (VB->TexCoordPtr[0]->size >= 2)
+	    /* The chromium menu emits some 3D tex coords even though no
+	     * 3D texture is enabled. Ignore the 3rd coordinate. */
+	    EMIT_ATTR( _TNL_ATTRIB_TEX0, EMIT_2F, SAVAGE_EMIT_ST0, SAVAGE_SKIP_ST0 );
+	 else if (VB->TexCoordPtr[0]->size == 1) {
+	    EMIT_ATTR( _TNL_ATTRIB_TEX0, EMIT_1F, SAVAGE_EMIT_S0, SAVAGE_SKIP_S0 );
+	    EMIT_PAD( 4 );
+	 } else
+	    EMIT_PAD( 8 );
+      } else
+	 EMIT_PAD( 8 );
+      skip &= ~SAVAGE_SKIP_ST0;
+   } else {
+      skip = SAVAGE_SKIP_ALL_S4;
+      skip &= ~SAVAGE_SKIP_Z; /* all mesa vertices have a z coordinate */
+
+      /* EMIT_ATTR's must be in order as they tell t_vertex.c how to
+       * build up a hardware vertex.
+       */
+      if ((index & _TNL_BITS_TEX_ANY) || !(ctx->_TriangleCaps & DD_FLATSHADE))
+	 EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, SAVAGE_EMIT_W, SAVAGE_SKIP_W );
       else
-	 EMIT_ATTR( _TNL_ATTRIB_TEX0, EMIT_1F, SAVAGE_EMIT_S0, SAVAGE_HW_NO_U0 );
-   }
-   if (index & _TNL_BIT_TEX(1)) {
-      if (VB->TexCoordPtr[1]->size == 4)
-	 /* Projective textures are not supported by the hardware */
-	 assert (0); /* should be caught by savageCheckPTexHack */
-      else if (VB->TexCoordPtr[1]->size >= 2)
-	 EMIT_ATTR( _TNL_ATTRIB_TEX1, EMIT_2F, SAVAGE_EMIT_ST1, SAVAGE_HW_NO_UV1 );
-      else
-	 EMIT_ATTR( _TNL_ATTRIB_TEX1, EMIT_1F, SAVAGE_EMIT_S1, SAVAGE_HW_NO_U1 );
+	 EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, 0, 0 );
+
+      /* t_context.c always includes a diffuse color */
+      EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, SAVAGE_EMIT_C0, SAVAGE_SKIP_C0 );
+      
+      if (index & (_TNL_BIT_COLOR1|_TNL_BIT_FOG)) {
+	 if ((index & _TNL_BIT_COLOR1))
+	    EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_3UB_3F_BGR, SAVAGE_EMIT_C1, SAVAGE_SKIP_C1 );
+	 else
+	    EMIT_PAD( 3 );
+	 if ((index & _TNL_BIT_FOG))
+	    EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1UB_1F, SAVAGE_EMIT_FOG, SAVAGE_SKIP_C1 );
+	 else
+	    EMIT_PAD( 1 );
+      }
+
+      if (index & _TNL_BIT_TEX(0)) {
+	 if (ptexHack)
+	    EMIT_ATTR( _TNL_ATTRIB_TEX0, EMIT_3F_XYW, SAVAGE_EMIT_STQ0, SAVAGE_SKIP_ST0);
+	 else if (VB->TexCoordPtr[0]->size == 4)
+	    assert (0); /* should be caught by savageCheckPTexHack */
+	 else if (VB->TexCoordPtr[0]->size >= 2)
+	    /* The chromium menu emits some 3D tex coords even though no
+	     * 3D texture is enabled. Ignore the 3rd coordinate. */
+	    EMIT_ATTR( _TNL_ATTRIB_TEX0, EMIT_2F, SAVAGE_EMIT_ST0, SAVAGE_SKIP_ST0 );
+	 else
+	    EMIT_ATTR( _TNL_ATTRIB_TEX0, EMIT_1F, SAVAGE_EMIT_S0, SAVAGE_SKIP_S0 );
+      }
+      if (index & _TNL_BIT_TEX(1)) {
+	 if (VB->TexCoordPtr[1]->size == 4)
+	    /* projective textures are not supported by the hardware */
+	    assert (0); /* should be caught by savageCheckPTexHack */
+	 else if (VB->TexCoordPtr[1]->size >= 2)
+	    EMIT_ATTR( _TNL_ATTRIB_TEX1, EMIT_2F, SAVAGE_EMIT_ST1, SAVAGE_SKIP_ST1 );
+	 else
+	    EMIT_ATTR( _TNL_ATTRIB_TEX1, EMIT_1F, SAVAGE_EMIT_S1, SAVAGE_SKIP_S1 );
+      }
    }
 
    /* Need to change the vertex emit code if the SetupIndex changed or
     * is set for the first time (indicated by vertex_size == 0). */
    if (setupIndex != imesa->SetupIndex || imesa->vertex_size == 0) {
-      imesa->vertex_size = 
+      GLuint hwVertexSize;
+      imesa->vertex_size =
 	 _tnl_install_attrs( ctx, 
 			     imesa->vertex_attrs, 
 			     imesa->vertex_attr_count,
 			     imesa->hw_viewport, 0 );
       imesa->vertex_size >>= 2;
       imesa->SetupIndex = setupIndex;
+      imesa->skip = skip;
 
-      imesa->DrawPrimitiveCmd = drawCmd;
-      imesa->HwVertexSize = imesa->vertex_size;
-
+      hwVertexSize = imesa->vertex_size;
       if (setupIndex & SAVAGE_EMIT_Q0) {
 	 /* The vertex setup code emits homogenous texture
 	  * coordinates. They are converted to normal 2D coords by
@@ -984,10 +1046,28 @@ static void savageRenderStart( GLcontext *ctx )
 	  * vertex sizes. Functions that emit vertices to the hardware
 	  * need to use HwVertexSize, anything that manipulates the
 	  * vertices generated by t_vertex uses vertex_size. */
-	 imesa->HwVertexSize--;
+	 hwVertexSize--;
 	 assert (imesa->ptexHack);
       } else
 	 assert (!imesa->ptexHack);
+
+      if (hwVertexSize != imesa->HwVertexSize) {
+	 /* Changing the vertex size: flush vertex and command buffer and
+	  * discard the DMA buffer, if we were using one. */
+	 savageFlushVertices(imesa);
+	 savageFlushCmdBuf(imesa, GL_TRUE);
+	 if (hwVertexSize == 8) {
+	    if (SAVAGE_DEBUG & DEBUG_DMA)
+	       fprintf (stderr, "Using DMA, skip=0x%02x\n", skip);
+	    /* we can use vertex dma */
+	    imesa->vtxBuf = &imesa->dmaVtxBuf;
+	 } else {
+	    if (SAVAGE_DEBUG & DEBUG_DMA)
+	       fprintf (stderr, "Not using DMA, skip=0x%02x\n", skip);
+	    imesa->vtxBuf = &imesa->clientVtxBuf;
+	 }
+	 imesa->HwVertexSize = hwVertexSize;
+      }
    }
 
    /* Update hardware state and get the lock */
@@ -1040,7 +1120,6 @@ void savageFallback( GLcontext *ctx, GLuint bit, GLboolean mode )
       imesa->Fallback |= bit;
       if (oldfallback == 0) {
 	 /* the first fallback */
-	 FLUSH_BATCH( imesa );
 	 _swsetup_Wakeup( ctx );
 	 imesa->RenderIndex = ~0;
       }

src/mesa/drivers/dri/savage/server/savage_dri.h

@@ -29,6 +29,11 @@
 #include "xf86drm.h"
 #include "drm.h"
 
+/* Totals 2 Mbytes which equals 2^16 32-byte vertices divided among up
+ * to 32 clients. */
+#define SAVAGE_NUM_BUFFERS 32
+#define SAVAGE_BUFFER_SIZE (1 << 16) /* 64k */
+
 #define SAVAGE_DEFAULT_AGP_MODE     1
 #define SAVAGE_MAX_AGP_MODE         4
 
@@ -68,14 +73,17 @@ typedef struct _server{
    unsigned int frontOffset;
    unsigned int frontPitch;
    unsigned int frontbufferSize;
+   unsigned int frontBitmapDesc;
    
    unsigned int backOffset;
    unsigned int backPitch;
    unsigned int backbufferSize;
+   unsigned int backBitmapDesc;
 
    unsigned int depthOffset;
    unsigned int depthPitch;
    unsigned int depthbufferSize;
+   unsigned int depthBitmapDesc;
 
    unsigned int textureOffset;
    int textureSize;
@@ -89,12 +97,7 @@ typedef struct _server{
    drmRegion status;
 
    /* AGP mappings */
-#if 0
-   drmRegion warp;
-   drmRegion primary;
    drmRegion buffers;
-#endif
-
    drmRegion agpTextures;
    int logAgpTextureGranularity;
 
@@ -114,71 +117,41 @@ typedef struct {
    int cpp;
    int zpp;
 
-   int agpMode;
+   int agpMode; /* 0 for PCI cards */
+
+   unsigned int sarea_priv_offset;
+
+   unsigned int bufferSize; /* size of DMA buffers */
    
-   drm_handle_t frontbuffer;
    unsigned int frontbufferSize;
    unsigned int frontOffset;
-   unsigned int frontPitch;
-   unsigned int frontBitmapDesc;   /*Bitmap Descriptior*/ 
-   unsigned int IsfrontTiled;
 
-   drm_handle_t backbuffer;
    unsigned int backbufferSize;
    unsigned int backOffset;
-   unsigned int backPitch;
-   unsigned int backBitmapDesc;   /*Bitmap Descriptior*/
 
-   drm_handle_t depthbuffer;
    unsigned int depthbufferSize;
    unsigned int depthOffset;
-   unsigned int depthPitch;
-   unsigned int depthBitmapDesc;   /*Bitmap Descriptior*/
 
-
-
-   drm_handle_t textures;
-   drm_handle_t xvmcSurfHandle;
    unsigned int textureOffset;
    unsigned int textureSize;
    int logTextureGranularity;
 
-   /* Allow calculation of setup dma addresses.
-    */
-   unsigned int agpBufferOffset;
-
-   unsigned int agpTextureOffset;
-   unsigned int agpTextureSize;
-   drmRegion agpTextures;
-   int logAgpTextureGranularity;
-
-/*   unsigned int mAccess;*/
-
-   drmRegion aperture;
+   /* Linear aperture */
+   drm_handle_t apertureHandle;
+   unsigned int apertureSize;
    unsigned int aperturePitch;    /* in byte */
 
+   /* Status page (probably not needed, but no harm, read-only) */
+   drm_handle_t statusHandle;
+   unsigned int statusSize;
 
-   drmRegion registers;
-   drmRegion BCIcmdBuf;
-   drmRegion status;
+   /* AGP textures */
+   drm_handle_t agpTextureHandle;
+   unsigned int agpTextureSize;
+   int logAgpTextureGranularity;
 
-#if 0
-   drmRegion primary;
-   drmRegion buffers;
-#endif
-  /*For shadow status*/
-  unsigned long sareaPhysAddr;
-
-   unsigned int sarea_priv_offset;
-  int shadowStatus;
+   /* Not sure about this one */
+   drm_handle_t xvmcSurfHandle; /* ? */
 } SAVAGEDRIRec, *SAVAGEDRIPtr;
 
 #endif
-
-
-
-
-
-
-
-

src/mesa/drivers/dri/trident/trident_context.c

@@ -40,7 +40,7 @@
 #include "matrix.h"
 #include "extensions.h"
 #if defined(USE_X86_ASM)
-#include "X86/common_x86_asm.h"
+#include "x86/common_x86_asm.h"
 #endif
 #include "simple_list.h"
 #include "mm.h"

src/mesa/drivers/dri/trident/trident_dd.c

@@ -27,7 +27,7 @@
 #include "trident_context.h"
 #include "trident_lock.h"
 #if defined(USE_X86_ASM)
-#include "X86/common_x86_asm.h"
+#include "x86/common_x86_asm.h"
 #endif
 
 #include "swrast/swrast.h"

src/mesa/drivers/dri/unichrome/Makefile

@@ -29,7 +29,6 @@ DRIVER_SOURCES = \
 	via_texmem.c \
 	via_texstate.c \
 	via_tris.c \
-	via_vb.c \
 	via_texcombine.c \
 	xf86drmVIA.c
 

src/mesa/drivers/dri/unichrome/via_context.c

@@ -52,7 +52,6 @@
 #include "via_tex.h"
 #include "via_span.h"
 #include "via_tris.h"
-#include "via_vb.h"
 #include "via_ioctl.h"
 #include "via_fb.h"
 #include "via_regs.h"
@@ -257,6 +256,8 @@ static const char * const card_extensions[] =
    "GL_EXT_texture_env_combine",
    "GL_EXT_texture_env_dot3",
    "GL_EXT_texture_lod_bias",
+   "GL_EXT_secondary_color",
+   "GL_EXT_fog_coord",
    "GL_NV_blend_square",
    NULL
 };
@@ -288,7 +289,7 @@ AllocateDmaBuffer(const GLvisual *visual, viaContextPtr vmesa)
         return GL_FALSE;
 
     vmesa->dmaLow = 0;
-    vmesa->dmaCliprectAddr = 0;
+    vmesa->dmaCliprectAddr = ~0;
     return GL_TRUE;
 }
 
@@ -356,6 +357,7 @@ viaCreateContext(const __GLcontextModes *mesaVis,
        vmesa->depth_max = (GLfloat)0xffff;
        vmesa->depth_clear_mask = 0xf << 28;
        vmesa->ClearDepth = 0xffff;
+       vmesa->polygon_offset_scale = 1.0 / vmesa->depth_max;
        break;
     case 24:
        vmesa->hasDepth = GL_TRUE;
@@ -370,6 +372,7 @@ viaCreateContext(const __GLcontextModes *mesaVis,
        vmesa->have_hw_stencil = GL_TRUE;
        vmesa->stencilBits = mesaVis->stencilBits;
        vmesa->stencil_clear_mask = 0x1 << 28;
+       vmesa->polygon_offset_scale = 2.0 / vmesa->depth_max;
        break;
     case 32:
        vmesa->hasDepth = GL_TRUE;
@@ -380,6 +383,7 @@ viaCreateContext(const __GLcontextModes *mesaVis,
        vmesa->depth_clear_mask = 0;
        vmesa->ClearDepth = 0xffffffff;
        vmesa->depth_clear_mask = 0xf << 28;
+       vmesa->polygon_offset_scale = 2.0 / vmesa->depth_max;
        break;
     default:
        assert(0); 
@@ -465,6 +469,7 @@ viaCreateContext(const __GLcontextModes *mesaVis,
 
     vmesa->texHeap = mmInit(0, viaScreen->textureSize);
     vmesa->renderIndex = ~0;
+    vmesa->setupIndex = ~0;
 
     make_empty_list(&vmesa->TexObjList);
     make_empty_list(&vmesa->SwappedOut);
@@ -489,7 +494,6 @@ viaCreateContext(const __GLcontextModes *mesaVis,
     viaInitTriFuncs(ctx);
     viaInitSpanFuncs(ctx);
     viaInitIoctlFuncs(ctx);
-    viaInitVB(ctx);
     viaInitState(ctx);
         
 #ifdef DEBUG
@@ -552,7 +556,6 @@ viaDestroyContext(__DRIcontextPrivate *driContextPriv)
         _tnl_DestroyContext(vmesa->glCtx);
         _ac_DestroyContext(vmesa->glCtx);
         _swrast_DestroyContext(vmesa->glCtx);
-        viaFreeVB(vmesa->glCtx);
 	FreeBuffer(vmesa);
         /* free the Mesa context */
 	_mesa_destroy_context(vmesa->glCtx);
@@ -658,8 +661,16 @@ viaMakeCurrent(__DRIcontextPrivate *driContextPriv,
                             (GLframebuffer *)driDrawPriv->driverPrivate,
                             (GLframebuffer *)driReadPriv->driverPrivate);
 	if (VIA_DEBUG) fprintf(stderr, "Context %d MakeCurrent\n", vmesa->hHWContext);
-        viaXMesaWindowMoved(vmesa);
-
+	
+	/* These are probably needed only the first time a context is
+	 * made current:
+	 */
+	viaXMesaWindowMoved(vmesa);
+	ctx->Driver.Scissor(ctx, 
+			    ctx->Scissor.X,
+			    ctx->Scissor.Y,
+			    ctx->Scissor.Width,
+			    ctx->Scissor.Height);
 
     }
     else {

src/mesa/drivers/dri/unichrome/via_context.h

@@ -35,11 +35,13 @@ typedef struct via_texture_object_t *viaTextureObjectPtr;
 #include "mtypes.h"
 #include "drm.h"
 #include "mm.h"
+#include "tnl/t_vertex.h"
 
 #include "via_screen.h"
 #include "via_tex.h"
 #include "via_common.h"
 #include "xf86drmVIA.h"
+
 #define VIA_FALLBACK_TEXTURE           	0x1
 #define VIA_FALLBACK_DRAW_BUFFER       	0x2
 #define VIA_FALLBACK_READ_BUFFER       	0x4
@@ -51,9 +53,10 @@ typedef struct via_texture_object_t *viaTextureObjectPtr;
 #define VIA_FALLBACK_BLEND_EQ          	0x200
 #define VIA_FALLBACK_BLEND_FUNC        	0x400
 #define VIA_FALLBACK_USER_DISABLE      	0x800
+#define VIA_FALLBACK_PROJ_TEXTURE      	0x1000
 
-#define VIA_DMA_BUFSIZ                  5000
-#define VIA_DMA_HIGHWATER               (VIA_DMA_BUFSIZ - 256)
+#define VIA_DMA_BUFSIZ                  4096
+#define VIA_DMA_HIGHWATER               (VIA_DMA_BUFSIZ - 128)
 
 #define VIA_NO_CLIPRECTS 0x1
 
@@ -103,6 +106,7 @@ struct via_context_t {
    GLuint depth_clear_mask;
    GLuint stencil_clear_mask;
    GLfloat depth_max;
+   GLfloat polygon_offset_scale;
 
     GLubyte    *dma;
     viaRegion tex;
@@ -120,11 +124,15 @@ struct via_context_t {
      */
     GLuint Fallback;
 
-    /* State for via_vb.c and via_tris.c.
+    /* State for via_tris.c.
      */
     GLuint newState;            /* _NEW_* flags */
     GLuint newEmitState;            /* _NEW_* flags */
-    GLuint setupNewInputs;
+    GLuint newRenderState;            /* _NEW_* flags */
+
+    struct tnl_attr_map vertex_attrs[VERT_ATTRIB_MAX];
+    GLuint vertex_attr_count;
+
     GLuint setupIndex;
     GLuint renderIndex;
     GLmatrix ViewportMatrix;
@@ -203,15 +211,18 @@ struct via_context_t {
     GLuint regHTXnTBLRAa_1;
     GLuint regHTXnTBLRFog_1;
 
-    /* Hardware state 
-     */
-    GLuint dirty;             
     int vertexSize;
-    int vertexFormat;
+    int hwVertexSize;
+    GLboolean ptexHack;
+    int coloroffset;
+    int specoffset;
+
     GLint lastStamp;
 
     GLenum TexEnvImageFmt[2];
     GLuint ClearColor;
+    GLuint ClearMask;
+
     /* DRI stuff
      */
     GLuint needClip;

src/mesa/drivers/dri/unichrome/via_ioctl.c

@@ -3,7 +3,7 @@
  * Copyright 2001-2003 S3 Graphics, Inc. All Rights Reserved.
  *
  * Permission is hereby granted, free of charge, to any person obtaining a
-v * copy of this software and associated documentation files (the "Software"),
+ * copy of this software and associated documentation files (the "Software"),
  * to deal in the Software without restriction, including without limitation
  * the rights to use, copy, modify, merge, publish, distribute, sub license,
  * and/or sell copies of the Software, and to permit persons to whom the
@@ -90,77 +90,61 @@ void viaCheckDma(viaContextPtr vmesa, GLuint bytes)
 } while (0)
 
 
-static void viaBlit(viaContextPtr vmesa, GLuint bpp,GLuint srcBase,
-		    GLuint srcPitch,GLuint dstBase,GLuint dstPitch,
-		    GLuint w,GLuint h,int xdir,int ydir, GLuint blitMode, 
+static void viaBlit(viaContextPtr vmesa, GLuint bpp,
+		    GLuint srcBase, GLuint srcPitch, 
+		    GLuint dstBase, GLuint dstPitch,
+		    GLuint w, GLuint h, 
+		    GLuint blitMode, 
 		    GLuint color, GLuint nMask ) 
 {
 
-    GLuint dwGEMode = 0, srcY=0, srcX, dstY=0, dstX;
-    GLuint cmd;
+    GLuint dwGEMode, srcX, dstX, cmd;
     RING_VARS;
 
     if (VIA_DEBUG)
-       fprintf(stderr, "%s bpp %d src %x/%x dst %x/%x w %d h %d dir %d,%d mode: %x color: 0x%08x mask 0x%08x\n",
-	       __FUNCTION__, bpp, srcBase, srcPitch, dstBase, dstPitch, w,h, xdir, ydir, blitMode, color, nMask);
+       fprintf(stderr, "%s bpp %d src %x/%x dst %x/%x w %d h %d  mode: %x color: 0x%08x mask 0x%08x\n",
+	       __FUNCTION__, bpp, srcBase, srcPitch, dstBase, dstPitch, w,h, blitMode, color, nMask);
 
 
     if (!w || !h)
         return;
 
-    srcX = srcBase & 31;
-    dstX = dstBase & 31;
     switch (bpp) {
     case 16:
-        dwGEMode |= VIA_GEM_16bpp;
-	srcX >>= 1;
-	dstX >>= 1;
+        dwGEMode = VIA_GEM_16bpp;
+	srcX = (srcBase & 0x1f) >> 1;
+	dstX = (dstBase & 0x1f) >> 1;
         break;
     case 32:
-        dwGEMode |= VIA_GEM_32bpp;
-	srcX >>= 2;
-	dstX >>= 2;
+        dwGEMode = VIA_GEM_32bpp;
+	srcX = (srcBase & 0x1f) >> 2;
+	dstX = (dstBase & 0x1f) >> 2;
 	break;
     default:
-        dwGEMode |= VIA_GEM_8bpp;
-        break;
-    }
-
-
-    cmd = 0; 
-
-    if (xdir < 0) {
-        cmd |= VIA_GEC_DECX;
-        srcX += (w - 1);
-        dstX += (w - 1);
-    }
-    if (ydir < 0) {
-        cmd |= VIA_GEC_DECY;
-        srcY += (h - 1);
-        dstY += (h - 1);
+        return;
     }
 
     switch(blitMode) {
     case VIA_BLIT_FILL:
-        BEGIN_RING((2 + 9) * 2);
-	SetReg2DAGP(VIA_REG_GEMODE, dwGEMode);
-	SetReg2DAGP( VIA_REG_FGCOLOR, color);
-	cmd |= VIA_GEC_BLT | VIA_GEC_FIXCOLOR_PAT | (VIA_BLIT_FILL << 24);
+	cmd = VIA_GEC_BLT | VIA_GEC_FIXCOLOR_PAT | (VIA_BLIT_FILL << 24);
 	break;
     case VIA_BLIT_COPY:
-        BEGIN_RING((2 + 9) * 2);
-	SetReg2DAGP(VIA_REG_GEMODE, dwGEMode);
-	SetReg2DAGP( VIA_REG_KEYCONTROL, 0x0);
-	cmd |= VIA_GEC_BLT | (VIA_BLIT_COPY << 24);
+	cmd = VIA_GEC_BLT | (VIA_BLIT_COPY << 24);
+	break;
+    default:
+        return;
     }	
 
+    BEGIN_RING(22);
+    SetReg2DAGP( VIA_REG_GEMODE, dwGEMode);
+    SetReg2DAGP( VIA_REG_FGCOLOR, color);
     SetReg2DAGP( 0x2C, nMask);
-    SetReg2DAGP( VIA_REG_SRCBASE, (srcBase & ~31) >> 3);
-    SetReg2DAGP( VIA_REG_DSTBASE, (dstBase & ~31) >> 3);
+    SetReg2DAGP( VIA_REG_SRCBASE, (srcBase & ~0x1f) >> 3);
+    SetReg2DAGP( VIA_REG_DSTBASE, (dstBase & ~0x1f) >> 3);
     SetReg2DAGP( VIA_REG_PITCH, VIA_PITCH_ENABLE |
-	       (srcPitch >> 3) | (((dstPitch) >> 3) << 16));
-    SetReg2DAGP( VIA_REG_SRCPOS, ((srcY << 16) | srcX));
-    SetReg2DAGP( VIA_REG_DSTPOS, ((dstY << 16) | dstX));
+	       (srcPitch >> 3) | ((dstPitch >> 3) << 16));
+    SetReg2DAGP( VIA_REG_SRCPOS, srcX);
+    SetReg2DAGP( VIA_REG_DSTPOS, dstX);
     SetReg2DAGP( VIA_REG_DIMENSION, (((h - 1) << 16) | (w - 1)));
     SetReg2DAGP( VIA_REG_GECMD, cmd);
     SetReg2DAGP( 0x2C, 0x00000000);
@@ -192,7 +176,6 @@ static void viaFillBuffer(viaContextPtr vmesa,
 	      offset, buffer->pitch,
 	      offset, buffer->pitch, 
 	      w, h,
-	      0, 0, 
 	      VIA_BLIT_FILL, pixel, mask); 
    }
 }
@@ -204,7 +187,6 @@ static void viaClear(GLcontext *ctx, GLbitfield mask, GLboolean all,
 {
    viaContextPtr vmesa = VIA_CONTEXT(ctx);
    __DRIdrawablePrivate *dPriv = vmesa->driDrawable;
-   const GLuint colorMask = *((GLuint *)&ctx->Color.ColorMask);
    int flag = 0;
    GLuint i = 0;
    GLuint clear_depth_mask = 0xf << 28;
@@ -212,12 +194,12 @@ static void viaClear(GLcontext *ctx, GLbitfield mask, GLboolean all,
 
    VIA_FLUSH_DMA(vmesa);
 
-   if ((mask & DD_FRONT_LEFT_BIT) && colorMask == ~0) {
+   if (mask & DD_FRONT_LEFT_BIT) {
       flag |= VIA_FRONT;
       mask &= ~DD_FRONT_LEFT_BIT;
    }
 
-   if ((mask & DD_BACK_LEFT_BIT) && colorMask == ~0) {
+   if (mask & DD_BACK_LEFT_BIT) {
       flag |= VIA_BACK;	
       mask &= ~DD_BACK_LEFT_BIT;
    }
@@ -239,9 +221,9 @@ static void viaClear(GLcontext *ctx, GLbitfield mask, GLboolean all,
 	    mask &= ~DD_STENCIL_BIT;
 	 }
 	 else {
-	    fprintf(stderr, "XXX: Clear stencil writemask %x -- need triangles (or a ROP?)\n", 
-		    ctx->Stencil.WriteMask[0]);
-	    /* Fixme - clear with triangles */
+	    if (VIA_DEBUG)
+	       fprintf(stderr, "XXX: Clear stencil writemask %x -- need triangles (or a ROP?)\n", 
+		       ctx->Stencil.WriteMask[0]);
 	 }
       }
    }
@@ -293,11 +275,11 @@ static void viaClear(GLcontext *ctx, GLbitfield mask, GLboolean all,
       }
 	    
       if (flag & VIA_FRONT) {
-	 viaFillBuffer(vmesa, &vmesa->front, boxes, nr, vmesa->ClearColor, 0);
+	 viaFillBuffer(vmesa, &vmesa->front, boxes, nr, vmesa->ClearColor, vmesa->ClearMask);
       } 
 		
       if (flag & VIA_BACK) {
-	 viaFillBuffer(vmesa, &vmesa->back, boxes, nr, vmesa->ClearColor, 0); /* FIXME: masks */
+	 viaFillBuffer(vmesa, &vmesa->back, boxes, nr, vmesa->ClearColor, vmesa->ClearMask);
       }
 
       if (flag & VIA_DEPTH) {
@@ -341,7 +323,7 @@ static void viaDoSwapBuffers(viaContextPtr vmesa,
 	      src, back->pitch,
 	      dest, front->pitch,
 	      w, h,
-	      0,0,VIA_BLIT_COPY, 0, 0); 
+	      VIA_BLIT_COPY, 0, 0); 
    }
 }
 
@@ -513,11 +495,9 @@ static void via_emit_cliprect(viaContextPtr vmesa,
    GLuint pitch = buffer->pitch;
    GLuint offset = buffer->orig;
 
-   GLuint clipL = b->x1 - vmesa->drawX;
-   GLuint clipR = b->x2 - vmesa->drawX;
-   GLuint clipT = b->y1 - vmesa->drawY;
-   GLuint clipB = b->y2 - vmesa->drawY;
-	    
+   if (0)
+      fprintf(stderr, "emit cliprect for box %d,%d %d,%d\n", b->x1, b->y1, b->x2, b->y2);
+
    vb[0] = HC_HEADER2;
    vb[1] = (HC_ParaType_NotTex << 16);
 
@@ -526,8 +506,8 @@ static void via_emit_cliprect(viaContextPtr vmesa,
       vb[3] = (HC_SubA_HClipLR << 24) | 0x0;
    }
    else {
-      vb[2] = (HC_SubA_HClipTB << 24) | (clipT << 12) | clipB;
-      vb[3] = (HC_SubA_HClipLR << 24) | (clipL << 12) | clipR;
+      vb[2] = (HC_SubA_HClipTB << 24) | (b->y1 << 12) | b->y2;
+      vb[3] = (HC_SubA_HClipLR << 24) | (b->x1 << 12) | b->x2;
    }
 	    
    vb[4] = ((HC_SubA_HDBBasL << 24) | (offset & 0xFFFFFF));
@@ -543,6 +523,11 @@ static int intersect_rect(drm_clip_rect_t *out,
                           drm_clip_rect_t *b)
 {
     *out = *a;
+    
+    if (0)
+       fprintf(stderr, "intersect %d,%d %d,%d and %d,%d %d,%d\n", 
+	       a->x1, a->y1, a->x2, a->y2,
+	       b->x1, b->y1, b->x2, b->y2);
 
     if (b->x1 > out->x1) out->x1 = b->x1;
     if (b->x2 < out->x2) out->x2 = b->x2;
@@ -575,6 +560,9 @@ void viaFlushDmaLocked(viaContextPtr vmesa, GLuint flags)
    int i;
    RING_VARS;
 
+   if (VIA_DEBUG)
+      fprintf(stderr, "%s\n", __FUNCTION__);
+
    if (*(GLuint *)vmesa->driHwLock != (DRM_LOCK_HELD|vmesa->hHWContext) &&
        *(GLuint *)vmesa->driHwLock != (DRM_LOCK_HELD|DRM_LOCK_CONT|vmesa->hHWContext)) {
       fprintf(stderr, "%s called without lock held\n", __FUNCTION__);
@@ -640,35 +628,42 @@ void viaFlushDmaLocked(viaContextPtr vmesa, GLuint flags)
       dump_dma( vmesa );
 
    if (flags & VIA_NO_CLIPRECTS) {
-      assert(vmesa->dmaCliprectAddr == 0);
+      if (0) fprintf(stderr, "%s VIA_NO_CLIPRECTS\n", __FUNCTION__);
+      assert(vmesa->dmaCliprectAddr == ~0);
       fire_buffer( vmesa );
    }
-   else if (!vmesa->dmaCliprectAddr) {
+   else if (vmesa->dmaCliprectAddr == ~0) {
       /* Contains only state.  Could just dump the packet?
        */
-      if (0) fprintf(stderr, "no dmaCliprectAddr\n");
+      if (0) fprintf(stderr, "%s: no dmaCliprectAddr\n", __FUNCTION__);
       if (0) fire_buffer( vmesa );
    }
    else if (vmesa->numClipRects) {
-      int ret;
       drm_clip_rect_t *pbox = vmesa->pClipRects;
+      if (0) fprintf(stderr, "%s: %d cliprects\n", __FUNCTION__, vmesa->numClipRects);
       
       for (i = 0; i < vmesa->numClipRects; i++) {
-	 if (vmesa->glCtx->Scissor.Enabled) {
-	    drm_clip_rect_t b;
-	    if (!intersect_rect(&b, &pbox[i], &vmesa->scissorRect)) 
-	       continue;
-	    via_emit_cliprect(vmesa, &b);	 
-	 }
-	 else {
-	    via_emit_cliprect(vmesa, &pbox[i]);
-	 }
-	 
-	 ret = fire_buffer(vmesa);
-	 if (ret)
+	 drm_clip_rect_t b;
+
+	 b.x1 = pbox[i].x1 - (vmesa->drawX + vmesa->drawXoff);
+	 b.x2 = pbox[i].x2 - (vmesa->drawX + vmesa->drawXoff);
+	 b.y1 = pbox[i].y1 - vmesa->drawY;
+	 b.y2 = pbox[i].y2 - vmesa->drawY;
+
+	 if (vmesa->glCtx->Scissor.Enabled &&
+	     !intersect_rect(&b, &b, &vmesa->scissorRect)) 
+	    continue;
+
+	 b.x1 += vmesa->drawXoff;
+	 b.x2 += vmesa->drawXoff;
+
+	 via_emit_cliprect(vmesa, &b);
+
+	 if (fire_buffer(vmesa) != 0)
 	    goto done;
       }
    } else {
+      if (0) fprintf(stderr, "%s: no cliprects\n", __FUNCTION__);
       UNLOCK_HARDWARE(vmesa);
       sched_yield();
       LOCK_HARDWARE(vmesa);
@@ -678,7 +673,7 @@ void viaFlushDmaLocked(viaContextPtr vmesa, GLuint flags)
    /* Reset vmesa vars:
     */
    vmesa->dmaLow = 0;
-   vmesa->dmaCliprectAddr = 0;
+   vmesa->dmaCliprectAddr = ~0;
    vmesa->newEmitState = ~0;
 }
 
@@ -752,6 +747,8 @@ GLuint *viaAllocDmaFunc(viaContextPtr vmesa, int bytes, const char *func, int li
 
    {
       GLuint *start = (GLuint *)(vmesa->dma + vmesa->dmaLow);
+      if (0)
+	 fprintf(stderr, "%s %04x 0x%x bytes\n", func, vmesa->dmaLow, bytes);
       vmesa->dmaLow += bytes;
       return start;
    }
@@ -760,6 +757,9 @@ GLuint *viaAllocDmaFunc(viaContextPtr vmesa, int bytes, const char *func, int li
 
 GLuint *viaExtendPrimitive(viaContextPtr vmesa, int bytes)
 {
+   if (0)
+      fprintf(stderr, "%s %d\n", __FUNCTION__, bytes);
+
    assert(vmesa->dmaLastPrim);
    if (vmesa->dmaLow + bytes > VIA_DMA_HIGHWATER) {
       viaWrapPrimitive(vmesa);
@@ -767,6 +767,8 @@ GLuint *viaExtendPrimitive(viaContextPtr vmesa, int bytes)
 
    {
       GLuint *start = (GLuint *)(vmesa->dma + vmesa->dmaLow);
+      if (0)
+	 fprintf(stderr, "%s %04x 0x%x bytes\n", __FUNCTION__, vmesa->dmaLow, bytes);
       vmesa->dmaLow += bytes;
       return start;
    }

src/mesa/drivers/dri/unichrome/via_render.c

@@ -38,7 +38,6 @@
 #include "via_context.h"
 #include "via_tris.h"
 #include "via_state.h"
-#include "via_vb.h"
 #include "via_ioctl.h"
 
 /*
@@ -52,7 +51,7 @@
 #define HAVE_LINE_LOOP   1
 #define HAVE_TRIANGLES   1
 #define HAVE_TRI_STRIPS  1
-#define HAVE_TRI_STRIP_1 0      /* has it, template can't use it yet */
+#define HAVE_TRI_STRIP_1 0  
 #define HAVE_TRI_FANS    1
 #define HAVE_POLYGONS    1
 #define HAVE_QUADS       0
@@ -73,7 +72,7 @@
     viaExtendPrimitive( vmesa, (nr) * vmesa->vertexSize * 4)
 
 #define EMIT_VERTS(ctx, j, nr, buf) \
-    via_emit_contiguous_verts(ctx, j, (j) + (nr), buf)
+    _tnl_emit_vertices_to_buffer(ctx, j, (j)+(nr), buf )  
     
 #define FLUSH() VIA_FINISH_PRIM( vmesa )
 
@@ -94,24 +93,18 @@ static GLboolean via_run_fastrender(GLcontext *ctx,
     struct vertex_buffer *VB = &tnl->vb;
     GLuint i;
     
-    /* Don't handle clipping or indexed vertices.
-     */
-    if (VB->ClipOrMask || 
-	vmesa->renderIndex != 0 || 
-	!via_fastvalidate_render( ctx, VB )) {
-	if (VIA_DEBUG) { 
-	    fprintf(stderr, "slow path\n");    
-	    fprintf(stderr, "ClipOrMask = %08x\n", VB->ClipOrMask);
-	    fprintf(stderr, "renderIndex = %08x\n", vmesa->renderIndex);	
-	    fprintf(stderr, "Elts = %08x\n", (GLuint)VB->Elts);	
-	}	    
-        return GL_TRUE;
-    }
-    if (VIA_DEBUG) fprintf(stderr, "%s - in\n", __FUNCTION__);    
-    vmesa->setupNewInputs = VERT_BIT_CLIP;
 
     tnl->Driver.Render.Start(ctx);
     
+    if (VB->ClipOrMask || 
+	vmesa->renderIndex != 0 || 
+	!via_fastvalidate_render( ctx, VB )) {
+	tnl->Driver.Render.Finish(ctx);
+        return GL_TRUE;
+    }
+
+    tnl->clipspace.new_inputs |= VERT_BIT_POS;
+
     for (i = 0; i < VB->PrimitiveCount; ++i) {
         GLuint mode = VB->Primitive[i].mode;
         GLuint start = VB->Primitive[i].start;
@@ -121,31 +114,14 @@ static GLboolean via_run_fastrender(GLcontext *ctx,
     }
 
     tnl->Driver.Render.Finish(ctx);
-    
-    if (VIA_DEBUG) fprintf(stderr, "%s - out\n", __FUNCTION__);    
+
     return GL_FALSE;            /* finished the pipe */
 }
 
 
 static void via_check_fastrender(GLcontext *ctx, struct tnl_pipeline_stage *stage)
 {
-    GLuint inputs = 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;
+   stage->inputs = TNL_CONTEXT(ctx)->render_inputs;
 }
 
 

src/mesa/drivers/dri/unichrome/via_state.c

@@ -36,7 +36,6 @@
 #include "via_context.h"
 #include "via_state.h"
 #include "via_tex.h"
-#include "via_vb.h"
 #include "via_tris.h"
 #include "via_ioctl.h"
 
@@ -615,7 +614,8 @@ static void viaScissor(GLcontext *ctx, GLint x, GLint y,
     if (!vmesa->driDrawable)
        return;
 
-    if (VIA_DEBUG) fprintf(stderr, "%s in\n", __FUNCTION__);    
+    if (VIA_DEBUG)
+       fprintf(stderr, "%s %d,%d %dx%d, drawH %d\n", __FUNCTION__, x,y,w,h, vmesa->driDrawable->h);
 
     if (ctx->Scissor.Enabled) {
         VIA_FLUSH_DMA(vmesa); /* don't pipeline cliprect changes */
@@ -628,6 +628,19 @@ static void viaScissor(GLcontext *ctx, GLint x, GLint y,
     if (VIA_DEBUG) fprintf(stderr, "%s out\n", __FUNCTION__);    
 }
 
+static void viaEnable(GLcontext *ctx, GLenum cap, GLboolean state)
+{
+   viaContextPtr vmesa = VIA_CONTEXT(ctx);
+
+   switch (cap) {
+   case GL_SCISSOR_TEST:
+      VIA_FLUSH_DMA(vmesa);
+      break;
+   default:
+      break;
+   }
+}
+
 
 
 /* Fallback to swrast for select and feedback.
@@ -684,6 +697,26 @@ static void viaClearColor(GLcontext *ctx, const GLfloat color[4])
 	
 }
 
+#define WRITEMASK_ALPHA_SHIFT 31
+#define WRITEMASK_RED_SHIFT   30
+#define WRITEMASK_GREEN_SHIFT 29
+#define WRITEMASK_BLUE_SHIFT  28
+
+static void viaColorMask(GLcontext *ctx,
+			 GLboolean r, GLboolean g,
+			 GLboolean b, GLboolean a)
+{
+   viaContextPtr vmesa = VIA_CONTEXT( ctx );
+
+   if (VIA_DEBUG)
+      fprintf(stderr, "%s r(%d) g(%d) b(%d) a(%d)\n", __FUNCTION__, r, g, b, a);
+
+   vmesa->ClearMask = (((!r) << WRITEMASK_RED_SHIFT) |
+		       ((!g) << WRITEMASK_GREEN_SHIFT) |
+		       ((!b) << WRITEMASK_BLUE_SHIFT) |
+		       ((!a) << WRITEMASK_ALPHA_SHIFT));
+}
+
 
 /* =============================================================
  */
@@ -730,7 +763,7 @@ void viaInitState(GLcontext *ctx)
     viaContextPtr vmesa = VIA_CONTEXT(ctx);
 
     vmesa->regCmdA = HC_ACMD_HCmdA;
-    vmesa->regCmdB = HC_ACMD_HCmdB | HC_HVPMSK_X | HC_HVPMSK_Y | HC_HVPMSK_Z;
+    vmesa->regCmdB = HC_ACMD_HCmdB;
     vmesa->regEnable = HC_HenCW_MASK;
 
    /* Mesa should do this for us:
@@ -828,7 +861,7 @@ static void viaChooseTextureState(GLcontext *ctx)
                 FALLBACK(vmesa, VIA_FALLBACK_TEXTURE, GL_TRUE);
                 return;
             }
-            vmesa->regCmdB |= HC_HVPMSK_S | HC_HVPMSK_T | HC_HVPMSK_W | HC_HVPMSK_Cs;
+
             vmesa->regEnable |= HC_HenTXMP_MASK | HC_HenTXCH_MASK | HC_HenTXPP_MASK;
    
             switch (texObj->MinFilter) {
@@ -885,9 +918,6 @@ static void viaChooseTextureState(GLcontext *ctx)
 
 	    viaTexCombineState( vmesa, texUnit0->_CurrentCombine, 0 );
         }
-	else {
-	/* Should turn Cs off if actually no Cs */
-	}
 
         if (texUnit1->_ReallyEnabled) {
             struct gl_texture_object *texObj = texUnit1->_Current;
@@ -965,10 +995,6 @@ static void viaChooseTextureState(GLcontext *ctx)
 	}
     }
     else {
-	if (ctx->Fog.Enabled)
-    	    vmesa->regCmdB &= (~(HC_HVPMSK_S | HC_HVPMSK_T));	
-	else	    
-    	    vmesa->regCmdB &= (~(HC_HVPMSK_S | HC_HVPMSK_T | HC_HVPMSK_W));
         vmesa->regEnable &= (~(HC_HenTXMP_MASK | HC_HenTXCH_MASK | HC_HenTXPP_MASK));
     }
     if (VIA_DEBUG) fprintf(stderr, "%s - out\n", __FUNCTION__);    
@@ -1256,12 +1282,13 @@ static void viaChooseColorState(GLcontext *ctx)
         vmesa->regHROP = HC_HROP_P;
 
     vmesa->regHFBBMSKL = (*(GLuint *)&ctx->Color.ColorMask[0]) & 0xFFFFFF;
-    vmesa->regHROP |= ((*(GLuint *)&ctx->Color.ColorMask[0]) >> 24) & 0xFF;
+    vmesa->regHROP |= ctx->Color.ColorMask[3];
 
-    if ((GLuint)((GLuint *)&ctx->Color.ColorMask[0]) & 0xFF000000)
+    if (ctx->Color.ColorMask[3])
         vmesa->regEnable |= HC_HenAW_MASK;
     else
-        vmesa->regEnable &= (~HC_HenAW_MASK);
+        vmesa->regEnable &= ~HC_HenAW_MASK;
+
     if (VIA_DEBUG) fprintf(stderr, "%s - out\n", __FUNCTION__);    
 }
 
@@ -1272,7 +1299,6 @@ static void viaChooseFogState(GLcontext *ctx)
     if (ctx->Fog.Enabled) {
         GLubyte r, g, b, a;
 
-        vmesa->regCmdB |= (HC_HVPMSK_Cd | HC_HVPMSK_Cs | HC_HVPMSK_W);
         vmesa->regEnable |= HC_HenFOG_MASK;
 
         /* Use fog equation 0 (OpenGL's default) & local fog.
@@ -1287,10 +1313,6 @@ static void viaChooseFogState(GLcontext *ctx)
         vmesa->regHFogCH = a;
     }
     else {
-        if (!ctx->Texture._EnabledUnits) {
-            vmesa->regCmdB &= ~ HC_HVPMSK_W;
-    	    vmesa->regCmdB &= ~ HC_HVPMSK_Cs;
-	}	    
         vmesa->regEnable &= ~HC_HenFOG_MASK;
     }
 }
@@ -1299,7 +1321,6 @@ static void viaChooseDepthState(GLcontext *ctx)
 {
     viaContextPtr vmesa = VIA_CONTEXT(ctx);
     if (ctx->Depth.Test) {
-        vmesa->regCmdB |= HC_HVPMSK_Z;
         vmesa->regEnable |= HC_HenZT_MASK;
         if (ctx->Depth.Mask)
             vmesa->regEnable |= HC_HenZW_MASK;
@@ -1309,10 +1330,6 @@ static void viaChooseDepthState(GLcontext *ctx)
 	
     }
     else {
-        /* Still need to send parameter Z.
-         */
-        
-	vmesa->regCmdB |= HC_HVPMSK_Z;
         vmesa->regEnable &= ~HC_HenZT_MASK;
         
         /*=* [DBG] racer : can't display cars in car selection menu *=*/
@@ -1330,11 +1347,9 @@ static void viaChooseLightState(GLcontext *ctx)
 
     if (ctx->Light.ShadeModel == GL_SMOOTH) {
         vmesa->regCmdA |= HC_HShading_Gouraud;
-        vmesa->regCmdB |= HC_HVPMSK_Cd;
     }
     else {
         vmesa->regCmdA &= ~HC_HShading_Gouraud;
-        vmesa->regCmdB |= HC_HVPMSK_Cd;
     }
 }
 
@@ -1515,17 +1530,6 @@ static void viaChooseTriangle(GLcontext *ctx)
 void viaValidateState( GLcontext *ctx )
 {
     viaContextPtr vmesa = VIA_CONTEXT(ctx);
-    struct gl_texture_unit *texUnit0 = &ctx->Texture.Unit[0];
-    struct gl_texture_unit *texUnit1 = &ctx->Texture.Unit[1];
-    if (VIA_DEBUG) fprintf(stderr, "%s - in\n", __FUNCTION__);    
-
-
-    if (texUnit0->_ReallyEnabled || texUnit1->_ReallyEnabled || ctx->Fog.Enabled) {
-	vmesa->regCmdB |= HC_HVPMSK_Cs;
-    }
-    else {
-    	vmesa->regCmdB &= ~ HC_HVPMSK_Cs;
-    }
     
     if (vmesa->newState & _NEW_TEXTURE) {
         viaChooseTextureState(ctx);
@@ -1555,10 +1559,13 @@ void viaValidateState( GLcontext *ctx )
     if (vmesa->newState & _NEW_STENCIL)
         viaChooseStencilState(ctx);
     
+    if (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR)
+        vmesa->regEnable |= HC_HenCS_MASK;
+    else
+        vmesa->regEnable &= ~HC_HenCS_MASK;
+
     vmesa->newEmitState |= vmesa->newState;
     vmesa->newState = 0;
-            
-    if (VIA_DEBUG) fprintf(stderr, "%s - out\n", __FUNCTION__);        
 }
 
 static void viaInvalidateState(GLcontext *ctx, GLuint newState)
@@ -1585,11 +1592,13 @@ void viaInitStateFuncs(GLcontext *ctx)
     ctx->Driver.BlendEquationSeparate = viaBlendEquationSeparate;
     ctx->Driver.BlendFuncSeparate = viaBlendFuncSeparate;
     ctx->Driver.ClearColor = viaClearColor;
+    ctx->Driver.ColorMask = viaColorMask;
     ctx->Driver.DrawBuffer = viaDrawBuffer;
     ctx->Driver.RenderMode = viaRenderMode;
     ctx->Driver.Scissor = viaScissor;
     ctx->Driver.DepthRange = viaDepthRange;
     ctx->Driver.Viewport = viaViewport;
+    ctx->Driver.Enable = viaEnable;
 
     /* Pixel path fallbacks.
      */

src/mesa/drivers/dri/unichrome/via_texcombine.c

@@ -42,7 +42,6 @@
 #include "via_context.h"
 #include "via_state.h"
 #include "via_tex.h"
-#include "via_vb.h"
 #include "via_tris.h"
 #include "via_ioctl.h"
 

src/mesa/drivers/dri/unichrome/via_tris.c

@@ -41,14 +41,13 @@
 #include "via_tris.h"
 #include "via_state.h"
 #include "via_span.h"
-#include "via_vb.h"
 #include "via_ioctl.h"
 
 /***********************************************************************
  *                    Emit primitives as inline vertices               *
  ***********************************************************************/
 
-#if 0
+#if 1
 #define COPY_DWORDS(vb, vertsize, v)					\
     do {								\
         int j;								\
@@ -69,58 +68,108 @@
     } while (0)
 #endif
 
-static void __inline__ via_draw_triangle(viaContextPtr vmesa,
-                                         viaVertexPtr v0,
-                                         viaVertexPtr v1,
-                                         viaVertexPtr v2)
+static void via_draw_triangle(viaContextPtr vmesa,
+			      viaVertexPtr v0,
+			      viaVertexPtr v1,
+			      viaVertexPtr v2)
 {
-    GLuint vertsize = vmesa->vertexSize;
-    GLuint *vb = viaExtendPrimitive(vmesa, 3 * 4 * vertsize);
-/*     fprintf(stderr, "%s: %p %p %p\n", __FUNCTION__, v0, v1, v2); */
-    COPY_DWORDS(vb, vertsize, v0);
-    COPY_DWORDS(vb, vertsize, v1);
-    COPY_DWORDS(vb, vertsize, v2);
+   GLuint vertsize = vmesa->vertexSize;
+   GLuint *vb = viaExtendPrimitive(vmesa, 3 * 4 * vertsize);
+
+   COPY_DWORDS(vb, vertsize, v0);
+   COPY_DWORDS(vb, vertsize, v1);
+   COPY_DWORDS(vb, vertsize, v2);
 }
 
 
-static void __inline__ via_draw_quad(viaContextPtr vmesa,
-                                     viaVertexPtr v0,
-                                     viaVertexPtr v1,
-                                     viaVertexPtr v2,
-                                     viaVertexPtr v3)
+static void via_draw_quad(viaContextPtr vmesa,
+			  viaVertexPtr v0,
+			  viaVertexPtr v1,
+			  viaVertexPtr v2,
+			  viaVertexPtr v3)
 {
-    GLuint vertsize = vmesa->vertexSize;
-    GLuint *vb = viaExtendPrimitive(vmesa, 6 * 4 * vertsize);
+   GLuint vertsize = vmesa->vertexSize;
+   GLuint *vb = viaExtendPrimitive(vmesa, 6 * 4 * vertsize);
 
-/*     fprintf(stderr, "%s: %p %p %p %p\n", __FUNCTION__, v0, v1, v2, v3); */
-    COPY_DWORDS(vb, vertsize, v0);
-    COPY_DWORDS(vb, vertsize, v1);
-    COPY_DWORDS(vb, vertsize, v3);
-    COPY_DWORDS(vb, vertsize, v1);
-    COPY_DWORDS(vb, vertsize, v2);
-    COPY_DWORDS(vb, vertsize, v3);
+   COPY_DWORDS(vb, vertsize, v0);
+   COPY_DWORDS(vb, vertsize, v1);
+   COPY_DWORDS(vb, vertsize, v3);
+   COPY_DWORDS(vb, vertsize, v1);
+   COPY_DWORDS(vb, vertsize, v2);
+   COPY_DWORDS(vb, vertsize, v3);
 }
 
-static __inline__ void via_draw_line(viaContextPtr vmesa,
-                                     viaVertexPtr v0,
-                                     viaVertexPtr v1)
+static void via_draw_line(viaContextPtr vmesa,
+			  viaVertexPtr v0,
+			  viaVertexPtr v1)
 {
-    GLuint vertsize = vmesa->vertexSize;
-    GLuint *vb = viaExtendPrimitive(vmesa, 2 * 4 * vertsize);
-    COPY_DWORDS(vb, vertsize, v0);
-    COPY_DWORDS(vb, vertsize, v1);
+   GLuint vertsize = vmesa->vertexSize;
+   GLuint *vb = viaExtendPrimitive(vmesa, 2 * 4 * vertsize);
+   COPY_DWORDS(vb, vertsize, v0);
+   COPY_DWORDS(vb, vertsize, v1);
 }
 
 
-static __inline__ void via_draw_point(viaContextPtr vmesa,
-                                      viaVertexPtr v0)
+static void via_draw_point(viaContextPtr vmesa,
+			   viaVertexPtr v0)
 {
-    GLuint vertsize = vmesa->vertexSize;
-    GLuint *vb = viaExtendPrimitive(vmesa, 4 * vertsize);
-    COPY_DWORDS(vb, vertsize, v0);
+   GLuint vertsize = vmesa->vertexSize;
+   GLuint *vb = viaExtendPrimitive(vmesa, 4 * vertsize);
+   COPY_DWORDS(vb, vertsize, v0);
 }
 
 
+/* Fallback drawing functions for the ptex hack.
+ */
+#define PTEX_VERTEX( tmp, vertex_size, v)	\
+do {							\
+   GLuint j;						\
+   GLfloat rhw = 1.0 / v->f[vertex_size];		\
+   for ( j = 0 ; j < vertex_size ; j++ )		\
+      tmp.f[j] = v->f[j];				\
+   tmp.f[3] *= v->f[vertex_size];			\
+   tmp.f[vertex_size-2] *= rhw;				\
+   tmp.f[vertex_size-1] *= rhw;				\
+} while (0)
+
+static void via_ptex_tri (viaContextPtr vmesa,
+			  viaVertexPtr v0,
+			  viaVertexPtr v1,
+			  viaVertexPtr v2)
+{
+   GLuint vertsize = vmesa->hwVertexSize;
+   GLuint *vb = viaExtendPrimitive(vmesa, 3*4*vertsize);
+   viaVertex tmp;
+
+   PTEX_VERTEX(tmp, vertsize, v0); COPY_DWORDS(vb, vertsize, &tmp);
+   PTEX_VERTEX(tmp, vertsize, v1); COPY_DWORDS(vb, vertsize, &tmp);
+   PTEX_VERTEX(tmp, vertsize, v2); COPY_DWORDS(vb, vertsize, &tmp);
+}
+
+static void via_ptex_line (viaContextPtr vmesa,
+			   viaVertexPtr v0,
+			   viaVertexPtr v1)
+{
+   GLuint vertsize = vmesa->hwVertexSize;
+   GLuint *vb = viaExtendPrimitive(vmesa, 2*4*vertsize);
+   viaVertex tmp;
+
+   PTEX_VERTEX(tmp, vertsize, v0); COPY_DWORDS(vb, vertsize, &tmp);
+   PTEX_VERTEX(tmp, vertsize, v1); COPY_DWORDS(vb, vertsize, &tmp);
+}
+
+static void via_ptex_point (viaContextPtr vmesa,
+			    viaVertexPtr v0)
+{
+   GLuint vertsize = vmesa->hwVertexSize;
+   GLuint *vb = viaExtendPrimitive(vmesa, 1*4*vertsize);
+   viaVertex tmp;
+
+   PTEX_VERTEX(tmp, vertsize, v0); COPY_DWORDS(vb, vertsize, &tmp);
+}
+
+
+
 
 
 /***********************************************************************
@@ -129,7 +178,6 @@ static __inline__ void via_draw_point(viaContextPtr vmesa,
 
 #define TRI(a, b, c)                                \
     do {                                            \
-        if (VIA_DEBUG) fprintf(stderr, "hw TRI\n"); \
         if (DO_FALLBACK)                            \
             vmesa->drawTri(vmesa, a, b, c);         \
         else                                        \
@@ -138,7 +186,6 @@ static __inline__ void via_draw_point(viaContextPtr vmesa,
 
 #define QUAD(a, b, c, d)                            \
     do {                                            \
-        if (VIA_DEBUG) fprintf(stderr, "hw QUAD\n");\
         if (DO_FALLBACK) {                          \
             vmesa->drawTri(vmesa, a, b, d);         \
             vmesa->drawTri(vmesa, b, c, d);         \
@@ -149,7 +196,6 @@ static __inline__ void via_draw_point(viaContextPtr vmesa,
 
 #define LINE(v0, v1)                                \
     do {                                            \
-        if(VIA_DEBUG) fprintf(stderr, "hw LINE\n");\
         if (DO_FALLBACK)                            \
             vmesa->drawLine(vmesa, v0, v1);         \
         else                                        \
@@ -158,7 +204,6 @@ static __inline__ void via_draw_point(viaContextPtr vmesa,
 
 #define POINT(v0)                                    \
     do {                                             \
-        if (VIA_DEBUG) fprintf(stderr, "hw POINT\n");\
         if (DO_FALLBACK)                             \
             vmesa->drawPoint(vmesa, v0);             \
         else                                         \
@@ -222,7 +267,7 @@ static struct {
     } while (0)
 
 
-#define DEPTH_SCALE (1.0 / 0xffff)
+#define DEPTH_SCALE vmesa->polygon_offset_scale
 #define UNFILLED_TRI unfilled_tri
 #define UNFILLED_QUAD unfilled_quad
 #define VERT_X(_v) _v->v.x
@@ -244,7 +289,7 @@ do {								\
 
 #define VERT_SET_SPEC( v0, c )					\
 do {								\
-   if (havespec) {						\
+   if (specoffset) {						\
       UNCLAMPED_FLOAT_TO_UBYTE(v0->v.specular.red, (c)[0]);	\
       UNCLAMPED_FLOAT_TO_UBYTE(v0->v.specular.green, (c)[1]);	\
       UNCLAMPED_FLOAT_TO_UBYTE(v0->v.specular.blue, (c)[2]);	\
@@ -252,7 +297,7 @@ do {								\
 } while (0)
 #define VERT_COPY_SPEC( v0, v1 )			\
 do {							\
-   if (havespec) {					\
+   if (specoffset) {					\
       v0->v.specular.red   = v1->v.specular.red;	\
       v0->v.specular.green = v1->v.specular.green;	\
       v0->v.specular.blue  = v1->v.specular.blue; 	\
@@ -262,16 +307,16 @@ do {							\
 
 #define VERT_SAVE_RGBA( idx )    color[idx] = v[idx]->ui[coloroffset]
 #define VERT_RESTORE_RGBA( idx ) v[idx]->ui[coloroffset] = color[idx]
-#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 VERT_SAVE_SPEC( idx )    if (specoffset) spec[idx] = v[idx]->ui[specoffset]
+#define VERT_RESTORE_SPEC( idx ) if (specoffset) v[idx]->ui[specoffset] = spec[idx]
 
 
 #define LOCAL_VARS(n)                                                   \
     viaContextPtr vmesa = VIA_CONTEXT(ctx);                             \
     GLuint color[n], spec[n];                                           \
-    GLuint coloroffset = (vmesa->vertexSize == 4 ? 3 : 4);              \
-    GLboolean havespec = (vmesa->vertexSize > 4);                       \
-    (void)color; (void)spec; (void)coloroffset; (void)havespec;
+    GLuint coloroffset = vmesa->coloroffset;              \
+    GLuint specoffset = vmesa->specoffset;                       \
+    (void)color; (void)spec; (void)coloroffset; (void)specoffset;
 
 
 /***********************************************************************
@@ -409,9 +454,9 @@ via_fallback_tri(viaContextPtr vmesa,
 {    
     GLcontext *ctx = vmesa->glCtx;
     SWvertex v[3];
-    via_translate_vertex(ctx, v0, &v[0]);
-    via_translate_vertex(ctx, v1, &v[1]);
-    via_translate_vertex(ctx, v2, &v[2]);
+    _swsetup_Translate(ctx, v0, &v[0]);
+    _swsetup_Translate(ctx, v1, &v[1]);
+    _swsetup_Translate(ctx, v2, &v[2]);
     viaSpanRenderStart( ctx );
     _swrast_Triangle(ctx, &v[0], &v[1], &v[2]);
     viaSpanRenderFinish( ctx );
@@ -425,8 +470,8 @@ via_fallback_line(viaContextPtr vmesa,
 {
     GLcontext *ctx = vmesa->glCtx;
     SWvertex v[2];
-    via_translate_vertex(ctx, v0, &v[0]);
-    via_translate_vertex(ctx, v1, &v[1]);
+    _swsetup_Translate(ctx, v0, &v[0]);
+    _swsetup_Translate(ctx, v1, &v[1]);
     viaSpanRenderStart( ctx );
     _swrast_Line(ctx, &v[0], &v[1]);
     viaSpanRenderFinish( ctx );
@@ -439,7 +484,7 @@ via_fallback_point(viaContextPtr vmesa,
 {
     GLcontext *ctx = vmesa->glCtx;
     SWvertex v[1];
-    via_translate_vertex(ctx, v0, &v[0]);
+    _swsetup_Translate(ctx, v0, &v[0]);
     viaSpanRenderStart( ctx );
     _swrast_Point(ctx, &v[0]);
     viaSpanRenderFinish( ctx );
@@ -540,6 +585,13 @@ static void viaFastRenderClippedPoly(GLcontext *ctx, const GLuint *elts,
 
 
 
+
+#define _VIA_NEW_VERTEX (_NEW_TEXTURE |                         \
+                         _DD_NEW_SEPARATE_SPECULAR |            \
+                         _DD_NEW_TRI_UNFILLED |                 \
+                         _DD_NEW_TRI_LIGHT_TWOSIDE |            \
+                         _NEW_FOG)
+
 #define _VIA_NEW_RENDERSTATE (_DD_NEW_LINE_STIPPLE |            \
                               _DD_NEW_TRI_UNFILLED |            \
                               _DD_NEW_TRI_LIGHT_TWOSIDE |       \
@@ -556,71 +608,190 @@ static void viaFastRenderClippedPoly(GLcontext *ctx, const GLuint *elts,
 #define ANY_FALLBACK_FLAGS (POINT_FALLBACK|LINE_FALLBACK|TRI_FALLBACK)
 #define ANY_RASTER_FLAGS (DD_TRI_LIGHT_TWOSIDE|DD_TRI_OFFSET|DD_TRI_UNFILLED)
 
-void viaChooseRenderState(GLcontext *ctx)
+static void viaChooseRenderState(GLcontext *ctx)
 {
-    TNLcontext *tnl = TNL_CONTEXT(ctx);
-    viaContextPtr vmesa = VIA_CONTEXT(ctx);
-    GLuint flags = ctx->_TriangleCaps;
-    GLuint index = 0;
-    if (VIA_DEBUG) fprintf(stderr, "%s - in\n", __FUNCTION__);
-    
-    if (VIA_DEBUG) fprintf(stderr, "_TriangleCaps = %x\n", flags);    
-    if (flags & (ANY_FALLBACK_FLAGS|ANY_RASTER_FLAGS)) {
-        if (flags & ANY_RASTER_FLAGS) {
-            if (flags & DD_TRI_LIGHT_TWOSIDE)    index |= VIA_TWOSIDE_BIT;
-            if (flags & DD_TRI_OFFSET)           index |= VIA_OFFSET_BIT;
-            if (flags & DD_TRI_UNFILLED)         index |= VIA_UNFILLED_BIT;
-        }
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   viaContextPtr vmesa = VIA_CONTEXT(ctx);
+   GLuint flags = ctx->_TriangleCaps;
+   GLuint index = 0;
 
-        vmesa->drawPoint = via_draw_point;
-        vmesa->drawLine = via_draw_line;
-        vmesa->drawTri = via_draw_triangle;
+   if (vmesa->ptexHack) {
+      vmesa->drawPoint = via_ptex_point;
+      vmesa->drawLine = via_ptex_line;
+      vmesa->drawTri = via_ptex_tri;
+      index |= VIA_FALLBACK_BIT;
+   }
+   else {
+      vmesa->drawPoint = via_draw_point;
+      vmesa->drawLine = via_draw_line;
+      vmesa->drawTri = via_draw_triangle;
+   }
 
-        /* Hook in fallbacks for specific primitives.
-         */
-        if (flags & ANY_FALLBACK_FLAGS) {
-            if (flags & POINT_FALLBACK)
-                vmesa->drawPoint = via_fallback_point;
+   if (flags & (ANY_FALLBACK_FLAGS|ANY_RASTER_FLAGS)) {
+      if (flags & DD_TRI_LIGHT_TWOSIDE)    index |= VIA_TWOSIDE_BIT;
+      if (flags & DD_TRI_OFFSET)           index |= VIA_OFFSET_BIT;
+      if (flags & DD_TRI_UNFILLED)         index |= VIA_UNFILLED_BIT;
+      if (flags & ANY_FALLBACK_FLAGS)      index |= VIA_FALLBACK_BIT;
 
-            if (flags & LINE_FALLBACK)
-                vmesa->drawLine = via_fallback_line;
+      /* Hook in fallbacks for specific primitives.
+       */
+      if (flags & POINT_FALLBACK)
+	 vmesa->drawPoint = via_fallback_point;
+      
+      if (flags & LINE_FALLBACK)
+	 vmesa->drawLine = via_fallback_line;
 
-            if (flags & TRI_FALLBACK)
-                vmesa->drawTri = via_fallback_tri;
+      if (flags & TRI_FALLBACK)
+	 vmesa->drawTri = via_fallback_tri;
+   }
 
-            index |= VIA_FALLBACK_BIT;
-        }
-    }
-    if (VIA_DEBUG) {
-	fprintf(stderr, "index = %x\n", index);    
-	fprintf(stderr, "renderIndex = %x\n", vmesa->renderIndex);
-    }	
-    if (vmesa->renderIndex != index) {
-        vmesa->renderIndex = index;
+   if (vmesa->renderIndex != index) {
+      vmesa->renderIndex = index;
 
-        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.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;
 
-        tnl->Driver.Render.Quad = rast_tab[index].quad;
-
-        if (index == 0) {
-            tnl->Driver.Render.PrimTabVerts = via_fastrender_tab_verts;
-            tnl->Driver.Render.PrimTabElts = via_fastrender_tab_elts;
-            tnl->Driver.Render.ClippedLine = line; /* from tritmp.h */
-            tnl->Driver.Render.ClippedPolygon = viaFastRenderClippedPoly;
-        }
-        else {
-            tnl->Driver.Render.PrimTabVerts = _tnl_render_tab_verts;
-            tnl->Driver.Render.PrimTabElts = _tnl_render_tab_elts;
-            tnl->Driver.Render.ClippedLine = viaRenderClippedLine;
-            tnl->Driver.Render.ClippedPolygon = viaRenderClippedPoly;
-        }
-    }
-    if (VIA_DEBUG) fprintf(stderr, "%s - out\n", __FUNCTION__);    
+      if (index == 0) {
+	 tnl->Driver.Render.PrimTabVerts = via_fastrender_tab_verts;
+	 tnl->Driver.Render.PrimTabElts = via_fastrender_tab_elts;
+	 tnl->Driver.Render.ClippedLine = line; /* from tritmp.h */
+	 tnl->Driver.Render.ClippedPolygon = viaFastRenderClippedPoly;
+      }
+      else {
+	 tnl->Driver.Render.PrimTabVerts = _tnl_render_tab_verts;
+	 tnl->Driver.Render.PrimTabElts = _tnl_render_tab_elts;
+	 tnl->Driver.Render.ClippedLine = viaRenderClippedLine;
+	 tnl->Driver.Render.ClippedPolygon = viaRenderClippedPoly;
+      }
+   }
 }
 
 
+#define VIA_EMIT_TEX1	0x01
+#define VIA_EMIT_TEX0	0x02
+#define VIA_EMIT_PTEX0	0x04
+#define VIA_EMIT_RGBA	0x08
+#define VIA_EMIT_SPEC	0x10
+#define VIA_EMIT_FOG	0x20
+#define VIA_EMIT_W	0x40
+
+#define EMIT_ATTR( ATTR, STYLE, INDEX, REGB )				\
+do {									\
+   vmesa->vertex_attrs[vmesa->vertex_attr_count].attrib = (ATTR);	\
+   vmesa->vertex_attrs[vmesa->vertex_attr_count].format = (STYLE);	\
+   vmesa->vertex_attr_count++;						\
+   setupIndex |= (INDEX);						\
+   regCmdB |= (REGB);							\
+} while (0)
+
+#define EMIT_PAD( N )							\
+do {									\
+   vmesa->vertex_attrs[vmesa->vertex_attr_count].attrib = 0;		\
+   vmesa->vertex_attrs[vmesa->vertex_attr_count].format = EMIT_PAD;	\
+   vmesa->vertex_attrs[vmesa->vertex_attr_count].offset = (N);		\
+   vmesa->vertex_attr_count++;						\
+} while (0)
+
+
+
+static void viaChooseVertexState( GLcontext *ctx )
+{
+   viaContextPtr vmesa = VIA_CONTEXT(ctx);
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   GLuint index = tnl->render_inputs;
+   GLuint regCmdB = HC_HVPMSK_X | HC_HVPMSK_Y | HC_HVPMSK_Z;
+   GLuint setupIndex = 0;
+
+   vmesa->vertex_attr_count = 0;
+ 
+   /* EMIT_ATTR's must be in order as they tell t_vertex.c how to
+    * build up a hardware vertex.
+    */
+   if (index & (_TNL_BITS_TEX_ANY|_TNL_BIT_FOG)) {
+      EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_4F_VIEWPORT, VIA_EMIT_W, HC_HVPMSK_W );
+      vmesa->coloroffset = 4;
+   }
+   else {
+      EMIT_ATTR( _TNL_ATTRIB_POS, EMIT_3F_VIEWPORT, 0, 0 );
+      vmesa->coloroffset = 3;
+   }
+
+   /* t_context.c always includes a diffuse color */
+   EMIT_ATTR( _TNL_ATTRIB_COLOR0, EMIT_4UB_4F_BGRA, VIA_EMIT_RGBA, HC_HVPMSK_Cd );
+      
+   vmesa->specoffset = 0;
+   if (index & (_TNL_BIT_COLOR1|_TNL_BIT_FOG)) {
+      if ((index & _TNL_BIT_COLOR1)) {
+	 vmesa->specoffset = vmesa->coloroffset + 1;
+	 EMIT_ATTR( _TNL_ATTRIB_COLOR1, EMIT_3UB_3F_BGR, VIA_EMIT_SPEC, HC_HVPMSK_Cs );
+      }
+      else
+	 EMIT_PAD( 3 );
+
+      if ((index & _TNL_BIT_FOG))
+	 EMIT_ATTR( _TNL_ATTRIB_FOG, EMIT_1UB_1F, VIA_EMIT_FOG, HC_HVPMSK_Cs );
+      else
+	 EMIT_PAD( 1 );
+   }
+
+   if (index & _TNL_BIT_TEX(0)) {
+      if (vmesa->ptexHack)
+	 EMIT_ATTR( _TNL_ATTRIB_TEX0, EMIT_3F_XYW, VIA_EMIT_PTEX0, (HC_HVPMSK_S | HC_HVPMSK_T) );
+      else 
+	 EMIT_ATTR( _TNL_ATTRIB_TEX0, EMIT_2F, VIA_EMIT_TEX0, (HC_HVPMSK_S | HC_HVPMSK_T) );
+   }
+
+   if (index & _TNL_BIT_TEX(1)) {
+      EMIT_ATTR( _TNL_ATTRIB_TEX1, EMIT_2F, VIA_EMIT_TEX1, 0 );	/* how does the hardware find out about this? */
+   }
+
+   if (setupIndex != vmesa->setupIndex) {
+      vmesa->vertexSize = _tnl_install_attrs( ctx, 
+					       vmesa->vertex_attrs, 
+					       vmesa->vertex_attr_count,
+					       vmesa->ViewportMatrix.m, 0 );
+      vmesa->vertexSize >>= 2;
+      vmesa->setupIndex = setupIndex;
+      vmesa->regCmdB &= ~HC_HVPMSK_MASK;
+      vmesa->regCmdB |= regCmdB;
+
+      if (vmesa->ptexHack) 
+	 vmesa->hwVertexSize = vmesa->vertexSize - 1;
+      else
+	 vmesa->hwVertexSize = vmesa->vertexSize;
+   }
+}
+
+
+
+
+/* Check if projective texture coordinates are used and if we can fake
+ * them. Fallback to swrast we can't. Returns GL_TRUE if projective
+ * texture coordinates must be faked, GL_FALSE otherwise.
+ */
+static GLboolean viaCheckPTexHack( GLcontext *ctx )
+{
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   struct vertex_buffer *VB = &tnl->vb;
+   GLuint index = tnl->render_inputs;
+   GLboolean fallback = GL_FALSE;
+   GLboolean ptexHack = GL_FALSE;
+
+   if (index & _TNL_BIT_TEX(0) && VB->TexCoordPtr[0]->size == 4) {
+      if ((index & _TNL_BITS_TEX_ANY) == _TNL_BIT_TEX(0))
+	 ptexHack = GL_TRUE; 
+      else
+	 fallback = GL_TRUE;
+   }
+   if ((index & _TNL_BIT_TEX(1)) && VB->TexCoordPtr[1]->size == 4)
+      fallback = GL_TRUE;
+
+   FALLBACK(VIA_CONTEXT(ctx), VIA_FALLBACK_PROJ_TEXTURE, fallback);
+   return ptexHack;
+}
+
 
 
 
@@ -628,27 +799,40 @@ void viaChooseRenderState(GLcontext *ctx)
 /*                 High level hooks for t_vb_render.c                 */
 /**********************************************************************/
 
-static void viaRunPipeline(GLcontext *ctx)
-{
-    viaContextPtr vmesa = VIA_CONTEXT(ctx);
-    
-    if (vmesa->newState) {
-       viaValidateState( ctx );
-       if (!vmesa->Fallback) {
-	  viaChooseVertexState(ctx);
-	  viaChooseRenderState(ctx);
-       }
-    }
-
-    _tnl_run_pipeline(ctx);
-}
 
 static void viaRenderStart(GLcontext *ctx)
 {
-    /* Check for projective texturing.  Make sure all texcoord
-     * pointers point to something.  (fix in mesa?)
-     */
-    viaCheckTexSizes(ctx);
+   viaContextPtr vmesa = VIA_CONTEXT(ctx);
+   TNLcontext *tnl = TNL_CONTEXT(ctx);
+   struct vertex_buffer *VB = &TNL_CONTEXT(ctx)->vb;
+
+   {
+      GLboolean ptexHack = viaCheckPTexHack( ctx );
+      if (ptexHack != vmesa->ptexHack) {
+	 vmesa->ptexHack = ptexHack;
+	 vmesa->newRenderState |= _VIA_NEW_RENDERSTATE;
+      }
+   }
+
+   if (vmesa->newState) {
+      vmesa->newRenderState |= vmesa->newState;
+      viaValidateState( ctx );
+   }
+
+   if (vmesa->Fallback) {
+      tnl->Driver.Render.Start(ctx);
+      return;
+   }
+
+   if (vmesa->newRenderState) {
+      viaChooseVertexState(ctx);
+      viaChooseRenderState(ctx);
+      vmesa->newRenderState = 0;
+   }
+
+   /* Important:
+    */
+   VB->AttribPtr[VERT_ATTRIB_POS] = VB->NdcPtr;
 }
 
 static void viaRenderFinish(GLcontext *ctx)
@@ -745,9 +929,8 @@ void viaRasterPrimitive(GLcontext *ctx,
     
 /*     assert((vmesa->dmaLow & 0x4) == 0); */
 
-    if (vmesa->dmaCliprectAddr == 0) {
+    if (vmesa->dmaCliprectAddr == ~0) {
        if (VIA_DEBUG) fprintf(stderr, "reserve cliprect space at %x\n", vmesa->dmaLow);
-       assert(vmesa->dmaLow);
        vmesa->dmaCliprectAddr = vmesa->dmaLow;
        BEGIN_RING(8);
        OUT_RING( HC_HEADER2 );    
@@ -790,9 +973,11 @@ static void viaRenderPrimitive( GLcontext *ctx, GLuint prim )
 
 void viaFinishPrimitive(viaContextPtr vmesa)
 {
-   if (VIA_DEBUG) fprintf(stderr, "%s\n", __FUNCTION__);
+   if (VIA_DEBUG)
+      fprintf(stderr, "%s\n", __FUNCTION__);
 
-   if (!vmesa->dmaLastPrim) {
+   if (!vmesa->dmaLastPrim || vmesa->dmaCliprectAddr == ~0) {
+      assert(0);
    }
    else if (vmesa->dmaLow != vmesa->dmaLastPrim) {
       GLuint cmdA = vmesa->regCmdA_End | HC_HPLEND_MASK | HC_HPMValidN_MASK | HC_HE3Fire_MASK;    
@@ -818,6 +1003,9 @@ void viaFinishPrimitive(viaContextPtr vmesa)
 	 viaFlushDma( vmesa );
    }
    else {
+      if (VIA_DEBUG)
+	 fprintf(stderr, "remove empty primitive\n");
+
       /* Remove the primitive header:
        */
       vmesa->dmaLastPrim = 0;
@@ -827,7 +1015,7 @@ void viaFinishPrimitive(viaContextPtr vmesa)
        */
       if (vmesa->dmaCliprectAddr == vmesa->dmaLow - 8 * sizeof(GLuint)) {
 	 vmesa->dmaLow -= 8 * sizeof(GLuint);
-	 vmesa->dmaCliprectAddr = 0;
+	 vmesa->dmaCliprectAddr = ~0;
       }
    }
 
@@ -853,7 +1041,6 @@ void viaFallback(viaContextPtr vmesa, GLuint bit, GLboolean mode)
     if (mode) {
         vmesa->Fallback |= bit;
         if (oldfallback == 0) {
-            if (VIA_DEBUG) fprintf(stderr, "ENTER FALLBACK\n");
 	    VIA_FLUSH_DMA(vmesa);
             _swsetup_Wakeup(ctx);
             vmesa->renderIndex = ~0;
@@ -862,15 +1049,26 @@ void viaFallback(viaContextPtr vmesa, GLuint bit, GLboolean mode)
     else {
         vmesa->Fallback &= ~bit;
         if (oldfallback == bit) {
-            if (VIA_DEBUG) fprintf(stderr, "LEAVE FALLBACK\n");
+	    _swrast_flush( ctx );
+
 	    tnl->Driver.Render.Start = viaRenderStart;
             tnl->Driver.Render.PrimitiveNotify = viaRenderPrimitive;
             tnl->Driver.Render.Finish = viaRenderFinish;
-            tnl->Driver.Render.BuildVertices = viaBuildVertices;
+
+	    tnl->Driver.Render.BuildVertices = _tnl_build_vertices;
+	    tnl->Driver.Render.CopyPV = _tnl_copy_pv;
+	    tnl->Driver.Render.Interp = _tnl_interp;
+
+	    _tnl_invalidate_vertex_state( ctx, ~0 );
+	    _tnl_invalidate_vertices( ctx, ~0 );
+	    _tnl_install_attrs( ctx, 
+				vmesa->vertex_attrs, 
+				vmesa->vertex_attr_count,
+				vmesa->ViewportMatrix.m, 0 ); 
+
             vmesa->newState |= (_VIA_NEW_RENDERSTATE|_VIA_NEW_VERTEX);
         }
-    }
-    
+    }    
 }
 
 
@@ -881,6 +1079,7 @@ void viaFallback(viaContextPtr vmesa, GLuint bit, GLboolean mode)
 
 void viaInitTriFuncs(GLcontext *ctx)
 {
+    viaContextPtr vmesa = VIA_CONTEXT(ctx);
     TNLcontext *tnl = TNL_CONTEXT(ctx);
     static int firsttime = 1;
 
@@ -889,10 +1088,18 @@ void viaInitTriFuncs(GLcontext *ctx)
         firsttime = 0;
     }
 
-    tnl->Driver.RunPipeline = viaRunPipeline;
+    tnl->Driver.RunPipeline = _tnl_run_pipeline;
     tnl->Driver.Render.Start = viaRenderStart;
     tnl->Driver.Render.Finish = viaRenderFinish;
     tnl->Driver.Render.PrimitiveNotify = viaRenderPrimitive;
     tnl->Driver.Render.ResetLineStipple = _swrast_ResetLineStipple;
-    tnl->Driver.Render.BuildVertices = viaBuildVertices;
+    tnl->Driver.Render.BuildVertices = _tnl_build_vertices;
+    tnl->Driver.Render.CopyPV = _tnl_copy_pv;
+    tnl->Driver.Render.Interp = _tnl_interp;
+
+    _tnl_init_vertices( ctx, ctx->Const.MaxArrayLockSize + 12, 
+			(6 + 2*ctx->Const.MaxTextureUnits) * sizeof(GLfloat) );
+   
+    vmesa->verts = (char *)tnl->clipspace.vertex_buf;
+
 }

src/mesa/drivers/dri/unichrome/via_vb.c

@@ -1,380 +0,0 @@
-/*
- * Copyright 1998-2003 VIA Technologies, Inc. All Rights Reserved.
- * Copyright 2001-2003 S3 Graphics, Inc. All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * 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
- * VIA, S3 GRAPHICS, AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
- * DEALINGS IN THE SOFTWARE.
- */
-
-#include "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 "via_context.h"
-#include "via_vb.h"
-#include "via_ioctl.h"
-#include "via_tris.h"
-#include "via_state.h"
-
-static struct {
-    void                (*emit)(GLcontext *, GLuint, GLuint, void *, GLuint);
-    tnl_interp_func      interp;
-    tnl_copy_pv_func     copy_pv;
-    GLboolean           (*check_tex_sizes)(GLcontext *ctx);
-    GLuint               vertex_size;
-    GLuint               vertex_format;
-} setup_tab[VIA_MAX_SETUP];
-
-#define TINY_VERTEX_FORMAT 	1
-#define NOTEX_VERTEX_FORMAT 	2
-#define TEX0_VERTEX_FORMAT 	3
-#define TEX1_VERTEX_FORMAT 	4
-			    
-#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 & VIA_XYZW_BIT)
-#define DO_RGBA (IND & VIA_RGBA_BIT)
-#define DO_SPEC (IND & VIA_SPEC_BIT)
-#define DO_FOG  (IND & VIA_FOG_BIT)
-#define DO_TEX0 (IND & VIA_TEX0_BIT)
-#define DO_TEX1 (IND & VIA_TEX1_BIT)
-#define DO_TEX2 0
-#define DO_TEX3 0
-#define DO_PTEX (IND & VIA_PTEX_BIT)
-
-#define VERTEX viaVertex
-#define VERTEX_COLOR via_color_t
-#define GET_VIEWPORT_MAT() VIA_CONTEXT(ctx)->ViewportMatrix.m
-#define GET_TEXSOURCE(n)  n
-#define GET_VERTEX_FORMAT() VIA_CONTEXT(ctx)->vertexFormat
-#define GET_VERTEX_SIZE() VIA_CONTEXT(ctx)->vertexSize * sizeof(GLuint)
-#define GET_VERTEX_STORE() VIA_CONTEXT(ctx)->verts
-#define INVALIDATE_STORED_VERTICES()
-
-#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
-
-#define UNVIEWPORT_VARS  \
-   viaContextPtr vmesa = VIA_CONTEXT(ctx); \
-   GLfloat h = vmesa->driDrawable->h, depth_max = vmesa->depth_max, xoff = vmesa->drawXoff;
-
-#define UNVIEWPORT_X(x)  x - (SUBPIXEL_X + xoff)
-#define UNVIEWPORT_Y(y)  - y + h + SUBPIXEL_Y
-#define UNVIEWPORT_Z(z)  z * (float)depth_max
-
-#define PTEX_FALLBACK() FALLBACK(VIA_CONTEXT(ctx), VIA_FALLBACK_TEXTURE, 1)
-
-#define INTERP_VERTEX setup_tab[VIA_CONTEXT(ctx)->setupIndex].interp
-#define COPY_PV_VERTEX setup_tab[VIA_CONTEXT(ctx)->setupIndex].copy_pv
-
-
-/***********************************************************************
- *         Generate  pv-copying and translation functions              *
- ***********************************************************************/
-
-#define TAG(x) via_##x
-#include "tnl_dd/t_dd_vb.c"
-
-/***********************************************************************
- *             Generate vertex emit and interp functions               *
- ***********************************************************************/
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT)
-#define TAG(x) x##_wg
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_SPEC_BIT)
-#define TAG(x) x##_wgs
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_TEX0_BIT)
-#define TAG(x) x##_wgt0
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_TEX0_BIT | VIA_TEX1_BIT)
-#define TAG(x) x##_wgt0t1
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_TEX0_BIT | VIA_PTEX_BIT)
-#define TAG(x) x##_wgpt0
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_TEX0_BIT | VIA_TEX1_BIT |\
-             VIA_PTEX_BIT)
-#define TAG(x) x##_wgpt0t1
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_SPEC_BIT | VIA_TEX0_BIT)
-#define TAG(x) x##_wgst0
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_SPEC_BIT | VIA_TEX0_BIT |\
-             VIA_TEX1_BIT)
-#define TAG(x) x##_wgst0t1
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_SPEC_BIT | VIA_TEX0_BIT |\
-             VIA_PTEX_BIT)
-#define TAG(x) x##_wgspt0
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_SPEC_BIT | VIA_TEX0_BIT |\
-             VIA_TEX1_BIT | VIA_PTEX_BIT)	     
-#define TAG(x) x##_wgspt0t1
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_FOG_BIT)
-#define TAG(x) x##_wgf
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_FOG_BIT | VIA_SPEC_BIT)
-#define TAG(x) x##_wgfs
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_FOG_BIT | VIA_TEX0_BIT)
-#define TAG(x) x##_wgft0
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_FOG_BIT | VIA_TEX0_BIT |\
-             VIA_TEX1_BIT)
-#define TAG(x) x##_wgft0t1
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_FOG_BIT | VIA_TEX0_BIT |\
-             VIA_PTEX_BIT)
-#define TAG(x) x##_wgfpt0
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_FOG_BIT | VIA_TEX0_BIT |\
-             VIA_TEX1_BIT | VIA_PTEX_BIT)
-#define TAG(x) x##_wgfpt0t1
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_FOG_BIT | VIA_SPEC_BIT |\
-             VIA_TEX0_BIT)
-#define TAG(x) x##_wgfst0
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_FOG_BIT | VIA_SPEC_BIT |\
-             VIA_TEX0_BIT | VIA_TEX1_BIT)
-#define TAG(x) x##_wgfst0t1
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_FOG_BIT | VIA_SPEC_BIT |\
-             VIA_TEX0_BIT | VIA_PTEX_BIT)
-#define TAG(x) x##_wgfspt0
-#include "tnl_dd/t_dd_vbtmp.h"
-
-#define IND (VIA_XYZW_BIT | VIA_RGBA_BIT | VIA_FOG_BIT | VIA_SPEC_BIT |\
-             VIA_TEX0_BIT | VIA_TEX1_BIT | VIA_PTEX_BIT)
-#define TAG(x) x##_wgfspt0t1
-#include "tnl_dd/t_dd_vbtmp.h"
-
-static void init_setup_tab(void) {
-
-    init_wg();
-    init_wgs();
-    init_wgt0();
-    init_wgt0t1();
-    init_wgpt0();
-    init_wgpt0t1();    
-    init_wgst0();
-    init_wgst0t1();
-    init_wgspt0();
-    init_wgspt0t1();
-    init_wgf();
-    init_wgfs();
-    init_wgft0();
-    init_wgft0t1();
-    init_wgfpt0();
-    init_wgfpt0t1();
-    init_wgfst0();
-    init_wgfst0t1();
-    init_wgfspt0();
-    init_wgfspt0t1();
-}
-
-void viaPrintSetupFlags(char *msg, GLuint flags) {
-    if (VIA_DEBUG) fprintf(stderr, "%s(%x): %s%s%s%s%s%s\n",
-            msg,
-            (int)flags,
-            (flags & VIA_XYZW_BIT)     ? " xyzw," : "",
-            (flags & VIA_RGBA_BIT)     ? " rgba," : "",
-            (flags & VIA_SPEC_BIT)     ? " spec," : "",
-            (flags & VIA_FOG_BIT)      ? " fog," : "",
-            (flags & VIA_TEX0_BIT)     ? " tex-0," : "",
-            (flags & VIA_TEX1_BIT)     ? " tex-1," : "");
-}
-
-void viaCheckTexSizes(GLcontext *ctx) 
-{
-    TNLcontext *tnl = TNL_CONTEXT(ctx);
-    viaContextPtr vmesa = VIA_CONTEXT(ctx);
-
-    if (!setup_tab[vmesa->setupIndex].check_tex_sizes(ctx)) {
-        /* Invalidate stored verts
-         */
-        vmesa->setupNewInputs = ~0;
-        vmesa->setupIndex |= VIA_PTEX_BIT;
-
-        if (!vmesa->Fallback &&
-            !(ctx->_TriangleCaps & (DD_TRI_LIGHT_TWOSIDE|DD_TRI_UNFILLED))) {
-            tnl->Driver.Render.Interp = setup_tab[vmesa->setupIndex].interp;
-            tnl->Driver.Render.CopyPV = setup_tab[vmesa->setupIndex].copy_pv;
-        }
-
-	if (vmesa->Fallback)
-	   tnl->Driver.Render.Start(ctx);
-    }
-    if (VIA_DEBUG) fprintf(stderr, "%s - out\n", __FUNCTION__);    
-}
-
-void viaBuildVertices(GLcontext *ctx,
-                      GLuint start,
-                      GLuint count,
-                      GLuint newinputs) 
-{
-    viaContextPtr vmesa = VIA_CONTEXT(ctx);
-    GLuint stride = vmesa->vertexSize * sizeof(GLuint);
-    GLubyte *v = (GLubyte *)vmesa->verts + start * stride;
-
-    newinputs |= vmesa->setupNewInputs;
-    vmesa->setupNewInputs = 0;
-
-    if (!newinputs)
-        return;
-
-    if (newinputs & VERT_BIT_POS) {
-        setup_tab[vmesa->setupIndex].emit(ctx, start, count, v, stride);
-    }
-    else {
-        GLuint ind = 0;
-
-        if (newinputs & VERT_BIT_COLOR0)
-            ind |= VIA_RGBA_BIT;
-
-        if (newinputs & VERT_BIT_COLOR1)
-            ind |= VIA_SPEC_BIT;
-
-        if (newinputs & VERT_BIT_TEX0)
-            ind |= VIA_TEX0_BIT;
-
-        if (newinputs & VERT_BIT_TEX1)
-            ind |= VIA_TEX1_BIT;
-
-        if (newinputs & VERT_BIT_FOG)
-            ind |= VIA_FOG_BIT;
-
-        if (vmesa->setupIndex & VIA_PTEX_BIT)
-            ind = ~0;
-	
-        ind &= vmesa->setupIndex;
-	ind |= VIA_XYZW_BIT;
-        
-	if (ind) {
-            setup_tab[ind].emit(ctx, start, count, v, stride);
-        }
-    }
-}
-
-void viaChooseVertexState(GLcontext *ctx) {
-    TNLcontext *tnl = TNL_CONTEXT(ctx);
-    viaContextPtr vmesa = VIA_CONTEXT(ctx);
-    GLuint ind = VIA_XYZW_BIT | VIA_RGBA_BIT;
-
-    if (ctx->_TriangleCaps & DD_SEPARATE_SPECULAR)
-        ind |= VIA_SPEC_BIT;
-
-    if (ctx->Fog.Enabled)
-        ind |= VIA_FOG_BIT;
-
-    if (ctx->Texture._EnabledUnits > 1)
-        ind |= VIA_TEX1_BIT | VIA_TEX0_BIT;
-    else if (ctx->Texture._EnabledUnits == 1)
-        ind |= VIA_TEX0_BIT;
-
-    vmesa->setupIndex = ind;
-    viaPrintSetupFlags(__FUNCTION__, ind);
-
-    if (ctx->_TriangleCaps & (DD_TRI_LIGHT_TWOSIDE|DD_TRI_UNFILLED)) {
-        tnl->Driver.Render.Interp = via_interp_extras;
-        tnl->Driver.Render.CopyPV = via_copy_pv_extras;
-    }
-    else {
-        tnl->Driver.Render.Interp = setup_tab[ind].interp;
-        tnl->Driver.Render.CopyPV = setup_tab[ind].copy_pv;
-    }
-
-    vmesa->vertexSize = setup_tab[ind].vertex_size;
-    vmesa->vertexFormat = setup_tab[ind].vertex_format;
-
-    if (VIA_DEBUG) fprintf(stderr, "%s, size %d\n", __FUNCTION__, vmesa->vertexSize);
-}
-
-void *via_emit_contiguous_verts(GLcontext *ctx,
-				GLuint start,
-				GLuint count,
-				void *dest) 
-{
-    viaContextPtr vmesa = VIA_CONTEXT(ctx);
-    GLuint stride = vmesa->vertexSize * 4;    
-    setup_tab[vmesa->setupIndex].emit(ctx, start, count, dest, stride);
-    return (void *)((char *)dest + (count - start) * stride); 
-}
-
-
-void viaInitVB(GLcontext *ctx) 
-{
-    viaContextPtr vmesa = VIA_CONTEXT(ctx);
-    GLuint size = TNL_CONTEXT(ctx)->vb.Size;
-
-    vmesa->verts = ALIGN_MALLOC(size * 4 * 16, 32);
-
-    {
-        static int firsttime = 1;
-        if (firsttime) {
-            init_setup_tab();
-            firsttime = 0;
-        }
-    }
-}
-
-void viaFreeVB(GLcontext *ctx) {
-    viaContextPtr vmesa = VIA_CONTEXT(ctx);
-
-    if (vmesa->verts) {
-        ALIGN_FREE(vmesa->verts);
-        vmesa->verts = 0;
-    }
-}

src/mesa/drivers/dri/unichrome/via_vb.h

@@ -1,70 +0,0 @@
-/*
- * Copyright 1998-2003 VIA Technologies, Inc. All Rights Reserved.
- * Copyright 2001-2003 S3 Graphics, Inc. All Rights Reserved.
- *
- * Permission is hereby granted, free of charge, to any person obtaining a
- * copy of this software and associated documentation files (the "Software"),
- * to deal in the Software without restriction, including without limitation
- * 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
- * VIA, S3 GRAPHICS, AND/OR ITS SUPPLIERS BE LIABLE FOR ANY CLAIM, DAMAGES OR
- * OTHER LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE,
- * ARISING FROM, OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER
- * DEALINGS IN THE SOFTWARE.
- */
-
-#ifndef _VIAVB_H
-#define _VIAVB_H
-
-#include "mtypes.h"
-#include "swrast/swrast.h"
-
-#define VIA_TEX1_BIT	0x0001
-#define VIA_TEX0_BIT	0x0002
-#define VIA_RGBA_BIT	0x0004
-#define VIA_SPEC_BIT	0x0008
-#define VIA_FOG_BIT	0x0010
-#define VIA_XYZW_BIT	0x0020
-#define VIA_PTEX_BIT	0x0040
-#define VIA_MAX_SETUP	0x0080
-
-#define _VIA_NEW_VERTEX (_NEW_TEXTURE |                         \
-                         _DD_NEW_SEPARATE_SPECULAR |            \
-                         _DD_NEW_TRI_UNFILLED |                 \
-                         _DD_NEW_TRI_LIGHT_TWOSIDE |            \
-                         _NEW_FOG)
-
-
-extern void viaChooseVertexState(GLcontext *ctx);
-extern void viaCheckTexSizes(GLcontext *ctx);
-extern void viaBuildVertices(GLcontext *ctx,
-                             GLuint start,
-                             GLuint count,
-                             GLuint newinputs);
-
-
-extern void *via_emit_contiguous_verts(GLcontext *ctx,
-				       GLuint start,
-				       GLuint count,
-				       void *dest);
-
-extern void via_translate_vertex(GLcontext *ctx,
-                                 const viaVertex *src,
-                                 SWvertex *dst);
-
-extern void viaInitVB(GLcontext *ctx);
-extern void viaFreeVB(GLcontext *ctx);
-
-extern void via_print_vertex(GLcontext *ctx, const viaVertex *v);
-extern void viaPrintSetupFlags(char *msg, GLuint flags);
-
-#endif

src/mesa/glapi/glX_proto_send.py

@@ -181,6 +181,15 @@ generic_%u_byte( GLint rop, const void * ptr )
 		return offset
 
 
+	def large_emit_begin(self, indent, f):
+		print '%s    const GLint op = %s;' % (indent, f.opcode_real_name())
+		print '%s    const GLuint cmdlenLarge = cmdlen + 4;' % (indent)
+		print '%s    GLubyte * const pc = __glXFlushRenderBuffer(gc, gc->pc);' % (indent)
+		print '%s    (void) memcpy((void *)(pc + 0), (void *)(&cmdlenLarge), 4);' % (indent)
+		print '%s    (void) memcpy((void *)(pc + 4), (void *)(&op), 4);' % (indent)
+		return
+
+
 	def common_func_print_just_header(self, f):
 		print '#define %s %d' % (f.opcode_name(), f.opcode_value())
 
@@ -316,11 +325,8 @@ generic_%u_byte( GLint rop, const void * ptr )
 		if f.can_be_large:
 			print '%s}' % (indent)
 			print '%selse {' % (indent)
-			print '%s    const GLint op = %s;' % (indent, f.opcode_real_name())
-			print '%s    const GLuint cmdlenLarge = cmdlen + 4;' % (indent)
-			print '%s    GLubyte * const pc = __glXFlushRenderBuffer(gc, gc->pc);' % (indent)
-			print '%s    (void) memcpy((void *)(pc + 0), (void *)(&op), 4);' % (indent)
-			print '%s    (void) memcpy((void *)(pc + 4), (void *)(&cmdlenLarge), 4);' % (indent)
+
+			self.large_emit_begin(indent, f)
 			offset = self.common_emit_args(f, "pc", indent, 8, 1)
 			
 			p = f.variable_length_parameter()

src/mesa/shader/descrip.mms

@@ -19,6 +19,7 @@ LIBDIR = [---.lib]
 CFLAGS = /include=($(INCDIR),[])/define=(PTHREADS=1)/name=(as_is,short)
 
 SOURCES = \
+	atifragshader.c \
 	arbfragparse.c \
 	arbprogparse.c \
 	arbprogram.c \
@@ -32,6 +33,7 @@ SOURCES = \
 	shaderobjects.c
 
 OBJECTS = \
+	atifragshader.obj,\
 	arbfragparse.obj,\
 	arbprogparse.obj,\
 	arbprogram.obj,\
@@ -58,7 +60,7 @@ clean :
 	purge
 	delete *.obj;*
 
-
+atifragshader.obj : atifragshader.c
 arbfragparse.obj : arbfragparse.c
 arbprogparse.obj : arbprogparse.c
 arbprogram.obj : arbprogram.c

src/mesa/swrast/descrip.mms

@@ -24,11 +24,11 @@ SOURCES = s_aaline.c s_aatriangle.c s_accum.c s_alpha.c s_alphabuf.c \
         s_drawpix.c s_feedback.c s_fog.c s_imaging.c s_lines.c s_logic.c \
 	s_masking.c s_nvfragprog.c s_pixeltex.c s_points.c s_readpix.c \
 	s_span.c s_stencil.c s_texstore.c s_texture.c s_triangle.c s_zoom.c \
-	s_auxbuffer.c
+	s_auxbuffer.c s_atifragshader.c
  
 OBJECTS = s_aaline.obj,s_aatriangle.obj,s_accum.obj,s_alpha.obj,\
 	s_alphabuf.obj,s_auxbuffer.obj,s_bitmap.obj,s_blend.obj,\
-	s_buffers.obj,s_context.obj,\
+	s_buffers.obj,s_context.obj,s_atifragshader.obj,\
 	s_copypix.obj,s_depth.obj,s_drawpix.obj,s_feedback.obj,s_fog.obj,\
 	s_imaging.obj,s_lines.obj,s_logic.obj,s_masking.obj,s_nvfragprog.obj,\
 	s_pixeltex.obj,s_points.obj,s_readpix.obj,s_span.obj,s_stencil.obj,\
@@ -47,6 +47,7 @@ clean :
 	purge
 	delete *.obj;*
 
+s_atifragshader.obj : s_atifragshader.c
 s_aaline.obj : s_aaline.c
 s_aatriangle.obj : s_aatriangle.c
 s_accum.obj : s_accum.c

src/mesa/x86/3dnow_normal.S

@@ -1,4 +1,4 @@
-/* $Id: 3dnow_normal.S,v 1.8 2004/04/26 10:10:25 alanh Exp $ */
+/* $Id: 3dnow_normal.S,v 1.9 2005/01/04 14:33:47 brianp Exp $ */
 
 /*
  * Mesa 3-D graphics library
@@ -41,6 +41,7 @@
 
 ALIGNTEXT16
 GLOBL GLNAME(_mesa_3dnow_transform_normalize_normals)
+HIDDEN(_mesa_3dnow_transform_normalize_normals)
 GLNAME(_mesa_3dnow_transform_normalize_normals):
 
 #define FRAME_OFFSET 12
@@ -218,6 +219,7 @@ LLBL (G3TN_end):
 
 ALIGNTEXT16
 GLOBL GLNAME(_mesa_3dnow_transform_normalize_normals_no_rot)
+HIDDEN(_mesa_3dnow_transform_normalize_normals_no_rot)
 GLNAME(_mesa_3dnow_transform_normalize_normals_no_rot):
 
 #undef FRAME_OFFSET
@@ -357,6 +359,7 @@ LLBL (G3TNNR_end):
 
 ALIGNTEXT16
 GLOBL GLNAME(_mesa_3dnow_transform_rescale_normals_no_rot)
+HIDDEN(_mesa_3dnow_transform_rescale_normals_no_rot)
 GLNAME(_mesa_3dnow_transform_rescale_normals_no_rot):
 
 #undef FRAME_OFFSET
@@ -428,6 +431,7 @@ LLBL (G3TRNR_end):
 
 ALIGNTEXT16
 GLOBL GLNAME(_mesa_3dnow_transform_rescale_normals)
+HIDDEN(_mesa_3dnow_transform_rescale_normals)
 GLNAME(_mesa_3dnow_transform_rescale_normals):
 
 #undef  FRAME_OFFSET
@@ -525,6 +529,7 @@ LLBL (G3TR_end):
 
 ALIGNTEXT16
 GLOBL GLNAME(_mesa_3dnow_transform_normals_no_rot)
+HIDDEN(_mesa_3dnow_transform_normals_no_rot)
 GLNAME(_mesa_3dnow_transform_normals_no_rot):
 
 #undef  FRAME_OFFSET
@@ -591,6 +596,7 @@ LLBL (G3TNR_end):
 
 ALIGNTEXT16
 GLOBL GLNAME(_mesa_3dnow_transform_normals)
+HIDDEN(_mesa_3dnow_transform_normals)
 GLNAME(_mesa_3dnow_transform_normals):
 
 #undef  FRAME_OFFSET
@@ -675,6 +681,7 @@ LLBL (G3T_end):
 
 ALIGNTEXT16
 GLOBL GLNAME(_mesa_3dnow_normalize_normals)
+HIDDEN(_mesa_3dnow_normalize_normals)
 GLNAME(_mesa_3dnow_normalize_normals):
 
 #undef  FRAME_OFFSET
@@ -786,6 +793,7 @@ LLBL (G3N_end):
 
 ALIGNTEXT16
 GLOBL GLNAME(_mesa_3dnow_rescale_normals)
+HIDDEN(_mesa_3dnow_rescale_normals)
 GLNAME(_mesa_3dnow_rescale_normals):
 
 #undef  FRAME_OFFSET

src/mesa/x86/3dnow_xform1.S

@@ -1,4 +1,4 @@
-/* $Id: 3dnow_xform1.S,v 1.2 2004/04/26 10:10:25 alanh Exp $ */
+/* $Id: 3dnow_xform1.S,v 1.3 2005/01/04 14:33:47 brianp Exp $ */
 
 /*
  * Mesa 3-D graphics library
@@ -35,6 +35,7 @@
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points1_general )
+HIDDEN(_mesa_3dnow_transform_points1_general)
 GLNAME( _mesa_3dnow_transform_points1_general ):
 
     PUSH_L    ( ESI )
@@ -99,6 +100,7 @@ LLBL( G3TPGR_3 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points1_identity )
+HIDDEN(_mesa_3dnow_transform_points1_identity)
 GLNAME( _mesa_3dnow_transform_points1_identity ):
 
     PUSH_L    ( ESI )
@@ -146,6 +148,7 @@ LLBL( G3TPIR_4 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points1_3d_no_rot )
+HIDDEN(_mesa_3dnow_transform_points1_3d_no_rot)
 GLNAME( _mesa_3dnow_transform_points1_3d_no_rot ):
 
     PUSH_L    ( ESI )
@@ -203,6 +206,7 @@ LLBL( G3TP3NRR_3 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points1_perspective )
+HIDDEN(_mesa_3dnow_transform_points1_perspective)
 GLNAME( _mesa_3dnow_transform_points1_perspective ):
 
     PUSH_L    ( ESI )
@@ -256,6 +260,7 @@ LLBL( G3TPPR_3 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points1_2d )
+HIDDEN(_mesa_3dnow_transform_points1_2d)
 GLNAME( _mesa_3dnow_transform_points1_2d ):
 
     PUSH_L    ( ESI )
@@ -311,6 +316,7 @@ LLBL( G3TP2R_3 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points1_2d_no_rot )
+HIDDEN(_mesa_3dnow_transform_points1_2d_no_rot)
 GLNAME( _mesa_3dnow_transform_points1_2d_no_rot ):
 
     PUSH_L    ( ESI )
@@ -364,6 +370,7 @@ LLBL( G3TP2NRR_3 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points1_3d )
+HIDDEN(_mesa_3dnow_transform_points1_3d)
 GLNAME( _mesa_3dnow_transform_points1_3d ):
 
     PUSH_L    ( ESI )

src/mesa/x86/3dnow_xform2.S

@@ -1,4 +1,4 @@
-/* $Id: 3dnow_xform2.S,v 1.2 2004/04/26 10:10:25 alanh Exp $ */
+/* $Id: 3dnow_xform2.S,v 1.3 2005/01/04 14:33:47 brianp Exp $ */
 
 /*
  * Mesa 3-D graphics library
@@ -35,6 +35,7 @@
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points2_general )
+HIDDEN(_mesa_3dnow_transform_points2_general)
 GLNAME( _mesa_3dnow_transform_points2_general ):
 
     PUSH_L    ( ESI )
@@ -115,6 +116,7 @@ LLBL( G3TPGR_3 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points2_perspective )
+HIDDEN(_mesa_3dnow_transform_points2_perspective)
 GLNAME( _mesa_3dnow_transform_points2_perspective ):
 
     PUSH_L    ( ESI )
@@ -170,6 +172,7 @@ LLBL( G3TPPR_3 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points2_3d )
+HIDDEN(_mesa_3dnow_transform_points2_3d)
 GLNAME( _mesa_3dnow_transform_points2_3d ):
 
     PUSH_L    ( ESI )
@@ -246,6 +249,7 @@ LLBL( G3TP3R_3 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points2_3d_no_rot )
+HIDDEN(_mesa_3dnow_transform_points2_3d_no_rot)
 GLNAME( _mesa_3dnow_transform_points2_3d_no_rot ):
 
     PUSH_L    ( ESI )
@@ -304,6 +308,7 @@ LLBL( G3TP3NRR_3 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points2_2d )
+HIDDEN(_mesa_3dnow_transform_points2_2d)
 GLNAME( _mesa_3dnow_transform_points2_2d ):
 
     PUSH_L    ( ESI )
@@ -367,6 +372,7 @@ LLBL( G3TP2R_3 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points2_2d_no_rot )
+HIDDEN(_mesa_3dnow_transform_points2_2d_no_rot)
 GLNAME( _mesa_3dnow_transform_points2_2d_no_rot ):
 
     PUSH_L    ( ESI )
@@ -422,6 +428,7 @@ LLBL( G3TP2NRR_3 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points2_identity )
+HIDDEN(_mesa_3dnow_transform_points2_identity)
 GLNAME( _mesa_3dnow_transform_points2_identity ):
 
     PUSH_L    ( ESI )

src/mesa/x86/3dnow_xform3.S

@@ -1,4 +1,4 @@
-/* $Id: 3dnow_xform3.S,v 1.3 2004/04/26 10:10:25 alanh Exp $ */
+/* $Id: 3dnow_xform3.S,v 1.4 2005/01/04 14:33:47 brianp Exp $ */
 
 /*
  * Mesa 3-D graphics library
@@ -35,6 +35,7 @@
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points3_general )
+HIDDEN(_mesa_3dnow_transform_points3_general)
 GLNAME( _mesa_3dnow_transform_points3_general ):
 
     PUSH_L    ( ESI )
@@ -119,6 +120,7 @@ LLBL( G3TPGR_2 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points3_perspective )
+HIDDEN(_mesa_3dnow_transform_points3_perspective)
 GLNAME( _mesa_3dnow_transform_points3_perspective ):
 
     PUSH_L    ( ESI )
@@ -199,6 +201,7 @@ LLBL( G3TPPR_2 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points3_3d )
+HIDDEN(_mesa_3dnow_transform_points3_3d)
 GLNAME( _mesa_3dnow_transform_points3_3d ):
 
     PUSH_L    ( ESI )
@@ -286,6 +289,7 @@ LLBL( G3TP3R_2 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points3_3d_no_rot )
+HIDDEN(_mesa_3dnow_transform_points3_3d_no_rot)
 GLNAME( _mesa_3dnow_transform_points3_3d_no_rot ):
 
     PUSH_L    ( ESI )
@@ -361,6 +365,7 @@ LLBL( G3TP3NRR_2 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points3_2d )
+HIDDEN(_mesa_3dnow_transform_points3_2d)
 GLNAME( _mesa_3dnow_transform_points3_2d ):
 
     PUSH_L    ( ESI )
@@ -433,6 +438,7 @@ LLBL( G3TP2R_3 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points3_2d_no_rot )
+HIDDEN(_mesa_3dnow_transform_points3_2d_no_rot)
 GLNAME( _mesa_3dnow_transform_points3_2d_no_rot ):
 
     PUSH_L    ( ESI )
@@ -499,6 +505,7 @@ LLBL( G3TP2NRR_2 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points3_identity )
+HIDDEN(_mesa_3dnow_transform_points3_identity)
 GLNAME( _mesa_3dnow_transform_points3_identity ):
 
     PUSH_L    ( ESI )

src/mesa/x86/3dnow_xform4.S

@@ -1,4 +1,4 @@
-/* $Id: 3dnow_xform4.S,v 1.3 2004/04/26 10:10:25 alanh Exp $ */
+/* $Id: 3dnow_xform4.S,v 1.4 2005/01/04 14:33:47 brianp Exp $ */
 
 /*
  * Mesa 3-D graphics library
@@ -35,6 +35,7 @@
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points4_general )
+HIDDEN(_mesa_3dnow_transform_points4_general)
 GLNAME( _mesa_3dnow_transform_points4_general ):
 
     PUSH_L    ( ESI )
@@ -127,6 +128,7 @@ LLBL( G3TPGR_2 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points4_perspective )
+HIDDEN(_mesa_3dnow_transform_points4_perspective)
 GLNAME( _mesa_3dnow_transform_points4_perspective ):
 
     PUSH_L    ( ESI )
@@ -206,6 +208,7 @@ LLBL( G3TPPR_2 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points4_3d )
+HIDDEN(_mesa_3dnow_transform_points4_3d)
 GLNAME( _mesa_3dnow_transform_points4_3d ):
 
     PUSH_L    ( ESI )
@@ -295,6 +298,7 @@ LLBL( G3TP3R_2 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points4_3d_no_rot )
+HIDDEN(_mesa_3dnow_transform_points4_3d_no_rot)
 GLNAME( _mesa_3dnow_transform_points4_3d_no_rot ):
 
     PUSH_L    ( ESI )
@@ -367,6 +371,7 @@ LLBL( G3TP3NRR_2 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points4_2d )
+HIDDEN(_mesa_3dnow_transform_points4_2d)
 GLNAME( _mesa_3dnow_transform_points4_2d ):
 
     PUSH_L    ( ESI )
@@ -441,6 +446,7 @@ LLBL( G3TP2R_2 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points4_2d_no_rot )
+HIDDEN(_mesa_3dnow_transform_points4_2d_no_rot)
 GLNAME( _mesa_3dnow_transform_points4_2d_no_rot ):
 
     PUSH_L    ( ESI )
@@ -508,6 +514,7 @@ LLBL( G3TP2NRR_3 ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_3dnow_transform_points4_identity )
+HIDDEN(_mesa_3dnow_transform_points4_identity)
 GLNAME( _mesa_3dnow_transform_points4_identity ):
 
     PUSH_L    ( ESI )

src/mesa/x86/assyntax.h

@@ -1730,4 +1730,11 @@ SECTION _DATA public align=16 class=DATA use32 flat
 #define TLBL(a)		CONCAT(a,$)
 #endif
 
+/* hidden symbol visibility support */
+#ifdef GNU_ASSEMBLER
+#define HIDDEN(a)       .hidden a
+#else
+#define HIDDEN(a)
+#endif
+
 #endif /* __ASSYNTAX_H__ */

src/mesa/x86/common_x86_asm.S

@@ -46,6 +46,7 @@
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_x86_has_cpuid)
+HIDDEN(_mesa_x86_has_cpuid)
 GLNAME(_mesa_x86_has_cpuid):
 
 	/* Test for the CPUID command.  If the ID Flag bit in EFLAGS
@@ -69,6 +70,7 @@ GLNAME(_mesa_x86_has_cpuid):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_x86_cpuid)
+HIDDEN(_mesa_x86_cpuid)
 GLNAME(_mesa_x86_cpuid):
 
 	MOV_L	(REGOFF(4, ESP), EAX)		/* cpuid op */
@@ -92,6 +94,7 @@ GLNAME(_mesa_x86_cpuid):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_x86_cpuid_eax)
+HIDDEN(_mesa_x86_cpuid_eax)
 GLNAME(_mesa_x86_cpuid_eax):
 
 	MOV_L	(REGOFF(4, ESP), EAX)		/* cpuid op */
@@ -104,6 +107,7 @@ GLNAME(_mesa_x86_cpuid_eax):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_x86_cpuid_ebx)
+HIDDEN(_mesa_x86_cpuid_ebx)
 GLNAME(_mesa_x86_cpuid_ebx):
 
 	MOV_L	(REGOFF(4, ESP), EAX)		/* cpuid op */
@@ -117,6 +121,7 @@ GLNAME(_mesa_x86_cpuid_ebx):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_x86_cpuid_ecx)
+HIDDEN(_mesa_x86_cpuid_ecx)
 GLNAME(_mesa_x86_cpuid_ecx):
 
 	MOV_L	(REGOFF(4, ESP), EAX)		/* cpuid op */
@@ -130,6 +135,7 @@ GLNAME(_mesa_x86_cpuid_ecx):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_x86_cpuid_edx)
+HIDDEN(_mesa_x86_cpuid_edx)
 GLNAME(_mesa_x86_cpuid_edx):
 
 	MOV_L	(REGOFF(4, ESP), EAX)		/* cpuid op */
@@ -155,6 +161,7 @@ GLNAME(_mesa_x86_cpuid_edx):
  */
 ALIGNTEXT4
 GLOBL GLNAME( _mesa_test_os_sse_support )
+HIDDEN(_mesa_test_os_sse_support)
 GLNAME( _mesa_test_os_sse_support ):
 
 	XORPS	( XMM0, XMM0 )
@@ -169,6 +176,7 @@ GLNAME( _mesa_test_os_sse_support ):
  */
 ALIGNTEXT4
 GLOBL GLNAME( _mesa_test_os_sse_exception_support )
+HIDDEN(_mesa_test_os_sse_exception_support)
 GLNAME( _mesa_test_os_sse_exception_support ):
 
 	PUSH_L	( EBP )

src/mesa/x86/mmx_blendtmp.h

@@ -13,7 +13,7 @@
  */
 ALIGNTEXT16
 GLOBL GLNAME( TAG(_mesa_mmx_blend) )
-
+HIDDEN( TAG(_mesa_mmx_blend) )
 GLNAME( TAG(_mesa_mmx_blend) ):
 
     PUSH_L     ( EBP )

src/mesa/x86/read_rgba_span_x86.S

@@ -76,6 +76,7 @@ mask:
  */
 
 .globl _generic_read_RGBA_span_BGRA8888_REV_MMX
+.hidden _generic_read_RGBA_span_BGRA8888_REV_MMX
 	.type	_generic_read_RGBA_span_BGRA8888_REV_MMX, @function
 _generic_read_RGBA_span_BGRA8888_REV_MMX:
 	pushl	%ebx
@@ -171,6 +172,7 @@ _generic_read_RGBA_span_BGRA8888_REV_MMX:
  */
 
 .globl _generic_read_RGBA_span_BGRA8888_REV_SSE
+.hidden _generic_read_RGBA_span_BGRA8888_REV_SSE
 	.type	_generic_read_RGBA_span_BGRA8888_REV_SSE, @function
 _generic_read_RGBA_span_BGRA8888_REV_SSE:
 	pushl	%esi
@@ -330,6 +332,7 @@ _generic_read_RGBA_span_BGRA8888_REV_SSE:
 
 	.text
 .globl _generic_read_RGBA_span_BGRA8888_REV_SSE2
+.hidden _generic_read_RGBA_span_BGRA8888_REV_SSE2
 	.type	_generic_read_RGBA_span_BGRA8888_REV_SSE2, @function
 _generic_read_RGBA_span_BGRA8888_REV_SSE2:
 	pushl	%esi
@@ -508,6 +511,7 @@ alpha:	.long	0x00000000
 
 	.text
 	.globl	_generic_read_RGBA_span_RGB565_MMX
+        .hidden _generic_read_RGBA_span_RGB565_MMX
 	.type	_generic_read_RGBA_span_RGB565_MMX, @function
 
 _generic_read_RGBA_span_RGB565_MMX:

src/mesa/x86/sse_normal.S

@@ -1,4 +1,4 @@
-/* $Id: sse_normal.S,v 1.4 2004/04/26 10:10:25 alanh Exp $ */
+/* $Id: sse_normal.S,v 1.5 2005/01/04 14:33:47 brianp Exp $ */
 
 /*
  * Mesa 3-D graphics library
@@ -45,6 +45,7 @@
 
 ALIGNTEXT16
 GLOBL GLNAME(_mesa_sse_transform_rescale_normals_no_rot)
+HIDDEN(_mesa_sse_transform_rescale_normals_no_rot)
 GLNAME(_mesa_sse_transform_rescale_normals_no_rot):
 
 #define FRAME_OFFSET 8
@@ -106,6 +107,7 @@ LLBL(K_G3TRNNRR_finish):
 
 ALIGNTEXT16
 GLOBL GLNAME(_mesa_sse_transform_rescale_normals)
+HIDDEN(_mesa_sse_transform_rescale_normals)
 GLNAME(_mesa_sse_transform_rescale_normals):
 
 #define FRAME_OFFSET 8
@@ -198,6 +200,7 @@ LLBL(K_G3TRNR_finish):
 
 ALIGNTEXT16
 GLOBL GLNAME(_mesa_sse_transform_normals_no_rot)
+HIDDEN(_mesa_sse_transform_normals_no_rot)
 GLNAME(_mesa_sse_transform_normals_no_rot):
 
 #define FRAME_OFFSET 8

src/mesa/x86/sse_xform1.S

@@ -1,4 +1,4 @@
-/* $Id: sse_xform1.S,v 1.2 2004/04/26 10:10:25 alanh Exp $ */
+/* $Id: sse_xform1.S,v 1.3 2005/01/04 14:33:47 brianp Exp $ */
 
 /*
  * Mesa 3-D graphics library
@@ -44,6 +44,7 @@
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points1_general)
+HIDDEN( _mesa_sse_transform_points1_general )
 GLNAME( _mesa_sse_transform_points1_general ):
 
 #define FRAME_OFFSET 8
@@ -100,6 +101,7 @@ LLBL(K_GTP1GR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points1_identity)
+HIDDEN(_mesa_sse_transform_points1_identity)
 GLNAME( _mesa_sse_transform_points1_identity ):
 
 #define FRAME_OFFSET 8
@@ -151,6 +153,7 @@ LLBL(K_GTP1IR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points1_3d_no_rot)
+HIDDEN(_mesa_sse_transform_points1_3d_no_rot)
 GLNAME(_mesa_sse_transform_points1_3d_no_rot):
 
 #define FRAME_OFFSET 8
@@ -211,6 +214,7 @@ LLBL(K_GTP13DNRR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points1_perspective)
+HIDDEN(_mesa_sse_transform_points1_perspective)
 GLNAME(_mesa_sse_transform_points1_perspective):
 
 #define FRAME_OFFSET 8
@@ -269,6 +273,7 @@ LLBL(K_GTP13PR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points1_2d)
+HIDDEN(_mesa_sse_transform_points1_2d)
 GLNAME(_mesa_sse_transform_points1_2d):
 
 #define FRAME_OFFSET 8
@@ -323,6 +328,7 @@ LLBL(K_GTP13P2DR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points1_2d_no_rot)
+HIDDEN(_mesa_sse_transform_points1_2d_no_rot)
 GLNAME(_mesa_sse_transform_points1_2d_no_rot):
 
 #define FRAME_OFFSET 8
@@ -379,6 +385,7 @@ LLBL(K_GTP13P2DNRR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points1_3d)
+HIDDEN(_mesa_sse_transform_points1_3d)
 GLNAME(_mesa_sse_transform_points1_3d):
 
 #define FRAME_OFFSET 8

src/mesa/x86/sse_xform2.S

@@ -1,4 +1,4 @@
-/* $Id: sse_xform2.S,v 1.2 2004/04/26 10:10:25 alanh Exp $ */
+/* $Id: sse_xform2.S,v 1.3 2005/01/04 14:33:47 brianp Exp $ */
 
 /*
  * Mesa 3-D graphics library
@@ -44,6 +44,7 @@
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points2_general)
+HIDDEN (_mesa_sse_transform_points2_general)
 GLNAME( _mesa_sse_transform_points2_general ):
 
 #define FRAME_OFFSET 8
@@ -104,6 +105,7 @@ LLBL(K_GTP2GR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points2_identity)
+HIDDEN(_mesa_sse_transform_points2_identity)
 GLNAME( _mesa_sse_transform_points2_identity ):
 
 #define FRAME_OFFSET 8
@@ -156,6 +158,7 @@ LLBL(K_GTP2IR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points2_3d_no_rot)
+HIDDEN(_mesa_sse_transform_points2_3d_no_rot)
 GLNAME(_mesa_sse_transform_points2_3d_no_rot):
 
 #define FRAME_OFFSET 8
@@ -215,6 +218,7 @@ LLBL(K_GTP23DNRR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points2_perspective)
+HIDDEN(_mesa_sse_transform_points2_perspective)
 GLNAME(_mesa_sse_transform_points2_perspective):
 
 #define FRAME_OFFSET 8
@@ -273,6 +277,7 @@ LLBL(K_GTP23PR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points2_2d)
+HIDDEN(_mesa_sse_transform_points2_2d)
 GLNAME(_mesa_sse_transform_points2_2d):
 
 #define FRAME_OFFSET 8
@@ -335,6 +340,7 @@ LLBL(K_GTP23P2DR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points2_2d_no_rot)
+HIDDEN(_mesa_sse_transform_points2_2d_no_rot)
 GLNAME(_mesa_sse_transform_points2_2d_no_rot):
 
 #define FRAME_OFFSET 8
@@ -391,6 +397,7 @@ LLBL(K_GTP23P2DNRR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points2_3d)
+HIDDEN(_mesa_sse_transform_points2_3d)
 GLNAME(_mesa_sse_transform_points2_3d):
 
 #define FRAME_OFFSET 8

src/mesa/x86/sse_xform3.S

@@ -1,4 +1,4 @@
-/* $Id: sse_xform3.S,v 1.2 2004/04/26 10:10:25 alanh Exp $ */
+/* $Id: sse_xform3.S,v 1.3 2005/01/04 14:33:47 brianp Exp $ */
 
 /*
  * Mesa 3-D graphics library
@@ -44,6 +44,7 @@
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points3_general)
+HIDDEN(_mesa_sse_transform_points3_general)
 GLNAME( _mesa_sse_transform_points3_general ):
 
 #define FRAME_OFFSET 8
@@ -113,6 +114,7 @@ LLBL(K_GTPGR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points3_identity)
+HIDDEN(_mesa_sse_transform_points3_identity)
 GLNAME( _mesa_sse_transform_points3_identity ):
 
 #define FRAME_OFFSET 8
@@ -167,6 +169,7 @@ LLBL(K_GTPIR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points3_3d_no_rot)
+HIDDEN(_mesa_sse_transform_points3_3d_no_rot)
 GLNAME(_mesa_sse_transform_points3_3d_no_rot):
 
 #define FRAME_OFFSET 8
@@ -233,6 +236,7 @@ LLBL(K_GTP3DNRR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points3_perspective)
+HIDDEN(_mesa_sse_transform_points3_perspective)
 GLNAME(_mesa_sse_transform_points3_perspective):
 
 #define FRAME_OFFSET 8
@@ -305,6 +309,7 @@ LLBL(K_GTP3PR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points3_2d)
+HIDDEN(_mesa_sse_transform_points3_2d)
 GLNAME(_mesa_sse_transform_points3_2d):
 
 #define FRAME_OFFSET 8
@@ -369,6 +374,7 @@ LLBL(K_GTP3P2DR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points3_2d_no_rot)
+HIDDEN(_mesa_sse_transform_points3_2d_no_rot)
 GLNAME(_mesa_sse_transform_points3_2d_no_rot):
 
 #define FRAME_OFFSET 8
@@ -429,6 +435,7 @@ LLBL(K_GTP3P2DNRR_finish):
 
 ALIGNTEXT4
 GLOBL GLNAME(_mesa_sse_transform_points3_3d)
+HIDDEN(_mesa_sse_transform_points3_3d)
 GLNAME(_mesa_sse_transform_points3_3d):
 
 #define FRAME_OFFSET 8

src/mesa/x86/sse_xform4.S

@@ -1,4 +1,4 @@
-/* $Id: sse_xform4.S,v 1.2 2004/04/26 10:10:25 alanh Exp $ */
+/* $Id: sse_xform4.S,v 1.3 2005/01/04 14:33:47 brianp Exp $ */
 
 /*
  * Mesa 3-D graphics library
@@ -41,6 +41,7 @@
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_sse_transform_points4_general )
+HIDDEN(_mesa_sse_transform_points4_general)
 GLNAME( _mesa_sse_transform_points4_general ):
 
 	PUSH_L( ESI )
@@ -112,6 +113,7 @@ LLBL( sse_general_done ):
 
 ALIGNTEXT4
 GLOBL GLNAME( _mesa_sse_transform_points4_3d )
+HIDDEN(_mesa_sse_transform_points4_3d)
 GLNAME( _mesa_sse_transform_points4_3d ):
 
 	PUSH_L( ESI )
@@ -183,6 +185,7 @@ LLBL( K_GTP43P3DR_finish ):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_sse_transform_points4_identity )
+HIDDEN(_mesa_sse_transform_points4_identity)
 GLNAME( _mesa_sse_transform_points4_identity ):
 
 	PUSH_L( ESI )

src/mesa/x86/x86_cliptest.S

@@ -1,4 +1,4 @@
-/* $Id: x86_cliptest.S,v 1.10 2002/06/11 01:26:58 brianp Exp $ */
+/* $Id: x86_cliptest.S,v 1.11 2005/01/04 14:33:47 brianp Exp $ */
 
 /*
  * Mesa 3-D graphics library
@@ -101,6 +101,7 @@ clip_table:
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_cliptest_points4 )
+HIDDEN(_mesa_x86_cliptest_points4)
 GLNAME( _mesa_x86_cliptest_points4 ):
 
 #ifdef ELFPIC
@@ -275,6 +276,7 @@ LLBL(ctp4_finish):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_cliptest_points4_np )
+HIDDEN(_mesa_x86_cliptest_points4_np)
 GLNAME( _mesa_x86_cliptest_points4_np ):
 
 #ifdef ELFPIC

src/mesa/x86/x86_xform2.S

@@ -1,4 +1,4 @@
-/* $Id: x86_xform2.S,v 1.2 2002/03/07 21:40:08 brianp Exp $ */
+/* $Id: x86_xform2.S,v 1.3 2005/01/04 14:33:47 brianp Exp $ */
 
 /*
  * Mesa 3-D graphics library
@@ -66,6 +66,7 @@
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points2_general )
+HIDDEN(_mesa_x86_transform_points2_general)
 GLNAME( _mesa_x86_transform_points2_general ):
 
 #define FRAME_OFFSET 8
@@ -156,6 +157,7 @@ LLBL(x86_p2_gr_done):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points2_perspective )
+HIDDEN(_mesa_x86_transform_points2_perspective)
 GLNAME( _mesa_x86_transform_points2_perspective ):
 
 #define FRAME_OFFSET 12
@@ -221,6 +223,7 @@ LLBL(x86_p2_pr_done):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points2_3d )
+HIDDEN(_mesa_x86_transform_points2_3d)
 GLNAME( _mesa_x86_transform_points2_3d ):
 
 #define FRAME_OFFSET 8
@@ -302,6 +305,7 @@ LLBL(x86_p2_3dr_done):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points2_3d_no_rot )
+HIDDEN(_mesa_x86_transform_points2_3d_no_rot)
 GLNAME( _mesa_x86_transform_points2_3d_no_rot ):
 
 #define FRAME_OFFSET 12
@@ -371,6 +375,7 @@ LLBL(x86_p2_3dnrr_done):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points2_2d )
+HIDDEN(_mesa_x86_transform_points2_2d)
 GLNAME( _mesa_x86_transform_points2_2d ):
 
 #define FRAME_OFFSET 8
@@ -443,6 +448,7 @@ LLBL(x86_p2_2dr_done):
 
 ALIGNTEXT4
 GLOBL GLNAME( _mesa_x86_transform_points2_2d_no_rot )
+HIDDEN(_mesa_x86_transform_points2_2d_no_rot)
 GLNAME( _mesa_x86_transform_points2_2d_no_rot ):
 
 #define FRAME_OFFSET 8
@@ -507,6 +513,7 @@ LLBL(x86_p2_2dnrr_done):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points2_identity )
+HIDDEN(_mesa_x86_transform_points2_identity)
 GLNAME( _mesa_x86_transform_points2_identity ):
 
 #define FRAME_OFFSET 12

src/mesa/x86/x86_xform3.S

@@ -1,4 +1,4 @@
-/* $Id: x86_xform3.S,v 1.2 2002/03/07 21:40:08 brianp Exp $ */
+/* $Id: x86_xform3.S,v 1.3 2005/01/04 14:33:47 brianp Exp $ */
 
 /*
  * Mesa 3-D graphics library
@@ -66,6 +66,7 @@
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points3_general )
+HIDDEN(_mesa_x86_transform_points3_general)
 GLNAME( _mesa_x86_transform_points3_general ):
 
 #define FRAME_OFFSET 8
@@ -172,6 +173,7 @@ LLBL(x86_p3_gr_done):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points3_perspective )
+HIDDEN(_mesa_x86_transform_points3_perspective)
 GLNAME( _mesa_x86_transform_points3_perspective ):
 
 #define FRAME_OFFSET 12
@@ -252,6 +254,7 @@ LLBL(x86_p3_pr_done):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points3_3d )
+HIDDEN(_mesa_x86_transform_points3_3d)
 GLNAME( _mesa_x86_transform_points3_3d ):
 
 #define FRAME_OFFSET 8
@@ -345,6 +348,7 @@ LLBL(x86_p3_3dr_done):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points3_3d_no_rot )
+HIDDEN(_mesa_x86_transform_points3_3d_no_rot)
 GLNAME( _mesa_x86_transform_points3_3d_no_rot ):
 
 #define FRAME_OFFSET 8
@@ -417,6 +421,7 @@ LLBL(x86_p3_3dnrr_done):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points3_2d )
+HIDDEN(_mesa_x86_transform_points3_2d)
 GLNAME( _mesa_x86_transform_points3_2d ):
 
 #define FRAME_OFFSET 12
@@ -494,6 +499,7 @@ LLBL(x86_p3_2dr_done):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points3_2d_no_rot )
+HIDDEN(_mesa_x86_transform_points3_2d_no_rot)
 GLNAME( _mesa_x86_transform_points3_2d_no_rot ):
 
 #define FRAME_OFFSET 12
@@ -564,6 +570,7 @@ LLBL(x86_p3_2dnrr_done):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points3_identity )
+HIDDEN(_mesa_x86_transform_points3_identity)
 GLNAME(_mesa_x86_transform_points3_identity ):
 
 #define FRAME_OFFSET 16

src/mesa/x86/x86_xform4.S

@@ -1,4 +1,4 @@
-/* $Id: x86_xform4.S,v 1.2 2002/03/07 21:40:08 brianp Exp $ */
+/* $Id: x86_xform4.S,v 1.3 2005/01/04 14:33:47 brianp Exp $ */
 
 /*
  * Mesa 3-D graphics library
@@ -66,6 +66,7 @@
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points4_general )
+HIDDEN(_mesa_x86_transform_points4_general)
 GLNAME( _mesa_x86_transform_points4_general ):
 
 #define FRAME_OFFSET 8
@@ -179,6 +180,7 @@ LLBL(x86_p4_gr_done):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points4_perspective )
+HIDDEN(_mesa_x86_transform_points4_perspective)
 GLNAME( _mesa_x86_transform_points4_perspective ):
 
 #define FRAME_OFFSET 12
@@ -261,6 +263,7 @@ LLBL(x86_p4_pr_done):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points4_3d )
+HIDDEN(_mesa_x86_transform_points4_3d)
 GLNAME( _mesa_x86_transform_points4_3d ):
 
 #define FRAME_OFFSET 12
@@ -363,6 +366,7 @@ LLBL(x86_p4_3dr_done):
 
 ALIGNTEXT16
 GLOBL GLNAME(_mesa_x86_transform_points4_3d_no_rot)
+HIDDEN(_mesa_x86_transform_points4_3d_no_rot)
 GLNAME(_mesa_x86_transform_points4_3d_no_rot):
 
 #define FRAME_OFFSET 12
@@ -443,6 +447,7 @@ LLBL(x86_p4_3dnrr_done):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points4_2d )
+HIDDEN(_mesa_x86_transform_points4_2d)
 GLNAME( _mesa_x86_transform_points4_2d ):
 
 #define FRAME_OFFSET 16
@@ -528,6 +533,7 @@ LLBL(x86_p4_2dr_done):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points4_2d_no_rot )
+HIDDEN(_mesa_x86_transform_points4_2d_no_rot)
 GLNAME( _mesa_x86_transform_points4_2d_no_rot ):
 
 #define FRAME_OFFSET 16
@@ -605,6 +611,7 @@ LLBL(x86_p4_2dnrr_done):
 
 ALIGNTEXT16
 GLOBL GLNAME( _mesa_x86_transform_points4_identity )
+HIDDEN(_mesa_x86_transform_points4_identity)
 GLNAME( _mesa_x86_transform_points4_identity ):
 
 #define FRAME_OFFSET 12